Split (1)

train · 1.47k rows

filename stringlengths 13 34	markdown stringlengths 0 359k
10.1261_rna.079608.123	Comparison of TRIBE and STAMP for identifying targets of RNA binding proteins in human and Drosophila cells KATHARINE C. ABRUZZI,1 CORRIE RATNER,1 and MICHAEL ROSBASH Howard Hughes Medical Institute, Brandeis University, Waltham, Massachusetts 02454, USA ABSTRACT RNA binding proteins (RBPs) perform a myriad of functions and are implicated in numerous neurological diseases. To iden- tify the targets of RBPs in small numbers of cells, we developed TRIBE, in which the catalytic domain of the RNA editing enzyme ADAR (ADARcd) is fused to an RBP. When the RBP binds to an mRNA, ADAR catalyzes A to G modifications in the target mRNA that can be easily identified in standard RNA sequencing. In STAMP, the concept is the same except the ADARcd is replaced by the RNA editing enzyme APOBEC. Here we compared TRIBE and STAMP side-by-side in human and Drosophila cells. The goal is to learn the pros and cons of each method so that researchers can choose the method best suited to their RBP and system. In human cells, TRIBE and STAMP were performed using the RBP TDP-43. Although they both identified TDP-43 target mRNAs, combining the two methods more successfully identified high-con- fidence targets. In Drosophila cells, RBP–APOBEC fusions generated only low numbers of editing sites, comparable to the level of control editing. This was true for two different RBPs, Hrp48 and Thor (Drosophila EIF4E-BP), indicating that STAMP does not work well in Drosophila. Keywords: RNA binding proteins; ADAR; APOBEC INTRODUCTION RNAs are bound by RNA binding proteins (RBPs), even in the nucleus before transcription is complete. For example, the association of RBPs with pre-mRNA affects all aspects of nuclear RNA processing, from capping, splicing, and 3′-end formation to nuclear export (for review, see Hocine et al. 2010). Additional RBPs bind to cytoplasmic mRNAs and determine their subcellular localization and stability (Das et al. 2021). Yet, other RBPs act to regulate translation in- cluding the association of specific mRNAs with the ribo- some (Babitzke et al. 2009). The coordinated functioning of these RBPs is necessary to generate the right amount of the correct protein, at the right time and in the right place. Many human neurological diseases are linked to RBP mutations. This vulnerability may be due in part to the fact that most central brain neurons are long-lived and can- not divide themselves out of trouble. For example, muta- tions in FMRP are responsible for Fragile X syndrome (Penagarikano et al. 2007), dysregulation of TDP-43 under- lies amyotrophic lateral sclerosis (ALS; Ling et al. 2013; Gao et al. 2019), and mutations in SMN1 cause spinal mus- cular atrophy (SMA; Farrar and Kiernan 2015). It is therefore 1 These authors contributed equally to this work. Corresponding author: [email protected] Article is online at http://www.rnajournal.org/cgi/doi/10.1261/rna .079608.123. Freely available online through the RNA Open Access option. important to understand the mRNA targets of key neuronal RBPs. For many years, crosslinking immunoprecipitation (CLIP) has been the tried-and-true method for identifying the tar- gets of RBPs. CLIP is a powerful method because UV cross- linking is used to biochemically attach the RBP to the target mRNA, which makes precise binding site identification pos- sible. However, CLIP also has drawbacks: an efficient anti- body is needed, crosslinking is biased to guanosine and thymidine residues, crosslinking can capture weak or tran- sient interactions, and a very large amount of biological ma- terial is usually required (for review, see Hafner et al. 2021; Xu et al. 2022). As we began to understand more about the heterogeneity of tissues and even single cells, it became apparent that a method that can identify RBP targets in small amounts of material was needed. In the past seven years, two methods have been devel- oped that allow researchers to examine the targets of RBP in small discrete groups of cells. The first was from our lab- oratory in 2016 and is called targets of RNA binding pro- teins identified by editing (TRIBE; McMahon et al. 2016). In TRIBE, an RBP of interest is fused to the catalytic domain of ADAR (ADARcd). When the RBP binds its target mRNA, the ADARcd edits the mRNA. In 2018, we incorporated a © 2023 Abruzzi et al. This article, published in RNA, is available under a Creative Commons License (Attribution 4.0 International), as de- scribed at http://creativecommons.org/licenses/by/4.0/. 1230 RNA (2023) 29:1230–1242; Published by Cold Spring Harbor Laboratory Press for the RNA Society single point mutation in the ADARcd, which substantially increased editing activity and decreased local sequence preferences (Kuttan and Bass 2012); this variant has been used for all recent TRIBE experiments (HyperTRIBE; Xu et al. 2018). Editing sites are then identified computation- ally from RNA sequencing data. Importantly, this allows target identification in very small numbers of cells and neu- rons without requiring biochemistry. TRIBE has been use- ful to a number of other labs working in different fields and model organisms, ranging from humans to malaria parasites (Liu et al. 2019; Alizzi et al. 2020; Nguyen et al. 2020; Arribas-Hernández et al. 2021; Cheng et al. 2021; Singh et al. 2021; van Leeuwen et al. 2022). A very similar strategy, conceptually and in overall design, was recently published in which an RBP is fused to a different editing enzyme, APOBEC1 (apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like). This method was used in two different studies and named STAMP (Brannan et al. 2021) and Dart-seq (Meyer 2019); we will use the nomenclature STAMP in this paper. TRIBE and STAMP both offer distinct advantages and disadvan- tages. ADAR is an adenosine deaminase and makes A-to-I edits within a supposedly double-stranded region even without the RNA binding regions normally present in ADAR (Macbeth et al. 2005). APOBEC1 is a cytidine deam- inase and therefore catalyzes C-to-T edits in single-strand- ed RNA but also edits single-stranded DNA (Rosenberg et al. 2011; Smith et al. 2012; Salter et al. 2016). ADAR and APOBEC both have some local sequence preferences (UAG → ADAR; A/U flanked → APOBEC; Rosenberg et al. 2011; Kuttan and Bass 2012). TRIBE only fuses the ADARcd, which can be cleanly separated from the ADAR double-stranded RNA binding regions to the RBP. The lat- ter normally directs the specificity of full-length ADAR and is replaced by the RBP in TRIBE. In contrast, the entire APOBEC protein is fused in STAMP, perhaps because it is uncertain how the editing specificity of APOBEC is defined. For example, APOBEC-mediated editing may require dimerization as well as cofactors (e.g., A1CF and RBM47; for review, see Smith et al. 2012; Fossat et al. 2014). Given the similarity of TRIBE and STAMP, it is perhaps uncertain which method one should choose to identify tar- gets of an RBP of interest. We therefore examined these methods side-by-side in HEK-293 and Drosophila cells to identify the pros and cons of each method experimentally. In HEK cells, the human ADAR2cd (referred to in this man- uscript as ADAR; 44 kDa) and rat APOBEC1 (referred to in this manuscript as APOBEC; 25 kDa) were fused to Tar DNA binding protein-43 (TDP-43). We previously success- fully used this protein for TRIBE in a mammalian system (Herzog et al. 2020). As indicated by published STAMP results (Brannan et al. 2021), APOBEC worked well in HEK cells. TDP-43- APOBEC generated a similar number of editing sites as TDP-43-ADAR, >10-fold more sites than expressing the Comparison of TRIBE and STAMP editing enzymes alone. TDP-43-ADAR and TDP-43- APOBEC both identified substantial numbers of target genes, although TDP-43-ADAR identified 50% more genes than TDP-43-APOBEC. Moreover, 70% of the TDP-43-APOBEC target genes were also identified by TDP-43-ADAR. We also compared STAMP and TRIBE in Drosophila, with two RBPs, Hrp48, and Thor (Drosophila EIF4E-BP). Targets of both proteins had been successfully identified with TRIBE (McMahon et al. 2016; Jin et al. 2020). To test STAMP in Drosophila, we replaced the ADARcd with APOBEC to generate Hrp48 and Thor STAMP. Although TRIBE worked well with both of these RBPs, as expected there was no substantial editing of tar- get mRNAs with STAMP. RESULTS To directly compare the efficacy of ADAR (TRIBE) with that of APOBEC (STAMP) in HEK-293 cells, we took advantage of two existing TDP-43-ADAR TRIBE constructs (Herzog et al. 2020): the cytomegalovirus (CMV) promoter ex- pressed either the TDP-43 coding sequence (cds) followed by the ADARcd or the ADARcd alone (pCMV-TDP-43- ADARcd and pCMV-ADARcd). Parallel constructs were generated in which the ADARcd was replaced with rat APOBEC1 as used in STAMP experiments (Fig. 1A; Brannan et al. 2021; pCMV-TDP-43-APOBEC and pCMV- APOBEC). These plasmids were transfected into HEK-293 cells along with a pCMV-eGFP plasmid. As a control, HEK cells were transformed with pCMV-EGFP only. Western blot analysis revealed that similar levels of TDP-43 fusion proteins were generated for TRIBE and STAMP, although TDP-43-APOBEC had slightly higher protein levels (Supplemental Fig. 1). Twenty-four hours after transfection, GFP-positive transfected cells were isolated using a Melody Fluorescence Activated Cell sorter (BD Melody FACS). RNA sequencing libraries were generated from two different bi- ological replicates using Smart-seq2 (see Materials and Methods). Existing TRIBE computational pipelines were used to identify both A to G editing events (TDP-43-ADAR and ADAR-only) as well as C to T editing events (TDP-43- in the RNA sequencing APOBEC and APOBEC-only) data (McMahon et al. 2016; Rahman et al. 2018; see Materials and Methods). In short, libraries were trimmed and mapped to the human genome (GRChg38.p13). Differential gene expression analysis was performed to verify that expression of TDP-43-ADAR and TDP-43- APOBEC did not cause dramatic changes to the transcrip- tomes of the HEK cells (Supplemental Fig. 2). To identify editing sites, the numbers of A, G, C, and T at every posi- tion in the genome were uploaded to a MySQL database. To be considered an editing site, a particular genomic lo- cation needed to be encoded by predominantly the non- edited base (adenosine [ADAR] or cytosine [APOBEC]) in www.rnajournal.org 1231 B Abruzzi et al. A C edit similar mRNA regions of all three genes. These editing clusters are often colocalized with putative TDP- 43 binding sites discovered by CLIP (bottom panel Fig. 1C; lines indicate known CLIP target sites; Hallegger et al. 2021). Expression of the editing enzymes alone (APOBEC, green; ADAR, orange; two biological repli- cates each) resulted in many fewer ed- iting sites on these target mRNAs (Fig. 1B,C). Although the number of RNA edit- ing sites generated by TDP-43-ADAR and TDP-43-APOBEC were quite sim- ilar, the two enzymes show different editing characteristics. First, express- ing APOBEC alone generated fivefold higher levels of editing than express- ing ADARcd alone (Fig. 1B). This may be because the entire APOBEC cod- ing sequence was used compared to only the catalytic domain of ADAR (see Discussion). Second, ADAR more often edits the exact same nucle- replicates otides in two biological (48% of sites are identical; another 17% are within 100 bp), whereas APOBEC is more likely to edit nearby nucleotides (only 31% of sites are iden- tical and 37% are within 100 bp; compare Fig. 2A,B). Third, ADAR gen- erates an overall higher level of editing at each site (Fig. 2C). The average per- centage editing of a TDP-43-ADAR site is 15.5% compared to 9.8% for TDP-43-APOBEC. This is undoubtedly related to the observation that ADAR often edits the same nucleotide in replicate experiments, that is, RBP–ADAR is more likely to edit the same A in anoth- er copy of the same transcript rather than nearby locations, which results in more reproducibility and a higher editing percentage. Fourth, TDP-43-ADAR yields fewer edited nu- cleotides per mRNA (mean of 4.8) compared to TDP-43- APOBEC (mean of 7; Fig. 2D). This observation is also because TDP-43-ADAR is more likely to edit the same nu- cleotide, which yields higher editing percentages at this lo- cation but fewer overall edited nucleotides on a target mRNA. We hypothesized that this more localized preference of ADAR may be because it has more restricted site choice rules than APOBEC. To compare these rules, we computa- tionally extracted the neighboring nucleotide 5′-and 3′ of the edited sites (Fig. 2E,F). Surprisingly, TDP-43-ADAR had a less restricted site choice than TDP-43-APOBEC (Fig. 2E,F). Eighty-four percent of TDP-43-APOBEC FIGURE 1. Both TDP-43 TRIBE (ADAR) and TDP-43 STAMP (APOBEC) identify candidate TDP-43 targets in HEK-293 cells. (A) Design of constructs for the expression of TDP-43- ADAR, TDP-43-APOBEC and their respective controls in HEK-293 cells. All transgenes were expressed using the cytomegalovirus (CMV) promoter. (B) The average number of editing sites identified in HEK-293 cells expressing TDP-43-ADAR, TDP-43-APOBEC, ADAR, or APOBEC alone. The graph quantifies A to G changes for ADAR and C to T changes for APOBEC. Two biological replicates are shown. Error bars indicate standard deviation. (C) Visualization of RNA editing sites generated by TDP-43-ADAR (red) and TDP-43-APOBEC (blue) on TARDBP (the gene encoding TDP-43), NET1 and CD59. The Integrated Genomics Viewer (IGV; Robinson et al. 2011; Thorvaldsdottir et al. 2013) visualization of two biological replicates of TDP-43-ADAR (red), TDP-43-APOBEC (blue), ADAR alone (orange), and APOBEC alone (green). Y-axis is 20% for all editing site tracks. RNA sequencing data is shown on top in gray; y-axis is 1000 fpkm. TDP-43-CLIP sites are shown in the bottom track; each line represents a CLIP site (Hallegger et al. 2021). The direction of transcription is denoted by the arrows on the refseq genes. Exons are shown as closed blocks and introns as lines. the HEK cells expressing only EGFP (see Materials and Methods). If this criterion was met, the same genomic loca- tion was examined in the experimental samples. To score as edited, a location required at least 20 reads and >6% A to G editing (ADAR) or C to T editing (APOBEC; see Materials and Methods). Expression of TDP-43-ADAR and TDP-43-APOBEC in HEK cells resulted in 35,000–45,000 editing sites in both bi- ological replicates of the sequencing libraries (Fig. 1B). The TDP-43-editing enzyme fusions generated substantially more editing than expressing the editing enzymes alone (in gray), showing that RNA editing events are substantially increased by fusing these enzymes to an RBP. As a prelimi- nary investigation of editing site location, we examined these sites on three TDP-43 target transcripts: a known tar- get of TDP-43 (TDP-43 mRNA itself; the gene encoding this mRNA is TARDBP; Ayala et al. 2011; Polymenidou et al. 2011) as well as two putative targets, NET1 and CD59 (Fig. 1C). TDP-43-ADAR (red; two biological replicates) and TDP-43-APOBEC (blue; two biological replicates) 1232 RNA (2023) Vol. 29, No. 8 Comparison of TRIBE and STAMP A C E B D F FIGURE 2. TDP-43 TRIBE and TDP-43 STAMP identify editing sites with different characteristics on TDP-43 target transcripts. (A,B) Pie charts showing the relationship between editing sites in two biological replicates. Editing sites identified in both experiments at: identical locations (dark blue), two different locations that are in close proximity in the biological replicates (green and gray; 50–100 bp), within 200 bp (yellow), or >200 bp (dark red). (C) Mean percentage editing for all editing sites identified in TDP-43-ADAR (∼15%) and TDP-43-APOBEC (∼10%). Mean values indicated by X and median values by lines (P-value <0.0001; Wilcoxon rank-sum test). (D) Mean editing sites per transcript generated by TDP-43-APOBEC (∼7) and TDP-43-ADAR (∼5) (mean indicated by X and median indicated by a line; P-value <0.0001; Wilcoxon rank-sum test). (E,F) The near neighbors of all edited nucleotides were identified and quantified (percentage of all editing sites). TDP-43-ADAR (E) preferentially edited adenosines followed by guanosine. TDP-43-APOBEC (F ) preferentially edited cytosines that were flanked by A or T. editing events occurred on a C flanked by only A or T (Fig. 2F; ACA, ACT, TCT, and TCA). Editing sites were rarely ob- served on cytosines with a 5′-C or 5′-G. TDP-43-ADAR also showed specific site choice but to a lesser extent: 64% of the TDP-43-ADAR editing sites were concentrated on AAG, CAG, TAC, and TAG (Fig. 2E). These observations are consistent with previous work showing that the ADAR2-E488Q catalytic domain prefers T or A 5′-of the editing sites (Kuttan and Bass 2012). The higher restrictive site choice of APOBEC suggests that the localized prefer- ences of ADAR is due to RNA secondary structure (see Discussion). To accommodate these differences in editing between TDP-43-ADAR and TDP-43-APOBEC and to make a pipe- line useful for both enzymes, we made several modifica- tions to our previous TRIBE bioinformatics pipeline. To account for the lower reproducibility of the APOBEC edit- ing sites, we required an APOBEC editing site to be within 100 bp of a second site in the biological replicate instead of requiring the same site to be identified in two biological replicates as done for ADAR. In addition, we ignored any nucleotide edited at a frequency above 1% in either repli- cate of the enzyme-only samples. This conservative filter- ing step accommodates the higher background editing www.rnajournal.org 1233 Abruzzi et al. by the APOBEC-only construct and has been previously used in mammalian cells to ensure that editing sites iden- tified with fusion proteins are not false positives (Biswas et al. 2020). After incorporating these two additional filter- ing steps, a final set of editing sites and target mRNAs were identified for TDP-43-ADAR and TDP-43-APOBEC (Fig. 3A). Although a nearly identical number of editing sites were identified for the two fusion proteins, TDP-43- APOBEC generated more sites per mRNA and therefore fewer target transcripts (Fig. 3B). CLIP experiments indicate that TDP-43 binds to introns as well as to the 3′-UTRs of target transcripts (Polymenidou et al. 2011; Hallegger et al. 2021). Since the Smart-seq libraries generated in this study are generat- ed from poly(A) mRNA, introns should and do represent a very small portion of sequenced transcripts. Editing sites were however enriched in the 3′-UTRs of target transcripts. This was shown by mapping TDP-43-ADAR or TDP-43- APOBEC editing sites as well as TDP-43 CLIP binding sites (Hallegger et al. 2021) to the genome and calculating the B A C percentage of sites within the 5′-UTR, cds, and 3′-UTR. This distribution was compared to the transcriptome distribu- tion and graphed as a fold change relative to that distribu- tion (Fig. 3C). All three samples show strong enrichment in 5′-UTRs as well as 3′-UTRs with an underrepresentation of sites in the cds. A higher percentage of TDP-43-ADAR editing sites map to the cds than those from TDP-43-APOBEC and TDP-43-CLIP. However, if the coordinates of the TDP-43- ADAR editing site are expanded by 50 bp, 16% of the TDP-43-ADAR editing sites shift from being the cds to the 3′-UTR (Fig. 3C; right). It is therefore possible that many of these binding events occur in the 3′-UTR near the cds-3′-UTR border. The 3′-UTR and 5′-UTR enrichment of TDP-43-ADAR, TDP-43-APOBEC, and TDP-43-CLIP sites are quite similar. To determine whether these three methods identify the same binding regions, we expanded each editing or CLIP site by 100 bp in both directions and then compared the three regions (Fig. 4A,B). Forty-seven percent of the TDP-43-APOBEC and 30% of the TDP-43-ADAR sites are in the same genes and overlapping (Fig. 4A; blue), whereas the other edited re- gions were unique to TDP-43- APOBEC and TDP-43-ADAR (Fig. 4A; green). We then asked whether editing sites that are present in both TDP-43-ADAR and TDP-43-APOBEC (common sites) are more likely to also be identified by CLIP than sites found by either TDP-43 TRIBE or STAMP (unique sites). Indeed, there is a three- to fourfold increase in the percentage of common sites that are also identified in CLIP experiments (blue; Fig. 4B). FIGURE 3. TDP-43-ADAR and TDP-43-APOBEC generated similar numbers of editing sites that were preferentially located in the 3′-UTR and 5′-UTR. (A) Quantification of the number of editing sites generated by TDP-43-ADAR and TDP-43-APOBEC (A to G changes are indicat- ed for ADAR and C to T changes for APOBEC). This graph shows only editing sites that were consistent between two biological replicates not found in enzyme-only controls. The number of editing sites identified was normalized per million reads to adjust for sequencing library depth. (B) The number of target transcripts identified by TDP-43-ADAR and TDP-43- APOBEC. (C) The distribution of editing sites in the 5′-UTR, coding sequence (cds), and 3′- UTR was calculated for TDP-43-ADAR and TDP-43-APOBEC as well as TDP-43-CLIP. The graphs show the fold enrichment relative to the read distribution of the RNA sequencing li- brary. In all cases, the 3′-UTR and 5′-UTR enrichment is significant using a proportion test; P- value <0.0001. CLIP data is from Hallegger et al. (2021). (Right) TDP-43-ADAR editing sites were shifted by 50 bp toward the end of the transcripts and the localization of the sites was reexamined. There was a 16% increase in the number of sites in the 3′-UTR. 1234 RNA (2023) Vol. 29, No. 8 Because RBP–ADAR fusions often edit nucleotides as far as 500 nt from the RBP binding site (Xu et al. 2018), TDP-43-ADAR and TDP-43-APOBEC may identify similar transcripts even if their editing sites are not within 200 bp of one another. We therefore compared the lists of mRNAs identi- fied as TDP-43 targets by TDP-43- ADAR and TDP-43-APOBEC: 73% of the TDP-43-APOBEC transcripts and 46% of the TDP-43-ADAR transcripts are identified by both methods (Fig. 4C; blue). Because TDP-43-ADAR and TDP- 43-APOBEC identify a substantially overlapping set of the two methods can be combined to transcripts, A C B D Comparison of TRIBE and STAMP Comparison of TRIBE and STAMP in Drosophila We decided to follow the success of combining the ADAR and APOBEC editing enzymes to identify higher confidence TDP-43 targets in mam- malian cells by testing the two en- zymes side-by-side in Drosophila Schneider 2 cells (Drosophila S2 cells). To this end, we adopted the same strategy used above in mammalian cells and replaced the ADARcd with the rat APOBEC1 cds. TRIBE was orig- inally developed in Drosophila, and we assayed the same two RBPs previ- ously used to identify TRIBE targets in S2 cells, namely, Hrp48 and Thor (Drosophila EIF4E-BP; McMahon et al. 2016; Jin et al. 2020). This result- ed in a panel of plasmids that con- tained the metallothionein inducible promoter (MT) driving an RBP-editing enzyme fusion followed by a p2A self-cleaving peptide and dsRed for vi- sualization (Fig. 5A). We also generat- ed enzyme-only and dsRed only controls using the same general strat- egy. We transfected these expression plasmids into Drosophila S2 cells and isolated dsRed-positive cells via FACS (BD Melody). Poly(A) mRNA was isolated from 400 cells and was used as input for RNA sequencing li- braries generated using Smart-seq2 (Picelli et al. 2014). RNA editing sites were identified using the new TRIBE computational pipeline described above adapted for Drosophila (see Materials and Methods). As previously shown, expression of both Hrp48-ADAR and Thor-ADAR dramatically increased the number of edit- ing sites above the low level of editing achieved by express- ing the ADARcd alone (Fig. 5B; orange and light gray; McMahon et al. 2016; Jin et al. 2020). In contrast and unlike in human cells (Fig. 1B), expression of Hrp48-APOBEC and Thor-APOBEC did not cause an increase in editing above enzyme-only levels (Fig. 4B; blue and dark gray). This was not due to the instability of the APOBEC fusion proteins; western blotting could easily detect Hrp48-APOBEC (Supplemental Fig. 3). Moreover, the few C to T transitions appear due to APOBEC; these editing events are much more prevalent than other nucleotide changes and show the characteristic site choice of this enzyme (flanking A or T nucleotides; Supplemental Fig. 4). This suggests that both RBP–APOBEC fusions only inefficiently edit target Indeed, the mRNAs in Drosophila tissue culture cells. www.rnajournal.org 1235 FIGURE 4. TDP-43-ADAR and TDP-43-APOBEC identified common edited regions and tran- scripts. (A) The coordinates of identified editing sites found in TDP-43-ADAR and TDP-43- APOBEC was expanded by 100 bp in each direction and examined for overlapping regions (>1 bp was considered overlapping). Edited regions found in both TDP-43 TRIBE and TDP- 43 STAMP are indicated in blue, and those editing regions that were unique to either TDP- 43-ADAR or TDP-43-APOBEC are shown in green. (B) The editing sites identified by TDP- 43-APOBEC and TDP-43-ADAR were split into two groups: common sites (those found by both TDP-43-ADAR and TDP-43-APOBEC; blue) or unique sites (found only by one of the ed- iting enzymes; green). These two subsets of editing sites were then overlapped with TDP-43 CLIP sites (any overlap >1 bp was considered overlapping). The percentage of the editing sites overlapping regions containing CLIP peaks is shown. (C) The transcripts identified by TDP-43- ADAR and TDP-43-APOBEC were compared. The graph indicates the percentage of tran- scripts that are found by both TRIBE and STAMP (blue) or were unique to either TRIBE or STAMP (green). (D) The 100 bp region surrounding RNA editing sites generated by TDP-43- ADAR, TDP-43-APOBEC, and TDP-43-CLIP sites was analyzed for binding motifs using Xstreme (Grant and Bailey 2021). The table indicates the percentage of edited regions that contain GU/GT-rich motifs with high significance. identify a higher confidence set of TDP-43 targets; this is also suggested by the increased overlap between CLIP and the common edited regions. We therefore examined the 100 bp regions surrounding TDP-43-ADAR, TDP-43- APOBEC, and TDP-43-CLIP sites for the GU-rich motif that is associated with TDP-43-binding (Polymenidou et al. 2011; Tollervey et al. 2011). Since the GUGUGU or GTGTGT motif was originally identified in CLIP experi- ments, we first examined the 100 bp region surrounding TDP-43-CLIP sites. Fifty-seven percent of these sites con- tain a GTGTGT motif, a highly significant enrichment (Fig. −1867). The TDP-43-ADAR and TDP-43- 4D; q = 6.7 × 10 APOBEC identified regions were also significantly enriched for GTGTGT motifs; a similar percentage of these regions, 56% and 62%, contain GTGTGT motifs (Fig. 4D). For re- gions identified by both methods, the GTGTGT motif was identified in 86% of the binding regions with even higher statistical significance (Fig. 4D). This suggests that higher confidence binding regions can be identified by combining TRIBE and STAMP. Abruzzi et al. number of RBP–APOBEC editing sites and mRNA targets that passed threshold (consistent between the two biolog- ical replicates and not edited at >1% editing in the enzyme- only controls) was more than 10-fold less than the same RBPs fused to ADAR (Fig. 5C,D). is significantly higher The experiments in human cells shown above indicated that RBP–APOBEC fusions were likely to edit more sites but with a lower percentage of editing (Fig. 2C). This same trend is observed in Drosophila; the mean editing percentage on Hrp48-APOBEC editing sites is 8.8%, while the mean editing of Hrp48-ADAR sites (26.1%; Supplemental Fig. 5). To test the possibility that requiring 6% editing was preventing the identification of RBP– APOBEC editing sites, we reanalyzed the data with a 4% ed- iting cutoff (Supplemental Fig. 6A). Although the number of editing sites identified by Thor-APOBEC and Hrp48- APOBEC increased approximately threefold, the number of APOBEC-only editing sites increased 15-fold. This suggests that a too high editing threshold is not the reason for the weakRBP–APOBECeditingrelativetoAPOBEC-onlyediting. Another possible explanation for the lack of APOBEC ed- iting in Drosophila S2 cells was proposed by a recent paper posted on bioRxiv (Doll et al. 2022). It indicates that APOBEC deaminase activity is poor at temperatures com- monly used for S2 cells (18°–24°) but functional at 29°C; to test this possibility, we repeated the Hrp48-Apobec ex- periment in S2 cells grown at 28°C. Raising the temperature did not increase APOBEC editing activity when fused to Hrp48 in Drosophila cells; on the contrary, overall editing levels by APOBEC alone and by Hrp48-APOBEC were fur- ther reduced at 28°C (Supplemental Fig. 6B). These results suggest that STAMP may not be a viable option for detect- ing RBP target mRNAs in nonmammalian systems (see Discussion). DISCUSSION RNA binding proteins guide RNAs throughout their life, by binding to nascent RNAs as they emerge from the poly- merase, by facilitating the removal of introns and nuclear export, and then by modulating mRNA turnover, localiza- tion, and translation in the cytoplasm (Babitzke et al. 2009; Hocine et al. 2010; Das et al. 2021). Mutations in over 1000 RBPs are linked to human diseases including Fragile X and ALS (for review, see Gebauer et al. 2021). Therefore, it is critical to have reliable tools to identify in specific cell types and the mRNA targets of RBPs, even single cells. To this end, we directly compared in this manuscript TRIBE (RBP–ADAR) and STAMP (RBP– APOBEC) in both human cells and in Drosophila cells. In human cells, TDP-43 targets were successfully identified using both methods, and ∼70% of the STAMP targets were identified by TRIBE. The results also indicated that a higher confidence set of RBP targets is identified by de- fining the common targets identified with both methods. 1236 RNA (2023) Vol. 29, No. 8 Consistent with our finding that TRIBE and STAMP both work well and similarly is a very recent study that utilized both TRIBE and STAMP simultaneously, which the authors dubbed TRIBE-STAMP (Flamand et al. 2022). In this study, Flamand et al. investigated the mRNA targets of the m6A reader proteins, YTHDF1, YTHDF2, and YTHDF3 and showed that these three proteins identified similar target mRNAs with both TRIBE and STAMP. Although the overlap between these two methods was greater than observed here, this may be because the m6A reader proteins bound a very large percentage of the transcriptome, perhaps gen- erating a higher likelihood of common targets. These au- thors also showed that TRIBE and STAMP can be combined to determine if two RPBs bind to the same tran- script, another application of using both TRIBE and STAMP in parallel. Although we were hopeful STAMP would also become a second tool for use in Drosophila, we were unable to observe significant RNA editing by APOBEC fused to two different RBPs in Drosophila S2 cells. The Drosophila constructs mirror those successfully used in human cells and in the original STAMP study, that is, APOBEC is fused to the carboxyl termi- nus of the RBP. The STAMP fusion proteins are transcribed and translated, and some APOBEC-derived editing is still de- tected (Fig. 5B; Supplemental Fig. 3). However, the Thor- APOBEC and Hrp48-APOBEC editing levels are similar to those observed in the APOBEC-only controls (Fig. 5B). This poor editing is not due to reduced enzyme activity at lower temperatures; similar results were obtained at 28°C (Supplemental Fig. 6B). These results indicate that STAMP is not a good choice for RBP target identification in Drosophila and perhaps in other invertebrate systems. Does either TRIBE or STAMP pose a significant advan- tage for RBP target identification in mammalian cells? Does one method identify substantially more false nega- tives or false positives than the other? As far as false negatives are concerned, each enzyme has intrinsic features that bias the editing of an RBP-bound RNA. TDP-43-APOBEC has a strong nearest neighbor pref- erence as nearly all cytosines edited are flanked by A or T (84%; Fig. 2F). A previous study argued that APOBEC is ide- al for RBP target mRNA identification. This is because it can edit cytosines in single-stranded mRNA, which should be between 25% and 35% of nucleotides in mammalian tran- scripts (Brannan et al. 2021). However, the strong nearest neighbor preference observed both here and in a previous study (Rosenberg et al. 2011) suggests that RBP–APOBEC fusion proteins edit a more limited set of cytosines. None- theless, the observation that TDP-43-APOBEC tends to edit different cytosines on the same transcript or multiple copies of the same transcript suggests that this nearest neighbor preference is not an issue for the identification of most mRNA targets and target regions. In contrast, TDP-43-ADAR is more likely to edit the same adenosine resulting more often in the editing of the exact A C B D Comparison of TRIBE and STAMP ing the E488Q mutation (Xu et al. 2018). This HyperTRIBE method gives rise to much more editing, due to fast- er editing speed and less preference for an UAG neighboring sequence surrounding the editing site (Kuttan and Bass 2012). In fact, as shown here and elsewhere, TRIBE identifies target transcripts with editing sites containing from 1 to 43 nt and also edits cds (Fig. 2D; Cheng et al. 2021). What about false positives, editing sites that do not reflect RBP mRNA binding? TRIBE only uses the ADAR catalytic domain (ADARcd), avoiding its RNA binding regions. Probably as a consequence, the ADARcd alone generates a low level of A to G editing, suggesting that most TDP-43-ADARcd editing sites and target transcripts are bona fide positives. In contrast, the cat- alytic and RNA binding regions of APOBEC are not defined, requiring use of the entire protein in STAMP. This likely explains the fivefold increase in editing by APOBEC-only compared to ADAR-only (Fig. 1B). FIGURE 5. STAMP does not work well in Drosophila S2 cells. (A) Schematic of the expression constructs used to test TRIBE and STAMP in Drosophila S2 cells. (B) The average number of editing sites identified using Hrp48 and Thor TRIBE (orange) and STAMP (blue) in two biolog- ical replicates. ADAR and APOBEC-only controls are also shown to illustrate the background editing of each enzyme (gray). Error bars indicate standard deviation. (C ) Editing sites identi- fied in both biological replicates and not found in enzyme-only controls were quantified and normalized relative to the total number of reads in the RNA sequencing library. (D) The number of target transcripts identified by applying TRIBE and STAMP fused to Hrp48 and Thor are indicated. same nucleotide in multiple experiments (Fig. 2A). Since the ADARcd has less stringent nearest neighbor preferenc- es than APOBEC (Fig. 2E,F), the repeat editing of specific adenosines by the ADARcd is more likely due to its dou- ble-stranded RNA requirement (Macbeth et al. 2005) and may be responsible for the fewer TRIBE edits compared to STAMP edits: an average of 4.8 sites/transcript for the ADARcd versus 7 sites/transcript for APOBEC (Fig. 2D). Despite this modest difference, TRIBE identifies somewhat more target transcripts than STAMP despite being ex- pressed at lower levels (Fig. 3B; Supplemental Fig. 1). This indicates that the ADARcd double-stranded RNA re- quirement is not an obstacle to target identification as pre- viously discussed (McMahon et al. 2016) and that TRIBE does not suffer from a severe false negative problem rela- tive to STAMP. This previous study has also purported that TRIBE is not well suited for RBP target identification compared to STAMP because the ADARcd only generates very few ed- its, only one to two per target transcript, and that TRIBE is incapable of editing coding regions (cds) due to their sin- gle-stranded nature (Brannan et al. 2021). This comment was based on results with Fmr1 using the first version of TRIBE prior to the adoption in 2018 of an ADARcd contain- The ability of APOBEC to edit sin- gle-stranded DNA could also impact the level of detectable background editing. Indeed, rat APOBEC1 used here and in the original STAMP study has also been used successfully in combination with CRISPR for genome edit- ing (Komor et al. 2016, 2017). Other studies have shown that APOBEC1 can drive off-target DNA editing and RNA editing even when fused to Cas9 (Grunewald et al. 2019; Jin et al. 2019; Kanca et al. 2019; McGrath et al. 2019). This observation was true even with transient trans- fection of the Cas9–Apobec fusions (McGrath et al. 2019). To date, genomic DNA has not been examined in STAMP, making it uncertain whether APOBEC-driven single- stranded DNA editing is contributing to false positive ed- iting sites and transcripts. Although current computational approaches cannot distinguish DNA editing from RNA ed- iting, we attempted to ameliorate this issue by eliminating from the final list of editing sites any nucleotide that has >1% editing in enzyme-only control samples. Other possible sources of false positive editing sites and target transcripts are shared by TRIBE and STAMP. First, these methods currently rely on overexpression, which can cause RBPs to bind to and edit secondary, low efficien- cy targets as well as primary high efficiency targets. Second, there are false-positive sources of editing that are intrinsic to living cells, endogenous editing by ADAR and APOBEC as well as genomic variation in the form of www.rnajournal.org 1237 Abruzzi et al. sNPs. Our computational pipeline is set up to handle these issues and is also comprehensive for both methods. Unlike computational approaches that identify editing sites by comparing the experimental RNA sample to publicly avail- able genomic sequences, we directly compare RNA se- quences from TRIBE and STAMP samples to the same cell line expressing only GFP and therefore eliminate most of these false positive sources, i.e., sites that are ed- ited by endogenous ADAR and APOBEC as well as sNPs line (see Materials and are present in the control cell Methods). A third source of false positives is technical er- ror, whether from PCR during RNA sequencing library gen- eration or from sequencing itself. To control for this, we require edited nucleotides to have 20 reads of coverage with greater than one edited read to ensure that a candi- date altered nucleotide is genuine. Although this require- ment requires additional sequencing depth to detect editing events in less abundant transcripts (at least 12 mil- lion uniquely mapped reads for each sample), it prevents the identification of an editing site due to a single A to G (ADAR) or C to T (APOBEC) change that could be due to technical error. Although our previous TRIBE analyses have been done with editing percentage cutoffs as low as 5% (Biswas et al. 2020; Herzog et al. 2020) and as high as 10% (McMahon et al. 2016; Xu et al. 2018), we used in this work an editing cutoff (6%) dictated by the low- er STAMP editing percentages as well as the by similar pa- rameters used in the previous STAMP study (Brannan et al. 2021). In summary, our pipeline is generally conservative and designed to reduce the contribution of false positives while maintaining bona fide targets. The topics of false positives and false negatives warrant a return to considering the lack of overlap between TRIBE, STAMP, and CLIP (Fig. 4). There is unfortunately no ground truth; just like TRIBE and STAMP, CLIP is also prone to false positives and false negatives (for review, see Xu et al. 2022). One conservative approach is to identify a set of high-confidence targets by overlapping methods. A posi- tive hit in more than one method will decrease the likeli- hood that it is a false positive. This should be possible in mammalian systems by performing TRIBE and STAMP and moving forward with common targets. This will be es- In pecially useful when CLIP is not easy to apply. Drosophila and in other systems in which applying more than one method is not feasible, the use of multiple bio- logical replicates is still helpful in identifying the most con- sistent and reproducible RBP targets. MATERIALS AND METHODS Plasmids The details of human and Drosophila plasmid construction are be- low and listed in Supplemental Table 1. For all plasmids, we used either Q5 (NEB) or Ex Taq (TaKaRa) DNA polymerases to PCR am- 1238 RNA (2023) Vol. 29, No. 8 plify inserts. All inserts and vectors were gel purified using the Gel Extraction Kit (Qiagen). We inserted all fragments into vectors us- ing either Gibson Assembly (NEB) or NEBuilder (NEB) unless oth- erwise noted. All plasmids were transformed into NEBalpha DH5 high efficiency competent cells. We verified candidate plasmids using colony PCR using primers listed in Supplemental Table 2 and REDTaq ReadyMix (Sigma-Aldrich). All plasmids were mini- prepped (Qiagen) and sequenced with Plamidsaurus (https ://www.plasmidsaurus.com). HEK-293 expression vectors Plasmids for expressing TDP-43-ADAR (pCMV-hADARcd-E488Q) as well as the ADAR-only control (pCMV-hADARcd-E488Q) were previously published (Herzog et al. 2020). To generate pCMV- APOBEC (pCR25), we amplified rat APOBEC1 from pMT- Hrp48-APOBEC-P2A-dsRed (pCR8; see below) with primers CR60 and CR61 and inserted it into pCMV-ADARcd-E488Q di- gested with EcoRI and KpnI. To clone pCMV-TDP-43-APOBEC (pCR27), we amplified APOBEC from pMT-Hrp48-APOBEC- P2A-dsRed (pCR8) with primers CR73 and CR74 and inserted it into pCMV-TDP-43-ADARcd-E488Q linearized using PCR with primers CR70 and CR71. Drosophila plasmids A Drosophila HyperTRIBE plasmid was generated by including a self-cleaving peptide, p2A, followed by dsRed downstream from the Hrp48-ADAR sequence (pMT-Hrp48-ADAR-E488Q-P2A- dsRed; pCR1). This plasmid and all related plasmids have a V5- tag 5′ of the editing enzyme. The plasmid pMT-Tyf-ADAR- E488Q-P2A-dsRed was digested with PmeI and NotI to liberate ADAR-E488Q-P2A-dsRed. This fragment was ligated into PmeI and Not1-digested pMT-Hrp48-ADAR-E488Q using T4 ligase (NEB). pMT-Hrp48-APOBEC-P2A-dsRed (pCR8) was made by am- plifying rat APOBEC1 from pCMV-BE1 (Addgene #73019) using primers CR26 and CR27. Gibson Assembly was then used to insert APOBEC1 into pMT-Hrp48-ADAR-E488Q-P2A-dsRED (pCR1) di- gested with ApaI and NotI to remove ADAR-E488Q. The resulting plasmid was then cut with NotI to insert a linker (made by annealing primers CR29 and CR30) using Gibson Assembly. To clone the (pMT-APOBEC-p2A-dsRed; pCR10), we APOBEC-only control amplified APOBEC from pCMV-BE1 (Addgene #73019) using primers CR28 and CR27. Plasmid pMT-Hrp48-ADAR-P2A-dsRed (pCR1) was digested with ApaI and KpnI to liberate Hrp48-ADAR and APOBEC was inserted using Gibson Assembly. To clone an ADAR-only control (pMT-ADAR-P2A-dsRed; pCR12), we digested pMT_Hrp48_ADAR_E488Q_P2A_dsRed (pCR1) with NotI and KpnI to remove Hrp48. We blunted the resulting sticky ends using Klenow end-blunting (NEB) and recircularized the plasmid using T4 blunt-end ligation (NEB). To clone pMT-Thor-APOBEC-P2A- dsRed (pCR26), we amplified Thor from pMT-Thor-Linker- HyperTRIBE (Jin et al. 2020) with primers CR66 and CR67 and in- serted it into pMT-Hrp48-Linker-APOBEC-P2A-dsRed (pCR8) di- gested with KpnI and NotI to remove Hrp48 using NEBuilder (NEB). To clone pMT-Thor-ADAR-P2A-dsRed (pCR30), we ampli- fied Thor from pMT-Thor-Linker-HyperTRIBE (Jin et al. 2020) using primers CR80 and CR81 and inserted it into pMT-Hrp48- ADAR-E488Q-P2A-dsRed (pCR1) digested with KpnI and NotI to remove Hrp48. Insertion was done using NEBuilder. Cell culture HEK293T cells (ATCC CRL-3216) were cultured at 37°C in Gibco DMEM, high glucose, GlutaMAX supplement, pyruvate (Thermo Fisher) with 10% Fetalgro synthetic FBS (RMBio, FGR-BBT) and 1% Penicillin–Streptomycin (Genesee Scientific). HEK cells were transiently cotransfected with pCMV-EGFP and the relevant plas- mid of interest in a six well plate using the Lipofectamine3000 protocol (Thermo Fisher). The CMV promoter is constitutively ex- pressed and 24 h after transfection, GFP-positive cells were col- lected with BD FACS Melody. Drosophila S2 cells were cultured at 23°C or 28°C in Schneider’s media with 10% Gibco HI-FBS (Thermo Fisher) and 1% Gibco Antibiotic-Antimycotic (Thermo Fisher). Drosophila S2 cells were transiently transfected using the Mirus TransIT- 2020 transfection protocol (Mirus Bio) in a six well plate for ∼24 h. Cells were induced for 24 h by inducing the metallothionein promoter (pMT) using 0.5 mM copper sulfate (Sigma). DsRed- positive cells were collected with the BD FACS Melody. RNA library preparation Four-hundred cells were sorted directly into aliquots of 100 µL ly- sis buffer (Invitrogen Dynabeads mRNA Direct Kit; Thermo Fisher). Poly(A)-plus RNA was isolated using Invitrogen Dynabeads mRNA Direct Kit (Thermo Fisher, 61012) and se- quencing libraries were prepared following Smart-Seq2 opti- mized protocol (Picelli et al. 2014). cDNA was quantified with D5000HS TapeStation (Agilent) and final tagmented libraries were quantified on D1000HS TapeStation (Agilent). Libraries were sequenced on Illumina NextSeq500 with 75 cycle High Output Kit v2.5. TRIBE pipeline To identify editing sites, the TRIBE pipeline was used as previous- ly described (Rahman et al. 2018; https://github.com/rosbashlab/ HyperTRIBE). Briefly, custom scripts were used to trim (Cutadapt; Martin 2011) and align (STAR; Dobin et al. 2013) reads to the ap- propriate genome (Human [GRChg38.p13] or Drosophila [dm6]). Mapped reads were used to generate gene expression data (see below) as well as bigwig files for visualization on the Integrated Genomics Viewer (IGV) (Robinson et al. 2011). The mapped reads were then converted to a matrix file listing the number of As, Gs, Cs, and Ts found in sequencing reads at each genomic coordi- nate. This data was loaded into a MySQL database for efficient querying. Editing sites were identified by identifying coordinates in the control samples (EGFP control for HEK-293 cells and the dsRed control for Drosophila S2 cells) that meet three criteria: the reads were an A (ADAR) or a C (1) At (APOBEC), (2) <0.5% of the reads at that location were the edited base, i.e., a G (ADAR) or a T (APOBEC), and (3) there were at least nine reads covering the location. If these three conditions were met, then these coordinates were examined in the experimental samples. The coordinate is considered an editing site if: (1) there are at least 20 reads in the experimental sample, and (2) >6% of the reads have an edited base (G [ADAR] and T [APOBEC]) at that location (see Rahman et al. 2018 and custom scripts). Note that although our original TRIBE studies used a 10% editing cutoff least 80% of Comparison of TRIBE and STAMP (McMahon et al. 2016), more recent TRIBE studies have imple- mented a lower editing threshold of 5% (Biswas et al. 2020; Herzog et al. 2020). We opted for a 6% editing cutoff for these studies so that our analysis was consistent with the previous STAMP analysis (Brannan et al. 2021). Supplemental Table 3 illus- trates the effect of each filter on the total number of editing sites identified. By comparing experimental samples to a control RNA sample from the same cell line rather than publicly available ge- nomic sequences, the majority of sNPs and endogenous editing events are filtered out in the initial editing site identification step. Once editing sites were identified, a final list of editing sites for each TRIBE-RBP and STAMP-RBP was generated by identifying editing sites consistent between two biological replicates for RBP–ADAR and within 100 bp of each other in two biological rep- licates RBP–APOBEC. Both approaches utilized bedtools intersect and bedtools closest (Quinlan 2014). Editing sites with at least 1% editing in the editing enzyme-only control samples were identified using the custom script (Threshold_editsites_20reads.py) with ed- iting threshold set to 0.01. The list of all sites identified with >1% editing in either biological replicate of the enzyme-only control li- braries was then subtracted from the list of putative editing sites identified in the experimental samples using bedtools. The number of editing sites per transcript was determined using a custom script (summarize_results.pl) which generates a list of all transcripts, the number of editing sites in that transcript (Rahman et al. 2018). Quantification of the distribution of RNA sequencing reads in the human and Drosophila RNA libraries was performed with read_distribution.py script in RSeQC4.0.0 (Wang et al. 2012). The distribution of editing sites between the 5′-UTR, cds, and 3′-UTR was determined using bedtools and custom scripts and compared to the distribution of reads in the sequencing li- brary. Near neighbor preference was determined using custom scripts. To examine the overlap of TDP-43-TRIBE, TDP-43-STAMP, and TDP-43-CLIP binding regions, the editing or CLIP site was ex- panded by 100 bp in both directions using bedtools slop. Then the overlap (>1 bp) between the different regions was determined us- ing bedtools intersect. All custom scripts generated for this study are annotated and deposited at https://github.com/rosbashlab/ Comparison-of-TRIBE-and-STAMP. To examine TDP-43-TRIBE, TDP-43-STAMP, and TDP-43-CLIP sites for the presence of GU-rich motifs preferentially bound by TDP-43, the region surrounding the editing site or CLIP site was ex- panded by 50 bp using bedtools slop. The resulting bed file was used to retrieve sequence information for these regions to use as input for motif analysis using Xstreme (Grant and Bailey 2021). CLIP data analysis CLIP data was downloaded from ArrayExpress with accession number E-MTAB-9436 (Hallegger et al. 2021). CLIP data from two biological replicates was analyzed using iCount at iMAPS (https://imaps.goodwright.com/history/). CLIP peaks identified in both biological replicates were identified using bedtools and used as a high-confidence set of putative TDP-43-CLIP sites. Gene expression analysis Raw gene count values were generated using HTseq.scripts. count during initial mapping of RNA sequencing (Anders et al. www.rnajournal.org 1239 Abruzzi et al. 2015). Two replicates of either TDP-43-ADAR or TDP-43- APOBEC were compared to two replicates of the negative control cells (HEK-293 cells expressing only EGFP). Prior to differential ex- pression analysis, data was filtered to remove transcripts that were expressed at lower levels (expressed at less than 5 FPKM reads in more than two of the four samples). Differential gene expression upon overexpression of TRIBE and STAMP RBPs was analyzed us- ing EdgeR (Robinson et al. 2010). The resulting smearplots are shown. Western blot analysis HEK cells transfected with pCMV-TDP-43-hADAR-E488Q, pCMV-3HA-hADAR-E488Q, pCMV-TDP-43-APOBEC, pCMV- APOBEC, and pCMV-EGFP were lysed in RIPA buffer (10 mM HEPES pH 7.5, 5 mM Tris pH 7.5, 50 mM KCl, 10% glycerol, 2 mM EDTA, 1% Triton X-100, 0.4% NP-40, and protease cOmplete Mini tablets [Roche]). After centrifugation (15,000 rpm at 4°C for 10 min), samples were boiled in Laemmli buffer and loaded on a 10% MOPS NuPage Bis-Tris gel (Invitrogen) with the Novex Sharp Pre-stained Protein Standard (Invitrogen). S2 cells transfected with pMT-Hrp48-APOBEC-P2A-dsRed (pCR8) and pMT-APOBEC-P2A-dsRED (pCR10) in wells of a six well plate and induced with 0.5 mM copper sulfate for 24 h. Two-hundred and fifty microliters of RIPA buffer was added to each well; cells were scraped and lysed at 4°C for 15 min. After centrifugation (15,000 rpm at 4°C for 10 min), samples were boiled in Laemmli buffer and loaded into a 10% MES NuPage Bis-Tris gel (Invitrogen) with the Novex Sharp Pre-stained Protein Standard (Invitrogen). The gels were transferred to nitrocellulose using iBLOT2 (Invitrogen) following manufacturer’s protocol. The blots were blocked with 5% milk in Tris-buffered saline with 1% Tween 20 (TBST) for 1 h before being incubated overnight at 4°C in primary antibodies (1:1000 mouse anti-V5 Tag Monoclonal Antibody [Invitrogen], 1:1000 anti-Actin monoclonal antibody [Invitrogen], or 1:500 anti-TDP-43 [Proteintech] in 5% milk/TBST). Primary an- tibodies were removed with six washes of TBST each for 5 min. The blots were incubated with secondary antibodies (Cytiva Life Science anti-Mouse IgG, peroxidase-linked antibody from sheep [Fisher Scientific] or Cytiva Life Science anti-Rabbit IgG, peroxi- dase-linked antibody from donkey [Fisher Scientific]) diluted in 5% milk/TBST for 1 h on a rocker at room temperature. The blots were then washed 6× for 5 min in TBST prior to being developed with Clarity Western ECL Substrate Kit (Bio-Rad). To examine actin levels as a loading control blots were stripped using Restore Western Blot Stripping Buffer (Thermo Scientific) for 15 min and reblocked with 5% milk for 1 h at RT on rocker. Blots were imaged using a ChemiDoc (Bio-Rad). DATA DEPOSITION All sequencing data generated in this study is deposited in the Gene Expression Omnibus (GEO) under accession number GSE223557. Human data is in subseries GSE223555 and Drosophila data is in subseries GSE223556. The TRIBE analysis pipeline is publicly available at https://github.com/rosbashlab/ HyperTRIBE. All scripts used in this manuscript are available at https://github.com/rosbashlab/Comparison-of-TRIBE-and-STAMP. 1240 RNA (2023) Vol. 29, No. 8 SUPPLEMENTAL MATERIAL Supplemental material is available for this article. ACKNOWLEDGMENTS We would like to thank all members of the Rosbash laboratory for helpful comments on this work. We thank Shlesha Richhariya and Don Rio for comments on the written manuscript. We thank Daniel Shin for laboratory management and administrative sup- port. This work was supported by the Howard Hughes Medical Institute (NIH) (R01DA37721 and R01NS130670). Institutes of Health and the National Received January 28, 2023; accepted April 28, 2023. REFERENCES Alizzi RA, Xu D, Tenenbaum CM, Wang W, Gavis ER. 2020. The ELAV/ Hu protein found in neurons regulates cytoskeletal and ECM adhe- sion inputs for space-filling dendrite growth. PLoS Genet 16: e1009235. doi:10.1371/journal.pgen.1009235 Anders S, Pyl PT, Huber W. 2015. HTSeq—a Python framework to work with high-throughput sequencing data. Bioinformatics 31: 166–169. doi:10.1093/bioinformatics/btu638 Arribas-Hernández L, Rennie S, Köster T, Porcelli C, Lewinski M, Staiger D, Andersson R, Brodersen P. 2021. Principles of mRNA targeting via the Arabidopsis m6A-binding protein ECT2. Elife 10. doi:10.7554/eLife.72375 Ayala YM, De Conti L, Avendano-Vazquez SE, Dhir A, Romano M, D’Ambrogio A, Tollervey J, Ule J, Baralle M, Buratti E, et al. 2011. TDP-43 regulates its mRNA levels through a negative feed- back loop. EMBO J 30: 277–288. doi:10.1038/emboj.2010.310 Babitzke P, Baker CS, Romeo T. 2009. Regulation of translation initia- tion by RNA binding proteins. Annu Rev Microbiol 63: 27–44. doi:10.1146/annurev.micro.091208.073514 Biswas J, Rahman R, Gupta V, Rosbash M, Singer RH. 2020. MS2- TRIBE evaluates both protein-RNA interactions and nuclear orga- nization of transcription by RNA editing. iScience 23: 101318. doi:10.1016/j.isci.2020.101318 Brannan KW, Chaim IA, Marina RJ, Yee BA, Kofman ER, Lorenz DA, Jagannatha P, Dong KD, Madrigal AA, Underwood JG, et al. 2021. Robust single-cell discovery of RNA targets of RNA binding proteins and ribosomes. Nat Methods 18: 507–519. doi:10.1038/ s41592-021-01128-0 Cheng Y-L, Hsieh H-Y, Tu S-L. 2021. A new method to identify global targets of RNA-binding proteins in plants. bioRxiv doi:10.1101/ 2021.06.11.448000 Das S, Vera M, Gandin V, Singer RH, Tutucci E. 2021. Intracellular mRNA transport and localized translation. Nat Rev Mol Cell Biol 22: 483–504. doi:10.1038/s41580-021-00356-8 Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. 2013. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29: 15–21. doi:10.1093/bioinformatics/ bts635 Doll RM, Boutros M, Port F. 2022. A temperature-tolerant CRISPR base editor mediates highly efficient and precise gene inactivation in vivo. bioRxiv doi:10.1101/2022.12.13.520203 Farrar MA, Kiernan MC. 2015. The genetics of spinal muscular atro- phy: progress and challenges. Neurotherapeutics 12: 290–302. doi:10.1007/s13311-014-0314-x Flamand MN, Ke K, Tamming R, Meyer KD. 2022. Single-molecule identification of the target RNAs of different RNA binding proteins simultaneously in cells. Genes Dev 36: 1002–1015. doi:10.1101/ gad.349983.122 Fossat N, Tourle K, Radziewic T, Barratt K, Liebhold D, Studdert JB, Power M, Jones V, Loebel DA, Tam PP. 2014. C to U RNA editing mediated by APOBEC1 requires RNA-binding protein RBM47. EMBO Rep 15: 903–910. doi:10.15252/embr.201438450 Gao R, Asano SM, Upadhyayula S, Pisarev I, Milkie DE, Liu T-L, Singh V, Graves A, Huynh GH, Zhao Y, et al. 2019. Cortical column and whole brain imaging with molecular contrast and nanoscale reso- lution. Science 363: eaau8302. doi:10.1126/science.aau8302 Gebauer F, Schwarzl T, Valcarcel J, Hentze MW. 2021. RNA-binding proteins in human genetic disease. Nat Rev Genet 22: 185–198. doi:10.1038/s41576-020-00302-y Grant CE, Bailey TL. 2021. XSTREME: comprehensive motif analysis of biological sequence datasets. bioRxiv doi:10.1101/2021.09.02 .458722 Grunewald J, Zhou R, Garcia SP, Iyer S, Lareau CA, Aryee MJ, Joung JK. 2019. Transcriptome-wide off-target RNA editing in- duced by CRISPR-guided DNA base editors. Nature 569: 433– 437. doi:10.1038/s41586-019-1161-z Hafner M, Katsantoni M, Köster T, Marks J, Mukherjee J, Staiger D, Ule J, Zavolan M. 2021. CLIP and complementary methods. Nat Rev Methods Primers 1: 20. doi:10.1038/s43586-021-00018-1 Hallegger M, Chakrabarti AM, Lee FCY, Lee BL, Amalietti AG, Odeh HM, Copley KE, Rubien JD, Portz B, Kuret K, et al. 2021. TDP-43 condensation properties specify its RNA-binding and reg- ulatory repertoire. Cell 184: 4680–4696.e22. doi:10.1016/j.cell .2021.07.018 Herzog JJ, Xu W, Deshpande M, Rahman R, Suib H, Rodal AA, Rosbash M, Paradis S. 2020. TDP-43 dysfunction restricts dendritic complexity by inhibiting CREB activation and altering gene ex- pression. Proc Natl Acad Sci 201917038. doi:10.1073/pnas .1917038117 Hocine S, Singer RH, Grunwald D. 2010. RNA processing and export. Cold Spring Harb Perspect Biol 2: a000752. doi:10.1101/cshper spect.a000752 Jin S, Zong Y, Gao Q, Zhu Z, Wang Y, Qin P, Liang C, Wang D, Qiu JL, Zhang F, et al. 2019. Cytosine, but not adenine, base editors in- duce genome-wide off-target mutations in rice. Science 364: 292–295. doi:10.1126/science.aaw7166 Jin H, Xu W, Rahman R, Na D, Fieldsend A, Song W, Liu S, Li C, Rosbash M. 2020. TRIBE editing reveals specific mRNA targets of eIF4E-BP in Drosophila and in mammals. Sci Adv 6: eabb8771. doi:10.1126/sciadv.abb8771 Kanca O, Zirin J, Garcia-Marques J, Knight SM, Yang-Zhou D, Amador G, Chung H, Zuo Z, Ma L, He Y, et al. 2019. An efficient CRISPR-based strategy to insert small and large fragments of DNA using short homology arms. Elife 8: e51539. doi:10.7554/ eLife.51539 Komor AC, Kim YB, Packer MS, Zuris JA, Liu DR. 2016. Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature 533: 420–424. doi:10.1038/nature17946 Komor AC, Zhao KT, Packer MS, Gaudelli NM, Waterbury AL, Koblan LW, Kim YB, Badran AH, Liu DR. 2017. Improved base ex- cision repair inhibition and bacteriophage Mu Gam protein yields C:G-to-T:A base editors with higher efficiency and product purity. Sci Adv 3: eaao4774. doi:10.1126/sciadv.aao4774 Kuttan A, Bass BL. 2012. Mechanistic insights into editing-site specif- icity of ADARs. Proc Natl Acad Sci 109: E3295–E3304. doi:10 .1073/pnas.1212548109 Ling H, Kara E, Bandopadhyay R, Hardy J, Holton J, Xiromerisiou G, Lees A, Houlden H, Revesz T. 2013. TDP-43 pathology in a patient carrying G2019S LRRK2 mutation and a novel p.Q124E MAPT. Comparison of TRIBE and STAMP 34: Neurobiol .neurobiolaging.2013.04.011 Aging e2885–e2889. doi:10.1016/j Liu M, Lu B, Fan Y, He X, Shen S, Jiang C, Zhang Q. 2019. TRIBE un- covers the role of Dis3 in shaping the dynamic transcriptome in malaria parasites. Front Cell Dev Biol 7. doi:10.3389/fcell.2019 .00264 Macbeth MR, Schubert HL, Vandemark AP, Lingam AT, Hill CP, Bass BL. 2005. Inositol hexakisphosphate is bound in the ADAR2 core and required for RNA editing. Science 309: 1534–1539. doi:10.1126/science.1113150 Martin M. 2011. Cutadapt removes adapter sequences from high- throughput sequencing reads. EMBnet J 17: 3. doi:10.14806/ej .17.1.200 McGrath E, Shin H, Zhang L, Phue JN, Wu WW, Shen RF, Jang YY, Revollo J, Ye Z. 2019. Targeting specificity of APOBEC-based cy- tosine base editor in human iPSCs determined by whole genome sequencing. Nat Commun 10: 5353. doi:10.1038/s41467-019- 13342-8 McMahon AC, Rahman R, Jin H, Shen JL, Fieldsend A, Luo W, Rosbash M. 2016. TRIBE: hijacking an RNA-editing enzyme to identify cell-specific targets of RNA-binding proteins. Cell 165: 742–753. doi:10.1016/j.cell.2016.03.007 Meyer KD. 2019. DART-seq: an antibody-free method for global m6A detection. Nat Methods 16: 1275–1280. doi:10.1038/s41592- 019-0570-0 Nguyen DTT, Lu Y, Chu KL, Yang X, Park SM, Choo ZN, Chin CR, Prieto C, Schurer A, Barin E, et al. 2020. HyperTRIBE uncovers in- creased MUSASHI-2 RNA binding activity and differential regula- tion in leukemic stem cells. Nat Commun 11: 2026. doi:10.1038/ s41467-020-15814-8 Penagarikano O, Mulle JG, Warren ST. 2007. The pathophysiology of Fragile X syndrome. Annu Rev Genomics Hum Genet 8: 109–129. doi:10.1146/annurev.genom.8.080706.092249 Picelli S, Faridani OR, Bjorklund AK, Winberg G, Sagasser S, Sandberg R. 2014. Full-length RNA-seq from single cells using Smart-seq2. Nat Protoc 9: 171–181. doi:10.1038/nprot .2014.006 Polymenidou M, Lagier-Tourenne C, Hutt KR, Huelga SC, Moran J, Liang TY, Ling SC, Sun E, Wancewicz E, Mazur C, et al. 2011. Long pre-mRNA depletion and RNA missplicing contribute to neuronal vulnerability from loss of TDP-43. Nat Neurosci 14: 459–468. doi:10.1038/nn.2779 Quinlan AR. 2014. BEDTools: the Swiss-army tool for genome feature analysis. Curr Protoc Bioinformatics 47: 11–34. doi:10.1002/ 0471250953.bi1112s47 Rahman R, Xu W, Jin H, Rosbash M. 2018. Identification of RNA-bind- ing protein targets with HyperTRIBE. Nat Protoc 13: 1829–1849. doi:10.1038/s41596-018-0020-y Robinson MD, McCarthy DJ, Smyth GK. 2010. edgeR: a bioconductor package for differential expression analysis of digital gene expres- sion data. Bioinformatics 26: 139–140. doi:10.1093/bioinfor matics/btp616 Robinson JT, Thorvaldsdottir H, Winckler W, Guttman M, Lander ES, Getz G, Mesirov JP. 2011. Integrative Genomics Viewer. Nat Biotechnol 29: 24–26. doi:10.1038/nbt.1754 Rosenberg BR, Hamilton CE, Mwangi MM, Dewell S, Papavasiliou FN. 2011. numerous APOBEC1 mRNA-editing targets in transcript 3′ UTRs. Nat Struct Mol Biol 18: 230–236. doi:10.1038/nsmb.1975 sequencing reveals Transcriptome-wide Salter JD, Bennett RP, Smith HC. 2016. The APOBEC protein family: united by structure, divergent in function. Trends Biochem Sci 41: 578–594. doi:10.1016/j.tibs.2016.05.001 Singh A, Hulsmeier J, Kandi AR, Pothapragada SS, Hillebrand J, D, Thiagarajan Petrauskas Agrawal A, Rt K, K, www.rnajournal.org 1241 Abruzzi et al. Jayaprakashappa D, et al. 2021. Antagonistic roles for Ataxin-2 structured and disordered domains in RNP condensation. Elife 10: e60326. doi:10.7554/eLife.60326 Smith HC, Bennett RP, Kizilyer A, McDougall WM, Prohaska KM. 2012. Functions and regulation of the APOBEC family of proteins. Semin Cell Dev Biol 23: 258–268. doi:10.1016/j.semcdb.2011.10.004 Integrative Genomics Viewer (IGV): high-performance genomics data visuali- zation and exploration. Brief Bioinform 14: 178–192. doi:10 .1093/bib/bbs017 Thorvaldsdottir H, Robinson JT, Mesirov JP. 2013. Tollervey JR, Curk T, Rogelj B, Briese M, Cereda M, Kayikci M, König J, Hortobágyi T, Nishimura AL, Župunski V, et al. 2011. Characteriz- ing the RNA targets and position-dependent splicing regulation by TDP-43. Nat Neurosci 14: 452–458. doi:10.1038/nn.2778 van Leeuwen W, VanInsberghe M, Battich N, Salmen F, van Oudenaarden A, Rabouille C. 2022. the stress granule transcriptome via RNA-editing in single cells and in vivo. Cell Rep Methods 2: 100235. doi:10.1016/j.crmeth.2022 .100235 Identification of Wang L, Wang S, Li W. 2012. RSeQC: quality control of RNA-seq ex- periments. Bioinformatics 28: 2184–2185. doi:10.1093/bioinfor matics/bts356 Xu W, Rahman R, Rosbash M. 2018. Mechanistic implications of en- hanced editing by a HyperTRIBE RNA-binding protein. RNA 24: 173–182. doi:10.1261/rna.064691.117 Xu W, Biswas J, Singer RH, Rosbash M. 2022. Targeted RNA editing: novel tools to study post-transcriptional regulation. Mol Cell 82: 389–403. doi:10.1016/j.molcel.2021.10.010 1242 RNA (2023) Vol. 29, No. 8
10.1371_journal.pbio.3002472	RESEARCH ARTICLE Dynamics of bacterial recombination in the human gut microbiome Zhiru Liu1, Benjamin H. GoodID 1,2,3* 1 Department of Applied Physics, Stanford University, Stanford, California, United States of America, 2 Department of Biology, Stanford University, Stanford, California, United States of America, 3 Chan Zuckerberg Biohub–San Francisco, San Francisco, California, United States of America * [email protected] Abstract AU : Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly: Horizontal gene transfer (HGT) is a ubiquitous force in microbial evolution. Previous work has shown that the human gut is a hotspot for gene transfer between species, but the more subtle exchange of variation within species—also known as recombination—remains poorly characterized in this ecosystem. Here, we show that the genetic structure of the human gut microbiome provides an opportunity to measure recent recombination events from sequenced fecal samples, enabling quantitative comparisons across diverse commensal species that inhabit a common environment. By analyzing recent recombination events in the core genomes of 29 human gut bacteria, we observed widespread heterogeneities in the rates and lengths of transferred fragments, which are difficult to explain by existing mod- els of ecological isolation or homology-dependent recombination rates. We also show that natural selection helps facilitate the spread of genetic variants across strain backgrounds, both within individual hosts and across the broader population. These results shed light on the dynamics of in situ recombination, which can strongly constrain the adaptability of gut microbial communities. Introduction The horizontal exchange of genetic material—also known as horizontal gene transfer (HGT)— is a pervasive force in microbial ecology and evolution [1]. HGT is particularly important within the human gut microbiota, where hundreds of species coexist with each other in close physical proximity [2–4]. HGT is often associated with the acquisition of new genes or path- ways, which can confer resistance to antibiotics [3–8] or enable novel metabolic capabilities [3,9–14]. Genetic material can also be transferred between more closely related strains, where it can overwrite existing regions of the genome via homologous recombination [15,16]. This more subtle form of horizontal exchange acts to reshuffle genetic variants within species, simi- lar to meiotic recombination in sexual organisms. Homologous recombination plays a crucial role in microbial evolution, from the emergence of new bacterial species [17–20] to the transi- tion between clonal and quasi-sexual evolution [21–23]. Homologous recombination can also serve as a scaffold for the incorporation of novel genetic material, which can facilitate the spread of accessory genes across different strain backgrounds [24]. However, while numerous studies have established the pervasiveness of bacterial recombination [21,25–28], the a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Liu Z, Good BH (2024) Dynamics of bacterial recombination in the human gut microbiome. PLoS Biol 22(2): e3002472. https:// doi.org/10.1371/journal.pbio.3002472 Academic Editor: Manimozhiyan Arumugam, University of Copenhagen Faculty of Health and Medical Sciences: Kobenhavns Universitet Sundhedsvidenskabelige Fakultet, DENMARK Received: November 21, 2022 Accepted: December 14, 2023 Published: February 8, 2024 Copyright: © 2024 Liu, Good. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The raw sequencing reads for the metagenomic samples used in this study were downloaded from public repositories listed in the following publications: 10.1038/ nature11209, 10.1038/nature11450, 10.1016/j. cels.2016.10.004, and 10.1101/gr.233940.117. Data underlying all figures, such as the numerical values of bar plots, can be found in 10.5281/ zenodo.10304481. All other metadata, as well as the source code for the sequencing pipeline, downstream analyses, and figure generation are available at Zenodo (10.5281/zenodo.10368227) or PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 1 / 22 Recombination in the gut microbiome GitHub (https://github.com/zhiru-liu/microbiome_ evolution). evolutionary dynamics of this process are still poorly understood in natural populations like the gut microbiota. Funding: This work was supported in part by a Stanford Bio-X Bowes Fellowship (to Z.L.), the Alfred P. Sloan Foundation grant FG-2021-15708 (B.H.G.), National Institutes of Health Grant No. R35GM146949 (B.H.G.), and a Terman Fellowship from Stanford University (B.H.G.). B.H.G. is a Chan Zuckerberg Biohub - San Francisco Investigator. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Multiple methods have been developed for inferring in situ recombination from the fine- scale diversity of natural bacterial isolates [26,27,29–33]. The key challenge lies in disentangl- ing the effects of recombination from the other evolutionary forces (e.g., mutation, selection, and genetic drift) that shape genetic diversity over the same timescales. Existing studies often address this problem using an inverse approach, by fitting the observed data to simple parametric models from microbial population genetics. Examples range from simple summary statistics like linkage disequilibrium [26,27,34,35] and related metrics [21,28,32,33,36–40] to complete probabilistic reconstructions of the genealogies of the sampled genomes [29–31]. Previous applications of these methods have provided extensive evidence for ongoing recom- bination within the core genomes of many bacterial species [25,41]—including many species of human gut bacteria [27]. HGT, horizontal gene transfer; Abbreviations: AU : Anabbreviationlisthasbeencompiledforthoseusedinthetext:Pleaseverifythatallentriesarecorrect: HMM, hidden Markov model; RND, resistance- nodulation-division; SNV, single-nucleotide variant. However, many of these existing methods rely on simplified evolutionary scenarios, which ignore the effects of natural selection, and make additional restrictive assumptions about the demographic structure of the population. Recent work has shown that these simplified models often fail to capture key features of microbial genetic diversity [26–28,42], which can strongly bias estimates of the underlying recombination parameters. Our limited understanding of these effects makes it difficult to answer key questions about the role of recombination in natu- ral populations like the gut microbiota: Is recombination fast enough to allow local adaptations to persist within a host, e.g., during fecal microbiota transplants [43] or sudden dietary shifts [44]? Does natural selection tend to promote or hinder the spread of genetic variants across different strain backgrounds? And can the rates and lengths of transferred fragments shed light on the underlying mechanisms of recombination in situ? Here, we show that the genetic structure of the human gut microbiome provides a unique opportunity to address these questions. Using strain-resolved metagenomics, we show that the large sample sizes and host colonization structure of this ecosystem enable systematic compari- sons of strains across a broad range of distance and timescales, from the scale of individual hosts to the diversity of the broader global population. We show that some of these strains are closely related enough that one can resolve homologous recombination events directly, with- out requiring restrictive modeling assumptions or explicit phylogenetic inference. We use these observations to develop a nonparametric approach for identifying large numbers of recent recombination events within 29 prevalent species of human gut bacteria. This compara- tive data set allows us to systematically explore the landscape of homologous recombination in this host-associated ecosystem. Our results reveal extensive heterogeneity in rates and lengths of transferred fragments— both among different species and between different strains of the same species—which are dif- ficult to explain by ecological isolation or reduced efficiencies of recombination. We also find that natural selection can play an important role in facilitating the spread of transferred frag- ments into different strain backgrounds. Our results suggest that in situ recombination events are shaped by a combination of evolutionary processes, which may strongly depend on the ecological context of their host community. Results Partially recombined genomes underlie the broad range of genetic diversity in many species of gut bacteria To quantify the dynamics of homologous recombination across different timescales, we ana- lyzed shotgun metagenomic data from a collection of healthy human gut microbiomes that we PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 2 / 22 PLOS BIOLOGYRecombination in the gut microbiome collated in a previous study [27]. This data set comprises 932 fecal samples from 693 subjects from North America, Europe, and China (S1 Table). We used a reference-based approach to identify single-nucleotide variants (SNVs) in the core genome of each species in each sample (Section 1 in S1 Text). These metagenomic variants reflect a complex mixture of the global genetic diversity within a given species, as well as the specific combination of lineages that are present within a given host. While it is difficult to resolve the underlying lineages in the most general case, we previously showed [27] that the lineage structure in many human gut meta- genomes is simple enough that the core genome of the dominant strain can be inferred with a high degree of confidence (Fig 1A and Fig A in S1 Text). Using this approach, we obtained a total of 5,416 “quasi-phased” genomes from 43 different species in 541 unique hosts. The genetic differences between these strains provide a window into the long-term evolutionary forces that operate in these species over multiple host colonization cycles. Previous work has shown that the genetic diversity within many species of gut bacteria spans a broad range of timescales [27,45]. For example, in Alistipes putredinis (a prominent gut commensal), the synonymous divergence between a typical pair of strains is d�2%, but some pairs of strains are separated by just a handful of SNVs (Fig 1C). Similar pairs of closely related strains have been observed in other bacterial species, where they are often associated with clinical outbreaks or other local transmission processes [21,36,46]. In our case, the breadth of sampling of the human gut microbiome allows us to rule out many of these microe- pidemic factors, since closely related strains are frequently observed in unrelated hosts from different countries [27]. Population genetic theory predicts that similar patterns can also arise due to the local nature of bacterial recombination [28,39]. Since the length of a typical recombined segment (ℓr) is usually much shorter than the total genome length (L), there is a broad range of timescales between the first recombination event (Tr) and the time required for the genome to be completely overwritten by imported fragments (Tmosaic � Tr � L=‘r; Fig 1B). In quasi-sexual bacterial populations, most pairs of strains will share a common ancestor Tmrca�Tmosaic generations ago, so that their present-day genomes comprise a mosaic of over- lapping recombination events. However, in a large enough sample, some pairs of strains will inevitably share a common ancestor on timescales much shorter than Tmosaic (Fig 1A). Among these “closely related” strains, recombination will not have had enough time to completely cover the ancestral genome with DNA from other, more typically diverged strains. Rather, individual recombination events will be visible as “blocks” of typical genetic divergence against a backdrop of nearly identical DNA sequence [28,39]. These partially recombined genomes have previously been observed in other bacterial species—most notably in Escherichia coli [28,39] and other bacterial pathogens [42,47,48]. Fig 1D shows that similar examples can be observed within the A. putredinis population as well. To test whether this pattern holds more broadly, we divided the core genome of each pair of strains into blocks of 1,000 synonymous sites and calculated the fraction of blocks with zero SNV differences within them. In a pair of partially recombined genomes, we would expect to see a neg- ative correlation between the fraction of identical blocks (a proxy for the fraction of clonal ances- try) and the overall genetic divergence across the genome (Fig 1E and Fig B in S1 Text). One can observe such a trend in A. putredinis (Fig 1C)—well beyond that expected from the randomness of individual mutations. Instead, we find that a simple model of accumulated transfers (red line; Section 2 in S1 Text) can account for a large fraction of the spread in genome-wide divergence in A. putredinis, consistent with the partial recombination model in Fig 1B. Similar patterns can be observed in many other gut commensals (Figs C and D in S1 Text). Some species exhibit some variation in the divergence of the most distantly related strains (e.g., Bacteroides vulgatus and Eubacterium rectale), consistent with the presence of subspecies PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 3 / 22 PLOS BIOLOGYRecombination in the gut microbiome Fig 1. Partially recombined genomes underlie the broad range of genetic divergence in many species of gut bacteria. (A) Genetic differences between the core genomes of the dominant strain of a given species (>80% within-host frequency) are inferred from pairwise comparisons of metagenomes from unrelated hosts (Section 1 in S1 Text). (B) Timescales of recombination in a quasi-sexual bacterial population: most strains share a common ancestor � Tmosaic generations ago, so their present-day genomes are completely overwritten by recombination; in large samples, some pairs of strains will share a common ancestor � Tmosaic generations ago, and recombination events will be visible as blocks of local divergence against a shared clonal background. (C) Average synonymous divergence vs. fraction of identical blocks for pairs of A. putredinis strains from unrelated hosts (Section 2 in S1 Text). Points denote individual pairs, while the marginal distribution is shown on the right; red line shows the expectation from a simple model of accumulated transfers (Section 2 in S1 Text), while gray line shows the expectation when mutations are randomly distributed across the genome. (D) Spatial distribution of synonymous SNVs for 3 example pairs from panel C (symbols); only a portion of the core genome is shown. Points denote individual SNVs, while lines show the local divergence in sliding 300 bp windows. (E) Analogous version of C for neutral simulations (Section 5.3 in S1 Text, Fig B in S1 Text). (F) Inferred values of Tmrca/Tmosaic for the partial recombination model in Section 2 in S1 Text; 2 additional species (H. pylori and M. tuberculosis) are shown on the right for comparison. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g001 or other forms of population structure [18,27,49]. Yet even in these cases, we find that partially recombined genomes can still account for much of the variation among more closely related strains. Across species, we find that our simple model of accumulated transfers can explain more than 50% of the weighted variation in pairwise divergence within 36 of the 43 species we PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 4 / 22 PLOS BIOLOGYRecombination in the gut microbiome examined (Fig F in S1 Text). The implied amounts of recombination are often quite large (Tmrca/Tmosaic≳10, Fig 1F) and are comparable to highly recombinant species like Helicobacter pylori. These estimates suggest that typical pairs of strains have been completely overwritten by recombination events (Fig 1B). Despite the generality of this trend, the total number of closely related strains can vary sub- stantially between species (S2 Table). For example, many Alistipes and Bacteroides species con- tain hundreds of closely related pairs, while other species like Prevotella copri have only a handful. While the causes of these differences are currently unclear, the simplified patterns of recombination among these strains suggest that we can use them to directly resolve individual recombination events within a range of different species. Measuring individual recombination events that accumulate between closely related strains in different hosts To identify individual recombination events across a diverse range of human gut species, we turned to an automated approach for analyzing the spatial distribution of genetic differences along the core genomes of closely related pairs of strains. We chose to focus on the core genome to limit the impact of plasmids and other mobile genetic elements, which can be hori- zontally transmitted at much higher rates than normal chromosomal DNA [50–52]. By restricting our attention to core genes, we sought to infer the baseline rates of recombination that shape the evolution of the larger genome, which involve the permanent replacement of existing sequences in addition to successful transfers. Our pairwise model assumes that the genetic differences along the core genome arise through a mixture of 2 processes: (i) point mutations (which alter individual sites); and (ii) homologous recombination events (which replace longer stretches of DNA with a correspond- ing fragment sampled from another strain in the populations). For sufficiently close pairs, the mutation and recombination processes have a negligible chance of overlapping, which means that they can be captured by a simple hidden Markov model (HMM) that transitions between clonal and recombined regions at different locations along the genome (Fig 2B and Fig G in S1 Text; Section 3.1 in S1 Text). The corresponding transition rates between these states will vary between different pairs of strains, due to the differences in their time-aggregated rates of recombination. Since the genealogies of close pairs are particularly simple, these pairwise esti- mates can implicitly capture various forms of selection, non-equilibrium demography, and other deviations from the simplest neutral null models, even when there is insufficient data for a complete phylogenetic reconstruction. In contrast to previous approaches [28–31,53], we used the empirical distribution of local divergence to model the number of SNVs imported by each recombined fragment (Section 3.1 in S1 Text). This allows us to capture the broad variation observed in different transfers (Fig H in S1 Text) in a way that is directly informed by the available data. We validated the perfor- mance of our algorithm (CP-HMM) through simulations and found that it can reliably iden- tify individual recombination events across a range of genetic divergence scales (Figs I–L in S1 Text; Section 3.2 in S1 Text). Fig 2 shows an example of this approach applied to B. vulgatus, one of the most abundant and prevalent species in the human gut. Previous work [27] has shown that this species pos- sesses a strong population structure with 2 major clades (corresponding to the vulgatus and dorei subspecies [54]), such that the within-clade divergence is approximately 10-fold smaller than the divergence between clades (Fig 2A). We exploited this structure to further resolve the recombination events into within- and between-clade transfers based on their local sequence divergence (Fig 2B, Section 3.1 in S1 Text). By applying our HMM algorithm to the 210 pairs PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 5 / 22 PLOS BIOLOGYRecombination in the gut microbiome Fig 2. Measuring individual recombination events that accumulate between closely related strains. (A) Schematic illustration for B. vulgatus, which has a strong population structure with 2 major clades. (B) Our pairwise hidden Markov model (CP-HMM) classifies the core enome of each pair of closely related strains into clonal regions (gray) and recombined regions (blue = within-clade, orange = between-clade) based on their local synonymous divergence; points denote individual SNVs, while lines show the local divergence in sliding 1,000 bp windows. Data from 2 example pairs are shown. (C) The observed number of recombination events in all pairs of closely related B. vulgatus strains as a function of the synonymous divergence in their inferred clonal regions (Section 3.1 in S1 Text). These events are further partitioned into within-clade and between-clade transfers (top and bottom). Lines indicate the average trend computed using a local regression technique, while shaded regions indicate the local spread (Section 3.3 in S1 Text). (D) Distribution of the estimated transfer lengths for each of the recombination events in panel C. These data show that the rates and lengths of successful transfers strongly depend on the divergence of the imported fragments. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g002 of closely related B. vulgatus strains in our cohort, we identified a total of �1,700 recombined regions with a mean length of �20 kb (S3 Table). We also applied our algorithm to a separate collection of B. vulgatus isolate genomes (Fig O in S1 Text; Section 3.10 in S1 Text) to verify that our conclusions were robust to the quasi-phasing approach employed in Fig 2. We observed an overall trend toward larger numbers of recombination events in strains with higher clonal divergence (Fig 2C), consistent with the gradual accumulation of successful transfers over time. However, the larger sample reveals that this is not a simple linear relationship: Some strains have anomalously large numbers of transfers even at low clonal divergence, while others have anomalously few transfers even at high clonal divergence (Fig 2C). Similar results are also observed when considering the cumulative length of the recombined genome for each pair (Fig M in S1 Text), which confirms that this variation is not an artifact of the event detection algorithm. Instead, these data suggest that successful transfers in B. vulgatus do not accumulate at a fixed recombination rate, as assumed under the simplest models of neutral evolution. We also found that recombination between the major B. vulgatus clades occurred much less frequently than recombination within clades, with a ~5-fold reduction in the total number of detected transfers as a function of their genetic divergence (Fig 2C and Fig J in S1 Text). This genetic isolation could arise from several factors, ranging from reduced opportunities for recombination (e.g., due to ecological isolation [2] or fewer homologous flanking regions for initiating strand invasion [55,56]) to greater downstream incompatibilities in the acquired fragments (e.g., epistatic interactions [57,58] or mismatch-repair-mediated proofreading [59,60]). In this case, the larger ensemble of detected transfers allows us to further distinguish between these scenarios. Beyond the reduction in the number of detected recombination events, we also observed a systematic difference in the lengths of the individual transfers, with a ~7-fold reduction in the median transfer length between clades (Fig 2D and Fig J in S1 Text). These differences indicate that the greater genetic isolation of the B. vulgatus clades cannot be captured by a simple rescaling of the recombination rate and that additional factors like epista- sis or mismatch-repair-mediated proofreading are necessary to explain the data. Variation of recombination rates within and across gut species To understand how these results for B. vulgatus extend to other members of the gut micro- biome, we applied the same approach to the other species in our data set with a sufficient PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 6 / 22 PLOS BIOLOGYRecombination in the gut microbiome number of closely related strains. This pairwise analysis yielded a total of 228,078 recombined regions in 7,383 closely related pairs from 29 different species. These data revealed systematic variations in the rates and lengths of transferred fragments across many prevalent gut species (Fig 3 and Figs P–R in S1 Text), similar to E. coli and other bacterial pathogens [16,61–63]. We found that some of these trends were consistent with the phylogenetic relationships between species. For example, species in the Rikenellaceae family tended to have relatively fre- quent and short transfers, while Bacteroidaceae family tended to have lower rates and longer transfers. However, we also observed large differences within individual genera. For example, Bacteroides massiliensis has a relatively linear accumulation of transfers over time (Fig 3B), while most pairs of Bacteroides caccae strains have few detected recombination events (Fig 3A). The typical transfer length varies among Bacteroides species as well (6 to 35 kb), spanning a larger range than Alistipes (3 to 6 kb). Fig 3. Heterogeneous recombination rates within and between prevalent gut species. (A–C) Analogous versions of Fig 2C for 3 example species, which were chosen to illustrate a range of characteristic behaviors. Gray regions denote the points that were excluded by our filtering steps (Section 3 in S1 Text). (D) Apparent recombination rates (number of transfers/clonal divergence/core genome length) for all species with a sufficient number of closely related strains (Section 3.3 in S1 Text). For species with > 100 close pairs, we plot the average recombination rate at 4 characteristic divergence times (dc = 2.5,5,7,5,10×10−5, highlighted as points along the trend lines in panels A–C) using the trend lines in panels A–C; estimates are connected by lines to aid visualization. For species with < 100 close pairs, we plot the distribution of apparent recombination rates for all individual pairs; box plots indicate the median and inter-quartile range. (E) Lengths of recombined fragments for each of the species in panel D. Symbols show the lengths of all detected transfer events across all pairs of closely related strains; box plots indicate the median and inter-quartile range. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g003 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 7 / 22 PLOS BIOLOGYRecombination in the gut microbiome Zooming in further, we also observed considerable variation within individual species. Some of these differences could be attributed to the presence of strong population structure (similar to B. vulgatus), with a reduction in both the rates and lengths of successful transfers between highly diverged clades (e.g., Alistipes Shahii; Fig V in S1 Text). However, we also observed substantial variation even in the absence of population structure. For example, A. putredinis contains many closely related strains with an anomalously large number of trans- fers, as well as an excess of more diverged strains with few recombined segments (Fig 3C and Fig K in S1 Text). Other species (e.g., B. caccae; Fig 3A) exhibited bimodal distributions of transferred fragments. None of these behaviors can be captured by a single underlying recom- bination rate. Interestingly, apart from the handful of species with strong population structure, we observed no systematic trend between the frequency of recombination and the divergence of the transferred fragments (Fig 4 and Fig W in S1 Text), as expected under certain models of homologous recombination [19,64]. This observation, in combination with the large number of species in our data set, helps shed further light on the mechanisms that could be responsible for the lower recombination rates we observe between clades. For example, B. thetaiotaomi- cron and B. stercoris both maintain high recombination rates at synonymous divergences com- parable to the genetically isolated clades observed in B. vulgatus and B. finegoldii (Fig 4 and Fig W in S1 Text). This suggests that the genetic isolation of these clades is not a product of their underlying recombination machinery (which should be similar in different Bacteroides spe- cies) but rather by genetic incompatibilities that have accumulated between the 2 clades, or related scenarios like incompatible restriction-modification systems [65–68]. Understanding the ecological and evolutionary forces that caused these incompatibilities to emerge within some Bacteroides species but not others is an interesting avenue for future work. Signatures of within-host recombination in co-colonized hosts Our preceding analysis focused on the successful transfers that have accumulated between closely related strains in unrelated hosts. How do these long-term dynamics—which aggregate over multiple host colonization cycles—emerge from the local processes of competition and colonization within individual hosts? Some of this recombination could occur when multiple strains of the same species are pres- ent within the same host [69]. While examples of co-colonization are less common in the human gut [27,45], we can still identify many individual hosts in our larger cohort in which 2 diverged strains were present at intermediate frequencies, based on the frequencies of SNVs within their corresponding metagenomes (Fig A in S1 Text). Recombination between these strains will generate hybrid genomes that contain a short fragment from their donor (Fig 5A). Each of these hybrid strains will originate as a single cell and will not be visible in a mixed sam- ple unless they later rise to appreciable frequencies. Such a shift could occur through a single- cell bottleneck, e.g., if the hybrid strain is lucky enough to found a new population in naive host. Alternatively, if the transferred fragment provides a fitness benefit to the recipient strain, it can rapidly increase in frequency within its host and eventually displace its parent. These “gene- specific sweeps” will lead to a characteristic depletion of SNVs within the donated region in a mixed population sample, while preserving the remaining genetic variation elsewhere along the genome (Fig 5A). The higher frequencies of the resulting hybrids will make them substantially more likely to seed future colonization events in other hosts, suggesting that they could play an important role in generating the recombination events we observed in Figs 2 and 3. Fig 5 shows an example of this scenario in a longitudinally sampled host who was co-colo- nized by a pair of typically diverged B. vulgatus strains (d�1%). We observed a sudden PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 8 / 22 PLOS BIOLOGYRecombination in the gut microbiome Fig 4. Quantifying the frequency of recombination as a function of the genetic divergence between donor and recipient DNA sequences. (A) Schematic illustration showing the genetic divergence of 2 recombined fragments relative to the focal pair of genomes. The synonymous divergence of each detected transfer is computed and aggregated across all closely related pairs within a species. (B–E) Distribution of donor-recipient divergence for all detected transfers in 4 example species. Orange lines show the observed data, while the blue lines show a null expectation obtained by randomly drawing segments from the observed collection of genomes (Section 3.6 in S1 Text). Insets show the corresponding complementary cumulative distribution functions. For species with a strong clade structure (D and E), the average between-clade divergence is indicated by dashed vertical lines. (F) Differences between the observed and simulated divergence distributions for all of the species in Fig 3, summarized by the Kolmogorov–Smirnov (K-S) distance (inset). Solid bars indicate statistically significant differences (P<10−3; one-sided K-S test), while arrows indicate the example species in panels B–E. Together, these data show that many species exhibit only small differences between their observed and expected divergence distributions (K-S distance ≲0.1), even when their overall sequence divergence is comparable to counterexamples like D and E (Figs W and X in S1 Text). The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g004 depletion of within-host SNVs within a ~20 kb region during the ~6-month interval between samples (Fig 5B and 5C), while the SNV patterns across the rest of the genome were largely preserved. This local depletion of diversity cannot be explained by a large deletion event in one of the 2 strains, since the estimated copy number of the recombined region remained close to one at both timepoints (Fig 5C and Fig Y in S1 Text). This region spanned a total of 25 core and accessory genes on the reference genome, including a resistance-nodulation-division (RND) family efflux pump (S5 Table and Fig EE in S1 Text); at present, it is not clear which of these genes was responsible for driving the sweep, or if the recombined fragment was simply hitchhiking alongside a different causative mutation. With limited longitudinal data from co-colonized hosts, it is difficult to find many contem- poraneous examples like the one illustrated above. However, we reasoned that the remnants of these gene-specific sweeps would still be visible even in metagenomic data from a single time- point. Previous work suggests that conspecific strains can coexist within their hosts for years at a time [27,70,71]. Any gene-specific sweeps that occur during this interval will produce an extended run of zero SNVs against the backdrop of an otherwise diverse metagenome. We identified many such runs of shared ancestry among the co-colonized hosts in our cohort PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 9 / 22 PLOS BIOLOGYRecombination in the gut microbiome Fig 5. Signatures of within-host recombination among co-colonizing strains. (A) Schematic illustration of a potential recombination scenario: (i) a single host is colonized by a pair of diverged strains; (ii) recombination generates hybrid strains that initially reside at low frequencies; (iii) if a hybrid replaces its parent (e.g., due to a selective sweep), it will lead to a depletion of genetic diversity within the transferred region. (B) Within host SNVs are identified by aligning metagenomic sequencing reads to the reference genome. The frequencies and coverages of these SNVs can be used to identify gene-specific sweeps by hybrid strains (Fig OO in S1 Text). (C) An example of a hybrid sweep in a B. vulgatus population in a longitudinally sampled host. Top and bottom parts show metagenomic data collected from the same host at timepoints T0 (top) and T1 (bottom); Δt~6 mo. In the top panel of each timepoint, solid lines denote the local coverage, estimated from a moving average of the local read depth. In the bottom panel of each timepoint, symbols denote the frequencies of within-host SNVs in the highlighted region of panel D (orange arrow), which are polarized such that the reference alleles have frequency >0.5 at T0; for comparison, the genome-wide distribution of SNV frequencies is shown on the right, illustrating the coexistence between 2 dominant strains at both timepoints (black dotted lines, bar plots). Gray regions denote non- core genes. These data show a sudden depletion of SNVs within a ~20 kb region. The consistent coverage around the genome-wide average (gray dashed lines) at both timepoints indicates that the depletion of SNVs in the highlighted regions is not caused by large deletion in one of the coexisting strains. (D) Tracts of shared ancestry between B. vulgatus strains. Top panels show the spatial distribution of within-host SNVs (green vertical lines) and tracts of shared ancetry (white regions of 0 SNV) from the host in (C); orange arrow highlights the putative within-host sweep event in (C). For comparison, the bottom 2 panels show analogous distributions computed for pairs of strains from different hosts. In these examples, long sharing tracts similar to the within-host sweep in (C) are visible along the genome. (E) Distribution of the longest sharing tract in each co-colonized host for 2 example species (Section 4.3 in S1 Text). Gray dashed lines indicate the mean transfer length inferred in Fig 3E. The total number of co-colonized samples and the P-value under the one-sided Kolmogorov–Smirnov test are shown. The B. vulgatus distribution is PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 10 / 22 PLOS BIOLOGYnot significantly different from its between-host counterpart, while E. rectale displays a significantly higher rate of within-host sharing. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g005 Recombination in the gut microbiome (Section 4.2 in S1 Text), including several other examples in the B. vulgatus population above (Fig 5D). These runs can extend for thousands of base pairs and are significantly longer than we would expect if the mutations were randomly scattered across the genome (P<10−10, Fig Z in S1 Text). This suggests that they could be candidates for previous gene-specific sweeps that occurred within the host’s lifetime. However, it is important to distinguish this scenario from older recombination events that were inherited by the strains before they colonized their current host (Fig 5A). Estimates sug- gest that a 10 kb fragment will require hundreds of years on average to accumulate its first mutation [72], which implies that any given run could be consistent with a broad range of pos- sible ages. Consistent with this expectation, we also observed many long runs of shared ances- try when comparing strains from unrelated hosts—some of which extended for as long as the within-host examples above (Fig 5D and Fig Z in S1 Text). This suggests that the true signal of within-host recombination must be distinguished from this baseline level of sharing. We reasoned that if within-host recombination was prevalent, we should still expect to see longer runs of shared ancestry in co-colonizing strains compared to random pairs of strains obtained from unrelated hosts. To test this idea, we used the length of the longest run as a test statistic, and asked how the distribution of this quantity differed between co-colonizing strains of the same species and random pairs of strains selected from unrelated hosts. We observed a strong enrichment of long runs in co-colonizing strains of E. rectale (Fig 5E), which suggests that they were likely caused by previous within-host recombination events similar to the B. vulgatus example above. Similar results were obtained when we examined the total length of runs that exceeded a given length threshold (Fig AA in S1 Text). In contrast, we found that some of the other species with high rates of recombination across hosts (e.g., A. putredinis; Fig 3C) did not show any enrichment in within-host sharing (Fig AA in S1 Text). This negative result could imply that co-colonizing strains recombine less fre- quently in these species or that fewer hybrid strains manage to sweep to high frequencies. It could also occur if the background levels of between-host sharing are sufficiently frequent that they overwhelm any signature of within-host sweeps. This scenario could be particularly rele- vant for species like B. vulgatus (Fig 5), in which nearly half of all random strain pairs share identical sequences longer than the typical transfer length in Fig 2. These results show how understanding the population genetic patterns between hosts can be important for resolving the evolutionary forces within individual host communities. Distribution of shared DNA segments across hosts reveals selection on recent transfers The high levels of between-host sharing in species like B. vulgatus raise a natural question: Why do random pairs of strains share so many stretches of identical DNA within their core genomes? Population genetic theory predicts that such tracts of shared ancestry can emerge even in simple neutral scenarios due to the joint action of recombination, mutation, and genetic drift [73]. For a random pair of strains, the expected number of shared fragments lon- ger than ℓ scales as � L=�d‘2ð1 þ r=mÞ2, where �d is the average divergence between typical pairs of strains (Fig CC in S1 Text; Section 5.1 in S1 Text). The slow decay with ℓ and r implies that this number will often be larger than one, even for tracts as long as ℓ~10 kb. This suggests that the presence of shared segments alone is not surprising. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 11 / 22 PLOS BIOLOGYRecombination in the gut microbiome Fig 6. Global distribution of shared DNA segments reveals selection on recent transfers. (A) Schematic of pairwise haplotype sharing metric: For each position in the core genome, we compute the fraction of strain pairs from different hosts that have identical genotypes across a window of D‘ � 1=�d synonymous sites (Section 5.2 in S1 Text). (B) Observed sharing landscape for B. vulgatus (middle panel); separate comparisons are performed for strains from the same clade (blue, Δℓ�1,500 synonymous sites � 10 kb) or different clades (orange, Δℓ�220 synonymous sites � 1.5 kb). The top panel shows the average synonymous divergence computed in sliding windows of size Δl = 3,000. These landscapes reveal regions of elevated sharing across hosts (e.g., shaded region) that cannot be explained by local reductions in diversity. Red shaded region indicates the within-host recombination event in Fig 5B and 5C. The bottom panel shows analogous sharing landscapes from neutral simulations (Section 5.3 in S1 Text), which display more even rates of sharing across the genome. Gray lines denote 100 simulation runs with the same parameters, while the blue line highlights 1 typical run. (C) Sharing landscape for E. rectale, computed for pairs of strains in different hosts (top) and co-colonizing strains from the same host (bottom). (D) Heterogeneous sharing landscapes across 27 species. Blue points show the coefficient of variation of the sharing probability across the genome for all species with sufficient between-host comparisons. B. vulgatus (within clade) and E. rectale are highlighted as pink triangles. Gray points show analogous values derived from neutral simulations across a range of parameter values (Section 5.3 in S1 Text); each point denotes the mean of 100 simulation runs, while lines show the standard deviation. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.10304481. https://doi.org/10.1371/journal.pbio.3002472.g006 However, this simple neutral scenario makes strong predictions about how often a given region is shared across multiple pairs of strains. To test whether this scenario could recapitu- late our data, we scanned across the genome of each species and calculated the probability that each position was involved in a long shared segment (‘ � �d > 15, Fig 6A, Section 5.2 in S1 Text). This analysis revealed a systematic variation in the probability of shared segments at dif- ferent genomic locations (Fig 6B–6D). An example of this behavior is shown for B. vulgatus in Fig 6B. A typical site in the B. vulga- tus genome has a 3% chance of being shared in a segment longer than approximately 10 kb. However, we observed that many local regions were shared much more frequently than the genome-wide average, despite having comparable levels of genetic diversity (Fig 6B and Fig DD in S1 Text). Some of these peaks are driven by the expansion of a single dominant PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 12 / 22 PLOS BIOLOGYRecombination in the gut microbiome haplotype, while others correspond to multiple distinct haplotypes that are shared by different sets of strains (Figs FF and GG in S1 Text). Similar “sharing hotspots” can be observed in other prevalent gut species as well (Fig 6C and 6D). This high degree of heterogeneity is inconsistent with simple neutral models of bacterial evolution. Simulations show that neutral models generate significantly tighter correlations between the average and maximum levels of sharing across the genome (P<10−8; Student’s t test; Fig 6B and 6D and Fig HH in S1 Text). We also asked whether this heterogeneity could be explained by varying recombination rates along the genome [51,74,75]. However, our simula- tions showed that the sharing hot spots in Fig 6 are qualitatively distinct from traditional recombination hot spots. Local increases in the recombination rate actually decreased the probability of sharing longer segments (Fig II in S1 Text), since recombination tends to pro- duce larger numbers of haplotypes with different combinations of mutations. Consistent with this finding, we observe few systematic correlations between the haplotype sharing landscapes in Fig 6 and the recombination hot spots inferred from Fig 3 (Fig JJ in S1 Text). These analyses suggest that the heterogeneous sharing probabilities in Fig 6B are likely driven by positive selection on fragments that are spreading through the population via recom- bination. Consistent with this hypothesis, we found that the regions with the highest levels of sharing are statistically enriched for certain functional genes (e.g., glycosyltransferases) that have previously been shown to be under selection in the gut [72] (Section 5.5 in S1 Text). We also found that the sharing landscape qualitatively differs for fragments that are shared within versus between the major B. vulgatus clades (Fig 6B). This provides further evidence that the selection pressures are specific to the identities of the donated and recipient DNA sequences. Finally, we asked how these global selection pressures were related to the within-host sweeps we detected in Fig 5. For example, we found that the within-host sweep event in Fig 5C occurred within one of the most prominent sharing hotspots in B. vulgatus (Fig 6B), which is peaked around 3 RND efflux pump genes (Fig EE in S1 Text). This suggests that both events were likely driven by a common set of selection pressures. However, this parallelism did not arise through selection of the same DNA sequences: while the sweeping haplotype in Fig 5C was also present in a few other hosts in our panel, we found that several other distinct haplo- types contributed to the global sharing hotspot at this location (Fig FF in S1 Text). This sug- gests that natural selection has promoted the transfer of multiple genetic variants at these loci —similar to a soft selective sweep [76]. Even larger differences were observed within the E. rectale populations in Fig 5E. In this case, while we observed some overlap in the sharing hotspots within versus between hosts, we also identified several new hotspots that were only present among co-colonizing strains (Fig 6C, Fig KK in S1 Text). These significant differences in the locations of the within-host sharing events (P<0.001, permutation test; Section 5.4 in S1 Text) provide further evidence that they were likely driven by selection on recent transfers within their hosts. More broadly, these results show that within-host sweeps are not always local versions of ongoing global sweeps, but may reflect distinct and repeatable selection pressures that are specific to the within-host environment (e.g., competition- versus colonization-related traits [77]). Understanding the tradeoffs that give rise to these different selection pressures is an interesting topic for future work. Discussion Recombination is a ubiquitous force in bacterial evolution, but dynamics of this process are still poorly understood in many natural microbial populations. Here, we sought to quantify these dynamics by leveraging the broad range of timescales inherent in the human gut PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 13 / 22 PLOS BIOLOGYRecombination in the gut microbiome microbiome ecosystem. By analyzing recent recombination events within a panel of 29 gut commensals, we were able to identify general trends across diverse bacterial species that inhabit a common host-associated environment. At a birds-eye view, the rates of recombination we observed across hosts (Fig 3) are compa- rable to other bacterial species [25,33,47] and are consistent with the strong decay in linkage disequilibrium that has been observed in global samples of gut bacteria [27,78]. Across species, we found that recombination is responsible for introducing >10-fold as much variation as mutation (Tmrca/Tmosaic≳10; Fig R in S1 Text), which implies that the genomes of typical cir- culating strains are almost completely overwritten by recombination. These values are broadly consistent with previous observations in bacterial pathogens, though their different sampling strategies can make it difficult to perform detailed numerical comparisons (Fig LL in S1 Text; Section 3.10.1 in S1 Text). The observation of such high rates of genetic exchange in commen- sal gut bacteria poses challenges for efforts to identify signals of parallel evolution in strains sampled from different hosts [72], or signals of codiversification across host populations [79,80], since they imply that individual variants can frequently decouple from the genome- wide phylogeny. In this case, more elaborate methods like the haplotype sharing metric in Fig 6 could be useful for resolving common selection pressures across hosts. Although the long-term recombination rates in Fig 3 represent an average over multiple host colonization cycles, it is useful to consider their implications when extrapolated down to the scale of a single host community. If we assume the recombination events in Fig 3 accumu- late largely neutrally (or via neutral hitchhiking [81]), then the rates implied by these data sug- gest that every site in the genome will be involved in more than a thousand recombination events within a single day (Section 3.8 in S1 Text). These ballpark estimates suggest that there will be numerous opportunities for adaptive mutations to spread between co-colonizing strains within a host (e.g., during a fecal microbiota transplant), even if the donor or recipient strain is present at a low frequency (e.g., approximately 0.1%). However, since each recombi- nation event originates in a single cell, it can still take tens of thousands of generations (approximately 5 to 50 years) before a typical ancestral lineage will be involved in a single de novo recombination event. The large gap between these timescales can help explain why recombination can be an important driver of adaptation in the gut (Fig 5) [27], while also pre- serving the largely clonal structure observed in individual host populations [27,45,70,82]. We emphasize that these extrapolations should be treated with a degree of caution, since they assume that most of the recombination events in Fig 3 are effectively neutral. If the vast major- ity of these events were locally adaptive, then the true rate of recombination could be smaller than the apparent rates in Fig 3 (Section 3.8 in S1 Text). In addition to the overall rates, the enhanced resolution of our approach also provided new insights into the dynamics of recombination within the gut microbiota. Extending previous findings in other bacterial species [28,61,83–85] (see [16] for a review), we observed wide- spread strain-level variation in recombination rates within many commensal gut species—at least some of which could be attributed to existing population structure (e.g., “subspecies” [49] or “ecotypes” [86]). In these handful of examples, the comparative nature of our data set helps illuminate the potential causes of this genetic isolation. By comparing the rates and lengths of successful transfers in species with different levels of genetic diversity, we obtained new evi- dence that the barriers to recombination are likely driven by negative selection on the recom- bined fragments (e.g., due to genetic incompatibilities), or related scenarios like incompatible restriction-modification systems [65–68], rather than passive mechanisms like ecological isola- tion or homology-dependent recombination rates (Fig 4). Our results suggest that understand- ing the causes and extent of these incompatibilities will be important for predicting the genetic cohesion and structure of bacterial species. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 14 / 22 PLOS BIOLOGYRecombination in the gut microbiome While our underlying approach relied on the presence of closely related strains to resolve individual recombination events, the widespread occurrence of these partially recombined genomes is still an interesting evolutionary puzzle. We previously showed [27] that the ecologi- cal structure of the human gut microbiome allows us to rule out common sampling biases (e.g., microepidemics or clonal blooms) that have been conjectured to play a role in other microbial species [21,36,46]. We also observed considerable variation across different com- mensal gut bacteria, with more than a quarter of the species in our panel containing just a handful of closely related strains from unrelated hosts. How could the same sample of hosts generate such a broad range of closely related strains in different species? The simplest neutral models predict a characteristic relationship between the mosaic timescale (Tmosaic/Tmrca) and the fraction of partially recombined genome pairs in the sample (Fig MMB in S1 Text) [42]. However, we found that the observed fractions are often much higher than this baseline expec- tation and show little correlation with the estimated recombination rates (Fig MMA in S1 Text). This suggests that new evolutionary models will be necessary to understand this puz- zling feature of many natural bacterial populations. Our results suggest that at least some of the long-term recombination dynamics across hosts arise from within-host sweeps of transferred fragments in hosts with multiple co-coloniz- ing strains. This could provide a potential mechanism for the strain-level variation in recombi- nation rates we observed in many species, since both the colonization structure and propensity for sweeps can vary dramatically in different hosts [27,70,82,87]. It remains unclear whether non-sweeping transfers could also play an important role in generating the long-term rates of recombination across hosts. Our results highlight the challenges involved in detecting these events, since we found that even unrelated strains can frequently share long stretches of DNA that are likely spreading through the global population via natural selection. These scenarios could potentially be distinguished with denser longitudinal sampling or larger samples of clonal isolates (e.g., using single-cell techniques [88]), which would allow us to distinguish between preexisting and in situ transfers [69]. While our present data do not provide direct information about the underlying mecha- nisms of horizontal DNA exchange in these species, our findings impose some interesting con- straints on the potential mechanisms that might be involved. Many of the species in our panel (e.g., Bacteroides) are not known to be naturally competent [89], but still have long-term recombination rates that are as high as other species that are (e.g., Streptococcus pneumoniae [47,90]). Many gut commensals are known to engage in conjugative transfer, both in vitro and in vivo [91]. However, the time required for bacterial conjugation carries a substantial oppor- tunity cost in the high growth regimes of the large intestine, and would need to be ameliorated by a corresponding fitness benefit or residence in a privileged spatial location [92]. Moreover, we observe little correlation between the overall rates of recombination in different species and their frequency of apparent multi-colonization (Fig NN in S1 Text). This suggests that these and other mechanisms that require physical proximity between strains are not the major driver of the long-term recombination rates we observed across hosts. It is possible that other species (e.g., phage or another commensal bacterium in the larger gut community) could serve as intermediate vectors for horizontal transfer between strains that are physically segregated in different hosts. Such inter-species transfer events have recently been observed within individ- ual gut microbiomes [3,11,14]. It remains to be seen whether the rates of this process are suffi- cient to generate the long-term recombination rates we observe within species. An important limitation of our metagenomic approach is that it is primarily restricted to recombination events within the core genome. While this provides important information about the long-term rates of recombination within gut commensal species, it is possible that much of this core-genome hybridization could be driven by positive selection on linked PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 15 / 22 PLOS BIOLOGYRecombination in the gut microbiome accessory genes (e.g., antibiotic resistance genes). Future applications of our methods on grow- ing collections of clonal isolates [93] could shed light on these functional targets of horizontal transfer [94], and thereby provide a fuller picture of the landscape of bacterial recombination within the gut microbiota. Supporting information S1 Table. Metadata of metagenomic samples used in this study. We analyzed a collection of 932 samples from 693 individuals, collated in a previous study [27]. This included samples from 250 individuals from the Human Microbiome Project (HMP) [95,96], 185 individuals from [97], 250 individuals from [98], and 8 individuals from [99]. Listed are the subject identi- fiers, sample identifiers, run accessions, country of the study, continent of the study, visit num- ber, and study. (TXT) S2 Table. Number of close pairs across species. This table contains statistics of closely related strains across 43 species in our cohort. For each species, we computed the fraction of identical genome blocks for all pairs of genomes from unique hosts and recorded the number of pairs with >20%, >50%, >80% identical blocks. This table also contains the number of genomes in each species (“num_qp_samples”). Some species (e.g., Prevotella copri, Roseburia inulinivorans) have substantially fewer closely related pairs than others with comparable number of genomes. (CSV) S3 Table. Detected transfers in the closely related pairs of 29 species. This table contains all the locations and divergences of recombination transfers shown in Figs 2 and 3. Listed are the species names, sample identifiers for each pair of strains, if the transfer is between-clade (“Y,” “N,” “NA”), if the transfer is included in Fig 2 (“TRUE,” “FALSE”), divergences (all sites or synonymous sites only), locations of transferred regions, and if the transfer is a potential dupli- cate of other detected transfers (“TRUE,” “FALSE”) (see Section 3.9 in S1 Text). (CSV) S4 Table. Species with high-quality dual-colonized samples. Listed are species with >5 high- quality dual-colonized samples that passed the filters described in Section 4.1 in S1 Text. (CSV) S5 Table. Annotations for genes in the within-host sweep example of Bacteroides vulgatus. Listed here are genes involved in the within-host sweep example in Fig 5 that have within-host SNVs at the first time point. Gene annotations are downloaded from PATRIC [100]. (CSV) S6 Table. Clonal divergence thresholds d* and clonal fraction thresholds f ∗ thresholds d* and clonal fraction thresholds f ∗ tion 3.3 in S1 Text). (CSV) c . Clonal fraction c for selecting close pairs in certain species (Sec- S7 Table. Metadata of isolate genomes used in Section 3.10 in S1 Text. Listed are the species names, species types (commensal or pathogen), genome accessions, and other information compiled in the Unified Human Gastrointestinal Genome (UHGG) collection [93]. (CSV) S1 Text. Methods and supplemental information. (PDF) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 16 / 22 PLOS BIOLOGYRecombination in the gut microbiome Acknowledgments We thank S. Maslov and G. Birzu for useful discussions, and D. Wong, S. Walton, and J. Fer- rare for comments and feedback on the manuscript. Author Contributions Conceptualization: Zhiru Liu, Benjamin H. Good. Formal analysis: Zhiru Liu, Benjamin H. Good. Funding acquisition: Benjamin H. Good. Investigation: Zhiru Liu, Benjamin H. Good. Writing – original draft: Zhiru Liu, Benjamin H. Good. Writing – review & editing: Zhiru Liu, Benjamin H. Good. References 1. Soucy SM, Huang J, Gogarten JP. Horizontal gene transfer: building the web of life. Nat Rev Genet. 2015; 16(88):472–482. https://doi.org/10.1038/nrg3962 PMID: 26184597 2. Smillie CS, Smith MB, Friedman J, Cordero OX, David LA, Alm EJ. Ecology drives a global network of gene exchange connecting the human microbiome. Nature. 2011; 480(73767376):241–244. https:// doi.org/10.1038/nature10571 PMID: 22037308 3. Groussin M, Poyet M, Sistiaga A, Kearney SM, Moniz K, Noel M, et al. Elevated rates of horizontal gene transfer in the industrialized human microbiome. Cell. 2021; 184(8):2053–2067.e18. https://doi. org/10.1016/j.cell.2021.02.052 PMID: 33794144 4. Kent AG, Vill AC, Shi Q, Satlin MJ, Brito IL. Widespread transfer of mobile antibiotic resistance genes within individual gut microbiomes revealed through bacterial Hi-C. Nat Commun. 2020; 11(11):4379. https://doi.org/10.1038/s41467-020-18164-7 PMID: 32873785 5. Smith HW. Transfer of antibiotic resistance from animal and human strains of Escherichia coli to resi- dent E. coli in the alimentary tract of man. Lancet. 1969; 293(7607):1174–1176. 6. von Wintersdorff CJH, Penders J, van Niekerk JM, Mills ND, Majumder S, van Alphen LB, et al. Dis- semination of Antimicrobial Resistance in Microbial Ecosystems through Horizontal Gene Transfer. Front Microbiol. 2016:7. https://doi.org/10.3389/fmicb.2016.00173 PMID: 26925045 7. Sheinman M, Arkhipova K, Arndt PF, Dutilh BE, Hermsen R, Massip F. Identical sequences found in distant genomes reveal frequent horizontal transfer across the bacterial domain. Elife. 2021; 10: e62719. https://doi.org/10.7554/eLife.62719 PMID: 34121661 8. Zlitni S, Bishara A, Moss EL, Tkachenko E, Kang JB, Culver RN, et al. Strain-resolved microbiome sequencing reveals mobile elements that drive bacterial competition on a clinical timescale. Genome Med. 2020; 12(1):1–17. https://doi.org/10.1186/s13073-020-00747-0 PMID: 32471482 9. Pa´l C, Papp B, Lercher MJ. Adaptive evolution of bacterial metabolic networks by horizontal gene transfer. Nat Genet. 2005; 37(1212):1372–1375. https://doi.org/10.1038/ng1686 PMID: 16311593 10. Hehemann JH, Correc G, Barbeyron T, Helbert W, Czjzek M, Michel G. Transfer of carbohydrate- active enzymes from marine bacteria to Japanese gut microbiota. Nature. 2010; 464(72907290):908– 912. https://doi.org/10.1038/nature08937 PMID: 20376150 11. Coyne MJ, Zitomersky NL, McGuire AM, Earl AM, Comstock LE. Evidence of Extensive DNA Transfer between Bacteroidales Species within the Human Gut. MBio. 2014; 5(3):e01305–e01314. https://doi. org/10.1128/mBio.01305-14 PMID: 24939888 12. Frazão N, Sousa A, La¨ssig M, Gordo I. Horizontal gene transfer overrides mutation in Escherichia coli colonizing the mammalian gut. Proc Natl Acad Sci U S A. 2019; 116(36):17906–17915. https://doi.org/ 10.1073/pnas.1906958116 PMID: 31431529 13. Pudlo NA, Pereira GV, Parnami J, Cid M, Markert S, Tingley JP, et al. Diverse events have transferred genes for edible seaweed digestion from marine to human gut bacteria. Cell Host Microbe. 2022; 30 (3):314–328.e11. https://doi.org/10.1016/j.chom.2022.02.001 PMID: 35240043 14. Garcı´a-Bayona L, Coyne MJ, Comstock LE. Mobile Type VI secretion system loci of the gut Bacteroi- dales display extensive intra-ecosystem transfer, multi-species spread and geographical clustering. PLoS Genet. 2021; 17(4):e1009541. https://doi.org/10.1371/journal.pgen.1009541 PMID: 33901198 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 17 / 22 PLOS BIOLOGYRecombination in the gut microbiome 15. Thomas CM, Nielsen KM. Mechanisms of, and Barriers to, Horizontal Gene Transfer between Bacte- ria. Nat Rev Microbiol. 2005; 3(99):711–721. https://doi.org/10.1038/nrmicro1234 PMID: 16138099 16. Hanage WP. Not So Simple After All: Bacteria, Their Population Genetics, and Recombination. Cold Spring Harb Perspect Biol. 2016; 8(7):a018069. https://doi.org/10.1101/cshperspect.a018069 PMID: 27091940 17. Fraser C, Hanage WP, Spratt BG. Recombination and the Nature of Bacterial Speciation. Science. 2007; 315(5811):476–480. https://doi.org/10.1126/science.1127573 PMID: 17255503 18. Bobay LM, Ochman H. Biological Species Are Universal across Life’s Domains. Genome Biol Evol. 2017; 9(3):491–501. https://doi.org/10.1093/gbe/evx026 PMID: 28186559 19. Dixit PD, Pang TY, Maslov S. Recombination-Driven Genome Evolution and Stability of Bacterial Spe- cies. Genetics. 2017; 207(1):281–295. https://doi.org/10.1534/genetics.117.300061 PMID: 28751420 20. Olm MR, Crits-Christoph A, Diamond S, Lavy A, Matheus Carnevali PB, Banfield JF. Consistent meta- genome-derived metrics verify and delineate bacterial species boundaries. Msystems. 2020; 5(1): e00731–e00719. https://doi.org/10.1128/mSystems.00731-19 PMID: 31937678 21. Smith JM, Smith NH, O’Rourke M, Spratt BG. How clonal are bacteria? Proc Natl Acad Sci U S A. 1993; 90(10):4384–4388. https://doi.org/10.1073/pnas.90.10.4384 PMID: 8506277 22. Neher RA, Shraiman BI. Competition between recombination and epistasis can cause a transition from allele to genotype selection. Proc Natl Acad Sci U S A. 2009; 106(16):6866–6871. https://doi.org/ 10.1073/pnas.0812560106 PMID: 19366665 23. Bendall ML, Stevens SL, Chan LK, Malfatti S, Schwientek P, Tremblay J, et al. Genome-wide selective sweeps and gene-specific sweeps in natural bacterial populations. ISME J. 2016; 10(77):1589–1601. https://doi.org/10.1038/ismej.2015.241 PMID: 26744812 24. Schubert S, Darlu P, Clermont O, Wieser A, Magistro G, Hoffmann C, et al. Role of Intraspecies Recombination in the Spread of Pathogenicity Islands within the Escherichia coli Species. PLoS Pathog. 2009; 5(1):e1000257. https://doi.org/10.1371/journal.ppat.1000257 PMID: 19132082 25. Vos M, Didelot X. A comparison of homologous recombination rates in bacteria and archaea. ISME J. 2009; 3(22):199–208. https://doi.org/10.1038/ismej.2008.93 PMID: 18830278 26. Rosen MJ, Davison M, Bhaya D, Fisher DS. Fine-scale diversity and extensive recombination in a qua- sisexual bacterial population occupying a broad niche. Science. 2015; 348(6238):1019–1023. https:// doi.org/10.1126/science.aaa4456 PMID: 26023139 27. Garud NR, Good BH, Hallatschek O, Pollard KS. Evolutionary dynamics of bacteria in the gut micro- biome within and across hosts. PLoS Biol. 2019; 17(1):e3000102. https://doi.org/10.1371/journal.pbio. 3000102 PMID: 30673701 28. Sakoparnig T, Field C, van Nimwegen E. Whole genome phylogenies reflect the distributions of recombination rates for many bacterial species. Elife. 2021; 10:e65366. https://doi.org/10.7554/eLife. 65366 PMID: 33416498 29. Croucher NJ, Page AJ, Connor TR, Delaney AJ, Keane JA, Bentley SD, et al. Rapid phylogenetic anal- ysis of large samples of recombinant bacterial whole genome sequences using Gubbins. Nucleic Acids Res. 2015; 43(3):e15–e15. https://doi.org/10.1093/nar/gku1196 PMID: 25414349 30. Didelot X, Falush D. Inference of Bacterial Microevolution Using Multilocus Sequence Data. Genetics. 2007; 175(3):1251–1266. https://doi.org/10.1534/genetics.106.063305 PMID: 17151252 31. Didelot X, Wilson DJ. ClonalFrameML: Efficient Inference of Recombination in Whole Bacterial Genomes. PLoS Comput Biol. 2015; 11(2):e1004041. https://doi.org/10.1371/journal.pcbi.1004041 PMID: 25675341 32. 33. Lin M, Kussell E. Correlated Mutations and Homologous Recombination Within Bacterial Populations. Genetics. 2017; 205(2):891–917. https://doi.org/10.1534/genetics.116.189621 PMID: 28007887 Lin M, Kussell E. Inferring bacterial recombination rates from large-scale sequencing datasets. Nat Methods. 2019; 16(22):199–204. https://doi.org/10.1038/s41592-018-0293-7 PMID: 30664775 34. Good BH. Linkage Disequilibrium between Rare Mutations. Genetics. 2022; 220(4):iyac004. https:// doi.org/10.1093/genetics/iyac004 PMID: 35100407 35. Arnold B, Sohail M, Wadsworth C, Corander J, Hanage WP, Sunyaev S, et al. Fine-Scale Haplotype Structure Reveals Strong Signatures of Positive Selection in a Recombining Bacterial Pathogen. Mol Biol Evol. 2020; 37(2):417–428. https://doi.org/10.1093/molbev/msz225 PMID: 31589312 36. Fraser C, Hanage WP, Spratt BG. Neutral microepidemic evolution of bacterial pathogens. Proc Natl Acad Sci U S A. 2005; 102(6):1968–1973. https://doi.org/10.1073/pnas.0406993102 PMID: 15684071 37. Hanage WP, Fraser C, Spratt BG. The impact of homologous recombination on the generation of diversity in bacteria. J Theor Biol. 2006; 239(2):210–219. https://doi.org/10.1016/j.jtbi.2005.08.035 PMID: 16236325 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 18 / 22 PLOS BIOLOGYRecombination in the gut microbiome 38. Johnson PLF, Slatkin M. Inference of Microbial Recombination Rates from Metagenomic Data. PLoS Genet. 2009; 5(10):e1000674. https://doi.org/10.1371/journal.pgen.1000674 PMID: 19798447 39. Dixit PD, Pang TY, Studier FW, Maslov S. Recombinant transfer in the basic genome of Escherichia coli. Proc Natl Acad Sci U S A. 2015; 112(29):9070–9075. https://doi.org/10.1073/pnas.1510839112 PMID: 26153419 40. Lynch M, Xu S, Maruki T, Jiang X, Pfaffelhuber P, Haubold B. Genome-Wide Linkage-Disequilibrium Profiles from Single Individuals. Genetics. 2014; 198(1):269–281. https://doi.org/10.1534/genetics. 114.166843 PMID: 24948778 41. Preska Steinberg A, Lin M, Kussell E. Core genes can have higher recombination rates than acces- sory genes within global microbial populations. Elife. 2022; 11:e78533. https://doi.org/10.7554/eLife. 78533 PMID: 35801696 42. Cui Y, Yang X, Didelot X, Guo C, Li D, Yan Y, et al. Epidemic Clones, Oceanic Gene Pools, and Eco- LD in the Free Living Marine Pathogen Vibrio parahaemolyticus. Mol Biol Evol. 2015; 32(6):1396– 1410. https://doi.org/10.1093/molbev/msv009 PMID: 25605790 43. Borody TJ, Khoruts A. Fecal microbiota transplantation and emerging applications. Nat Rev Gastroen- terol Hepatol. 2012; 9(22):88–96. https://doi.org/10.1038/nrgastro.2011.244 PMID: 22183182 44. David LA, Maurice CF, Carmody RN, Gootenberg DB, Button JE, Wolfe BE, et al. Diet rapidly and reproducibly alters the human gut microbiome. Nature. 2014; 505(74847484):559–563. https://doi. org/10.1038/nature12820 PMID: 24336217 45. Truong DT, Tett A, Pasolli E, Huttenhower C, Segata N. Microbial strain-level population structure and genetic diversity from metagenomes. Genome Res. 2017; 27(4):626–638. https://doi.org/10.1101/gr. 216242.116 PMID: 28167665 46. Roach DJ, Burton JN, Lee C, Stackhouse B, Butler-Wu SM, Cookson BT, et al. A Year of Infection in the Intensive Care Unit: Prospective Whole Genome Sequencing of Bacterial Clinical Isolates Reveals Cryptic Transmissions and Novel Microbiota. PLoS Genet. 2015; 11(7):e1005413. https://doi.org/10. 1371/journal.pgen.1005413 PMID: 26230489 47. Croucher NJ, Harris SR, Fraser C, Quail MA, Burton J, van der Linden M, et al. Rapid Pneumococcal Evolution in Response to Clinical Interventions. Science. 2011; 331(6016):430–434. https://doi.org/10. 1126/science.1198545 PMID: 21273480 48. Calland JK, Pascoe B, Bayliss SC, Mourkas E, Berthenet E, Thorpe HA, et al. Quantifying bacterial evolution in the wild: A birthday problem for Campylobacter lineages. PLoS Genet. 2021; 17(9): e1009829. https://doi.org/10.1371/journal.pgen.1009829 PMID: 34582435 49. Costea PI, Coelho LP, Sunagawa S, Munch R, Huerta-Cepas J, Forslund K, et al. Subspecies in the global human gut microbiome. Mol Syst Biol. 2017; 13(12):960. https://doi.org/10.15252/msb. 20177589 PMID: 29242367 50. Licht TR, Wilcks A. Conjugative gene transfer in the gastrointestinal environment. Adv Appl Microbiol. 2005; 58:77–95. https://doi.org/10.1016/S0065-2164(05)58002-X PMID: 16543030 51. Oliveira PH, Touchon M, Cury J, Rocha EPC. The chromosomal organization of horizontal gene trans- fer in bacteria. Nat Commun. 2017; 8(11):841. https://doi.org/10.1038/s41467-017-00808-w PMID: 29018197 52. Carr VR, Shkoporov A, Hill C, Mullany P, Moyes DL. Probing the Mobilome: Discoveries in the Dynamic Microbiome. Trends Microbiol. 2021; 29(2):158–170. https://doi.org/10.1016/j.tim.2020.05. 003 PMID: 32448763 53. Zhou Z, McCann A, Weill FX, Blin C, Nair S, Wain J, et al. Transient Darwinian selection in Salmonella enterica serovar Paratyphi A during 450 years of global spread of enteric fever. Proc Natl Acad Sci U S A. 2014; 111(33):12199–12204. https://doi.org/10.1073/pnas.1411012111 PMID: 25092320 54. Bakir MA, Sakamoto M, Kitahara M, Matsumoto M, Benno Y. Bacteroides dorei sp. nov., isolated from human faeces. Int J Syst Evol Microbiol. 2006; 56(7):1639–1643. https://doi.org/10.1099/ijs.0.64257-0 PMID: 16825642 55. Shen P, Huang HV. Homologous recombination in Escherichia coli: dependence on substrate length and homology. Genetics. 1986; 112(3):441–457. https://doi.org/10.1093/genetics/112.3.441 PMID: 3007275 56. Majewski J, Cohan FM. DNA Sequence Similarity Requirements for Interspecific Recombination in Bacillus. Genetics. 1999; 153(4):1525–1533. https://doi.org/10.1093/genetics/153.4.1525 PMID: 10581263 57. Arnold BJ, Gutmann MU, Grad YH, Sheppard SK, Corander J, Lipsitch M, et al. Weak Epistasis May Drive Adaptation in Recombining Bacteria. Genetics. 2018; 208(3):1247–1260. https://doi.org/10. 1534/genetics.117.300662 PMID: 29330348 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 19 / 22 PLOS BIOLOGYRecombination in the gut microbiome 58. Wiedenbeck J, Cohan FM. Origins of bacterial diversity through horizontal genetic transfer and adap- tation to new ecological niches. FEMS Microbiol Rev. 2011; 35(5):957–976. https://doi.org/10.1111/j. 1574-6976.2011.00292.x PMID: 21711367 59. Zahrt TC, Maloy S. Barriers to recombination between closely related bacteria: MutS and RecBCD inhibit recombination between Salmonella typhimurium and Salmonella typhi. Proc Natl Acad Sci U S A. 1997; 94(18):9786–9791. https://doi.org/10.1073/pnas.94.18.9786 PMID: 9275203 60. Vulic M, Lenski RE, Radman M. Mutation, recombination, and incipient speciation of bacteria in the laboratory. Proc Natl Acad Sci U S A. 1999; 96(13):7348–7351. https://doi.org/10.1073/pnas.96.13. 7348 PMID: 10377417 61. Didelot X, Me´ric G, Falush D, Darling AE. Impact of homologous and non-homologous recombination in the genomic evolution of Escherichia coli. BMC Genomics. 2012; 13(1):256. https://doi.org/10.1186/ 1471-2164-13-256 PMID: 22712577 62. Didelot X, Bowden R, Street T, Golubchik T, Spencer C, McVean G, et al. Recombination and Popula- tion Structure in Salmonella enterica. PLoS Genet. 2011; 7(7):e1002191. https://doi.org/10.1371/ journal.pgen.1002191 PMID: 21829375 63. Dillon MM, Thakur S, Almeida RND, Wang PW, Weir BS, Guttman DS. Recombination of ecologically and evolutionarily significant loci maintains genetic cohesion in the Pseudomonas syringae species complex. Genome Biol. 2019; 20(1):3. https://doi.org/10.1186/s13059-018-1606-y PMID: 30606234 64. Vulić M, Dionisio F, Taddei F, Radman M. Molecular keys to speciation: DNA polymorphism and the control of genetic exchange in enterobacteria. Proc Natl Acad Sci U S A. 1997; 94(18):9763–9767. https://doi.org/10.1073/pnas.94.18.9763 PMID: 9275198 65. Budroni S, Siena E, Hotopp JCD, Seib KL, Serruto D, Nofroni C, et al. Neisseria meningitidis is struc- tured in clades associated with restriction modification systems that modulate homologous recombina- tion. Proc Natl Acad Sci U S A. 2011; 108(11):4494–4499. https://doi.org/10.1073/pnas.1019751108 PMID: 21368196 66. Claus H, Friedrich A, Frosch M, Vogel U. Differential Distribution of Novel Restriction-Modification Sys- tems in Clonal Lineages of Neisseria meningitidis. J Bacteriol. 2000; 182(5):1296–1303. https://doi. org/10.1128/JB.182.5.1296-1303.2000 PMID: 10671450 67. Oliveira PH, Touchon M, Rocha EPC. Regulation of genetic flux between bacteria by restriction–modi- fication systems. Proc Natl Acad Sci U S A. 2016; 113(20):5658–5663. https://doi.org/10.1073/pnas. 1603257113 PMID: 27140615 68. Nandi T, Holden MTG, Didelot X, Mehershahi K, Boddey JA, Beacham I, et al. Burkholderia pseudo- mallei sequencing identifies genomic clades with distinct recombination, accessory, and epigenetic profiles. Genome Res. 2015; 25(1):129–141. https://doi.org/10.1101/gr.177543.114 PMID: 25236617 69. Cao Q, Didelot X, Wu Z, Li Z, He L, Li Y, et al. Progressive genomic convergence of two Helicobacter pylori strains during mixed infection of a patient with chronic gastritis. Gut. 2015; 64(4):554–561. https://doi.org/10.1136/gutjnl-2014-307345 PMID: 25007814 70. Roodgar M, Good BH, Garud NR, Martis S, Avula M, Zhou W, et al. Longitudinal linked-read sequenc- ing reveals ecological and evolutionary responses of a human gut microbiome during antibiotic treat- ment. Genome Res. 2021; 31(8):1433–1446. https://doi.org/10.1101/gr.265058.120 PMID: 34301627 71. Aggarwala V, Mogno I, Li Z, Yang C, Britton GJ, Chen-Liaw A, et al. Precise quantification of bacterial strains after fecal microbiota transplantation delineates long-term engraftment and explains outcomes. Nat Microbiol. 2021; 6(1010):1309–1318. https://doi.org/10.1038/s41564-021-00966-0 PMID: 34580445 72. Zhao S, Lieberman TD, Poyet M, Kauffman KM, Gibbons SM, Groussin M, et al. Adaptive Evolution within Gut Microbiomes of Healthy People. Cell Host Microbe. 2019; 25(5):656–667.e8. https://doi. org/10.1016/j.chom.2019.03.007 PMID: 31028005 73. Harris K, Nielsen R. Inferring Demographic History from a Spectrum of Shared Haplotype Lengths. PLoS Genet. 2013; 9(6):e1003521. https://doi.org/10.1371/journal.pgen.1003521 PMID: 23754952 74. Yahara K, Didelot X, Ansari MA, Sheppard SK, Falush D. Efficient Inference of Recombination Hot Regions in Bacterial Genomes. Mol Biol Evol. 2014; 31(6):1593–1605. https://doi.org/10.1093/ molbev/msu082 PMID: 24586045 75. Yahara K, Didelot X, Jolley KA, Kobayashi I, Maiden MCJ, Sheppard SK, et al. The Landscape of Realized Homologous Recombination in Pathogenic Bacteria. Mol Biol Evol. 2016; 33(2):456–471. https://doi.org/10.1093/molbev/msv237 PMID: 26516092 76. Hermisson J, Pennings PS. Soft sweeps and beyond: understanding the patterns and probabilities of selection footprints under rapid adaptation. Methods Ecol Evol. 2017; 8(6):700–716. https://doi.org/10. 1111/2041-210X.12808 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 20 / 22 PLOS BIOLOGYRecombination in the gut microbiome 77. Livingston G, Matias M, Calcagno V, Barbera C, Combe M, Leibold MA, et al. Competition–coloniza- tion dynamics in experimental bacterial metacommunities. Nat Commun. 2012; 3(11):1234. https:// doi.org/10.1038/ncomms2239 PMID: 23212363 78. Shi ZJ, Dimitrov B, Zhao C, Nayfach S, Pollard KS. Fast and accurate metagenotyping of the human gut microbiome with GT-Pro. Nat Biotechnol. 2022; 40(44):507–516. https://doi.org/10.1038/s41587- 021-01102-3 PMID: 34949778 79. Suzuki TA, Fitzstevens JL, Schmidt VT, Enav H, Huus KE, Mbong Ngwese M, et al. Codiversification of gut microbiota with humans. Science. 2022; 377(6612):1328–1332. https://doi.org/10.1126/ science.abm7759 PMID: 36108023 80. Good BH. Limited codiversification of the gut microbiota with humans. bioRxiv. 2023. https://doi.org/ 10.1101/2022.10.27.514143 81. Good BH, Desai MM. Deleterious Passengers in Adapting Populations. Genetics. 2014; 198(3):1183– 1208. https://doi.org/10.1534/genetics.114.170233 PMID: 25194161 82. Poyet M, Groussin M, Gibbons SM, Avila-Pacheco J, Jiang X, Kearney SM, et al. A library of human gut bacterial isolates paired with longitudinal multiomics data enables mechanistic microbiome research. Nat Med. 2019; 25(99):1442–1452. https://doi.org/10.1038/s41591-019-0559-3 PMID: 31477907 83. Yang C, Pei X, Wu Y, Yan L, Yan Y, Song Y, et al. Recent mixing of Vibrio parahaemolyticus popula- tions. ISME J. 2019; 13(1010):2578–2588. https://doi.org/10.1038/s41396-019-0461-5 PMID: 31235840 84. Arevalo P, VanInsberghe D, Elsherbini J, Gore J, Polz MF. A Reverse Ecology Approach Based on a Biological Definition of Microbial Populations. Cell. 2019; 178(4):820–834.e14. https://doi.org/10. 1016/j.cell.2019.06.033 PMID: 31398339 85. Wielgoss S, Didelot X, Chaudhuri RR, Liu X, Weedall GD, Velicer GJ, et al. A barrier to homologous recombination between sympatric strains of the cooperative soil bacterium Myxococcus xanthus. ISME J. 2016; 10(1010):2468–2477. https://doi.org/10.1038/ismej.2016.34 PMID: 27046334 86. Cohan FM. Transmission in the Origins of Bacterial Diversity, From Ecotypes to Phyla. Microbiology. Spectrum. 2017; 5(5):5.5.13. https://doi.org/10.1128/microbiolspec.MTBP-0014-2016 PMID: 29027519 87. Chen DW, Garud NR. Rapid evolution and strain turnover in the infant gut microbiome. Genome Res. 2022:gr.276306.121. https://doi.org/10.1101/gr.276306.121 PMID: 35545448 88. Zheng W, Zhao S, Yin Y, Zhang H, Needham DM, Evans ED, et al. High-throughput, single-microbe genomics with strain resolution, applied to a human gut microbiome. Science. 2022; 376(6597): eabm1483. https://doi.org/10.1126/science.abm1483 PMID: 35653470 89. Husain F, Tang K, Veeranagouda Y, Boente R, Patrick S, Blakely G, et al. Novel large-scale chromo- somal transfer in Bacteroides fragilis contributes to its pan-genome and rapid environmental adapta- tion. Microbial Genomics. 2017; 3(11):e000136. https://doi.org/10.1099/mgen.0.000136 PMID: 29208130 90. Salvadori G, Junges R, Morrison DA, Petersen FC. Competence in Streptococcus pneumoniae and Close Commensal Relatives: Mechanisms and Implications. Frontiers in Cellular and Infection. Micro- biology. 2019:9. 91. Neil K, Allard N, Grenier F, Burrus V, Rodrigue S. Highly efficient gene transfer in the mouse gut micro- biota is enabled by the Incl2 conjugative plasmid TP114. Commun Biol. 2020; 3(11):1–9. https://doi. org/10.1038/s42003-020-01253-0 PMID: 32963323 92. Ghosh OM, Good BH. Emergent evolutionary forces in spatial models of luminal growth and their application to the human gut microbiota. Proc Natl Acad Sci U S A. 2022; 119(28):e2114931119. https://doi.org/10.1073/pnas.2114931119 PMID: 35787046 93. Almeida A, Nayfach S, Boland M, Strozzi F, Beracochea M, Shi ZJ, et al. A unified catalog of 204,938 reference genomes from the human gut microbiome. Nat Biotechnol. 2021; 39(11):105–114. https:// doi.org/10.1038/s41587-020-0603-3 PMID: 32690973 94. Mehta RS, Petit RA, Read TD, Weissman DB. Detecting Patterns of Accessory Genome Coevolution in Staphylococcus Aureus Using Data from Thousands of Genomes. BMC Bioinformatics. 2023; 24 (1):243. https://doi.org/10.1186/s12859-023-05363-4 PMID: 37296404 95. Consortium HMP. A framework for human microbiome research. Nature. 2012; 486(74027402):215– 221. https://doi.org/10.1038/nature11209 PMID: 22699610 96. Lloyd-Price J, Mahurkar A, Rahnavard G, Crabtree J, Orvis J, Hall AB, et al. Strains, functions and dynamics in the expanded Human Microbiome Project. Nature. 2017; 550(7674):61–66. https://doi. org/10.1038/nature23889 PMID: 28953883 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 21 / 22 PLOS BIOLOGYRecombination in the gut microbiome 97. Qin J, Li Y, Cai Z, Li S, Zhu J, Zhang F, et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature. 2012; 490(7418):55–60. https://doi.org/10.1038/nature11450 PMID: 23023125 98. Xie H, Guo R, Zhong H, Feng Q, Lan Z, Qin B, et al. Shotgun metagenomics of 250 adult twins reveals genetic and environmental impacts on the gut microbiome. Cell Syst. 2016; 3(6):572–584. https://doi. org/10.1016/j.cels.2016.10.004 PMID: 27818083 99. Korpela K, Costea P, Coelho LP, Kandels-Lewis S, Willemsen G, Boomsma DI, et al. Selective mater- nal seeding and environment shape the human gut microbiome. Genome Res. 2018; 28(4):561–568. https://doi.org/10.1101/gr.233940.117 PMID: 29496731 100. Davis JJ, Wattam AR, Aziz RK, Brettin T, Butler R, Butler RM, et al. The PATRIC Bioinformatics Resource Center: expanding data and analysis capabilities. Nucleic Acids Res. 2020; 48(D1):D606– D612. https://doi.org/10.1093/nar/gkz943 PMID: 31667520 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002472 February 8, 2024 22 / 22 PLOS BIOLOGY
10.1371_journal.pcbi.1011739	RESEARCH ARTICLE Explaining the flaws in human random generation as local sampling with momentum Lucas CastilloID 1, Pablo Leo´ n-Villagra´ ID 2, Nick Chater3, Adam Sanborn1 1 Department of Psychology, University of Warwick, Coventry, United Kingdom, 2 Cognitive, Linguistic & Psychological Sciences, Brown University, Providence, Rhode Island, United States of America, 3 Warwick Business School, University of Warwick, Coventry, United Kingdom [email protected] Abstract In many tasks, human behavior is far noisier than is optimal. Yet when asked to behave ran- domly, people are typically too predictable. We argue that these apparently contrasting observations have the same origin: the operation of a general-purpose local sampling algo- rithm for probabilistic inference. This account makes distinctive predictions regarding ran- dom sequence generation, not predicted by previous accounts—which suggests that randomness is produced by inhibition of habitual behavior, striving for unpredictability. We verify these predictions in two experiments: people show the same deviations from random- ness when randomly generating from non-uniform or recently-learned distributions. In addi- tion, our data show a novel signature behavior, that people’s sequences have too few changes of trajectory, which argues against the specific local sampling algorithms that have been proposed in past work with other tasks. Using computational modeling, we show that local sampling where direction is maintained across trials best explains our data, which sug- gests it may be used in other tasks too. While local sampling has previously explained why people are unpredictable in standard cognitive tasks, here it also explains why human ran- dom sequences are not unpredictable enough. Author summary When explicitly asked to be random, people are not random enough. Previous accounts of these random generation tasks have argued that people are effortfully trying not to be pre- dictable. In many other tasks, however, people also show random behavior, even when it is unnecessary or outright disadvantageous. Here, we try to bridge this apparent gap. We hypothesize that the randomness people produce when trying to be random and the ran- domness that they display when trying to make the best choice has the same common mechanism: drawing mental samples to make judgments and decisions. In two experi- ments, we compare previous random generation accounts, which are task-specific in nature, to the more general account of mental sampling that has been used to explain how people behave in many other domains. We find that the flexibility of human random gen- eration in our data is better explained by the mental sampling account. We also find a novel empirical signature of momentum in random generation, which points to a new a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Castillo L, Leo´n-Villagra´ P, Chater N, Sanborn A (2024) Explaining the flaws in human random generation as local sampling with momentum. PLoS Comput Biol 20(1): e1011739. https://doi.org/10.1371/journal.pcbi.1011739 Editor: Ulrik R. Beierholm, Durham University, UNITED KINGDOM Received: June 14, 2023 Accepted: December 5, 2023 Published: January 5, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pcbi.1011739 Copyright: © 2024 Castillo et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All data and computational code are available on OSF at https:// osf.io/dw8ez/. Funding: LC, PLV and ANS were funded by a European Research Council (https://erc.europa.eu/) PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 1 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Consolidator grant (817492-SAMPLING) awarded to ANS. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. kind of mental sampling algorithm. If mental sampling governs behavior in random gen- eration tasks and elsewhere, then this task has great promise in helping to understand wider human behavior. Introduction In many tasks, people behave with some degree of randomness, even when they are not required to do so. For example, in a task involving several repeated gambles, participants might make a different choice when presented with the same set of options for a second time [1]. Surprisingly, people behave randomly even to their disadvantage: when given options with different reward probabilities, participants will choose each alternative proportionally to the probability of being rewarded, rather than always choose the most advantageous option [2, 3]. To account for these inconsistencies in people’s behavior, almost all models of cognition postulate that mental mechanisms include sources of randomness, typically modeled as independent, identically distributed (iid) samples. This source of randomness within cogni- tion is used to explain the noisiness of behavior, whether in higher-level cognitive processes such as categorization or decision-making [4–7], and in lower-level processes such as per- ception [8, 9]. Whether people can produce randomness has also been studied more directly, by asking participants to generate sequences of items unpredictably. Having the ability to behave ran- domly is important in adversarial situations, where being unpredictable is the optimal behav- ior [10, 11], and perceiving someone as unpredictable is seen as an indicator of free will [12, 13]. In addition, departures from randomness have helped model the cognitive architectures of neurotypical and neurodivergent populations [14, 15]. In a typical random generation experiment, participants are given a set of items (usually numbers from 1 to 10) and are asked to produce them unpredictably, which instructions will often exemplify as drawing items ‘out of a hat’ with replacement. Variations of the task have involved doing the task vocally or using a keyboard or mouse [16], performing the task at dif- ferent speeds [17], while multi-tasking [18], or collaboratively [19]. Paradoxically, despite people’s tendency to behave randomly in a myriad of domains that do not require them to, this body of work has found that when asked to be unpredictable, peo- ple are not random enough. Across experiments, the same picture emerges: people’s sequences are typically more compressible than truly random sequences [15] (c.f. [20]) and display pre- dictable patterns of serial dependence [21, 22], thus deviating significantly from the iid sam- pling that many cognitive models include. Previous accounts of people’s behavior in random generation tasks make no connection between people’s excessively unpredictable behavior in many perceptual and cognitive tasks from their performance when explicitly generating random sequences. Instead, they character- ize being random as the product of effortful behavior. For example, in their network modula- tion model Jahanshahi et al. [23] theorize that in a random generation task people create an associative network with each possible response as a node, and with links (representing the probability of transitioning from one item to another) having weights proportional to the items’ strength of association. If operating alone, this network would produce greatly stereo- typed responses, but a second component of their model—the controller—inhibits the stron- gest links to enable variability, while monitoring the output to modulate its intervention on the network. Similarly, another popular account, Baddeley’s schema account [24], postulates that in random generation tasks people follow deterministic, learned action sets (schemas), PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 2 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum and switch between sets based on a monitoring process that evaluates how unpredictable the resulting sequence is and changes strategy if randomness is perceived to decline. According to these accounts, people’s deviations from randomness can be explained by biases regarding which schemas are preferred and limitations in how often schemas are changed and how well randomness is monitored; or by limitations in the controller’s ability to inhibit links in the associative network. While these approaches can explain people’s flaws when producing random sequences, they are, however, specific to the requirements of the ran- dom generation task, and would not apply to other tasks in which randomness has been observed. Recent work has, however, raised the possibility that a single mechanism—local sampling— might explain both the excessive noisiness of many aspects of human behavior and the exces- sive predictability of random sequence generation. It turns out that randomness in other tasks is also typically not iid [25]. Instead, the noise in people’s behavior has a rich structure, includ- ing long-term autocorrelations [26, 27]. Moreover, it has been suggested that these patterns arise from a general-purpose approximation to probabilistic inference [28] widely used in sta- tistics and machine learning. These local sampling algorithms generate new samples from the previous one, creating sequential dependencies [29]. Local sampling algorithms have been used to explain how people reason with probabilities [28], including the characteristic judg- ment errors people make [30, 31], and have also been proposed in causal learning [32], bistable perception [8], memory retrieval [33], and elsewhere [34]. Here, we postulate that local sampling underpins the attempt to generate sequences. This would have implications for the domains in which random generation has been employed, that is, for how we understand people’s behavior in adversarial situations, free will, and neuro- diverse populations. In addition, random generation tasks could reveal undiscovered aspects of the underlying local sampling mechanism, shedding light on how people perform a wide range of tasks involving probabilistic inference. Previous accounts of random generation have been constructed to deal with cases where people must choose from a uniform distribution over a single dimension (e.g. numbers), often with ordered items, and expect that people will draw samples uniformly: the network modula- tion model achieves unpredictability by attempting to make each possible bigram equally likely, while the schema model does so by changing which schemas are used based on unpre- dictability of the sequence alone. Crucially, the local sampling account is much more general, predicting that people will be able to draw samples from any distribution while matching their probabilities, including non-uniform distributions or multivariate distributions. This leads to a crucial differential prediction between our proposal and previous models, which center on uniform distributions and which postulate that people strive for unpredictability only. In con- trast, local sampling accounts assume that sampling will match the underlying distribution density learned by the participant. A second differential prediction is that, according to the schema account and the network modulation model, many of the patterns that arise from human random generation do so from the existence of habitual behaviors that must be inhibited. In case of the network modula- tion model, the associations between items have different strengths, and the controller evens these out to increase variability. In case of the schema account, transitions between items are due to the application of well-learned transformations (schemas). These accounts have been applied to tasks where the items to randomize were known beforehand, and so it is unclear from these accounts how participants would perform when trying to randomize sets that have recently been learned. Here, we devised two novel random generation tasks exploring whether people can gener- ate non-uniformly-distributed items, or items they have recently been learned. Importantly, PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 3 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum we wanted to see whether people generated items in these novel tasks while still displaying the typical departures from randomness that are often seen in the typical random generation para- digm, where items are well-known and uniformly distributed, as that would point to the same cognitive mechanism being used throughout. In Experiment 1, we asked participants to pro- duce a sequence of random heights, and tested whether their sequences reflected the true, approximately Gaussian, distribution of heights. In Experiment 2, we taught participants a set of items configured in either a one-dimensional or a two-dimensional display, and tested whether they could generate random items in these domains. We found that people could generate random items in these new tasks, and that their sequences exhibited the same systematic deviations from randomness found elsewhere in the random generation literature [35], pointing to a common mechanism across all these tasks. Crucially, as the local sampling account predicts, participants were sensitive to the distribu- tional properties of the domain, being able to reproduce non-uniform distributions in their samples. Finally, we observed a key systematic deviation from iid sampling—that people follow the same trajectory for multiple trials over and above what would be expected from their mak- ing small transitions only. We computationally modeled these qualitative observations, identifying which forms of local sampling explain human data best and contrasting them with Cooper’s [36] schema model. Because no computational model is available for the network modulation account, we do not include it in our model comparisons, but return to it in our discussion. We found that data were most closely matched by a local sampling algorithm with “recycled momentum”, an algorithm which has not previously been suggested to underlie human sampling. Our analysis also shows that random generation tasks are useful for identifying subtle differences between different candidate algorithms, opening the door to more such experiments in future. Results In Experiment 1, we tested whether participants could sample random items non-uniformly while displaying the same deviations from iid. Participants produced a random sequence of heights of either men or women in the United Kingdom. In one sequence, they sampled heights as distributed according to a uniform distribution (Uniform condition); in the other sequence, heights were distributed following their actual distribution (which is roughly Gauss- ian [37], and so we term this the Gaussian condition). The order in which participants pro- duced these sequences was counterbalanced. In Experiment 2, we tested whether participants could sample from a set of novel items while displaying the same deviations from iid. Partici- pants first learned a set of syllables arranged in either a single row (one-dimensional condi- tion) or a grid (two-dimensional condition; see Fig 1B), then produced two random sequences for the same display. Deviations from iid sampling To evaluate whether people deviated from iid sampling in the same way as in previous random generation experiments, we evaluated several properties of the sequences commonly investi- gated in past research [21] and whether these deviated from the properties expected from a random sequence [38] (see Fig 1 for examples of the first three measures). We focused on devi- ations from serial independence, as they can easily be applied to non-uniform distributions and are more interpretable than compressibility measures, which can in turn inform building better cognitive models. The indices were: • Repetitions: The proportion of transitions where the new item was a repetition of the last. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 4 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Fig 1. Example measures. Examples of a repetition (blue), an adjacency (yellow) and a turning point (red) in (A) a sequence of normally-distributed heights (in cm) in the Gaussian condition of Experiment 1, and (B) a sequence of syllables whose arrangement had recently been learned, for the two- dimensional condition of Experiment 2. Woman’s outline image source: https://commons.wikimedia.org/wiki/File:Black_-_replace_this_image_ female.svg. https://doi.org/10.1371/journal.pcbi.1011739.g001 • Adjacencies: The proportion of transitions where the new item was one unit distance away to the one prior. To ensure that this analysis reflected people’s transition patterns, we analyzed this measure after removing all repeated items from the sequence. By applying this correction, we avoided having values for this measure depend on how often repetitions occur (for exam- ple, without this correction, Repetitions and Adjacencies would be negatively correlated). • Turning Points: The proportion of transitions that did not follow the previous direction. In previous random generation experiments using the number line, this has been defined as a transition that begins a descending run after having followed an ascending run (e.g. “1, 4, 2”), or vice versa. For Experiment 2, we generalize this measure to describe turning points in the spatial displays: we define it as a transition for which the absolute difference between the current and previous direction is larger than 90 degrees (see Fig 1). Again, we analyzed this measure after removing all repeated items from the sequence. • Distances: The average Euclidean distance traveled in each transition. We analyzed this mea- sure after removing all repeated items from the sequence. In previous random generation literature, people generating numbers or letters have been found to deviate from iid sampling in that they repeat items infrequently and transition between items making small jumps and following the same trajectory for multiple utterances. For this reason, if the same mechanism was used to generate items in these novel tasks, we would expect higher Adjacencies and lower Distances, lower Repetitions, and lower Turning Points, than iid sampling. To compare participant’s values to those that would be expected from iid sampling, we reshuffled each participant’s sequence 104 times and obtained the average value of each index across reshuffled sequences. If an index required removal of repetitions, we first reshuffled the original sequence with all items, and then removed items that were a repetition of the last in the new reordering. We ran generalized linear mixed-effects models predicting the observed values, using a logit link function for the first three measures and the identity link function for Distances (see Eq 1). We included the iid expectation as an offset variable (coefficient set to 1), so that the value of β0 represented the difference between observed and expected values. We also included a random intercept per participant (ui). ObservedValue ¼ b0 þ 1 � ExpectedValue þ ui ð1Þ PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 5 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum We also compared these results to a model that included a regressor on experimental condi- tion (Eq 2), which allowed us to examine whether potential deviations from iid sampling depended on the domain participants were producing from (β1). ObservedValue ¼ b0 þ b1 � Condition þ 1 � iidValue þ ui ð2Þ We only report condition differences where the conditions displayed different qualitative trends, relegating other analyses to S2 Text. Finally, because each participant produced two sequences, we added Order and Order × Condition terms to the model above to ensure that the above results did not depend on whether the sequence was their first or their second. We found no qualitative differences due to either term, and so we relegate the report on those analyses to S3 Text. Experiment 1. We found that participants deviated from their reshuffled sequences in the same systematic way as in previous random generation experiments. Compared to their reshuffled sequences, participants’ values were lower for Repetitions (Observed = .015, Expected = .043, Z = −2.61, p = .009, d = −0.67, BF10 = 4), Turning Points (Obs. = .47, Exp. = .65, Z = −11.12, p < .001, d = −0.64, BF10 = 4.2 × 106) and Distances (Obs. = 9.98, Exp. = 18.42, t(18.98) = −3.36, p = .003, d = −0.42, BF10 = 7), and higher for Adjacencies (Obs. = .22, Exp. = .07, Z = 6.15, p < .001, d = 0.97, BF10 = 2.6 × 103). Experiment 2. Likewise, in Experiment 2, participants deviated from their reshuffled sequences in the same systematic way as in previous random generation experiments. Both in the one-dimensional and two-dimensional conditions, participants had lower Repetitions (Obs. = .04, Exp. = .15, Z = −9.55, p < .001, d = −1.31, BF10 = 4.6 × 107) and lower Turning Points (Obs. = .66, Exp. = 0.70, Z = −3.70; p < .001, d = −0.17, BF10 = 5). In the one-dimen- sional condition, they also had lower Distances (Obs. = 2.22, Exp. = 2.70, t(19.01) = −6.58, p < .001, d = −0.32, BF10 = 703) and higher Adjacencies (Obs. = .46, Exp. = .29, Z = 6.27, p < .001, d = 0.61, BF10 = 340) than iid, but these did not differ in the two-dimensional condition (Obs. = 1.47, Exp. = 1.37, t(18.99) = 1.60, p = .13, d = 0.13, BF10 = 1/8; and Obs. = .58, Exp. = .55, Z = 1.61, p = .11, d = 0.08, BF10 = 1/12; respectively). Distributional sensitivity A key distinction between the predictions of the schema and local sampling accounts is whether people can generate random sequences that do not follow a uniform distribution. Comparing aggregate and individual distributions corresponding to the random sequences, we found that, in aggregate, Uniform and Gaussian conditions produced different item distributions (see Fig 2A). This difference was not merely due to data aggregation, since most of the individual partic- ipants also produced different distributions in the two conditions (see Fig 2C). In order to quantitatively study whether participants’ sequences resembled a uniform or a Gaussian distri- bution more, we developed a new measure, which we term Shape, which was calculated as: S ¼ 1 N XN n¼1 lpdfGðxnÞ (cid:0) lpdfUðxnÞ where lpdfG(�) and lpdfU(�) are the log densities of the best fitting Gaussian and uniform distri- bution respectively, xn is each item in the sequence, and N is the total length of the sequence. Shape values are positive when a sequence is better described by a Gaussian distribution rather than a uniform distribution, and vice versa. We did not remove repeated items before calculat- ing this measure. Initially, we pre-registered that we would use a different measure: we would fit sequences to a normal and a uniform distribution, and use BIC values to classify participants’ sequences. However, we decided against using this measure in favor of the Shape measure. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 6 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Fig 2. Distributional Sensitivity Results. Distribution of items for participants in the Gaussian and Uniform conditions in Experiment 1. To ease visualization, we normalized each participant’s values. (A) We calculated the average proportion of values of each participant in each of thirty bins, so that each participant had equal weight in the resulting plot. Comparing the aggregate histograms shows that in the uniform condition participants had a flatter distribution. (B) Shape measure of each participant for the Gaussian and Uniform conditions, with the dashed line representing equal values for both conditions. Error bars are 95% confidence intervals (obtained via bootstrapping). Most participants had a more Gaussian sequence in the Gaussian condition (participants below the dashed line). Although most participants lie in the second quadrant (shaded), meaning that they had a Gaussian sequence in the Gaussian condition and a uniform sequence in the Uniform condition, several other participants lie in the first quadrant, meaning their sequences were Gaussian in both conditions. (C) Histogram of each participant’s normalized values for each condition. For most participants, there’s a clear difference between conditions, with the Gaussian values being more concentrated. https://doi.org/10.1371/journal.pcbi.1011739.g002 When simulating normal and uniform sequences, we found that the Type II error rate was three times smaller (0.47%) for the Shape measure than for our pre-registered measure (1.46%). In addition, we found the Shape measure more interpretable, as results from different condi- tions can be compared (e.g. Fig 2B). We then ran a generalized linear mixed-effects model Shape = β0 + β1 × Distribution with a random intercept per participant. The average value of Shape was 0.18 for the Gaussian condi- tion (SD = 0.23) and −0.05 for the Uniform condition (SD = 0.21), which constitutes decisive evidence for a difference among conditions (t(19.00) = 4.15, p < .001, d = 1.01, BF10 = 61). Sur- prisingly, several participants uttered sequences that were best fit by a Gaussian distribution in both conditions (i.e., that had a positive Shape value; see Fig 2B), but the vast majority of par- ticipants had a higher Shape value when the target was Gaussian. Turns at the center In both experiments, we showed how little people changed direction compared to the iid expectation, as reflected by their low values of Turning Points. Using all items to study this PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 7 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum phenomenon, however, may not be fully indicative of people’s sampling process. This is because participants transitioned between items in smaller jumps than iid, and so domain boundaries and uneven mass may not have influenced their sequences as greatly as they do in iid sampling. To illustrate this, in Fig 3 we plot the expected proportion of turns relative to the location of the last item in a sequence, showing that making small jumps is sufficient for low Turning Points: The middle row shows an iid sampler that was modified to never make greater transitions than one standard deviation, and as a result had a Turning Points value of .52 on average (as opposed to .66 for iid). For these reasons, as a novel analysis, we focused our analysis on turns from the center of the uttered domain, as it is in this region—where no concerns about mass or boundaries exist —that the measure is most diagnostic. In central regions, standard local sampling models and iid sampling predict that the proportion of direction changes will approximately be 50%. In Experiment 1, we restricted our analyses to items where the last utterance was between the 37.5 and 62.5 percentiles (the region of the distribution where 25% of the mass lies). This represented 26.44% and 27.63% of the data for the Gaussian and uniform condition respec- tively. In Experiment 2, to increase power, we limited our analysis to the three central hexes in the one-dimensional condition (37.45% of data) but this was not possible in the two-dimen- sional condition, where we limited our analysis to the central hex only (11.81% of data). Fig 3. Theoretical and Empirical Turn Proportions. Proportion of turns relative to the location of the last-uttered item, for iid sampling (top row), a “small-transitions” iid sampler that discards items more than 1 SD away from the last item (middle row), and the observed values from participants (bottom row). While small Distances alone would lead to lower Turning Points values than iid sampling, the expected turns would be the same at the center of the distribution (i.e. around 50%). We show that people follow their current trajectory over and above what would be expected from small transitions only, going below the 50% threshold at the center of the distribution. Note: To be able to compare different participants in Experiment 1, we standardized their sequences and calculated a Z score of the heights uttered. Then we divided these scores into 13 possible bins, with midpoints at −3, −2.5, . . ., 0, . . ., 3, and counted the relative frequency of turns in each bin. We show bins -2 to 2 only to aid visualization. The shaded area represents the region we defined as the center in each condition. Simulated plots were generated from 100 sequences each. https://doi.org/10.1371/journal.pcbi.1011739.g003 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 8 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum In both experiments we found that people show low Turning Points even in this limited domain, but only in univariate domains: In Experiment 1, participants had lower Turning Points than iid (Obs. = .41, Exp. = .51, Z = −3.37, p < .001, d = −0.34, BF10 = 4), with no differ- ence between the two target distributions (Z = 2.31, p = .02, d = 0.20, BF10 = 1/3). In Experi- ment 2, participant’s Turning Points differed between conditions (Z = 3.46, p = .001, d = 0.44, BF10 = 4), with participants in the two-dimensional condition having Turning Points values similar to iid (Obs. = .50, Exp. = .50, Z = 0.21, p = .84, d = 0.01, BF10 = 1/35) but with strong evidence for smaller Turning Points values in the one-dimensional condition (Obs. = .41, Exp. = .54, Z = −3.88, p < .001, d = −0.40, BF10 = 10). Model comparison Models We compared the performance of Cooper’s schema model [36], an iid sampler, and six local sampling algorithms. We used Cooper’s model as a stand-in for all schema models, as, to our knowledge, it is the only computational implementation of a schema account, along with Sex- ton and Cooper’s [39] model from which it derives. As for the local sampling algorithms, we obtained them by choosing a simple algorithm (Metropolis-Hastings) and adding modifica- tions that may approximate qualitative features of the data, as explained below (further details about models can be found in S5 Text). This expands on the sets of sampling algorithms com- pared in [27] and [40]. iid Sampling. This model draws independent, identically distributed random samples according to the true distribution (i.e., the ideal baseline against which people are compared). Schema. Cooper’s [36] schema model, derived from Sexton and Cooper’s [39] model, generates a sequence by iteratively applying one of a set of pre-learned schemas to the previous item, with some schemas being more likely than others, and with the active schema changing over time. The schema model directly applies to Experiment 1, and for Experiment 2 we made the assumption that schemas could be quickly generated for the novel representation on which participants were trained. This is a generous assumption, as it is unlikely that schemas, which are habitual in nature, could be developed for this task in such a short period of time. However, we choose to make it in order to have a computational model other than local sampling model with which to compare participants’ data with. Local Sampling. We include six Markov Chain Monte Carlo (MCMC) algorithms in our comparison; a family of algorithms that are widely used in statistics [41], and which have pre- viously been compared to human data before [27, 30, 31, 33, 42]. They operate by creating a chain of states that are only dependent on the previous one. In each iteration, an update to the current state is proposed, with more likely states being favored (thus the chain approximating a distribution). The details of how state updates are proposed and accepted is what differentiates MCMC algorithms. We use Metropolis-Hastings [43] as the base algorithm: in each iteration, it pro- poses a new state by adding random noise, then it evaluates whether to transition to that pro- posed state based on the likelihood ratio of the proposed and current locations, with a preference for more likely locations. We create the other five local sampling algorithms by adding qualitative features to Metrop- olis-Hastings in a semi-factorial way (see Table 1): • Multiple chains: This feature involves running multiple chains that swap places stochasti- cally, with some chains transitioning to lower-likelihood states more frequently. This makes PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 9 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Table 1. Qualitative features of the local sampling algorithms compared. Qualitative Modifications (# Parameters) Multiple chains (3) Gradient-based proposals (1) Recycled momentum (1) MH ✘ ✘ ✘ MC3 ✔ ✘ ✘ HMC ✘ ✔ ✘ Samplers REC* ✘ ✔ ✔ MCHMC* MCREC* ✔ ✔ ✘ ✔ ✔ ✔ Sampler abbreviations are MH: Metropolis-Hastings, MC3: Metropolis-coupled Markov Chain Monte Carlo, HMC: Hamiltonian Monte Carlo, REC: Recycled- momentum Hamiltonian Monte Carlo. MCHMC and MCREC are the Metropolis-coupled versions of HMC and REC. Starred algorithms are those that have not been compared to human data in past work. Which parameters govern a sampler’s behavior is determined by these qualitative features (see S5 Text for details). https://doi.org/10.1371/journal.pcbi.1011739.t001 exploration more efficient in multimodal domains (as the sampler is more likely to traverse low-likelihood valleys). • Gradient-based proposals: This feature involves having samplers propose new states in a way that utilizes the gradient of the posterior distribution, by simulating a physical system using Hamiltonian dynamics, where the current state is the position and the momentum is drawn randomly in every iteration [44]. This makes the sampler more efficient in multi-dimen- sional domains. • Recycled momentum: This feature changes how the momentum is chosen in each iteration of gradient-based samplers, obtaining it by partially ‘recycling’ the previous momentum rather than drawing it randomly. This feature can only be added if proposals are gradient-based (hence the semi-factorial design), and can be used to make exploration more directed (avoiding the back and forth of random walks). We generated 105 sequences for each model, each time drawing parameters from a uniform prior. For Experiment 1, where the true distribution is unknown, we estimated the best-fitting Beta or Normal distribution for each participant; iid and local sampling models sampled from a distribution defined by the average parameters of participants’ best fitting distributions. The schema model uses the range of possible responses to sample, which we obtained by calculat- ing the average minimum and maximum value of participants’ sequences. To compare people’s sequences to the performance of these generative models, we used Approximate Bayesian Computation (ABC, [45]), a simulation-based technique that can be used to perform model comparison in cases where no likelihood function is available (see Methods). We used the above summary measures to compare these models to people’s perfor- mance, restricting Turning Points to the center of the distribution as defined above. Because Cooper’s model assumes the distribution to be uniform, the author also uses mea- sures of entropy in his analyses. To ensure that our comparison is fair, we also analyze the sequences from uniform distributions using the entropy measures used in Cooper [36], but for simplicity, and because results are qualitatively the same, we report those analyses in S7 Text. Because the schema model cannot sample in two-dimensional domains, and because this con- dition proved undiagnostic between local sampling features, we also relegate the discussion of the different models in the two-dimensional condition of Experiment 2 to S6 Text. Results Experiment 1, Uniform condition. The average best fitting distribution was Beta(1.27, 1.43), and the average range of possible values was from 122 to 219 cm (97 possible values). PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 10 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Model recovery results were good, with the iid and schema models being correctly classified 99% and 95% of the time respectively, and with local sampling models being correctly classified 86% of the time. REC was misclassified as HMC and MCREC as MCHMC 21% and 27% of the time respectively: This is to be expected somewhat, as the behavior of models with recycled momentum becomes more similar to models without it the lower the amount of recycling is. Local sampling algorithms approximated participants’ Shape best, while the schema model generated consistently more uniformly-distributed sequences. All models produced too small Adjacencies and too large Distances overall, and only the gradient-based samplers with multi- ple chains (MCHMC and MCREC) and the schema model could approximate participants’ Distances in some simulations. All samplers but the schema model produced too large values for Repetitions, and only the samplers with recycled momentum (REC and MCREC) and the schema model could consistently match people’s low Turning Points (see Fig 4 for distribu- tions of summary values). Quantitatively, local sampling algorithms performed better than the iid (BF10 = 2.0 × 10105) and schema (BF10 = 5.1 × 1030) models in this condition, and predicted 14 out of 20 partici- pants best (the schema model predicted 6 participants best). Regarding local sampling qualita- tive features, we found support for models running multiple chains (BF10 = 2.2 × 1031), having gradient-based proposals (BF10 = 1.4 × 1043), and recycling their momentum (BF10 = 8.0 × 108; see Fig 5, first column). Experiment 1, Gaussian condition. Models had a Gaussian distribution with a mean of 176.4cm and a standard deviation of 12cm as the target (irrespective of whether the participant sampled male or female heights), which were the average parameters of participants’ individual best fitting distributions. The estimated range of responses (used by the schema model) was smaller here, 83cm (from 131cm to 213cm). Model recovery results were good: the iid and schema models were correctly categorized 99.9% of the time, and local samplers were correctly categorized 78% of the time, with REC being misclassified as HMC and MCREC as MCHMC 19% and 21% of the time respectively. All models except the schema model (see Fig 4) could replicate the fact that participants reproduced the target distribution in the Gaussian condition (similar Shape values). Here, local sampling algorithms matched participants’ Repetitions, but this was more due to an increase in people’s frequencies than to a change in sampler behavior (on average, people’s Repetitions were .07 in the Gaussian condition and .02 in the Uniform condition; Z = 10.05, p < .001, d = 0.98, BF10 = 6.8 × 108). Again, only samplers with recycled momentum and the schema model matched people’s low Turning Points. Once more, local sampling algorithms replicated participants’ data better than the iid (BF10 = 2.1 × 10105) and schema (BF10 = 3.7 × 10120) models, and predicted 15 participants best (with iid and schema predicting 2 and 3 participants best respectively). As for qualitative local sampling features, we found support for running multiple chains (1.3 × 1021), using gra- dient-based proposals (3.9 × 1025) and recycling momentum (BF10 = 4.6 × 105; see Fig 5, sec- ond column). Experiment 2, One-dimensional condition. The average best fitting distribution was Beta(1, 1) (i.e., a uniform distribution), and so this was the target distribution for both the iid sampler and the local sampling algorithms. Model recovery results were poor for the iid model and the local sampling models: while the prior error rate for schema models was 27%, the iid model and local sampling models had error rates of 64% and 82%, with iid being misclassified as a local sampling model 63% of the time, and local sampling models being misclassified as another local sampling model 52% of the time. All models were able to consistently replicate participants’ Shape values, but their values for Adjacencies were too low and their values for Distances too large. Regarding Turning PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 11 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Fig 4. Model Summaries. Distribution of values for each of the computed measures for the eight candidate models, in three tasks, with parameter values drawn from the prior. Further details are provided in the main text. Notice that in the Gaussian condition of Experiment 1 the iid sampler had fewer repetitions than participants, yet in the main text we reported that people had fewer repetitions than expected. This is because there we compared their performance to reshuffled sequences, not to iid sampling from the distribution, and people have fewer unique items than iid sampling. https://doi.org/10.1371/journal.pcbi.1011739.g004 Points, only the local sampling algorithms that recycled their momentum (REC and MCREC) and the schema model replicated the low number of Turning Points at the central hex that people produced. Finally, only the schema model had as low Repetitions as partici- pants (see Fig 4). PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 12 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Fig 5. Posteriors per participant. Posteriors per participant in three tasks, with each column representing one participant. While in the one- dimensional condition the schema model performed best, local sampling algorithms replicated participants’ data better in the Uniform and Gaussian conditions. Within local sampling features, we found decisive evidence for samplers running multiple chains, using gradient-based proposals, and recycling their momentum. Note: Conditions in Experiment 1 varied within participants, and so the nth bar in the Uniform condition is the same participant as the nth bar in the Gaussian condition. https://doi.org/10.1371/journal.pcbi.1011739.g005 For these reasons, the schema model was closest to people’s performance in this condition, predicting 19 out of 20 participants best (with iid predicting one participant best), and with a BF10 = 9.2 × 1089 over the local sampling class of models and BF10 = 3.6 × 10136 over the iid sampler. However, this is a generous interpretation of the schema model, assuming that people are able to very quickly learn and apply schemas to novel representations. Despite the schema model outperforming local sampling models in this condition, we still compared the qualitative features of local sampling algorithms: we found evidence against models running multiple chains (BF10 = 4.0 × 10−5), evidence for gradient-based proposals (BF10 = 2.7 × 103), and evidence for recycled momentum (BF10 = 4.8 × 1016; see Fig 5, third column), although the model recovery results for this condition reveal that local sampling models were not particularly distinctive among themselves. Combined Posterior Probabilities. Finally, we combined the obtained posterior proba- bilities for the three conditions by multiplying them together, in order to carry out joint com- parisons. When aggregating over conditions, we found decisive evidence for local sampling algorithms over the schema model (BF10 = 2.9 × 1058). We also found decisive evidence for local samplers running multiple chains (BF10 = 1.3 × 1048), using gradient-based proposals (BF10 = 1.4 × 1085), and recycling their momentum (BF10 = 7.2 × 1026). Comparing the individual models across all three conditions, we found that the best fitting model was MCREC, with a BF10 = 7.2 × 1026 over the next best model, MCHMC, and a BF10 = 1.8 × 1059 over the schema model (see Table 2). Discussion In the current study, we expanded the random generation paradigm to ask participants to gen- erate sequences from non-uniformly distributed domains and from recently-learned displays. We found that participants displayed the same systematic deviations from iid sampling as PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 13 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Table 2. Model Bayes Factors. Model MCREC MCHMC REC Schema HMC MC3 MH iid E1: Unif Condition 1.2 × 10106 1.5 × 1097 5.4 × 1074 4.0 × 1074 1.2 × 1059 1.1 × 1054 7.5 × 10−15 1 E1: Gauss Condition 1.2 × 10106 2.7 × 10100 8.6 × 1084 5.6 × 10−16 9.2 × 1083 6.9 × 1074 6.1 × 1021 1 E2: 1D Condition 9.4 × 1042 4.9 × 1030 2.4 × 1047 3.6 × 10136 3.9 × 1025 1.9 × 1014 1.8 × 1027 1 Combined Posteriors 1.4 × 10255 2.0 × 10228 1.1 × 10207 8.0 × 10195 4.2 × 10168 1.4 × 10143 8.2 × 1034 1 Bayes Factors for the eight candidate models, in each separate task and considering the joint posterior probabilities. Models have been arranged in order of Bayes Factor over iid when combining the three tasks, from largest to smallest. Metropolis-coupled Recycled-momentum Hamiltonian Monte Carlo (MCREC) is the best-fitting model, but other local sampling algorithms are also competitive. https://doi.org/10.1371/journal.pcbi.1011739.t002 found in previous random generation experiments, pointing to a common mechanism under- lying their performance across tasks. We also showed that participants could flexibly change the distribution from which the generated samples, being able to generate the same items— heights, in our case—in a uniform or Gaussian fashion depending on the given instructions. Finally, we identified a key qualitative feature of people’s random generation, people’s ten- dency to maintain their trajectory for many samples, and showed that this pattern does not arise only due to the fact that people make small transitions. These findings directly contradict schema accounts of human random generation [16, 36], which predict that people generate random items by striving for serial independence only, and that the systematic deviations from random sampling that people display arise from the presence of habitual responses (sche- mas) and the attempt to suppress them. The schema account could be relaxed to allow for quick learning of schemas (as we assumed when modeling the one-dimensional condition of Experiment 2), or more sophisticated and task-specific schemas could be postulated for the domains examined here. However, these hypothetical schema accounts would still not repro- duce target distributions, and this account would need substantial modifications in order to do so: for example, additional monitoring processes on the response histogram, not its unpredict- ability, would need to be included, as well as a policy on how to trade off deviations from unpredictability and from the optimal histogram. Although no computational model was available for the network modulation account of random number generation [23], the results here presented can qualitatively be compared to what this account would predict. The network modulation account postulates that people gen- erate random items by creating an associative network with each possible response as a node, and with links representing their strength of association. While this network alone produces habitual, stereotyped responses, a controller inhibits the strongest links to allow for more unpredictable behavior. If it were modified to have bidirectional links between items, with the strength of each direction being proportional to the density ratio between the receiving node and the origin, this account would be able to reproduce the fact that people can generate random items in a Gaussian fashion. Reproducing the finding that people make fewer turns than expected would require a very specific kind of interaction between the associative network and the controller: because the associative network cannot produce fewer turns (as transitions only depend on the current state and not the previous state), the controller would have to modify the network’s links frequently and in a very distinct way in order to achieve this result. In addition, this PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 14 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum account would also not be able to explain the fact that the same deviations from iid sampling appear when the items have recently been learned, as no habitual responses need to be inhib- ited in this case. Instead, data are consistent with the alternative we proposed: that people are not using a surrogate process to sample items, but instead sampling from the domain directly using their general cognitive ability to produce samples to perform inferences. We also quantitatively modeled people’s behavior to the performance of a schema model [36], an iid sampler, and six MCMC algorithms that varied in three qualitative features. In one dataset—the one-dimen- sional condition of Experiment 2—the schema model performed best. This was due to it being able to reproduce the low level of repetitions participants displayed in such a small range of possible items, as such a range is closest to the tasks it was designed for. The fact that we assumed that schemas could be applied to newly-learned items, however, should be noted when interpreting these results. In the other three datasets, as well as when combining posteri- ors across datasets, we found that local sampling algorithms were best at replicating human performance, thus linking people’s behavior in random generation tasks to their performance in other domains. We identified several features, such as recycling momentum, that had not been previously compared to human data, showing that these features allowed local sampling algorithms to better fit the data than the sampling algorithms that have performed best in past comparisons (i.e., MC3 [27, 40]). Recent research using process-tracing techniques suggests that how samples come to mind and what the task at hand is are largely independent. For example, Mills and Phillips [46] ask their participants to generate a list of animals as they come to mind, and find that doing so with no other purpose, or in order to answer a specific question, does not change the types of items participants produce. Similarly, Hardisty, Johnson and Weber [47] find that reporting ideas that come to mind while making a decision makes no qualitative difference on the result- ing choice, compared to a condition without thought listing. If how samples are accrued is task-independent, then having identified these local sampling features in the random generation task also has implications for how people engage in the other tasks where local sampling algorithms have been applied: These sampling approaches have been used to explain how people come up with ideas in a semantic fluency task [33], how they estimate temporal duration [27], how they perceive visual stimuli with multiple interpretations [8] and why they present multiple biases in how they reason with probabilities [28, 30, 31]. Although more and more research is being done comparing human performance to local sampling algorithms, the set of available algorithms in the computer science literature is incredibly vast, and efforts to identify the features of the human mental algorithm are in their infancy. An exciting conclusion of the current work is that random generation tasks can dis- tinguish fine-grained differences between algorithms, an endeavor that can be expanded on in future work. Limitations and Future Directions Despite their success, local samplers displayed too high Repetitions in most conditions. Future research may investigate additional qualitative features that can improve on this fact: for exam- ple, ‘unadjusted’ algorithms [48], which do not evaluate the relative goodness of the proposed and last state before transitioning, would repeat less by not rejecting proposals. Alternatively, a post-sampling mechanism that explicitly eliminates some of the repetitions could be imple- mented, akin to participants choosing not to utter their sampled item if it is identical to the last; or sequences could be thinned to only report the nth sample, following research suggest- ing that people use few, but more than one, samples when making probability judgments [6]. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 15 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum An exciting feature of the current data is that people were able to replicate the target distri- bution in Experiment 1, despite never being told participants what the true distribution of heights in the United Kingdom is. Another avenue of future research, therefore, is to use ran- dom generation as a belief elicitation method. In many domains, knowing what experts believe is essential to build models of possible future outcomes [49], and in related work, we have shown that random generation can be used to elicit beliefs as a complement to other more established techniques [50]. This was true for additional distributions to the ones shown here: distributions with high skewness (gross earnings from films) or where some values are extremely unlikely (American football scores). Another avenue of future research will be to apply local sampling models to previous ran- dom generation results: most notably, random generation research on neurodivergent popula- tions might benefit from a sampling interpretation. A vast literature has shown that different neurodiverse populations show differences with neurotypical controls for some measures of randomness but not others. For example, patients with schizophrenia and Parkinson’s disease will display even higher rates of adjacent items than neurotypical controls but the same bias against repeating items [51, 52], while patients with multiple sclerosis will make even fewer turns than healthy controls [14]. Conversely, patients with unilateral frontal lobe lesions may show a lesser bias against repeating items than neurotypical adults [53]. These differences can be framed as differences in the qualitative features of the sampling model or as differences in the parameter values used. For example, a local sampling model without multiple chains will have more adjacent items, and increasing the degree of recycled momentum will lead to fewer turns. The models here presented might also need to be expanded to account for the many previ- ous findings on how neurotypical adults perform the random generation task. For example, many studies have manipulated how participants input their responses: rather than say items out loud participants may press keys on a keyboard [54], select items with a mouse [18], or fill squares in a grid [55], with differences in how random resulting sequences are. Analogously, changing how the task of being random is described to participants may influence their sequences (for example, asking them to simulate a coin toss mentally will lead to more random sequences than explicitly asking for a random choice between heads and tails [56]). Partici- pants are also more random when participating in a zero-sum game like matching pennies or rock paper scissors [20, 57]. While it is possible that changes in the input mechanism may influence the mental representation participants have of the task (in the same way that our spa- tial training of syllables did), and that different instructions or a competitive setting may result in differences in effort while doing the task, future work will be needed to incorporate these findings into local sampling models. It is possible, however, that some other findings in the random generation literature not here explored would be already replicable by the local sampling models we use in this study: previous research has shown that people are less random and say more adjacent items when production rates are high [16, 21, 58], which could be achieved by a local sampling model that makes smaller proposals between utterances. Similarly, if people have a second task to perform while they’re generating a random sequence, they display even fewer repetitions and fewer turning points [18], which a model with an increased degree of recycled momentum would reproduce. Conclusion We devised novel tasks to study volitional random generation and found that people can gen- erate random items from a wider range of domains than previously studied, while still PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 16 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum displaying the characteristic deviations from iid sampling observed in previous tasks. We showed that local sampling algorithms replicate people’s data more successfully, linking peo- ple’s randomness in the random generation paradigm to behavioral noise in other tasks. We were also able to identify several qualitative features that mirror people’s sequences, showing that random generation can be a useful paradigm to reverse-engineer people’s sampling algorithm. Methods Ethics statement For both experiments, ethical approval was given by the Humanities and Social Sciences Research Ethics Committee (HSSREC) at the University of Warwick. Written informed con- sent was obtained from all participants. Experiment 1 In this experiment (preregistration at https://osf.io/ux5tp), we asked participants to produce a random sequence of people’s heights, either from a distribution in which all possible heights are equally distributed, or from the actual distribution of heights in the United Kingdom (which is roughly Gaussian for both adult men and adult women [37]). All participants gener- ated heights from both distributions and the order of the distributions was counterbalanced. Participants. Participants were recruited from the University of Warwick participant pool. Being fluent in English was the only inclusion criterion. To ensure that participants accessed the task with an appropriate microphone and stable internet connection, a pre- screening task was run in which candidates recorded themselves reading a short text. Partici- pants were paid £0.50 for completing the pre-screen task, irrespective of the outcome. The first two authors independently rated whether the recording was audible, and the participants were invited to perform the main task if at least one rater had deemed their recording to be valid. Raters reached high agreement (87.2%, Cohen’s κ = .63). 85% of the participants who partici- pated in the pre-screen were invited to the main experiment. We collected data from 21 partic- ipants (Mean Age = 23.8, SD = 4.71; 14 male, 7 female). Following pre-registered criteria we calculated a measure of how predictable sequences were (sequence determinism [59], see S9 Text]), and data from one participant (5%) was excluded from analysis (91% of their items were one inch taller than the previous). Participants received payment of £2.5 plus a bonus of up to £1.35 depending on performance, which we measured by how well they kept to the given pace (mean total payment = £3.71). The experiment lasted approximately 20 minutes. Design and Procedure. The experiment was conducted via video call. First, we intro- duced participants to the task, and then they had to produce random heights from that distri- bution for five minutes. In one condition, heights were introduced as distributed according to a uniform distribution, whereas in the other condition, heights followed the true distribution for adults in the UK for the target gender. Following [37], we consider the true distribution of UK heights to be Gaussian for each gender, with a mean of 176.4cm and standard deviation of 7.02 for men and a mean of 163.6cm and a standard deviation of 6.03 for women (we did not disclose this information to participants). We will refer to these conditions as Uniform and Gaussian, respectively. Participants produced heights at random in two blocks, one for each condition, in counter- balanced order. The target gender was fixed for each participant throughout the experiment. Participants could express heights in feet and inches or meters and centimeters, and we ana- lyzed all randomness measures with the units they had used (For simplicity, we use centimeters only throughout the current text). PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 17 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum At the beginning of the experiment, participants were asked what they believed the height of the shortest and tallest adult was in the UK (for the gender in their condition). Some partici- pants spontaneously asked whether they should consider people with restricted growth, which they were told not to. Participants estimated the minimum and maximum values reasonably well, below and above the 1st and 99th percentiles of the true distribution: their median mini- mum was 136 cm (SD = 17) for men and 125 cm for women (SD = 23.6), and their median maximum was 217 cm for men (SD = 20.3) and 200 cm for women (SD = 35.9). After these preliminary questions, participants were told a cover story matching the experi- mental condition for that block. In the Gaussian condition, participants were asked to imagine that a photographer wanted to take a picture representing the heights of adults (of the gender they had been allocated), taking a picture of 10,000 people so that each possible height appeared as often as it does in the population (as in the ‘living histograms’ of [60]). They were told to imagine that each person who had been in the picture wrote their height on a piece of paper, and that that paper was put in a bag. In the Uniform condition, they were told to imag- ine that each possible height within the height boundaries they had previously specified had been written on a piece of paper and that all papers were put in a bag. After the learning stage, participants were asked to produce random heights, as if they were drawing a random paper from the bag that had been described, saying the height out loud, put- ting the paper back in the bag, and reshuffling the papers. They repeated this process for five minutes. While saying items out loud, they were asked to look at the screen, where a dot flashed at 30 times per minute, and were instructed to say a height every time the dot appeared. The pace of production was chosen to be slower than in Experiment 1 because pilot testing revealed that participants required more time to utter the multi-syllable heights. After five minutes, participants were allowed a short rest, and then the Learning and Pro- duction stages were repeated for the other condition. Participants produced both tasks at a similar pace—the median temporal gap between successive items was 2.09s (SD = .27) and 2.02s (SD = .24) for the first and second sequence participants produced, a difference that was significant but ambiguous (F(1, 19) = 5.55, p = .03, d = −0.53, BF10 = 1). Experiment 2 Preregistered (https://osf.io/q3yrj) analyses for this experiment were reported in [61]. Analyses here followed the analysis plan for Experiment 1. In this experiment, participants first learned a display of syllables, arranged in either a one- dimensional row or a two-dimensional configuration (see Fig 1), by moving virtually through the display, revealing the syllable at their current location. Then, once they could reproduce these spatial arrangements from memory, they were asked to utter a random sequence of sylla- bles from that set. Participants. We recruited 42 participants (Mean Age = 24.95, SD = 9.67; 14 male, 27 female, 1 non-binary) from the University of Warwick participant pool. The only inclusion criterion was that participants had English as their first language. This was a stricter language requirement than in Experiment 1, as we wanted to ensure that the syllables were meaningless to participants. Two participants (5%) did not learn the syllables in the allocated time, and were excluded from analysis following our preregistered criteria. Participants received a pay- ment of £3.5 plus a bonus of up to £1.8, which depended on their performance in learning the syllables (the average total payment was £4.64). The experiment lasted approximately 30 minutes. Materials. We chose seven syllables of two letters each, all ending in a to ensure consistent ease when uttering consecutive items. To select them, we considered both the frequencies of PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 18 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum the syllables and syllable pairs in the Brown corpus [62], aiming for a homogeneous set. The resulting selection was: ca, ha, la, ma, na, pa, and ta. Design and Procedure. The experiment was conducted via video call. Participants first learned the display they had been allocated (Learning Stage), then they uttered syllables at ran- dom (Production Stage) for five minutes. The two blocks, learning the display and producing a random sequence, were repeated after a short break (with the same display in both blocks). The key experimental manipulation, which varied between participants, was whether the dis- play of syllables they learned was the one- or two-dimensional arrangement. How the syllables were arranged within the one- or two-dimensional display followed one of five possible config- urations and was chosen randomly for each participant. In the learning stage, participants were presented with a display consisting of seven hexa- gons, arranged in either a single row or a two-dimensional grid, depending on the experimen- tal condition. The hexagons were oriented so that the vertex was on top, and the two- dimensional grid consisted of three rows of two, three, and two hexagons, respectively. Each hexagon contained a hidden syllable, and participants’ task was to view and learn which sylla- ble each hexagon displayed. To do so, they selected a hexagon whose syllable they wanted to reveal, which made the previous syllable disappear, and the syllable in the chosen hexagon appeared. They could freely choose any hexagon as their starting one, but subsequent choices were constrained to adjacent hexagons only, which made the learning process akin to ‘spatially exploring’ the display. To promote active learning, we included a delay of one second between the disappearance of a syllable and the appearance of the next, and instructed participants to announce which syllable they expected in the hexagon they had selected before it appeared. As soon as participants felt confident that they had learned the display, their knowledge was tested by asking them to name the syllables displayed on the seven hexagons in random order. If participants answered all seven queries correctly in two consecutive tests, they proceeded to the production stage, or else they returned to learn the display. Participants were excluded, and the experiment was terminated if they failed the test four times, or if they exceeded the maximum learning time of 10 minutes. Participants spent an average of 5.7 minutes (SD = 2.4) learning the syllables, and no participant spent more than ten minutes. Two participants failed the test four times and were excluded (on average, participants failed 0.65 times, SD = .86. The two excluded participants’ average learning time was 9m and 54s). In the production stage, participants uttered syllables from the set they had learned at ran- dom for five minutes. To instruct participants to produce random sequences, we asked them to imagine that they were drawing the syllables out of a hat each time, and putting the syllable back before shuffling and drawing the next, following standard practice in previous random generation experiments [16, 17]. During this stage, participants did not see the display they had learned, but instead saw a dot flashing on screen, appearing at a pace of 80 times per min- ute (once every 750ms). Participants produced a slightly lower pace than targeted (M = 71.21 syllables per minute, SD = 14.32). After completing this stage, participants were allowed a short break. Then, both learning and production stages were repeated, using the same display of syllables in the same arrange- ment: participants had the opportunity to revise the display, and after testing they uttered sylla- bles for another five minutes. The average time spent revising and testing was 104s (SD = 47s), and the average number of failed attempts was 0.18 (SD = .38). Both were much lower than in the initial learning stage (t(39) = −11.61, p < .001, BF10 = 1.2 × 107 and t(39) = −3.48, p = .001, BF10 = 9, respectively), which suggests that participants had no difficulty remembering the dis- play after their first random generation block. Participants were slower at producing items in the first block they produced, with the median temporal gap between the items they uttered being larger for the first block participants produced (median = 843ms, SD = 146) than for the PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 19 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum second (median = 799ms, SD = 142). The evidence for this difference was ambiguous, with a significant p-value but anecdotal Bayes factor (F(1, 39) = 11.97, p = .001, d = −.33, BF10 = 2.46). All participants named each syllable at least once in each sequence. Model comparison Method. To compare people’s sequences to the performance of the several generative models, we used Approximate Bayesian Computation (ABC [45]), a simulation-based tech- nique that can be used to perform model comparison in cases where no likelihood function is available. In our case (many approaches to ABC are available), we first generated vast volumes of artificial data from the candidate models, sampling random values for their parameters from their prior distributions; then obtained summary measures for the observed and synthetic data. Finally, we used a machine learning tool for data classification, random forests, to compute the posterior belief on each of the candidate models [63]. In short, a set of classification decision trees is trained on the simulated data and learns to categorize it into the different candidate models it originated from. Then, it classifies the observed data into which candidate model is more likely to have produced it. To obtain the probability that the classification model makes an error, a separate regression forest is trained on the same training data and the classification error during training, and is applied to the observed data. This regression forest only computes the probability that the classification label provided was incorrect, and so to obtain a full poste- rior distribution over all candidate models, we ran forests recursively for each uttered sequence, each time removing from the candidate models the model that had been considered best in the previous iteration, until a posterior for each model was obtained. Unlike more tradi- tional ABC approaches, the random-forest approach requires fewer simulations for each candi- date model, and is more robust to the choice of summary statistics (for completeness, however, we carry out a more typical ABC analysis in S8 Text, with similar results). Here, we fit canonical distributions to participants’ data in each experiment, which would be the target distributions the models would sample from. We then simulated 105 sequences of 400 items each for each candidate model and each condition, choosing the prior over the parameters to be as uninformative as possible (we describe model parameters and their associ- ated priors in S5 Text). Because the iid sampler and the local sampling algorithms sample in continuous space, but participants produced whole numbers, we rounded the values the sam- plers produced. To evaluate the resulting data, we used the summary statistics described in the main text, choosing to focus on Turning Points from the central region of the distribution. Because the schema model can only sample from univariate distributions, we do not consider the two-dimensional condition of Experiment 2 in the main text, but we did fit local sampling models there too, and we describe those results in the S6 Text (this subset of the data, however, yielded uninformative results, with all local sampling algorithms and the iid sampler perform- ing equally well). In order to compare the three higher-level approaches to random generation (iid sampling, local sampling, and schema accounts), we carried out Bayesian Model Averaging [64] to com- pute Bayes factors (i.e., the ratio of the average posteriors of each candidate class). We also used these to compare qualitative features of the candidate local sampling models, comparing models that were ‘matched’: we only included models to our comparisons that had an identical equivalent in the other side of the comparison but for the factor of interest (e.g., to compute Bayes factors of inclusion for gradient-based proposals, the average of the posteriors for HMC and MCHMC was compared to that of MH and MC3. Because no model exists that has recy- cled momentum but not gradient-based proposals, REC was not added to either side of the comparison; see Table 1). PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 20 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum Supporting information S1 Text. Additional Exploratory Plots. (PDF) S2 Text. Condition Effects. (PDF) S3 Text. Order Effects. (PDF) S4 Text. Other pre-registered analyses. (PDF) S5 Text. Model Details. (PDF) S6 Text. Model comparison in the two-dimensional condition of Experiment 2. (PDF) S7 Text. Entropy Measures. (PDF) S8 Text. Standard Approximate Bayesian Computation. (PDF) S9 Text. Exclusion Criteria. (PDF) Author Contributions Conceptualization: Lucas Castillo, Pablo Leo´n-Villagra´, Nick Chater, Adam Sanborn. Data curation: Lucas Castillo. Formal analysis: Lucas Castillo. Funding acquisition: Adam Sanborn. Investigation: Lucas Castillo. Methodology: Lucas Castillo, Pablo Leo´n-Villagra´, Nick Chater, Adam Sanborn. Project administration: Lucas Castillo, Adam Sanborn. Software: Lucas Castillo. Supervision: Adam Sanborn. Visualization: Lucas Castillo, Pablo Leo´n-Villagra´. Writing – original draft: Lucas Castillo, Pablo Leo´n-Villagra´, Adam Sanborn. Writing – review & editing: Lucas Castillo, Pablo Leo´n-Villagra´, Nick Chater, Adam Sanborn. References 1. Mosteller F, Nogee P. An Experimental Measurement of Utility. The Journal of Political Economy. 1951; 59(5):371–404. https://doi.org/10.1086/257106 2. Neimark ED, Shuford EH. Comparison of Predictions and Estimates in a Probability Learning Situation. Journal of Experimental Psychology. 1959; 57(5):294–298. https://doi.org/10.1037/h0043064 PMID: 13654638 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 21 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum 3. Vulkan N. An Economist’s Perspective on Probability Matching. Journal of Economic Surveys. 2000; 14 (1):101–118. https://doi.org/10.1111/1467-6419.00106 4. Nosofsky RM. Attention, Similarity, and the Identification-Categorization Relationship. Journal of Exper- imental Psychology: General. 1986; 115(1):39–57. https://doi.org/10.1037/0096-3445.115.1.39 PMID: 2937873 5. Costello F, Watts P. Surprisingly Rational: Probability Theory plus Noise Explains Biases in Judgment. Psychological Review. 2014; 121(3):463–480. https://doi.org/10.1037/a0037010 PMID: 25090427 6. Sundh J, Zhu JQ, Chater N, Sanborn A. A Unified Explanation of Variability and Bias in Human Probabil- ity Judgments: How Computational Noise Explains the Mean–Variance Signature. Journal of Experi- mental Psychology: General. 2023; 152(10):2842–2860. https://doi.org/10.1037/xge0001414 PMID: 37199970 7. Levy RP, Reali F, Griffiths TL. Modeling the Effects of Memory on Human Online Sentence Processing with Particle Filters. In: Advances in Neural Information Processing Systems; 2009. p. 937–944. 8. Gershman SJ, Vul E, Tenenbaum JB. Multistability and Perceptual Inference. Neural Computation. 2012; 24(1):1–24. https://doi.org/10.1162/NECO_a_00226 PMID: 22023198 9. Ratcliff R. A Theory of Memory Retrieval. Psychological review. 1978; 85(2):59. https://doi.org/10.1037/ 0033-295X.85.2.59 10. Misirlisoy E, Haggard P. Asymmetric Predictability and Cognitive Competition in Football Penalty Shoot- outs. Current Biology. 2014; 24(16):1918–1922. https://doi.org/10.1016/j.cub.2014.07.013 PMID: 25088554 11. Brockbank E, Vul E. Humans Fail to Outwit Adaptive Rock, Paper, Scissors Opponents. In: Proceed- ings of the Annual Meeting of the Cognitive Science Society. vol. 43; 2021. p. 1740–1746. 12. Ebert JP, Wegner DM. Mistaking Randomness for Free Will. Consciousness and Cognition. 2011; 20 (3):965–971. https://doi.org/10.1016/j.concog.2010.12.012 PMID: 21367624 13. Nichols S, Knobe J. Moral Responsibility and Determinism: The Cognitive Science of Folk Intuitions. Nouˆs (Detroit, Mich). 2007; 41(4):663–685. doi: 10.1111/j.1468-0068.2007.00666.x 14. Geisseler O, Pflugshaupt T, Buchmann A, Bezzola L, Reuter K, Schuknecht B, et al. Random Number Generation Deficits in Patients with Multiple Sclerosis: Characteristics and Neural Correlates. Cortex; a journal devoted to the study of the nervous system and behavior. 2016; 82:237–243. https://doi.org/10. 1016/j.cortex.2016.05.007 PMID: 27403852 15. Hornero R, Abasolo D, Jimeno N, Sanchez CI, Poza J, Aboy M. Variability, Regularity, and Complexity of Time Series Generated by Schizophrenic Patients and Control Subjects. IEEE Transactions on Bio- medical Engineering. 2006; 53(2):210–218. https://doi.org/10.1109/TBME.2005.862547 PMID: 16485749 16. Baddeley AD. The Capacity for Generating Information by Randomization. Quarterly Journal of Experi- mental Psychology. 1966; 18(2):119–129. https://doi.org/10.1080/14640746608400019 PMID: 5935121 17. Towse JN. On Random Generation and the Central Executive of Working Memory. British Journal of Psychology. 1998; 89(1):77–101. https://doi.org/10.1111/j.2044-8295.1998.tb02674.x PMID: 9532724 18. Cooper RP, Karolina W, Davelaar EJ. Differential Contributions of Set-Shifting and Monitoring to Dual- Task Interference. Quarterly Journal of Experimental Psychology. 2012; 65(3):587–612. https://doi.org/ 10.1080/17470218.2011.629053 PMID: 22182315 19. Towse JN, Towse AS, Saito S, Maehara Y, Miyake A. Joint Cognition: Thought Contagion and the Con- sequences of Cooperation When Sharing the Task of Random Sequence Generation. PLOS ONE. 2016; 11(3):e0151306. https://doi.org/10.1371/journal.pone.0151306 PMID: 26977923 20. Wong A, Merholz G, Maoz U. Characterizing Human Random-Sequence Generation in Competitive and Non-Competitive Environments Using Lempel–Ziv Complexity. Scientific Reports. 2021; 11 (1):20662. https://doi.org/10.1038/s41598-021-99967-6 PMID: 34667239 21. 22. 23. Towse JN, Neil D. Analyzing Human Random Generation Behavior: A Review of Methods Used and a Computer Program for Describing Performance. Behavior Research Methods, Instruments, & Comput- ers. 1998; 30(4):583–591. https://doi.org/10.3758/BF03209475 Towse JN, Valentine JD. Random Generation of Numbers: A Search for Underlying Processes. Euro- pean Journal of Cognitive Psychology. 1997; 9(4):381–400. https://doi.org/10.1080/713752566 Jahanshahi M, Profice P, Brown RG,Ridding, Dirnberger G, Rothwell JC. The Effects of Transcranial Magnetic Stimulation over the Dorsolateral Prefrontal Cortex on Suppression of Habitual Counting dur- ing Random Number Generation. Brain : a journal of neurology. 1998; 121(8):1533–1544. https://doi. org/10.1093/brain/121.8.1533 PMID: 9712014 24. Baddeley AD. Exploring the Central Executive. The Quarterly Journal of Experimental Psychology. 1996; 49(1):5–28. https://doi.org/10.1080/713755608 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 22 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum 25. Sanborn AN, Zhu JQ, Spicer J, Leon-Villagra P, Castillo L, Falben J, et al. Noise in Cognition: Bug or Feature? PsyArXiv; 2022. 26. Gilden DL, Thornton T, Mallon MW. 1/f Noise in Human Cognition. Science (New York, NY). 1995; 267 (5205):1837–1839. https://doi.org/10.1126/science.7892611 PMID: 7892611 27. Zhu JQ, Leo´n-Villagra´ P, Chater N, Sanborn AN. Understanding the Structure of Cognitive Noise. PLoS Computational Biology. 2022; 18(8):e1010312. https://doi.org/10.1371/journal.pcbi.1010312 PMID: 35976980 28. Chater N, Zhu JQ, Spicer J, Sundh J, Leo´ n-Villagra´ P, Sanborn AN. Probabilistic Biases Meet the Bayesian Brain. Current Directions in Psychological Science. 2020; 29(5):506–512. https://doi.org/10. 1177/0963721420954801 29. MacKay DJC. Information Theory, Inference, and Learning Algorithms. Cambridge, UK; New York: Cambridge University Press; 2003. 30. Dasgupta I, Schulz E, Gershman SJ. Where Do Hypotheses Come From? Cognitive Psychology. 2017; 96:1–25. https://doi.org/10.1016/j.cogpsych.2017.05.001 PMID: 28586634 31. Lieder F, Griffiths TL, M Huys QJ, Goodman ND. The Anchoring Bias Reflects Rational Use of Cognitive Resources. Psychonomic Bulletin & Review. 2018; 25(1):322–349. https://doi.org/10.3758/s13423- 017-1286-8 PMID: 28484952 32. Bramley NR, Dayan P, Griffiths TL, Lagnado DA. Formalizing Neurath’s Ship: Approximate Algorithms for Online Causal Learning. Psychological Review. 2017; 124(3):301–338. https://doi.org/10.1037/ rev0000061 PMID: 28240922 33. Zhu JQ, Sanborn AN, Chater N. Mental Sampling in Multimodal Representations. Advances in Neural Information Processing Systems. 2018; 31:5748–5759. 34. Griffiths TL, Vul E, Sanborn AN. Bridging Levels of Analysis for Probabilistic Models of Cognition. Cur- rent Directions in Psychological Science. 2012; 21(4):263–268. https://doi.org/10.1177/ 0963721412447619 35. Wagenaar WA. Generation of Random Sequences by Human Subjects: A Critical Survey of Literature. Psychological Bulletin. 1972; 77(1):65. https://doi.org/10.1037/h0032060 36. Cooper RP. Executive Functions and the Generation of “Random” Sequential Responses: A Computa- tional Account. Journal of Mathematical Psychology. 2016; 73:153–168. https://doi.org/10.1016/j.jmp. 2016.06.002 37. Freeman JV, Cole TJ, Chinn S, Jones PR, White EM, Preece MA. Cross Sectional Stature and Weight Reference Curves for the UK, 1990. Archives of Disease in Childhood. 1995; 73(1):17–24. https://doi. org/10.1136/adc.73.1.17 PMID: 7639543 38. Nickerson RS. The Production and Perception of Randomness. Psychological Review. 2002; 109 (2):330–357. https://doi.org/10.1037/0033-295X.109.2.330 PMID: 11990321 39. Sexton NJ, Cooper RP. An Architecturally Constrained Model of Random Number Generation and Its Application to Modeling the Effect of Generation Rate. Frontiers in Psychology. 2014; 5. https://doi.org/ 10.3389/fpsyg.2014.00670 PMID: 25071644 40. Spicer J, Zhu JQ, Chater N, Sanborn AN. How Do People Predict a Random Walk? Lessons for Models of Human Cognition. PsyArXiv; 2022. 41. Brooks S, Andrew Gelman, Jones G, Meng XL, editors. Handbook for Markov Chain Monte Carlo. Boca Raton: Taylor & Francis; 2011. 42. Sanborn AN, Griffiths TL, Navarro DJ. Rational Approximations to Rational Models: Alternative Algo- rithms for Category Learning. Psychological Review. 2010; 117(4):1144–1167. https://doi.org/10.1037/ a0020511 PMID: 21038975 43. Hastings WK. Monte Carlo Sampling Methods Using Markov Chains and Their Applications. Biome- trika. 1970; 57(1):97–109. https://doi.org/10.1093/biomet/57.1.97 44. Neal RM. MCMC Using Hamiltonian Dynamics. In: Brooks S, Gelman A, Jones G, Meng XL, editors. Handbook of Markov Chain Monte Carlo. 1st ed. Chapman and Hall/CRC; 2011. p. 113–162. 45. Turner BM, Van Zandt T. Approximating Bayesian Inference through Model Simulation. Trends in Cognitive Sciences. 2018; 22(9):826–840. https://doi.org/10.1016/j.tics.2018.06.003 PMID: 30093313 46. Mills T, Phillips J. Locating What Comes to Mind in Empirically Derived Representational Spaces. Cog- nition. 2023; 240:105549. https://doi.org/10.1016/j.cognition.2023.105549 PMID: 37647741 47. Hardisty DJ, Johnson EJ, Weber EU. A Dirty Word or a Dirty World?: Attribute Framing, Political Affilia- tion, and Query Theory. Psychological Science. 2010; 21(1):86–92. https://doi.org/10.1177/ 0956797609355572 PMID: 20424028 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 23 / 24 PLOS COMPUTATIONAL BIOLOGYExplaining the flaws in human random generation as local sampling with momentum 48. Roberts GO, Tweedie RL. Exponential Convergence of Langevin Distributions and Their Discrete Approximations. Bernoulli Official Journal of the Bernoulli Society for Mathematical Statistics and Prob- ability. 1996; 2(4):341. https://doi.org/10.2307/3318418 49. Mikkola P, Martin OA, Chandramouli S, Hartmann M, Pla OA, Thomas O, et al. Prior Knowledge Elicita- tion: The Past, Present, and Future; 2021. 50. Leo´ n-Villagra´ P, Castillo L, Chater N, Sanborn AN. Eliciting Human Beliefs Using Random Generation. In: Proceedings of the Annual Meeting of the Cognitive Science Society. vol. 44; 2022. 51. Salame´ P, Danion JM. Inhibition of Inappropriate Responses Is Preserved in the Think-No-Think and Impaired in the Random Number Generation Tasks in Schizophrenia. Journal of the International Neuropsychological Society. 2007; 13(02). https://doi.org/10.1017/S1355617707070300 PMID: 17286885 52. Spatt J, Goldenberg G. Components of Random Generation by Normal Subjects and Patients with Dys- executive Syndrome. Brain and Cognition. 1993; 23(2):231–242. https://doi.org/10.1006/brcg.1993. 1057 PMID: 8292327 53. Brugger P, Landis T, Regard M. The Brain as a Random Generator: The Relevance of Subjective Ran- domization for Neuropsychology. Journal of Clinical and Experimental Neuropsychology. 1992; 14 (1):84. https://doi.org/10.1080/01688639208403061 54. Baddeley AD, Emslie H, Kolodny J, Duncan J. Random Generation and the Executive Control of Work- ing Memory. 1998; p. 36. 55. Gauvrit N, Zenil H, Soler-Toscano F, Delahaye JP, Brugger P. Human Behavioral Complexity Peaks at Age 25. PLOS Computational Biology. 2017; 13(4):e1005408. https://doi.org/10.1371/journal.pcbi. 1005408 PMID: 28406953 56. Guseva M, Bogler C, Allefeld C, Haynes JD. Instruction Effects on Randomness in Sequence Genera- tion. Frontiers in Psychology. 2023; https://doi.org/10.3389/fpsyg.2023.1113654 PMID: 37034908 57. Rapoport A, Budescu DV. Generation of Random Series in Two-Person Strictly Competitive Games. Journal of Experimental Psychology: General. 1992; 121(3):352–363. https://doi.org/10.1037/0096- 3445.121.3.352 58. Castillo L, Leo´ n-Villagra´ P, Chater N, Sanborn AN. Production Rate Effects on Random Generation in a Naturalistic Domain;. 59. Marwan N, Romano MC, Thiel M, Kurths J. Recurrence Plots for the Analysis of Complex Systems. Physics reports. 2007; 438(5-6):237–329. https://doi.org/10.1016/j.physrep.2006.11.001 60. Blakeslee AF. Corn and Men: The Interacting Influence of Heredity and Environment—Movements for Betterment of Men, or Corn, or Any Other Living Thing, One-sided Unless They Take Both Factors into Account. Journal of Heredity. 1914; 5(11):511–518. https://doi.org/10.1093/oxfordjournals.jhered. a107785 61. Castillo L, Leo´ n-Villagra´ P, Chater N, Sanborn AN. Local Sampling with Momentum Accounts for Human Random Sequence Generation. In: Proceedings of the Annual Meeting of the Cognitive Science Society. vol. 43; 2021. 62. Francis WN, Kucera H. Brown Corpus Manual. Manual of Information to Accompany A Standard Cor- pus of Present-Day Edited American English, for Use with Digital Computers. Providence, Rhode Island: Brown University; 1979. 63. Pudlo P, Marin JM, Estoup A, Cornuet JM, Gautier M, Robert CP. Reliable ABC Model Choice via Ran- dom Forests. Bioinformatics (Oxford, England). 2016; 32(6):859–866. https://doi.org/10.1093/ bioinformatics/btv684 PMID: 26589278 64. Hinne M, Gronau QF, van den Bergh D, Wagenmakers EJ. A Conceptual Introduction to Bayesian Model Averaging. Advances in Methods and Practices in Psychological Science. 2020; 3(2):200–215. https://doi.org/10.1177/2515245919898657 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011739 January 5, 2024 24 / 24 PLOS COMPUTATIONAL BIOLOGY
10.1371_journal.pdig.0000467	RESEARCH ARTICLE Clinical gait analysis using video-based pose estimation: Multiple perspectives, clinical populations, and measuring change Jan StenumID 1,2, Melody M. Hsu1,3, Alexander Y. Pantelyat4, Ryan T. RoemmichID 1,2* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Stenum J, Hsu MM, Pantelyat AY, Roemmich RT (2024) Clinical gait analysis using video-based pose estimation: Multiple perspectives, clinical populations, and measuring change. PLOS Digit Health 3(3): e0000467. https:// doi.org/10.1371/journal.pdig.0000467 Editor: Mengling Feng, National University Singapore Saw Swee Hock School of Public Health, SINGAPORE Received: May 4, 2023 Accepted: February 12, 2024 Published: March 26, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pdig.0000467 Copyright: © 2024 Stenum et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The dataset of unimpaired gait is available from http://bytom.pja. edu.pl/projekty/hm-gpjatk. The stroke and PD 1 Center for Movement Studies, Kennedy Krieger Institute, Baltimore, Maryland, United States of America, 2 Department of Physical Medicine and Rehabilitation, The Johns Hopkins University School of Medicine, Baltimore, Maryland, United States of America, 3 Department of Neuroscience, The Johns Hopkins University School of Medicine, Baltimore, Maryland, United States of America, 4 Department of Neurology, The Johns Hopkins University School of Medicine, Baltimore, Maryland, United States of America * [email protected] Abstract Gait dysfunction is common in many clinical populations and often has a profound and dele- terious impact on independence and quality of life. Gait analysis is a foundational compo- nent of rehabilitation because it is critical to identify and understand the specific deficits that should be targeted prior to the initiation of treatment. Unfortunately, current state-of-the-art approaches to gait analysis (e.g., marker-based motion capture systems, instrumented gait mats) are largely inaccessible due to prohibitive costs of time, money, and effort required to perform the assessments. Here, we demonstrate the ability to perform quantitative gait anal- yses in multiple clinical populations using only simple videos recorded using low-cost devices (tablets). We report four primary advances: 1) a novel, versatile workflow that lever- ages an open-source human pose estimation algorithm (OpenPose) to perform gait analy- ses using videos recorded from multiple different perspectives (e.g., frontal, sagittal), 2) validation of this workflow in three different populations of participants (adults without gait impairment, persons post-stroke, and persons with Parkinson’s disease) via comparison to ground-truth three-dimensional motion capture, 3) demonstration of the ability to capture clinically relevant, condition-specific gait parameters, and 4) tracking of within-participant changes in gait, as is required to measure progress in rehabilitation and recovery. Impor- tantly, our workflow has been made freely available and does not require prior gait analysis expertise. The ability to perform quantitative gait analyses in nearly any setting using only low-cost devices and computer vision offers significant potential for dramatic improvement in the accessibility of clinical gait analysis across different patient populations. Author summary People that experience a stroke or are diagnosed with Parkinson’s disease often have mobility impairments such as slow walking speeds, shortened steps and abnormal move- ment of the legs during walking. It is a challenge for clinicians to measure and track the PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 1 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation datasets contain videos with identifiable information and are therefore not available. Code for our workflow is available at https://github.com/ janstenum/GaitAnalysis-PoseEstimation/tree/ Multiple-Perspectives. Funding: We acknowledge funding from the NIH (grant R21 HD110686 to RTR), RESTORE Center Pilot Project Award (to RTR via NIH grant P2CHD101913), the American Parkinson Disease Association (grant 964604 to RTR), and the Sheikh Khalifa Stroke Institute at Johns Hopkins Medicine to RTR. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. multitude of walking parameters that can indicate recovery or progression of disease in an objective and quantitative manner. We present a new workflow that allows a user to ana- lyze the gait pattern of a person walking recorded with only a single video obtained with a smartphone or other digital recording device. We test our workflow is 3 groups of partici- pants: persons with no gait impairment, persons post-stroke, and persons with Parkin- son’s disease. We show that a user can perform these video-based gait analyses by recording videos with views from either the side or the front, which is important given the space restrictions in most clinical areas. Our workflow can produce accurate results as compared with a gold standard three-dimensional motion capture system. Furthermore, the workflow can track changes in gait, which is needed to measure changes in mobility over time that may occur because of recovery or progression of disease. This work offers potential for dramatic improvement in the accessibility of clinical gait analysis across dif- ferent patient populations. Introduction Walking is the primary means of human locomotion. Many clinical conditions–including neu- rologic damage or disease (e.g., stroke, Parkinson’s disease (PD), cerebral palsy), orthopedic injury, and lower extremity amputation–have a debilitating effect on the ability to walk [1–3]. Quantitative gait analysis is the foundation for effective gait rehabilitation [4]: it is critical that we objectively measure and identify specific deficits in a patient’s gait and track changes. Unfortunately, there are significant limitations with the current state-of-the-art. Marker-based motion capture laboratories are considered the gold standard measurement technique, but they are prohibitively costly and available largely to select hospitals and research institutions. Other commercially available technologies (e.g., gait mats, wearable systems) only provide pre- defined parameters (e.g., spatiotemporal data or step counts), are relatively costly, and require specific hardware. There is a clear need for new technologies that can lessen these barriers and provide accessible and clinically useful gait analysis with minimal costs of time, money, and effort. Recent developments in computer vision have enabled the exciting prospect of quantitative movement analysis using only digital videos recorded with low-cost devices such as smart- phones or tablets [5–7]. These pose estimation technologies leverage computer vision to iden- tify specific “keypoints” on the human body (e.g., knees, ankles) automatically from simple digital videos [8,9]. The number of applications of pose estimation for human health and per- formance has increased exponentially in recent years due to the potential for dramatic improvement in the accessibility of quantitative movement assessment [6,7,10]. We have pre- viously used OpenPose [8]–a freely available pose estimation algorithm–to develop and test a comprehensive video-based gait analysis workflow, demonstrating the ability to measure a variety of spatiotemporal gait parameters and lower-limb joint kinematics from only short (<10 seconds) sagittal (side view) videos of individuals without gait impairment [11]. Others have also used a variety of approaches to combine pose estimation outputs and neural net- works to estimate different aspects of mobility [5,12–16]. This foundational work in using pose estimation for video-based gait analysis has demon- strated significant potential of this emerging technology. There are now prime opportunities to build upon what has already been developed and progress toward direct clinical applica- tions. In moving toward clinical application, we considered the needs for: 1) flexible approaches that can accommodate different perspectives based on the space constraints of the PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 2 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Fig 1. Conceptual overview. We recorded three-dimensional (3D) motion capture and digital videos of gait trials performed by persons post-stroke and persons with Parkinson’s disease (A). We analyzed digital videos of the frontal (CFront) and sagittal plane (CSag) with OpenPose to track anatomical keypoints (B). We developed workflows to perform a gait analysis, independently, for videos of the frontal and sagittal plane (C). See Methods section for detailed information about the frontal and sagittal plane post-processing workflows. Note that the ‘Calculate depth-change time-series’ step in the frontal workflow contains multiple sub-steps including tracking the pixel size of the torso and low-pass filtering (see S4 Fig for justification of tracking method and smoothing). We compared spatiotemporal gait parameters and joint kinematics from our workflows to parameters obtained with 3D motion capture (D). https://doi.org/10.1371/journal.pdig.0000467.g001 end user (e.g., a clinician may only have access to a long, narrow hallway or hospital corridor where a sagittal recording of the patient is not possible), 2) testing and validation directly in clinical populations with gait dysfunction, 3) measurement of clinically relevant gait parame- ters that are of particular relevance to specific populations, and 4) the ability to measure changes in gait that occur in response to a change in speed. Here, we present a novel, versatile approach for performing clinical gait analysis using only simple digital videos. First, we developed and tested a novel workflow that performs a gait analysis using frontal plane recordings of a person walking either away from or toward the camera (Fig 1). Our approach is based on tracking the size of the person as they appear in the video image (measured with keypoints from OpenPose) and using trigonometric relationships to estimate depth and, ultimately, spatial parameters such as step length and PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 3 / 23 CFront3D motioncaptureCSagOriginal videoOpenPoseCFrontCSagChange coordi-nate systemCorrect left-right limb identificationGap fill and low-pass filterCalculate scaling factorFind eventsChange coordi-nate systemCorrect left-right limb identificationGap fillCalculate depth-change Find eventsGait analysisPost-processingDigitalvideosABCSagittal workflowFrontal workflow• Spatiotemporal gait parameters(cid:129) Spatiotemporal gait parameters(cid:129) 2D joint kinematicsMotion captureDGait analysis(cid:129) Spatiotemporal gait parameters(cid:129) 2D/3D joint kinematicsPLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Fig 2. Diagram of frontal plane analysis to obtain spatiotemporal gait parameters. A person of size (height) s stands at two distances from a frontal plane camera (CFront; panel A): an initial reference depth (dRef) and at a depth-change (Δdi). The size in pixels of the person at each depth are denoted by sRef and si. From trigonometric relationships we derive a relationship between pixel size and depth-change (B, see Methods for detailed explanation; f, focal length of camera; xIP, position of image plane of camera; xCam, position of camera lens; xRef, initial position of person; xi, position of person following depth-change). The predicted pixel sizes of a person standing at increasing depths closely tracks manually annotated pixel sizes, which shows that we can use pixel size to estimate depth-changes (C). Summary of our frontal plane workflow (D): OpenPose tracks anatomical keypoints, we find gait cycle events, calculate a time-series of pixel size, and calculate depth-change at which point step lengths and step times can be derived. https://doi.org/10.1371/journal.pdig.0000467.g002 gait speed (Fig 2; see expanded description in Methods). Second, we test both our frontal and sagittal workflows directly in two clinical populations with gait impairments that result from neurologic damage or disease (persons post-stroke or with Parkinson’s disease). PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 4 / 23 ssisRef fsisRefsxCamxRefxixIPImage planeLens02.557.51012.51517.520∆di (m)00.10.20.30.40.50.60.70.80.91sRatioPredicted (dRef = 4.88 m)Manual annotationyxsRefsisisiABC0 m4.27 m8.53 m18.29 mTrackedFind gait eventsCalculatesize ratioTime (s)−50−2502550Time (s)00.20.40.60.81sRatioTime (s)012345∆di (m)Left gait cycleRight gait cycleTimeLengthOriginalCalculatedepth-changeD012345012345012345Vertical dist. betweenankle keypoints (pixels)dRef∆disRatio = si × sRef−1sRatio = dRef × (dRef + ∆di)−1CFrontCalculate spatiotemporal gait parametersxRefxiPLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Results Development and testing of a novel approach for gait videos recorded in the frontal plane We first validated our frontal plane approach during overground walking in a group of young participants without gait impairment (we have previously demonstrated the accuracy of obtaining gait parameters using sagittal plane videos in the same dataset of unimpaired partici- pants [11]). We then compared spatiotemporal gait parameters (step time, step length and gait speed; averaged values for a single walking bout) simultaneously obtained with 3D motion capture and with frontal plane videos positioned to capture the person walking away from one camera and toward the other camera (data collection setup shown in Fig 3A). Step time showed average differences (negative values denote greater values in video data; positive values denote greater values in motion capture data) and errors (absolute difference) up to one and two motion capture frames (motion capture recorded at 100 Hz; 0.01 and 0.02 s), respectively, between motion capture and frontal plane video (Fig 3B and S1 Table). The 95% limits of agreement between motion capture and frontal plane videos ranged from Fig 3. Testing of a novel approach for spatiotemporal gait analysis from videos of unimpaired adults recorded in the frontal plane. We recorded digital videos of the frontal plane where the person walked toward one camera and away from the other camera (A). We compared spatiotemporal gait parameters (B, step time; C, step length; D, gait speed) between the two digital videos and 3D motion capture (see S1 Table). https://doi.org/10.1371/journal.pdig.0000467.g003 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 5 / 23 0.40.60.81Step time (s) MC0.40.60.81Step time (s)r=0.973P<0.0010.40.60.81Step time (s) MC0.40.60.81Step time (s)r=0.981P<0.0010.40.60.81Step time (s)0.40.60.81Step time (s)r=0.956P<0.001−0.1−0.0500.050.1Diff. step time (s)0.20.40.60.811.2Step length (m)0.20.40.60.811.2Step length (m)r=0.908P<0.0010.20.40.60.811.2Step length (m)0.20.40.60.811.2Step length (m)r=0.872P<0.0010.20.40.60.811.2Step length (m)0.20.40.60.811.2Step length (m)r=0.917P<0.001−0.25−0.12500.1250.25Diff. step length (m)0.60.811.21.41.61.8Gait speed (m s−1)0.60.811.21.41.61.8Gait speed (m s−1)r=0.952P<0.0010.60.811.21.41.61.8Gait speed (m s−1)0.60.811.21.41.61.8Gait speed (m s−1)r=0.938P<0.0010.60.811.21.41.61.8Gait speed (m s−1)0.60.811.21.41.61.8Gait speed (m s−1)r=0.945P<0.001−0.25−0.12500.1250.25Diff. gait speed (m s−1)BCDCFrontTowardCFrontAwayA3D motioncaptureCFrontAwayCFrontAwayCFrontTowardCFrontToward MC MCCFrontAwayCFrontAwayCFrontTowardCFrontToward MC MCCFrontAwayCFrontAwayCFrontTowardCFrontTowardMC−CFrontAwayMC−CFrontTowardCFront−CFrontAwayTowardMC−CFrontAwayMC−CFrontTowardCFront−CFrontAwayTowardMC−CFrontAwayMC−CFrontTowardCFront−CFrontAwayTowardUnimpairedPLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation −0.03 to 0.05 s, suggesting that 95% of differences with motion capture fell within this interval. Step length showed average differences and errors up to about 0.02 and 0.03 m, respectively, between motion capture and frontal plane videos (Fig 3C). The 95% limits of agreement between motion capture and frontal plane videos ranged from −0.052 to 0.094 m. Gait speed showed average differences and error up to 0.04 and 0.06 m s−1, respectively, with 95% limits of agreement ranging between −0.11 and 0.17 m s−1 (Fig 3D). Correlations for all spatiotempo- ral gait parameters between motion capture and frontal plane videos were strong (all r values between 0.872 and 0.981, all P<0.001; Fig 3B–3D). Testing of video-based gait analysis in persons with neurologic damage or disease Next, we evaluated both our sagittal and frontal plane workflows in two patient populations with neurologic damage or disease (persons post-stroke and persons with PD). We compared spatiotemporal gait parameters (step time, step length, and gait speed), lower-limb sagittal plane joint kinematics, and condition-specific, clinically relevant parameters (stroke: step time asymmetry and step length asymmetry; PD: trunk inclination) simultaneously obtained with 3D motion capture and with sagittal and frontal plane videos (data collection setup shown in Fig 4A). Note that frontal videos are limited to spatiotemporal gait parameters and that joint kinematics and trunk inclination can only be obtained from sagittal videos within our current workflows. We present gait parameters as averaged values across four overground walking bouts each at 1) preferred and 2) fast speeds (see S2 Table for values of gait parameters). For preferred speed trials we instructed participants to walk at their preferred speed; for fast speed trials we instructed participants to walk at the fastest speed that they felt comfortable. Of the four trials at each speed, there were two trials of the participants walking away from the frontal camera (with the left side against the sagittal camera) and two trials walking toward the frontal camera (with the right side against the sagittal camera). We intend our workflows to have clinical applications and therefore present values as session-level values (i.e., the results that would be obtained as if the four walking trials were treated as a single clinical gait analysis); we report more detailed comparisons at the level of single trial averages and step-by-step comparisons in the supplement (S3 and S4 Tables). Testing in persons post-stroke We then tested how well our workflows could measure gait parameters in persons post-stroke. Step time showed average differences and errors of zero and one motion capture frames (recorded at 100 Hz; 0 and 0.01 s), respectively, between motion capture and sagittal videos; and average differences and errors of two and five motion capture frames (0.02 and 0.05 s), respectively, between motion capture and frontal videos (Fig 4B and Table 1). The 95% limits of agreement spanned a narrower interval (−0.04 to 0.04 s) for sagittal videos than frontal vid- eos (−0.09 to 0.10 s). Correlations of step time between motion capture and videos were strong (Fig 4B; all r�0.980). Step length showed average differences and errors of about 1 and 3 cm between motion capture and sagittal videos and average differences and errors of about −3 and 7 cm between motion capture and frontal videos (Fig 4C and Table 1). The 95% limits of agreement spanned intervals of −0.058 to 0.079 m for sagittal videos and −0.154 to 0.087 m for frontal videos. Cor- relations of step length between motion capture and videos were strong (Fig 4C; r�0.922). Gait speed showed average differences and errors of 0.02 and 0.04 m s−1 between motion capture and sagittal videos and average differences and errors of −0.07 and 0.10 m s−1 between PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 6 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Fig 4. Video-based gait analysis from frontal and sagittal views in persons post-stroke. We recorded digital videos of the frontal and sagittal plane during gait trials (A). We compared spatiotemporal gait parameters (B, step time; C, step length; D, gait speed) and gait asymmetry (E, step time asymmetry; F, step length asymmetry) between the two digital videos and 3D motion capture. We also compared lower-limb joint kinematics at the hip, knee and ankle obtained with sagittal videos and motion capture for the paretic (G) and non-paretic (H) limbs (MAE, mean absolute error). Gait parameters are calculated as session-level averages of four gait trials at either preferred or fast speeds (see Table 1). https://doi.org/10.1371/journal.pdig.0000467.g004 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 7 / 23 Step time (s) MC0123Step time (s) CSagr=0.997P<0.001Step time (s) MC0123Step time (s) CFrontr=0.980P<0.001Step time (s) CSag0123Step time (s) CFrontr=0.984P<0.001−0.200.2Diff. step time (s)00.51Step length (m)00.51Step length (m)r=0.967P<0.00100.51Step length (m)00.51Step length (m)r=0.922P<0.00100.51Step length (m)00.51Step length (m) r=0.918P<0.001−0.200.2Diff. step length (m)Gait speed (m s−1)012Gait speed (m s−1)r=0.981P<0.001Gait speed (m s−1)012Gait speed (m s−1)r=0.989P<0.001Gait speed (m s−1)012Gait speed (m s−1)r=0.987P<0.001−0.500.5Diff. gait speed (m s−1)−0.200.20.40.6Step time asym.−0.200.20.40.6Step time asym.r=0.977 P<0.001−0.200.20.40.6Step time asym.−0.200.20.40.6Step time asym.r=0.865P<0.001−0.200.20.40.6Step time asym.−0.200.20.40.6Step time asym.r=0.886P<0.001−0.200.2Diff. step time asym.−0.200.20.40.6Step length asym.−0.200.20.40.6Step length asym.r=0.890P<0.001−0.200.20.40.6Step length asym.−0.200.20.40.6Step length asym.r=0.230P=0.033−0.200.20.40.6Step length asym.−0.200.20.40.6Step length asym.r=0.461P<0.001−0.6−0.4−0.200.2Diff. step length asym.Gait cycle (%)−20020Hip angle (°) PareticGait cycle (%)0204060Knee angle (°) PareticGait cycle (%)−20020Ankle angle (°) PareticHipKneeAnkle0102030MAE (°) MC vs CSagGait cycle (%)−20020Hip angle (°) Non-pareticGait cycle (%)0204060Knee angle (°) Non-paretic80100Gait cycle (%)−20020Ankle angle (°) Non-pareticHipKneeAnkle0102030MAE (°) MC vs CSagBCDEFGHCFrontA3D motioncaptureCSag MC CSag MC CFront CSag CFront MC CSag MC CFront CSag CFront MC CSag MC CFront CSag CFront MC CSag MC CFront CSag CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFront6040200801006040200801006040200801006040200801006040200801006040200012012012012301230123 Fast Preferred Paretic Non-pareticStrokeMCCSagPLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Table 1. Comparison of video-based and motion capture measurements of spatiotemporal gait parameters in the stroke and Parkinson’s disease groups a. Gait Parameter MC−CS Difference (Mean±SD) MC−CF CF−CS \|MC−CS\| Error (Mean±SD) \|MC−CF\| 95% Limits of Agreement \|CS−CF\| MC−CS MC−CF CF−CS Step time (s) Step length (m) b Gait speed (m s−1) b Step time asym. Step length asym. b 0.00±0.02 0.010±0.035 0.02±0.06 0.01±0.03 −0.002±0.072 0.01±0.05 −0.033±0.061 −0.07±0.07 0.02±0.06 −0.042±0.127 0.01±0.04 −0.050±0.063 −0.09±0.07 0.01±0.06 −0.025±0.107 Stroke 0.02±0.01 0.028±0.024 0.04±0.05 0.03±0.02 0.050±0.053 0.05±0.04 0.072±0.037 0.10±0.06 0.07±0.05 0.106±0.097 0.05±0.04 0.079±0.044 0.12±0.06 0.07±0.04 0.100±0.073 −0.04; 0.04 −0.058; 0.079 −0.11; 0.14 −0.04; 0.07 −0.142; 0.138 −0.09; 0.10 −0.154; 0.087 −0.20; 0.06 −0.10; 0.14 −0.291; 0.208 −0.08; 0.09 −0.173; 0.073 −0.22; 0.04 −0.10; 0.12 −0.235; 0.186 Step time (s) Step length (m) b Gait speed (m s−1) b Trunk incl. (˚) c −0.00±0.01 −0.010±0.017 −0.02±0.02 −0.0±1.5 0.01±0.02 −0.051±0.050 −0.12±0.08 . . . 0.01±0.02 −0.041±0.055 −0.10±0.09 . . . Parkinson’s disease 0.01±0.00 0.021±0.009 0.03±0.02 1.5±0.7 0.03±0.01 0.074±0.042 0.15±0.07 . . . 0.03±0.01 0.075±0.040 0.15±0.06 . . . −0.02; 0.02 −0.044; 0.023 −0.07; 0.03 −3.0; 2.9 −0.03; 0.05 −0.150; 0.048 −0.28; 0.04 . . . −0.03; 0.05 −0.149; 0.068 −0.27; 0.07 . . . MC, motion capture; CS, sagittal plane camera; CF, frontal plane camera a Values of spatiotemporal gait parameters are calculated as session-level averages. b Parameter depending on step length: comparisons of MC and CS, step length calculated as distance between ankles at heel-strike; comparisons of MC and CF and of CS and CF, step length calculated as distance travelled by torso between consecutive heel-strikes. c Missing values because trunk inclination cannot be calculated from CF. https://doi.org/10.1371/journal.pdig.0000467.t001 motion capture and frontal videos (Fig 4D and Table 1). The 95% limits of agreement spanned intervals of −0.11 to 0.14 m s−1 for sagittal videos and −0.20 to 0.06 m s−1 for frontal videos. Correlations of gait speed between motion capture and videos were strong (Fig 4D; r�0.981). Step time asymmetry showed average differences and errors of 0.01 and 0.03 between motion capture and sagittal videos and average differences and errors of 0.02 and 0.07 between motion capture and frontal videos (Fig 4E and Table 1). The 95% limits of agree- ment spanned intervals of −0.04 to 0.07 for sagittal videos and −0.10 to 0.14 for sagittal vid- eos. Correlations of step time asymmetry between motion capture and videos were strong (Fig 4E; all r�0.865). Step length asymmetry showed average differences and errors of −0.002 and 0.050 between motion capture and sagittal videos and average differences and errors of −0.042 and 0.106 between motion capture and frontal videos (Fig 4F and Table 1). The 95% limits of agreement spanned intervals of −0.142 to 0.138 for sagittal videos and −0.291 to 0.208 for frontal videos. Correlations of step length asymmetry were strong between motion capture and sagittal videos (Fig 4F; r = 0.890) but weak between motion capture and frontal videos (Fig 4F; r = 0.230). The average mean absolute errors of lower-limb sagittal plane joint kinematics of the paretic and non-paretic limbs were 3.3˚, 4.0˚, and 6.3˚ at the hip, knee, and ankle, respectively, between motion capture and sagittal videos (Fig 4G and 4H). Testing in persons with Parkinson’s disease We next evaluated the performance of the video-based gait analysis in persons with PD (Fig 5A). Step time showed average differences and errors of zero and one motion capture frames (0 and 0.01 s) between motion capture and sagittal videos and average differences and errors of one and three motion capture frames (0.01 and 0.03 s) between motion capture and frontal videos (Fig 5B and Table 1). The 95% limits of agreement spanned intervals of −0.02 to 0.02 s for sagittal videos and −0.03 to 0.05 s for frontal videos. Correlations of step time between motion capture and videos were strong (Fig 5B; all r�0.961). PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 8 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Fig 5. Video-based gait analysis from frontal and sagittal views in persons with Parkinson’s disease. We recorded digital videos of the frontal and sagittal plane during gait trials (A). We compared spatiotemporal gait parameters (B, step time; C, step length; D, gait speed) between the two digital videos and 3D motion capture. We compared trunk inclination between sagittal plane videos and motion capture (E). We also compared lower-limb joint kinematics at the hip, knee and ankle obtained with sagittal videos and motion capture for the right (F) and non-paretic (G) limbs (MAE, mean absolute error). Gait parameters are calculated as session-level averages of four gait trials at either preferred or fast speeds (see Table 1). https://doi.org/10.1371/journal.pdig.0000467.g005 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 9 / 23 0.40.60.80.40.60.8r=0.993P<0.0010.40.60.80.40.60.8r=0.961P<0.0010.40.60.80.40.60.8r=0.959P<0.001−0.200.20.20.40.60.811.20.20.40.60.811.2r=0.987P<0.0010.20.40.60.811.20.20.40.60.811.2r=0.959P<0.0010.20.40.60.811.20.20.40.60.811.2r=0.940P<0.001−0.200.20.511.522.50.511.522.5r=0.996P<0.0010.511.522.50.511.522.5r=0.982P<0.0010.511.522.50.511.522.5r=0.972P<0.001−0.500.55060708090Trunk incl. (°)5060708090Trunk incl. (°)r=0.964P<0.001−505Diff. trunk incl. (°)BCDEFGGait cycle (%)−20020Hip angle (°)RightGait cycle (%)0204060Knee angle (°)RightGait cycle (%)−20020Ankle angle (°)RightHipKneeAnkle0102030MAE (°) MC vs CSagGait cycle (%)−20020Hip angle (°)LeftGait cycle (%)0204060Knee angle (°)Left80100Gait cycle (%)−20020Ankle angle (°) LeftHipKneeAnkle0102030MAE (°) MC vs CSag6040200801006040200801006040200801006040200801006040200801006040200Step time (s) MCStep time (s) CSagStep time (s) MCStep time (s) CFrontStep time (s) CSagStep time (s) CFrontDiff. step time (s)Step length (m)Step length (m)Step length (m)Step length (m)Step length (m)Step length (m) MC CSag MC CFront CSag CFrontGait speed (m s−1)Gait speed (m s−1)Gait speed (m s−1)Gait speed (m s−1)Gait speed (m s−1)Gait speed (m s−1)Diff. gait speed (m s−1) MC CSag MC CFront CSag CFrontDiff. step length (m) MC CSagMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSagMC−CFrontCSag−CFrontMC−CSag Fast Preferred RightLeftCFrontA3D motioncaptureCSagParkinson's diseaseMCCSagPLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Step length showed average differences and errors of about −1 and 2 cm between motion capture and sagittal videos and average differences and errors of −5 and 7 cm between motion capture and frontal videos (Fig 5C and Table 1). The 95% limits of agreement spanned intervals of −0.044 to 0.023 m for sagittal videos and −0.150 to 0.048 m for frontal videos. Correlations of step length between motion capture and videos were strong (Fig 5C; all r�0.959). Gait speed showed average differences and errors of −0.02 and 0.03 m s−1 between motion capture and sagittal videos and average differences and errors of −0.12 and 0.15 m s−1 between motion capture and frontal videos (Fig 5D and Table 1). The 95% limits of agreement spanned intervals of −0.07 to 0.03 m s−1 for sagittal videos and −0.28 to 0.04 m s−1 for frontal videos. Correlations of gait speed between motion capture and videos were strong (Fig 5D; all r�0.982). Trunk inclination showed average differences and errors of 0˚ and 1.5˚ between motion capture and sagittal videos (Fig 5E and Table 1; trunk inclination can only be extracted from sagittal videos, not frontal videos). The average mean absolute errors of left and right lower-limb sagittal plane joint kinematics were 2.7˚, 3.5˚, and 4.8˚ at the hip, knee, and ankle, respectively, between motion capture and sagittal videos (Fig 5F and 5G). Measuring changes in gait that occur due to changes in gait speed Next, to evaluate how accurately video analysis can track within-participant gait changes, we calculated the changes in spatiotemporal gait parameters that accompanied the increase in gait speed from preferred to fast speed gait trials in persons post-stroke and with PD (Fig 6A). The change in step time as a result of faster walking in persons post-stroke showed average differ- ences and errors of zero and two motion capture frames (0 and 0.02 s) when compared between motion capture and sagittal videos and average differences and errors of zero and four motion capture frames (0 and 0.04 s) when compared between motion capture and fron- tal videos (Fig 6B and Table 2). The 95% limits of agreement of change in step time of post- stroke walking spanned intervals of −0.03 to 0.03 s for sagittal videos and −0.08 to 0.07 s for frontal videos. In persons with PD, the change in step time showed average differences and error of zero and two motion capture frames (0 and 0.02 s) between motion capture and sagittal videos and average differences and errors of zero and three motion capture frames (0 and 0.03 s) between motion capture and frontal videos (Fig 6B and Table 2). The 95% limits of agreement of change in step time of PD walking spanned intervals of −0.02 to 0.02 s for sagittal videos and −0.05 to 0.04 s for frontal videos. Correlations of change in step time between motion capture and videos were strong (Fig 6B; all r�0.828). The change in step length as a result of faster walking in persons post-stroke showed aver- age differences and errors of about 0 and 2 cm between motion capture and sagittal videos and average differences and errors of about −1 and 5 cm between motion capture and frontal vid- eos (Fig 6C and Table 2). The 95% limits of agreement of change in step length of post-stroke walking spanned intervals of −0.031 to 0.037 m for sagittal videos and −0.088 to 0.075 m for frontal videos. Change in step length in persons with PD showed average differences and errors of about 0 and 2 cm between motion capture and sagittal videos and average differences and errors of about −3 and 7 cm between motion capture and frontal videos (Fig 6C and Table 2). The 95% limits of agreement of change in step length of PD walking spanned intervals of −0.022 to PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 10 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Fig 6. Measuring changes in gait that occur due to changes in gait speed in persons post-stroke and persons with Parkinson’s disease. We recorded digital videos of the frontal and sagittal plane during gait trials at preferred and fast speeds (A). We compared spatiotemporal gait parameters (B, step time; C, step length; D, gait speed) between the two digital videos and 3D motion capture. Subscripts Δv of gait parameters denote changes in the gait parameter due to speed-increases from preferred to fast speed walking trials. We calculated gait parameters as the difference between the session-level averages of preferred and fast speed trials (see Table 2). https://doi.org/10.1371/journal.pdig.0000467.g006 0.028 m for sagittal videos and −0.122 to 0.070 m for frontal videos. Correlations of change in step length between motion capture and videos were strong (Fig 6C; all r�0.763). The change in gait speed from preferred to fast speed gait trials in persons post-stroke showed average differences and errors of 0.01 and 0.04 m s−1 between motion capture and sag- ittal videos and average differences and errors of −0.02 and 0.06 m s−1 between motion capture and frontal videos (Fig 6D and Table 2). The 95% limits of agreement of change in gait speed of post-stroke walking spanned intervals of −0.09 to 0.11 m s−1 for sagittal videos and −0.14 to 0.11 m s−1 for frontal videos. Finally, in persons with PD, measured change in gait speed showed average differences and errors of 0 and 0.03 m s−1 between motion capture and sagittal videos and average dif- ferences and errors of −0.07 and 0.11 m s−1 between motion capture and frontal videos (Fig 6D and Table 2). The 95% limits of agreement of change in gait speed of PD walking spanned intervals of −0.04 to 0.04 m s−1 for sagittal videos and −0.19 to 0.06 m s−1 for PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 11 / 23 −0.4−0.20−0.4−0.20r=0.963P<0.001−0.4−0.20−0.4−0.20r=0.828P<0.001−0.4−0.20−0.4−0.20r=0.826P<0.001−0.200.200.20.400.20.4r=0.942P<0.00100.20.400.20.4r=0.763P<0.00100.20.400.20.4r=0.729P<0.001−0.200.200.20.40.600.20.40.6r=0.965P<0.00100.20.40.600.20.40.6r=0.949P<0.00100.20.40.600.20.40.6r=0.965P<0.001−0.200.2BCDStep time∆v (s) MCStep time∆v (s) CSagStep time∆v (s) MCStep time∆v (s) CSagStep time∆v (s) CFrontDiff. step time∆v (s)MC−CSagMC−CFrontCSag−CFrontStep length∆v (m)Step length∆v (m)Step length∆v (m)Step length∆v (m)Step length∆v (m)Step length∆v (m) MC CSag MC CFront CSag CFrontDiff. step length∆v (m)MC−CSagMC−CFrontCSag−CFrontGait speed∆v (m s−1)Gait speed∆v (m s−1)Gait speed∆v (m s−1)Gait speed∆v (m s−1)Gait speed∆v (m s−1)Gait speed∆v (m s−1)Diff. gait speed∆v (m s−1) MC CSag MC CFront CSag CFrontMC−CSagMC−CFrontCSag−CFrontStep time∆v (s) CFront PD Stroke Paretic/Right Non-paretic/LeftCFrontA3D motioncaptureCSagMeasuring change in stroke and Parkinson's diseasePLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation Table 2. Comparison of video-based and motion capture measurements of speed-related changes of spatiotemporal gait parameters of stroke and PD groups a. Gait Parameter MC−CS Difference (Mean±SD) MC−CF CF−CS \|MC−CS\| Error (Mean±SD) \|MC−CF\| 95% Limits of Agreement \|CS−CF\| MC−CS MC−CF CF−CS Stroke Step time (s) Step length (m) b Gait speed (m s−1) b −0.00±0.01 0.003±0.017 0.01±0.05 −0.00±0.04 −0.007±0.042 −0.02±0.06 −0.00±0.04 −0.001±0.044 −0.02±0.05 0.02±0.01 0.021±0.012 0.04±0.04 0.04±0.03 0.054±0.027 0.06±0.05 0.05±0.03 0.058±0.029 0.06±0.04 −0.03; 0.03 −0.031; 0.037 −0.09; 0.11 −0.08; 0.07 −0.088; 0.075 −0.14; 0.11 −0.07; 0.07 −0.087; 0.085 −0.12; 0.08 Step time (s) Step length (m) b Gait speed (m s−1) b −0.00±0.01 0.003±0.013 0.00±0.02 −0.00±0.02 −0.026±0.049 −0.07±0.06 −0.00±0.02 −0.015±0.056 −0.04±0.06 Parkinson’s disease 0.02±0.01 0.019±0.007 0.03±0.01 0.03±0.02 0.067±0.035 0.11±0.06 0.04±0.02 0.073±0.039 0.10±0.05 −0.02; 0.02 −0.022; 0.028 −0.04; 0.04 −0.05; 0.04 −0.122; 0.070 −0.19; 0.06 −0.05; 0.04 −0.125; 0.094 −0.16; 0.08 MC, motion capture; CS, sagittal plane camera; CF, frontal plane camera a Speed-changes are differences between preferred and fast speed walking trials; gait parameters are calculated as session-level averages. b Parameter depending on step length: comparisons of MC and CS, step length calculated as distance between ankles at heel-strike; comparisons of MC and CF and of CS and CF, step length calculated as distance travelled by torso between consecutive heel-strikes. https://doi.org/10.1371/journal.pdig.0000467.t002 frontal videos. Correlations of change in gait speed between motion capture and videos were strong (Fig 6D; all r�0.949). Factors that affect accuracy of the frontal video-based gait analysis workflow We noted that step length errors were occasionally large when calculated from frontal videos (up to nearly 30% of the average step length). We have previously described factors such as the position of the person relative to the camera that influence step length errors when calculated from sagittal videos [11]. Similarly, we wanted to identify and understand factors that influ- ence step length errors from videos recorded in the frontal plane. First, we considered that greater depth of the person relative to the frontal plane camera may lead to less precise step length estimates (S1 Fig). We partitioned the analysis of step length errors into videos from the frontal plane where the person walks away from the camera or toward the camera because OpenPose may track keypoints differently when viewing the front of the person (when walking toward) or the back of the person (when walking away). We found that step length errors increased with greater depth from the camera, so that the per- son’s size appeared smaller in the image. Step length errors were more affected by depth when the person walked away from the camera compared to walking toward the camera: from aver- age step length errors of about 7 cm nearest the camera (beginning of trial when the person walks away from the camera; end of the trial when the person walks toward the camera), aver- age errors increased up to about 16 cm when the person walked away, with a more modest increase of up to 11 cm when the person walked toward the camera. This suggests that preci- sion may decrease as the person appears smaller, likely due to less precise keypoint tracking by OpenPose. We also considered whether a scaling effect influenced step length errors so that longer steps had greater errors. We found that step length errors were not influenced by the magni- tude of step length (S2 Fig). We noted time-lags in the gait cycle detection of the frontal videos relative to motion cap- ture that could have influenced step length errors (this analysis could only be performed for the unimpaired participant dataset, in which motion capture and video recordings were syn- chronized). The timing of gait cycle detection differed depending on walking direction: when PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 12 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation the person walked away from the camera, gait cycle timings were, on average, four motion cap- ture frames (~0.04 s) before the timing detected from motion capture, and 15 motion capture frames (~0.15 s) after motion capture when the person walked toward the camera (S3 Fig, panel A). Using gait event timings from motion capture to calculate step lengths from frontal videos, there was a statistical difference in step length errors when the person walked away from the camera (P = 0.024), but not when the person walked toward the camera (P = 0.501; S3 Fig, panel B). The average step length error decreased from about 2 to 1 cm in the unim- paired participant dataset when using gait event timing from the motion capture data in the videos where the person walked away from the camera. Last, we considered that walking direction relative to the frontal plane camera may have influenced the accuracy of gait parameters. In the unimpaired participant dataset, in which two frontal plane cameras simultaneously captured the same walking trial from different van- tage points (see Fig 3A), we noted a minor overestimation of gait speed by an average of 0.04 m s−1 from the camera that the person walked away from compared to the camera that the person walked toward (S1 Table). We observed similar, albeit exaggerated, trends in the stroke and PD datasets. When comparing the average gait speed differences between motion capture and the frontal plane camera, gait speed was overestimated by 0.13 and 0.21 m s−1 for stroke and PD, respectively, when the person walked toward the frontal plane camera; the overes- timation was only minor at 0.01 and 0.03 m s−1 for stroke and PD, respectively, when the person walked toward the camera (S3 Table). The overestimation of gait speed was accompanied by greater errors when comparing the frontal camera to motion capture: average errors were 0.14 and 0.23 m s−1 for stroke and PD, respectively, when the person walked away from the camera; errors were only 0.06 and 0.08 m s−1 when the person walked toward the camera (S3 Table). The trends of overestimation and greater errors from frontal plane recordings where the person walked away from the camera were mirrored in the results of step length: there were greater overestimations and errors of step length when the person walked away from the cam- era (average overestimations of 0.056 and 0.082 m and errors of 0.084 and 0.092 m for stroke and PD, respectively) compared to when the person walked toward the camera (S3 Table; aver- age overestimations of 0.013 and 0.021 m and errors of 0.062 and 0.055 m for stroke and PD, respectively). This suggests that spatial gait parameters obtained from a frontal plane camera are influenced by walking direction and that the greatest precision was obtained when the per- son walked toward the camera. Furthermore, this also suggests that the accuracy of gait param- eters presented here, when calculated as session-level averages, can be improved if using only gait trials with the same walking direction. Discussion In this study, we demonstrated a new approach for performing clinical gait analyses using sim- ple videos recorded using low-cost devices and a workflow that leverages a freely available pose estimation algorithm (OpenPose) for video-based movement tracking. We showed that this novel approach can perform accurate gait analyses 1) from videos recorded from multiple perspectives (e.g., frontal or sagittal viewpoints), 2) across a diverse range of persons with and without gait impairment, 3) that capture clinically relevant and condition-specific aspects of gait, and 4) that measure within-participant changes in gait as a result of changes in walking speed. These findings demonstrate the versatility and accessibility of video-based gait analysis and have significant potential for clinical applications. Interest in video-based, markerless gait analysis has accelerated rapidly. Previous studies have used various approaches to move quantitative clinical gait analysis outside of the labora- tory or research center and directly into the home or clinic [5,6,13–15,17]. Here, we aimed to PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 13 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation develop a single approach that addressed several outstanding needs, including the needs to accommodate multiple different types of environments/viewing perspectives, use of datasets in multiple clinical populations with gait impairment, measurement of both spatiotemporal gait parameters and lower extremity two-dimensional kinematics, and measurement of within- participant changes in gait. It is also notable that we achieved accurate results using multiple different video recording devices with different sampling rates. By comparing our results against gold standard motion capture measurements, we provide data about the accuracy of all findings with respect to the current state-of-the-art. Our findings also enable us to progress toward development of a series of best practices for video-based clinical gait analysis. Unsurprisingly, we found that video-based gait analyses gen- erated from videos recorded using a sagittal viewpoint generally led to stronger correlations with motion capture data and lower error when compared to videos recorded from frontal viewpoints. This was particularly evident in gait parameters that require especially high levels of precision (e.g., step length asymmetry in persons post-stroke). Similar to our previous work [11], we also found that video-based measurements of ankle kinematics were generally less accurate (relative to motion capture) than measurements of hip or knee kinematics in persons with or without gait impairment. Therefore, when using the current iteration of our workflow, a user is likely to obtain best results by recording a sagittal video (if possible) and targeting measurement of spatiotemporal gait parameters and more proximal lower limb kinematics. We emphasize that our single-camera, video-based approach is not intended to reach marker- based motion capture levels of accuracy that other multi-camera approaches may target [6,18,19] or that may be required by various scientific disciplines (e.g., biomechanics, human motor control), but rather offers clinicians and other end-users access to a reasonably accurate approach for clinical gait analysis that requires minimal time and only a single video recording device. It is informative to consider the accuracy of our workflow relative to reported test-retest minimal detectable change or minimal clinical important difference values of the population of interest. For example, a meaningful change in gait speed is often reported as 0.10 m s−1 [20], but may vary from 0.05 up to 0.30 m s−1 depending on the population studied [21–32]. The average errors of our video-based measurements relative to motion capture generally fall within these margins, suggesting that gait speed is likely to be reliably measured in many popu- lations (e.g., older adults, post-stroke, PD, following hip fracture, cerebral palsy, multiple scle- rosis) using our workflow. Minimal detectable changes in gait kinematics may also be dependent on the population of interest, with estimates ranging from about 4˚ to 11˚ of lower- limb sagittal plane kinematics [26,28,33–36]. Average errors of sagittal plane hip and knee kinematics in our study were less than 4˚, while errors at the ankle were up to 6.8˚, suggesting that hip and knee kinematics from our workflow can be accurately tracked while continued improvement in measurement of ankle angles is needed. There remain additional significant hurdles to widespread implementation of video-based clinical gait analyses. There is a crucial unmet need for improved ease of use, as the user cur- rently must have access to specific computing hardware (i.e., pose estimation is most efficient when using a graphics processing unit (GPU)), download all relevant software, record the vid- eos, and manually process each video through the workflow. This generates an output that is contained within the software. This process is not well-suited for users without some level of technical expertise; there is an important need for new technologies that can streamline these steps and remove much of the technical know-how and burden of manual processing. Further- more, there is a need for validation in additional adult and pediatric clinical populations, as previous work has shown that existing pose estimation algorithms have difficulty with tracking patient populations with anatomical structures that likely differ significantly from the images PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 14 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation used to train the algorithms [13]. Thirteen of the participants with stroke used a cane; we did not observe instances where OpenPose mistakenly identified the cane as a limb. Lastly, it is likely that accuracy will continue to improve in the future as both computer vision algorithms and methods for data post-processing continue to advance. In this study, we used a pre-trained network [8], while a different network that was trained to be specific to both gait and clinical condition may further improve accuracy (the challenges of existing pre-trained networks for human pose estimation in movement science have been well-documented [37]). In this study, we developed and tested a novel approach for video-based clinical gait analy- sis. We showed that this approach accommodates multiple viewing perspectives, provides accurate and clinically relevant gait analyses (as compared to 3D motion capture) across multi- ple participant populations with and without gait impairment, and tracks within-participant changes in gait that are relevant to rehabilitation and recovery outcomes. All software needed to perform these analyses is freely available at https://github.com/janstenum/GaitAnalysis- PoseEstimation/tree/Multiple-Perspectives, where we also provide a series of detailed instruc- tions to assist the user. There is an urgent need to begin to move these emerging technologies with potential for significant clinical applications toward more user-friendly solutions. Materials and methods Participants We recruited 44 individuals post-stroke (15 female, 29 male; age 61±11 years (mean±SD); body mass 90±23 kg; height 1.73±0.11 m) and 19 individuals with PD (6 female, 13 male; age 67±7 years; body mass 77±14 kg; height 1.71±0.09 m) to participate in the study; all partici- pants were capable of walking independently with or without an assistive device. All partici- pants gave written informed consent before enrolling in the study in accordance with the protocol approved by The Johns Hopkins School of Medicine Institutional Review Board (Pro- tocol IRB00255175). Additionally, we used a publicly available dataset [38] of overground walking sequences from 32 unimpaired participants (10 women, 22 men) made available at http://bytom.pja.edu.pl/projekty/hm-gpjatk. The dataset included synchronized 3D motion capture files and digital video recordings of the walking sequences. The publicly available data- set does not contain identifiable participant information and faces have been blurred in the video recordings. Our analysis of the publicly available videos was deemed exempt by The Johns Hopkins University School of Medicine Institutional Review Board. Protocol and data collection Participants visited our laboratory for one day of testing. They first performed ten-meter walk tests at their preferred speed and the fastest speed at which they felt comfortable walking. Par- ticipants then performed eight overground walking trials (four trials at each preferred and fast speeds) across a walkway of 4.83 m. We mounted two commercially available tablets (Samsung Galaxy Tab A7) on tripods posi- tioned to capture frontal (CFront) and sagittal (CSag) plane views of the overground walking trials (video recordings occurred at a 30-Hz sampling rate; see Fig 1 for overview). Of the eight total walking trials, the participant walked away from the frontal plane camera with the left side turned to the sagittal plane camera during four of the trials; during the other four trials, the par- ticipant walked toward the frontal plane camera with the right side turned to the sagittal plane camera. Tablet cameras obtained videos with 1920 × 1080 pixel resolution. The frontal-view tab- let was positioned 1.52 m behind the start/end of the walkway and the sagittal-view tablet was positioned 3.89 m to the side of the midpoint of the walkway. The tablet positions were chosen to achieve the longest walkway in which the person remained visible to both frontal and sagittal PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 15 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation tablets, given the space restrictions of the laboratory. The frontal- and sagittal-view tablets were rotated to capture portrait and landscape views, respectively. The height of the frontal-view camera was set so that the entire participant remained visible when they were nearest the cam- era (about 0.85 m). The height of the sagittal-view camera was about 1.18 m so that the partici- pant appeared in the middle of the image as they travelled across the walkway. We simultaneously recorded walking trials using ten cameras (Vicon Vero, Denver, CO, USA) as part of a marker-based, 3D motion capture system at 100 Hz. We placed reflective markers on the seventh cervical vertebrae (C7), tenth thoracic vertebrae, jugular notch, xiphoid process, and bilaterally over the second and fifth metatarsal heads, calcaneus, medial and lateral malleoli, shank, medial and lateral femoral epicondyles, thigh, greater trochanter, iliac crest, and anterior and posterior superior iliac spines (ASIS and PSIS, respectively). In the previously published dataset of unimpaired adults without gait impairment, we used a subset of the data (sequences labelled s1) that consisted of a single walking bout of approxi- mately 5 m that included gait initiation and termination. We excluded data for one participant because the data belonged to another subset with diagonal walking sequences. We used data from two digital cameras (Basler Pilot piA1900-32gc, Ahrensburg, Germany) that simulta- neously recorded frontal plane views of the person walking away from one camera and toward the other camera (see Fig 3A for overview). The digital cameras obtained videos with 960 × 540 pixel resolution captured at 25 Hz. The average distance from the starting position of the participants to the cameras were 2.50 and 7.28 m for the camera that recorded the partic- ipant walking away and toward, respectively. Cameras were mounted on tripods and the height was about 1.3 m. Motion capture cameras (Vicon MX-T40, Denver, CO, USA) recorded 3D marker positions at 100 Hz. Markers were placed on the seventh cervical vertebrae, tenth thoracic vertebrae (T10), manubrium, sternum, right upper back and bilaterally on the front and back of the head, shoulder, upper arm, elbow, forearm, wrist (at radius and ulna), middle finger, ASIS, PSIS, thigh, knee, shank, ankle, heel, and toe. Data processing and analysis Motion capture data from the participants with stroke or PD were smoothed using a zero-lag 4th order low-pass Butterworth filter with a cutoff frequency of 7 Hz. The motion capture data from the participants without gait impairment in the publicly available dataset had already been smoothed. We identified left and right heel-strikes and toe-offs as the positive and negative peaks, respectively, of the anterior-posterior left or right ankle markers relative to the torso [39]. All digital video data were processed in two steps: 1) using OpenPose to automatically detect and label two-dimensional coordinates of various anatomical keypoints, 2) post-pro- cessing in MATLAB using custom-written code. The OpenPose analysis was similar for all video data, whereas we divided the post-processing workflows into two separate pipelines for videos capturing frontal or sagittal plane views. 1. OpenPose Analysis a. We ran the OpenPose demo over sequences of the video recordings that contained each walking bout. We have previously used a cloud-based service to run OpenPose with remote access to GPUs. Here we used a local computer with a GPU (NVIDIA GeForce RTX 3080) so that videos containing identifiable participant information were not shared with any third-party services. b. Videos were analyzed in OpenPose using the BODY_25 keypoint model that tracks the following 25 keypoints: nose, neck, mid-hip and bilateral keypoints at the eyes, ears, shoulders, elbows, wrists, hips, knees, ankles, heels, halluces, and fifth toes. PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 16 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation c. The output of the OpenPose analysis yielded: 1) JSON files for every video frame con- taining pixel coordinates of each keypoint detected in the frame, and 2) a new video file in which a stick figure that represents the detected keypoints is overlaid on the original video recording. 2. MATLAB Post-processing We created custom-written MATLAB code to process the JSON files that were output from the OpenPose analysis (https://github.com/janstenum/GaitAnalysis-PoseEstimation/tree/ Multiple-Perspectives). As an initial step, we checked whether multiple persons had been detected by OpenPose in the video (this can be the case when multiple people are visible or when OpenPose incorrectly detects keypoints in inanimate objects). Note that OpenPose has an optional flag to track only a single person; however, we did not use this option to avoid scenarios where the participant had not been tracked in favor of other persons (e.g., the experimenter). If multiple persons were detected, three MATLAB scripts were called that 1) required user input to identify the participant in a single frame of the video, 2) auto- matically identified the participant throughout the video and 3) allowed the user to visually inspect that the participant had been identified and correct any errors. Following the per- son-identification step, MATLAB workflows were different depending on whether the cam- era captured a frontal or sagittal plane view of the walking trial. We describe each workflow below. a. Frontal plane videos i. We changed the pixel coordinate system so that the positive vertical was directed upward and that positive horizontal was directed toward the participant’s left side. ii. We visually inspected and corrected errors in left-right identification of the limbs. In all, 362 (less than 1% of the 131,519 frames in total) frontal video frames were corrected. iii. We gap-filled keypoint trajectories using linear interpolation for gaps spanning to up 0.12 s. iv. We identified events of left and right gait cycles by local maxima and minima of the vertical distance between the left and right ankle keypoints. Gait events on the left limb were detected at positive peaks and gait events on the right limb were detected at nega- tive peaks in trials where the participants walked away from the frontal plane camera; and vice versa in trials where the participants walked toward the camera. In order to unify the nomenclature of gait events across motion capture data and sagittal and fron- tal plane video data, we refer to the gait events of the frontal plane analysis as heel- strikes. v. Last, we calculated a time-series of depth-change of the torso relative to the initial start- ing depth. We used the following equation to calculate depth-change (Δdi): Ddi ¼ dRef sRatio (cid:0) dRef; ðEq1Þ where dRef is the initial reference depth of the person relative to the frontal camera posi- tion and sRatio is the ratio of the pixel size of the person relative to the pixel size of the torso at the initial reference depth. Eq 1 is derived from trigonometric relations between the actual size of the person and the pixel size of the person as they appear on the image plane of the camera (see Fig 2A and 2B for an overview). We assume a fixed position of a pinhole camera with no lens distortion. We know the following relation when the PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 17 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation person is at an initial reference depth from the camera: sRef f ¼ s dRef ; ðEq2Þ where f is the focal length, sRef is the pixel size of the person at the reference distance and s is the actual size of the person. With a depth-change Δdi we obtain the following relationship: si f ¼ s dRef þ Ddi ; ðEq3Þ where si is the pixel size of the person as they appear with a depth-change. From Eqs 2 and 3, we obtain: sRatio ¼ si sRef ¼ dRef dRef þ Ddi : ðEq4Þ Using Eq 4 we obtain the expression in Eq 1. From Eq 1 we can estimate depth changes using only information about the reference depth of the person and the pixel size of the person. We validated this approach in Fig 2C by comparing the predicted value of sRatio based on Eq 4 (with a reference depth of 4.88 m) with values of sRatio found by manually tracking the pixel size of images of a person standing at depth-changes up to 18.29 m. The predicted relationship closely tracks the manually annotated pixel sizes in Fig 2C, suggesting that Eq 1 can be used to accurately calculate depth-changes in the frontal plane. Next, we considered methodological factors that may affect accuracy of the calculated depth-changes. We chose to track the size of the torso because there are only minor rotations in the transverse plane during the gait cycle, which ensures a consistent per- spective during a gait trial [40]. Torso size can be represented by 1) torso height (vertical distance between neck and midhip keypoints), 2) shoulder width (horizontal distance between left and right shoulder keypoints) and 3) the torso area (calculated as the square root of the product of torso height and shoulder width to ensure that size scales appro- priately with Eq 1). We evaluated the best tracking and smoothing method from the combination that yielded the lowest step length error and SD of step length differences between motion capture and frontal plane videos (See S4 Fig). Based on the evaluation, we chose to track torso size and low-pass filter size ratio using a cutoff frequency at 0.4 Hz. b. Sagittal plane videos i. We changed the pixel coordinate system so that positive vertical was direction upward and positive horizontal was the direction of travel. ii. We visually inspected and corrected errors in left-right identification of the limbs. In all, 5,369 (about 3.5% of the 153,669 frames in total) of sagittal video frames were corrected. iii. We gap-filled keypoint trajectories using linear interpolation for gaps spanning up to 0.12 s. iv. We smoothed trajectories using a zero-lag 4th order low-pass Butterworth filter with a cutoff frequency at 5 Hz. PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 18 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation v. We calculated a scaling factor to dimensionalize pixel distance. The scaling factor was as a ratio of a known distance in the line of progression relative to the pixel distance. We used the distance between strips of tape on the walkway. vi. We identified left and right heel-strikes and toe-offs as the positive and negative peaks, respectively, of the horizontal trajectories of the left or right ankle keypoints relative to the mid-hip keypoint. We cross-referenced gait events that had independently been identified in motion capture data and sagittal or frontal plane video data to ensure that all gait parameters were obtained based on the same gait cycles. We calculated the following spatiotemporal gait parameters: • Step time: duration between consecutive bilateral heel-strikes. • Step length (we used two methods to calculate step lengths): 1) as the horizontal distance between ankle markers or keypoints at instants of heel-strike and 2) as the distance travelled by the torso between consecutive bilateral heel-strikes. We used the distance travelled by the torso because the distances between the ankles at a heel-strike instant cannot be obtained from frontal plane videos. When comparing step lengths between motion capture and sagit- tal plane videos, we used the distance between the ankles; all step length comparisons with frontal plane data used the distance travelled by the torso. Step length methods were highly correlated (r = 0.938) with an average difference of −0.069 m, suggesting that the distance travelled by the torso was about 7 cm longer than the distance between the ankles (S5 Fig). • Gait speed: step length divided by step time. In stroke and PD data, we calculated paretic/non-paretic or left/right step times and step lengths, respectively. Paretic/left step time is the duration from non-paretic/right heel-strike until paretic/left heel-strike; vice versa for non-paretic/right step times. Paretic/left step length, calculated as the distance between the ankles, is the distance at paretic/left heel-strike; vice versa for non-paretic/right step lengths. Paretic/left step length, calculated as the distance trav- elled by the torso, is the distance travelled from non-paretic/right heel-strike to paretic/left heel-strike; vice versa for non-paretic/right step lengths. We calculated the changes in spatiotemporal gait parameters that accompany speed- changes (i.e., shorter step times, longer step lengths, and faster gait speeds) from the preferred and fast speed trials in the stroke and PD data. This allowed us to test how well gait changes can be tracked using video recordings. There are several commonly observed, clinically relevant gait impairments in stroke (e.g., gait asymmetry [41]) and PD (e.g., stooped posture [42])–thus, for each population we calcu- lated condition-specific gait parameters. We calculated step time asymmetry and step length asymmetry (difference between steps divided by sum of steps) in stroke gait and trunk inclina- tion in PD gait. Trunk inclination was calculated as the angle relative to vertical between the mid-hip and neck keypoints at heel-strikes in the sagittal plane videos and the angle between the C7 and right PSIS markers at heel-strikes in the motion capture data. During initial com- parisons we found an offset (mean±SD 12.0˚±1.5˚) between motion capture and sagittal plane video data; we subtracted a fixed offset of 12˚ from trunk inclination in the sagittal plane video data in order to create a better numeric comparison with the motion capture data. The offset is a consequence of the fact that video-based keypoints and markers track similar anatomical regions, but do not track the exact same anatomical locations [37,43]. We calculated sagittal plane lower limb joint kinematics at the hip, knee, and ankle using two-dimensional coordinates from the motion capture data and the sagittal plane video data. PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 19 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation We used markers at the greater trochanter and lateral femoral epicondyles and keypoints at the hip and knee to calculate hip angles; markers at the greater trochanter, lateral femoral epi- condyles and lateral malleoli and keypoints at the hip, knee, and ankle to calculate knee angles; markers at the lateral femoral epicondyles, lateral malleoli, and 5th metatarsal and keypoints at the knee, ankle, and hallux to calculate ankle angles. From our stroke and PD datasets, we compared gait parameters at three levels of compari- sons: at the step level calculating parameters for individual steps, as averages across single gait trials, and at the session level calculated as averages across several gait trials. In total there were 2,684 individual gait cycles (1,790 for stroke, 709 for PD and 185 for unimpaired), 527 gait tri- als (352 for stroke, 144 for PD and 31 for unimpaired) and 124 session level averages (88 for stroke and 36 for PD). We present session level gait parameters for stroke and PD and trial level for unimpaired data in the main text of the manuscript; we show results at the trial and step level in the S3 and S4 Tables. In the stroke and PD datasets, we compared gait parameters obtained during trials that were simultaneously recorded by motion capture, sagittal plane videos, and frontal plane vid- eos (see Fig 1 for overview). Note that some parameters (joint kinematics and trunk inclina- tion) can only be obtained with motion capture data and sagittal plane videos. In the dataset with unimpaired participants, we compared spatiotemporal gait parameters obtained during trials that were simultaneously captured with motion capture data and with two frontal cameras positioned to capture the participant walking away from one camera and toward the other camera (see Fig 3A for overview). Statistical analyses We compared gait parameters obtained with motion capture and video by calculating differ- ences, errors (absolute differences) and 95% limits of agreement (mean differences ± 1.96 × SD). We assessed correlations by calculating Pearson correlation coefficients. Supporting information S1 Fig. Step length errors and differences of frontal plane workflow relative to person’s dis- tance to camera. (PDF) S2 Fig. Step length errors and differences of frontal plane workflow relative to magnitude of step length. (PDF) S3 Fig. Influence of gait event timings on step length errors when using frontal plane work- flow. (PDF) S4 Fig. Evaluation of tracking methods and smoothing using frontal plane workflow. (PDF) S5 Fig. Comparison of two methods to calculate step length. (PDF) S1 Table. Comparison of spatiotemporal gait parameters of the unimpaired group. (PDF) PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 20 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation S2 Table. Spatiotemporal gait parameters for stroke and PD groups. (PDF) S3 Table. Comparison of spatiotemporal gait parameters of stroke and PD groups calcu- lated as trial averages. (PDF) S4 Table. Comparison of spatiotemporal gait parameters of stroke and PD groups calcu- lated for individual steps. (PDF) Author Contributions Conceptualization: Jan Stenum, Ryan T. Roemmich. Data curation: Jan Stenum, Melody M. Hsu. Formal analysis: Jan Stenum. Funding acquisition: Ryan T. Roemmich. Investigation: Jan Stenum. Methodology: Jan Stenum, Ryan T. Roemmich. Project administration: Ryan T. Roemmich. Resources: Alexander Y. Pantelyat. Software: Jan Stenum. Supervision: Ryan T. Roemmich. Visualization: Jan Stenum. Writing – original draft: Jan Stenum, Ryan T. Roemmich. Writing – review & editing: Jan Stenum, Melody M. Hsu, Alexander Y. Pantelyat, Ryan T. Roemmich. References 1. Olney S.J. & Richards C. Hemiparetic gait following stroke. Part I: Characteristics. Gait & Posture 4, 136–148 (1996). 2. Morris M.E., Iansek R., Matyas T.A. & Summers J.J. The pathogenesis of gait hypokinesia in Parkin- son’s disease. Brain 117, 1169–81 (1994). https://doi.org/10.1093/brain/117.5.1169 PMID: 7953597 3. Armand S., Decoulon G. & Bonnefoy-Mazure A. Gait analysis in children with cerebral palsy. EFORT Open Reviews 1, 448–460 (2016). https://doi.org/10.1302/2058-5241.1.000052 PMID: 28698802 4. Perry J. & Burnfield J. M. Gait Analysis: Normal and Pathological Function. 2nd ed. New Jersey: SLACK Inc. (2010). 5. Kidziński Ł., Yang B., Hicks J.L., Rajagopal A., Delp S.L. & Schwartz M.H. Deep neural networks enable quantitative movement analysis using single-camera videos. Nature Communications 11, 4054 (2020). https://doi.org/10.1038/s41467-020-17807-z PMID: 32792511 6. Uhlrich S.D., Falisse A., Kidziński Ł., Muccini J., Ko M., Chaudhari A.S., et al. OpenCap: Human move- ment dynamics from smartphone videos. PLoS Computational Biology 19, e1011462 (2023). https:// doi.org/10.1371/journal.pcbi.1011462 PMID: 37856442 7. Stenum J., Cherry-Allen K.M., Pyles C.O., Reetzke R.D., Vignos M.F. & Roemmich R.T. Applications of Pose Estimation in Human Health and Performance across the Lifespan. Sensors 21, 7315 (2021). https://doi.org/10.3390/s21217315 PMID: 34770620 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 21 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation 8. Cao Z., Hidalgo G., Simon T., Wei S.E. & Sheikh Y. OpenPose: Realtime Multi-Person 2D Pose Estima- tion Using Part Affinity Fields. IEEE transactions on pattern analysis and machine intelligence 43, 172– 186 (2021). https://doi.org/10.1109/TPAMI.2019.2929257 PMID: 31331883 9. Nath T., Mathis A., Chen A.C., Patel A., Bethge M. & Mathis M.W. Using DeepLabCut for 3D markerless pose estimation across species and behaviors. Nature Protocols 14, 2152–2176 (2019). https://doi.org/ 10.1038/s41596-019-0176-0 PMID: 31227823 10. Moro M., Pastore V.P., Tacchino C., Durand P., Blanchi I., Moretti P. et al. A markerless pipeline to ana- lyze spontaneous movements of preterm infants. Computer Methods and Programs in Biomedicine 226, 107119 (2022). https://doi.org/10.1016/j.cmpb.2022.107119 PMID: 36137327 11. Stenum J., Rossi C. & Roemmich R.T. Two-dimensional video-based analysis of human gait using pose estimation. PLoS computational biology 17, e1008935 (2021). https://doi.org/10.1371/journal. pcbi.1008935 PMID: 33891585 12. Needham L., Evans M., Wade L., Cosker D.P., McGuigan M.P., Bilzon J.L. et al. The development and evaluation of a fully automated markerless motion capture workflow. Journal of Biomechanics 144, 111338 (2022). https://doi.org/10.1016/j.jbiomech.2022.111338 PMID: 36252308 13. Cimorelli A., Patel A., Karakostas T. & Cotton R. J. Validation of portable in-clinic video-based gait anal- ysis for prosthesis users. Scientific Reports 14, 3840 (2024). https://doi.org/10.1038/s41598-024- 53217-7 PMID: 38360820 14. Lonini L., Moon Y., Embry K., Cotton R.J., McKenzie K., Jenz S. et al. Video-Based Pose Estimation for Gait Analysis in Stroke Survivors during Clinical Assessments: A Proof-of-Concept Study. Digital Bio- markers 6, 9–18 (2022). https://doi.org/10.1159/000520732 PMID: 35224426 15. Mehdizadeh S., Nabavi H., Sabo A., Arora T., Iaboni A. & Taati B. Concurrent validity of human pose tracking in video for measuring gait parameters in older adults: a preliminary analysis with multiple track- ers, viewing angles, and walking directions. Journal of NeuroEngineering and Rehabilitation 18, 139 (2021). https://doi.org/10.1186/s12984-021-00933-0 PMID: 34526074 16. Washabaugh E.P., Shanmugam T.A., Ranganathan R. & Krishnan C. Comparing the accuracy of open-source pose estimation methods for measuring gait kinematics. Gait & Posture 97, 188–195 (2022). https://doi.org/10.1016/j.gaitpost.2022.08.008 PMID: 35988434 17. Sabo A., Mehdizadeh S., Iaboni A., & Taati B. Estimating Parkinsonism Severity in Natural Gait Videos of Older Adults With Dementia. IEEE Journal of Biomedical and Health Informatics 26, 2288–2298 (2022). https://doi.org/10.1109/JBHI.2022.3144917 PMID: 35077373 18. Kanko R.M., Laende E.K., Strutzenberger G., Brown M., Selbie W.S., DePaul V. et al. Assessment of spatiotemporal gait parameters using a deep learning algorithm-based markerless motion capture sys- tem. Journal of Biomechanics 122, 110414 (2021). https://doi.org/10.1016/j.jbiomech.2021.110414 PMID: 33915475 19. Kanko R.M., Laende E.K., Davis E.M., Selbie W.S. & Deluzio K.J. Concurrent assessment of gait kine- matics using marker-based and markerless motion capture. Journal of Biomechanics 127, 110665 (2021). https://doi.org/10.1016/j.jbiomech.2021.110665 PMID: 34380101 20. Studenski S., Perera S., Patel K., Rosano C., Faulkner K., Inzitari M. et al. Gait Speed and Survival in Older Adults. JAMA 305, 50–58 (2011). https://doi.org/10.1001/jama.2010.1923 PMID: 21205966 21. Hass C.J., Bishop M., Moscovich M., Stegemo¨ ller E.L., Skinner J., Malaty I.A. et al. Defining the Clini- cally Meaningful Difference in Gait Speed in Persons With Parkinson Disease. Journal of Neurologic Physical Therapy 38, 233–238 (2014). https://doi.org/10.1097/NPT.0000000000000055 PMID: 25198866 22. Perera S., Mody S.H., Woodman R.C. & Studenski S.A. Meaningful Change and Responsiveness in Common Physical Performance Measures in Older Adults. Journal of the American Geriatrics Society 54, 743–749 (2006). https://doi.org/10.1111/j.1532-5415.2006.00701.x PMID: 16696738 23. Goldberg A. & Schepens S. Measurement error and minimum detectable change in 4-meter gait speed in older adults. Aging Clinical and Experimental Research 23, 406–412 (2011). https://doi.org/10.1007/ BF03325236 PMID: 22526072 24. Palombaro K.M., Craik R.L., Mangione K.K. & Tomlinson J.D. Determining Meaningful Changes in Gait Speed After Hip Fracture. Physical Therapy 86, 809–816 (2006). PMID: 16737406 25. Tilson J.K., Sullivan K.J., Cen S.Y., Rose D.K., Koradia C.H., Azen S.P. et al. Meaningful Gait Speed Improvement During the First 60 Days Poststroke: Minimal Clinically Important Difference. Physical Therapy 90, 196–208 (2010). https://doi.org/10.2522/ptj.20090079 PMID: 20022995 26. Kesar T.M., Binder-Macleod S.A., Hicks G.E. & Reisman D.S. Minimal detectable change for gait vari- ables collected during treadmill walking in individuals post-stroke. Gait & Posture 33, 314–317 (2011). https://doi.org/10.1016/j.gaitpost.2010.11.024 PMID: 21183350 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 22 / 23 PLOS DIGITAL HEALTHClinical gait analysis using video-based pose estimation 27. Lewek M.D. & Sykes R.I. Minimal Detectable Change for Gait Speed Depends on Baseline Speed in Individuals With Chronic Stroke. Journal of Neurologic Physical Therapy 43, (2019). https://doi.org/10. 1097/NPT.0000000000000257 PMID: 30702510 28. Geiger M., Supiot A., Pradon D., Do M.-C., Zory R. & Roche N. Minimal detectable change of kinematic and spatiotemporal parameters in patients with chronic stroke across three sessions of gait analysis. Human Movement Science 64, 101–107 (2019). https://doi.org/10.1016/j.humov.2019.01.011 PMID: 30710860 29. Andreopoulou G., Mahad D.J., Mercer T.H. & van der Linden M.L. Test-retest reliability and minimal detectable change of ankle kinematics and spatiotemporal parameters in MS population. Gait & Posture 74, 218–222 (2019). https://doi.org/10.1016/j.gaitpost.2019.09.015 PMID: 31561120 30. 31. Levin I., Lewek M.D., Giuliani C., Faldowski R. & Thorpe D.E. Test-retest reliability and minimal detect- able change for measures of balance and gait in adults with cerebral palsy. Gait & Posture 72, 96–101 (2019). https://doi.org/10.1016/j.gaitpost.2019.05.028 PMID: 31177021 Lang J.T., Kassan T.O., Devaney L.L., Colon-Semenza C. & Joseph M.F. Test-Retest Reliability and Minimal Detectable Change for the 10-Meter Walk Test in Older Adults With Parkinson’s disease. Jour- nal of Geriatric Physical Therapy 39, 165–170 (2016). https://doi.org/10.1519/JPT.0000000000000068 PMID: 26428902 32. Strouwen C., Molenaar E.A.L.M., Keus S.H.J., Mu¨ nks L., Bloem B.R. & Nieuwber A. Test-Retest Reli- ability of Dual-Task Outcome Measures in People With Parkinson Disease. Physical Therapy 96, 1276–1286 (2016). https://doi.org/10.2522/ptj.20150244 PMID: 26847010 33. McGinley J.L., Baker R., Wolfe R. & Morris M.E. The reliability of three-dimensional kinematic gait mea- surements: A systematic review. Gait & Posture 29, 360–369 (2009). https://doi.org/10.1016/j.gaitpost. 2008.09.003 PMID: 19013070 34. Fernandes R., Armada-da-Silva P., Pool-Goudaazward A., Moniz-Pereira V. & Veloso A.P. Three dimensional multi-segmental trunk kinematics and kinetics during gait: Test-retest reliability and mini- mal detectable change. Gait & Posture 46, 18–25 (2016). 35. Wilken J.M., Rodriguez K.M., Brawner M. & Darter B.J. Reliability and minimal detectible change values for gait kinematics and kinetics in healthy adults. Gait & Posture 35, 301–307 (2012). https://doi.org/10. 1016/j.gaitpost.2011.09.105 PMID: 22041096 36. Meldrum D., Shouldice C., Conroy R., Jones K. & Forward M. Test–retest reliability of three dimensional gait analysis: Including a novel approach to visualising agreement of gait cycle waveforms with Bland and Altman plots. Gait & Posture 39, 265–271 (2014). https://doi.org/10.1016/j.gaitpost.2013.07.130 PMID: 24139682 37. Seethapathi N., Wang S., Saluja R., Blohm G. & Kording K.P. Movement science needs different pose tracking algorithms. arXiv, 10.48550/arXiv.1907.10226 (2019). 38. Kwolek B., Michalczuk A., Krzeszowski T., Switonski A., Josinski H. & Wojciechowski K. Calibrated and synchronized multi-view video and motion capture dataset for evaluation of gait recognition. Multimedia Tools and Applications 78, 32437–32465 (2019). 39. Zeni J.A., Richards J.G. & Higginson J.S. Two simple methods for determining gait events during tread- mill and overground walking using kinematic data. Gait & Posture 27, 710–714 (2008). https://doi.org/ 10.1016/j.gaitpost.2007.07.007 PMID: 17723303 40. Chung C., Park M., Lee S., Kong S. & Lee K. Kinematic aspects of trunk motion and gender effect in normal adults. Journal of NeuroEngineering and Rehabilitation 7, 9 (2010). https://doi.org/10.1186/ 1743-0003-7-9 PMID: 20156364 41. Patterson K.K., Parafianowicz I., Danells C.J., Closson V., Verrier M.C., Staines W.R. et al. Gait Asym- metry in Community-Ambulating Stroke Survivors. Archives of Physical Medicine and Rehabilitation 89, 304–10 (2008). https://doi.org/10.1016/j.apmr.2007.08.142 PMID: 18226655 42. Termoz N., Halliday S.E., Winter D.A., Frank J.S., Patla A.E. & Prince F. The control of upright stance in young, elderly and persons with Parkinson’s disease. Gait & Posture 27, 463–470 (2008). https://doi. org/10.1016/j.gaitpost.2007.05.015 PMID: 17644337 43. Cronin N. J. Using deep neural networks for kinematic analysis: Challenges and opportunities. Journal of Biomechanics 123, 110460 (2021). https://doi.org/10.1016/j.jbiomech.2021.110460 PMID: 34029787 PLOS Digital Health \| https://doi.org/10.1371/journal.pdig.0000467 March 26, 2024 23 / 23 PLOS DIGITAL HEALTH
10.1126_science.abn5887	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 May 15. Published in final edited form as: Science. 2023 April 28; 380(6643): eabn5887. doi:10.1126/science.abn5887. This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding authors. [email protected] (KMM). §Consortium authors and affiliations listed at end of manuscript. Author contributions: Conceptualization: HJH, GS, EKK, BS Data Acquisition: KLM, HJH, BS, GS Analysis: KLM, HJH, KM, MSW, XL, KS, EKK Writing: KLM, HJH, KM, XL, EKK, BS Competing interests: Authors declare no competing interests. Consortium list: Gregory Andrews1, Joel C. Armstrong2, Matteo Bianchi3, Bruce W. Birren4, Kevin R. Bredemeyer5, Ana M. Breit6, Matthew J. Christmas3, Hiram Clawson2, Joana Damas7, Federica Di Palma8,9, Mark Diekhans2, Michael X. Dong3, Eduardo Eizirik10, Kaili Fan1, Cornelia Fanter11, Nicole M. Foley5, Karin Forsberg-Nilsson12,13, Carlos J. Garcia14, John Gatesy15, Steven Gazal16, Diane P. Genereux4, Linda Goodman17, Jenna Grimshaw14, Michaela K. Halsey14, Andrew J. Harris5, Glenn Hickey18, Michael Hiller19,20,21, Allyson G. Hindle11, Robert M. Hubley22, Graham M. Hughes23, Jeremy Johnson4, David Juan24, Irene M. Kaplow25,26, Elinor K. Karlsson1,4,27, Kathleen C. Keough17,28,29, Bogdan Kirilenko19,20,21, Klaus-Peter Koepfli30,31,32, Jennifer M. Korstian14, Amanda Kowalczyk25,26, Sergey V. Kozyrev3, Alyssa J. Lawler4,26,33, Colleen Lawless23, Thomas Lehmann34, Danielle L. Levesque6, Harris A. Lewin7,35,36, Xue Li1,4,37, Abigail Lind28,29, Kerstin Lindblad-Toh3,4, Ava Mackay-Smith38, Voichita D. Marinescu3, Tomas Marques-Bonet39,40,41,42, Victor C. Mason43, Jennifer R. S. Meadows3, Wynn K. Meyer44, Jill E. Moore1, Lucas R. Moreira1,4, Diana D. Moreno-Santillan14, Kathleen M. Morrill1,4,37, Gerard Muntané24, William J. Murphy5, Arcadi Navarro39,41,45,46, Martin Nweeia47,48,49,50, Sylvia Ortmann51, Austin Osmanski14, Benedict Paten2, Nicole S. Paulat14, Andreas R. Pfenning25,26, BaDoi N. Phan25,26,52, Katherine S. Pollard28,29,53, Henry E. Pratt1, David A. Ray14, Steven K. Reilly38, Jeb R. Rosen22, Irina Ruf54, Louise Ryan23, Oliver A. Ryder55,56, Pardis C. Sabeti4,57,58, Daniel E. Schäffer25, Aitor Serres24, Beth Shapiro59,60, Arian F. A. Smit22, Mark Springer61, Chaitanya Srinivasan25, Cynthia Steiner55, Jessica M. Storer22, Kevin A. M. Sullivan14, Patrick F. Sullivan62,63, Elisabeth Sundström3, Megan A. Supple59, Ross Swofford4, Joy-El Talbot64, Emma Teeling23, Jason Turner-Maier4, Alejandro Valenzuela24, Franziska Wagner65, Ola Wallerman3, Chao Wang3, Juehan Wang16, Zhiping Weng1, Aryn P. Wilder55, Morgan E. Wirthlin25,26,66, James R. Xue4,57, Xiaomeng Zhang4,25,26 Affiliations: 1Program in Bioinformatics and Integrative Biology, UMass Chan Medical School; Worcester, MA 01605, USA. 2Genomics Institute, University of California Santa Cruz; Santa Cruz, CA 95064, USA. 3Department of Medical Biochemistry and Microbiology, Science for Life Laboratory, Uppsala University; Uppsala, 751 32, Sweden. 4Broad Institute of MIT and Harvard; Cambridge, MA 02139, USA. 5Veterinary Integrative Biosciences, Texas A&M University; College Station, TX 77843, USA. 6School of Biology and Ecology, University of Maine; Orono, ME 04469, USA. 7The Genome Center, University of California Davis; Davis, CA 95616, USA. 8Genome British Columbia; Vancouver, BC, Canada. 9School of Biological Sciences, University of East Anglia; Norwich, UK. 10School of Health and Life Sciences, Pontifical Catholic University of Rio Grande do Sul; Porto Alegre, 90619–900, Brazil. 11School of Life Sciences, University of Nevada Las Vegas; Las Vegas, NV 89154, USA. 12Biodiscovery Institute, University of Nottingham; Nottingham, UK. 13Department of Immunology, Genetics and Pathology, Science for Life Laboratory, Uppsala University; Uppsala, 751 85, Sweden. 14Department of Biological Sciences, Texas Tech University; Lubbock, TX 79409, USA. 15Division of Vertebrate Zoology, American Museum of Natural History; New York, NY 10024, USA. 16Keck School of Medicine, University of Southern California; Los Angeles, CA 90033, USA. 17Fauna Bio Incorporated; Emeryville, CA 94608, USA. 18Baskin School of Engineering, University of California Santa Cruz; Santa Cruz, CA 95064, USA. 19Faculty of Biosciences, Goethe-University; 60438 Frankfurt, Germany. 20LOEWE Centre for Translational Biodiversity Genomics; 60325 Frankfurt, Germany. 21Senckenberg Research Institute; 60325 Frankfurt, Germany. 22 Institute for Systems Biology; Seattle, WA 98109, USA. 23School of Biology and Environmental Science, University College Dublin; Belfield, Dublin 4, Ireland. 24Department of Experimental and Health Sciences, Institute of Evolutionary Biology (UPF-CSIC), Universitat Pompeu Fabra; Barcelona, 08003, Spain. 25Department of Computational Biology, School of Computer Science, Carnegie Mellon University; Pittsburgh, PA 15213, USA. 26Neuroscience Institute, Carnegie Mellon University; Pittsburgh, PA 15213, USA. 27Program in Molecular Medicine, UMass Chan Medical School; Worcester, MA 01605, USA. 28Department of Epidemiology & Biostatistics, University of California San Francisco; San Francisco, CA 94158, USA. 29Gladstone Institutes; San Francisco, CA 94158, USA. 30Center for Species Survival, Smithsonian’s National Zoo and Conservation Biology Institute; Washington, DC 20008, USA. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 2 Comparative genomics of Balto, a famous historic dog, captures lost diversity of 1920s sled dogs Katherine L. Moon1,2,, Heather J. Huson3, Kathleen Morrill4,5,6, Ming-Shan Wang1,2, Xue Li4,5,6, Krishnamoorthy Srikanth3, Zoonomia Consortium§, Kerstin Lindblad-Toh6,7, Gavin J. Svenson8, Elinor K. Karlsson4,5, Beth Shapiro1,2 1 Department of Ecology and Evolutionary Biology, University of California Santa Cruz, Santa Cruz, CA, USA 2 Howard Hughes Medical Institute, University of California Santa Cruz, Santa Cruz, CA, USA 3 Department of Animal Sciences, Cornell University College of Agriculture and Life Sciences, Ithaca, NY, 14853, USA 31Computer Technologies Laboratory, ITMO University; St. Petersburg 197101, Russia. 32Smithsonian-Mason School of Conservation, George Mason University; Front Royal, VA 22630, USA. 33Department of Biological Sciences, Mellon College of Science, Carnegie Mellon University; Pittsburgh, PA 15213, USA. 34Senckenberg Research Institute and Natural History Museum Frankfurt; 60325 Frankfurt am Main, Germany. 35Department of Evolution and Ecology, University of California Davis; Davis, CA 95616, USA. 36John Muir Institute for the Environment, University of California Davis; Davis, CA 95616, USA. 37Morningside Graduate School of Biomedical Sciences, UMass Chan Medical School; Worcester, MA 01605, USA. 38Department of Genetics, Yale School of Medicine; New Haven, CT 06510, USA. 39Catalan Institution of Research and Advanced Studies (ICREA); Barcelona, 08010, Spain. 40CNAG-CRG, Centre for Genomic Regulation, Barcelona Institute of Science and Technology (BIST); Barcelona, 08036, Spain. 41Department of Medicine and LIfe Sciences, Institute of Evolutionary Biology (UPF-CSIC), Universitat Pompeu Fabra; Barcelona, 08003, Spain. 42 Institut Català de Paleontologia Miquel Crusafont, Universitat Autònoma de Barcelona; 08193, Cerdanyola del Vallès, Barcelona, Spain. 43 Institute of Cell Biology, University of Bern; 3012, Bern, Switzerland. 44Department of Biological Sciences, Lehigh University; Bethlehem, PA 18015, USA. 45BarcelonaBeta Brain Research Center, Pasqual Maragall Foundation; Barcelona, 08005, Spain. 46CRG, Centre for Genomic Regulation, Barcelona Institute of Science and Technology (BIST); Barcelona, 08003, Spain. 47Department of Comprehensive Care, School of Dental Medicine, Case Western Reserve University; Cleveland, OH 44106, USA. 48Department of Vertebrate Zoology, Canadian Museum of Nature; Ottawa, Ontario K2P 2R1, Canada. 49Department of Vertebrate Zoology, Smithsonian Institution; Washington, DC 20002, USA. 50Narwhal Genome Initiative, Department of Restorative Dentistry and Biomaterials Sciences, Harvard School of Dental Medicine; Boston, MA 02115, USA. 51Department of Evolutionary Ecology, Leibniz Institute for Zoo and Wildlife Research; 10315 Berlin, Germany. 52Medical Scientist Training Program, University of Pittsburgh School of Medicine; Pittsburgh, PA 15261, USA. 53Chan Zuckerberg Biohub; San Francisco, CA 94158, USA. 54Division of Messel Research and Mammalogy, Senckenberg Research Institute and Natural History Museum Frankfurt; 60325 Frankfurt am Main, Germany. 55Conservation Genetics, San Diego Zoo Wildlife Alliance; Escondido, CA 92027, USA. 56Department of Evolution, Behavior and Ecology, School of Biological Sciences, University of California San Diego; La Jolla, CA 92039, USA. 57Department of Organismic and Evolutionary Biology, Harvard University; Cambridge, MA 02138, USA. 58Howard Hughes Medical Institute; Chevy Chase, MD, USA. 59Department of Ecology and Evolutionary Biology, University of California Santa Cruz; Santa Cruz, CA 95064, USA. 60Howard Hughes Medical Institute, University of California Santa Cruz; Santa Cruz, CA 95064, USA. 61Department of Evolution, Ecology and Organismal Biology, University of California Riverside; Riverside, CA 92521, USA. 62Department of Genetics, University of North Carolina Medical School; Chapel Hill, NC 27599, USA. 63Department of Medical Epidemiology and Biostatistics, Karolinska Institutet; Stockholm, Sweden. 64 Iris Data Solutions, LLC; Orono, ME 04473, USA. 65Museum of Zoology, Senckenberg Natural History Collections Dresden; 01109 Dresden, Germany. 66Allen Institute for Brain Science; Seattle, WA 98109, USA Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 3 4 Bioinformatics and Integrative Biology, UMass Chan Medical School, Worcester, MA 01655, USA 5 Morningside Graduate School of Biomedical Sciences, UMass Chan Medical School, Worcester, MA 01655, USA 6 Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA 7 Department of Medical Biochemistry and Microbiology, Science for Life Laboratory, Uppsala University; Uppsala, 751 32, Sweden. 8 Cleveland Museum of Natural History, Cleveland, OH 44106, USA Abstract We reconstruct the phenotype of Balto, the heroic sled dog renowned for transporting diphtheria antitoxin to Nome, Alaska in 1925, using evolutionary constraint estimates from the Zoonomia alignment of 240 mammals and 682 genomes from dogs and wolves of the 21st century. Balto shares just part of his diverse ancestry with the eponymous Siberian husky breed. Balto’s genotype predicts a combination of coat features atypical for modern sled dog breeds, and a slightly smaller stature. He had enhanced starch digestion compared with Greenland sled dogs and a compendium of derived homozygous coding variants at constrained positions in genes connected to bone and skin development. We propose that Balto’s population of origin, which was less inbred and genetically healthier than modern breeds, was adapted to the extreme environment of 1920s Alaska. One-Sentence Summary: Comparative genomics uncovers genotype-phenotype links between Balto, famed sled dog of the 1925 Serum Run, and modern dogs. Technological advances in the recovery of ancient DNA make it possible to generate high-coverage nuclear genomes from historic and fossil specimens, but interpreting genetic data from past individuals is difficult without data from their contemporaries. Comparative genomic analysis offers a solution: by combining population-level genomic data and catalogs of trait associations in modern populations, we can infer the genetic and phenotypic features of long-dead individuals and the populations from which they were born. Zoonomia is a new comparative resource that addresses limitations of previous datasets (1) to support interpretation of paleogenomics data. With 240 placental mammal species, Zoonomia has sufficient power to distinguish individual bases under evolutionary constraint - a useful predictor of functional importance (2) - in coding and regulatory elements (3). Zoonomia’s reference-free genome alignment (4, 5) allows evolutionary constraint to be scored in any of its 240 species, including dogs. Here, we generate a genome for Balto, the famous sled dog who delivered diphtheria serum to the children of Nome, Alaska, during a 1925 outbreak. Following his death, Balto was taxidermied and his remains are held by the Cleveland Museum of Natural History. We generated a 40.4-fold coverage nuclear genome from Balto’s underbelly skin using protocols for degraded samples. His DNA was well preserved, with an average endogenous content of Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 4 87.7% in sequencing libraries, low (<1%) damage rates (fig. S1) and short (68bp) average fragment sizes, consistent with the age of the sample. Balto was born in the kennel of sled dog breeder Leonard Seppala in 1919. Although Seppala’s small fast dogs were known as Siberian huskies (6), they were a working population that differed from the dog breed recognized by the American Kennel Club (AKC) today. Modern dog breeds are genetically closed populations that conform to a tightly delineated physical standard (7). Balto’s relationship to AKC-recognized sled dog breeds like the Siberian husky (established in 1930) and Alaskan malamute (1935) (8) is unclear. Balto himself was neutered at six months of age and had no offspring. Working populations of sled dogs survive. Alaskan sled dogs are bred solely for physical performance, including outcrossing with various breeds (9). Greenland sled dogs are an indigenous land-race breed that have been used for hunting and sledging by Inuit in Greenland for 850 years, where they have been isolated from contact with other dogs (10). Here, we use the term “breed” exclusively to refer to modern breeds recognized by the AKC or other kennel clubs (e.g. sled dog breeds), as distinct from the less rigidly defined populations of Greenland sled dogs and Alaskan sled dogs (working sled dogs). This is a genetic distinction; AKC-registered dogs can be successful working sled dogs. We compared Balto to working sled dogs, sled dog breeds, other breeds, village dogs (free- breeding dogs without known breed ancestry), and other canids. Our whole genome dataset comprised 688 dogs (table S1) representing 135 breeds/populations, including three Alaskan sled dogs and five Greenland sled dogs (10). We identified evolutionarily constrained bases using phyloP evolutionary constraint scores from the dog-referenced version of the 240 species Zoonomia alignment (3). Ancestry analysis places Balto in a clade of sled dog breeds and working sled dogs and closest to the Alaskan sled dogs (Fig. 1A,B). Most of his ancestry is assigned to clades of Arctic-origin dogs (68%) and, to a lesser extent, Asian-origin dogs (24%) in an unsupervised admixture analysis with 2166 dogs and 116 clusters (Fig. 1C, table S2, S3). He carried no discernible wolf ancestry. The more recently established Alaskan sled dog population (9) did not fall into a distinct ancestry cluster in the unsupervised analysis, but comprised 34% of Balto’s ancestry in a supervised analysis defining them as a cluster (fig. S2). Balto was more genetically diverse than breed dogs today and similar to working sled dogs (Fig. 1D). Balto had shorter runs of homozygosity than any breed dog, and fewer runs of homozygosity than all but one Tibetan mastiff (table S4). When inbreeding is calculated from runs of homozygosity, Balto and the two working sled dog populations are lower than almost any breed dog (fig. S3). When inbreeding is calculated using an allele frequency approach (method-of-moment), Greenland sled dogs have high inbreeding coefficients, reflecting their long genetic isolation in Greenland (fig. S3). To evaluate the genetic health of Balto’s population of origin, we developed an analytical approach that leveraged the Zoonomia 240 species constraint scores and required only a single dog from each population (necessary since Balto is the only available representative of his population). Briefly, we selected one individual at random from each breed or Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 5 population (57 dogs total) and scored variant positions as either evolutionarily constrained (and more likely to be damaging (2)) or not using the Zoonomia phyloP scores (3). We also identified variants likely to be “rare” (low frequency) in each dog’s breed or population. Because we couldn’t directly measure population allele frequencies with only a single representative dog, we defined “rare” variants as heterozygous or homozygous variants unique to that dog among all 57 representative dogs. This metric effectively identifies variants occurring at unusually low frequencies (fig. S4). Balto and modern working sled dogs had a lower burden of rare, potentially damaging variation, indicating they represent genetically healthier populations (11) than breed dogs. Balto and the working sled dogs had significantly fewer potentially damaging variants (missense or constrained) than any breed dog, including the sled dog breeds (Fig. 1E). The pattern persists even in the less genetically diverse Greenland sled dog. Selection for fitness in working sled dog populations appears more effective in removing damaging genetic variation than selection to meet a breed standard. Balto’s physical appearance predicted from his genome sequence (Fig. 2A, table S5) matches historical photos (Fig. 2B) and his taxidermied remains, indicating that the same variants shaping modern breed phenotypes also explained natural variation in his pre-breed working population. We predict that he stood 55cm tall at his shoulders (12)(Fig. 2C), within the acceptable range for today’s Siberian husky breed (53–60cm (8)), and had a double layered coat (13) that was mostly black with only a little bit of white (14). He was homozygous for an allele conferring tan points (15) and one for blue eyes (16), but both were masked by his melanistic facial mask (17), and his predicted light-tan pigmentation (18) may have been indistinguishable from white. He carried neither the “wolf agouti” nor “Northern domino” patterns that are common in the Siberian husky and other sled dog breeds today (19). Both Balto and Alaskan sled dogs had unexpected evidence of adaptation to starch-rich diets. They carry the dog version of MGAM, a gene involved in starch processing that is differentiated between dogs and wolves (20) and one of fourteen regions analyzed for evidence of selective pressure in Balto’s lineage using a gene tree analysis (table S6). In earlier work, the high frequency of the wolf version of MGAM in Greenland sled dogs prompted speculation that reduced starch digestion might be a working sled dog trait (10). Our findings suggest this phenomenon is specific to Greenland sled dogs. Gene tree analysis places one of Balto’s chromosomes in the ancestral wolf cluster, and one to the derived dog cluster (fig. S5). Most Alaskan sled dogs carry the dog version (frequency=0.83). However, read coverage of the gene AMY2B suggests Balto had fewer copies of this gene than many modern dogs, and thus comparatively lower production of the starch-digesting enzyme amylase (21, 22). Taken together, we suggest Balto’s ability to digest starch was enhanced compared to wolves and Greenland sled dogs, but reduced compared to modern breeds. Of the other 14 regions tested, most (10/14) lacked sufficient diversity in dogs to resolve phylogenetic relationships. Bootstrap support was weak for two other genes selected in Greenland sled dogs (CACNA1A and MAGI2). As expected, Balto did not carry versions of EPAS1 associated with high altitude adaptation (23). Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 6 We found an enrichment for unusual function variation in Balto’s population consistent with adaptation to the extreme environments in which early 20th century sled dogs worked. We identified variants in Balto’s genome that were new (not seen in wolves) and likely to be common in his population (homozygous in Balto; fig. S4). We further filtered for variants that were both protein-altering (missense) and evolutionarily constrained (FDR<0.01), and thus likely to be functional. Balto was no more likely to carry such variants than dogs from 54 other populations (fig. S6), but in Balto these variants tended to disrupt tissue development genes (GO:0009888; 24 genes; 3.02-fold enrichment; pFDR=0.013)(table S7). This enrichment was unique to Balto (Fig. 2D, fig. S7), and most of the variants were rare or missing in other dog populations (fig. S8). Even when all GO biological process gene sets are tested in all 57 dogs, Balto’s enrichment in tissue development genes is highly unusual. It ranks 4th out of 888,573 dog/set pairs tested (fig. S7, table S8). Phenotype associations from human disease studies suggest that these variants could have influenced skeletal and epithelial development including joint formation, body weight, coordination, and skin thickness (table S9)(24). Modern sled dog breeds and working sled dogs are only slightly more similar to Balto than other dogs at these variants (fig. S9). Balto was part of a famed population of small, fast, and fit sled dogs imported from Siberia. Following his famous run, the Siberian husky breed was recognized by the AKC. By sequencing his genome from his taxidermied remains and analyzing it in the context of large comparative and canine datasets, we show that Balto shared only part of his ancestry with today’s Siberian huskies. Balto’s working sled dog contemporaries were healthier and more genetically diverse than modern breeds, and may have carried variants that helped them survive the harsh conditions of 1920s Alaska (6). Further work is still needed to assess the impact of the evolutionarily constrained missense variants that Balto carried. While the era of Balto and his fellow huskies has passed, comparative genomics, supported by a growing collection of modern and past genomes, can provide a snapshot of individuals and populations from the past, as well as insights into the selective pressures that shaped them. Materials and Methods: Assembly of comparative canid genetic variants We collated a reference set of comparative canid genetic variants starting from the curated Broad-UMass Canid Variant set (https://data.broadinstitute.org/DogData/) and comprising whole genome sequencing data for 531 dogs of known breed ancestry distributed among 132 breeds, 28 dogs of mixed breed ancestry, 12 dogs of unknown ancestry, 69 worldwide indigenous or village dogs, 33 wolves, and 1 coyote (see stable S1). Ancient DNA extraction, library preparation, and genome assembly We extracted DNA from a ~5mm × 5mm piece of Balto’s underbelly skin tissue, in two replicates (HM246 and HM247) with an extraction negative, using the ancient DNA specific protocol in Dabney et al. 2013 (28). We prepared 32 ~1pmol input Illumina libraries from these extracts following the Santa Cruz library preparation method (29), including positive and negative controls. All 32 libraries passed quality control (QC), and so we sequenced Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 7 them to a depth of ~2.3 billion on a NovaSeq 6000 platform 150bp paired end (see table S11 for the number of reads produced per library). We used SeqPrep v.1.1 (30) to trim adapters, remove reads shorter than 28bp, and merge remaining paired-end reads with a minimum overlap of 15 bp. We then used the Burrows-Wheeler Aligner (BWA) v.0.7.12 (31) with a minimum quality cut off of 20 to align reads to the Canis lupus familiaris (dog) reference genome (CanFam3.1) (NCBI: GCA_000002285.2). All 32 bam files (one for each library) were merged into one with PCR duplicates removed. We used both Qualimap (v2.2.1) and samtools (v1.7) to calculate metrics and assess the quality of the alignment (see table S12). Variant calling We used GATK HaplotypeCaller to call variants in Balto as well as 10 previously published Greenland sled dogs (10) and 3 Alaskan sled dogs sequenced for this study (see Supplementary Methods for details on sampling, DNA extraction, and sequencing) against the UMass-Broad Canid Variant set using parameter --genotyping-mode GENOTYPE_GIVEN_ALLELES --alleles (known alleles). Then, we merged variant call records from these 14 dogs with records from the UMass-Broad Candid Variants set, for variant calls in a full set of 688 individuals: Balto (this study), 3 modern Alaskan sled dogs (this study), 10 modern Greenland sled dogs (10), 531 dogs from modern breeds, 40 dogs of unknown or admixed ancestry, 69 village or indigenous dogs, 33 wolves, and 1 coyote. Phylogenetic analysis and neighbor-joining trees Using a dataset of 100 representative canids (see table S1 for samples selected in the `Phylogenetic Analysis`) we confirmed Balto’s phylogenetic position by generating a neighbor-joining (NJ) phylogenetic tree and conducting a principal component analysis (PCA). We converted the variant calls into a FASTA file and used MEGA-CC(33) with 1000 bootstraps to assess tree topology. We also ran a PCA on this set using PLINK (v1.9), and then visualized the first two principal components in R (v. 3.6.3) using the `ggplot2` package. Global ancestry inference We inferred Balto’s ancestral similarity to modern dog breeds, sled dog type breeds, and working sled dogs using a custom built reference panel of modern dogs and canids of the 21st century (table S3). In PLINK (v2.00a3LM) (35), we identified 4,267,732 biallelic single nucleotide polymorphisms with <10% missing genotypes, and calculated Wright’s F-statistics using Hudson method (36, 37) for (1) each dog breed and sled dog population versus all other dogs; (2) all village dogs versus all other dogs; (3) each regional village dog population; (4) all wolves versus all other dogs; (5) all coyotes versus all other canids; and (6) North American wolves versus Eurasian wolves. We selected 1,858,634 SNPs with FST>0.5 across all comparisons, and performed LD-based pruning in 250kb windows for r2>0.2 to extract 136,779 markers for global ancestry inference. We merged Balto’s genotypes for these SNPs with genotypes from the reference samples. For reference samples also represented in the whole genome dataset, population labels used in the admixture analysis are given in the `Representative in Global Ancestry Inferencè column of table S1. Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 8 We performed global ancestry inference using ADMIXTURE (38) in both supervised mode (random seed: 43) with 20 bootstrap replicates to estimate parameter standard errors, and in unsupervised mode for the same number of populations (K=116), which showed low levels of error (0.3) in ten-fold cross-validation analysis of chromosome 1 for K clusters between 50 and 150 (table S13). Homozygosity and inbreeding metrics We removed samples with any missing data from the dataset of 100 representative individuals used in the phylogenetic analyses, leaving 86 individuals (see table S1 for samples selected in the ‘Homozygosity Analysis’). Using this pruned dataset, we detected runs of homozygosity (RoH) using a window-based approach implemented in PLINK (v1.9) (35). We calculated two measures of inbreeding: the method-of-moments coefficient in PLINK (FMoM) and the metric based on runs-of-homozygosity (FRoH), as recommended by Zhao et al. 2020 (40) (table S4). Using the R (v. 3.6.3) function `cor.test`, we confirmed that FRoH and FMoM are significantly correlated (RPearson= 0.6752819, p= 9.958e-13, t= 8.3913, df= 84). Population representative sampling As Balto is the sole representative of his population, we randomly selected one representative sample from each of 57 populations for the discovery of individually- represented, population-relevant genetic variants (see table S1 for samples selected in the `Population Variants Analysis`) among 67,085,518 biallelic single nucleotide polymorphisms. These populations included Balto, 1 Alaskan sled dog, 1 Greenland sled dog, and 54 modern purebred dogs, including 1 Siberian husky and 1 Alaskan malamute. Likewise, we selected, where available, another 5 to 11 random samples from 10 modern breeds, and all remaining Greenland sled dog samples, to assess the population-wide allele frequency of these variants (see table S1 `Population Frequency Analysis`). Dog-referenced mammalian evolutionary constraint We selected biallelic SNPs under evolutionary constraint by examining sites overlapping phyloP evolutionary constraint scores from the dog-referenced version of the 240 species Cactus alignment (3). We calculated the constraint score cutoffs at various false discovery rates (FDR). Unique, rare, and potentially deleterious variants We first identified all “population-unique” variants, defined as those observed in the representative dog from a population (either once or twice) and not observed in representatives from any of the other populations. With this method, we identified 206,164 population-unique variants for Balto, 120,279 for the Alaskan sled dog, 119,482 variants for the Greenland sled dog, 120,780 unique to the Alaskan malamute, and 133,200 unique to the Siberian husky. We confirmed that population-unique variants tend to be uncommon by calculating the allele frequencies in its population. We used Zoonomia PhyloP scores and SnpEff(42) annotations to identify which population-unique variants were either “evolutionarily constrained” (phyloP score above the FDR 0.05 cutoff of 2.56) or a Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 9 missense mutation and thus more likely to have functional consequences (table S15). We grouped the dogs into working dog groups including Balto, Alaskan sled dog, and Greenland sled dog, and modern breeds including all the other 54 dogs. We then applied Student’s t-test on the percentage of “evolutionarily constrained” or missense mutation for the two groups. Derived, common, and potentially beneficial variants We identified “homozygous derived” variants, defined as those observed twice in the representative dog from a population and not observed in wolves, for each of the populations. With this method, we identified 176,135 homozygous derived variants for Balto, 148,036 variants for Alaskan sled dog, 260,457 variants for Greenland sled dog, 225,270 variants for Alaskan Malamute, and 189,188 variants for Siberian husky. We confirmed that homozygous variants in each representative dog tend to be “common” in their population by calculating the allele frequency of the homozygous derived variants in its own breed. We also used a Wilcox test against randomly selected SNPs to show that population-unique SNPs are rare, whereas homozygous derived SNPs are rather common, among their population. We further defined variants likely to be functional as those that were both “highly evolutionarily constrained” (defined by phyloP score above the FDR>0.01 cutoff of 3.52) and a missense mutation. We annotated the variant by genes, and performed gene set enrichment against all Gene Ontology Biological Process gene sets (http:// geneontology.org/) using the R package rbioapi v. 0.7.4 (43, 44) (table S7, S8). We also tested for overlap between Balto’s variant genes and genes implicated in particular phenotypes in human studies using the Human Phenotype Ontology (24) and the “Investigate gene sets” feature provided by GSEA (http://www.gsea-msigdb.org/) (table S9). Prediction of Balto’s aesthetic phenotypes We extracted Balto’s genotypes for a panel of 27 genetic variants associated with physical appearance in domestic dogs (table S5) to infer his coat coloration, patterning, and type. We also phased haplotypes from Balto’s genotypes using EAGLE (v.2.4.1) (51) with reference haplotypes from the phased UMass-Broad Canid Variants and constructed the haplotype consensus sequences of the MITF-M promoter length polymorphism locus (chr20: 21,839,331 – 21,839,366) and upstream SINE insertion locus (chr20: 21,836,232 – 21,836,429) using BCFtools in order to investigate the MITF variants that putatively affect white spotting. We also ran a body size prediction for Balto using a random forest model (R packages `caret` and `randomForest`) built on the relative heights (defined as where a dog’s shoulders fall relative to an “average person”, and surveyed on a Likert scale from ankle-high and shorter, or survey option 0, to hip-high and taller, or survey option 4) of 1,730 modern pet dogs surveyed and 2,797 size-associated SNPs genotyped by the Darwin’s Ark project described previously (12) (see supporting files for model and scripts used to run prediction). Balto’s physiological adaptations We examined the genotypes underlying 14 regions (table S6), which included 1 region under selection in high altitude individuals (53) (Endothelial PAS domain-containing protein Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 10 1-EPAS1), 2 regions previously identified as under selection in sled dogs (10) (Calcium Voltage-Gated Channel Subunit Alpha1 A - CACNA1A and Maltase-Glucoamylase - MGAM), 8 regions identified by population branch statistics as potentially under selection in sled dog breeds (12), and 3 regions responsible for aesthetic phenotypes described previously in domestic dogs (Melanocortin 1 Receptor - MC1R (45), Agouti Signaling Protein - ASIP (52), and a chr28 cis-regulatory region associated with single-layered coats (13)). Following the method outlined in Bergström et al. 2020 (21), we also investigated the number of Amylase Alpha 2B (AMY2B) copies Balto had by quantifying the ratio of reads (reads/total length of region) mapping to the AMY2B regions in CanFam3.1 (ratio: 0.20) to the number of reads mapping to 75 randomly chosen 1kb windows of the genome (ratio: 0.59), given that higher copy numbers are suggested for dog adaptation to starch-rich diets (22). Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgments: We thank the Cleveland Museum of Natural History for their contributions to Balto’s preservation and history, and the owners of the three working Alaskan sled dogs sequenced for this work (IACUC #2014–0121). Funding: NIH grant R01 HG008742 (EKK) NIH grant U19 AG057377 (EKK) The Siberian Husky Club of America Data and materials availability: Raw sequencing reads for Balto and Alaskan sled dogs have been deposited to the NCBI Sequence Read Archive under BioProject accession PRJNA786530. References 1. Lindblad-Toh K, Garber M, Zuk O, Lin MF, Parker BJ, Washietl S, Kheradpour P, Ernst J, Jordan G, Mauceli E, Ward LD, Lowe CB, Holloway AK, Clamp M, Gnerre S, Alföldi J, Beal K, Chang J, Clawson H, Cuff J, Di Palma F, Fitzgerald S, Flicek P, Guttman M, Hubisz MJ, Jaffe DB, Jungreis I, Kent WJ, Kostka D, Lara M, Martins AL, Massingham T, Moltke I, Raney BJ, Rasmussen MD, Robinson J, Stark A, Vilella AJ, Wen J, Xie X, Zody MC, Broad Institute Sequencing Platform and Whole Genome Assembly Team, Baldwin J, Bloom T, Chin CW, Heiman D, Nicol R, Nusbaum C, Young S, Wilkinson J, Worley KC, Kovar CL, Muzny DM, Gibbs RA, Baylor College of Medicine Human Genome Sequencing Center Sequencing Team, Cree A, Dihn HH, Fowler G, Jhangiani S, Joshi V, Lee S, Lewis LR, Nazareth LV, Okwuonu G, Santibanez J, Warren WC, Mardis ER, Weinstock GM, Wilson RK, Genome Institute at Washington University, Delehaunty K, Dooling D, Fronik C, Fulton L, Fulton B, Graves T, Minx P, Sodergren E, Birney E, Margulies EH, Herrero J, Green ED, Haussler D, Siepel A, Goldman N, Pollard KS, Pedersen JS, Lander ES, Kellis M, A high-resolution map of human evolutionary constraint using 29 mammals. Nature. 478, 476–482 (2011). [PubMed: 21993624] 2. Meadows J, Gazal S, Sullivan P, Zoonomia Consortium, Karlsson EK, Lindblad-Toh K, Leveraging Base Pair Mammalian Constraint to Understand Genetic Variation and Human Disease. Science. Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 11 3. Christmas MJ, Kaplow IM, Genereux DP, Dong MX, Hughes GM, Li X, Sullivan PF, Hindle AG, Andrews G, Armstrong JC, Bianchi M, Breit AM, Diekhans M, Fanter C, Foley NM, Goodman L, Keough KC, Kirilenko B, Kowalczyk A, Lawless C, Lind A, Meadows JRS, Moreira L, Ryan L, Swofford R, Valenzuela A, Wagner F, Wallerman O, Damas J, Fan K, Grimshaw J, Johnson J, Kozyrev SV, Lawler AJ, Marinescu VD, Osmanski A, Paulat NS, Phan BN, Reilly SK, Schäffer DE, Steiner C, Supple MA, Wilder AP, Wirthlin ME, Xue JR, Birren BW, Gazal S, Hubley RM, Koepfli K-P, Marques-Bonet T, Meyer W, Nweeia M, Shapiro B, Smit AFA, Springer M, Teeling E, Weng Z, Hiller M, Levesque DL, Lewin H, Murphy WJ, Navarro A, Paten B, Pollard KS, Ray DA, Ruf I, Ryder OA, Pfenning AR, Lindblad-Toh K, Karlsson EK, Evolutionary constraint and innovation across hundreds of placental mammals. Science. 4. Armstrong J, Hickey G, Diekhans M, Fiddes IT, Novak AM, Deran A, Fang Q, Xie D, Feng S, Stiller J, Genereux D, Johnson J, Marinescu VD, Alföldi J, Harris RS, Lindblad-Toh K, Haussler D, Karlsson E, Jarvis ED, Zhang G, Paten B, Progressive Cactus is a multiple-genome aligner for the thousand-genome era. Nature. 587, 246–251 (2020). [PubMed: 33177663] 5. Zoonomia Consortium A comparative genomics multitool for scientific discovery and conservation. Nature. 587, 240–245 (2020). [PubMed: 33177664] 6. Salisbury G, Salisbury L, The Cruelest Miles: The Heroic Story of Dogs and Men in a Race Against an Epidemic (W. W. Norton & Company, 2003). 7. Sutter NB, Mosher DS, Gray MM, Ostrander EA, Morphometrics within dog breeds are highly reproducible and dispute Rensch’s rule. Mamm. Genome. 19, 713–723 (2008). [PubMed: 19020935] 8. American Kennel Club, The Complete Dog Book: 20th Edition (Random House Publishing Group, 2007). 9. Huson HJ, Parker HG, Runstadler J, Ostrander EA, A genetic dissection of breed composition and performance enhancement in the Alaskan sled dog. BMC Genet. 11, 71 (2010). [PubMed: 20649949] 10. Sinding M-HS, Gopalakrishnan S, Ramos-Madrigal J, de Manuel M, Pitulko VV, Kuderna L, Feuerborn TR, Frantz LAF, Vieira FG, Niemann J, Samaniego Castruita JA, Carøe C, Andersen- Ranberg EU, Jordan PD, Pavlova EY, Nikolskiy PA, Kasparov AK, Ivanova VV, Willerslev E, Skoglund P, Fredholm M, Wennerberg SE, Heide-Jørgensen MP, Dietz R, Sonne C, Meldgaard M, Dalén L, Larson G, Petersen B, Sicheritz-Pontén T, Bachmann L, Wiig Ø, Marques-Bonet T, Hansen AJ, Gilbert MTP, Arctic-adapted dogs emerged at the Pleistocene–Holocene transition. Science. 368, 1495–1499 (2020). [PubMed: 32587022] 11. Shindyapina AV, Zenin AA, Tarkhov AE, Santesmasses D, Fedichev PO, Gladyshev VN, Germline burden of rare damaging variants negatively affects human healthspan and lifespan. Elife. 9 (2020), doi:10.7554/eLife.53449. 12. Morrill K, Hekman J, Li X, McClure J, Logan B, Goodman L, Gao M, Dong Y, Alonso M, Carmichael E, Snyder-Mackler N, Alonso J, Noh HJ, Johnson J, Koltookian M, Lieu C, Megquier K, Swofford R, Turner-Maier J, White ME, Weng Z, Colubri A, Genereux DP, Lord KA, Karlsson EK, Ancestry-inclusive dog genomics challenges popular breed stereotypes. Science (2022). 13. Whitaker DT, Ostrander EA, Hair of the Dog: Identification of a Cis-Regulatory Module Predicted to Influence Canine Coat Composition. Genes. 10 (2019), doi:10.3390/genes10050323. 14. Karlsson EK, Baranowska I, Wade CM, Salmon Hillbertz NHC, Zody MC, Anderson N, Biagi TM, Patterson N, Pielberg GR, Kulbokas EJ 3rd, Comstock KE, Keller ET, Mesirov JP, von Euler H, Kämpe O, Hedhammar A, Lander ES, Andersson G, Andersson L, Lindblad-Toh K, Efficient mapping of mendelian traits in dogs through genome-wide association. Nat. Genet. 39, 1321–1328 (2007). [PubMed: 17906626] 15. Dreger DL, Parker HG, Ostrander EA, Schmutz SM, Identification of a mutation that is associated with the saddle tan and black-and-tan phenotypes in Basset Hounds and Pembroke Welsh Corgis. J. Hered. 104, 399–406 (2013). [PubMed: 23519866] 16. Deane-Coe PE, Chu ET, Slavney A, Boyko AR, Sams AJ, Direct-to-consumer DNA testing of 6,000 dogs reveals 98.6-kb duplication associated with blue eyes and heterochromia in Siberian Huskies. PLoS Genet. 14, e1007648 (2018). [PubMed: 30286082] 17. Schmutz SM, Berryere TG, Ellinwood NM, Kerns JA, Barsh GS, MC1R studies in dogs with melanistic mask or brindle patterns. J. Hered. 94, 69–73 (2003). [PubMed: 12692165] Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 12 18. Slavney AJ, Kawakami T, Jensen MK, Nelson TC, Sams AJ, Boyko AR, Five genetic variants explain over 70% of hair coat pheomelanin intensity variation in purebred and mixed breed domestic dogs. PLoS One. 16, e0250579 (2021). [PubMed: 34043658] 19. Anderson H, Honkanen L, Ruotanen P, Mathlin J, Donner J, Comprehensive genetic testing combined with citizen science reveals a recently characterized ancient MC1R mutation associated with partial recessive red phenotypes in dog. Canine Med Genet. 7, 16 (2020). [PubMed: 33292722] 20. Axelsson E, Ratnakumar A, Arendt M-L, Maqbool K, Webster MT, Perloski M, Liberg O, Arnemo JM, Hedhammar A, Lindblad-Toh K, The genomic signature of dog domestication reveals adaptation to a starch-rich diet. Nature. 495, 360–364 (2013). [PubMed: 23354050] 21. Bergström A, Frantz L, Schmidt R, Ersmark E, Lebrasseur O, Girdland-Flink L, Lin AT, Storå J, Sjögren K-G, Anthony D, Antipina E, Amiri S, Bar-Oz G, Bazaliiskii VI, Bulatović J, Brown D, Carmagnini A, Davy T, Fedorov S, Fiore I, Fulton D, Germonpré M, Haile J, Irving-Pease EK, Jamieson A, Janssens L, Kirillova I, Horwitz LK, Kuzmanovic-Cvetković J, Kuzmin Y, Losey RJ, Dizdar DL, Mashkour M, Novak M, Onar V, Orton D, Pasarić M, Radivojević M, Rajković D, Roberts B, Ryan H, Sablin M, Shidlovskiy F, Stojanović I, Tagliacozzo A, Trantalidou K, Ullén I, Villaluenga A, Wapnish P, Dobney K, Götherström A, Linderholm A, Dalén L, Pinhasi R, Larson G, Skoglund P, Origins and genetic legacy of prehistoric dogs. Science. 370, 557–564 (2020). [PubMed: 33122379] 22. Arendt M, Fall T, Lindblad-Toh K, Axelsson E, Amylase activity is associated with AMY 2B copy numbers in dog: implications for dog domestication, diet and diabetes. Animal Genetics. 45 (2014), pp. 716–722. [PubMed: 24975239] 23. Gou X, Wang Z, Li N, Qiu F, Xu Z, Yan D, Yang S, Jia J, Kong X, Wei Z, Lu S, Lian L, Wu C, Wang X, Li G, Ma T, Jiang Q, Zhao X, Yang J, Liu B, Wei D, Li H, Yang J, Yan Y, Zhao G, Dong X, Li M, Deng W, Leng J, Wei C, Wang C, Mao H, Zhang H, Ding G, Li Y, Whole-genome sequencing of six dog breeds from continuous altitudes reveals adaptation to high-altitude hypoxia. Genome Res. 24, 1308–1315 (2014). [PubMed: 24721644] 24. Köhler S, Gargano M, Matentzoglu N, Carmody LC, Lewis-Smith D, Vasilevsky NA, Danis D, Balagura G, Baynam G, Brower AM, Callahan TJ, Chute CG, Est JL, Galer PD, Ganesan S, Griese M, Haimel M, Pazmandi J, Hanauer M, Harris NL, Hartnett MJ, Hastreiter M, Hauck F, He Y, Jeske T, Kearney H, Kindle G, Klein C, Knoflach K, Krause R, Lagorce D, McMurry JA, Miller JA, Munoz-Torres MC, Peters RL, Rapp CK, Rath AM, Rind SA, Rosenberg AZ, Segal MM, Seidel MG, Smedley D, Talmy T, Thomas Y, Wiafe SA, Xian J, Yüksel Z, Helbig I, Mungall CJ, Haendel MA, Robinson PN, The Human Phenotype Ontology in 2021. Nucleic Acids Res. 49, D1207–D1217 (2021). [PubMed: 33264411] 25. Thomas B, Thomas P, Leonhard Seppala: The Siberian Dog and the Golden Age of Sleddog Racing 1908–1941 (Pictorial Histories Publishing Company, Incorporated, 2015). 26. The Cleveland Museum of Natural History, Balto FAQs, (available at https://www.cmnh.org/ science-news/blog/march-2020/balto-faq). 27. Sled dog central: The Inuit sled dog by sue Hamilton, (available at http://www.sleddogcentral.com/ inuit.htm). 28. Dabney J, Knapp M, Glocke I, Gansauge M-T, Weihmann A, Nickel B, Valdiosera C, García N, Pääbo S, Arsuaga J-L, Meyer M, Complete mitochondrial genome sequence of a Middle Pleistocene cave bear reconstructed from ultrashort DNA fragments. Proc. Natl. Acad. Sci. U. S. A. 110, 15758–15763 (2013). [PubMed: 24019490] 29. Kapp JD, Green RE, Shapiro B, A Fast and Efficient Single-stranded Genomic Library Preparation Method Optimized for Ancient DNA. J. Hered. 112, 241–249 (2021). [PubMed: 33768239] 30. John JS, SeqPrep: tool for stripping adaptors and/or merging paired reads with overlap into single reads. URL: https://githubcom/jstjohn/SeqPrep (2011). 31. Li H, Durbin R, Fast and accurate long-read alignment with Burrows–Wheeler transform. Bioinformatics. 26, 589–595 (2010). [PubMed: 20080505] 32. Plassais J, Kim J, Davis BW, Karyadi DM, Hogan AN, Harris AC, Decker B, Parker HG, Ostrander EA, Whole genome sequencing of canids reveals genomic regions under selection and variants influencing morphology. Nat. Commun. 10, 1489 (2019). [PubMed: 30940804] Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 13 33. Kumar S, Stecher G, Peterson D, Tamura K, MEGA-CC: computing core of molecular evolutionary genetics analysis program for automated and iterative data analysis. Bioinformatics. 28, 2685–2686 (2012). [PubMed: 22923298] 34. Hayward JJ, Castelhano MG, Oliveira KC, Corey E, Balkman C, Baxter TL, Casal ML, Center SA, Fang M, Garrison SJ, Kalla SE, Korniliev P, Kotlikoff MI, Moise NS, Shannon LM, Simpson KW, Sutter NB, Todhunter RJ, Boyko AR, Complex disease and phenotype mapping in the domestic dog. Nat. Commun. 7, 10460 (2016). [PubMed: 26795439] 35. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, Maller J, Sklar P, de Bakker PIW, Daly MJ, Sham PC, PLINK: a tool set for whole-genome association and population- based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007). [PubMed: 17701901] 36. Weir BS, Cockerham CC, ESTIMATING F-STATISTICS FOR THE ANALYSIS OF POPULATION STRUCTURE. Evolution. 38, 1358–1370 (1984). [PubMed: 28563791] 37. Bhatia G, Patterson N, Sankararaman S, Price AL, Estimating and interpreting FST: the impact of rare variants. Genome Res. 23, 1514–1521 (2013). [PubMed: 23861382] 38. Alexander DH, Lange K, Enhancements to the ADMIXTURE algorithm for individual ancestry estimation. BMC Bioinformatics. 12, 246 (2011). [PubMed: 21682921] 39. Foote AD, Hooper R, Alexander A, Baird RW, Baker CS, Ballance L, Barlow J, Brownlow A, Collins T, Constantine R, Dalla Rosa L, Davison NJ, Durban JW, Esteban R, Excoffier L, Martin SLF, Forney KA, Gerrodette T, Gilbert MTP, Guinet C, Hanson MB, Li S, Martin MD, Robertson KM, Samarra FIP, de Stephanis R, Tavares SB, Tixier P, Totterdell JA, Wade P, Wolf JBW, Fan G, Zhang Y, Morin PA, Runs of homozygosity in killer whale genomes provide a global record of demographic histories. Mol. Ecol. (2021), doi:10.1111/mec.16137. 40. Zhao G, Zhang T, Liu Y, Wang Z, Xu L, Zhu B, Gao X, Zhang L, Gao H, Liu GE, Li J, Xu L, Genome-Wide Assessment of Runs of Homozygosity in Chinese Wagyu Beef Cattle. Animals (Basel). 10 (2020), doi:10.3390/ani10081425. 41. McQuillan R, Leutenegger A-L, Abdel-Rahman R, Franklin CS, Pericic M, Barac-Lauc L, Smolej-Narancic N, Janicijevic B, Polasek O, Tenesa A, Macleod AK, Farrington SM, Rudan P, Hayward C, Vitart V, Rudan I, Wild SH, Dunlop MG, Wright AF, Campbell H, Wilson JF, Runs of homozygosity in European populations. Am. J. Hum. Genet. 83, 359–372 (2008). [PubMed: 18760389] 42. Cingolani P, snpEff: Variant effect prediction (2012). 43. Rezwani M, Pourfathollah AA, Noorbakhsh F, rbioapi: User-Friendly R Interface to Biologic Web Services’ API. Bioinformatics (2022), doi:10.1093/bioinformatics/btac172. 44. Mi H, Ebert D, Muruganujan A, Mills C, Albou L-P, Mushayamaha T, Thomas PD, PANTHER version 16: a revised family classification, tree-based classification tool, enhancer regions and extensive API. Nucleic Acids Res. 49, D394–D403 (2021). [PubMed: 33290554] 45. Schmutz SM, Berryere TG, Goldfinch AD, TYRP1 and MC1R genotypes and their effects on coat color in dogs. Mamm. Genome. 13, 380–387 (2002). [PubMed: 12140685] 46. Cargill EJ, Famula TR, Schnabel RD, Strain GM, Murphy KE, The color of a Dalmatian’s spots: linkage evidence to support the TYRP1 gene. BMC Vet. Res. 1, 1 (2005). [PubMed: 16045797] 47. Kiener S, Kehl A, Loechel R, Langbein-Detsch I, Müller E, Bannasch D, Jagannathan V, Leeb T, Novel Brown Coat Color (Cocoa) in French Bulldogs Results from a Nonsense Variant in HPS3. Genes. 11 (2020), doi:10.3390/genes11060636. 48. Kerns JA, Newton J, Berryere TG, Rubin EM, Cheng J-F, Schmutz SM, Barsh GS, Characterization of the dog Agouti gene and a nonagouti mutation in German Shepherd Dogs. Mamm. Genome. 15, 798–808 (2004). [PubMed: 15520882] 49. Kerns JA, Cargill EJ, Clark LA, Candille SI, Berryere TG, Olivier M, Lust G, Todhunter RJ, Schmutz SM, Murphy KE, Barsh GS, Linkage and segregation analysis of black and brindle coat color in domestic dogs. Genetics. 176, 1679–1689 (2007). [PubMed: 17483404] 50. Monteagudo LV, Tejedor MT, The b(c) allele of TYRP1 is causative for the recessive brown (liver) colour in German Shepherd dogs. Anim. Genet. 46, 588–589 (2015). [PubMed: 26370740] 51. Loh P-R, Danecek P, Palamara PF, Fuchsberger C, A Reshef Y, K Finucane H, Schoenherr S, Forer L, McCarthy S, Abecasis GR, Durbin R, L Price A, Reference-based phasing using the Haplotype Reference Consortium panel. Nat. Genet. 48, 1443–1448 (2016). [PubMed: 27694958] Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 14 52. Berryere TG, Kerns JA, Barsh GS, Schmutz SM, Association of an Agouti allele with fawn or sable coat color in domestic dogs. Mamm. Genome. 16, 262–272 (2005). [PubMed: 15965787] 53. vonHoldt B, Fan Z, Ortega-Del Vecchyo D, Wayne RK, EPAS1 variants in high altitude Tibetan wolves were selectively introgressed into highland dogs. PeerJ. 5, e3522 (2017). [PubMed: 28717592] Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 15 Figure 1. Balto clusters most closely with Alaskan sled dogs, but had high genetic diversity and a lower burden of potentially damaging variants. (A) Neighbor-joining tree clusters Balto (★) most closely with the outbred, working population of Alaskan sled dogs, and a part of a clade of sled dog populations. (B) Similarly, principal component analysis puts Balto near, but not in, a cluster of Alaskan sled dogs. (C) Unsupervised admixture analysis of Balto alongside the Alaskan sled dogs and other dogs and canids (K= 116 putative populations and N= 2166 individuals) infers substantial ancestral similarity to Siberian huskies, Greenland sled dogs, and outbred dogs from Asia (table S2). The remainder of his ancestry (8%) matches poorly (<5%) to any other clusters. Balto and working sled dogs (D) had lower levels of inbreeding, and (E) carried fewer constrained (pwilcox=0.0019) and missense (pwilcox= 0.0023) rare variants than modern dog breeds (table S10). Science. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Moon et al. Page 16 Figure 2. Genomic recreation of Balto’s physical appearance. (A) Prediction of Balto’s coat features based on his genome sequence with details on each trait and genotype in blue boxes. (B) A photo of Balto with musher Gunnar Kaasen. From the photo and his taxidermied remains, Balto was a black dog with dark eyes and some white patches on his chest and legs. He had a double-layered coat, and stood just under knee-high relative to Kaasen. Photo credit: Cleveland Museum of Natural History. (C) Using a random forest model based on 1,730 dogs and 2,797 height-associated genetic variants (12), we predicted that Balto would stand around 55 cm tall (value: 2.3) at his withers, close to the average height for the Siberian husky breed. Circles show dogs from other breeds. (D) Gene set enrichment testing of genes with common and constrained missense variants in 57 different dog populations shows a significant enrichment (pFDR=0.013) in the GO Tissue Development pathway only for Balto’s population. Science. Author manuscript; available in PMC 2023 May 15.
10.1126_science.adh1720	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 August 04. Published in final edited form as: Science. 2023 July 28; 381(6656): eadh1720. doi:10.1126/science.adh1720. Deploying synthetic coevolution and machine learning to engineer protein-protein interactions Aerin Yang1, Kevin M. Jude1,2, Ben Lai3, Mason Minot4, Anna M. Kocyla1, Caleb R. Glassman1, Daisuke Nishimiya1, Yoon Seok Kim1, Sai T. Reddy4, Aly A. Khan3,5, K. Christopher Garcia1,2,6,* 1Department of Molecular and Cellular Physiology, Stanford University School of Medicine, Stanford, CA 94305, USA. 2Howard Hughes Medical Institute, Stanford University School of Medicine, Stanford, CA 94305, USA. 3Toyota Technological Institute at Chicago, Chicago, IL 60637, USA 4Department of Biosystems Science and Engineering, ETH Zurich, Basel, Switzerland. 5Departments of Pathology, and Family Medicine, University of Chicago, Chicago, IL 60637, USA 6Department of Structural Biology, Stanford University School of Medicine, Stanford, CA 94305, USA. Abstract Fine-tuning of protein-protein interactions occurs naturally through coevolution, but this process is difficult to recapitulate in the laboratory. We describe a synthetic platform for protein-protein coevolution that can isolate matched pairs of interacting muteins from complex libraries. This large dataset of coevolved complexes drove a systems-level analysis of molecular recognition between Z domain-affibody pairs spanning a wide range of structures, affinities, cross-reactivities, and orthogonalities, and captured a broad spectrum of coevolutionary networks. Furthermore, we harnessed pre-trained protein language models to expand, in silico, the amino acid diversity of our coevolution screen, predicting remodeled interfaces beyond the reach of the experimental This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding author. [email protected]. Author contributions Conceptualization: KCG Methodology: AY, KMJ, CRG, DN, KCG Investigation: AY, KMJ, BL, MM, AMK Visualization: AY, KMJ, BL, MM, YSK Funding acquisition: KCG Supervision: KCG Writing – original draft: AY, KMJ, AAK Writing – review & editing: AY, KMJ, BL, MM, YSK, STR, AAK, KCG Competing interests: Authors declare that they have no competing interests. Data and materials availability: Diffraction images have been deposited at the SBGrid databank. Protein structures and reflection files for LL1 and LL2 complex structures have been deposited at the RCSB Protein Databank with PDB IDs 8DA3, 8DA4, 8DA5, 8DA6, 8DA7, 8DA8, 8DA9, 8DAA, 8DAB, and 8DAC. NGS data is deposited at DRYAD (46). Data and code for MI, coevolution analysis, and deep learning model are available at https://github.com/akds/CoevolveML and archived at Zenodo (47). A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 2 library. The integration of these approaches provides a means of generating protein complexes with diverse molecular recognition properties as tools for biotechnology and synthetic biology. One-Sentence Summary: Integration of synthetic coevolution with machine learning enables prediction of remodeled protein-protein interfaces. In evolutionary biology, the concept of coevolution underscores the compensatory relationships between biological systems that occur as a result of evolutionary pressures. Coevolution refers to reciprocal changes that occur under selective pressures between pairs of biomolecules or living organisms to fine-tune functions. Charles Darwin introduced the concept of coevolution by observing the relationship between the length of insects’ proboscis and the size of orchids’ spur, which led him to predict the evolutionary changes of insects that could suck from the deep spur of Darwin’s orchid (1). By analogy, interacting proteins often undergo coupled mutations within, or proximal to their interfaces to maintain or refine their functional interactions (2-4). Phylogenetic sequence information reveals correlated mutations accumulate through natural evolution, suggestive of compensatory changes occurring between interacting residues (5, 6). Protein coevolution has been difficult to study experimentally in the laboratory using reconstituted systems. Although directed evolution via phage or yeast surface display has enabled efficient screening to discover binders with improved affinity and specificity toward a fixed target protein (7), it has been more challenging to execute “library-on-library” selections to coevolve both sides of a protein-protein interface concurrently (8-10). In vitro, co-selection by mixing separate libraries is limited by the inability to isolate discrete coevolved pairs from complex mixtures, thereby losing connectivity between the sequences of members of interacting pairs (8). Coevolution studies using both in vivo functional selections such as bacterial in vivo screening (11, 12) or yeast two-hybrid systems (13) and in vitro screening strategies including yeast mating systems (9, 10) or compartmentalized two-hybrid system (14) have been reported, but these systems are limited by small library sizes resulting in acquisition of sparse information rather than a broad evolutionary spectrum. An additional practical limitation to developing a synthetic coevolution system is that the diversity of experimental combinatorial libraries is limited, which makes experimental exploration of the entire sequence space required to fully sample a protein-protein interface impossible. However, recent advances in protein language models (15, 16), and transfer learning offer the possibility of employing transfer learning to “transfer” knowledge learned from a subset of combinations to predict the binding affinity of a larger set of amino acid combinations that have not been experimentally tested. This enables effective exploration of a much larger space of combinations and identification of those that perform the desired function, A high-throughput system for coevolving protein-protein interfaces could have practical utility for protein engineering in biotechnology and serve as a powerful basic tool to Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 3 interrogate fundamental properties of molecular recognition. Here, we describe a strategy to achieve coevolution of protein-protein pairs using a high-throughput screening platform for library-on-library based directed evolution. We adopted the Z domain of staphylococcal protein A and its affibody binder dimer complex as a model system (17). The large dataset of interacting mutant sequences was subjected to systems-level analysis of molecular recognition. High resolution crystal structures of orthogonal mutant pairs elaborated compensatory changes in predicted co-varying residues and structural adaptations. By tracking the mutational trajectories of coevolved mutants, we observed continuous changes in the connectivity and specificity between mutants. We show that the set of coevolved protein pairs can inform machine learning algorithms to predict new complexes with amino acid compositions not encoded within the experimental libraries. Results Design of inter-protein coevolution and validation of selection strategy To develop a platform for protein-protein coevolution using yeast surface display, we adapted the yeast display α-agglutinin system to display two different proteins expressed as a single chain connected by a flexible linker (Fig. 1A). A 3C protease site (-LEVLFQGP-) was inserted within the linker to enable 3C protease cleavage of the connected proteins. Following proteolytic cleavage, the first protein and its associated c-Myc tag remain covalently attached to the yeast cell surface while the second protein and HA-tag are liberated. The surviving non-covalently connected interacting pairs, together with the associated yeast clones, can then be isolated with C-terminal HA-tag binding antibodies. The identities of both interacting proteins can then be determined by DNA sequencing of the enriched yeast clones. We wished to execute proof-of-concept experiments for this strategy using a simple system of small stable proteins, so we chose the complex (KD = 10 nM) of Z domain and its affibody binder, ZpA963 (PDB: 2M5A). This is a model system with an interface idealized through phage display (18). We tested the cleavage-capture efficiency of three forms of fluorescently labeled anti-HA tag antibodies with different valency (Fab, IgG mAb, Fab+streptavidin (SA) complex) to determine whether their fluorescence was maintained after 3C protease cleavage of the linker between two proteins (Fig. S1A). We found that both bivalent IgG mAb labeled cells and tetrameric complex (Fab+SA), but not monovalent Fab labeled cells, maintained their staining levels of 45.8% and 69.9% of uncleaved cells respectively after 3C cleavage (Fig. 1A). We then chose six key residues forming the central hydrophobic portion of the interface based on the NMR structure of dimeric Z+ZpA963, accounting for 406 Å2 of the 1662 Å2 buried solvent-accessible surface area (BSA) on the two protein chains (Fig. 1B). When these six residues (F13, L17, and I31 in Z, and F17, I31, and L35 in ZpA963) were each mutated to alanine (6xAla), the antibody-stained yeast cells quickly lost their fluorescence to 0.63% within 10 minutes after 3C protease cleavage, whereas interacting pair displaying cells still retained fluorescence to 68.7% after an hour (Fig. 1C). We optimized different linker lengths (18, 22, 26 AA) and various components (1 copy or 2 copies of 3C protease site, HA-tag in the linker or at C-terminus), and magnetic-activated cell sorting (MACS) selection in the cleavage-capture assay (Fig. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 4 S1B-S1F). These general considerations and optimization strategies can be applied to other protein-protein complexes one wishes to implement into this coevolution platform. Notably, the on-yeast cleavage-capture assay is highly correlated with the dimer binding affinity, showing a log-linear relationship (R2 = 0.8382) at submicromolar affinity range (Fig. 1D). As an initial test, we asked whether the high-affinity Z-domain complex with ZpA963 would converge back to its phage-display idealized interface through coevolution (Fig. 1B). We generated libraries by randomizing the aforementioned six positions with two sets of degenerate codons: one set with only minimal hydrophobic amino acids (F, I, L, V, and M) and the other set with a more diverse set amino acids (F, I, L, V, H, K, N, Q, Y, D, and E) (Fig. 1B). After each round of selection, library evolution was monitored by cleavage- capture assay and flow cytometry. After four or five rounds of positive MACS selections along with interspersed negative selections, both HL1 and HL2 libraries clearly enriched higher HA-tag fluorescence after 30 min of 3C protease cleavage (Fig. 1E). We isolated cells displaying interacting pairs by FACS from each round of MACS for further next-generation sequencing (NGS). The NGS results showed that the libraries converged to the original sequences exactly or with very few differences (Fig. 1F). Leu17 in Z domain (A) and Ile31 in ZpA963 (B) were replaceable with Ile or Val, while other sites strongly converged to the original amino acids (Fig. 1F). Each clone was assessed by the cleavage-capture assay and reached different levels of steady-state binding of HA-tag fluorescence during 3C protease cleavage (Fig. 1G). Using surface plasmon resonance (SPR), we measured binding affinities that ranged from 7.9 nM to 34.1 nM, similar to the original template dimer affinity of 10 nM (Fig. S2A and S2B). These data suggest that the coevolution strategy was able to remodel the protein interface to its original “optimal” state from non-ideal starting points represented in the complex libraries. Coevolution of a low affinity dimer creates optimized new interfaces. We next generated libraries at the interface of a weakly associating dimer (Z+ ZSPA-1) with micromolar affinity to determine if we could affinity-mature the interface by coevolution (19, 20) (Fig. 2A). Consistent with its low affinity, the Z+ ZSPA-1 pair rapidly lost its HA-tag fluorescence within 15min of 3C protease treatment (Fig. S2C). Based on the crystal structure of the complex (PDB: 1LP1), nine interfacial positions located in a central hydrophobic patch were selected for library randomization: five positions (Q9, F13, L17, I31, K35) from Z domain and four positions (L9, V17, I31, F32) from ZSPA-1 domain. The first library, LL1, was designed to use minimal codon sets encoding both polar and hydrophobic amino acids (F, L, I, K, H, N, Q, and Y) for five positions on the Z domain and hydrophobic amino acids (F, L, I, V, and M) for four positions on the affibody ZSPA-1 (Fig. 2A). The second library, LL2, used a more diverse codon set encoding mixed amino acids (F, I, L, V, H, K, N, Q, Y, D, and E) for four randomized positions on each of the Z domain and the affibody, so that the functional diversity (1.91 × 109) of the yeast library almost reached the theoretical nucleotide diversity (4.29 × 109). After six rounds of positive MACS and two rounds of FACS selections, more than 90% of the populations enriched into the upper right quadrant of flow cytometry dot plots (Fig. 2B). NGS data collected at each step of selection clearly revealed the appearance of consensus sequences as the selection proceeded (Fig. 2C). Based on the sequencing data, we tested 11 clones Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 5 from LL1 and 22 clones from LL2 using the cleavage-capture assay, and all reached varying levels of steady-state binding during 1-hour 3C protease treatment (Fig. 2D). In contrast to the result of coevolution from the high affinity pair, the enriched mutants from both LL1 and LL2 libraries have only a few conserved residues shared with the original template: on average, 6 to 7 mutations were enriched (Fig. 2C and 2D). The highest affinity mutant from each library achieved approximately three-log enhanced affinity, KD of 1.99 nM for LL1.c1 (LIFFK/FILF) and 1.86 nM for LL2.c3 (LVLF/FIIV) compared to the original dimer (KD = 2.92 μM) (Fig. S2D-S2G). Synthetic coevolution yields pairs with different specificities and cross-reactivities To characterize the relationship between coevolved protein sequences in our screen, we visualized the sequencing data as a network. We used statistical enrichment to identify the sequence pairs with the strongest likelihood of binding from the enriched library sequencing data, based on the overall count of the individual sequences in the screen. We used a hypergeometric test (see Methods) to calculate this enrichment statistic, which compares the observed frequency of a particular protein pair in a screening library to the expected frequency of the pair based on the overall count of the individual proteins in the library. If the observed frequency is significantly higher than the expected frequency, it suggests that the protein pair is enriched for interaction. We extracted sequences with a p-value < 0.05 for further visualization and analysis of cross-reactivity and specificity. The enriched sequences accurately predicted the binding specificity of each Z-A sequence, matching well with its actual binding specificity (Fig. 3A). The sequence similarity network (SSN) is an efficient way to observe relationships among large sets of evolutionarily related proteins (21). We constructed SSNs using the concatenated Z-A and Z-B full-length 8 amino acid sequences collected from all screening rounds of the LL2 library. The SSN revealed clear connectivity between sequences from later rounds (rounds 5 to 7) when an edit distance threshold of 2 was applied (Fig. 3B, left). This analysis validates that our co-evolution platform progressively enriched communities of discrete recognition clusters. When sequences from round 7 were mapped with edit distance threshold 1, the sequences formed two large, disconnected groups and several smaller clusters (Fig. 3B, right). Several notable Z-A sequences were colored in the sequence similarity network. This revealed that the nodes with the same Z-A but differing Z-B were closely connected in the same cluster, and closely related Z-A sequences which differ by one amino acid could be clustered either together (e.g., VFLV and IFLV) or separately (e.g., LVLV and LVLF). The specificity similarity network (SpSN) of Z-A sequences which connects nodes when two Z-A sequences have common Z-B partners was illustrated, and the Z-A sequences that are clustered closely in the sequence similarity network were also closely connected in the specificity similarity network (Fig. S3). For example, VFLV, LVLV, and IFLV, clustered together in a big group in the SSN, are also closely connected in the SpSN, and LVLF is clustered separately in both the SSN and the SpSN (Fig. S3). This implies that the sequence similarity network can capture the specificity of Z-A sequences from our coevolutionary sequence data. The cluster graphs, which merge each clustered community into a single node, can efficiently show such relationships between co-evolved mutants and the structure of coevolutionary networks throughout the different screening Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 6 rounds (Fig. S4). Collectively this network level analysis reveals the extreme sensitivity of the specificity and cross-reactivity properties of Z-A and Z-B proteins to even single amino acid changes. In addition to the SSN, we utilized another visualization method to depict the cross- reactivity profiles of the NGS data in our coevolutionary libraries. The Circos plot shows the pairwise relationships, highlighting the relative cross-reactivity and orthogonality of the Z-A and Z-B proteins in both LL1 and LL2 libraries (Fig. 3C and Fig. S5-S6). We sampled 100 representative pairs to present in the plot, normalizing each pair to equal area in order to visualize the approximate cross-reactivity of each sequence. A series of Circos plots spanning all screening rounds (naïve, R2, R4, R5, R6, R7, and R8) reveals the progressive shifts in cross-reactivity during the selection process. For example, we observe the emergence of poly-specificity among certain dominant Z-A sequences and increased cross-reactivity between sequences in later rounds of selection in both the LL1 and LL2 libraries (Fig. S5 and S6). This result illustrates a broad range of specificity and orthogonality within our library. We next attempted to track the mutational pathways of specific coevolved pairs to assess how these dimer interfaces were diversified along the course of coevolution. We generated single mutational evolutionary pathways connecting the original sequence (QFLI/LVIF) with the prominent LL2 library mutants (Fig. 3D). First, we traced the Z-A pathway from the cluster graphs to identify the connected intermediates starting from the original sequence (QFLI) to the late mutants (LVFF, IVFF) (Fig. S7). The connectivity between early mutants (QFLI-VFLI), mid mutants (VFLI-VFLV-VFLF-VVLF-LVLF), and late mutants (LVLF-LVFF-IVFF) can be visualized from cluster graphs at different screening rounds (Fig. S7B). The ability to trace mutational pathways suggest this platform could be useful for simulating natural protein-protein evolution trajectories. To investigate the structural energetic mechanism mediating the changes in specificity during coevolution, we measured thermodynamic binding signatures by performing isothermal titration calorimetry (ITC) of several Z domain-affibody pairs along an evolutionary pathway (Fig. 3E, S8A, and S9). We see clear evidence for enthalpy-entropy compensation over the course of coevolution, and a trend where early strongly favorable enthalpy and unfavorable entropy transition to produce a less favorable binding enthalpy compensated by a more neutral entropy (Figs. 3E, S8). For example, we sampled representatives from the LL2 mutational pathway from the ‘founder’ pair (QFLI/LVIF) to IVFF/FILV (Figs. 3D, 3E). Although the overall free-energy landscape of this trajectory is flat, we see changes when examining the entropic and enthalpic terms. Binding of VFLV/IVVY and LVLF/FIIV are highly enthalpically favored and entropically disfavored, but by the end of the trajectory we see a more moderate enthalpy of binding coupled with a moderately disfavored entropy in IVFF/FILV. Although we could not observe any structural features that distinguish cross-reactive versus selective complexes, the thermodynamic properties of cross-reactive mutants (A-VFLV and A-LVLF) and specific mutant (A-IVFF) differed. We also followed a single mutational three-step evolutionary trajectory from LILFK/FIVM to LIFFK/FILF which are the two high affinity orthogonal pairs from LL1 library showing similar thermodynamic trends (Fig. S8A). A dramatic Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 7 thermodynamic transition occurs when Leu17A was mutated to Phe, to produce a less favorable binding enthalpy compensated by a more neutral entropy in specific mutant (LIFFK/FILF). Phenylalanine is often conserved in protein-protein binding sites, and aromatic residues frequently serve as anchor residues to mediate protein-protein interactions (22). The common mutation in both mutants, Leu17APhe, may act as a new anchor residue, thus leading to more entropically favored interactions between two proteins (23). We next verified by cleavage-capture assay the relative specificities of each Z-A sequence toward Z-B sequences from this evolutionary pathway (Fig. 3F). Starting from the early mutants, the specificity matrix clearly indicates gradual and continuous compensatory changes of binding preference between variants along the mutational pathway for both LL1 and LL2 libraries (Fig. 3F and Fig. S8B). Thus, we could systematically track the diversification of specificities and cross-reactivities within our library by mapping of the coevolutionary network. Direct coupling analysis and structural adaptations in coevolved complexes We sought to evaluate the accuracy of coevolutionary patterns between residues in predicting protein interaction contacts (Fig. 4). The coevolution of residues in protein sequences is affected by epistatic couplings, which may or may not match with structural contacts (24). First, we used mutual information (MI), a measure of the statistical coupling between any two positions in a protein pair, which can reflect structural interactions. To do this, we again filtered protein pairs to statistically enrich for those pairs occurring significantly more often than expected, mirroring our approach in the SSN analysis (see Methods). Using these filtered pairs, we calculated pairwise MI between all residues in the LL1 and LL2 screens (Fig. S10). MI serves as a local information theoretical metric, enabling us to determine the level of dependence between two positions. Our results showed that the top-ranked inferred coupling (17A-31B) was consistent with known contacts in the 3D structure of the original pair, indicating that structural constraints are captured in the sequence coevolution. Next, we applied a direct coupling analysis framework to the unfiltered LL2 sequences, which constitute a larger and more complex library (25). Our goal was to determine if direct interactions could be inferred with the increased size and complexity of the LL2 library and a global statistical method. We used the inverse covariance matrix to infer direct contacts. The columns in the matrix represent residues from one protein, rows represent residues from another protein, and elements represent the statistical dependencies between residues (Fig. 4A). By analyzing the inverse covariance matrix, we identified 13A-9B and 17A-31B as strongly interacting pairs, which supports their direct contact with each other in 3D structures. The top 5 highly correlated residues were close in the original structure, but the overall relationship between inter-residue distance and DCA score was weak (Fig. 4B). To clarify these inter-residue co-variations discovered from the sequence data, we determined crystal structures of 10 coevolved pairs that spanned a range of cross-reactivities and orthogonalities (5 from LL1 and 5 from LL2). All structures were solved at high resolution (ranging from 1.00 to 1.92 Å resolution) (Fig. S11-S13 and Tables S1 and S2). From the structures, we could verify clear compensatory changes between the residues showing the most significant covariations (13A-9B / 17A-31B) in both LL1 and LL2 library Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 8 mutants. Phe13B of the Z domain, which is a core residue of the central hydrophobic patch in the original dimer, was mutated to the smaller Ile or Val in both LL1 and LL2 library mutants, and this was compensated by mutation of the opposing residue Leu9A to the larger Phe (Fig. 4C). We also observed another highly correlated opposing residue pair (17A-31B) mutated in a compensatory manner in both libraries. Interestingly, here Leu17APhe is rotated outward, accommodating Ile31BLeu to fill the cavity between the two proteins (Fig. 4D). The extent of interface structural remodeling in all complexes due to the coevolution selection pressure is made clear in Fig. 4E, where the non-mutated residues Q10A and W35B accommodate the library mutations at positions 9A and 32B by adopting completely different positions and local environments (Fig. 4E). The two library positions (9A-32B) and proximal residues, Gln10A and Trp35B, kept close contact in all mutant structures, albeit with different interactions. The ability to rearrange at these positions allows decoupling of mutations despite close proximity (Fig. 4E). These results indicate that the protein interfaces of both specific and cross-reactive complexes were completely remodeled in different ways to improve affinities up to three logs (KD of LL1.c2 = 1.86 nM, original = 2.92μM) and bias specificities (KD of Z-ALL1.c4 (FILFK) with Z-BLL1.c4 (FIVM) = 2.53 nM, and with ZSPA-1 (LVIF) = 21.9μM). Cross-reactivity and orthogonality in coevolved dimer structures The availability of a large panel of coevolved mutants allows us to ask questions about their relative cross-reactivity versus specificity. For example, A-LILFK has more Z-B binding partners (B=77) than A-LIFFK (B=15) from the LL1 library, and A-LVLF (B=53) and A-VFLV (B=42) are also more cross-reactive than A-IVFF (B=3) from LL2 library sequence data. To answer the question of what causes differences in the cross-reactivity of certain Z-A sequences and to clarify specificity-determining residues, we compared high-affinity mutant structures of each Z-A sequence (Fig. 5). First, the binding preferences of the two highest affinity pairs from the LL1 library, LL1.c1 (LIFFK/FILF) and LL1.c2 (LILFK/FIVM), are nearly completely orthogonal, so we focused on investigating specificity-determining positions of the two variants (Fig. 5A and 5B). The two mutants differ by only three amino acid positions (positions 17A, 31B, and 32B) but have virtually no cross-reactivity with each other. Z-A sequences of LL1.c1 and LL1.c2 bind to their own Z-B sequence 500-fold stronger than when mixed with the other’s Z-B sequence (Fig. 5B). On the other hand, B-FIVF, which has only a single amino acid change from B-FILF or B-FIVM, has poly-specificity and binds to both Z-A sequences (A-LIFFK and A-LILFK) with moderate affinity (Fig. 5B). Thus, the mutant LL1.c6 (LILFK/FIVF) can be a “bridging” intermediate to help explain the structural evolution of orthogonality through cross-reactivity. Comparing structures of LL1.c2 and LL1.c6 revealed that the single mutation Met32BPhe in LL1.c6 induced a noticeable geometric change by forming an enhanced hydrophobic cluster within the binding interface (Leu9A, Leu13A, Lys35A, Phe5B, Phe32B, and Trp35B) (Fig. 5C). Due to the rotation of Trp35B and Trp35B-centered hydrophobic packing induced by Phe32B, the N-terminal end of helix 1 and C-terminal end of helix 2 of Z-A tilted 17° closer to Z-B. Furthermore, the compensatory relationship between position 17A and position 31B is clearly revealed from the structures of LL1.c1 and LL1.c2 (Fig. 4D). Taken together, the synergistic effects of the geometric change and compensatory mutations found from Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 9 these three specificity-determining positions resulted in biased specificities of the two high- affinity variants evolved from the same library. The monomers from the high affinity mutants from the LL2 library (three Z-A mutants and five Z-B mutants) are even more orthogonal to one another (Fig. S2F and S2G). The three orthogonal mutants, LL2.c17 (VFLV/IVVY), LL2.c7 (LVLF/FIVK) and LL2.c22 (IVFF/FILV), were selected to compare differences in their affinity and structures (Fig. 5D). Each Z-A mutant binds to its binding partner with nanomolar affinity (3.98 to 44.2 nM) but has minimal cross-reactivity with other monomers (Fig. 5E). The backbone structures of the three mutants are relatively similar (Cα r.m.s.d. of Z-A after aligning Z-B ranges from 0.447 Å to 0.641 Å). The dimer interactions of LL2.c17 have sharply diverged from the other two LL2 mutants, with the Phe13A-centered hydrophobic patch surrounded by multiple rewired hydrogen bonds, an additional hydrogen bond between Asn11A and Phe32BTyr, and α-helix 2 of Z-A was slightly shifted to generate new interactions with helix 1 of Z-B, which explains the significantly improved affinity between these two proteins compared to the original pair (Fig. 5F and S14). The other two mutants, LL2.c7 and c22, have clustered pi-pi and pi-cation interactions at the interface (Phe31A, Lys35A, Phe9B, and Trp35B) (Fig. 5F). Additionally, the same compensatory mutations (positions 17A and 31B) as LL1 mutants are also seen from LL2.c7 and c22 mutants (Fig. 4D). Contrary to the LL1 library, LL2 library mutants have less dramatic change in backbone orientation, but interfaces are more diverse due to the broader amino acids available to be mutated in library positions. We do not observe systematic differences in the structural parameters of the interfaces mediating specific (A-LIFFK in LL1 and A-IVFF in LL2) versus cross-reactive (A-LILFK in LL1 and A-VFLV, A-LVLF in LL2) complexes. All mutants except for one (LL2.c1) had a higher fraction of nonpolar BSA than the original dimer (58%), and all mutants except for one (LL2.c7) had a higher packing score (PackStat) than the original dimer (PackStat of Z/ZSPA-1 = 0.640) (Table S3). The cross-reactive complexes did not show evidence of poorly packed interfaces, non-ideal bonding that might predispose them to promiscuity; in this sense they are indistinguishable from protein interfaces of the specific complexes. Using protein language models to predict novel dimer interactions from co-evolved protein sequences The large database of coevolved complexes led us to ask if this information could be used to inform predictions through machine learning. One limitation of our experimental screen was that we used a limited set of amino acid codons in our experimental screen in order to fully sample the diversity of the yeast display libraries. But this raised the question of how to predict the binding affinity of larger diversity libraries containing more diverse amino acids without exceeding the practical diversity limits of the screening platform. One solution is the use of protein language models, which are self-supervised machine learning models pre- trained on large protein sequence databases (15, 26-28). We used protein language models to expand the set of amino acids in our screen through the process of transfer learning (Fig. 6A). Transfer learning involves applying knowledge gained from one problem to solve a related problem. By using a common protein language model to embed pairs of protein sequences, we can learn complex patterns that predict protein-protein interactions using a Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 10 limited set of amino acids and then apply this knowledge to predict binding affinity for novel pairs using a broader set of amino acids. Our two coevolution libraries, LL1 and LL2, used different subsets of amino acids to mutate library positions and yielded differently enriched sequences after screening (Fig. 2). The LL2 library (11 AA) has an expanded amino acid diversity compared to the LL1 library (8AA for Z-A and 5AA for Z-B), and only 3.2% of LL2 library sequencing data is LL1-type sequences (compatible with LL1 degenerate codon sets) while LL1 data has 40% of LL2-type sequences on average (Fig. 6B). Therefore, these two libraries are appropriate model systems to test how LL1 sequence data trained model can expand sequence space and predict new interactions only possible from LL2 sequence data. To determine if protein language models could be used to model Z-affibody pairs in our screens, we used the pre-trained ESM protein language model (16), which incorporates knowledge of all amino acids from large evolutionary sequence datasets, to generate embeddings of the sequences observed in LL1 (Fig. 6C). We predict dimer interactions using the outer product of individual protein sequence embeddings, and the resulting outer product matrix is then used as input into a convolutional neural network for further processing. This approach allows us to model complex interactions of protein-protein interface sequences, including those featuring an expanded set of amino acids than those used in our experimental screen. We first encoded the individual protein sequences from each protein pair in the LL1 screen and trained a deep neural network to identify positive interacting pairs. Positive interactions were defined as enriched or filtered protein pairs occurring significantly more often than expected in the enriched library (rounds 6 and 7), reflecting our methodology in the SSN and DCA analyses (see Methods). Negative interactions were defined as protein pairs that were present in the naïve library NGS data but absent in rounds 6 and 7. Given that each round captured 7-10 times more cells than the observed diversity of sequences in the naïve library, we reason the absence of these interactions is most likely due to being outcompeted during co-evolution. Next, we sought to evaluate the performance of the LL1-trained model in classifying held-out positive and negative LL2 interactions, which contain an extended amino acid library (Fig. 6D). The LL1-trained model could classify 5,565 LL2 sequences in our held-out test set (2,794 positive and 2,771 negative) with AUC of .88 (Fig. 6D). We then specifically examined the LL1-trained model’s ability to generalize by assessing its capacity to handle a progressively expanding amino acid library. We binned the LL2 test data based on the number of previously unused amino acids incorporated in the held-out test sequences (0, 1, 2, 3, 4 or more) relative to the LL1 sequence training data. These amino acids, which were not included in the LL1 training library, serve as indicators of the difference between the test and training sequences. Despite a decrease in performance trend as more amino acids are introduced, the LL1-trained model still achieves an AUC of 0.8 and an AP of 0.7, even when up to 3 out of 8 amino acids are not part of the LL1 training library. These tests demonstrate the model’s robustness in handling sequence variations and making reliable predictions. Finally, we applied our LL1-trained model to all LL2 screening rounds from naïve to final round 8 to assess the ability of the model to predict interactions of pairs in different Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 11 selection stages (Fig. 6E). The predicted binding scores of each round increases as screening proceeds, and the mean of the predicted scores at each round is highly correlated with actual %HA-tag fluorescence level after protease cleavage (Fig. 6E and 6F). We also compared the predicted scores with experimentally validated pairs in Fig. 3. The 28 validated interacting pairs (% of max HA-tag > 1) in Fig. 3G showed elevated predicted binding scores (Fig. 6E). The model could even moderately predict affinity changes between pairs along the mutational pathway in Fig. 3E (Fig. 6G). Even though two intermediates (VFLF+IIVY and VVLF+FIIY) were predicted to have higher affinities than their actual affinities, overall trends are similar between prediction and affinity throughout the pathway. We also evaluated the accuracy of our model in identifying hits among the top-ranked sequences. We conducted experimental validation on the binding of the 11 highest-ranked sequences and found that 6 out of the 11 sequences (hit rate = 54.5%) demonstrated affinities within the detectable range (submicromolar) as confirmed by the on-yeast cleavage-capture assay (Fig. 6H). These data demonstrate that we can use a protein language model to expand sequence space from the experimental sequence data of LL1 and predict the new interactions that we observed from LL2 screening data (Fig. 6I). Discussion We have developed a facile method for protein-protein coevolution as a solution to the problem of linking phenotype to genotype in large-scale library-on-library selections (29-31). The large collection of interacting Z-domain/affibody pairs we generated enabled a systems-level structure-function analysis of molecular recognition within this model system. We observed important characteristics of natural protein-level coevolution, including compensatory mutations between residues and hydrophobic core repacking. Acquiring compensatory mutations between directly interacting proteins is the simplest molecular mechanism that can cause epistasis between two genes (32, 33). Based on direct coupling analysis and high-resolution crystal structures, we could successfully infer epistatic interactions between Z domain-affibody dimer interfaces. The crystal structures of coevolved mutants revealed that when a key residue of the original central hydrophobic patch, Phe13A, was mutated to smaller amino acids like leucine or valine, Phe9B or Trp35B newly form the core of the central hydrophobic patch, presumably rearranging an existing hot spot or creating new ones. We infer that coevolving contact residues can fundamentally change binding interfaces to have different specificities and affinities by reinforcing or rearranging hot spots. The remodeling of the dimer interface of the Z domain and affibody was similar to the repacking of the hydrophobic core of widely-studied proteins such as Rop, T4 lysozyme, and λ Repressor-GCN4 Leucine Zipper Fusions (34-38). Thus, interface coevolution appears to follow principles of protein core repacking (39-42). In the course of our experimental studies of coevolution we identified a challenge in the form of a curse of dimensionality, where an exponential increase in experimental data is needed to test protein interactions as the number of mutating positions and amino acid alphabet increases. This issue is a major practical limitation to protein engineering using combinatorial libraries because full diversity libraries exceed the experimental diversity possible in yeast, phage, or ribosome display. To address this challenge, we used protein language models. Previously, protein language models have been limited to predicting Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 12 monomeric properties, and fine-grained variant effect analysis of protein-protein interactions has been difficult to evaluate due to a lack of data. Here, we demonstrate that by leveraging a shared sequence space learned from large-scale protein sequence databases, we can both extract informative representations of protein sequences and model their binding interactions. The amino acid composition of a protein encodes the information required to determine not just its structure but also its ability to negotiate interactions with other functional partners. Therefore, by using the information encoded in the latent protein embedding space, we can explore a larger space of protein-protein interactions than what is experimentally available. This approach combined with transfer learning can reduce data requirements and provide reliable predictions of binding interactions. This synthetic coevolution strategy can potentially be used in biotechnology applications. Although AlphaFold and RoseTTAFold are useful for predicting 3D protein structures from the amino acid sequence, predicting de novo protein-protein interactions remains a challenge (43). The experimental data generated from our coevolution strategy can be used as training data for machine learning algorithms to expand sequence space much wider than what can be obtained experimentally and to predict protein-protein interactions. The one-pot production of a large set of protein pairs with different specificity and cross-reactivity is also useful for synthetic biology. Orthogonal interfaces are essential components to build reliable and predictable orthogonal gene circuits to avoid undesirable crosstalk with the host or other machinery (44, 45). Our synthetic coevolution strategy can generate user-designed orthogonal protein complexes for such applications. Materials and Methods Protein expression and purification The DNA plasmids encoding for each affibody were cloned into pET28, a bacterial expression vector. The vector includes the affibody gene with either only C-terminal His6- tag or biotin-acceptor peptide tag (BAP tag, GLNDIFEAQKIEW) followed by His6-tag between the NcoI and XhoI sites of pET28b (Novagen). To express affibody monomers, the vector was transformed into E. coli BL21 (DE3), and the cells were grown at 37°C in TB medium supplemented with 50mg/l kanamycin. At 0.6 OD600, 0.5mM isopropyl- β-D-thiogalactoside (IPTG) was added to induce protein expression and the cell culture was incubated for overnight at 30°C before harvest. The proteins were purified by Ni2+-NTA agarose column chromatography (Ni-NTA, Qiagen) followed by size-exclusion chromatography with a Superdex S75 10/300GL Increase column (GE Healthcare). The proteins were stored in HEPES buffered saline (HBS, 20mM HEPES pH 7.5, 150mM sodium chloride). Affibody proteins used for surface plasmon resonance experiments were site-specifically biotinylated at the C-terminal BAP tag using BirA ligase and re-purified by size-exclusion chromatography. Yeast display of single-chain Z domain-affibody dimers Single chain affibody dimers were displayed on the surface of yeast S. cerevisiae strain EBY100 (Invitrogen, cat. no. C839-00) by fusion to the C-terminus of the Aga2 protein. Affibody dimers connected with a GS-linker and 3C protease cleavage site in the middle Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 13 were inserted between an N-terminal cMyc epitope and a C-terminal HA tag. N-cMyc-ZA- linker-ZB-HA-C insert was cloned into the pCT302 vector (Addgene #41845). Competent yeast cells were electroporated with affibody plasmids and recovered in YPD (Sigma, cat. no. Y1375) at 30°C for an hour. Next, recovered cells were grown in SDCAA media (pH 4.5, 20 g dextrose, 6.7 g yeast nitrogen base, 5 g bactocasamino acids, 10.4 g sodium citrate and 6.4 g citric acid monohydrate dissolved in 1 liter of deionized H2O, supplemented with 10 ml of Gibco™ Penicinillin-Stereptomycin, 10,000 U/ml) to OD600 10, and the cultures were induced at 20°C for 24 hours by diluting to OD600 1.0 in SGCAA (prepared as SDCAA, but use 20g galactose instead of dextrose) (7). The display level of proteins was confirmed by staining the cells with an Alexa Fluor 488-labeled anti-cMyc antibody (Cell Signaling Technology, cat. no. 2279S) and Alexa Fluor 647-labeled anti-HA antibody (1:50 dilution; Cell Signaling Technology, cat. no. 3444S), and fluorescence was monitored by flow cytometry (Beckman Coulter, CytoFLEX). Yeast displayed libraries Details of library assembly, sequences, and selection protocols are provided in Supplementary Methods. On-yeast cleavage-capture assay For single clone cleavage-capture assay, colonies were picked from transformed EBY100 cells plated on SDCAA plate. 5 × 105 induced yeast cells were stained with an Alexa Fluor 488-labeled anti-cMyc antibody and Alexa Fluor 647-labeled anti-HA antibody (1:50 dilution). Antibody-stained cells were washed with MACS buffer (autoMACS® Running Buffer, Miltenyi, cat. no. 130-091-221), then incubated in 20 μL 3C protease cleavage solution (lab-made 3C protease was diluted to 0.4 mg/mL in MACS buffer) at 4°C. At each time point, 2 μL was sampled and diluted in ice-cold 100μL MACS buffer, and fluorescence was measured by flow cytometry. The measured mean fluorescence intensity (MFI) was divided by MFI before cleavage to gain % of max MFI to represent an affinity between two interacting proteins. Cross-reactivity Circos Plots Circos plots were created via the circlize software package (48). In short, sequences with p-value < 0.05 were combined into separate data sets for LL1 and LL2 and further separated by screening round. A cross-reactivity score was calculated for each unique Z-A sequence by determining the number of its unique Z-B pairs per data set. Cross-reacitvity scores were then normalized to sum to 1. Finally, to facilitate visualization via circos plots, the data set was subsetted using the ‘train_test_split’ function of the python scikit-learn (version 1.2.2) package. To maintain the proportion of Z-A cross reactivity, the ‘stratify’ option was applied to the cross-reactivity score. Sequence Similarity Network, cluster graph and Specificity Similarity Network Sequence similarity networks and cluster graphs were created via the igraph software package (49). Nodes of the edit distance-based networks correspond to unique Z-A/Z-B pairs. Connections are present between nodes for instances in which the edit distance of Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 14 two Z-A/Z-B pairs is below a given threshold. Nodes of the Specificity Similarity Network correspond to unique Z-A sequences and connections are drawn between Z-A sequences when Z-A sequences share common Z-B sequences numbering above a certain threshold. Mutual Information To measure the coevolution relationship among interface residues, we computed the mutual information (MI) between two positions i, j as MIij = ∑AB f(Ai, Bj)log( f(Ai, Bj) f(Ai)f(Bj) ) following (Dunn et al., 2005) (48), where f(Ai, Bj) is the observed frequency of the amino acid pair (A, B) at position i, j, f(Ai) is the observed frequency of amino acid A at position i, and f(Bj) is the observed frequency of amino acid B at position j. Inverse covariance matrix AB = f(Ai , Bj) − f(Ai)f(Bj) where f(Ai , Bj) is the observed frequency of amino acid pair A, To uncover direct coupling signals from the MSAs, we used a method based on the estimation of the inverse covariance matrix following (Jones et al., 2012) (25). For position i, j and amino acid pair A, B we compute the empirical covariance matrix as Sij B at position i, j. f(Ai), f(Bj) are the observed frequency of amino acid A at position i and the observed frequency of amino acid B at position j respectively. Then we use the Graphical Lasso to estimate the inverse covariance matrix θ by maximizing the objective function log(det(θ)) − ∑ij = 1 ∣ θij ∣ ≤ α and θ ≻ 0 where Sijθij subject to the constraints ∑ij = 1 S is the empirical covariance matrix, θ is the inverse covariance matrix and α is the sparsity constraint parameter. We set α = 1 in all of our analysis. The optimization is performed with CVXPY v1.2 package in python. d d Data To train our deep learning model, we assembled positive and negative protein-protein pair examples from the oligopeptide pair dataset from the LL1 library. For enriched samples, we filtered the intermediate enriched library and applied the hypergeometric test described in Sequence library filter with a 0.05 p-value threshold, resulting in 14,491 pairs. For naive samples, we randomly sampled 14,471 pairs from the naive library that were not present in the intermediate enriched library. We then randomly split the data into training and validation sets with 80% and 20%, respectively. For the LL2 library, we applied the same method, resulting in 2,794 enriched and 2,763 naive samples as our test set. We also normalized the sequencing counts for our training label such that all naive samples scored 0 and all positive pairs are scored according to their observed sequencing counts then Min-Max normalized as Xs = log(count(X) + 100) − log(2) log(maxcount) − log(2) . Note that we added 100 base counts to all positive pairs to distinguish them from the naive pairs after normalization. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 15 Protein Language Model embeddings For each oligopeptide pair, we used the full chain sequence with the corresponding amino acid in the mutant position as sequence input to the protein language model for the latent vector representation generation. The vector representation is taken as the average position- wise embedding from the last layer of the protein language model with 1,280 dimensions. For each pair, we generate the sequence embeddings for each chain separately as V a, V b, and the outer product is computed across the vector representation of the two respective chains as a two-dimensional matrix representation for each oligopeptide pair as V ab = V a ⊗ V b. Model Architecture We designed and implemented a 3-layer 2D CNN model with kernel size (5,5) and channel size [64,128,256] followed by a two-layer fully connected network to predict the binding score of the input oligopeptide pairs. The model takes the two-dimensional oligopeptide pair representation V ab as the input and outputs a scalar P ab as the binding score. We also apply a max pooling layer and instance norm in-between each CNN layer. W l = InstanceNorm(ReLU(Maxpool(2Dconv(W l − 1)))) where W 0 = V ab P ab = F C(flatten(W f)) where W f is the output of the last CNN layer. We also applied sigmoid transformation to the fully connected network output for scaling. Sigmoid(X) = 1 1 − e−x Model training and testing All models are trained with squared L2 norm loss and the Adam optimizer with learning rate of 1e−4 on a NVIDIA 2080Ti machine for 100 epochs with the best saved checkpoint. Our implementation uses the PyTorch V1.11 compiled with CUDA 10.2. X-ray crystallography Details of crystallization and structure determination are provided in Supplementary Methods along with structure statistics Tables S1-S2. Surface plasmon resonance Dissociation constants (KD) of affibody dimers were acquired by surface plasmon resonance (SPR) using the BIAcore T100 instrument (GE Healthcare). Approximately 100 resonance units (RU) of biotinylated affibody were captured on a streptavidin-coated (SA) sensor chip (Cytiva), including a reference channel with an unrelated protein. HBS-P+ (Cytiva) was used for all SPR runs. All measurements were made with two-fold serial dilutions using 60-120 s association and 300-500 s dissociation at a flow rate of 30-50 μl/min. Regeneration Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 16 was performed using 0.02% SDS or 0.1M glycine, pH 2.5 after each analyte injection. The sensorgrams obtained were either fit to the 1:1 binding model or the steady-state affinity model using the BIAcore T100 evaluation software. Isothermal titration calorimetry For isothermal titration calorimetry experiments, proteins were dialyzed overnight against HBS buffer. After dialysis, concentrations were measured using the BCA assay kit (Thermo Fisher). Titrations of all mutants were performed in a Microcal VP-ITC instrument at 298 K with ZSPA-1 variants in the cell at 5 μM and Z variants in the syringe at 7-10× the cell concentration. The parent ZSPA-1 protein was used in the cell at 50 μM, with the parent Z protein in the syringe at 350 μM. Baseline subtraction was performed by titrating Z variants or Z parent into the dialysis buffer. All data were analyzed in Origin 7.0, fit to a 1-site model by fitting ΔH, K a, and the number of binding sites (n). Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgments: We thank X. Yang, M. Yen, R. A. Fernandes, D. Waghray, A. Velasco for their support. Use of the Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory, is supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences under Contract No. DE-AC02-76SF00515. The SSRL Structural Molecular Biology Program is supported by the DOE Office of Biological and Environmental Research, and by the National Institutes of Health, National Institute of General Medical Sciences (P30GM133894). The Berkeley Center for Structural Biology is supported in part by the Howard Hughes Medical Institute. The Advanced Light Source is a Department of Energy Office of Science User Facility under Contract No. DE-AC02-05CH11231. Funding: This work was supported by The Howard Hughes Medical Institute (KCG) The Emerson Collective (KCG) The Human Frontier Science Program (AY) References and notes 1. Darwin C , On the various contrivances by which British and foreign orchids are fertilised by insects, and on the good effects of intercrossing. (Murray J, London, 1862). 2. de Juan D, Pazos F, Valencia A, Nat. Rev. Genet 14, 249–261 (2013). [PubMed: 23458856] 3. Lockless SW, Ranganathan R, Science. 286, 295–299 (1999). [PubMed: 10514373] 4. Aakre CD et al., Cell. 163, 594–606 (2015). [PubMed: 26478181] 5. Zhang X, Perica T, Teichmann SA, Curr Opin Struct. Biol 6, 954–963 (2015). 6. Pazos F, Helmer-Citterich M, Ausiello G, Valencia A, J. Mol. Biol 271, 511–523 (1997). [PubMed: 9281423] 7. Chao G et al., Nat. Protoc 1, 755–768 (2006). [PubMed: 17406305] 8. Bowley DR, Jones TM, Burton DR, Lerner RA, Proc. Natl. Acad. Sci 106, 1380–1385 (2009). [PubMed: 19139405] 9. Wang L, Lan X, Cell Discov. 8, 30 (2022). [PubMed: 35379810] Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 17 10. Younger D, Berger S, Baker D, Klavins E, Proc. Natl. Acad. Sci 114, 12166–12171 (2017). [PubMed: 29087945] 11. Pelletier JN, Arndt KM, Plückthun A, Michnick SW, Nat. Biotechnol 17, 683–690 (1999). [PubMed: 10404162] 12. Arndt KM et al., J. Mol. Biol 295, 627–639 (2000). [PubMed: 10623552] 13. Yang F et al., Nucleic Acids Res. 46, 1–12 (2018). [PubMed: 29177436] 14. Siau JW et al., Nucleic Acids Res. 48, E128 (2020). [PubMed: 33104786] 15. Bepler T, Berger B, Cell Syst. 12, 654–669.e3 (2021). [PubMed: 34139171] 16. Lin Z et al., Science. 379, 1123–1130 (2023). [PubMed: 36927031] 17. Gebauer M, Skerra A, Annu. Rev. Pharmacol. Toxicol 60, 391–415 (2020). [PubMed: 31914898] 18. Lindborg M et al., Protein Eng. Des. Sel 26, 635–644 (2013). [PubMed: 23924760] 19. Eklund M, Axelsson L, Uhlén M, Nygren PÅ, Proteins Struct. Funct. Genet 48, 454–462 (2002). [PubMed: 12112671] 20. Högbom M, Eklund M, Åke Nygren P, Nordlund P, Proc. Natl. Acad. Sci 100, 3191–3196 (2003). [PubMed: 12604795] 21. Atkinson HJ, Morris JH, Ferrin TE, Babbitt PC, PLoS One. 4 (2009), doi:10.1371/ journal.pone.0004345. 22. Ma B, Elkayam T, Wolfson H, Nussinov R, Proc. Natl. Acad. Sci 100, 5772–5777 (2003). [PubMed: 12730379] 23. Rajamani D, Thiel S, Vajda S, Camacho CJ, Proc. Natl. Acad. Sci 101, 11287–11292 (2004). [PubMed: 15269345] 24. Hopf TA et al., Elife. 3, 713–724 (2014). 25. Jones DT, Buchan DWA, Cozzetto D, Pontil M, Bioinformatics. 28, 184–190 (2012). [PubMed: 22101153] 26. Chowdhury R et al., Nat. Biotechnol 40, 1617–1623 (2022). [PubMed: 36192636] 27. Meier J et al., Adv. Neural Inf. Process. Syst 35, 29287–29303 (2021). 28. Gligorijević V et al., Nat. Commun 12 (2021), doi:10.1038/s41467-021-23303-9. 29. Packer MS, Liu DR, Nat. Rev. Genet 16, 379–394 (2015). [PubMed: 26055155] 30. Leemhuis H, Stein V, Griffiths AD, Hollfelder F, Curr. Opin. Struct. Biol 15, 472–478 (2005). [PubMed: 16043338] 31. Tizei PAG, Csibra E, Torres L, Pinheiro VB, Biochem. Soc. Trans 44, 1165–1175 (2016). [PubMed: 27528765] 32. Lehner B, Trends Genet. 27, 323–331 (2011). [PubMed: 21684621] 33. Gregoret LM, Sauer RT, Proc. Natl. Acad. Sci. U. S. A 90, 4246–4250 (1993). [PubMed: 8483940] 34. Munson M, Regan L, O’Brien R, Sturtevant JM, Protein Sci. 3, 2015–2022 (1994). [PubMed: 7535612] 35. Willis MA, Bishop B, Regan L, Brunger AT, Structure. 8, 1319–1328 (2000). [PubMed: 11188696] 36. Hu JC, O’Shea EK, Kim PS, Sauer RT, Science. 250, 1400–1403 (1990). [PubMed: 2147779] 37. Matsumura M, Becktel WJ, Matthews BW, Nature. 334, 406–410 (1988). [PubMed: 3405287] 38. Xu J, Baase WA, Baldwin E, Matthews BW, Protein Sci. 7, 158–177 (1998). [PubMed: 9514271] 39. Sandberg WS, Terwilliger TC, Science. 245, 54–57 (1989). [PubMed: 2787053] 40. Ho SP, DeGrado WF, J. Am. Chem. Soc 109, 6751–6758 (1987). 41. Regan L, Degrado WF, Science. 241, 976–978 (1988). [PubMed: 3043666] 42. Huang PS et al., Science. 346, 481–485 (2014). [PubMed: 25342806] 43. Marchand A, Van Hall-Beauvais AK, Correia BE, Curr. Opin. Struct. Biol 74, 102370 (2022). [PubMed: 35405427] 44. Costello A, Badran AH, Trends Biotechnol. 39, 59–71 (2021). [PubMed: 32586633] 45. Chen Z, Elowitz MB, Cell. 184, 2284–2301 (2021). [PubMed: 33848464] 46. Yang et al. NGS Data for "Deploying synthetic coevolution and machine learning to engineer protein-protein interactions", Dryad, DOI: 10.5061/dryad.gf1vhhmv7 Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 18 47. Lai B and Khan AA, Data and code for "Deploying synthetic coevolution and machine learning to engineer protein-protein interactions" Zenodo, DOI: 10.5281/zenodo.8035449 48. Gu Z, Gu L, Eils R, Schlesner M, Brors B, Bioinformatics. 30, 2811–2812 (2014). [PubMed: 24930139] 49. Csardi G, Nepusz T, InterJournal. Complex Sy, 1695 (2006). 50. Gloor GB, Martin LC, Wahl LM, Dunn SD, Biochemistry. 44, 7156–7165 (2005). [PubMed: 15882054] 51. Walter TS et al., Structure. 14, 1617–1622 (2006). [PubMed: 17098187] 52. Kabsch W, Acta Crystallogr. Sect. D Biol. Crystallogr 66, 125–132 (2010). [PubMed: 20124692] 53. Winter G et al., Acta Crystallogr. Sect. D, Struct. Biol 74, 85–97 (2018). [PubMed: 29533234] 54. Winn MD et al., Acta Crystallogr. Sect. D Biol. Crystallogr 67, 235–242 (2011). [PubMed: 21460441] 55. Evans PR, Murshudov GN, Acta Crystallogr. Sect. D Biol. Crystallogr 69, 1204–1214 (2013). [PubMed: 23793146] 56. Evans PR, Acta Crystallogr. Sect. D Biol. Crystallogr 67, 282–292 (2011). [PubMed: 21460446] 57. McCoy AJ et al., J. Appl. Crystallogr 40, 658–674 (2007). [PubMed: 19461840] 58. Terwilliger TC et al., Acta Crystallogr D Biol Crystallogr. 64, 61–69 (2008). [PubMed: 18094468] 59. Emsley P, Lohkamp B, Scott WG, Cowtan K, Acta Crystallogr. Sect. D Biol. Crystallogr 66, 486–501 (2010). [PubMed: 20383002] 60. Echols N et al., J. Appl. Cryst 45, 581–586 (2012). [PubMed: 22675231] 61. V Afonine P et al., Acta Crystallogr. Sect. D Biol. Crystallogr 68, 352–367 (2012). [PubMed: 22505256] 62. Liebschner D et al., Acta Crystallogr. Sect. D Struct. Biol 75, 861–877 (2019). [PubMed: 31588918] 63. Bricogne G et al., Cambridge, United Kingdom Glob. Phasing Ltd (2017). 64. Headd JJ et al., Acta Crystallogr. Sect. D Biol. Crystallogr 70, 1346–1356 (2014). [PubMed: 24816103] 65. Karplus PA, Diederichs K, Science. 336, 1030–1033 (2012). [PubMed: 22628654] 66. Chen VB et al., Acta Crystallogr. Sect. D Biol. Crystallogr 66, 12–21 (2010). [PubMed: 20057044] 67. Sheffler W, Baker D, Protein Sci. 18, 229–239 (2009). [PubMed: 19177366] 68. Morin A et al., Elife. 2 (2013), doi:10.7554/eLife.01456. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 19 Figure 1. Design and validation of protein-protein coevolution strategy. (A) A schematic representation of protein-protein coevolution workflow. The α-agglutinin yeast surface display system was used to display two proteins connected by a flexible linker. A 3C protease site within the linker enabled cleavage, and the interacting proteins can be captured by C-terminally bound anti-HA antibody (red). (B) Close-up view of key residues in the hydrophobic cavity of Z domain (green) and affibody ZpA963 (blue) (PDB: 2M5A). Encoded amino acids are used for two separate libraries, HL1 and HL2 (bottom). (C) On-yeast cleavage-capture assay of interacting pair (Z+ZpA963) and non-interacting pair (6xAla). Data are mean ± SD; n = 3 independent replicates. (D) Correlation between on-yeast cleavage-capture assay and binding affinity measured of Z domain-affibody dimer mutants measured by SPR. Note that on-yeast cleavage-capture assay shows a strong semilog-linear relationship (R2 = 0.8382) with binding affinity (pKD). (E) Histogram of the flow cytometric analysis. Note that HA-tag fluorescence in the library shows strong enrichment after MACS (PM) and FACS (PF) for HL1 and HL2 libraries. (F) Sequence frequency logo of NGS data in the naïve library and post final round of FACS. The original sequence (FLI+FIL) is derived from Z domain (A) and ZpA963 (B) dimer. Note that the libraries converged back to the original sequences either exactly or with minimal variations. The color scheme represents hydrophobic (black), polar (green), basic (blue), acidic (red), and neutral (purple) amino acids. (G) On-yeast cleavage-capture assay of the six most frequent mutants from HL1 and HL2 NGS data. The sequence of each mutant (1: FII+FIL, 2: FLI+FIL, 3: FII+FVL, 4: FLI+FVL, 5: FLI+FII, 6: FII+FII) Note that all six mutants show different levels of steady-state binding of HA-tag fluorescence during 3C protease cleavage. Data are mean ± SD; n = 3 independent replicates. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 20 Figure 2. Engineering remodeled dimer interfaces by coevolution. (A) Library positions on the interface (top) from the complex of Z domain (green, chain A) and ZSPA-1 (blue, chain B) (PDB: 1LP1). Encoded amino acids used for making two separate libraries, LL1 and LL2 (bottom). (B) Flow cytometry dot plots showing enrichment of HA-tag fluorescence (red squares) in the library after rounds 6 to 8 (left). Antibody-labeled yeast cells were cleaved with 3C protease for 30 min. Cells were pre-gated on c-Myc+. Histograms showing elevation of HA-tag fluorescence during selection, from round 6 (green), to 7 (blue) and 8 (red) (right). (C) Sequence frequency logo of NGS data in naïve library, rounds 6, 7, and 8, revealing the appearance of consensus sequences as the selection proceeded in both LL1 and LL2 libraries. The original sequence (QFLIK+LVIF) is derived from Z domain (A) and ZSPA-1 (B) dimer. The color scheme represents hydrophobic (black), polar (green), basic (blue), acidic (red), and neutral (purple) amino acids. (D) On-yeast cleavage-capture assay of the mutants from LL1 (left) and LL2 (right) library. The altered positions compared to original amino acids are colored in red. Data are mean ± SD; n = 3 independent replicates. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 21 Figure 3. Visualization and mapping of coevolutionary networks. (A) The sequence logo of Z-B sequences paired with each Z-A sequence from the statistically enriched NGS data (p-value < 0.05) and actual binding specificity measured by on-yeast cleavage-capture assay, normalized to the highest affinity of each Z-A sequence (below). Filtered sequences accurately predicted binding specificity, matching the actual binding specificity of each Z-A sequence. (B) Sequence similarity networks (SSNs) of concatenated 8 amino acid Z-A/Z-B library position sequences from all screening rounds (left) and round 7 (right) of LL2 library. Notable Z-A sequences are colored and specified in the panel (right). The edit distance threshold for connecting nodes in the total library network is 2 and in the round 7 network is 1. The left SSN is colored by screening round and demonstrates connectivity among sequences from later screening rounds (rounds 5 to 7). The right SSN is colored by Z-A sequence and provides a detailed view of the enriched stage (round 7), showing cluster formation based on Z-A specificities. (C) Circos cross-reactivity plot of 100 sampled pairs from LL1 and LL2 round 7 sequence data. The Circos plots illustrate the pairwise relationships between the 100 sampled pairs of Z-A and Z-B proteins. Each pair is normalized to have equal area, providing a visual representation of the approximate cross-reactivity of each sequence. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 22 (D) A single mutational pathway of mutants from the LL2 library connecting the original sequence (QFLI/LVIF) with the prominent LL2 library mutants. Mutated positions are color-coded: red (one mutation), green (two mutations), and blue (three mutations). The number of mutations at each position is represented by a 4-digit number next to each Z-A and Z-B sequence (E) A plot illustrating the changes in Δ pathway (D) compared to the original pair (QFLI/LVIF). Mutations introduced in each step are highlighted in red. (F) A matrix to show binding specificity changes of the Z-A variants from the pathway. Binding affinities measured by on-yeast cleavage-capture assay were normalized based on the highest affinity in each Z-A sequence. The single mutation introduced at each step is indicated in red. The highest affinity pair in each column was boxed in green. Control is a mutant with all library positions mutated to alanines. Data are mean of n = 3 independent replicates. . ΔH, and −ΔTΔS for three mutants in the ΔG, Δ Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 23 Figure 4. Coupling analysis and structural adaptation of coevolved variants. (A) DCA matrix to predict inter-residue covariation of LL2 library sequences (round 6 and 7). The DCA scores are normalized between 0 and 1. The pairs with the highest DCA scores, 13A-9B and 17A-31B, are marked with red squares. The matrix rows represent residues from Z-A, columns represent residues from Z-B, and the elements represent the statistical dependencies between residues. Through the inverse covariance matrix analysis, the pairs 13A-9B and 17A-31B were identified as strongly interacting pairs, indicating their direct contact in the 3D structures. (B) Inter-residue contacts (left), and the relationship between DCA and inter-residue distance is measured from the original pair structure (right) (PDB: 1LP1). The dashed lines are color-coded (from purple to yellow) based on DCA matrix in panel (A). The top two highest DCA contacts (Leu 17A – Ile 31B, Phe 13A – Leu 9B) are colored in red. The overall relationship between inter-residue distance and DCA score was weak (R2 = 0.0203). (C-E) Close-up views of library positions to show local side chain rearrangements. Pairs of residues at the center of the dimer interface were mutated in a compensatory manner between 13A and 9B (C) and between 17A and 31B (D). Side chain substitutions from 4 different interacting pairs are shown as sticks (E) Library positions 9A and 32B are closely associated with proximal residues, Gln10A and Trp35B, maintaining the shape complementarity between two proteins. In the bottom left, B chains of seven interacting Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 24 pairs are aligned, with close up views of the boxed region shown for each pair. Coupled side chains are shown as sticks with transparent spheres to indicate packing interactions. Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 25 Figure 5. Specificity determinants of orthogonal high-affinity mutants. (A) The altered positions compared to the original amino acids are colored red, and varying positions between mutants are highlighted with green boxes. (B) A table of affinity between Z-A and Z-B monomers measured by SPR. LL1.c1 and c2 are orthogonal to each other and B-FIVF of LL1.c6 are cross-reactive to both Z-A mutants. (C) Comparison of LL1.c2 and LL1.c6 structures near position 32B shows how the single mutation M32BF induces large conformational changes by side chain rotation of Trp35B and increased hydrophobic interactions around it. Superposition of overall structures of LL1.c1, LL1.c2 and LL1.c6 (left). Close-up views of each mutant show Trp35-centered hydrophobic interactions with surrounding residues (right). Position 32 is highlighted with dashed circles. (D) A table showing amino acids in library positions of the three orthogonal LL2 mutants, LL2.c17 (VFLV/IVVY), LL2.c7 (LVLF/FIVK) and LL2.c22 (IVFF/FILV), that were selected to compare differences in their affinity and structures. (E) Binding affinities of each combination of Z-A and Z-B mutants of the three mutants. (F) Significant structural difference at the interface of LL2.c17 and other two mutants. Superposition of overall structures (left). Close-up views of interface (right). LL2.c17 has Phe13A as the core of a central hydrophobic patch surrounded by multiple hydrogen bonds. LL2.c7 and c22 have a Phe9B-centered hydrophobic patch composed of clustered pi-pi interactions and cation-pi interactions (F31A, K35A, F9B, and W35B). Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 26 Figure 6. Sequence space expansion using protein language model (A) A schematic representation of sequence space expansion through protein language model. (B) The fraction of LL1-type sequences (the Z-A and Z-B sequences can be encoded with LL1 degenerate codon sets) in LL2 sequencing data and vice versa. Fractions of each screening round (from naïve to R8) were represented in a Box plot with individual data points. A two-tailed Mann–Whitney test was used to analyze results. P < 0.001. (C) A schematic representation of our approach to predict dimer interactions with expanded set of amino acids using outer product-based convolutional neural network. (D) The classification efficiency of LL1-trained model on LL2 test set. (left) A violin plot representing predicted binding score of negative (n = 2,771) and positive (n = 2,794) data. Two-tailed Mann–Whitney test. P < 0.0001. (middle) A ROC plot and (right) a PR plot. Note that the sequences in test set were categorized into five groups based on the number of new amino acids compared to the LL1 sequence data, allowing an assessment of the impact of dissimilarity between the two libraries on predictions. The AUC (Area Under the ROC curve) and AP (Average Precision) values of total sequences and each subgroup are: all sequences (n = 5,565, AUC = 0.88, AP = 0.89), 0 AA (n = 508, AUC = 0.88, AP = 0.98), 1 AA (n = 1,332, AUC = 0.91, AP = 0.97), 2 AA (n = 1,509, AUC = 0.84, AP = 0.87), Science. Author manuscript; available in PMC 2023 August 04. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Yang et al. Page 27 3 AA (n = 1,091, AUC = 0.80, AP = 0.70), 4 and more AA (n = 1,125, AUC = 0.73, AP = 0.32). The diagonal dotted line in ROC plot represents AUC = 0.5. (E) The predicted binding scores of LL2 sequencing data of each screening round were represented in a violin plot. One-way ANOVA. P < 0.001, ***P < 0.0001. ns, not significant. (n = 28–10,000) (F) The correlation between predicted binding score of LL2 sequencing data and actual %HA-tag MFI after protease cleavage. Normalized %HA-tag MFI and predicted binding score of each round was compared by Spearman’s correlation test (r = 0.9643, P = 0.0028). Data are mean ± SD; n = 3 independent replicates for HA-tag MFI measurements. (G) The correlation between predicted binding score and relative affinity of the pairs from the mutational pathway in Fig. 6A. Normalized % of max HA-tag MFI from cleavage- capture assay and predicted binding score of each round was compared by Spearman’s correlation test (r = 0.5476, P = 0.0855). (H) Top 11 sequences by predicted binding score from LL2 NGS data. The binding of 6 out of the 11 sequences were verified by on-yeast cleavage-capture assay and their relative binding affinities were normalized to the high affinity LL2 pair, LL2.c3 (LVLF+FIIV). n.d. = not detectable affinity by the assay. (I) A cartoon representation depicting the expansion of sequence space from experimental LL1 data to the predicted LL2 sequence space using a protein language model and transfer learning. Science. Author manuscript; available in PMC 2023 August 04.
10.1371_journal.pbio.3002512	RESEARCH ARTICLE Cross-frequency coupling in cortico- hippocampal networks supports the maintenance of sequential auditory information in short-term memory Arthur Borderie1,2, Anne Caclin3, Jean-Philippe Lachaux3, Marcela Perrone-Bertollotti4, Roxane S. Hoyer1, Philippe Kahane5, He´ lène Catenoix3,6, Barbara Tillmann3,7, Philippe AlbouyID 1,2,3* 1 CERVO Brain Research Center, School of Psychology, Laval University, Que´ bec, Canada, 2 International Laboratory for Brain, Music and Sound Research (BRAMS), CRBLM, Montreal, Canada, 3 Universite´ Claude Bernard Lyon 1, CNRS, INSERM, Centre de Recherche en Neurosciences de Lyon CRNL U1028 UMR5292, Bron, France, 4 Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, LPNC, Grenoble, France, 5 Univ. Grenoble Alpes, Inserm, U1216, CHU Grenoble Alpes, Grenoble Institut Neurosciences, Grenoble, France, 6 Department of Functional Neurology and Epileptology, Lyon Civil Hospices, member of the ERN EpiCARE, and Lyon 1 University, Lyon, France, 7 Laboratory for Research on Learning and Development, LEAD– CNRS UMR5022, Universite´ de Bourgogne, Dijon, France * [email protected] Abstract AU : Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly: It has been suggested that cross-frequency coupling in cortico-hippocampal networks enables the maintenance of multiple visuo-spatial items in working memory. However, whether this mechanism acts as a global neural code for memory retention across sensory modalities remains to be demonstrated. Intracranial EEG data were recorded while drug- resistant patients with epilepsy performed a delayed matched-to-sample task with tone sequences. We manipulated task difficulty by varying the memory load and the duration of the silent retention period between the to-be-compared sequences. We show that the strength of theta-gamma phase amplitude coupling in the superior temporal sulcus, the infe- rior frontal gyrus, the inferior temporal gyrus, and the hippocampus (i) supports the short- term retention of auditory sequences; (ii) decodes correct and incorrect memory trials as revealed by machine learning analysis; and (iii) is positively correlated with individual short- term memory performance. Specifically, we show that successful task performance is asso- ciated with consistent phase coupling in these regions across participants, with gamma bursts restricted to specific theta phase ranges corresponding to higher levels of neural excitability. These findings highlight the role of cortico-hippocampal activity in auditory short-term memory and expand our knowledge about the role of cross-frequency coupling as a global biological mechanism for information processing, integration, and memory in the human brain. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Borderie A, Caclin A, Lachaux J-P, Perrone-Bertollotti M, Hoyer RS, Kahane P, et al. (2024) Cross-frequency coupling in cortico- hippocampal networks supports the maintenance of sequential auditory information in short-term memory. PLoS Biol 22(3): e3002512. https://doi. org/10.1371/journal.pbio.3002512 Academic Editor: Timothy D. Griffiths, Newcastle University Medical School, UNITED KINGDOM Received: May 23, 2023 Accepted: January 22, 2024 Published: March 5, 2024 Copyright: © 2024 Borderie et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Codes and preprocessed data are available at https://osf.io/ m7dta/. Note that raw SEEG and neuroimaging (T1-MPRAGE) data are protected and cannot be shared (CPP Sud-Est V, 2009-A00239-48). Funding: This work was conducted in the framework of the LabEx CeLyA ("Centre Lyonnais d’Acoustique", ANR-10-LABX-0060, https://celya. universite-lyon.fr/labex-celya-151124.kjsp) and of the LabEx Cortex ("Construction, Function and PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 1 / 24 Cognitive Function and Rehabilitation of the Cortex", ANR-11-LABX-0042, https://labex-cortex. universite-lyon.fr/) of Universite´ de Lyon, within the program "Investissements d’avenir" (ANR-11-IDEX- 0007, https://anr.fr/) operated by the French National Research Agency (ANR, https://anr.fr/). This work was supported a NSERC Discovery grant (https://www.nserc-crsng.gc.ca/) and a FRQS Junior 1 and 2 grants (https://frq.gouv.qc.ca/sante/ ) and a Brain Canada Future leaders Grant (https:// braincanada.ca/) to P.A. A.B. and R.S.H are funded by the CERVO Foundation (https://fondationcervo. com/, FRQS, https://frq.gouv.qc.ca/sante/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Cross-frequency coupling enables integration and memory of auditory information in the human brain Introduction It is well established that the medial temporal lobe, in particular the hippocampus, is involved in the formation of long-term memories (LTM; [1]). Notably, hippocampal lesions consis- tently entail LTM deficits (i.e., anterograde amnesia [2]). In contrast, numerous empirical data obtained with a variety of materials, such as words [3], digits [4,5], tones [5], or single-dot loca- tions [4], have led to the hypothesis that hippocampal lesions do not impact working memory (WM) and short-term memory (STM) functions [6,7]. These findings suggest that WM and STM functions rely on distinct processes from LTM (e.g., [8,9]; see also [10,11] for neuroimag- ing studies). However, this hypothesis has been challenged by (i) neuropsychological studies reporting that patients with hippocampal lesions experience difficulties in maintaining items in WM or STM [12–14]; and (ii) fMRI [15–17], intracranial EEG [18–21], or single-unit recordings [22,23] in humans reporting persistent, load-dependent, hippocampal activity during WM maintenance of visual information (see also [15] for evidence of hippocampal involvement during auditory STM and [24] for a review about hippocampal activity during general auditory processing). honest significant difference; Abbreviations: HSD, AU : Anabbreviationlisthasbeencompiledforthoseusedthroughoutthetext:Pleaseverifythatallentriesarecorrectlyabbreviated: IES, inverse efficiency score; IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; LMM, linear mixed model; LTM, long-term memory; PAC, phase amplitude coupling; PLV, phase locking value; RT, response time; STM, short-term memory; STS, superior temporal sulcus; SVM, support vector machine; WM, working memory. Hippocampal activity during WM and STM has been originally associated with mainte- nance-related increase of theta and gamma power [21,25–28]. Interestingly, recent studies went a step further by showing that successful visual memory performance requires the cou- pling of gamma activity to specific phases of the hippocampal theta (theta-gamma phase amplitude coupling (PAC) [29–32]). Theta-gamma PAC consists in gamma subcycles (local neural activity associated to the processing of each encoded item) that occur at specific theta phase ranges. It has been suggested that theta-gamma PAC plays a critical role in the mainte- nance of different items in memory and as well as their serial order [31–33]. To date, theta- gamma PAC has been observed in cortico-thalamo-cortical, cortico-cortical, and cortico-hip- pocampal networks for episodic, working, and long-term memory consolidation in the visual modality [28,34,35]. For the specific case of STM, hippocampal theta-gamma PAC has first been isolated with SEEG in a visual word recognition paradigm in humans: an increased syn- chronization between the phase of the theta band, and the power changes in the beta and gamma bands were observed when patients successfully remembered previously presented words [36]. Several studies have since confirmed the implication of PAC in STM and WM by showing that the simultaneous maintenance and/or manipulation of multiple visual items in memory is implemented under the form of hippocampal theta-gamma PAC [18,20,37,38]. Overall, previous results suggest that WM or STM maintenance, in which different items must be separately and sequentially maintained over a short period of time, is represented by an ordered activity of cell assemblies implemented under the form of theta-gamma PAC in human cortico-hippocampal networks [31]. However, to date, these studies have mainly focused on visuo-spatial processing, and very little is known about the potential role of theta- gamma PAC in auditory and hippocampal regions during the short-term retention of sequen- tial auditory information. Coupling across cortical oscillations of distinct frequencies in the auditory cortex has been assumed to enable the multiscale sensory analysis of speech (pho- nemes and syllables [39–41]). However, the direct contribution of auditory-hippocampal cross-frequency coupling for the short-term maintenance of sequential auditory information has not yet been demonstrated. In the present study, we recorded intracranial EEG data while drug-resistant patients with epilepsy performed a delayed matched-to-sample task with tone sequences. If theta-gamma PAC is a predictor of successful memory maintenance, its strength in the auditory and hippocampal regions should (i) be increased during short-term retention of tone sequences (as compared to simple perception); (ii) decode correct and incorrect PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 2 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain responses in the STM task using machine learning analysis; and, finally, (iii) be positively cor- related with individual auditory STM performance. Results Intracranial EEG recordings were obtained from 16 neurosurgical patients with focal drug- resistant epilepsy. The participants performed an auditory STM task, consisting in the compar- ison of tone sequences presented in pairs and separated by a silent retention period. In each block of the task, in 50% of the trials, the tone sequences were identical (expected response “same”) and 50% differed by one note (expected response “different”). To manipulate task dif- ficulty, in different conditions, we varied the memory load (3 or 6 to-be-encoded tones, with a tone duration of 250 ms) and the duration of the silent retention period between the to-be- compared sequences (2 s, 4 s, and 8 s; see Table 1 for a detailed description of the conditions and number of participants tested in each condition). Participants also performed a block of listening of the same trials with the instruction to not compare the tone sequences and were simply required to press a button as fast as possible at the end of the last tone of the second sequence (Perception task, 6 tones, 2 s silent period between the tone sequences; see Methods). Accuracy Task performance was evaluated using d prime (signal detection theory). To evaluate the impact of the duration of the silent retention period for 6-tone sequences, we performed a nonparametric repeated measures ANOVA (Friedman test) with duration (2 s, 4 s, and 8 s) as a within-participants factor (n = 6 participants, note that all participants did not perform all the tasks—see Table 1). The main effect of duration was significant χ2 (2) = 7.00, p = .03. Post hoc tests performed with Durbin–Conover pairwise comparisons revealed that performance in the 2 s duration condition was significantly better than performance in the 2 other duration conditions (4 s, p = 0.004; and 8 s, p = .03). Performance in the 4 s and 8 s conditions did not differ significantly (p = 0.24, Fig 1B, left panel). To evaluate the impact of memory load on accuracy (3 versusAU : PleasenotethatasperPLOSstyle; donotuse}vs:}exceptintablesandcaptions:Hence; allinstanceof }vs:}havebeenspelledoutto}versus}throughoutthetext: a Wilcoxon rank test revealing, as expected, that performance was increased for the 3-tone condition as compared to the 6-tone condition (W [5] = 21.0, p = 0.031; Fig 1B, right panel). 6 tones with a 4 s silent retention period, n = 6 participants), we performed Response times The same analyses were performed for response times of correct responses (RTsAU : PleasenotethatasperPLOSstyle; abbreviateanyinstanceofthefullword=phraseafterthefirstmention:Hence; allinstancesof }responsetime}or}responsetimes}havebeenchangedto}RT}or}RTs; }respectively: ; Fig 1C) in the same participants (n = 6). Nonparametric repeated measures ANOVA (Friedman test) Table 1. Description of the conditions. Conditions 6 tones—short retention 6 tones—medium retention 6 tones—long retention 3 tones—medium retention Task STM STM STM STM 6 tones -perception task Do not compare sequences and press 1 key at the end of the second sequence STM, short-term memory. https://doi.org/10.1371/journal.pbio.3002512.t001 Memory load Retention duration (s) Number of patients tested 6 tones (total sequence duration 1.5 s) 6 tones (total sequence duration 1.5 s) 6 tones (total sequence duration 1.5 s) 3 tones (total sequence duration 0.75 s) 6 tones (total sequence duration 2 4 8 4 2 16 6 16 6 16 1.5 s)AU : Pleaseconfirmthattheitalicized}6tonesðtotalsequenceduration1:5sÞ}underthe}Memoryload}columninTable1canbechangedtoregulartext: PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 3 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Fig 1. Paradigm, behavioral performance, and brain oscillations. (A) Auditory tasks (here with 6-tone sequences, 2 s retention): “Same” trials: After a delay, the first melody was repeated. “Different” trials: One tone was changed in the second melody of the pair in comparison to the first melody (red rectangle). Memory load (3 or 6 tones) and duration of the retention period (2, 4, 8 s) varied in separate blocks. Source data can be found at https://osf.io/m7dta/. (B) Accuracy in terms of d prime presented as a function of the duration of the retention period (left panel; N = 6) and memory load (right panel; N = 6). Colored circles depict participants (one color per participant). Asterisks indicate significance (p < 0.05, nonparametric tests; see text for details); NS, nonsignificant. Source data can be found at https:// osf.io/m7dta/. (C) Response time (s) presented as a function of the duration of the retention period (left panel; N = 6) and memory load (right panel; N = 6). Colored circles depict participants (one color per participant; same color coding as in Fig 1B). NS, nonsignificant. Source data can be found at https://osf.io/m7dta/. (D) Left panel: T-values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left Heschl’s gyrus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 5). Right panel shows the PSD, power spectrum density (zscore) average over a trial time window (0 to 5,000 ms) that was used to define frequency for phase and frequency for amplitude for the PAC analysis. Shaded error bars indicate SEM. Source data can be found at https://osf.io/m7dta/. (E) Left panel: T-values in the time- frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left hippocampus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 14). Right panel shows the PSD, power spectrum density (zscore) average over a trial time window (0 to 5,000 ms) that was used to define frequency for phase for the PAC analysis. Shaded error bars indicate SEM. Source data can be found at https://osf.io/m7dta/. (F) SEEG contacts modelled with 4 mm radius spheres (see Methods) in the MRI volume showing a significant increase in oscillatory power (FDR corrected) relative to baseline in theta (4 Hz) and gamma (30–90 Hz) ranges (Hilbert transform averaged over time) during encoding, retention, and retrieval in all memory conditions in all participants (n = 16). All results are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. https://doi.org/10.1371/journal.pbio.3002512.g001 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 4 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain with duration (2 s, 4 s, and 8 s) as a within-participants factor revealed that the main effect of duration of the silent retention period was not significant χ2 (2) = 0.33, p = .84. In addition, Wilcoxon rank test revealed no significant difference of RTs between the 3-tone condition and the 6-tone condition (4 s silent retention period, W [5] = 4.00, p = .21; Fig 1C, right panel). Spectral fingerprints of perception and short-term memory of auditory sequences Fig 1D and 1E show the oscillatory activity (t test relative to the baseline −1,000 to 0 ms before stimulus onset, FDR corrected in time and frequency) in the time-frequency domain for SEEG contacts located in the left and right Heschl’s Gyri (according to the AAL3 atlas; see Methods, Fig 1D, 9 SEEG contacts, n = 5 participants with one electrode in this area, S1 Table) and bilat- eral hippocampal and para-hippocampal regions (Fig 1E, 72 SEEG contacts, n = 14 partici- pants with one electrode in these areas, S2 Table) for a trial time window for the 6-tone condition, 2 s retention period. Note that the same figures using a logarithmic scale for the fre- quency axis are presented in S1 Fig. In the auditory cortex, for each tone during the encoding and retrieval periods, transient gamma activity (30 to 90 Hz) was observed. As expected, the encoding of the entire sequence in the auditory cortex was associated with sustained theta oscillations at 4 Hz (tone presentation rate) and at 8 Hz (harmonic; Fig 1D). Moreover, a sig- nificant alpha/beta (10 to 20 Hz) desynchronization (relative to baseline) was observed in the auditory cortex during encoding, retrieval, and at the beginning of the retention period (Fig 1D). In the hippocampal and para-hippocampal regions, sustained theta oscillations (4 to 8 Hz) were observed during the entire trial time window (Figs 1E and S1). We then aimed to evaluate the fluctuations of power relative to baseline in these frequency bands for all SEEG contacts in all participants and all memory conditions. We used Hilbert’s transform (to reduce the dimension of the data) to extract the magnitude of theta (4 Hz) and gamma (30 to 90 Hz) oscillations during encoding, retention, and retrieval periods of the dif- ferent conditions (averaged in time; see Table 1 for the relevant time periods) for each partici- pant, each SEEG contact, and each trial. A contrast with baseline (FDR corrected) revealed that gamma activity was increased bilaterally in primary and secondary auditory regions and in the hippocampus during encoding retention and retrieval (Fig 1F, top panel; see SupportingAU : PleasenotethatPLOSusestheterm}Supportinginformation:}Hence; }supplementaryinformation}hasbeenreplacedwith}Supportinginformation}throughoutthetext: information for details and coordinates). During memory retention, an increase in theta activity was observed in a distributed net- work including the hippocampal/para-hippocampal regions, inferior frontal gyrus, and several regions of the ventral auditory stream (see Supporting information for details and coordinates; Fig 1F, bottom panel). To investigate whether these fluctuations of oscillatory power were specific to the memory task, we contrasted memory trials (6 tones, 2 s silent retention delay) with perception trials (6 tones, 2 s silent delay) for each frequency band (theta, gamma) and for all time periods (encod- ing, retention, retrieval; note that period names apply to the memory task) with nonparametric permutation tests (see Methods and supporting results). To assess significance, we applied a cluster-based approach: We defined SEEG contacts as significant only when they were overlap- ping for at least 2 participants or 2 SEEG contacts (overlap estimated on an MRI volume where SEEG contacts are represented by spheres with a radius of 4 mm; see Methods). This analysis did not reveal any significant effect for the contrast memory versus perception for each of the periods of the task (encoding, retention, retrieval), all p-values > .05 (see S2 Fig plotting theta and gamma power for memory and perception conditions in all SEEG contacts located in regions showing increased theta and gamma power relative to baseline during the retention period). PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 5 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Theta-gamma PAC is associated with auditory STM retention Notwithstanding the fact that no effect was observed for the memory versus perception con- trast on theta and gamma power, we investigated whether theta-gamma PAC during memory retention could rather be a more specific marker of STM retention. For all PAC analyses, we adopted the following strategy: All analyses, except the memory versus perception contrast (see Table 1 and Fig 2), were done within subject, for all participants, using all data of the memory conditions. We then report only the significant SEEG contacts that were overlapping between participants or between electrodes using a cluster procedure (see below and Meth- ods). As expected, during encoding, clear transient gamma oscillations were nested in the theta cycle (Fig 2A for illustration) in the auditory cortex (Heschl’s gyrus, 9 SEEG contacts, n = 5 participants, S1 Table). To investigate whether this mechanism played a functional role during retention, we contrasted the theta-gamma PAC strength values of memory trials (6 tones, 2 s retention) with the theta-gamma PAC strength values of perception trials (6 tones, 2 Fig 2. Theta-gamma PAC during encoding and retention. (A) Top: Time-frequency plot of mean gamma power modulation time- locked to a 4-Hz (theta) oscillation during encoding in the right and left median belt (n = 7). Bottom: Theta (4 Hz) cycles for a 1-s time window. Source data can be found at https://osf.io/m7dta/. (B) Memory vs. perception contrast during retention. Top: SEEG contacts (left hippocampus (2 SEEG contacts, n = 2) and right auditory areas (15 SEEG contacts, n = 1)) showing a significant increase of theta (4 Hz)–gamma (30–90 Hz) PAC strength for memory trials as compared to perception trials during the silent (retention) delay (6 tones, 2 s retention period). All results are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. (C) Bar plot shows theta-gamma PAC values averaged over trials and participants for memory and perception conditions for the significant SEEG contacts displayed in (B). Circles show individual trials. Source data can be found at https://osf.io/ m7dta/. (D). T-values for the co-modulogram (in SEEG contacts identified in B) for memory versus perception contrast (p < .05, FDR corrected). Source data can be found at https://osf.io/m7dta/. https://doi.org/10.1371/journal.pbio.3002512.g002 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 6 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain s retention) during the retention period (permutation testing, 10,000 permutations), for each participant and each of their SEEG contacts (Fig 2B). After computing this analysis for each participant, we used the same cluster-based approach as for the analysis of oscillatory power (see Methods). This analysis revealed a clear increase in theta-gamma PAC in the left hippo- campus (2 SEEG contacts, n = 2) and right auditory regions (15 SEEG contacts, n = 1) in the memory condition compared to the perception condition (Fig 2B and 2C, all ps < 0.001; see S3 Table for coordinates). However, one can question whether this coupling was specific to theta and gamma oscilla- tions as theta-beta, alpha-gamma, and alpha-beta PAC have previously been reported during working memory [42]. To test whether this effect was specific to the phase of the theta and the amplitude of the gamma oscillations, we computed the same analysis in the SEEG contacts showing significant PAC increase in the memory versus perception contrast (displayed Fig 2B; see S3 Table for details and coordinates), but using multiple low frequencies as frequency for phase (4 to 11 Hz, i.e., theta to alpha) and multiple high frequencies as frequency for amplitude (15 to 140 Hz, i.e., beta to high gamma; see Fig 2D). Interestingly, the memory versus percep- tion contrast performed on these co-modulograms (p < .05, FDR corrected) revealed that the maximum increase in PAC strength for memory trials as compared to perception trials was observed between theta (4 to 6 Hz) as frequency for phase and gamma as frequency for ampli- tude (35 to 105 Hz). Note that we performed the same analysis in all SEEG contacts located in regions showing increased theta and gamma power relative to baseline during retention (Fig 1F, middle panel, coordinates in the Supporting information). This analysis revealed no significant difference of PAC strength between memory and perception trials after FDR cor- rection (see S3 Fig for illustration of the difference of PAC strength values between memory and perception trials) Theta-gamma PAC in fronto-temporal areas and hippocampus decodes correct and incorrect memory trials and correlates with auditory STM performance We then investigated whether the strength of theta-gamma PAC during memory retention can decode correct and incorrect memory trials and predict STM performance. To do so, we used the SEEG data and the behavioral data of all memory conditions for each participant. We first used a support vector machine (SVM) classifier with 3-fold cross-validation to classify correct and incorrect trials in all memory conditions, using only PAC strength in each SEEG contact as input features (see Methods). This approach was implemented for each participant: The model is trained only on data from 2/3 of the trials to predict whether a trial is correct or incorrect in the remaining 1/3 of the trials. The procedure is repeated 3 times, and the sum- mary of the SVM’s performance (average of all models) reflects, for each participant, the degree to which correct and incorrect STM trials can be discriminated based on PAC strength. As all participants had more correct than incorrect trials for all memory conditions, we made a random selection of the correct trials (to match the number of incorrect trials for each condi- tion) to train and test the classifier. Then, we repeated this analysis 100 times with 100 different random selection of correct trials for each participant. SVM’s performance was evaluated using the output of the 100 models (accuracy minus chance) for each participant. The models significantly classified correct and incorrect memory trials above chance in 12/ 16 participants (all ps < .03 as measured by a Wilcoxon rank test; Fig 3A; ROC curves for each participant are presented in Fig 3B). We then aimed to define the SEEG features (i.e., SEEG contacts) the models relied upon to discriminate correct and incorrect STM trials. For each participant with significant above chance decoding accuracy, we extracted the feature weights PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 7 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Fig 3. PAC as markers of correct vs. incorrect memory retention identified with machine learning. (A) SVM decoding accuracy (accuracy minus chance—chance level: 0%) for a 2-class decoding analysis of PAC strength and SEEG contacts as features (correct vs. incorrect memory retention in all memory conditions). The colored bars represent accuracy minus chance for each participant (sorted as a function of accuracy with a jet colormap). Orange shaded rectangle overlaps with participants showing decoding accuracy significantly above chance. Blue shaded rectangle overlaps with participants with decoding accuracy not significantly different from chance. Asterisk: significant, ns: nonsignificant. Source data can be found at https://osf.io/m7dta/ (B) ROC for each participant (same color code as in A). Black dashed line represents the chance level. Source data can be found at https://osf.io/m7dta/. (C) Normalized feature weights showing features (SEEG contacts) with the largest influence (z-score) for each participant with significant decoding accuracy. Source data can be found at https://osf.io/m7dta/. PAC, phase amplitude coupling; ROC, receiver operating characteristic curve; SVM, support vector machineAU : AbbreviationlistshavebeencompiledforthoseusedinFigs3 (cid:0) 5:Pleaseverifythatallentriesarecorrectlyabbreviated: . https://doi.org/10.1371/journal.pbio.3002512.g003 to estimate their relative importance (z-scored, normalized across features for each partici- pant) in the classification. We then extracted the SEEG contact showing the maximum zscore value (i.e., contributing more to the classification) for each participant and represented it on a MRI volume (Fig 3C). This analysis revealed that the right and left hippocampus, the right IFG, the right and left primary auditory cortices, the left STS, and the left ITG (see S4 Table for details) were the brain regions where PAC strength allowed to classify correct and incorrect memory trials. It is relevant to note, however, that this analysis does not allow to infer whether PAC strength in the identified brain regions was associated to good or poor performance. Indeed, the features weights shown in Fig 3C can be used only to infer that PAC strength in these given SEEG contacts can decode correct and incorrect memory trials. We thus investigated whether theta-gamma PAC during memory retention can be corre- lated to STM performance. To do so, we used the SEEG data and the behavioral data of all memory conditions for each participant. This allowed us to benefit from the variability in PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 8 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Fig 4. Theta-gamma PAC in the hippocampus and ventral auditory stream correlates with behavior. (A) Left panel: SEEG contacts showing a positive correlational relationship between theta-gamma PAC and performance (negative correlation with IES). Results are displayed on the single subject T1 in the MNI space provided by SPM12. Right panel: Scatter plot of IES (note that the scale is inverted for clarity: 5 corresponding to poor performance and 0 corresponding to good performance) against theta-gamma PAC strength for each significant SEEG contact. Each color depicts a different participant (N = 6). Source data can be found at https://osf.io/m7dta/. (B) Left panel: SEEG contacts showing a negative correlational relationship between theta-gamma PAC and performance (positive correlation with IES). Results are displayed on the single subject T1 in the MNI space provided by SPM12. Right panel: Scatter plot of IES (note that the scale is inverted for clarity: 5 corresponding to poor performance and 0 corresponding to good performance) against theta-gamma PAC strength for each significant SEEG contact. Colors show the different participant (N = 4). Source data can be found at https://osf.io/m7dta/. IES, inverse efficiency score; PAC, phase amplitude coupling. https://doi.org/10.1371/journal.pbio.3002512.g004 behavioral performance associated with the manipulation of the memory load and of the dura- tion of the retention period. As a significant effect of condition emerged for the accuracy data (Fig 1B), but not for the RT data (Fig 1C), we computed for each trial the inverse efficiency score (IES; correct RT at the single trial scale/percent correct in the corresponding condition; see [43] and Methods). This behavioral metric increased the variability of behavioral scores between memory conditions with a low score representing a rapid RT and a high percentage of correctness. We then performed a Pearson’s correlation between IES and PAC strength val- ues for each SEEG contact and each participant (across all conditions). This analysis revealed, after cluster correction, that theta-gamma PAC values in the left hippocampus (4 SEEG con- tacts, n = 2), left superior temporal sulcus (STS; 2 SEEG contacts, n = 2), right inferior tempo- ral gyrus (ITG; 2 SEEG contacts, n = 2), and left inferior frontal gyrus/insula (IFG; 2 SEEG contacts, n = 2) had a positive correlational relationship with performance (i.e., negatively cor- related with the IES; Fig 4A and see S5 Table). Moreover, this analysis also revealed that theta- gamma PAC in the left Heschl’s gyrus (4 SEEG contacts, n = 4) had a negative relationship with performance (positively correlated with the IES; Fig 4B and S6 Table). Note that we per- formed the same analysis only with the conditions that were performed by all 16 participants (see Table 1) and obtained similar results (see S4 Fig). Coupling phase is consistent across participants and trials The analyses presented in Figs 2 to 4 evaluated PAC strength for each participant (coupling consistent across trials, within participant). However, these analyses do not guarantee that the coupling occurred at the same phase for all participants: Different participants could show a preferred coupling at different phases of the theta oscillations. To investigate this question, we PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 9 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain further evaluated whether gamma bursts were consistently restricted to specific phase ranges of the theta oscillations across participants in regions identified in Fig 4A (using data of all conditions available for the participants showing significant effects in Fig 4A). We first com- puted the theta-gamma phase consistency across trials, for the SEEG contacts where the PAC strength was correlated with behavioral performance (see Fig 4A and S5 Table). For each trial, and each SEEG contact, we extracted the magnitude of gamma oscillations (30 to 90 Hz) as a function of the phase of the theta oscillation (4 Hz) (average over the entire retention period, theta phase divided into 8 bins; see Methods). In both memory (correct trials) and perception trials separately, we computed the intertrial phase locking value (PLV) as a measure of inter- trial phase consistency of the coupling. Then, this metric was contrasted between memory and perception trials (Wilcoxon rank test) for each region (grouping SEEG contacts as a function of their location in the AAL atlas; Fig 5A). As expected, this analysis revealed greater consis- tency in theta-gamma PAC for memory as compared to perception trials for all regions (all p- values < .0001; Fig 5B). Finally, we aimed to identify whether a specific coupling phase range between the phase of the theta oscillations and the amplitude of gamma oscillations can be identified in these regions across trials and participants. To do so, we used linear mixed models (LMM) and mod- eled the variability between participants by defining by-participant random intercepts. This analysis was done for each region with theta phase bin as fixed factors and participants as a random factor (using data of all memory conditions available for the participants showing sig- nificant effects in Fig 4A). For all regions, we observed a main effect of theta phase (all χ2 (7) > 18.7; all ps < .01) on the gamma power. Post hoc Tukey analysis revealed increased gamma power between −π/2 and 0 of the theta cycle as compared to other bins in all regions (Fig 5C, see S7–S10 Tables for detailed statistics). Discussion Using intracranial electrophysiological recordings in humans, we showed that (i) the strength of theta-gamma PAC in temporal regions and hippocampus was increased during the short-term retention of auditory sequences as compared to simple perception; (ii) the strength of theta-gamma PAC in STS, ITG, IFG, and hippocampus decode correct and incorrect memory trials as evaluated with machine learning; (iii) the strength of theta- gamma PAC in these regions was positively correlated with individual STM performance; and, finally, that (iv) the coupling phase was highly consistent in these regions across indi- vidual participants to enable successful memory performance (high-frequency oscillations consistently restricted to specific phase ranges of the theta oscillations). The implications of these findings are discussed below. Increasing memory load and duration of the silent retention period decrease performance In line with previous studies, the present behavioral findings indicated that participants’ STM abilities (as also observed for other materials, such as verbal or visuo-spatial) decreased with increasing duration of the silent retention period [44] and increasing memory load ([45]; see Fig 1B). In the present study, we used these manipulations to increase the variability in task difficulty (and, consequently, modulate participants’ behavioral performance) across condi- tions. By combining information from accuracy and response times, we extracted a behavioral measure for each trial (IES; see methods and [43]) that we used to investigate the link between PAC strength values and behavior for each participant. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 10 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Fig 5. Theta gamma PAC is consistent across trials and participants. (A) SEEG contacts identified in Fig 4A and grouped as a function of their location according to the AAL Atlas: green, left STS; red, left hippocampus; blue, right ITG; yellow, left IFG/insula. Regions are displayed on the single subject T1 in the MNI space provided by SPM12. Source data can be found at https://osf.io/m7dta/. (B) PAC intertrial phase consistency computed for each region. Bar plot shows intertrial phase locking values across participants and SEEG contacts for memory trials (correct responses, colored as a function of the regions) and perception trials in the same region. Error bars indicate SEM. Asterisk indicates significance. Source data can be found at https://osf.io/m7dta/. (C) Preferred coupling phase: gamma power presented as a function of theta phase bins for each region. Shading represents the standard deviation across trials and participants. Asterisks (*** p < .001; * p < .05) and grey shading indicate significance. Note that for clarity, we show only the results for the post hoc tests performed for the peak of gamma power for each region. Detailed post hoc statistics are reported in S7–S10 Tables. Source data can be found at https://osf.io/m7dta/. IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; PAC, phase amplitude coupling; STS, superior temporal sulcus. https://doi.org/10.1371/journal.pbio.3002512.g005 Brain networks of auditory perception and short-term memory Time-frequency analyses revealed that transient gamma activity was evoked by each tone of the sequence in the auditory cortex, secondary auditory regions, hippocampus, and several PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 11 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain areas of the ventral pathway during the encoding and retrieval periods of the STM task and the equivalent periods of the perception task (see Fig 1C and 1D). It is well established that gamma oscillations are marking bottom-up and local (intraregional) processes during both passive and active sensory integration [46,47]. Observing such transient bursts after each tone of the to-be-encoded sequence can thus be considered as a marker of the integration of tones’ fea- tures by the sensory system (bottom-up). In addition, sustained theta oscillations were observed in distributed regions of the ventral pathway, including STS, STG, IFG, and hippocampus (see Supporting information) during encoding, retention, and retrieval. Theta oscillations (4 to 8 Hz) are typically considered as markers of attention, arousal, or memory during demanding cognitive tasks [48–50]. Notably, theta oscillations are known to play a key role in ordering items that are presented sequentially in STM or WM [51]. Moreover, theta oscillations have been associated to long-range commu- nication between distant brain regions during memory maintenance [49,50,52–54]. In the present study, an increase relative to baseline in theta power was observed in the hippocampus, inferior frontal regions, and secondary auditory regions, a brain network that has been consis- tently reported as being recruited during auditory STM tasks [15,55–57] (Fig 1F). However, during all phases of the task (referred to as encoding, retention, and retrieval periods for the memory task and their equivalent for the perception task), we did not observe any significant differences of gamma and theta magnitude between memory and perception trials. This result contrasts with the studies reported above [49,50,52–54]. A possible interpre- tation would be that the participants have been carrying out a form of WM during the percep- tion task (always performed after the memory condition; see Methods) even if they were not instructed to do so. An alternative interpretation would be that the fluctuations in oscillatory magnitude in the theta and gamma frequency ranges extracted in the present study were not specific to memory and might rather be associated with the perception of the sequence and attention towards the auditory input (note that even in the perception task, participants had to pay attention to the sound sequences to push a button at the end of S2).We thus aimed to define whether more fine-grained oscillatory markers related to memory retention can be identified with the investigation of theta-gamma PAC. Theta-gamma PAC in auditory and hippocampal regions is associated to auditory short-term memory retention During encoding, we observed that gamma oscillations were nested in the theta cycle in the auditory cortex (see Fig 2A for illustration). This effect was expected as each tone of the sequence induced a time-locked (or evoked) increase in gamma power, and the phase of the theta oscillation was entrained by the tone presentation rate (4 Hz; see [49,54] for basic princi- ples of sensory entrainment). We then investigated whether this statistical dependency between the phase of theta oscillations and the amplitude of gamma oscillations was still pres- ent during the retention period, a time window for which no stimuli were presented. More specifically, we investigated whether PAC signals were increased during memory retention as compared to perception. In the left hippocampus and right temporal regions, the strength of theta-gamma PAC was indeed significantly higher during the retention delay in the memory condition compared to the perception condition (see Fig 2B and S3 Table). It is relevant to note that this effect was observed in a limited number of SEEG contacts and participants. This is related to the cluster correction procedure we have used that keep only SEEG contacts that overlap between participants or contacts. One possible interpretation is that PAC during memory retention could result from sustained PAC signals that originally emerged during encoding (see Fig 2A; PAC coming from bottom-up entrainment at 4 Hz). It can thus be PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 12 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain argued that the significant effect observed between memory retention and perception could result from attentional differences for memory and perception trials during encoding (partici- pants paying more attention during memory than perception trials). However, one can argue that attentional effects could not only be observed in PAC measures but could also affect theta and gamma magnitude [58]. As the contrast between memory trials and perception trials for theta and gamma magnitude was not significant in the present study, we propose that these PAC effects were specific to memory. These results thus suggest a role of the hippocampus in auditory STM. This is in line with several neuroimaging studies in the visual modality [16,18,19,38] and also with recent single- unit recording studies in humans reporting increased neural firing in the hippocampus during the maintenance of visual representations [22,23,59]. For auditory STM, hippocampal involve- ment has, however, been less frequently described in previous research. Using an auditory STM task during fMRI recordings, Kumar and colleagues [15] have shown sustained activity in both ventral and dorsal parts of the hippocampus during an auditory STM task. Here, we observed activity mainly in its ventral part (y = −4), a finding fitting well with the fact that the anterior portion of the hippocampus is anatomically and functionally connected to auditory areas [60,61]. Interestingly, Kumar and colleagues [15] also reported that the pattern of fMRI activity in hippocampal areas allows the decoding of the different sounds maintained in mem- ory. Our present study goes beyond these findings by identifying the neurophysiological mech- anism by which the hippocampus supports retention of auditory information in memory. Indeed, here we showed that theta-gamma PAC in the hippocampus and temporal regions (STS, ITG) decodes correct and incorrect memory trials (Fig 3A and S4 Table) and was posi- tively correlated with behavioral performance (negative correlation with IES; Fig 4A and 4B and S5 Table). This finding is well aligned with previous research showing that hippocampal theta-gamma PAC plays a functional role during memory retention for visual material [18,20,37,38]. In the present study, we show that the temporal and hippocampal regions imple- ment the same electrophysiological mechanism to allow for the maintenance of sequential auditory information, a finding that has, to our knowledge, never been reported before. This finding is also well aligned with a recent study showing cortico-hippocampal interplay in the theta range during both encoding and retention of a STM task with visually presented words [62]. Taken together, our results suggest a clear role of theta-gamma PAC in the temporal and hippocampal regions during auditory STM in the human brain. In addition to auditory and hippocampal regions, we observed that theta-gamma PAC strength in the left IFG decodes correct and incorrect memory trials (Fig 3A and S4 Table) and was positively correlated with behavioral performance (negative correlation with IES; Fig 4A and S5 Table). This is in line with the well-established role of the IFG in STM maintenance in humans [15,50,55–57,63–69].Interestingly, we also observed that theta-gamma PAC in Heschl’s gyrus during memory retention was negatively correlated with behavioral performance (positive corre- lational relationship with IES; Fig 4B). This result suggests that to perform successfully the STM task, PAC signals need to reach higher-level regions, namely, STS, ITG, hippocampus, and infe- rior frontal regions, to allow for efficient maintenance of the information. This hypothesis receives support in a recent study showing that theta and gamma activity in the human hippo- campus is associated with successful recall when extrahippocampal activation patterns shifted from perceptual toward mnemonic representations. This study also suggests that recurrent hip- pocampal–cortical interactions are then implemented to support memory processing [70]. From a more global perspective, our results are in agreement with the theta-gamma neural code hypothesis developed by Lisman and Jensen [31], proposing that cross-frequency signal- ing in cortico-hippocampal networks is a sophisticated mechanism implanted by the brain to hold sequentially organized information in memory [20,25,31]. This hypothesis assumes that PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 13 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain representations of individual encoded items (via high-frequency oscillations) do not occur during the entire cycle of low-frequency oscillations. Instead, these high-frequency oscillations are thought to be restricted to specific phase ranges of the slow oscillation that correspond to higher levels of neural excitability [20,31,71]. To test the validity of this model, we investigated for each region whether the gamma bursts in the present data were consistently restricted to a specific phase range of the theta oscillations across trials and participants. Consistent phase coupling across participants during successful memory performance We extracted the PAC consistency across trials and participants in the brain regions where PAC strength was positively predicting behavioural performance (see Fig 5A and S5 Table). Inter- trial-phase locking analysis on these signals revealed greater consistency in theta-gamma PAC for memory trials than for perception trials in all regions (Fig 5B). We then aimed to identify whether a preferred coupling phase range could be identified. We observed that, for correct memory trials, the gamma bursts were occurring consistently at a specific phase range of the theta cycle in the left STS, right ITG, left IFG, and the left hippocampus (see Fig 5C and S7–S10 Tables). This preferred phase is of interest because it suggests that similar mechanisms are implemented in this network across trials and participants. Interestingly, the gamma burst occurred from the trough of the theta cycle to its peak. As shown in earlier research, the phase of theta oscillation reflects rhythmic fluctuations of neural excitability [72]. Such cycles, occur- ring several times per second, represent fluctuations between (high-excitability) phases during which relevant information is amplified and (low-excitability) phases during which information is suppressed. Here, we observed high coupling consistency between −π/2 and 0 of the theta cycle, a phase range corresponding to a high-excitability period of the oscillation where infor- mation processing can be amplified [25,31,72]. Observing this effect only for correct memory trials is another important cue suggesting that fronto-auditory-hippocampal theta-gamma PAC allows successful integration and the retention of sequential auditory information in STM. Overall, our study provides new information about the neurophysiological mechanisms by which the fronto-temporal-hippocampal network encodes and maintains sequential auditory information. The findings provide crucial insights into the networks and brain dynamics involved in this fundamental process in the auditory modality. Methods Participants Intracranial recordings were obtained from 16 neurosurgical patients with drug-resistant focal epilepsy (8 females and 8 males, mean age: 32.6 +/− 8.73 years) at the Epilepsy Department of the Grenoble Neurological Hospital (Grenoble, France) and the Epilepsy Department of Lyon Neurological Hospital (Lyon, France). All patients were stereotactically implanted with multi- lead EEG depth electrodes. Data from all electrodes exhibiting pathological waveforms were discarded from the present study. All participants provided written informed consent, and the experimental procedures were approved by the appropriate regional ethics committee (CPP Sud-Est V, 2009-A00239-48). The study has been conducted according to the principles expressed in the Declaration of Helsinki. Task and conditions The participants were asked to perform an auditory STM task, consisting in the comparison of tone sequences presented in pairs and separated by a silent retention period. Participants also PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 14 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain performed a block of passive listening of these trials in which they were required to ignore the content of tone sequences and press a button as fast as possible at the end of S2. To manipulate task difficulty (only for the memory task), in different blocks, we varied the memory load (3 or 6 to-be-encoded items) as well as the duration of the silent retention period between the to-be- compared sequences (2 s, 4 s, and 8 s; see Table 1 for a detailed description of the conditions). All tone sequences were composed of 250-ms-long piano tones presented sequentially without interstimulus interval. The 2 sequences could be either the same or different (50% of each trial type). For “different” trials, the second sequence differed by a single tone altering the melodic contour (Fig 1A). For the 6-tone melodies, 120 different tone sequences were created using 8 piano tones differing in pitch height (Cubase software, melodies from [55]); all used tones belonged to the key of C Major (C3, D3, E3, F3, G3, A3, B3, C4). For the 3-tone sequences, 60 different tone sequences were created using the same pool of piano tones (material from [55,56]). Procedure Presentation software (Neurobehavioral Systems, Albany, CA, USA) was used for the delivery of the experimental protocol to present the auditory stimuli and to register button presses. For each trial, participants listened binaurally (presented with headphone at a comfortable listen- ing level) to the first 3- or 6-tone sequence with a total respective duration of 750 or 1,500 ms (encoding, S1), followed by a silent retention period (2, 4, or 8 s), and then the second sequence (retrieval, S2, 750 or 1,500 ms duration). Conditions were counterbalanced across participants. Participants were informed of the block order and were asked to indicate their answers by pressing one of 2 keys with their right hand after the end of S2. Their responses were recorded during the first 2 s of the intertrial interval, whose random duration was com- prised between 2.5 and 3 s. No feedback was given during the experiment. Each block of the task included 30 trials (15 “same” trials and 15 “different” trials for each condition). Within each block, the trials were presented in a pseudorandomized order; the same trial type (i.e., “same” or “different”) could not be repeated more than 3 times in a row. Before the first ses- sion, participants performed a set of 10 practice trials (with melodies not used in the main experiment). Analysis of behavioral data Task performance was measured with d prime (Signal Detection Theory). RTs were measured from the end of S2. Behavioral data were analyzed with nonparametric repeated measures ANOVA (Friedman) and Wilcoxon rank test (see Results). The IES was calculated for each trial. IES is computed by normalizing, at the single trial scale, the participant RT by their respective percentage of correct responses in each condition. As compared to RTs, this beha- vioural metric increases the variability of behavioural scores with a low score representing a short RT and a high percentage of correctness [43]. Correlation analysis between performance at the single trial level and brain data (PAC values; see below) were performed using IES. Localization of depth electrodes In each patient’s brain, 10 to 16 semirigid, multilead electrodes were stereotactically implanted. The SEEG electrodes had a diameter of 0.8 mm and, depending on the target structure, consist of 10 to 15 contact leads 2.0 mm wide and 1.5 mm apart (DIXI Medical Instruments). All par- ticipants underwent two 3D anatomical MPRAGE T1-weighted MRI scan on a 1.5T Siemens Sonata scanner or on a 3T Siemens Trio (Siemens AG, Erlangen, Germany) before implanta- tion and just after the SEEG implantation. The anatomical volume consisted of 160 sagittal PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 15 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain slices with 1 mm3 voxel, covering the whole brain. The scalp and cortical surfaces were extracted from the T1-weighted anatomical MRI. All electrode contacts were identified on the post-implantation MRI showing the electrodes and coregistered on a pre-implantation MRI (ImaGIN toolbox; https://f-tract.eu/software/imagin/). MNI coordinates were computed using the SPM (http://www.fil.ion.ucl.ac.uk/spm/) toolbox. In addition to MNI coordinates, we computed the localization of the SEEG contacts in the AAL3 atlas [73]. Intracranial recordings Intracranial recordings were conducted using a video-SEEG monitoring system (Micromed), which allowed the simultaneous data recording from 128 depth EEG electrode sites (identical acquisition system and acquisition parameters in the 2 recording sites). The data were band- pass filtered online from 0.1 to 200 Hz and sampled at 512 Hz for all patients. At the time of acquisition, data were recorded using a reference electrode located in white matter, and each electrode trace was subsequently re-referenced to its immediate neighbour (bipolar deriva- tions). This bipolar montage has several advantages over common referencing. It helps elimi- nating signal artifacts common to adjacent electrode contacts (such as the 50 Hz mains artifact or distant physiological artifacts) and achieves a high local specificity by cancelling out effects of distant sources that spread equally to both adjacent sites through volume conduction. The spatial resolution achieved by the bipolar SEEG is estimated to be on the order of 3 mm [74]. Preprocessing SEEG data were preprocessed and visually checked to reject contacts contaminated by patho- logical epileptic activity or environmental artifacts. Powerline contamination of the raw data (main 50 Hz, harmonics 100 and 150 Hz) was reduced using notch filtering. Then, data were epoched to create trials with a window of 1,000 ms before the onset of S1 and 500 ms after the end of the last stimulus of the S2 sequence. SEEG contacts showing signal values exceeding 1,500 μV during the trial time window were excluded from the analysis: As a result, between 17 and 30 trials were kept for each participant and condition. Time-frequency analysis in Heschl’s gyrus and hippocampus We first performed time-frequency Morlet analysis for the SEEG contacts located in the right and left Heschl’s gyrus and bilateral hippocampus (according to the AAL atlas). This analysis was done to define the frequency bands of interest for the whole brain Hilbert’s analysis and to define the frequency for phase and frequency for amplitude for the PAC analysis. Time-fre- quency Morlet analysis was computed based on a wavelet transform of the signals [75]. The wavelet family was defined by (f0 /sf) = 7 with f0 ranging from 1 to 150 Hz in 1 Hz steps. The time-frequency wavelet transform was applied to each SEEG contact, each trial, and then power was averaged across trials, resulting in an estimate of oscillatory power at each time sample and each frequency bin between 1 and 150 Hz. Note that both evoked and induced activity were estimated. We then performed a normalization (z-scoring) with −1,000 to 0 ms preceding the presentation of the S1 sequence as baseline. Time-frequency plots of SEEG con- tacts were regrouped in left and right Heschl’s gyrus and bilateral hippocampus across partici- pants using the AAL3 brain atlas. By doing so, we were able to investigate the data of several participants on one time-frequency map per area. Normalized and averaged time-frequency maps of the auditory cortex and hippocampus were used to define the frequency for phase and frequency for amplitude for the PAC analysis (see below). Frequency for amplitude was defined from 30 Hz to 90 Hz as it matched with the amplitude of time-frequency maps gamma bursts in the auditory cortex (see also [18] for similar parameters). Frequency for phase was PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 16 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain defined at 4 Hz because sustained theta power at 4 Hz was observed in the auditory cortex dur- ing encoding (Fig 1D), and this frequency matched the frequency of presentation of the stimuli. Hilbert transform Once the frequency bands of interest were defined, we aimed to investigate if fluctuation of theta and gamma power were associated to memory processes (as compared to perception). In order to perform this analysis at the whole brain level and to reduce the dimension of the data, we computed for each participant the Hilbert transform for correct trials for each period of the STM task (encoding, retention, and retrieval, average in time for each time period; see Table 1) and the corresponding periods of the perception task. We extracted the magnitude of theta activity at 4 Hz and gamma activity between 30 to 90 Hz for each trial for each SEEG contact. These data were then used to contrast brain activity in the memory conditions and baseline and to contrast brain activity in the memory and perception conditions using permutation tests as implemented in MATLAB. Contrasts with baseline were corrected for multiple com- parison using FDR corrections. Memory versus perception contrast were corrected with a cluster procedure (see below). Phase amplitude coupling Theta-gamma PAC was computed using the method developed by [76]. Frequency for phase and frequencies for amplitudes were defined by a power spectrum density analysis on SEEG contacts located in the auditory cortex and in the hippocampus and computed over the total duration of a trial time window (0 to 5.5 s for the 6 tones, 2 s memory condition as this condi- tion was performed by all 16 participants). Frequency for phase was selected as the frequency showing the highest peak in the theta band (4 to 8 Hz) in the auditory cortex and hippocampus (see Fig 1D and 1E) and frequency for amplitude was defined as a 60-Hz-width frequency band centered on the highest peak in the gamma band (peak at 60 Hz ± 30 Hz resulting in a band between 30 and 90 Hz) in the auditory cortex. Based on these results (see Fig 1D and 1E), we used 4 Hz as the frequency for phase (frequency of presentation of stimuli) and 30 to 90 Hz as the frequency for amplitude for the PAC analyses. As no high gamma peak emerged in this PSD analysis, we did not investigate PAC for frequencies above 90 Hz. 3D representation and cluster procedure For all PAC analyses and Hilbert data, significant SEEG contacts were plotted on a MNI MRI volume using marsbar and SPM functions [77]. To do so, we extracted the MNI coordinate of each SEEG contact and represent the oscillatory magnitude and PAC values on spheres of 4 mm radius in the MRI volume. PAC plots were corrected with a cluster approach: by consider- ing as significant only the contacts that were overlapping across at least 2 participants or 2 SEEG contacts in the MRI volume. Multivariate analyses Multivariate analyses were performed using MATLAB and SVM implementation (https:// www.mathworks.com/help/stats/fitcecoc.html). A linear classifier was chosen as SEEG data contains many more features than examples, and classification of such data is generally suscep- tible to overfitting. One way of alleviating the danger of overfitting is to choose a simple func- tion (such as a linear function) for classification, where each feature affects the prediction solely via its weight and without interaction with other features (rather than more complex PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 17 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain classifiers, such as nonlinear SVMs or artificial neural networks, which can let interactions between features and nonlinear functions thereof drive the prediction). Our strategy was to use the SVM classifier with 3-fold cross-validation to classify correct and incorrect memory trials of all memory conditions, using the SEEG contact as features. For each participant, the 1/ model is trained only on data 2/3 of the trials to predict whether each trial in the remainingAU : Pleasecheckandconfirmthattheeditto}Foreachparticipant; themodelistrainedonlyondata:::}didnotaltertheintendedmeaningofthesentence: 3 set of trials is correct or incorrect. The procedure is repeated 3 further times to estimate the classification performance across the full set folds. As all participants had more correct than incorrect trials for all memory conditions, we made a random selection of the correct trials (to match the number of incorrect trials for each condition) to train and test the classifier. Then, we repeated this analysis 100 times with 100 different random selection of correct trials for each participant. SVM’s performance was evaluated using the output of the 100 models (accu- racy minus chance) for each subject. For each subject with above chance decoding accuracy, we extracted the features weights (zscore) to evaluate the relative contribution of each feature (SEEG contact) in the classification. Phase consistency analysis We extracted the PAC consistency across trials and participants in the brain regions where the PAC strength was correlated with behavioural performance (see Figs 4A and 5A and S5 Table). For each trial, we extracted the magnitude of gamma oscillations (30 to 90 Hz) as a function of the phase of the theta oscillation (4 Hz; phase divided into 8 bins). We then extracted the intertrial phase locking (PLV) on these signals using PLV functions available in Brainstorm. To identify whether significant preferred coupling phase could be identified, we extracted for each SEEG contact the gamma power for 8 different phase bins of the theta cycle. To define if a preferred coupling phase can be identified across trials and participant for each region, we used LMMs. The variability between participants was modeled by defining by-par- ticipant random intercepts. LMMs were performed in R 3.4.1 using the lme4 [78] and car [79] packages. Both fixed and random factors were considered in statistical modeling. Wald chi- squared tests were used for fixed effects in LMM [79]. The fixed effect represents the mean effect across all participants after accounting for variability. We considered the results of the main analyses significant at p < .05. When we found a significant main effect, post hoc honest significant difference (HSD) tests were systematically performed using the R emmeans pack- age (emmeans version 1.6.3). P values were considered as significant at p < .05 and were adjusted for the number of comparisons performed. More precisely, to avoid increased Type I error when multiple comparisons were performed, the p-value of the Tukey HSD test was adjusted using the Tukey method for comparing the given number of estimates. Supporting information S1 Fig. Brain oscillations displayed with a logarithmic scale for the frequency axis. (A) T- values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left Heschl’s gyrus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 5). (B) T-values in the time-frequency domain (t test relative to baseline −1,000 to 0 before stimulus onset, FDR corrected in time and frequency domains) of SEEG contacts located in the right and left hippocampus (displayed on the single subject T1 in the MNI space provided by SPM12) for a trial time window (−1,000 to 6,000 ms) for the condition 6-tone memory load, 2 s retention period (n = 14). (PDF) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 18 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain S2 Fig. Theta (orange) and gamma (red) magnitude averaged over SEEG contacts located in regions showing increased power relative to baseline during retention presented as a function of task (memory, perception). NS, nonsignificant. (PDF) S3 Fig. Memory minus perception (the colormap represents the difference in PAC strength between memory and perception trial—note that the contrast is not significant) for the co- modulogram in SEEG contacts that had previously shown an increase in theta and gamma power identified in Fig 1F, retention period). (PDF) S4 Fig. Theta-gamma PAC in the hippocampus and ventral auditory stream correlates with behavior. Left panel: SEEG contacts showing a positive (hot colormap) and negative (blue colormap) relationship between theta-gamma PAC and performance using data from conditions performed by all 16 participants (6 tones encoding 2 s retention and 6 tones encod- ing 8 s retention). Results are displayed on the single subject T1 in the MNI space provided by SPM12. (PDF) S1 Table. Regions and coordinates Fig 1D: Heschl’s gyrus. (PDF) S2 Table. Regions and coordinates Fig 1E: Hippocampal regions. (PDF) S3 Table. Regions and coordinates Fig 2B: PAC memory vs. perception L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior. (PDF) S4 Table. Coordinates of the maximum value (zscore) of the features weights for each par- ticipant with significant above chance decoding accuracy—Fig 3C, L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior; Tri, Triangular. (PDF) S5 Table. Regions and coordinates Fig 4A: Correlation between PAC and IES, L, Left; R, Right; Sup, Superior; Mid, Middle; Inf, Inferior; Oper, Opercular. (PDF) S6 Table. Regions and coordinates Fig 4B: Correlation between PAC and IES. (PDF) S7 Table. Post hoc tests of Fig 5C: Left STS. (PDF) S8 Table. Post hoc tests of Fig 5C: Left IFG. (PDF) S9 Table. Post hoc tests of Fig 5C: Left hippocampus. (PDF) S10 Table. Post hoc tests of Fig 5C: Right ITG. (PDF) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 19 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain Acknowledgments We thank Luc H. Arnal for his insightful comments on a previous version of this manuscript. Author Contributions Conceptualization: Anne Caclin, Jean-Philippe Lachaux, Barbara Tillmann, Philippe Albouy. Data curation: Arthur Borderie, Anne Caclin, Marcela Perrone-Bertollotti, Barbara Tillmann, Philippe Albouy. Formal analysis: Arthur Borderie, Roxane S. Hoyer, Philippe Albouy. Funding acquisition: Barbara Tillmann, Philippe Albouy. Investigation: Arthur Borderie, Marcela Perrone-Bertollotti, Philippe Albouy. Methodology: Arthur Borderie, Jean-Philippe Lachaux, Philippe Kahane, He´lène Catenoix, Philippe Albouy. Project administration: Anne Caclin, Jean-Philippe Lachaux, Philippe Kahane, He´lène Cate- noix, Barbara Tillmann, Philippe Albouy. Resources: Anne Caclin, Jean-Philippe Lachaux, Philippe Kahane, He´lène Catenoix, Barbara Tillmann, Philippe Albouy. Software: Philippe Albouy. Supervision: Anne Caclin, Barbara Tillmann, Philippe Albouy. Validation: Philippe Albouy. Visualization: Arthur Borderie, Philippe Albouy. Writing – original draft: Arthur Borderie, Philippe Albouy. Writing – review & editing: Arthur Borderie, Anne Caclin, Jean-Philippe Lachaux, Roxane S. Hoyer, Philippe Kahane, He´lène Catenoix, Barbara Tillmann, Philippe Albouy. References 1. Scoville WB, Milner B. Loss of recent memory after bilateral hippocampal lesions. J Neurol Neurosurg Psychiatry. 1957; 20(1):11–21. https://doi.org/10.1136/jnnp.20.1.11 PMID: 13406589 2. Spiers HJ, Maguire EA, Burgess N. Hippocampal amnesia. Neurocase. 2001; 7(5):357–382. https://doi. org/10.1076/neur.7.5.357.16245 PMID: 11744778 3. Baddeley AD, Warrington EK. Amnesia and the distinction between long-and short-term memory. J Verb Learning Verb Behav. 1970; 9(2):176–189. 4. Cave CB, Squire LR. Intact verbal and nonverbal short-term memory following damage to the human hippocampus. Hippocampus. 1992; 2(2):151–163. https://doi.org/10.1002/hipo.450020207 PMID: 1308180 5. Wickelgren WA. Sparing of short-term memory in an amnesic patient: Implications for strength theory of memory. Neuropsychologia. 1968; 6(3):235–244. 6. Baddeley A, Jarrold C, Vargha-Khadem F. Working memory and the hippocampus. J Cogn Neurosci. 2011; 23(12):3855–3861. https://doi.org/10.1162/jocn_a_00066 PMID: 21671734 7. Jeneson A, Squire LR. Working memory, long-term memory, and medial temporal lobe function. Learn Mem. 2012; 19(1):15–25. https://doi.org/10.1101/lm.024018.111 PMID: 22180053 8. Atkinson RC, Shiffrin RM. Human memory: a proposed system and its control processes. In: Spence KW, editor. The Psychology of Learning and Motivation: Advances in Research and Theory. 2. New York: Academic Press; 1968. p. 89–195. 9. James W. The Principles of Psychology. Holt H, editor. New York; 1890. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 20 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain 10. Michels L, Bucher K, Luchinger R, Klaver P, Martin E, Jeanmonod D, et al. Simultaneous EEG-fMRI during a working memory task: modulations in low and high frequency bands. PLoS ONE. 2010; 5(4): e10298. https://doi.org/10.1371/journal.pone.0010298 PMID: 20421978 11. Zarahn E, Rakitin B, Abela D, Flynn J, Stern Y. Positive evidence against human hippocampal involve- ment in working memory maintenance of familiar stimuli. Cereb Cortex. 2005; 15(3):303–316. https:// doi.org/10.1093/cercor/bhh132 PMID: 15342440 12. Buffalo EA, Reber PJ, Squire LR. The human perirhinal cortex and recognition memory. Hippocampus. 1998; 8(4):330–339. https://doi.org/10.1002/(SICI)1098-1063(1998)8:4<330::AID-HIPO3>3.0.CO;2-L PMID: 9744420 13. Holdstock JS, Mayes AR, Gong QY, Roberts N, Kapur N. Item recognition is less impaired than recall and associative recognition in a patient with selective hippocampal damage. Hippocampus. 2005; 15 (2):203–215. https://doi.org/10.1002/hipo.20046 PMID: 15390152 14. Olson IR, Moore KS, Stark M, Chatterjee A. Visual working memory is impaired when the medial tempo- ral lobe is damaged. J Cogn Neurosci. 2006; 18(7):1087–1097. https://doi.org/10.1162/jocn.2006.18.7. 1087 PMID: 16839283 15. Kumar S, Joseph S, Gander PE, Barascud N, Halpern AR, Griffiths TD. A Brain System for Auditory Working Memory. J Neurosci. 2016; 36(16):4492–4505. https://doi.org/10.1523/JNEUROSCI.4341-14. 2016 PMID: 27098693 16. Ranganath C D’Esposito M. Medial temporal lobe activity associated with active maintenance of novel information. Neuron. 2001; 31(5):865–873. 17. Nichols EA, Kao YC, Verfaellie M, Gabrieli JD. Working memory and long-term memory for faces: Evi- dence from fMRI and global amnesia for involvement of the medial temporal lobes. Hippocampus. 2006; 16(7):604–616. https://doi.org/10.1002/hipo.20190 PMID: 16770797 18. Axmacher N, Henseler MM, Jensen O, Weinreich I, Elger CE, Fell J. Cross-frequency coupling supports multi-item working memory in the human hippocampus. Proc Natl Acad Sci U S A. 2010; 107(7):3228– 3233. https://doi.org/10.1073/pnas.0911531107 PMID: 20133762 19. Axmacher N, Mormann F, Fernandez G, Cohen MX, Elger CE, Fell J. Sustained neural activity patterns during working memory in the human medial temporal lobe. J Neurosci. 2007; 27(29):7807–7816. https://doi.org/10.1523/JNEUROSCI.0962-07.2007 PMID: 17634374 20. Bahramisharif A, Jensen O, Jacobs J, Lisman J. Serial representation of items during working memory maintenance at letter-selective cortical sites. PLoS Biol. 2018; 16(8):e2003805. https://doi.org/10.1371/ journal.pbio.2003805 PMID: 30110320 21. van Vugt MK, Schulze-Bonhage A, Litt B, Brandt A, Kahana MJ. Hippocampal gamma oscillations increase with memory load. J Neurosci. 2010; 30(7):2694–2699. https://doi.org/10.1523/JNEUROSCI. 0567-09.2010 PMID: 20164353 22. Boran E, Fedele T, Klaver P, Hilfiker P, Stieglitz L, Grunwald T, et al. Persistent hippocampal neural fir- ing and hippocampal-cortical coupling predict verbal working memory load. Sci Adv. 2019; 5(3): eaav3687. https://doi.org/10.1126/sciadv.aav3687 PMID: 30944858 23. Kornblith S, Quian Quiroga R, Koch C, Fried I, Mormann F. Persistent Single-Neuron Activity during Working Memory in the Human Medial Temporal Lobe. Curr Biol. 2017; 27(7):1026–1032. https://doi. org/10.1016/j.cub.2017.02.013 PMID: 28318972 24. Billig AJ, Lad M, Sedley W, Griffiths TD. The hearing hippocampus. Prog Neurobiol. 2022; 218:102326. https://doi.org/10.1016/j.pneurobio.2022.102326 PMID: 35870677 25. 26. 27. Lisman J, Buzsaki G, Eichenbaum H, Nadel L, Ranganath C, Redish AD. Viewpoints: how the hippo- campus contributes to memory, navigation and cognition. Nat Neurosci. 2017; 20(11):1434–1447. https://doi.org/10.1038/nn.4661 PMID: 29073641 Fell J, Klaver P, Lehnertz K, Grunwald T, Schaller C, Elger CE, et al. Human memory formation is accompanied by rhinal-hippocampal coupling and decoupling. Nat Neurosci. 2001; 4(12):1259–1264. https://doi.org/10.1038/nn759 PMID: 11694886 Fell J, Ludowig E, Staresina BP, Wagner T, Kranz T, Elger CE, et al. Medial temporal theta/alpha power enhancement precedes successful memory encoding: evidence based on intracranial EEG. J Neurosci. 2011; 31(14):5392–5397. https://doi.org/10.1523/JNEUROSCI.3668-10.2011 PMID: 21471374 28. Colgin LL, Moser EI. Gamma oscillations in the hippocampus. Physiology (Bethesda). 2010; 25 (5):319–329. https://doi.org/10.1152/physiol.00021.2010 PMID: 20940437 29. Canolty RT, Edwards E, Dalal SS, Soltani M, Nagarajan SS, Kirsch HE, et al. High gamma power is phase-locked to theta oscillations in human neocortex. Science. 2006; 313(5793):1626–1628. https:// doi.org/10.1126/science.1128115 PMID: 16973878 30. Canolty RT, Knight RT. The functional role of cross-frequency coupling. Trends Cogn Sci. 2010; 14 (11):506–515. https://doi.org/10.1016/j.tics.2010.09.001 PMID: 20932795 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 21 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain 31. 32. 33. Lisman JE, Jensen O. The theta-gamma neural code. Neuron. 2013; 77(6):1002–1016. Lisman JE, Idiart MA. Storage of 7 +/- 2 short-term memories in oscillatory subcycles. Science. 1995; 267(5203):1512–1515. https://doi.org/10.1126/science.7878473 PMID: 7878473 Fuentemilla L, Penny WD, Cashdollar N, Bunzeck N, Duzel E. Theta-coupled periodic replay in working memory. Curr Biol. 2010; 20(7):606–612. https://doi.org/10.1016/j.cub.2010.01.057 PMID: 20303266 34. Bergmann TO, Born J. Phase-Amplitude Coupling: A General Mechanism for Memory Processing and Synaptic Plasticity? Neuron. 2018; 97(1):10–13. https://doi.org/10.1016/j.neuron.2017.12.023 PMID: 29301097 35. Helfrich RF, Mander BA, Jagust WJ, Knight RT, Walker MP. Old Brains Come Uncoupled in Sleep: Slow Wave-Spindle Synchrony, Brain Atrophy, and Forgetting. Neuron. 2018; 97(1):221–30 e4. https:// doi.org/10.1016/j.neuron.2017.11.020 PMID: 29249289 36. Mormann F, Fell J, Axmacher N, Weber B, Lehnertz K, Elger CE, et al. Phase/amplitude reset and theta-gamma interaction in the human medial temporal lobe during a continuous word recognition mem- ory task. Hippocampus. 2005; 15(7):890–900. https://doi.org/10.1002/hipo.20117 PMID: 16114010 37. Chaieb L, Leszczynski M, Axmacher N, Hohne M, Elger CE, Fell J. Theta-gamma phase-phase cou- pling during working memory maintenance in the human hippocampus. Cogn Neurosci. 2015; 6 (4):149–157. https://doi.org/10.1080/17588928.2015.1058254 PMID: 26101947 38. 39. Leszczynski M, Fell J, Axmacher N. Rhythmic Working Memory Activation in the Human Hippocampus. Cell Rep. 2015; 13(6):1272–1282. https://doi.org/10.1016/j.celrep.2015.09.081 PMID: 26527004 Fontolan L, Morillon B, Liegeois-Chauvel C, Giraud AL. The contribution of frequency-specific activity to hierarchical information processing in the human auditory cortex. Nat Commun. 2014; 5:4694. https:// doi.org/10.1038/ncomms5694 PMID: 25178489 40. Hyafil A, Fontolan L, Kabdebon C, Gutkin B, Giraud AL. Speech encoding by coupled cortical theta and gamma oscillations. Elife. 2015; 4:e06213. 41. 42. Lakatos P, Shah AS, Knuth KH, Ulbert I, Karmos G, Schroeder CE. An oscillatory hierarchy controlling neuronal excitability and stimulus processing in the auditory cortex. J Neurophysiol. 2005; 94(3):1904– 1911. https://doi.org/10.1152/jn.00263.2005 PMID: 15901760 van der Meij R, Kahana M, Maris E. Phase-amplitude coupling in human electrocorticography is spa- tially distributed and phase diverse. J Neurosci. 2012; 32(1):111–123. https://doi.org/10.1523/ JNEUROSCI.4816-11.2012 PMID: 22219274 43. Bruyer R, Brysbaert M. Combining Speed and Accuracy in Cognitive Psychology: Is the Inverse Effi- ciency Score (IES) a Better Dependent Variable than the Mean Reaction Time (RT) and the Percentage Of Errors (PE)? Psychologica Belgica. 2011; 51(1):1–5. 44. Williamson VJ, McDonald C, Deutsch D, Griffiths TD, Stewart L. Faster decline of pitch memory over time in congenital amusia. Adv Cogn Psychol. 2010; 6:15–22. https://doi.org/10.2478/v10053-008- 0073-5 PMID: 20689638 45. Albouy P, Cousineau M, Caclin A, Tillmann B, Peretz I. Impaired encoding of rapid pitch information underlies perception and memory deficits in congenital amusia. Sci Rep. 2016; 6:18861. https://doi.org/ 10.1038/srep18861 PMID: 26732511 46. Siegel M, Donner TH, Engel AK. Spectral fingerprints of large-scale neuronal interactions. Nat Rev Neu- rosci. 2012; 13(2):121–134. https://doi.org/10.1038/nrn3137 PMID: 22233726 47. Fries P. Rhythms for Cognition: Communication through Coherence. Neuron. 2015; 88(1):220–235. https://doi.org/10.1016/j.neuron.2015.09.034 PMID: 26447583 48. Albouy P, Baillet S, Zatorre RJ. Driving working memory with frequency-tuned noninvasive brain stimu- lation. Ann N Y Acad Sci. 2018. https://doi.org/10.1111/nyas.13664 PMID: 29707781 49. Albouy P, Martinez-Moreno ZE, Hoyer RS, Zatorre RJ, Baillet S. Supramodality of neural entrainment: Rhythmic visual stimulation causally enhances auditory working memory performance. Sci Adv. 2022; 8(8):eabj9782. https://doi.org/10.1126/sciadv.abj9782 PMID: 35196074 50. Albouy P, Weiss A, Baillet S, Zatorre RJ. Selective Entrainment of Theta Oscillations in the Dorsal Stream Causally Enhances Auditory Working Memory Performance. Neuron. 2017; 94(1):193–206 e5. https://doi.org/10.1016/j.neuron.2017.03.015 PMID: 28343866 51. Roux F, Uhlhaas PJ. Working memory and neural oscillations: alpha-gamma versus theta-gamma codes for distinct WM information? Trends Cogn Sci. 2014; 18(1):16–25. 52. Backus AR, Schoffelen JM, Szebenyi S, Hanslmayr S, Doeller CF. Hippocampal-Prefrontal Theta Oscil- lations Support Memory Integration. Curr Biol. 2016; 26(4):450–457. https://doi.org/10.1016/j.cub.2015. 12.048 PMID: 26832442 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 22 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain 53. Violante IR, Li LM, Carmichael DW, Lorenz R, Leech R, Hampshire A, et al. Externally induced fronto- parietal synchronization modulates network dynamics and enhances working memory performance. Elife. 2017:6. https://doi.org/10.7554/eLife.22001 PMID: 28288700 54. Hanslmayr S, Axmacher N, Inman CS. Modulating Human Memory via Entrainment of Brain Oscilla- tions. Trends Neurosci. 2019; 42(7):485–499. https://doi.org/10.1016/j.tins.2019.04.004 PMID: 31178076 55. Albouy P, Mattout J, Bouet R, Maby E, Sanchez G, Aguera PE, et al. Impaired pitch perception and memory in congenital amusia: the deficit starts in the auditory cortex. Brain. 2013; 136(Pt 5):1639– 1661. https://doi.org/10.1093/brain/awt082 PMID: 23616587 56. Albouy P, Peretz I, Bermudez P, Zatorre RJ, Tillmann B, Caclin A. Specialized neural dynamics for ver- bal and tonal memory: fMRI evidence in congenital amusia. Hum Brain Mapp. 2019; 40(3):855–867. https://doi.org/10.1002/hbm.24416 PMID: 30381866 57. Malinovitch T, Albouy P, Zatorre RJ, Ahissar M. Training allows switching from limited-capacity manipu- lations to large-capacity perceptual processing. Cereb Cortex. 2023; 33(5):1826–1842AU : Pleasenotethatdetailshavebeenaddedtoref :57:Pleasecheckandconfirmthatthesearecorrect: 10.1093/cercor/bhac175 PMID: 35511687 . https://doi.org/ 58. Keller AS, Payne L, Sekuler R. Characterizing the roles of alpha and theta oscillations in multisensory attention. Neuropsychologia. 2017; 99:48–63. https://doi.org/10.1016/j.neuropsychologia.2017.02.021 PMID: 28259771 59. Kaminski J, Sullivan S, Chung JM, Ross IB, Mamelak AN, Rutishauser U. Erratum: Persistently active neurons in human medial frontal and medial temporal lobe support working memory. Nat Neurosci. 2017; 20(8):1189. https://doi.org/10.1038/nn0817-1189d PMID: 28745722 60. Poppenk J, Evensmoen HR, Moscovitch M, Nadel L. Long-axis specialization of the human hippocam- pus. Trends Cogn Sci. 2013; 17(5):230–240. https://doi.org/10.1016/j.tics.2013.03.005 PMID: 23597720 61. Strange BA, Witter MP, Lein ES, Moser EI. Functional organization of the hippocampal longitudinal axis. Nat Rev Neurosci. 2014; 15(10):655–669. https://doi.org/10.1038/nrn3785 PMID: 25234264 62. Dimakopoulos V, Megevand P, Stieglitz LH, Imbach L, Sarnthein J. Information flows from hippocam- pus to auditory cortex during replay of verbal working memory items. Elife. 2022:11. https://doi.org/10. 7554/eLife.78677 PMID: 35960169 63. Albouy P, Caclin A, Norman-Haignere SV, Leveque Y, Peretz I, Tillmann B, et al. Decoding Task- Related Functional Brain Imaging Data to Identify Developmental Disorders: The Case of Congenital Amusia. Front Neurosci. 2019; 13:1165. https://doi.org/10.3389/fnins.2019.01165 PMID: 31736698 64. Albouy P, Mattout J, Sanchez G, Tillmann B, Caclin A. Altered retrieval of melodic information in con- genital amusia: insights from dynamic causal modeling of MEG data. Front Hum Neurosci. 2015; 9:20. https://doi.org/10.3389/fnhum.2015.00020 PMID: 25698955 65. Samiee S, Vuvan D, Florin E, Albouy P, Peretz I, Baillet S. Cross-frequency brain network dynamics support pitch change detection. J Neurosci. 2022; 42(18):3823–3835. https://doi.org/10.1523/ JNEUROSCI.0630-21.2022 PMID: 35351829 66. 67. Zatorre RJ, Belin P, Penhune VB. Structure and function of auditory cortex: music and speech. Trends Cogn Sci. 2002; 6(1):37–46. https://doi.org/10.1016/s1364-6613(00)01816-7 PMID: 11849614 Zatorre RJ, Evans AC, Meyer E. Neural mechanisms underlying melodic perception and memory for pitch. J Neurosci. 1994; 14(4):1908–1919. https://doi.org/10.1523/JNEUROSCI.14-04-01908.1994 PMID: 8158246 68. Gaab N, Gaser C, Zaehle T, Jancke L, Schlaug G. Functional anatomy of pitch memory—an fMRI study with sparse temporal sampling. Neuroimage. 2003; 19(4):1417–1426. https://doi.org/10.1016/s1053- 8119(03)00224-6 PMID: 12948699 69. Schulze K, Gaab N, Schlaug G. Perceiving pitch absolutely: comparing absolute and relative pitch pos- sessors in a pitch memory task. BMC Neurosci. 2009; 10:106. https://doi.org/10.1186/1471-2202-10- 106 PMID: 19712445 70. Treder MS, Charest I, Michelmann S, Martin-Buro MC, Roux F, Carceller-Benito F, et al. The hippocam- pus as the switchboard between perception and memory. Proc Natl Acad Sci U S A. 2021; 118(50): e2114171118. https://doi.org/10.1073/pnas.2114171118 PMID: 34880133 71. Alekseichuk I, Turi Z, Amador de Lara G, Antal A, Paulus W. Spatial Working Memory in Humans Depends on Theta and High Gamma Synchronization in the Prefrontal Cortex. Curr Biol. 2016; 26 (12):1513–1521. https://doi.org/10.1016/j.cub.2016.04.035 PMID: 27238283 72. Buzsaki G, Draguhn A. Neuronal oscillations in cortical networks. Science. 2004; 304(5679):1926– 1929. https://doi.org/10.1126/science.1099745 PMID: 15218136 73. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, Crivello F, Etard O, Delcroix N, et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 23 / 24 PLOS BIOLOGYCross-frequency coupling enables integration and memory of auditory information in the human brain 74. 75. single-subject brain. Neuroimage. 2002; 15(1):273–289. https://doi.org/10.1006/nimg.2001.0978 PMID: 11771995 Jerbi K, Ossandon T, Hamame CM, Senova S, Dalal SS, Jung J, et al. Task-related gamma-band dynamics from an intracerebral perspective: review and implications for surface EEG and MEG. Hum Brain Mapp. 2009; 30(6):1758–1771. https://doi.org/10.1002/hbm.20750 PMID: 19343801 Tallon-Baudry C, Bertrand O. Oscillatory gamma activity in humans and its role in object representation. Trends Cogn Sci. 1999; 3(4):151–162. https://doi.org/10.1016/s1364-6613(99)01299-1 PMID: 10322469 76. Ozkurt TE, Schnitzler A. A critical note on the definition of phase-amplitude cross-frequency coupling. J Neurosci Methods. 2011; 201(2):438–443. https://doi.org/10.1016/j.jneumeth.2011.08.014 PMID: 21871489 77. Brett M, Anton JL, Valabregue R, Poline JB. Region of interest analysis using the MarsBar toolbox for SPM 99. Neuroimage. 2002; 16(Suppl 1:S497). 78. Bates D, Ma¨chler M, Bolker B, Walker S. Fitting Linear Mixed-Effects Models Using lme4. J Stat Softw. 2015; 67(1):1–48. 79. Fox J, Weisberg S. An R Companion to Applied Regression. Thousand Oaks, CA: Sage; 2019. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002512 March 5, 2024 24 / 24 PLOS BIOLOGY
10.1371_journal.pbio.3002555	RESEARCH ARTICLE Specification of distinct cell types in a sensory- adhesive organ important for metamorphosis in tunicate larvae Christopher J. Johnson1☯, Florian Razy-Krajka1☯, Fan Zeng2, Katarzyna M. Piekarz1, Shweta Biliya3, Ute Rothba¨ cher2, Alberto StolfiID 1 1 School of Biological Sciences, Georgia Institute of Technology, Atlanta, Georgia, United States of America, 2 Department of Zoology, University of Innsbruck, Innsbruck, Austria, 3 Molecular Evolution Core, Petit H. Parker Institute for Bioengineering and Bioscience, Georgia Institute of Technology, Atlanta, Georgia, United States of America ☯ These authors contributed equally to this work. * [email protected] (UR); [email protected] (AS) Abstract The papillae of tunicate larvae contribute sensory, adhesive, and metamorphosis-regulating functions that are crucial for the biphasic lifestyle of these marine, non-vertebrate chordates. We have identified additional molecular markers for at least 5 distinct cell types in the papil- lae of the model tunicate Ciona, allowing us to further study the development of these organs. Using tissue-specific CRISPR/Cas9-mediated mutagenesis and other molecular perturbations, we reveal the roles of key transcription factors and signaling pathways that are important for patterning the papilla territory into a highly organized array of different cell types and shapes. We further test the contributions of different transcription factors and cell types to the production of the adhesive glue that allows for larval attachment during settle- ment, and to the processes of tail retraction and body rotation during metamorphosis. With this study, we continue working towards connecting gene regulation to cellular functions that control the developmental transition between the motile larva and sessile adult of Ciona. Introduction Tunicates, the sister group to the vertebrates, comprise a diverse group of marine non-verte- brate chordates [1,2]. Most tunicate species are classified in the order Ascidiacea, commonly known as ascidians [3], although phylogenetic evidence suggests this is not a monophyletic group within Tunicata [4–6]. The majority of ascidians have a biphasic life cycle that alternates between a swimming larva and a sessile adult. The larva functions exclusively to disperse the species, not feeding until it has found a suitable location on which to settle and trigger meta- morphosis [7]. Recent work has started to reveal the cellular and molecular basis of larval settlement and metamorphosis. Key to the process of settlement and metamorphosis are the papillae, which comprise a set of 3 anterior sensory/adhesive organs in the laboratory model species of the a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Johnson CJ, Razy-Krajka F, Zeng F, Piekarz KM, Biliya S, Rothba¨cher U, et al. (2024) Specification of distinct cell types in a sensory- adhesive organ important for metamorphosis in tunicate larvae. PLoS Biol 22(3): e3002555. https:// doi.org/10.1371/journal.pbio.3002555 Academic Editor: Selene L Fernandez-Valverde, University of New South Wales - Kensington Campus: University of New South Wales, AUSTRALIA Received: May 2, 2023 Accepted: February 21, 2024 Published: March 13, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pbio.3002555 Copyright: © 2024 Johnson et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting information PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 1 / 31 files, and from OSF: https://osf.io/wzrdk/ and https://osf.io/sc7pr/. Raw sequencing reads available from the SRA database under accession PRJNA949791. Funding: This work was funded by National Science Foundation (NSF, www.nsf.gov) grant 1940743 and National Insititutes of Health (www. nih.gov) grant GM143326 to AS; an NSF graduate fellowship to CJJ; and by Austrian Science Fund (FWF, www.fwf.ac.at) grant P 35402-B to UR. The funders did not play any role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abbreviations: ACC, axial columnar cell; CEN, caudal epidermal neuron; FP, fluorescent protein; IC, inner collocyte; OC, outer collocyte; PN, papilla neuron; PNA, peanut agglutinin; PSC, palp sensory cell; RTEN, rostral trunk epidermal neuron; scRNAseq, single-cell RNA sequencing; sgRNA, single-chain guide RNA; TEM, transmission electron microscopy. Specification of distinct cell types in tunicate papillae genus Ciona and a majority of other ascidian genera as well (Fig 1) [8–11]. The papillae are composed of a few different cell types that have been characterized by both electron and fluo- rescence microscopy [9,12–14]. Several cells appear to secrete the “glue” or bioadhesive mate- rial required for the attachment of the larva to the substrate, termed “collocytes” [9,15]. Other cells are clearly neuronal (4 ciliated neurons per papilla) [9] and are required to trigger the onset of metamorphosis [16], which was also recently shown to depend on mechanical stimu- lation of the papillae [17]. Finally, at the very center of each papilla are 4 “Axial Columnar Cells” (ACCs), which have been suggested to possess chemosensory and contractile properties [11,18,19]. Although they have been called papilla “sensory cells” or “neurons,” they are not innervated and have little structural and molecular overlap with the other 2 cell types. Further- more, single-cell RNA sequencing (scRNAseq) revealed that they do not express genes typi- cally associated with neuronal function [20]. In Ciona, previous work had established that the 3 papillae likely arise from 3 clusters of Foxg+/Islet+ cells arranged roughly as a triangle—2 dorsal clusters (left and right) and single ventral cluster [21,22]. Although Foxg is initially activated in an entire row of cells at the very anterior of the neural plate, Sp6/7/8 (also known as Zfp220 or Buttonhead) is required to refine this swath of expression down to 3 “spots” of Foxg, which is required for expression of Islet in these cell clusters (Fig 1) [21]. MEK/ERK (e.g., MAPK) signaling also appears to play an important role in this refinement, as treatment with the MEK inhibitor U0126 results in a “U”-shaped band of Islet expression instead of 3 discrete foci (Fig 1) [22]. Similarly, BMP inhi- bition also causes a similar “U-shape” swath of Foxg/Islet expression, resulting in a single pro- trusion instead of the normal 3, termed the “cyrano” phenotype [23,24]. However, it has not been shown how these early specification events connect to the final cell type diversity and arrangement of the papillae. Here, we describe novel genetic markers and reporter constructs that allowed us to visualize each of the different cell type of the papillae and follow their development upon various molec- ular perturbations targeting specific transcription factors or signaling pathways. We show that different transcription factors contribute to the specification of the different cell types and that cell-cell signaling in the FGF/MAPK and Delta/Notch pathways are crucial for patterning and arranging these cells in the 3 papillae. Altering papilla development in different ways contrib- utes to different processes of post-settlement larval body plan rearrangements, revealing the complex molecular and cellular underpinning of tunicate larval metamorphosis. Methods Ciona handling Ciona robusta (intestinalis Type A) were shipped from San Diego (M-REP), while Ciona intes- tinalis (Type B) were shipped from Roscoff Biological Station, France. Eggs were fertilized in vitro, dechorionated, and electroporated following established protocols [25–27]. Staging at different temperatures was estimated based on the published C. robusta developmental table from TUNICANATO [28]. Unc-76 tags were used as a default for fluorescent proteins (FPs) for optimal cell labeling as previously described [29], which excludes the FPs from the nucleus and ensures transport down axons. Typically, 40 to 100 μg of untagged or Unc-76-tagged FP plasmids and 10 to 35 μg of histone (H2B) fusion FP plasmids were used per 700 μl of electro- poration solution. For CRISPR, typically 35 to 40 μg of Cas9 plasmid and 25 to 40 μg of each gRNA plasmid was used per 700 μl of electroporation solution, except when validating single- chain guide RNAs (sgRNAs) (see further below). For Sp6/7/8, Pou4, and Foxg, the 2 sgRNAs validated for each gene (S3 Fig) were used in combination. For Villin, all 3 validated sgRNAs were used in combination, while Tuba3 only had 1 validated sgRNA. For Islet, most PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 2 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 1. Development of the papillae of Ciona. (A) Diagram showing the early cell lineages that give rise to the papillae. The papillae invariantly derive from Foxc+ cells in the anterior neural plate, more specifically the anterior daughter cells of “Row 6” of the neural plate, which activate Foxg downstream of Foxc. Foxg is also activated in the posterior daughter cells of “Row 5,” which go on to give rise to part of the OSP. Numbers in each cell indicate their invariant identity according to the Conklin cell lineage nomenclature. Black bars indicate sibling cells born from the same mother cell. (B) Diagram of what is currently known about the later lineage and fates of the Foxg+ “Anterior Row 6” cells shown in panel A. As the cells divide mediolaterally, some cells up-regulate Sp6/7/8 and down-regulate Foxg (gray cells). Those cells that maintain Foxg expression turn on Islet and coalesce as 3 clusters of cells (pink with green outline): 1 medial, more ventral cluster, and 2 left/right, more dorsal clusters. Later, these 3 clusters organize the territory into the 3 protruding papillae of the larva, which contains several cell types described in detail by TEM [9]. Dashed cell outlines indicate uncertain number/provenance of cells. A-P: anterior-posterior. D-V: dorsal-ventral. Lineages and gene networks are based mostly on: [21,22,86,87]. OSP, oral siphon primordium; TEM, transmission electron microscopy. https://doi.org/10.1371/journal.pbio.3002555.g001 experiments used only the Islet.2 sgRNA, unless otherwise specified. Precise electroporation mixes for given perturbation experiments and controls are specified in the S1 File. C. robusta embryos were raised at 20 ˚C and C. intestinalis embryos were raised at 18 ˚C, unless otherwise specified. For U0126 treatment, U0126 stock solution resuspended in DMSO was diluted to 10 μm final concentration in artificial seawater prior to transferring embryos at stage 16 (approximately 7.5 hpf). Negative control embryos were transferred to seawater with the equivalent volume of DMSO vehicle. For DMH1 treatment, concentrated stock solution was diluted to 2.5 μm final concentration in artificial seawater prior to transferring embryos at stage 10 (4 hpf), as previously established [23]. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 3 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fixation, staining, imaging, scoring, and statistical analyses Embryos and larvae were fixed for fluorescent protein imaging in MEM-FA fixation solution (3.7% formaldehyde, 0.1 M MOPS (pH 7.4), 0.5 M NaCl, 1 mM EGTA, 2 mM MgSO4, 0.1% Triton-X100), rinsed in 1× PBS, 0.4% Triton-X100, 50 mM NH4Cl, and 1× PBS, 0.1% Triton- X100. For mRNA in situ hybridization, embryos/larvae were fixed in MEM-PFA fixation solu- tion (4% paraformaldehyde, 0.1 M MOPS (pH 7.4), 0.5 M NaCl, 1 mM EGTA, 2 mM MgSO4, 0.05% Tween-20) and in situ hybridization was carried out as previously described [30,31]. All probe template sequences are shown in the S1 File. Immunolabeling of Flag-tag (DYKDDDDK), β-galactosidase, mCherry (alone or in conjunction with mRNA in situ hybridization) was carried out as previously described [32], on embryos/larvae using mouse anti-DYKDDDDK Tag (Thermo Fisher catalog number MA1-91878, 1:1,000), mouse anti-β- gal (Promega catalog number Z3781, 1:1,000), and rabbit anti-mCherry (BioVision, accession number ACY24904, 1:500) primary antibodies. Specimens were mounted in 50% glycerol/1X PBS/2% DABCO mounting solution on slides with double-sided tape spacing between the slide and coverslip and imaged on Leica DM IL LED or DMI8 inverted epifluorescence micro- scopes, with maximum Z projection processing and cell measurements performed in LAS X. PNA staining was carried out on 4% PFA fixed larvae, using Tris-buffered saline (pH 8.0) supplemented with 5 mM CaCl2 and 0.1% Triton X-100 (TBS-T). Unspecific background was blocked by 3% BSA in TBS-T for 2 h at room temperature. Biotinylated peanut agglutinin (PNA; Vector Laboratories, B-1075) was diluted in BSA-TBS-T to a final concentration of 25 μg/ml and applied to the specimen overnight at 4 ˚C. After several washes in TBS-T over 2 h, larvae were incubated for 1 h in fluorescent streptavidin (Vector Laboratories, SA-5006) diluted 1:300 in BSA-TBS-T at room temperature. PNA stainings were mounted in Vectashield (Vector Laboratories, H-1000-10) and imaged using a Leica SP5 II confocal scanning micro- scope. Stacks were acquired sequentially and z-projected. Images were analyzed with ImageJ (Version 1.52 h). Only Foxc>H2B::mCherry+/lacZ+ embryos, larvae, and juveniles were scored, unless oth- erwise noted in results or figure and legend. For tests of proportion between 2 groups where there were 2 outcomes, Fisher’s exact test was used, while for tests of proportion between more than 2 groups/outcomes, chi-square test was performed. All results of tests of proportions shown in S4 Data. For continuous variable measurements, see “Quantitative image analyses” subsection below. CRISPR/Cas9 sgRNA design and validation The Cas9 [33] and Cas9::Geminin-Nterminus [34] protein-coding sequences have been described before. sgRNAs were designed using the CRISPOR website [35](crispor.tefor.net). Those sgRNAs with high Doench ‘16 score, high MIT specificity score, and not spanning known SNPs were selected for testing. Validation of sgRNAs was performed by co-electropo- ration 25 μg of Eef1a>Cas9 or Eef1a>Cas9::Geminin-Nterminus and 75 μg of the sgRNA plas- mid, per 700 μl of total electroporation volume. Genomic DNA was extracted from pooled larvae electroporated with a single sgRNA, using the QIAamp DNA micro kit (Qiagen). PCR products spanning each sgRNA target site were amplified from the corresponding genomic DNA, with primers designed so that the amplicon was to be 150 to 450 bp in size. Amplicons were purified by QIAquick PCR purification kit (Qiagen) and submitted for Amplicon-EZ Illumina-based sequencing by Azenta/Genewiz (New Jersey, United States of America), which returned mutagenesis rates and indel plots. CRISPR “rescue” cDNAs for Islet, Foxg, and Sp6/7/ 8 were designed with silent (i.e., synonymous) point mutations disrupting our sgRNA targets sites and/or their PAMs (see S1 File). In the case of the Islet.2 sgRNA, this one binds to a PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 4 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae sequence at an intron/exon boundary and therefore no mutation was needed for the rescue cDNA. RNA sequencing and analysis The scRNAseq data from Cao and colleagues were re-analyzed in Seurat [36]. Combined larva stage data was clustered and plotted using 30 dimensions (S1A Fig). Clusters 3 and 33 were determined to contain papilla cell types and were re-clustered separately, also using 30 dimen- sions (S1B Fig). Differential gene expression plots (S1C Fig) were explored to find candidate papilla cell type markers, which appeared to be enriched in subclusters 8 and 9 (S1 Data). Some were then confirmed by in situ hybridization (S 1D Fig) and/or reporter plasmids. All code and Seurat files can be downloaded from: https://osf.io/sc7pr/. An alternative filtering and clustering approach (https://osf.io/dbv42) used in parallel to find specifically papilla neu- rons (PNs) resulted in a different TSNE plot (https://osf.io/6cg4h). From this, clusters 8, 10, and 18 were selected based on known papilla cell type markers and re-clustered, which led to the identification of a new subcluster “10” enriched for both ACC and PN markers. ACC markers (cluster “J”) identified in Sharma and colleagues were subtracted from subcluster 10 markers to generate a list of potential PN-specific markers (https://osf.io/7xqp2). Bulk RNA integrity numbers were determined using the Agilent Bioanalyzer RNA 6000 Nano kit and used as a QC measure. All samples with RINs over 7 were used for library prepa- ration. mRNA was enriched using the NEBNext Poly(A) mRNA isolation module and Illu- mina compatible libraries were prepared using the NEBNext Ultra II RNA directional library preparation kit. QC on the libraries was performed on the Agilent Bioanalyzer 2100 and con- centrations were determined fluorometrically. The libraries were then pooled and sequenced on the NovaSeq 6000 with an SP Flow Cell to get PE100bp reads. The RNA-seq raw files were analyzed in Galaxy hub (usegalaxy.org) [37]. Firstly, the raw fastq files were inspected using FastQC Read Quality Reports (Galaxy Version 0.73+galaxy0) and MultiQC (Galaxy Version 1.11+galaxy0). The reads were then filtered and trimmed with Cutadapt (Galaxy Version 4.0+galaxy0). The minimum read length was set to 20 and the reads that did not meet the quality cutoff of 20 were discarded. Then, FastQC and MultiQC were used again to assess the resulting files after filtering and trimming. Next, the technical repli- cates were combined and used as the input to the mapping tool (RNA STAR, Galaxy Version 2.7.8a+galaxy0, length of the SA pre-indexing string of 12), together with the custom “KY21” version of the Ciona reference genome sequence and gene models (“Kyoto 2021”, obtained from the Ghost Database; http://ghost.zool.kyoto-u.ac.jp/download_ht.html) [38]. The counts were generated using featureCounts (Galaxy Version 2.0.1+galaxy2; minimum mapping qual- ity per gene was set to 10). Lastly, the differential gene expression analysis (S2 Data) was per- formed with DESeq2 (Galaxy Version 2.11.40.7+galaxy1). KY21 gene models were linked to KyotoHoya (“KH”) version gene models using the Ciona Gene Model Converter application https://github.com/katarzynampiekarz/ciona_gene_model_converter. Raw sequencing reads available from the SRA database under accession PRJNA949791. Analysis code and files can be found at: https://osf.io/wzrdk/. Quantitative image analyses Larvae subjected to papilla-specific knockout of Islet, Villin, or Tuba3 (using Foxc>Cas9, see S1 File for detailed electroporation recipes) and negative control larvae were fixed at 17 hpf, 20 ˚C and mounted as above. Islet intron 1 + -473/-9>Unc-76::GFP+ or CryBG>Unc-76::GFP+ cells were imaged using a K3M camera mounted on a Leica DMI8 inverted epifluorescence microscope and the greatest distance between the apical and the basal extremities of each GFP PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 5 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae + papilla (not individual cells) in LAS X, based on visible GFP fluorescence in the ACCs at a given focal plane (see example images with superimposed lines and measurements in S8D Fig). Sometimes, 2 or more papillae were GFP+ in the same larva. In these cases, each papilla was measured independently. Individual papilla length measurements are listed in S3 Data. For analysis of Islet perturbations on Villin reporter expression, fluorescence from Villin -1978/-1>Unc-76::GFP and Foxc>H2B::mCherry reporters were acquired as above but with fixed illumination intensity and exposure times (50 ms for GFP, 100 ms for mCherry). Mean fluorescence intensities in both channels (GFP, mCherry) were measured in mCherry+ areas corresponding to the papillae, and ratio of mean values (GFP/mCherry) was calculated. Results Identification of novel markers and reporters for specific cell types in the papillae We searched Ciona robusta (i.e., intestinalis Type A) whole-larva scRNAseq data [39] for evi- dence of the cell types described by transmission electron microscopy (TEM) of the papillae [9]. While a cell cluster annotated as “Palp Sensory Cells” (PSCs) appeared enriched for known markers of ACCs like CryBG (KH.S605.3) and KH.C3.516 [20,40], genes expressed in other papilla cell types were also enriched in this cluster as well, including Sp6/7/8 (KH.C13.22) [21,22] and Pou4 (KH.C2.42) [16,23]. Re-analysis and re-clustering of these data revealed novel potential markers for different cell types in and around the papillae (S1A–S1C Fig and S1 Data). We performed in situ mRNA hybridization for several of these candidate markers in C. robusta larvae (S1D Fig). As we had hoped, they appeared to label different cells in the papilla territory. Some appeared to label cells in the center of each papilla, while others were expressed in cells surrounding or on the outermost edges of each papilla. These vastly different expres- sion patterns supported the idea of mixed cell identities in the PSC scRNAseq cluster. To further confirm the expression patterns of these and other candidate markers, we made reporter plasmids from their upstream cis-regulatory sequences and electroporated these into Ciona embryos. None of the selected genes showed any appreciable homology to genes of known function in other organisms, but we reasoned that they might serve as useful markers for specific papilla cell types. First, the gene KH.L96.43, predicted to encode a secreted protein with TSP1 repeats and a trypsin-like serine protease domain (S1E Fig and S1 File), was expressed in cells surrounding and in between the 3 papillae (S1D Fig). This pattern was reca- pitulated by a KH.L96.43 reporter plasmid (“L96.43>GFP,” Fig 2A). Co-electroporation with the papilla-specific Foxg>mCherry reporter [39] showed clear, mutually exclusive expression between the 2 reporters. We propose that L96.43 marks a population of “peri-papillary” and/or “inter-papillary” cells previously identified as “basal cells” that are part of the larger papilla region but excluded from the 3 protruding, Foxg+ papillae sensu stricto [9]. Next, we further confirmed that the PNs are distinct from the ACCs [9]. Previously identi- fied as a potential PN marker by in situ hybridization [41], a TGFB reporter clearly labeled PNs (Fig 2B and S2A Fig), which are distinguished as the only papilla cell types bearing an axon [9]. However, co-electroporation of TGFB reporter with an ACC-specific CryBG reporter [40] appeared to result in “cross-talk,” or cross-plasmid transvection in which a cis-regulatory element in one plasmid activates the transcription of a reporter protein-encoding gene on another, distinct co-electroporated plasmid (S2B Fig). Indeed, other PN-specific reporters tested did not cross-talk with CryBG, including the previously published Gnrh1 reporter [42], and the novel reporter KH.C4.78 (“C4.78>GFP”) (S2C and S2D Fig). KH.C4.78 encodes a pre- dicted transmembrane protein with a single extracellular Sel1-like repeat (S1E Fig and S1 File). Interestingly, PN axons continued to extend posteriorly during the swimming phase to contact PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 6 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 2. Novel genetic markers label distinct cell types of the papillae. (A) GFP reporter plasmid (green) constructed using the cis-regulatory sequences from the KH.L96.43 gene labels basal cells in between and surrounding the protruding papillae labeled by Foxg reporter plasmid (pink). (B) TGFB>GFP reporter (green) labels PNs, the axons of which make contacts with BTN axons labeled by a BTN-specific Islet reporter (pink), at 23.5 h hpf, approximately corresponding to Hotta stage 30. (C) A KH.C4.78 reporter (C4.78>GFP) also labels PNs, which are also labeled by Foxg>H2B::mCherry (mCh) reporter (pink nuclei). (D) Lack of overlap between expression of C4.78>GFP (green) and a papilla-specific Islet reporter plasmid (pink nuclei) showing that PNs do not arise from Islet+ cells. (E, F) Co-electroporation of C11.360>GFP (green) with H2B::mCherry reporter plasmids (pink nuclei) indicates these cells come from Foxg-expressing cells that also express Islet. (G) C11.360>mCherry reporter (pink) labels centrally located ICs adjacent to ACCs labeled by CryBG>LacZ reporter (green). (H, I) L141.36>GFP reporter (green) labels OCs that arise from Foxg+ cells (pink nuclei) but do not express Islet (pink nuclei). (J) ICs and OCs are distinct cells as there is no overlap between C11.360 (green) and L141.36 (pink) reporter plasmid expression. (K) Ciona intestinalis (Type B) larva ICs labeled with a reporter plasmid made from the corresponding cis-regulatory sequence of the C. intestinalis Chr11.1038 gene, orthologous to C. robusta KH.C11.360. (L) C. intestinalis larva OCs labeled by a Chr7.130 reporter, corresponding to the C. robusta ortholog KH.L141.36. (M) Summary of the main marker genes and corresponding reporter plasmids used in this study to label different subsets of papilla progenitors and their derivative cell types. All GFP and mCherry reporters fused to the Unc-76 tag, unless specified (see Methods and supplement for details). Weaker Foxg -2863/-3 promoter used in panel A, all other Foxg reporters used the improved Foxg -2863/+54 sequence instead. All Islet reporters shown correspond to the Islet intron 1 + bpFOG>H2B::mCherry plasmid. White channel shows either DAPI (nuclei) and/or larva outline in brightfield, depending on the panel. All C. robusta raised at 20 ˚C to 18 hpf (roughly st. 28) except: panel B (23.5 hpf, ~st. 30); panels C–F (17 hpf, ~st. 27); panels H–J (20 hpf, ~st. 29). C. intestinalis raised at 18 ˚C to 20–22 hpf (Hotta stage 28). ACC, axial columnar cell; BTN, bipolar tail neuron; hpf, hours post-fertilization; IC, inner collocyte; OC, outer collocyte; PN, papilla neuron. https://doi.org/10.1371/journal.pbio.3002555.g002 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 7 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae the anterior axon branches of the bipolar tail neurons (Fig 2B), which project their posterior axon branches to the very tip of the tail [43]. This hints at a potential mechanism for trans- ducing sensory information from the papillae to the tail tip where tail retraction initiates, espe- cially during later time points when larvae are competent to settle [44]. Double electroporation with KH.C4.78 and Foxg reporters (Fig 2C and S2D Fig) revealed that, unlike the basal cells, PNs are specified from Foxg+ cells in the papillae. However, co-elec- troporation with a papilla-specific Islet reporter plasmid also revealed that PNs are adjacent to, but distinct from, the central Islet+ “core” of each papilla (Fig 2D and S2D Fig). In contrast, a KH.C11.360 reporter (“C11.360>GFP/mCherry”) labeled cells that were both Foxg+ and Islet +, but were clearly not the ACCs (Fig 2E–2G and S2C Fig). The KH.C11.360 gene encodes a predicted secreted/transmembrane protein with no other recognizable domains or motifs (S1 File). The C11.360+ cells were adjacent to the ACCs but lacked the thin protrusions into the hyaline cap that are typical of the ACCs and also lacked axons typical of the PNs. Therefore, these cells appear to be collocytes, proposed to be adhesive-secreting cells responsible for attachment to the substrate during larval settlement [9]. Previous characterization of the papillae by TEM described 12 collocytes in each papilla [9], yet the C11.360 reporter appeared to only label at most 4 cells per papilla. This suggested the exis- tence of cryptic collocyte subtypes. In fact, those same TEM images showed certain qualitative differences in cytoplasmic contents between peripheral collocytes and the more central collocytes [9]. Indeed, we identified another reporter, that of the gene KH.L141.36 (“L141.36>GFP”), that labeled Foxg+ but Islet-negative cells that are at the periphery of each papilla but that are not PNs as they do not have axons (Fig 2H and 2I, and S2D Fig). KH.L141.36 encodes a predicted trans- membrane protein with at least 4 extracellular Sushi/SCR/CCP domains (S1E Fig and S1 File). Co-electroporation of L141.36 and C11.360 reporters labeled mutually exclusive groups of cells (Fig 2J and S2D Fig). We propose that these respective reporters delineate more peripheral, or “outer” collocytes (OCs) versus more central, or “inner” collocytes (ICs). Interestingly, strong KH.L141.36 reporter expression was not visible in early larvae (approximately 17 hpf) like most of the other reporters described, suggesting a later onset of activation. When using these C. robusta reporter plasmids to electroporate the closely related C. intesti- nalis (i.e., Type B) sourced from Roscoff, France [45], we noticed that their expression was very weak (S2E and S2F Fig). This led us to re-cloning the orthologous sequences from the C. intestinalis Type B genome [46] (S1 File). Percent identity over the alignable portions of these noncoding sequences (disregarding large gaps or insertions) was 89% for C11.360 and 66% for L141.36. Electroporation of Type B embryos with Type B-specific reporter plasmids resulted in much stronger, reliable expression (Fig 2K and 2L). This suggests relatively significant changes to the cis-regulatory sequences of these cell type-specific genes in these otherwise nearly indistinguishable cryptic species. Although we also obtained additional reporters that labeled 1 or more different papilla cell types (S2G and S2H Fig), we now had a full set of papilla cell type-specific marker genes and reporter plasmids for a deeper investigation of papilla patterning and development (Fig 2M). Finally, it is also important to note that some of these reporters are also expressed in cell types outside the papillae (e.g., CryBG in the otolith and KH.C4.78 in the descending decussating neurons of the motor ganglion). Specification of ACCs, ICs, and OCs by Islet and Sp6/7/8 combinatorial logic How are the cell types of the papillae (ACCs, ICs, OCs, and PNs) specified? In situ mRNA hybridization previously revealed partially overlapping expression territories of 3 genes PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 8 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae encoding sequence-specific transcription factors (Fig 3A): a central domain of Islet+ cells, sur- rounded by a ring of cells that express both Islet and Sp6/7/8 (and Emx, though distinct from the earlier expression of Emx at neurula stages), and additional cells surrounding them expressing only Sp6/7/8 [22]. Additionally, overexpression of Islet had been previously shown to generate a single large papilla expressing the ACC reporter CryBG>GFP [22]. We therefore asked whether these transcription factors might be patterning the papillae into an ordered array of cell types (Fig 3A). First we asked, does Islet specify the centrally located ACCs and ICs? To test this, we turned to tissue-specific CRISPR/Cas9-mediated mutagenesis [33]. To knock out Islet in the papillae, we electroporated a previously validated sgRNA expression construct targeting its intron/exon 2 boundary (U6>Islet.2, 44% mutagenesis efficacy, S3 Fig) [47] together with Foxc>Cas9. Papilla-specific CRISPR-based knockout of Islet and resulting loss of ACC cell fate was con- firmed by loss of CryBG>GFP expression, compared to negative control individuals electropo- rated instead with previously published U6>Control sgRNA vector [33] targeting no sequence (Fig 3B and 3C, and S4A Fig). CryBG>GFP activation was rescued by expressing an Islet cDNA that is not targeted by our sgRNAs, demonstrating that its loss is not likely due to off- target CRISPR effects (S5A Fig). Therefore, we conclude that Islet is required for the specifica- tion and differentiation of ACCs. A smaller portion of larvae completely lost expression of the IC reporter, C11.360>GFP, but expression was still substantially reduced relative to the control (Fig 3B and 3C). This difference might be due to lower sensitivity of the IC reporter to Islet knockout, or might simply reflect the lower level of mosaicism of C11.360>GFP expression observed in the control. To test whether Islet is required for other cell types of the papillae, we repeated papilla-spe- cific Islet CRISPR knockout using our different reporters to monitor the specification or differ- entiation of OCs (L141.36>GFP) and PNs (TGFB>GFP). While Islet knockout altered the general morphology of the papillae (see further below), it did not cause significant loss of OC or PN reporter expression (Fig 3B and 3C, and S4A Fig). We therefore conclude that Islet is required for the specification and/or differentiation of ACCs and ICs, but not OCs or PNs. Because it was reported that an outer Emx+ “ring” of Islet+ cells in each papilla co-express Sp6/7/8 [22], we hypothesized that Sp6/7/8 might be required for a fate choice between ACCs and ICs. Corroborating the idea that these outer Islet+ cells are specified as ICs, we cloned an intronic cis-regulatory element from the Emx gene that is sufficient to drive late expression specifically in ICs (S2G Fig). This late ring of Emx expression is not to be confused with the earlier expression of Emx in Foxc+/Foxg-negative cells at the neurula stage [21], which repre- sent a distinct lineage (Fig 1). To test the role of Sp6/7/8 in IC versus ACC fate choice, we used the papilla-specific Islet cis-regulatory element to overexpress Islet or Sp6/7/8. While Islet>Islet did not reduce expression of either reporter (Fig 4A and 4B, with overexpression confirmed by immunostaining for a Flag tag epitope fused to Islet, S2I Fig), Islet>Sp6/7/8 specifically abol- ished ACC reporter expression, but not that of the IC reporter (Fig 4A and 4B). In fact, IC reporter expression appeared to be slightly expanded in approximately 29% of larvae electro- porated with Islet>Sp6/7/8, as assayed by perfect overlap with Islet>H2B::mCherry reporter (S5C Fig). Taken together, these results suggest that overexpression of Sp6/7/8 in the Islet + cells of the papillae is sufficient to abolish ACC fate and might convert the cells to an IC fate instead. To further show that the combination of Islet and Sp6/7/8 is sufficient to specify IC cell fate, we used the Foxc promoter to drive expression of Islet, Sp6/7/8, or a combination of both in the entire papilla territory. Foxc>Islet alone strongly promoted ACC reporter expression, as previously reported [22], but resulted in more scattered IC reporter expression (Fig 4C and 4D). In contrast, co-electroporation of Foxc>Islet and Foxc>Sp6/7/8 resulted more often in a PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 9 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 3. The transcription factor Islet is required for specification of ACCs and ICs. (A) Diagram depicting a partially overlapping expression patterns of Islet and Sp6/7/8, as originally shown by in situ mRNA hybridizations (Wagner and colleagues), and the correlation of these patterns with the later arrangement of ACCs, ICs, and OCs in the papillae. “Late” Emx expression in a ring of cells expressing both Islet and Sp6/7/8 appears to be distinct from earlier Emx expression in Foxg-negative cells (see text and S2 Fig for details). (B) Papilla lineage-specific CRISPR/Cas9-mediated mutagenesis of Islet using Foxc>Cas9 and a the U6>Islet.2 sgRNA plasmid shows reduction of larvae showing expression of reporters labeling ACCs and ICs, but not OCs or PNs. Results compared to a negative “control” condition using a negative control sgRNA (U6>Control, see text for details). Nuclei counterstained with DAPI (white). (C) Scoring data for larvae represented in panel B, averaged between biological duplicates. Foxc>H2B::mCherry+ larvae were scored for quantity of papillae showing visible expression of the corresponding GFP reporter plasmid. Due to mosaic uptake or retention of the plasmids after electroporation, number of papillae with GFP fluorescence is variable and rarely seen in all 3 papillae even in control larvae. Normally larvae have 3 papilla (GFP+ or not), but some mutants have more/fewer than 3. ACC/IC/OC subpanels in panel B at 20 hpf/20 ˚C (~st. 29), PN subpanels at 21 hpf/20 ˚C (~st. 29). Same applies to scoring data in Panel C. All experiments were performed in duplicate, with number of embryos ranging from 76 to 100 per condition per replicate. ** p < 0.0001 in both duplicates, as determined by chi-square test. ns = not statistically significant in at least 1 duplicate, also by chi-square test. See S4 Data for the data underlying the graphs and for statistical test details. ACC, axial columnar cell; IC, inner collocyte; OC, outer collocyte; PN, papilla neuron; sgRNA, single-chain guide RNA. https://doi.org/10.1371/journal.pbio.3002555.g003 large, single papilla expressing predominantly the IC reporter, not the ACC reporter (Fig 4C and 4D). Finally, we performed papilla-specific CRISPR knockout of Sp6/7/8, following the same strategy for Islet detailed above, using a combination of 2 new sgRNAs that were designed and validated (S3A Fig). Indeed, CRISPR/Cas9-mediated mutagenesis of Sp6/7/8 in the papilla ter- ritory resulted in loss of IC cell fate, as assayed by expression of C11.360>GFP (Fig 4C and 4E). Reduced IC reporter expression was also observed using either individual Sp6/7/8 sgRNAs independently (S4B Fig), and was overcome by Foxc promoter-driven overexpression of an Sp6/7/8 rescue cDNA (S5B Fig), suggesting this effect was not due to CRISPR off-targeting. In PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 10 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 4. Specification of ACCs, ICs, and OCs by a combinatorial logic of Islet and Sp6/7/8. (A) Overexpression of Sp6/7/8 (using the Islet>Sp6/7/8 plasmid) in all Islet+ papilla cells results in loss of ACCs (assayed by expression of CryBG>Unc-76::GFP, green), but not of ICs (assayed by expression of C11.360>Unc-76::GFP, green). Islet overexpression (with Islet>Flag::Islet-rescue) does not significantly impact the specification of ACCs or ICs. Larvae at 20 hpf/20 ˚C (~st. 29). (B) Scoring data showing presence or absence of ICs or ACCs in Foxc>H2B::mCherry+ larvae, as represented in panel A. Experiments were performed in duplicate with 99 or 100 larvae in each duplicate. (C) Cell type specification assayed by reporter plasmid expression (green) in larvae subjected to various Islet and/or Sp6/7/8 perturbation conditions (see main text for details). For ICs and ACCs, the “control” condition is negative control CRISPR (U6>Control), while for OCs it is Foxc>lacZ. Overexpression ACC/IC subpanels are at 18.5 hpf/20 ˚C (~st. 28), all CRISPR and OC panels at 20 hpf/20 ˚C (~st. 29). (D) Scoring data for most larvae represented in panel C. Foxc>H2B::mCherry+ larvae were scored for cell type-specific GFP reporter expression that was “heterogeneous” (mixed on/off GFP expression, with all “wild type” patterns of expression falling under this category), “predominant” (ectopic/supernumerary GFP+ cells), “sparse” (reduced frequency/intensity of GFP expression), or “absent” (no GFP visible). (E) IC or ACC reporter (C11.360>Unc-76::GFP, CryBG>Unc-76::GFP) expression scored in Foxc>H2B::mCherry+ larvae represented in top 2 panels of right-most column in C. Experiment was performed and scored in duplicate, with number of larvae in each duplicate ranging from 100 to 105. (F) OC-specific reporter (L141.36>Unc-76::GFP) expression scored in Foxc>H2B::mCherry+ larvae represented by the bottom/right-most subpanel in panel B. Scoring strategy same as in Fig 3. Asterisk denotes when a duplicate of the negative control condition was also used for plots in Fig 3, as multiple CRISPR experiments were performed in parallel. All experiments were performed in duplicate, with number of embryos ranging from 76 to 100 per duplicate. Foxc>Cas9 used for all CRISPR/Cas9 experiments. The Islet cis-regulatory sequence used (panels A and B) was always Islet intron 1 + -473/-9. For overexpression conditions, Foxc>lacZ or Islet>LacZ were used to normalize total amount of DNA (see S1 File for detailed electroporation recipes). All error bars indicate upper and lower limits. p < 0.0001 in both duplicates as determined by Fisher’s exact test (panels B and E) or chi- square test (panel F). ns = not statistically significant in at least 1 duplicate. See S4 Data for the data underlying the graphs and statistical test details. ACC, axial columnar cell; IC, inner collocyte; OC, outer collocyte. https://doi.org/10.1371/journal.pbio.3002555.g004 contrast, the same perturbation did not diminish the expression of the ACC reporter (Fig 4C and 4E). We noticed that Foxc>Sp6/7/8 alone resulted in a large proportion of larvae lacking either ACC or IC reporter expression (Fig 4D). This suggested the possibility that Sp6/7/8 alone PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 11 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae might be promoting another papilla cell fate. Indeed, we found that Sp6/7/8 knockout by CRISPR abolishes the expression of the OC reporter (L141.36>GFP), while Foxc>Sp6/7/8 expands it slightly (Fig 4C, 4D and 4F). In contrast, Foxc>Islet alone or in combination with Foxc>Sp6/7/8 suppressed OC reporter expression (Fig 4C and 4D), while Islet knockout did not affect it, as shown further above (Fig 3B and 3C, and S4A Fig). Taken together, these results suggest that a combinatorial transcriptional logic underlies papilla cell fate choices between ACCs (Islet alone), ICs (Islet + Sp6/7/8), and OCs (Sp6/7/8 alone). Identifying the adhesive-secreting cells of the papillae Previous data revealed PNA staining as a marker for glue-secreting cell granules, the adhesive papillary cap, and adhesive prints left by larvae on the substrate [9,15]. The delineation of 2 collocyte populations opened the question of whether both (ICs and OCs) are equally PNA- positive. To answer this question, we performed PNA stainings on larvae expressing IC or OC reporter plasmids (Fig 5A and 5B). Interestingly, ICs contained PNA-stained granules only at Fig 5. Both types of collocytes contribute to production of adhesive material. (A) PNA-stained granules (pink) are seen in the hyaline cap and the apical tip of ICs (left panel, white arrow) in a C. intestinalis larva labeled by the C. intestinalis C11.360>Unc-76::GFP reporter (green). PNA-stained granules are also seen in cells not labeled by the IC reporter (right subpanel, hollow arrowhead), suggesting they are localized in a different cell type. Left and right subpanels are from different focal planes of the same papilla. (B) OCs labeled with C. robusta L141.36>Unc-76::GFP (green) in a C. robusta larva, with PNA-stained granules (pink) in both apical and basal positions within the cell (white arrows). DAPI in blue. (C) PNA staining (pink) in C. robusta upon overexpression of Sp6/7/8 alone, showing reduction of IC specification as assayed by C11.360>Unc-76::GFP expression (green). Weak PNA staining and GFP expression are still visible in some papillae (solid arrow), but not others (open arrowhead). (D) PNA staining (pink) and C11.360>Unc-76::GFP expression (green) in C. robusta upon overexpression of both Islet and Sp6/7/8, showing expansion of IC fate in a single large papilla (arrow). PNA staining is similarly expanded over the entire IC cluster, confirming that ICs produce the adhesive glue. Foxc>lacZ expression (β- galactosidase immunostaining) shown in blue in both C and D. (E) Scoring of larvae represented in panels C and D, averaged across duplicates. Weak PNA staining is observed upon partial suppression of IC fate, but strong PNA staining is seen upon expansion of supernumerary ICs, confirming that this cell type is one of the major contributors of PNA-positive adhesive glue. Total larvae (duplicate 1) or β-galactosidase+ larvae (duplicate 2) were scored. p < 0.0001 in both duplicates, as determined by chi-square test. See S4 Data for sample size, statistical test details, and for the data underlying the graphs. C. intestinalis raised to 20–22 hpf at 18 ˚C (~st. 28), C. robusta raised to 20 hpf at 20 ˚C (~st. 29). See Supplemental Movies for full confocal stacks and S6 Fig for single-channel images. IC, inner collocyte; OC, outer collocyte; PNA, peanut agglutinin. https://doi.org/10.1371/journal.pbio.3002555.g005 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 12 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae the very apical tip, on top of which the strongest PNA staining is seen extracellularly (Fig 5A and S6 Fig and S1 Movie), while the majority of PNA-stained intracellular granules were not within the ICs at this stage (Fig 5A). Consistently, the OCs were the main cells showing PNA- stained granules located within the papillae (Fig 5B and S6 Fig and S2 Movie). This distribu- tion of PNA staining corresponds to the distribution of granules previously identified by high- pressure freezing electron microscopy [9], in which collocytes located in the central core of the papilla contain granules mostly at their apical end. Indeed, in cross-sections, granules were most abundant inside the papillary body, likely in cells identified here as OCs. To further investigate the contributions of both ICs and OCs to glue secretion, we per- formed PNA staining on larvae in distinct perturbation conditions. Namely, we electroporated larvae with Foxc>Sp6/7/8, which was shown above to suppress IC specification, or with Foxc>Islet and Foxc>Sp6/7/8 combined, which was shown to convert most of the papilla terri- tory into ICs. Although Foxc>Sp6/7/8 eliminated most IC reporter expression, PNA staining was still weakly present (Fig 5C and 5E), likely due to continued presence of OCs. In contrast, Foxc>Islet + Foxc>Sp6/7/8 resulted in a single enlarged papilla with supernumerary ICs, and the entire papilla was often covered by strong PNA staining (Fig 5D and 5E). Taken together, these results suggest that both ICs and OCs contribute to the production of adhesive material, but that the ICs (or their progenitors) are likely the more important contributors. Specification of PNs and OCs from cells that have down-regulated Foxg With the specification of ACCs/ICs/OCs explained in large part due to overlapping expression domains of Islet and Sp6/7/8, the precise developmental origins of the PNs and OCs still remained elusive. While it has become clear that the Islet+ cells at the core of each papilla give rise to ACCs and ICs, Papilla-specific CRISPR knockout of Islet did not abolish PNs or OCs, as shown above (Fig 3). This suggested they do not arise from these core Islet+ cells, consistent with their more lateral positions as shown previously by TEM [9]. Furthermore, recently pub- lished in situ hybridization data showing presumptive Pou4-expressing PN precursors sur- rounding Islet-expressing cells at late tailbud stage [23]. Indeed, co-electroporation of Islet reporter and PN- or OC-specific reporter plasmids clearly showed PNs and OCs immediately adjacent to, but distinct from, Islet+ cells (Fig 2D and 2I, and S2D Fig). Might PNs and OCs be arising from the cells in which Foxg is down-regulated (likely via repression by Sp6/7/8) and that do not go on to express Islet (Fig 6A) [21–23]? To test this, we used the MEK (MAPK kinase) inhibitor U0126 to expand Islet expression as previously done (Fig 6A) [22]. While treatment with 10 μm U0126 at 7.5 hpf (between stages 16 and 17, or late neurula and early tailbud) predictably expanded Islet reporter expression, it also eliminated expression of the PN reporter C4.78>GFP, as well as that of the OC reporter L141.36>GFP (Fig 6B and 6C). These results suggest that Foxg+ papilla cells that maintain Foxg expression go on to express Islet and give rise to ACCs and ICs, while the cells that activate Sp6/7/8 and down-regulate Foxg in response to MAPK signaling go on to give rise to OCs and PNs instead. PNs are specified by common peripheral neuron regulators Previous papilla-specific TALEN knockout of the neuronal transcription factor-encoding gene Pou4 successfully eliminated PNs and the larva’s tail resorption response to mechanical stimuli [16]. Pou4 has been previously implicated in a Myt1-dependent regulatory cascade that speci- fies the caudal epidermal neurons (CENs) of the tail, from neurogenic midline cells expressing the proneural bHLH transcription factor Ascl.a (KH.L9.13, sometimes called Ascl2 or Ascl.b previously) [23,48–51]. To precisely visualize the neurogenic cells of the papillae, we per- formed double (two-color) mRNA in situ hybridization for Ascl.a and Foxg at the mid-tailbud PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 13 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 6. Specification of PNs and OCs from Islet-negative cells by MAPK and Notch pathways. (A) Diagram showing effect of MAPK inhibition with the pharmacological MEK inhibitor U0126, based on findings from Wagner and colleagues and Liu and Satou. Inhibition of FGF/MAPK results in expansion of Foxg and Islet from 3 discrete foci to a large “U-shaped” swath, transforming 3 papillae into a single, enlarged papilla (similar results reported with BMP inhibition by Roure and colleagues). (B) The 10 μm U0126 treatment at 7.5 hpf/20 ˚C (st. 16) results in loss of PNs (assayed by C4.78>Unc-76::GFP at 17 hpf/20 ˚C, ~st. 27, green) upon expansion of Islet+ cells (pink nuclei), relative to DMSO alone. (C) The same treatment results in loss of OCs (L141.36>Unc-76::GFP at either 17 or 20 hpf/20 ˚C, ~st. 27–29 green) upon expansion of Islet+ cells (pink nuclei). Both U0126 experiments were performed in duplicate, with 100 larvae per condition per duplicate. (D) Two-color, whole-mount mRNA in situ hybridization for Foxg (green in merged image) and Ascl.a (KH. L9.13, pink). (E) Larva electroporated with Ascl.a>Unc-76::GFP labeling several papilla cells including PNs. (F) Myt1>mNeonGreen labeling PNs and other neurons including CENs. (G) Two-color in situ hybridization of Foxg (green) and Pou4 (pink), the latter labeling adjacent PNs and possibly RTENs. (H) Lineage-specific CRISPR/Cas9-mediated mutagenesis of Pou4 results in loss of PN reporter expression (C4.78>Unc-76::GFP, green). Asterisk denoted background/leaky expression in mesenchyme. Larvae at 17 hpf/20 ˚C (~st. 27). (I) Scoring of Foxc>H2B::mCherry+ larvae represented in panel H and in Sp6/7/8 CRISPR mutagenesis condition showing Sp6/7/8 does not appear to play a major role in PN reporter expression like Pou4. Experiments repeated in duplicate with 100 larvae in each. (J) Inhibition of Delta/Notch signaling using Foxc>SUH-DBM results in reduced expression of OC reporter (L141.36>Unc-76::GFP, green) at 21 hpf/20 ˚C (~st. 29). Experiment was repeated in duplicate with 100 larvae in each. (K) Notch inhibition also results in concomitant expansion of supernumerary PNs at 17 hpf/20 ˚C (~st. 27, labeled by C4.78>Unc-76::GFP, green) relative to Foxc>lacZ control. Experiment was repeated in duplicate, with 42 to 50 larvae in each. (L) Summary diagram and model of effects of Delta/Notch inhibition on PN/OC fate choice in Islet-negative (but formerly Foxg+) papilla progenitor cells. All Islet reporters are the Islet intron 1 + bpFOG>H2B::mCherry. All error bars indicate upper and lower limits. p < 0.0001, * p = 0.0003 in both duplicates as determined by Fisher’s exact test, ns = not statistically significant in at least 1 duplicate. See S4 Data for the data underlying the graphs and for statistical test details. CEN, caudal epidermal neuron; OC, outer collocyte; PN, papilla neuron; RTEN, rostral trunk epidermal neuron. https://doi.org/10.1371/journal.pbio.3002555.g006 stage. Indeed, Ascl.a expression was seen broadly in the papilla territory surrounding the 3 Foxg+ cell clusters (Fig 6D). This was confirmed by an Ascl.a fluorescent protein reporter plas- mid that labeled a broad set of papilla territory cells, including PNs and their axons (Fig 6E). Furthermore, a previously published Myt1 reporter [52] was also found to be expressed in the PNs (Fig 6F). Double in situ of Pou4 and Foxg revealed Pou4+ cells surrounding each Foxg+ PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 14 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae cluster, corroborating a recent report [23] (Fig 6G). It was not immediately clear which Pou4 + cells were PN precursors and which were nearby rostral trunk epidermal neuron (RTEN) precursors. Based on our images and those of the most recent study [23], we propose that there are initially 2 Pou4+ cells per papilla, later dividing to give rise to the 4 PNs per papilla as previously described [9]. This would mirror the development of the epidermal neurons of the tail, in which neurons are born side-by-side as pairs after a final cell division by a committed mother cell [49]. Papilla-specific CRISPR knockout of Pou4 with a combination of 2 newly val- idated sgRNAs (S3D Fig) recapitulated the loss of PN differentiation by the previously pub- lished TALEN knockout [16], as assayed by C4.78 and TGFB reporter expression (Fig 6H and 6I, and S4 Fig). In contrast, Pou4 knockout had no effect on the specification of ACCs or OCs, suggesting Pou4 function is specific for PN fate in the papillae (S4 Fig). Taken together, these results suggest that PNs are specified from interspersed neurogenic progenitors that are carved out by MAPK signaling. Interestingly, CRISPR knockout of Sp6/7/8 did not substantially affect PN specification (Fig 6I). This suggests that even though Sp6/7/8 down-regulates Foxg in these cells [21], it does not appear to be required for their neurogenic potential. Notch signaling regulates the fate choice between PNs and OCs Because both OCs and PNs appeared to arise from Foxg-down-regulating, Islet-negative cells, we sought to test whether an additional regulatory step is required for the fate choice between these 2 cell types. In the neurogenic midline territory of the tail epidermis, lateral inhibition by Delta/Notch signaling regulates the final number and spacing of CENs [48,49,53]. Delta/Notch limits the expression of Myt1, which in turn activates Pou4 expression. In the tail epidermis, the major ligand involved is the putative Delta like non-canonical Notch ligand homolog (encoded by gene KH.L50.6), which is also expressed in alternating pattern in the papillae (S7A Fig). We therefore decided to test whether a similar mechanism in controlling the num- ber of PNs and OCs surrounding each papilla. To test the requirement of Delta/Notch, we overexpressed a DNA-binding mutant of the Notch co-factor RBPJ/SUH (SUH-DBM) [54]. Indeed, electroporation with Foxc>SUH-DBM resulted in loss of OC reporter expression (Fig 6J), and concomitant expansion of PN reporter expression (Fig 6K). We conclude that Delta/ Notch signaling regulates PN versus OC fate choice in neurogenic progenitor cells surround- ing each presumptive papilla, with Notch delimiting the specification of supernumerary neu- rons, thus allowing OCs to form (Fig 6L). In contrast, SUH-DBM had minimal effect on IC/ ACC fate choice (S7B Fig), suggesting Delta/Notch might only regulate neurogenesis, and not cell fate choice in general, in the papilla territory. This common origin of PNs and OCs is also supported by the recent finding that the latter appear to have basal bodies like the PNs, but without the accompanying sensory cilia [9]. Inter- estingly, papilla-specific knockout of Foxg resulted in moderate loss of PN reporter expression (TGFB>GFP), and very little effect on the OC reporter (S4A Fig). This suggests differing requirement for Foxg in different cell type-specific branches of the papilla regulatory network, despite all these cell types arising from cells that initially express Foxg. Regulation of papilla morphogenesis by Islet It was previously shown that Foxg or Islet overexpression induces the formation of a single enlarged “megapapilla,” in which all cells are substantially elongated relative to the rest of the epidermis [21,22]. We have shown above that this appears to be driven by expansion of ACCs and/or ICs, which are atypically elongated in the apical-basal direction and form apical protru- sions and microvilli. Islet is sufficient for apical-basal elongation of epidermal cells [22], and morpholino-knockdown of Foxg (which is upstream of Islet) also impairs proper papilla PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 15 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae morphogenesis [21]. We asked if Islet is required for papilla morphogenesis, using papilla-spe- cific CRISPR knockout of Islet. Knocking out Islet in the papilla territory impaired the forma- tion of the typically “pointy-shaped” papillae, resulting instead in blunt cells with flat, broader apical surfaces and reduced cell length along the apical-basal axis (Fig 7A and 7B, S8C and S8D Fig). This result suggested that transcriptional targets downstream of Islet might be regu- lating the distinct cell shape of ACCs/ICs. To identify potential candidate effectors of morphogenesis downstream of Islet, we used bulk RNAseq to measure differential gene expression between different Islet perturbation con- ditions. We compared “negative control” embryos to (1) embryos in which Islet was overex- pressed in the whole territory using the Foxc promoter (Foxc>Islet); and (2) embryos in which Islet was knocked out specifically in the papilla lineage by CRISPR/Cas9. For this, we designed an additional sgRNA targeting the first exon of Islet, to be used in combination with the already published sgRNA to generate larger deletions. This new sgRNA vector, which we named U6>Islet.1, resulted in a mutagenesis efficacy of 20% (S3B Fig). Whole embryos from each condition were collected at 12 hpf (Islet conditions) at 20 ˚C in biological triplicate. RNA was extracted from pooled embryos in each sample, and RNAseq libraries were prepared from poly(A)-selected RNAs and sequenced by Illumina NovaSeq. This bulk RNAseq approach revealed that Islet overexpression results in the up-regulation of several ACC markers from previous scRNAseq analysis (S2 Data) [20]. With Islet overexpres- sion, this included ACC markers previously validated by mRNA in situ hybridization or reporter gene expression, such as CryBG (KH.S605.3) and Atp2a (KH.L116.40). Many ACC markers were conspicuously absent, but this may be due to the relatively early time point (12 hpf, late tailbud stage), well before hatching and ACC differentiation. This was a deliberate choice, as we were focused on papilla morphogenesis, which begins around this stage [22]. One resulting candidate Islet target revealed by RNAseq was Astl-related (KH.C9.850), and its expression in the Islet+ cells of the papillae was confirmed by in situ hybridization (S8A Fig). Indeed, Islet knockout by CRISPR eliminated Astl-related reporter expression, supporting our approach to identifying new targets of Islet (S8B Fig). Furthermore, the top up-regulated gene by Islet overexpression (and 17th most down-regulated by Islet CRISPR) was KY21.Chr10.318, which encodes a Fibrillin-related (Fbn) protein. This gene was previously shown to be specifi- cally expressed in the central Islet+ cells by in situ hybridization [23], further validating our approach to identifying putative Islet target genes. One particularly interesting ACC-specific candidate that was among the genes most highly up-regulated by Islet overexpression was Villin (KH.C9.512), an ortholog of the Villin family of genes encoding effectors of actin regulators [55]. The apical extensions of the ACCs are highly enriched for actin filaments and microtubules [9], suggesting that cytoskeletal modulation may be important for the extended length of these cells relative to surrounding cells. We con- firmed the expression of Villin in the papillae by in situ hybridization and reporter plasmids (Fig 7C and 7D). In the Islet CRISPR condition, Villin was the top down-regulated gene by Islet CRISPR knockout as well. Villin reporter expression was reduced in intensity but not completely lost upon knockout of Islet by CRISPR (Fig 7E and 7F), yet was dramatically up- regulated by Islet overexpression (Fig 7G and 7H). This suggests partially redundant activation of Villin by another factor, likely at earlier developmental stages (e.g., by Foxc or Foxg), and that Islet might be required for its sustained expression specifically in the central cells of the papilla throughout morphogenesis. This is consistent with the weak but broad expression of Villin>GFP in the entire papilla territory (Fig 7D), and the fact that papilla territory cells are already more elongated than epidermal cells in other parts of the embryo even at earlier stages [22]. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 16 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 7. Islet is also required for papilla morphogenesis. (A) Papilla shape is shortened and blunt at the apical end upon tissue-specific CRISPR/ Cas9-mediated mutagenesis of Islet. Embryos were electroporated with Islet intron 1 + -473/-9>Unc-76::GFP and Foxc>Cas9. Islet CRISPR was performed using U6>Islet.2 sgRNA plasmid and the negative control used U6>Control. Larvae were imaged at 20 hpf/20 ˚C (~st. 28). Right: Scoring of percentage of GFP+ larvae classified as having normal “protruding” or blunt papillae, as represented to the left. Experiment was performed and scored in duplicate, using 2 different GFP fusions: Unc-76::GFP and DcxΔC::GFP [88]. Replicate 1: n = 100 for either condition; replicate 2: n = 55 for either condition ** p < 0.0001 in both duplicates as determined by Fisher’s exact test. (B) Quantification of papilla cell (Islet intron 1 +-473/-9>Unc-76::GFP +) lengths along apical-basal axis in negative control and Islet CRISPR larvae at 18 hpf/20 ˚C (~st. 28). Both Islet.1 and Islet.2 sgRNAs used in combination. Statistical significance tested by unpaired t test (two-tailed). See S8 Fig for duplicate experiment. (C) In situ mRNA hybridization of Villin, showing expression in Foxg+/Islet+ central papilla cells at 10 hpf/20 ˚C (st. 21, left) and at larval stage (~st. 27, right). (D) Villin -1978/-1>Unc- 76::GFP showing expression in the papilla territory of electroporated larvae (~st. 28), strongest in the central cells. (E) Villin -1978/-1>Unc-76::GFP in st. 28 larvae is down-regulated by tissue-specific CRISPR/Cas9 mutagenesis of Islet (Foxc>Cas9 + U6>Islet.1 + U6>Islet.2, see text for details). (F) Quantification of effect of Islet CRISPR (as in panel E) on Villin -1978/-1>Unc 76::GFP/Foxc>H2B::mCherry mean fluorescence intensity ratios in ROIs defined by the mCherry+ nuclei (see Methods for details). Significance determined by Mann–Whitney test (two-tailed). (G) Villin reporter is up- regulated in st. 28 larvae by overexpressing Islet (Foxc>Islet, see text for details). (H) GFP/mCherry ratio quantification done in identical manner as in F, but comparing Islet overexpression (as in panel G) and control lacZ larvae. (I) Quantification of ACC lengths measured in negative control and papilla-specific Villin CRISPR larvae at 17 hpf/20 ˚C (~st. 27). Significance tested by unpaired t test (two-tailed). Although no statistically significant difference between control and CRISPR larvae was observed in this replicate, average ACC length was significantly shorter in the CRISPR condition in an additional replicate (S8 Fig). (J) mRNA in situ hybridization for Tuba3, showing enrichment in the central cells of the papillae in st. 27 larvae. (K) Tuba3>Unc-76::GFP reporter plasmid is broadly expressed in the papillae of st. 27 larvae but stronger in central cells. (L) Papilla-specific CRISPR knockout of Tuba3 does not result in decrease of average ACC apical-basal cell length compared to negative control CRISPR using U6>Control sgRNA instead. Significance tested by unpaired t test (two-tailed). ns = not significant. All large bars indicate medians and smaller bars indicate interquartile ranges. See S3 and S4 Data for the data underlying the graphs and for statistical test details. ACC, axial columnar cell; ROI, region of interest; sgRNA, single-chain guide RNA. https://doi.org/10.1371/journal.pbio.3002555.g007 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 17 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae To test whether Villin is required for proper morphogenesis of Islet+ cells in the papilla, we performed tissue-specific CRISPR knockout using a combination of 3 validated sgRNAs span- ning most of the coding sequence (S3E Fig). Because the functionally important “headpiece” domain is encoded by the last exon, we combined an sgRNA targeting this exon with 2 sgRNAs targeting more upstream exons. In one batch of Villin CRISPR larvae, ACCs were not significantly shorter on average along the apical-basal axis than in control larvae (Fig 7I). How- ever, average ACC length was significantly shorter in CRISPR larvae than control larvae in a duplicate experiment (S8E Fig). This contrast between replicates was found to be entirely due to variability between batches of control larvae, not the Villin CRISPR larvae (S8F Fig). Because the ACCs have been shown to dynamically extend or contract in length, possibly in response to external stimuli [56,57], we suspect that these differences in average length in dif- ferent batches of control animals are due to as of yet unidentified environmental conditions. In addition to actin regulators, we searched our list of putative Islet targets for microtubule components and regulators, since microtubule bundles were reported in the apical protrusions of the ACCs in Distaplia occidentalis [58]. We identified a gene encoding a divergent Tubulin alpha monomer (Tuba3, KH.C3.736) as one such potential target. Enrichment of Tuba3 expression in the central papillae was confirmed by in situ hybridization (Fig 7J) and a Tuba3 reporter plasmid (Fig 7K). However, papilla-specific CRISPR knockout of Tuba3 did not result in significantly shorter ACCs either (Fig 7L). Taken together, these results suggest that Islet is required for proper papilla morphogenesis, and that this may be due to its role in activating the expression of numerous effector genes. However, knocking out individual candidate effec- tor genes like Villin or Tuba3 has not yet revealed a key role for any one of these putative downstream targets. An investigation into the cell and molecular basis of larval settlement and metamorphosis With our different CRISPR knockouts affecting different cell types of the papillae, we asked how these different perturbations might affect larval metamorphosis. Only the involvement of the PNs in triggering metamorphosis has been demonstrated [16,17], but it is not yet known how the regulatory networks and cell types of the papillae affect different processes during metamorphosis. We performed papilla-specific CRISPR as above using the Foxc>Cas9 vector, targeting the 4 different transcription factors we have shown to be involved in patterning the cell types of the papillae: Pou4, Islet, Foxg, and Sp6/7/8. We assayed tail retraction and body rotation at the last stage of metamorphosis [28] (Fig 8A and 8B), as these are 2 processes that can be uncoupled in certain genetic perturbations or naturally occurring mutants [59]. Knockout of Pou4 recapitulated recent published results on this transcription factor [16]. Namely, both tail retraction and body rotation were blocked in the vast majority of individuals. This suggests that proper specification and/or differentiation of PNs by Pou4 is crucial for the ability of the larva to trigger the onset of metamorphosis. In contrast, Islet knockout did not affect tail retraction, but body rotation appeared somewhat impaired. This suggested that ACCs/ICs are not required for tail retraction, but might play a role in regulating body rotation downstream of it. Eliminating ACCs using Islet>Sp6/7/8 had no effect on either tail retraction or body rotation (Fig 8C), confirming that ACCs are not required for metamorphosis, but that perhaps certain Islet targets might specifically regulate body rotation. Unsurprisingly, Foxg knockout modestly impaired both tail retraction and body rotation (Fig 8B and 8D, and S9 Fig), but also resulted in a noticeable fraction (approximately 19% on average) of “tailed juve- niles” in which body rotation begins even in the absence of tail retraction. This unusual effect was seen even when repeating the experiment independently a third time, revealing consistent PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 18 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 8. Genetic perturbations of metamorphosis. (A) Ciona robusta juvenile undergoing metamorphosis, showing the retracted tail and rotated anterior-posterior body axis (dashed lines). PNs in the former papilla (now substrate attachment stolon, or holdfast) labeled by TGFB>Unc-76::GFP (green). Animal counterstained with DAPI (blue). (B) Scoring of Foxc>H2B::mCherry+ individuals showing tail retraction and/or body rotation at 48 hpf/20 ˚C in various papilla territory-specific (using Foxc>Cas9) CRISPR-based gene knockouts. Experiments were performed and scored in duplicate and percentages averaged, except for Foxg CRISPR for which a third replicate was performed (see S9 Fig). Number scored individuals in each replicate indicated underneath. “Tailed juveniles” have undergone body rotation but not tail retraction, whereas normally body rotation follows tail retraction. The sgRNA plasmids used for each condition were as follows- Control: U6>Control; Pou4: U6>Pou4.3.21 + U6>Pou4.4.106; Islet: U6>Islet.2; Foxg: U6>Foxg.1.116 + U6>Foxg.5.419; Sp6/7/8: U6>Sp6/7/8.4.29 + U6>Sp6/7/8.8.117. (C) Plot showing lack of any discernable metamorphosis defect after eliminating ACCs using Islet intron 1 + bpFOG>Sp6/7/8 (images not shown). Only Islet intron 1 + bpFOG>H2B::mCherry+ individuals were scored. Experiment was performed and scored in duplicate and averaged (n = 100 each duplicate). ACC specification was scored using the CryBG>Unc-76:: GFP reporter. (D) Example of “tailed juveniles” at 47 hpf/20 ˚C compared to a larva in which no tail retraction or body rotation has occurred, elicited by tissue-specific Foxg CRISPR (Foxc>Cas9 + U6>Foxg.1.116 + U6>Foxg5.419). See S9 Fig for scoring. All error bars denote upper and lower limits. p < 0.0001, * p < 0.0015 in both duplicates, ns = not significant in at least 1 duplicate, as determined by chi-square test comparing to the control conditions. See S4 Data for the data underlying the graphs and for statistical test details. ACC, axial columnar cell; PN, papilla neuron; sgRNA, single- chain guide RNA. https://doi.org/10.1371/journal.pbio.3002555.g008 uncoupling of these 2 processes upon Foxg knockout (S9 Fig). Finally, Sp6/7/8 CRISPR did not substantially alter either tail retraction or body rotation. Taken together, these results paint a more complex picture of regulation of metamorphosis by the papillae. Our findings suggest that different cell types of the papillae might play distinct roles in the regulation of metamor- phosis, perhaps interacting with one another to regulate different steps, or that certain tran- scription factors might be required for the expression of key rate-limiting components of these different processes. Further work will be required to disentangle these different cellular and genetic factors, which we hope will be aided by our cell type-specific reporters and CRISPR reagents. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 19 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Discussion Sensory systems are crucial for interactions between organisms and their environment. The concentration of sensory functions in the head is thought to have played a central role in verte- brate evolution, leading to a more active behavior emerging from early filter-feeding chordate ancestors [60–62]. The peripheral components of the sensory systems in vertebrates arise from 2 physically close but distinct ectodermal cell populations, the cranial sensory placodes and the neural crest [63]. Cranial sensory placodes are characterized by their common ontogenetic ori- gin from a crescent-shaped region surrounding the anterior neural plate. Our understanding of the evolutionary origins of structures long presented as vertebrate novelties has benefited from an increasing number of comparative studies with tunicates. Several discrete populations of peripheral sensory cells originating from distinct ectodermal regions in tunicates have respectively been linked to neural crest and cranial placodes, among them the sensory adhesive papillae [9,64–67]. Our results have confirmed the existence of molecularly distinct cell types in the Ciona papillae and the developmental pathways that specify them (summarized in Fig 9). Using CRISPR/Cas9-mediated mutagenesis, we have shown that different transcription factors are required for their specification, differentiation, and morphogenesis. Namely, ACCs and ICs are specified from Foxg+/Islet+ cells at the center of each of the 3 papillae, while OCs and PNs are specified from interleaved Islet-negative cells that nonetheless derive from initially Foxg + cells. While Sp6/7/8 specifies IC versus ACC fate among Islet+ cells, Delta/Notch signaling suppresses PN fate and promotes OC fate among Islet-negative cells. While there appear to be 2 molecularly distinct collocyte subtypes (OCs and ICs), both contain granules that are stained by PNA, and therefore both are likely to be involved in glue production. Where they differ might be in the timing of glue production and/or secretion, as they showed distinct subcellular localization of PNA+ granules, and PNA production was previously shown to start very early [9]. Our results also demonstrate a clear distinction between CryBG+ ACCs and Pou4+ PNs. Previously, these cells types have been confused and only recently distinguished by TEM and different molecular markers [9]. Here we show that, while both arise from Foxc+/Foxg+ cells, ACCs are not specified by Pou4, and PNs are not specified by Islet. However, because Pou4 can activate Foxg expression in a proposed feedback loop [68], overexpression of Pou4 might result in ectopic activation of ACC markers via ectopic Foxg and Islet activation. There are still unanswered questions that we hope future work will address: (1) How do the 3 “spots” of Foxg+/Islet+ cells form in an invariant manner? Ephrin-Eph signal- ing is often responsible for suppression of FGF/MAPK signaling in alternating cells in Ciona embryos, via asymmetric inheritance/activation of p120 RasGAP [69,70]. This is also true in the earlier patterning of the papilla territory, where EphrinA.d suppresses FGF/ MAPK to promote Foxg activation [21]. Curiously, later expression of EphrinA.d in the lineage appears to be stronger in medial Foxg+ cells than in lateral cells [21]. This distribu- tion would suffice to result in the alternating ON/OFF pattern of MAPK activation at the tailbud stage that results in the 3 foci of Foxg/Islet expression. Thus, it may be informative to test the ongoing functions of Ephrin-Eph signaling in this lineage throughout develop- ment. Interestingly, we did not observe substantial loss of PNs in Sp6/7/8 CRISPR larvae, even though Sp6/7/8 down-regulates Foxg in between the 3 “spots” [21] and is necessary for OC reporter expression (Fig 4F). It is possible that FGF/MAPK suppresses Islet in parallel, and that loss of Sp6/7/8 is not sufficient to expand Islet expression and ACC/IC progenitor fate (at the expense of PN/OC progenitors) in the same manner that the MEK inhibitor U0126 does. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 20 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Fig 9. Summary diagram. (A) Updated diagram of the development of the anterior descendants of Row 6 in the neural plate to show the proposed patterns of MAPK and Delta/Notch signaling that set up the 3 Foxg+ clusters and interleaved Foxg-negative neurogenic cells. (B) Diagram proposing the contributions of Foxg+ and Foxg-negative cells to later patterns of transcription factors that specify the different cell types found in each papilla, which is in turn is repeated 3 times, thanks to the process shown in panel A. (C) Papilla development shown as cell lineages, with dashed lines indicating uncertain cell divisions and lineage history. MAPK regulates binary fate choices, promoting Foxg expression in the papillae proper at first but later suppressing it (and Islet). Lastly, Delta-Notch signaling promotes OC fate and limits PN specification through lateral inhibition. Cell type numbers based on [9]. (D) Provisional gene regulatory network diagram of the transcription factors involved in specification and differentiation of the different papilla cell types. Arrowheads indicate activating gene expression or promoting cell fate, while blunt ends indicate repression of gene expression of cell fate. Solid lines indicate regulatory links (direct or indirect) that are supported by the current data and literature. Dashed lines indicate regulatory links that have not been tested, or need to be investigated in more detail. A-P, anterior-posterior; D-V, dorsal-ventral; OC, outer collocyte; OSP, oral siphon primordium; PN, papilla neuron. https://doi.org/10.1371/journal.pbio.3002555.g009 (2) How are PNs specified adjacent to the Islet+ cells? Since Delta/Notch signaling is involved in PN versus OC fate, we propose that there is something that biases Notch signaling to be activated preferentially in those cells not touching the Islet+ cells. This could be due to cell- autonomous activation of Notch signaling in the Islet+ cells, which in turn would allow for suppression of Notch in adjacent cells fated to become PNs. A recent study showed Pou4 expansion with concomitant “U”-shaped expansion of Islet in larvae with a single, enlarged papilla when inhibiting BMP signaling [23]. However, our expansion of Islet with treatment of U0126 (based on experiments from Wagner and colleagues) suggests the opposite, the elimination of Pou4+ PNs. Why the discrepancy? One possibility is that inhibiting BMP results in specification of supernumerary RTEN-like neurons from adjacent epidermis, not PNs. While Pou4 is expressed in all epidermal neurons including PNs and RTENs, our pre- ferred PN marker KH.C4.78 is not expressed in RTENs. However, we did not observe either loss or expansion of C4.78>GFP in larvae with enlarged, single papillae resulting from treatment with the BMP inhibitor DMH1 (S10 Fig). This suggests that inhibition of FGF/ MEK and BMP have slightly different effects on patterning and neurogenesis in the papillae. Further work will be needed to resolve these and other intriguing nuances. (3) What activates the expression of Sp6/7/8 in Islet+ cells, ultimately promoting IC specification? We do not yet know the exact mitotic history of the ACCs/ICs. How do the initially four Islet+ cells divide, and which daughter cells give rise to ACCs versus ICs? Are ACCs/ICs PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 21 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae specified in an invariant manner, or is there some variability? Finally, what allows the “creeping” activation of Sp6/7/8 in the outer ring of cells that likely become the ICs? Is this due to additional asymmetric FGF/MAPK activation downstream of Ephrin-Eph? Or could this be due to some other signaling pathway? Is there an inductive signal from adjacent cells, for instance PNs or common PN/OC progenitors? (4) How do the different papilla cell types regulate metamorphosis? We noticed some uncoupl- ing of tail resorption and body rotation upon targeting different transcription factors for deletion in the papillae (Fig 8). This was most apparent in the Foxg knockout, in which a substantial portion of individuals displayed the “tailed juvenile” phenotype in which body rotation proceeds even in the absence of tail resorption. From the Pou4 knockout, it is clear that PNs are upstream of both tail resorption and body rotation, but the partial uncoupling seen with the other manipulations were particularly intriguing. This uncoupling has been reported before in Cellulose synthase mutants, which results in similar tailed juveniles [71]. Additionally, perturbation of Gonadotropin-releasing hormone (GnRH) or the prohor- mone convertase enzyme (PC2) necessary for its processing similarly blocks tail resorption but not body rotation and further adult organ growth [72]. Thus, it is possible that while Pou4 disrupts PN specification altogether, Foxg might be more specifically required for GnRH or other neuropeptide expression/processing in the PNs. Supporting this idea, the Foxg CRISPR did not disrupt PN specification (as assayed by TGFB>GFP reporter expres- sion) as robustly as did Pou4 CRISPR. Alternatively, this uncoupling may also be a result of the different roles of Foxg in regulating different papilla cell types, which may be unequally and variably affected due to CRISPR knockout mosaicism in F0. Finally, the appearance of juveniles with resorbed tails but no further body rotation in the Islet CRISPR condition sug- gests a crucial role for the “core” cells of the papilla (ACCs/ICs) in metamorphosis down- stream of PNs. However, body rotation was not affected by eliminating either ICs (Sp6/7/8 CRISPR, Fig 8B) or ACCs (Islet>Sp6/7/8, Fig 8C), suggesting Islet is required for the expression of a “body rotation” factor independently of IC/ACC specification. Clearly, more work will be required to understand the contributions of different cell types, and potentially different molecular pathways in the same cell type, towards either activation or suppression of specific body plan rearrangement processes in tunicate larval metamorphosis. (5) Are the tunicate larval papillae homologous to vertebrate cement glands and/or olfactory pla- codes? The papillae have often been compared to the cement glands of fish and amphibian larvae, which are transient adhesive organs secreting sticky mucus [73]. Even though they are innervated by trigeminal fibers, the secreting cells from the cement gland differentiate from a surface ectoderm region anterior to the oral ectoderm and the panplacodal domain [74]. Therefore, they are usually not considered placodal derivatives. Despite their variabil- ity in size, number, and location, head adhesive organs are proposed to be homologous across vertebrate species based on their shared expression of Pitx1/2 and BMP4 genes, innervation by trigeminal fibers, and inhibiting mechanism of swimming behavior [73,74]. While recent papers have revealed an important role for BMP signaling in pattering the tunicate papilla territory [23,24], suggesting a potential evolutionary connection, additional work on the molecular basis of the papillary glue in tunicates will be required to answer questions of homology between these adhesive organs. Our identification of molecular sig- natures for both collocyte subtypes in the papillae of Ciona provides a starting point for future investigations and may allow for broader evolutionary comparisons between chor- dates and other bilaterians. If one considers the vertebrate cement gland and the trigeminal PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 22 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae neurons that innervate it as a functional unit but no homology to the tunicate papillae can be established, they might represent an interesting case of evolutionary convergence. Besides adhesion, a key role of the papilla is to act as an organ for bimodal mechano- and chemo-sensation regulating larval settlement and metamorphosis [75]. It has been proposed that the papilla territory may be a homolog of the olfactory placode, based on the expression of several regulators investigated in this study including Sp6/7/8, Islet, Foxc, and Foxg [18,21]. The roles of Foxg in specifying the papilla territory may also be linked evolutionarily to the functions of Foxg1 in vertebrate olfactory development [76,77]. Intriguingly, the specification of sensory PNs and probable adhesive-secreting OCs from shared Ascl.a+ progenitors (as revealed by Notch pathway inhibition experiments, Fig 6) suggests a close regulatory link between sensory/adhesive functions in Ciona papillae. In vertebrates, Ascl1 not only regulates sensory cell specification in developing olfactory epithelium and taste buds [78–80] but is also required for intestinal secretory cells [81,82]. Additionally, Ascl3 is expressed in vertebrate sali- vary gland duct progenitors, which are also highly enriched for orthologs of other tunicate papilla regulators like Foxc, Sp6/7/8, and Islet [83,84]. Thus, while the papillae of Ciona might represent a tunicate-specific evolutionary novelty, overlaying unrelated networks for sensory neurons and adhesive-secreting cells together in a single embryonic territory, it is also possible that they rely on a shared sensory/exocrine program that might have deeper evolutionary ori- gins instead [85]. Supporting information S1 Fig. Finding papilla cell type-specific markers in single-cell RNAseq data. (A) Cell clus- ters based from reanalysis and re-clustering of whole-larva single-cell RNA sequencing (scRNAseq) data from Cao and colleagues (see S1 Data). Dashed red box indicated clusters 3 and 33, which appeared to correspond to several papilla cell types. (B) Cells from clusters 3 and 33 from plot A set aside and re-clustered. (C) Differential expression plots showing exam- ples of candidate papilla cell type marker genes mapped onto clusters in B. (D) Fluorescent, whole-mount in situ mRNA hybridization (green) for certain genes plotted in C, labeling dif- ferent cells in the papillae of Ciona robusta (intestinalis Type A) hatched larvae. (E) Protein domain prediction diagrams for select cell type-specific marker proteins generated by SMART [89]. Unless specifically named, genes are indicated by KyotoHoya (KH) ID numbers (e.g., KH.L96.43). All larvae were fixed at 18 h post-fertilization (hpf), 20 ˚C, except for C11.360 and C2.1013 (18.5 hpf). Blue counterstain is DAPI. (TIF) S2 Fig. Additional marker genes and reporter plasmids expressed in papillae. (A) TGFB>Unc-76::GFP reporter (green) is not co-expressed in the same cells as the Islet intron 1 + -473/-9>mCherry reporter (pink) at 20.5 hpf (~st. 29). (B) Cross-talk between CryBG>Unc- 76::GFP and TGFB>Unc-76::mCherry reporter plasmids at 16 hpf (~st. 26), showing aberrant co-expression in ACCs and/or PNs only when co-electroporated. (C) Mutually exclusive expression of CryBG>lacZ in ACCs (cyan), Gnrh1>Unc-76::GFP in PNs (yellow), and C11.360>Unc-76::mCherry in ICs (magenta), with DAPI counterstained in gray. This larva is the same as in main Fig 2G, with an additional channel and different false coloring. (D) Images from Fig 2 with mCherry and GFP channels displayed separately. (E) C. intestinalis (Type B) larva electroporated with C. robusta C11.360>Unc-76::GFP reporter plasmid, showing specific but weak expression. (F) C. intestinalis (Type B) larva electroporated with C. robusta L96.43>Unc-76::GFP reporter plasmid, also showing weak expression. Papillae in panels E and F outlined by dashed lines. (G) Reporter plasmid containing the first intronic region of Emx PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 23 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae drives expression in ICs at 20 hpf (~st. 29), likely corresponding to the “ring” of late Emx expression in Islet+ cells reported in Wagner and colleagues and distinct from earlier Emx expression in the papilla lineage as described in Liu and Satou. (H) C14.116>Unc-76::mCherry reporter expressed in central cells (ACCs+ICs, pink) and basal cells around the 3 papillae at 20.5 hpf (~st. 29). (I) Immunostaining for the Flag epitope tag fused to the Islet-rescue protein used for Islet>Islet experiments in Fig 4. Flag immunostaining in green and Foxc>H2B:: mCherry in pink in merged image. Larvae fixed at 19 hpf (~st. 28). DAPI in gray. ACCs, axial columnar cells; PN, papilla neuron. (TIF) S3 Fig. Validation of sgRNAs for CRISPR/Cas9-mediated mutagenesis. Gene loci diagrams for the 4 transcription factor-encoding genes investigated in this study: Sp6/7/8, Foxg, Islet, and Pou4. Plots underneath each gene show validation by Illumina sequencing (“Next-gener- ation sequencing” or NGS) of amplicons, performed as “Amplicon-EZ” service by Azenta. Mutagenesis efficacies are calculated by this service, and histograms of mapped reads show specificity of indels elicited by each sgRNA. Negative control amplicons are amplified from samples that were electroporated with no sgRNA, U6>Control sgRNA, or sgRNAs targeting unrelated amplicon regions. Note different y axis scales for each plot. Asterisks in Villin exon 5 and Tuba3 amplicon plots indicate naturally occurring indels. Precise calculation of muta- genesis efficacy for Villin.5.105 and Tuba3.3.24 sgRNAs was not given due to these natural indels. (TIF) S4 Fig. Effect of various CRISPR knockouts on specification of ACCs, PNs, and OCs. (A) Scoring of effect of papilla-specific CRISPR knockout of Foxg or Pou4 on specification of ACCs and PNs. Embryos were electroporated with Foxc>H2B::mCherry, Foxc>Cas9, CryBG>Unc-76::GFP (ACC reporter), TGFB>Unc-76::GFP (PN reporter), or L141.36>Unc- 76::GFP (OC reporter), and gene-specific sgRNA combinations (see below for specific combi- nations). All were performed in duplicate and scores averaged, but some replicates and condi- tions are represented in Figs 3 and 4 also. Total embryos ranged between 76 and 100 per condition per replicate. Specific sgRNAs used: Foxg: U6>Foxg.1.116 + U6>Foxg.5.419; Pou4: U6>Pou4.3.21 + U6>Pou4.4.106; Sp6/7/8: U6>Sp6/7/8.4.29 + U6>Sp6/7/8.8.117; Islet: U6>Islet.2; Control: U6>Control. (B) Foxg, Pou4, and Sp6/7/8 sgRNAs were also tested alone (as opposed to pairs in combination) using reporter assays as in Figs 3 and 4. Those sgRNAs used further are highlighted in blue font. Additional sgRNAs abandoned due to low efficacy indicated in black font. For all plots, only larvae showing Foxc>H2B::mCherry expression in the papillae were scored. See S4 Data for the data underlying the graphs and for statistical test details. (TIF) S5 Fig. CRISPR rescues and possible expansion of ICs in Islet>Sp6/7/8. (A) CryBG>GFP expression in Islet (left) and Foxg (right) CRISPR larvae is rescued by co-electroporation with Islet intron 1 + bpFOG>Flag::Islet-rescue or Foxg>Foxg-rescue constructs, respectively, thanks to silent point mutations disrupting the sgRNA target binding sites. (B) Expression of C11.360>GFP is rescued in Sp6/7/8 CRISPR larvae upon co-electroporation with an Islet intron 1 + bpFOG>Sp6/7/8-rescue construct. (C) Example of expanded IC reporter (C11.360>Unc-76::GFP, green) in larvae (20 hpf/20 ˚C, ~st. 29) electroporated with Islet intron 1 + -473/-9>Sp6/7/8, as determined by perfect overlap with the Islet intron 1 + -473/-9>H2B:: mCherry reporter (pink). See text for more details. See S1 File for exact sequences and detailed electroporation recipes. See S4 Data for the data underlying the graphs and for statistical test PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 24 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae details. (TIF) S6 Fig. Single-channel images of PNA-stained larvae shown in Fig 5. (TIF) S7 Fig. Delta/Notch components and lack of SUH-DBM effect on ACC/IC fate choice. (A) Fluorescent whole-mount in situ mRNA hybridization for Delta like non-canonical Notch ligand (KH.L50.6) in a stage 23 embryo, marking epidermal sensory neurons including papilla neurons (arrows) and caudal epidermal neurons (CENs). (B) No significant difference in expression of either CryBG>mCherry or C11.360>GFP in larvae at 19 hpf at 20 ˚C (~st. 28) electroporated with the Delta-Notch pathway-inhibiting Foxc>SUH-DBM. Right: representa- tive panels showing expression of both reporters in control and SUH-DBM-expressing larvae. See S4 Data for the data underlying the graphs and statistical test details. (TIF) S8 Fig. Investigating the regulation of papilla morphogenesis by Islet and its putative tran- scriptional targets. (A) In situ mRNA hybridization (ISH) showing expression of Astl-related (green) specifically in the Islet+ cells of the papillae (labeled by Islet intron 1 + -473/- 9>mCherry, pink nuclei). (B) Tissue-specific CRISPR/Cas-mediated mutagenesis of Islet results in loss of Astl-related>Unc-76::GFP reporter expression in ACCs/ICs (green). Foxc>Cas9 was used to restrict CRISPR/Cas9 to the papilla territory (labeled by Foxc>H2B:: mCherry, pink nuclei). Asterisk denotes residual reporter expression in cells outside the papilla territory. Right: Scoring of larvae represented in panel B, following criteria used for Fig 3. ** p < 0.0001 using chi-square test. (C) Second duplicate of Islet CRISPR experiment in Fig 7B. Statistical significance was determined by unpaired t test (two-tailed). (D) Representative images of control and Islet CRISPR larvae used for measurements in Fig 7B, with example of apical-basal cell length measurements. (E) Both duplicates of Villin CRISPR experiments side by side. Statistical significance was calculated using Mann–Whitney test (two-tailed) for repli- cate 1 and unpaired t test (two-tailed) for replicate 2. ns = not significant. (F) Comparison of ACC lengths measured in control larvae from the 2 duplicate Villin CRISPR experiments, showing statistically significant difference in average ACC lengths between different batches of larvae, calculated using Mann–Whitney test (two-tailed). See S3 and S4 Data for the data underlying the graphs and for statistical test details. (TIF) S9 Fig. Third replicate of Foxg CRISPR effects on metamorphosis. Scoring of Foxc>H2B:: mCherry+ individuals as represented in Fig 7D, in third replicate of data in Fig 7B. See S1 File for detailed plasmid electroporation recipes. ** p < 0.0001 calculated by Fisher’s exact test. See S4 Data for the data underlying the graphs and for statistical test details. (TIF) S10 Fig. Pharmacological inhibition of BMP signaling and papilla neuron specification. Larva co-electroporated with Islet intron 1 + -473/-9>mCherry (pink) and PN reporter C4.78>Unc-76::GFP (green) showing lack of substantial loss or expansion of PNs in larvae treated with the BMP pathway inhibitor DMH1, in spite of expanded Islet reporter expression and a single, enlarged papilla. Left: scoring of PN reporter expression in Islet>mCherry + DMSO (negative control) and DMH1-treated larvae. All larvae raised to 19 hpf at 20 ˚C (~st. 28). See S4 Data for the data underlying the graphs and for statistical test details. (TIF) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 25 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae S1 Data. Differential gene expression table of re-analyzed whole larva single-cell RNAseq data from Cao and colleagues (new clusters 3 and 33 subclustered into 11 subclusters). (XLSX) S2 Data. Differential gene expression table of bulk RNAseq analysis of Islet overexpression or Islet CRISPR embryos vs. control embryos. (XLSX) S3 Data. Papilla length measurements in different CRISPR and control larvae. (XLSX) S4 Data. Summary of statistical tests of proportions (i.e., scoring). (XLSX) S1 Movie. Confocal stack represented by Fig 5A. (AVI) S2 Movie. Confocal stack represented by Fig 5B. (AVI) S1 File. All relevant DNA and protein sequences used in this study (supplemental sequence file). (DOCX) Acknowledgments We thank members of the labs at Georgia Tech and Innsbruck for critical feedback and sup- port. We thank Susanne Gibboney, Tanner Shearer, Alex Gurgis, Lindsey Cohen, Akhil Kulk- arni, and Eduardo Gigante for technical assistance. Author Contributions Conceptualization: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Ute Rothba¨cher, Alberto Stolfi. Data curation: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katarzyna M. Piekarz, Ute Rothba¨cher, Alberto Stolfi. Formal analysis: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katarzyna M. Pie- karz, Ute Rothba¨cher, Alberto Stolfi. Funding acquisition: Christopher J. Johnson, Ute Rothba¨cher, Alberto Stolfi. Investigation: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Ute Rothba¨cher, Alberto Stolfi. Methodology: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katarzyna M. Piekarz, Shweta Biliya, Ute Rothba¨cher, Alberto Stolfi. Project administration: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Ute Rothba¨- cher, Alberto Stolfi. Software: Katarzyna M. Piekarz. Supervision: Florian Razy-Krajka, Ute Rothba¨cher, Alberto Stolfi. Validation: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katarzyna M. Piekarz, Ute Rothba¨cher, Alberto Stolfi. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 26 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae Visualization: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katarzyna M. Piekarz, Ute Rothba¨cher, Alberto Stolfi. Writing – original draft: Katarzyna M. Piekarz, Shweta Biliya, Ute Rothba¨cher, Alberto Stolfi. Writing – review & editing: Christopher J. Johnson, Florian Razy-Krajka, Fan Zeng, Katar- zyna M. Piekarz, Shweta Biliya, Ute Rothba¨cher, Alberto Stolfi. References 1. Fodor ACA, Liu J, Turner L, Swalla BJ. Transitional chordates and vertebrate origins: Tunicates. Curr Top Dev Biol. 2021; 139:325–374. https://doi.org/10.1016/bs.ctdb.2020.10.001 PMID: 33602487 2. Lemaire P. Evolutionary crossroads in developmental biology: the tunicates. Development. 2011; 138 (11):2143–2152. https://doi.org/10.1242/dev.048975 PMID: 21558365 3. Satoh N. Developmental genomics of ascidians. John Wiley & Sons; 2013. 4. DeBiasse MB, Colgan WN, Harris L, Davidson B, Ryan JF. Inferring tunicate relationships and the evo- lution of the tunicate Hox cluster with the genome of Corella inflata. Genome Biol Evol. 2020; 12 (6):948–964. https://doi.org/10.1093/gbe/evaa060 PMID: 32211845 5. Kocot KM, Tassia MG, Halanych KM, Swalla BJ. Phylogenomics offers resolution of major tunicate rela- tionships. Mol Phylogenet Evol. 2018; 121:166–73. https://doi.org/10.1016/j.ympev.2018.01.005 PMID: 29330139 6. Delsuc F, Philippe H, Tsagkogeorga G, Simion P, Tilak M-K, Turon X, et al. A phylogenomic framework and timescale for comparative studies of tunicates. BMC Biol. 2018; 16(1):39. https://doi.org/10.1186/ s12915-018-0499-2 PMID: 29653534 7. Karaiskou A, Swalla BJ, Sasakura Y, Chambon JP. Metamorphosis in solitary ascidians. Genesis. 2015; 53(1):34–47. https://doi.org/10.1002/dvg.22824 PMID: 25250532 8. Caicci F, Zaniolo G, Burighel P, Degasperi V, Gasparini F, Manni L. Differentiation of papillae and rostral sensory neurons in the larva of the ascidian Botryllus schlosseri (Tunicata). J Comp Neurol. 2010; 518 (4):547–566. https://doi.org/10.1002/cne.22222 PMID: 20020541 9. 10. 11. Zeng F, Wunderer J, Salvenmoser W, Hess MW, Ladurner P, Rothba¨ cher U. Papillae revisited and the nature of the adhesive secreting collocytes. Dev Biol. 2019; 448(2):183–198. https://doi.org/10.1016/j. ydbio.2018.11.012 PMID: 30471266 Torrence SA, Cloney RA. Ascidian larval nervous system: primary sensory neurons in adhesive papil- lae. Zoomorphology. 1983; 102(2):111–123. Turon X. Morphology of the adhesive papillae of some ascidian larvae. Cah Biol Mar. 1991; 32:295– 309. 12. Pennati R, Groppelli S, De Bernardi F, Mastrototaro F, Zega G. Immunohistochemical analysis of adhe- sive papillae of Clavelina lepadiformis (Mu¨ller, 1776) and Clavelina phlegraea (Salfi, 1929) (Tunicata, Ascidiacea). Eur J Histochem. 2009; 53(1). 13. Dolcemascolo G, Pennati R, De Bernardi F, Damiani F, Gianguzza M. Ultrastructural comparative anal- ysis on the adhesive papillae of the swimming larvae of three ascidian species. Invertebr Surviv J. 2009; 6(1 (Suppl)):S77–S86. 14. Pennati R, Zega G, Groppelli S, De Bernardi F. Immunohistochemical analysis of the adhesive papillae of Botrylloides leachi (Chordata, Tunicata, Ascidiacea): Implications for their sensory function. Ital J Zool. 2007; 74(4):325–329. 15. Zeng F, Wunderer J, Salvenmoser W, Ederth T, Rothba¨ cher U. Identifying adhesive components in a model tunicate. Philos Trans R Soc B. 2019; 374(1784):20190197. 16. Sakamoto A, Hozumi A, Shiraishi A, Satake H, Horie T, Sasakura Y. The TRP channel PKD2 is involved in sensing the mechanical stimulus of adhesion for initiating metamorphosis in the chordate Ciona. Dev Growth Differ. 2022; 64(7):395–408. https://doi.org/10.1111/dgd.12801 PMID: 36053743 17. Wakai MK, Nakamura MJ, Sawai S, Hotta K, Oka K. Two-Round Ca2+ transient in papillae by mechani- cal stimulation induces metamorphosis in the ascidian Ciona intestinalis type A. Proc R Soc B. 1945; 2021(288):20203207. 18. Poncelet G, Shimeld SM. The evolutionary origins of the vertebrate olfactory system. Open Biol. 2020; 10(12):200330. https://doi.org/10.1098/rsob.200330 PMID: 33352063 19. Poncelet GJF, Parolini L, Shimeld S. A microfluidic device for controlled exposure of transgenic Ciona intestinalis larvae to chemical stimuli demonstrates they can respond to carbon dioxide. bioRxiv. 2022:2022–2008. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 27 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae 20. Sharma S, Wang W, Stolfi A. Single-cell transcriptome profiling of the Ciona larval brain. Dev Biol. 2019; 448(2):226–236. https://doi.org/10.1016/j.ydbio.2018.09.023 PMID: 30392840 21. Liu B, Satou Y. Foxg specifies sensory neurons in the anterior neural plate border of the ascidian embryo. Nat Commun. 2019; 10(1):1–10. 22. Wagner E, Stolfi A, Choi YG, Levine M. Islet is a key determinant of ascidian palp morphogenesis. Development. 2014; 141(15):3084–3092. https://doi.org/10.1242/dev.110684 PMID: 24993943 23. Roure A, Chowdhury R, Darras S. Regulation of anterior neurectoderm specification and differentiation by BMP signaling in ascidians. Development. 2023; 150(10). https://doi.org/10.1242/dev.201575 PMID: 37213081 24. Liu B, Ren X, Satou Y. BMP signaling is required to form the anterior neural plate border in ascidian embryos. Dev Genes Evol. 2023;1–11. 25. Christiaen L, Wagner E, Shi W, Levine M. Isolation of sea squirt (Ciona) gametes, fertilization, dechorio- nation, and development. Cold Spring Harb Protoc. 2009; 2009(12):pdb. prot5344. https://doi.org/10. 1101/pdb.prot5344 PMID: 20150091 26. Christiaen L, Wagner E, Shi W, Levine M. Electroporation of transgenic DNAs in the sea squirt Ciona. Cold Spring Harbor Protocols. 2009; 2009(12):pdb. prot5345. https://doi.org/10.1101/pdb.prot5345 PMID: 20150092 27. Kari W, Zeng F, Zitzelsberger L, Will J, Rothbaecher U. Embryo microinjection and electroporation in the chordate Ciona intestinalis. J Vis Exp. 2016; 116:e54313. https://doi.org/10.3791/54313 PMID: 27805579 28. Hotta K, Dauga D, Manni L. The ontology of the anatomy and development of the solitary ascidian Ciona: the swimming larva and its metamorphosis. Sci Rep. 2020; 10(1):17916. https://doi.org/10.1038/ s41598-020-73544-9 PMID: 33087765 29. Stolfi A, Levine M. Neuronal subtype specification in the spinal cord of a protovertebrate. Development. 2011; 138(5):995–1004. https://doi.org/10.1242/dev.061507 PMID: 21303852 30. Ikuta T, Saiga H. Dynamic change in the expression of developmental genes in the ascidian central ner- vous system: revisit to the tripartite model and the origin of the midbrain-hindbrain boundary region. Dev Biol. 2007: 312. https://doi.org/10.1016/j.ydbio.2007.10.005 PMID: 17996862 31. Stolfi A, Wagner E, Taliaferro JM, Chou S, Levine M. Neural tube patterning by Ephrin, FGF and Notch signaling relays. Development. 2011; 138(24):5429–5439. https://doi.org/10.1242/dev.072108 PMID: 22110057 32. Beh J, Shi W, Levine M, Davidson B, Christiaen L. FoxF is essential for FGF-induced migration of heart progenitor cells in the ascidian Ciona intestinalis. Development. 2007; 134(18):3297–3305. 33. Stolfi A, Gandhi S, Salek F, Christiaen L. Tissue-specific genome editing in Ciona embryos by CRISPR/Cas9. Development. 2014; 141(21):4115–4120. https://doi.org/10.1242/dev.114488 PMID: 25336740 34. Song M, Yuan X, Racioppi C, Leslie M, Stutt N, Aleksandrova A, et al. GATA4/5/6 family transcription factors are conserved determinants of cardiac versus pharyngeal mesoderm fate. Sci Adv. 2022; 8(10): eabg0834. https://doi.org/10.1126/sciadv.abg0834 PMID: 35275720 35. Haeussler M, Scho¨ nig K, Eckert H, Eschstruth A, Mianne´ J, Renaud J-B, et al. Evaluation of off-target and on-target scoring algorithms and integration into the guide RNA selection tool CRISPOR. Genome Biol. 2016; 17(1):1–12. https://doi.org/10.1186/s13059-016-1012-2 PMID: 27380939 36. Satija R, Farrell JA, Gennert D, Schier AF, Regev A. Spatial reconstruction of single-cell gene expres- sion data. Nat Biotechnol. 2015; 33(5):495. https://doi.org/10.1038/nbt.3192 PMID: 25867923 37. Afgan E, Nekrutenko A, Gru¨ ning BA, Blankenberg D, Goecks J, Schatz MC, et al. The Galaxy platform for accessible, reproducible and collaborative biomedical analyses: 2022 update. Nucleic Acids Res. 2022. https://doi.org/10.1093/nar/gkac247 PMID: 35446428 38. Satou Y, Tokuoka M, Oda-Ishii I, Tokuhiro S, Ishida T, Liu B, et al. A manually curated gene model set for an ascidian, Ciona robusta (Ciona intestinalis type A). Zoolog Sci. 2022; 39(3). https://doi.org/10. 2108/zs210102 PMID: 35699928 39. Cao C, Lemaire LA, Wang W, Yoon PH, Choi YA, Parsons LR, et al. Comprehensive single-cell tran- scriptome lineages of a proto-vertebrate. Nature. 2019; 571(7765):349–354. https://doi.org/10.1038/ s41586-019-1385-y PMID: 31292549 40. Shimeld SM, Purkiss AG, Dirks RPH, Bateman OA, Slingsby C, Lubsen NH. Urochordate βγ-crystallin and the evolutionary origin of the vertebrate eye lens. Curr Biol. 2005; 15(18):1684–1689. 41. Razy-Krajka F, Lam K, Wang W, Stolfi A, Joly M, Bonneau R, et al. Collier/OLF/EBF-dependent transcriptional dynamics control pharyngeal muscle specification from primed cardiopharyngeal progenitors. Dev Cell. 2014; 29(3):263–276. https://doi.org/10.1016/j.devcel.2014.04.001 PMID: 24794633 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 28 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae 42. Kusakabe TG, Sakai T, Aoyama M, Kitajima Y, Miyamoto Y, Takigawa T, et al. A conserved non-repro- ductive GnRH system in chordates. 2012. 43. Imai JH, Meinertzhagen IA. Neurons of the ascidian larval nervous system in Ciona intestinalis: II. Peripheral nervous system. J Comp Neurol. 2007; 501(3):335–352. https://doi.org/10.1002/cne.21247 PMID: 17245709 44. Matsunobu S, Sasakura Y. Time course for tail regression during metamorphosis of the ascidian Ciona intestinalis. Dev Biol. 2015; 405(1):71–81. https://doi.org/10.1016/j.ydbio.2015.06.016 PMID: 26102482 45. Pennati R, Ficetola GF, Brunetti R, Caicci F, Gasparini F, Griggio F, et al. Morphological Differences between Larvae of the Ciona intestinalis Species Complex: Hints for a Valid Taxonomic Definition of Distinct Species. PLoS ONE. 2015; 10(5):e0122879. https://doi.org/10.1371/journal.pone.0122879 PMID: 25955391 46. Satou Y, Sato A, Yasuo H, Mihirogi Y, Bishop J, Fujie M, et al. Chromosomal inversion polymorphisms in two sympatric ascidian lineages. Genome Biol Evol. 2021; 13(6):evab068. https://doi.org/10.1093/ gbe/evab068 PMID: 33822040 47. Gandhi S, Haeussler M, Razy-Krajka F, Christiaen L, Stolfi A. Evaluation and rational design of guide RNAs for efficient CRISPR/Cas9-mediated mutagenesis in Ciona. Dev Biol. 2017; 425(1):8–20. https:// doi.org/10.1016/j.ydbio.2017.03.003 PMID: 28341547 48. Tang WJ, Chen JS, Zeller RW. Transcriptional regulation of the peripheral nervous system in Ciona intestinalis. Dev Biol. 2013; 378(2):183–193. https://doi.org/10.1016/j.ydbio.2013.03.016 PMID: 23545329 49. Pasini A, Amiel A, Rothba¨ cher U, Roure A, Lemaire P, Darras S. Formation of the ascidian epidermal sensory neurons: insights into the origin of the chordate peripheral nervous system. PLoS Biol. 2006; 4 (7):e225. https://doi.org/10.1371/journal.pbio.0040225 PMID: 16787106 50. Roure A, Darras S. Msxb is a core component of the genetic circuitry specifying the dorsal and ventral neurogenic midlines in the ascidian embryo. Dev Biol. 2016; 409(1):277–287. https://doi.org/10.1016/j. ydbio.2015.11.009 PMID: 26592100 51. Waki K, Imai KS, Satou Y. Genetic pathways for differentiation of the peripheral nervous system in ascidians. Nat Commun. 2015; 6:8719. https://doi.org/10.1038/ncomms9719 PMID: 26515371 52. Tolkin T, Christiaen L. Rewiring of an ancestral Tbx1/10-Ebf-Mrf network for pharyngeal muscle specifi- cation in distinct embryonic lineages. Development. 2016; 143:3852–3862. https://doi.org/10.1242/dev. 136267 PMID: 27802138 53. Chen JS, Pedro MS, Zeller RW. miR-124 function during Ciona intestinalis neuronal development includes extensive interaction with the Notch signaling pathway. Development. 2011; 138(22):4943– 4953. https://doi.org/10.1242/dev.068049 PMID: 22028027 54. Hudson C, Yasuo H. A signalling relay involving Nodal and Delta ligands acts during secondary noto- chord induction in Ciona embryos. Development. 2006; 133(15):2855–2864. 55. Khurana S, George SP. Regulation of cell structure and function by actin-binding proteins: villin’s per- spective. FEBS Lett. 2008; 582(14):2128–2139. https://doi.org/10.1016/j.febslet.2008.02.040 PMID: 18307996 56. Johnson CJ, Razy-Krajka F, Stolfi A. Expression of smooth muscle-like effectors and core cardiomyo- cyte regulators in the contractile papillae of Ciona. EvoDevo. 2020; 11(1):1–18. https://doi.org/10.1186/ s13227-020-00162-x PMID: 32774829 57. Cloney R. Larval adhesive organs and metamorphosis in ascidians. II. The mechanism of eversion of the papillae of Distaplia occidentalis. Cell Tissue Res. 1979; 200(3):453–473. https://doi.org/10.1007/ BF00234856 PMID: 487411 58. Cloney RA. Larval adhesive organs and metamorphosis in ascidians: I. Fine structure of the everting papillae of Distaplia occidentalis. Cell Tissue Res. 1977; 183:423–444. 59. Nakayama-Ishimura A, Chambon Jp, Horie T, Satoh N, Sasakura Y. Delineating metamorphic path- ways in the ascidian Ciona intestinalis. Dev Biol. 2009; 326(2):357–367. https://doi.org/10.1016/j.ydbio. 2008.11.026 PMID: 19100250 60. Gans C, Northcutt RG. Neural crest and the origin of vertebrates: a new head. Science. 1983; 220 (4594):268–273. https://doi.org/10.1126/science.220.4594.268 PMID: 17732898 61. Diogo R, Kelly RG, Christiaen L, Levine M, Ziermann JM, Molnar JL, et al. A new heart for a new head in vertebrate cardiopharyngeal evolution. Nature. 2015; 520(7548):466. https://doi.org/10.1038/ nature14435 PMID: 25903628 62. Patthey C, Schlosser G, Shimeld SM. The evolutionary history of vertebrate cranial placodes–I: cell type evolution. Dev Biol. 2014; 389(1):82–97. https://doi.org/10.1016/j.ydbio.2014.01.017 PMID: 24495912 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 29 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae 63. Martik ML, Bronner ME. Riding the crest to get a head: neural crest evolution in vertebrates. Nat Rev Neurosci. 2021; 22(10):616–626. https://doi.org/10.1038/s41583-021-00503-2 PMID: 34471282 64. Abitua PB, Gainous TB, Kaczmarczyk AN, Winchell CJ, Hudson C, Kamata K, et al. The pre-vertebrate origins of neurogenic placodes. Nature. 2015. https://doi.org/10.1038/nature14657 PMID: 26258298 65. Abitua PB, Wagner E, Navarrete IA, Levine M. Identification of a rudimentary neural crest in a non-ver- tebrate chordate. Nature. 2012; 492(7427):104. https://doi.org/10.1038/nature11589 PMID: 23135395 66. Papadogiannis V, Pennati A, Parker HJ, Rothba¨cher U, Patthey C, Bronner ME, et al. Hmx gene con- servation identifies the origin of vertebrate cranial ganglia. Nature. 2022; 605(7911):701–705. https:// doi.org/10.1038/s41586-022-04742-w PMID: 35585239 67. Horie R, Hazbun A, Chen K, Cao C, Levine M, Horie T. Shared evolutionary origin of vertebrate neural crest and cranial placodes. Nature. 2018; 560(7717):228. https://doi.org/10.1038/s41586-018-0385-7 PMID: 30069052 68. Chacha PP, Horie R, Kusakabe TG, Sasakura Y, Singh M, Horie T, et al. Neuronal identities derived by misexpression of the POU IV sensory determinant in a protovertebrate. Proc Natl Acad Sci U S A. 2022; 119(4):e2118817119. https://doi.org/10.1073/pnas.2118817119 PMID: 35042818 69. Haupaix N, Abitua PB, Sirour C, Yasuo H, Levine M, Hudson C. Ephrin-mediated restriction of ERK1/2 activity delimits the number of pigment cells in the Ciona CNS. Dev Biol. 2014; 394(1):170–80. https:// doi.org/10.1016/j.ydbio.2014.07.010 PMID: 25062608 70. Haupaix N, Stolfi A, Sirour C, Picco V, Levine M, Christiaen L, et al. p120RasGAP mediates ephrin/ Eph-dependent attenuation of FGF/ERK signals during cell fate specification in ascidian embryos. Development. 2013; 140(21):4347–4352. https://doi.org/10.1242/dev.098756 PMID: 24067356 71. Sasakura Y, Nakashima K, Awazu S, Matsuoka T, Nakayama A, Azuma J, et al. Transposon-mediated insertional mutagenesis revealed the functions of animal cellulose synthase in the ascidian Ciona intes- tinalis. Proc Natl Acad Sci U S A. 2005; 102(42):15134. 72. Hozumi A, Matsunobu S, Mita K, Treen N, Sugihara T, Horie T, et al. GABA-Induced GnRH Release Triggers Chordate Metamorphosis. Curr Biol. 2020. https://doi.org/10.1016/j.cub.2020.02.003 PMID: 32220316 73. Re´taux S, Pottin K. A question of homology for chordate adhesive organs. Commun Integr Biol. 2011; 4 (1):75–77. https://doi.org/10.4161/cib.4.1.13926 PMID: 21509185 74. Pottin K, Hyacinthe C, Re´ taux S. Conservation, development, and function of a cement gland-like struc- ture in the fish Astyanax mexicanus. Proc Natl Acad Sci U S A. 2010; 107(40):17256–17261. https://doi. org/10.1073/pnas.1005035107 PMID: 20855623 75. Hoyer J, Kolar K, Athira A, van den Burgh M, Dondorp D, Liang Z, et al. Polymodal sensory perception of mechanical and chemical cues drives robust settlement and metamorphosis of a marine pre-verte- brate zooplanktonic larva. bioRxiv. 2023:2023–2007. 76. Duggan CD, DeMaria S, Baudhuin A, Stafford D, Ngai J. Foxg1 is required for development of the verte- brate olfactory system. J Neurosci. 2008; 28(20):5229–5239. https://doi.org/10.1523/JNEUROSCI. 1134-08.2008 PMID: 18480279 77. Kawauchi S, Santos R, Kim J, Hollenbeck PLW, Murray RC, Calof AL. The Role of Foxg1 in the Devel- opment of Neural Stem Cells of the Olfactory Epithelium. Ann N Y Acad Sci. 2009; 1170(1):21–7. https://doi.org/10.1111/j.1749-6632.2009.04372.x PMID: 19686101 78. Cau E, Casarosa S, Guillemot F. Mash1 and Ngn1 control distinct steps of determination and differenti- ation in the olfactory sensory neuron lineage. 2002. 79. Guillemot F, Lo L-C, Johnson JE, Auerbach A, Anderson DJ, Joyner AL. Mammalian achaete-scute homolog 1 is required for the early development of olfactory and autonomic neurons. Cell. 1993; 75 (3):463–476. https://doi.org/10.1016/0092-8674(93)90381-y PMID: 8221886 80. Seta Y, Oda M, Kataoka S, Toyono T, Toyoshima K. Mash1 is required for the differentiation of AADC- positive type III cells in mouse taste buds. Dev Dyn. 2011; 240(4):775–784. https://doi.org/10.1002/ dvdy.22576 PMID: 21322090 81. Flasse LC, Stern DG, Pirson JL, Manfroid I, Peers B, Voz ML. The bHLH transcription factor Ascl1a is essential for the specification of the intestinal secretory cells and mediates Notch signaling in the zebra- fish intestine. Dev Biol. 2013; 376(2):187–197. https://doi.org/10.1016/j.ydbio.2013.01.011 PMID: 23352790 82. Roach G, Wallace RH, Cameron A, Ozel RE, Hongay CF, Baral R, et al. Loss of ascl1a prevents secre- tory cell differentiation within the zebrafish intestinal epithelium resulting in a loss of distal intestinal motility. Dev Biol. 2013; 376(2):171–186. https://doi.org/10.1016/j.ydbio.2013.01.013 PMID: 23353550 83. Michael DG, Pranzatelli TJF, Warner BM, Yin H, Chiorini JA. Integrated epigenetic mapping of human and mouse salivary gene regulation. J Dent Res. 2019; 98(2):209–217. https://doi.org/10.1177/ 0022034518806518 PMID: 30392435 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 30 / 31 PLOS BIOLOGYSpecification of distinct cell types in tunicate papillae 84. Arany S, Catala´ n MA, Roztocil E, Ovitt CE. Ascl3 knockout and cell ablation models reveal complexity of salivary gland maintenance and regeneration. Dev Biol. 2011; 353(2):186–193. https://doi.org/10. 1016/j.ydbio.2011.02.025 PMID: 21377457 85. Arendt D, Benito-Gutierrez E, Brunet T, Marlow H. Gastric pouches and the mucociliary sole: setting the stage for nervous system evolution. Philos Trans R Soc Lond B Biol Sci. 2015; 370(1684):20150286. https://doi.org/10.1098/rstb.2015.0286 PMID: 26554050 86. Wagner E, Levine M. FGF signaling establishes the anterior border of the Ciona neural tube. Develop- ment. 2012; 139(13):2351–2359. https://doi.org/10.1242/dev.078485 PMID: 22627287 87. Nicol D, Meinertzhagen IA. Development of the central nervous system of the larva of the ascidian, Ciona intestinalis L: II. Neural plate morphogenesis and cell lineages during neurulation. Dev Biol. 1988; 130(2):737–766. 88. Kim K, Gibboney S, Razy-Krajka F, Lowe E, Wang W, Stolfi A. Regulation of neurogenesis by FGF sig- naling and Neurogenin in the invertebrate chordate Ciona. Front Cell Dev Biol. 2020; 8:477. https://doi. org/10.3389/fcell.2020.00477 PMID: 32656209 89. Letunic I, Khedkar S, Bork P. SMART: recent updates, new developments and status in 2020. Nucleic Acids Res. 2021; 49(D1):D458–D460. https://doi.org/10.1093/nar/gkaa937 PMID: 33104802 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002555 March 13, 2024 31 / 31 PLOS BIOLOGY
10.1371_journal.pclm.0000325	RESEARCH ARTICLE Climate change impact on Spodoptera frugiperda (Lepidoptera: Noctuidae) life cycle in Mozambique Telmo Cosme A. SumilaID 1, Simone E. T. Ferraz1, Angelica Durigon2 1 Physics Department, Federal University of Santa Maria, Santa Maria, RS, Brazil, 2 Department of Plant Science, Federal University of Santa Maria, Santa Maria, RS, Brazil [email protected], [email protected] Abstract Although different seasonal cues are important for fall armyworm (FAW, Spodoptera frugi- perda J.E. Smith) survival, it is known that the life cycle of this insect is strongly dependent on air temperature, means that its development rate proceeds faster when the weather is warm. To develops the insect needs to accumulate an amount of thermal units, as known as Growing Degree-Days (GDD). However, with the climate change driven by global warming, the GDD pattern must be changed and therefore, the life cycle of this new bug in Mozam- bique may be different from that observed in its native region. In the present study it is esti- mated the possible changes of FAW life cycle by applying the GDD method over Mozambique, under two representative scenarios of climate changes, RCP4.5 and RCP8.5 for 2070–2099 relative to present climate (1971–2000). For this purpose, dynamical down- scaling process through the regional model RegCM4, nested to global model HadGEM2 were used. The outputs of air temperature dataset from the simulations were used to com- pute the accumulated GDD and hence the FAW number of generations (NG) during the summer-season over the study domain. The findings indicate that there is a bipolar pattern of GDD accumulation, being negative over most of central and restricted areas in southern region, and positive in northern region, altitude-modified climate areas over central region, and over southernmost areas for both representative climate scenarios, relative to present climate. Meanwhile, there is an increase (decrease) in NG in the areas of higher (lower) increase in air temperature for both future scenarios relative to present climate. Introduction The life cycle and dynamical population of insects are strongly affected by air temperature, which influence on all ectothermic organisms, the so-called cold-blooded animals. Diapause is an expected physiological mechanism of the invertebrate small animals, but some of them, such as the cold-blooded insects have lack of this temperature adjustment process. The devel- opment rate of this insect is controlled by its biological clock directly correlated with air tem- perature. Thus, climate factors particularly air temperature, can impact the physiological a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Sumila TCA, Ferraz SET, Durigon A (2024) Climate change impact on Spodoptera frugiperda (Lepidoptera: Noctuidae) life cycle in Mozambique. PLOS Clim 3(1): e0000325. https:// doi.org/10.1371/journal.pclm.0000325 Editor: Djanaguiraman Maduraimuthu, Tamil Nadu Agricultural University, INDIA Received: August 17, 2023 Accepted: December 11, 2023 Published: January 18, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pclm.0000325 Copyright: © 2024 Sumila et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The CRU interpolated dataset can be found at “Dataset Record: CRU TS4.05: Climatic Research Unit (CRU) Time-Series (TS) version 4.05 of high-resolution gridded data of month-by-month variation in climate (January PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 1 / 25 PLOS CLIMATE1901–December 2020) (ceda.ac.uk)”; the weather stations dataset can be found at “INAM—National Institute of Meteorology”. The future climate scenarios can be found at IPCC home page. Funding: This research was funded by National Council for Scientific and Technological Development (TCAS) who, at the same time, received a salary from an institution of the government of Mozambique and, material resources for this work were provided by the Federal University of Santa Maria (AD and SETF). Angelica Durigon and Simone E.T. Ferraz are scientific productivity Fellows and professors at the Federal University of Santa Maria, mentors of this research. The funders had no role in the study design, data collection and analysis, decision to publish or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Climate change impact on Spodoptera frugiperda life cycle in Mozambique metabolism of these species [1–5]. Fall Armyworm, Spodoptera frugiperda, (J.E. Smith, 1797) (Lepidoptera: Noctuidae) hereafter FAW, an endemic and important agricultural pest native from tropical regions of North America, is a cold-blooded migratory insect. Means that, for itself the pest cannot survive under adverse climate conditions, particularly extreme cold or hot weather [3, 6, 7]. Since early 2016, the first record of FAW in West Africa (Nigeria, Ghana, Benin, Sao Tome´ and Principe and Togo) was confirmed and rapidly spread throughout the tropical and sub- tropical regions of sub-Saharan Africa [8–12]. Because of its ability to travel several hundred kilometers in a single night, in the following year, 2017, 28 sub-Saharan African countries including Mozambique, had confirmed the outbreaks of FAW [7, 11, 13]. According to some studies [e.g. 9, 14], the two strains of the species have been found in some African countries and, although the few records of the pest in Mozambique [e.g. 15, 16], it has been reported that the pest keeps spreading throughout the country. FAW may cause significant damage to agricultural crops with emphasis on maize, one of the two main FAW host crops. The agriculture activities in Mozambique are predominantly driven by smallholder farmers, with around 80% of Mozambican households practice agricul- ture as their main livelihood, and maize is one of the major staple crops of smallholder farmers in the country. The climate impact-related of FAW on maize (Zea mays L.) is difficult to proj- ect due to the complex interactions among insects and this host crop. Meanwhile, the presence of FAW associated with ongoing climate change increase the risk of agricultural activity and thus jeopardizing food security in the country, which since 2017 has recorded some estimated losses on this important host crop [17]. It is important to stress that, unlike the original moths, the African FAW infestation may represent a novel interstrain hybrid population and [18] mentioned uncertain behavioral characteristics of this population. Climate change is already affecting every inhabited region around the world, and the impacts are more severe in developing countries, such as those of sub-Saharan Africa. Throughout the Intergovernmental Panel for Climate Change-Assessment Reports (IPCC-AR) there is strengthened evidence that the projected climate change simulated by the global mod- els have shown the widespread increase in air temperature, particularly over sub-Saharan Afri- can region [19–21]. The southern Africa region, where Mozambique is located, is projected to experience an increase in spatio-temporal variability of air temperature, concurrent with mul- tiple changes in climate impact-drivers [19, 20]. Such as stated by [22] within Southern Africa, Mozambique is one of the hotspots, as it is particularly vulnerable to climate change com- pounded by high levels of poverty and strong reliance on the rainfed agricultural sector. Many terrestrial species have shifted their geographic ranges, seasonal activities, migration patterns, abundances and species interactions in response to new established climate environ- ment [23–25]. As stated before, FAW is strongly affected by climate factors and climate change may affect its geographical range, survival, mortality and number of generations per year [26– 29]. Some studies [13, 28] have shown that FAW can establish itself in almost all sub-Saharan African countries under current climate but, climate barriers, such as higher temperatures, may limit the spread of FAW to tropical regions close to equator. This statement supports the idea according to which, there are climate-related sensitive thresholds which if crossed have deleterious impact on FAW survival. Although a small portion is somewhat far, most of Mozambique territory are close to equator and the growing season is overwhelmingly hottest. Even though, there is a risk for FAW become transient and permanent population establish- ment in Mozambique under current climate conditions. Meanwhile, it is not yet known what will be the real impact of the predicted future climate on the dynamics and the life cycle of this insect in the country. So far, no study has been conducted to our knowledge to predict the local response of FAW under future climate change scenarios. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 2 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique This research aims to estimate the impact of climate change on FAW life cycle, as well as highlight the risk of outbreak due to future climate relative to present conditions in Mozam- bique. In addition, the study focuses on air temperature-dependence life cycle to map climati- cally month to seasonal thermal units needs for FAW development. Data and methodology Study domain and climatology Geographically, Mozambique is a country located in the coastal region of Southern Africa, bounded between 10˚ 27΄ S and 26˚ 57΄ S of latitude, and 30˚ 12΄ E and 40˚ 51΄ E of longitude, that favors tropical climate conditions (Fig 1). The country holds a long north-south coastline, covering about 2700 km and predominantly characterized by lowlands. However, below 20˚ S of latitude it is observed a marked east-west gradient of topographic caused by plateaus over Fig 1. Geographic location and topography of Mozambique. There are 10 provinces (Maputo, Gaza, Inhambane, Sofala, Manica, Tete, Zambe´zia, Nampula, Cabo Delgado and Niassa) and each capital city is represented by the black marks closed to respective symbol (MP, XX, IN, BR, CH, TT, QL, NP, PB and LC respectively). In the same time, these points represent the weather stations used here (left panel). The black dashed ellipse represents the section of Zambezi River basin. The three climatological sub-regions used in this study are shown in the 3 panels left and, the red dashed-bounded areas indicate the regions with FAW presence confirmed [30]. Direct link to the base layer of the map is https://www.gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g001 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 3 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique the central and northern regions of the country. There are important floodplains areas throughout the broadening Zambezi River valley in central region of Mozambique (area bounded by black shaded ellipse, Fig 1). This is an important agricultural region where farmers grow maize year-round. Therefore, besides to favorable climatic conditions the region may become a hotspot for FAW survival and overwintering, enabling seasonal migration between the main growing-season and the flood-recession crop. This migratory behavior probably allowed the expansion of the FAW habitat over the study domain, where by 2017 it had already been confirmed its presence in six provinces over the 3 regions of the country, namely: Maputo, Gaza, Manica [15, 16], Tete, Zambe´zia and Niassa provinces (Fig 1, red shaded) [30]. The tropical climate is overwhelming predominantly in the country but, as reported in pre- vious paragraph, the east-west gradient of topography, and because the regional and local char- acteristics it is possible to describe local climate diversity throughout the territory by Koppen- Geiger climate classification (Fig 2). Globally, there is an observed rainy season from October to March and, extended dry season from April to September (Figs 3 and 4) (INAM-Instituto Nacional de Meteorologia). The trend line of mean annual precipitation and air temperature, ranges from wetter and hotter, with up to 1500 mm and between ± 25 and ± 27˚C (during summer season), to drier and colder with around 300 mm and between ± 18 and ± 25˚C (throughout extended winter season), respectively. The central region of Mozambique concentrates the greatest amount of precipitation and the one that records in absolute and mean values, the highest air temperatures year-round (Figs 3 and 4). FAW biology and its dynamical distribution FAW is a polyphagous pest with a wide host range feeding on more than 186 plant species, sometimes considered one of the major pests of cereals and forage grasses [10, 11, 13]. The lar- val hatching occurs a few days after the female FAW moth lay eggs on the maize plant leaf. Usu- ally there are six larval instars throughout the FAW life cycle. In the first and second instars the larvae feed on leaves, but eventually they enter in the plant whorl and feed on the unfurled leaves causing extensive defoliation. Duration of the larval stage tends to be about 14 days dur- ing the summer and 30 days during cool weather. After completing the last larval stage, the lar- val pupate into the soil at a depth of 2 to 8 cm, which lasts from 8 to 9 days during the summer, but it may range from 20 to 30 days during the winter. After that the moths emerge from the ground. Under suitable climate conditions, the duration of adult life (moth phase) is estimated at about 10 days, with a range of about 7 to 21 days. Adults are nocturnal, most active during warm and humid evenings and able to migrate to other regions. In warm climate FAW is able to complete the life cycle between 3 to 4 weeks, but in cold weather conditions, sometimes lethal to its survival, may takes considerably longer, up to 45 days or more [3, 28, 31, 32]. One of the first studies focusing in air temperature impact-related of FAW throughout his metamorphosis was made by comparing the development period in each larval stage between summer and winter season [3]. Over the years, several other studies have been done similar analysis. For instance, [33] determined the development rate of FAW at different air tempera- ture levels and additionally computed the amount of thermal units required for the caterpillar to complete each of the larval instar. In the same study was stated that, unlike cold seasons, warmer seasons explicitly may induce to faster development rate, which is reflected in shorter time for each larval stage. Along the same line of research, [34] pointed out that, development time was longer at constant 25˚C than at a mean of 25˚C which fluctuated between 25˚C and 30˚C. Alongside, [35] reports the temporal and morphological parameters of the immature stages of FAW for larvae fed on artificial diet under controlled conditions, such as, 25˚C, 70% of relative humidity and 14 hour photophase. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 4 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 2. Koppen-Geiger climate classification for Mozambique domain. Source of data: International Institute for Applied System Analysis (IIASA). Direct link to the base layer of the map is https://www.gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g002 Under the perspective of the length of time to complete the life cycle it is observed that there is a significant difference in development time from winter to summer, meaning that colder conditions may delay up to 15 days to complete caterpillar stage Fig 5 [33]. Climate impact-related for FAW survival Under continue greenhouse gases emissions, global mean air temperature will continue to rise throughout 21st century. This trend of rising global mean air temperature, usually projected trough future climate scenarios or Representative Concentrations Pathways (RCPs), is PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 5 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 3. The right panel shows the mean annual trend of precipitation (mm, vertical green bars) and air temperature (˚C, red trend line) over the weather stations of Maputo (MP), Xai-xai (XX), Inhambane (IN), Beira (BR), Chimoio (CH), Tete (TT), Quelimane (QL), Nampula (NP), Pemba (PB), Lichinga (LC). Source of data: Instituto Nacional de Meteorologia (INAM). https://doi.org/10.1371/journal.pclm.0000325.g003 expected to reach and/or exceed 1.5˚C (2˚C) under RCP4.5 (RCP8.5) for 2081–2100 relative to 1985–2005 [20]. However, the referred increase of global mean air temperature will not be regionally uniform. The simulation of the numerical models indicates that over sub-Saharan African countries for example, it is virtually certain that there will be more records of extremes (warm) air temperature than colder as the global mean increases. Additionally, some regional- ized simulation have point out that, Mozambique as a country of tropical climate, is expected widespread increase in mean air temperature [22, 37, 38]. Besides to the likely shortening of aforementioned FAW life cycle, the hot and humid cli- mate over tropical and subtropical regions of Mozambique, may allow the insect overwinter and lives year-round, and becomes endemic in the country. These hypothesis can be supported by the field results of [15, 16] and from some few forecasting studies of FAW year-round, potential distribution and global extent, including over sub-Saharan Africa. For instance, besides to established climatic limits for FAW survival, [13] stated that much of sub-Saharan Africa may host the FAW populations year-round. Another important results found through Species Distribution Model (SDM) simulation, indicate the Indian Ocean coast in Southern Africa region (implicit indication of the present study domain) as the only area in southern Africa which has suitable climate environment for FAW survival [28]. It may also mean that, during the southern hemisphere summer season, FAW moths can migrate over long distances throughout the Mozambique regions, and establish transient populations which can jeopardize agricultural crops in the country. This behavior was observed over the native FAW region, and PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 6 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 4. Climatology of precipitation (shaded, mm/month) and air temperature (dashed lines,˚C) for 1971–2000 in Mozambique. Direct link to the base layer of the map is https://www.gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g004 reported by [6, 7, 13, 35]. According to these two studies, FAW moths can travel several hun- dred kilometers in a single night, behavior aimed at escaping adverse climate conditions, demand for host plants and ultimately the maintenance of the species. The previously highlighted findings are reinforced by the warm and humid tropical climate, the prevailing cli- mate feature in Mozambique. As known, the strongest adverse climatic factors for FAW sur- vival and distribution are extreme air temperatures, both minimal as well as maximum. But extreme maximum air temperatures are overwhelmingly more prevalent in the present study domain. Although there are regions of climate modified by altitude, where lower air tempera- tures are usual during winter season, thereby supporting the above-mentioned statements. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 7 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 5. Comparison between summer and winter duration for FAW development throughout caterpillar stage. Adapted by the author through the results presented by [3, 33 and 36]. https://doi.org/10.1371/journal.pclm.0000325.g005 The thermal units required for FAW development The range defined as the sum of daily mean air temperature is called thermal units or Growing Degree-Days (GDD) required for FAW development. However, there is some difference of air temperature required throughout the FAW development stages [33, 39–41]. For instance, [42] and [41] stated that when mean air temperature is between 15 and 25˚C then, daily fluctua- tions above and below these borders increase pupal and larval development rates and decrease adult deformity. Quoted by [42] the developmental time at constant air temperature in the lab- oratory ranged from 66.5 to 18.4 days at 18.3 to 35.0˚C respectively, with incomplete develop- ment at or below 15.6˚C. During the larval instars for FAW populations in the American regions, the feeding rate increase between 25–30˚C [40], while period until pupation decrease between 28.9–33.4˚C [43], larval developing rate increase between 21–30˚C [42] and, the low- est viability to pre-pupal stage occur at 32˚C with duration decreasing when temperature range between 18–32˚C [39]. Furthermore, considering the lack of diapause, the above-men- tioned studies were conducted under some defined lower and upper air temperature thresh- olds (cardinal temperatures) for FAW survival and development. For example, in [44] the lower development threshold was found to be equal to 7.4˚C, meanwhile [45] considered 10˚C, and [31] 13.8˚C. The most fairly conservative upper development threshold mentioned among the several studies is around 39.2˚C [44]. Recently, important result was published in the study of [33]. The authors pointed out that the development rate of FAW increase linearly with daily increasing temperatures between 18 and 30˚C. Otherwise, the larval mortality was highest (lowest) with temperature below 18˚C (between 26 and 30˚C). Note that from 30˚C and beyond, there is a trend to reverse larval mortality, means that the development rate becomes nonlinear (reversal trend) above these thresholds and FAW survival is limited (Fig 6). As can be seen in the same figure, the larval (egg-adult) development period is 34.39 days (71.4 days) at 18˚C, and 10.45 days (20 days) at PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 8 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 6. Correlation between mean development time (days) to egg-adult (dark-blue line, shaded), larval stages (dark-green line, shaded) and larval mortality (brown line, shaded). Source of data: Adapted by the author through the results presented by [33]. https://doi.org/10.1371/journal.pclm.0000325.g006 32˚C, supporting the results according to which FAW development rate is reduced during winter and increased throughout summer season [33]. As a consequence of the correlation described in previous paragraph, the response of FAW to air temperature can be described through a mathematical models and air temperature response functions, such as described briefly by [46–52] although its applications have been commonly used for agricultural crops. However, the physiological response to air temperature is not linear as suggested in most of mentioned papers, but there is a response function similar to parabolic trend line with downward-facing concavity. First of all, the cold-blooded insect develops following an asymmetric quadratic function through which is observed an exponen- tial increase of development rate, ranges from lower threshold (lower cardinal temperature, Tb) up to maximum development rate (optimum air temperature range, Top), around 25˚C. Then, decline sharply from Top to upper threshold (upper cardinal temperature, TB). If by any chance the maximum air temperature increases beyond TB, the insect will be exposed to the lethality conditions for its own survival (Fig 7). Climate data For a comprehensive assessment of the impact and implications of climate change on FAW life cycle, it was necessary to use the output of regional numerical model simulations that span a reasonable range of the likely climate change impact-related. The climate environment for FAW survival was simulated through regional model RegCM4 nested by global model Had- GEM2 for baseline (1971–2000) and future climate (2070–2099). The future climate PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 9 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 7. FAW development rate and its growth range limits, as a function of lower (Tb), optimal (Top) and higher (TB) cardinal temperatures in Celsius degrees. Source: Adapted by the author through the results presented by [53] by the author. https://doi.org/10.1371/journal.pclm.0000325.g007 simulations used here were run under two Representative Concentration Pathways, mid-radia- tive forcing (RCP4.5) and strong-radiative forcing (RCP8.5). These climate scenarios represent the assumption according to which, the increase of global mean surface air temperature by the end of twenty-first century (2081–2100) relative to 1985–2005 is likely to be 1.1˚C to 2.6˚C under RCP4.5 and 2.6˚C to 4.8˚C under RCP8.5 [20]. Although brief short description of the downscaling procedure and the regional model RegCM4 nested to global model HadGEM2 setup provided above, full details can be found in [54]. The GDD FAW requirements and number of generations For each development stage and therefore throughout the life cycle, FAW needs to accumulate amounts of thermal units. However, the insect grows and develop when air temperature is above Tb and below TB. The daily sum of air temperatures between this range and leads the insect to complete its life cycle is referred as thermal units or Growing Degree-Days (GDD,˚C day). This index indicates the thermal units above Tb and below TB to be accumulated for the specie during each day. Since there are no detailed studies for African climate environment and Mozambique in particular, the FAW life history and climate-related data from experimental studies conducted in other parts of the world were used to define the cardinal temperatures for computed the GDD and Number of Generations (NG) throughout the study domain. The conservative find- ings of air temperature thresholds range from 10 to 40˚C and, there is no hatching and no sur- vival at 40˚C or more. Some experiments indicate that when FAW is reared on an artificial diet there is no development between 35 and 37.8˚C [e.g., 28, 31, 33, 39, 41–43, 45]. Hence, based on the experimental data the lower and upper cardinal temperature for FAW population growth and development used in the present study, were left unchanged at Tb = 10˚C and TB = 40˚C, respectively. Meanwhile the optimum range (Top) is from 25˚C to 28˚C. As discussed PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 10 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique early, the development rate of FAW increase linearly with increasing of air temperature from Tb to optimum range, then declines when air temperature increases above this boundary up to TB, as can be seen in [54], Fig 6. Note that, although in Mozambique is not usual to record air temperatures below 10˚C, FAW overwinters only in warm and humid areas. Hence, even if temperatures are above Tb but below 20˚C for example, the development rate may slow down. So, for the purposes of cal- culation were used the bilinear method, which considers daily maximum and minimum tem- peratures, Tb and TB, in addition to two Top ranges, as described in the next section. Some of the crop models applied in most of the studies to simulate the phenology of crop or small animals use the canonical form of GDD method. This mathematical form described by Mcmaster and Wilhelm (1997), considers only the daily mean air temperature relative to lower cardinal temperature. But there are several studies using different mathematical forms [e.g., 41, 42]. For instance, [33] using the canonical method, found 390 GDD with Tb = 12.57˚C as the thermal requirements for FAW complete its life cycle. Similar results was found by [31], with Tb = 13.8˚C were estimated 346.2˚C day throughout FAW life cycle. To get minimum GDD required to complete a generation, [27] and [28] set Tb = 12.0˚C and found 559 and 400˚C day, respectively. Under suitable climate conditions, after the moth lay eggs, it will be hatching from 3 to 4 days and, during this period the estimated GDD accumulation is around 36˚C day. With the same climate conditions and available host crops, the larval (pupal) stage may take from 10 to 13 (7–14) days to complete the stage. During this period, it is expected to accumulate around 255˚C day (97˚C day). Hence, throughout its life cycle it is expected that FAW accumulate around 400˚C day (Fig 8). These estimates were made under Tb = 13˚C and TB�39˚C [31, 33, 39, 41, 42, 44, 45] (Fig 8). For the purpose of GDD calculation and considering the assumptions presented previously, the mathematical model used in this study is the bilinear model, according with: GDD ¼ ðTopt1 (cid:0) TbÞ � ðTarg (cid:0) TbÞ ðTopt1 (cid:0) TbÞ � 1day ; GDD ¼ ðTopt2 (cid:0) TbÞ � ðTavg (cid:0) TBÞ ðTopt2 (cid:0) TBÞ � 1day GDDsum ¼ Xn i¼1 GDDi; ð1Þ ð2Þ ð3Þ with n corresponding to the single day of GDD calculation and GDDsum is the sum of GDD for n days. If Tavg < Tb, then Tavg = Tb, and if, Topt1 < Tavg < Topt2, then Tavg = Topt, then Eq 1 is applied. When Topt2 < Tavg < TB, so Eq 2 is applied. Note that, in the same time where maxi- mum daily air temperature (Tmax) exceeds 40˚C is set equal to 40˚C, and minimum (Tmin) is set at 10˚C when it is less than 10˚C. The most common situation in the present climate condi- tions is observed when minimum Tmin and Tmax are above and below Tb and TB, respectively. Meanwhile, for the computation of the Number of Generations (NG) of FAW, it was applied the methodology established by Parra (1981) and Haddad et al. (1999), quoted in [39] and [55]. According to this method, the mathematical model is written as following: PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 Lc ¼ GDD �T (cid:0) Tb ; ð4Þ 11 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 8. Accumulated GDD (˚C day) to complete each development stage and the time span required under favorable weather conditions. Adapted by the author through the results presented by [31, 33, 39, 41, 42, 44 and 45] by the author. https://doi.org/10.1371/journal.pclm.0000325.g008 then NG ¼ N Lc : Replacing Lc from 4 in 4.1 is obtained: NG ¼ N � �T (cid:0) Tb GDD : ð4:1Þ ð5Þ Where Lc is the life cycle, �T is the daily mean air temperature, and N is the number of days which NG is calculated. The results presented here are regarding to summer season, because the main agroclimatic factors related to FAW survival are observed during this period. In other words, the growing season of the FAW host crops override to summer season in Mozambique, where the agricul- tural activity is predominantly rainfed; on the hand, the summer season over the study domain is in the same time the rainy season, during which is observed warm and humid climate, con- ditions that may favor or not the FAW survival in tropical region. Through the combination of GDD and NG results under the three experiments (baseline, RCP4.5 and RCP8.5), the five categories of GDD index linked to the possible NG were created, related to five levels of risk (very-low, low, medium, medium-high and high) and producing the colors scale for vulnerability. This index allows the delimitation of the risk and PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 12 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Table 1. GDD index and FAW risk and vulnerability. FAW risk and vulnerability GDD index 1 2 3 4 5 Risk Very-low Low Medium Medium-high High https://doi.org/10.1371/journal.pclm.0000325.t001 Vulnerability vulnerability to the FAW in Mozambique. The starting point to define the risk and vulnerabil- ity levels is the NG determined in the reference climatic conditions. The criterion for classify- ing risk and vulnerability levels is: very low for NG less than one, low for NG greater than or equal to one and less than 2, medium for NG greater than or equal to two and less than four, medium-high for NG greater than or equal to four and less than 6, high for NG equal or greater than 6. From there are determined five levels of GDD index, each one associated to the color scale (Table 1). These categorizations were adapted by taking into account two previous studies [26, 56]. Discussion and results Under current climate scenario and during the growing season, from October to March, the largest GDD accumulation (�2600˚C day) is concentrated throughout the coastline of north and central regions of the study domain (Fig 9A). Alongside to this result, there is a narrow hotspot through Zambezi River basin and around XX and IN weather stations. In the same summer season, the lowest GDD accumulation (<2540˚C day) is observed over the plateau areas, inside the central and northern regions. These results are closely matched with highest and lowest monthly and seasonal air temperature records over Mozambique. The lowest air temperature is recorded in the regions where the climate is modified by altitude, meanwhile in the low-lying areas the highest records occur. There are positive and negative responses for RCP4.5 radiative forcing relative to reference scenario. The differences range from less than -100 to more than 100˚C day but, the positive response prevailing (Fig 9B). Additionally, the regions with highest GDD in reference period are those with negative response under RCP4.5 scenario. Meanwhile, for the strongest repre- sentative scenario (RCP8.5), there is a widespread trend to reduction the GDD accumulation (Fig 9C). These negative responses ranges from -500 to just under 50˚C day relative to refer- ence period, and around XX and IN weather stations the signals are sharper and more signifi- cant over the north and central regions. The altitude-modified climate regions remain with a single response from the other regions and in this case, lesser significant GDD reduction. Important to note that the trend of this reduction is more significant along north-central coastline, around XX and IN weather stations, and throughout Zambezi River basin for both scenarios and also, more pronounced feature under RCP8.5 than RCP4.5 radiative forcing’s (Fig 9B and 9C). When the summer season is broken, the similar GDD accumulation pattern is observed during both OND and JFM, but smoothly higher in the first quarter (Fig 9D and 9G). The dif- ference between RCP4.5 relative to reference period allows to observe the come up of a bipo- larization among north and southernmost (positive response), against central-southern (negative response) regions (Fig 9E and 9H). Once again, the pattern of this bipolarization is more comprehensive during OND months (Fig 9E). Otherwise, the bipolarization pattern PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 13 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 9. Growing Degree-Days for present (reference) and future climate scenarios. Fig 9A, Fig 9B and Fig 9C illustrate GDD for summer-season (OND-JFM), but Fig 9A represent accumulated while Fig 9B and Fig 9C represent the difference between each RCP relative to reference period. Fig 9D, Fig 9E and Fig 9F shown the GDD for OND quarter, and Fig 9G, Fig 9H and Fig 9I for JFM. While Fig 9D and g are the total GDD, Fig 9E, Fig 9F, Fig 9H and Fig 9I corresponds to the difference between each scenario and respective quarters, relative to reference period. Direct link to the base layer of the map is https://www. gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g009 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 14 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique almost demise during OND months under RCP8.5 scenario, because the negative response is extended northward (Fig 9F). Although these findings do not be the same during JFM months, the pattern remains for both RCPs but with a westward GDD gradient occurring, being nega- tive through coastline and positive inland (Fig 9H and 9I). An important worth to highlight is that, for both spatio-temporal scale there is the more negative predominance along the Zam- bezi River basin and around XX and IN weather stations, and more positive in altitude-modi- fied climate regions. In addition, throughout the summer season, the OND months contribute more to the variation and establishment of this GDD feature relative to JFM (Fig 9E and 9F, against Fig 9H and 9I). The GDD spatial pattern showed in Fig 9, is closely related to the climatology of the mean air temperature during the extended summer season in Mozambique (Fig 10). As can be seen for reference period (1971–2000) and from October to March, the highest mean air tempera- ture is concentrated on the coastline of the central region and along the Zambezi River basin. Nevertheless, relatively similar values are observed in the coastline of north region and north- west areas of XX and IN weather stations. The lowest values occur in low-lying areas of the southernmost region, and over central and northern plateau regions (Fig 10A). Under both RCPs scenarios a generalized positive response is observed in the study domain. However, the overwhelming positive response occur over central region and northwest areas of XX and IN weather stations (Fig 10B and 10C). This reduction of GDD comes from projected increasing of daily air temperature above Top and TB. For instance, similar results was found by [27] and [28]. According to these studies, under climate change scenarios, areas of climatic suitability for FAW establishment are expected to gradually decrease over time mainly due to heat and dry stress. However, in the same studies is claimed that, if determined climate or weather conditions are met, the estab- lished persistent FAW populations could, in turn, serve as a source of seasonal invasions and migrate into less favorable climatic regions. This latter statement may be usual if the combina- tion of available GDD and irrigated crop fields in the suitable hotspot of Zambezi River basin allow seasonal FAW populations during projected warmer climate in Mozambique. FAW may Fig 10. Climatology of the mean air temperature for reference period (panel a) and the difference between RCP4.5 (panel b) and RCP8.5 (panel c) relative to reference period (1971–2000). The average is relative to extended summer season, from October to March (ONDJFM). Direct link to the base layer of the map is https://www.gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g010 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 15 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique still survive and become established in significant areas in Mozambique, because the suitable thermal units for its life cycle is still favorable. Moreover, the reduction (increase) of GDD in low-lying (plateau) regions may be as manifested shift of suitable area for FAW survival. This scenario may be the trigger to activate the migration behavior of the insect for the most suit- able new regions for FAW development in the country. Similar behavior was observed in native regions of central-north America, as well described by [6] and [7]. The NG of FAW throughout the extended summer season (ONDJFM) and in each quarter (OND and JFM) is shown below (Fig 11). The mean NG during the summer season ranges from around 0.6 to 1.2 under reference scenario. As can be seen, the spatial pattern of NG overlaps the accumulation of GDD in summer season, with greater NG (lower NG) occurring in warmer (colder) climate regions (Fig 11A). Relative to reference period, there is a positive response of both RCPs, means that the NG will increase under futures scenarios. Moreover, besides to the possible suitable agroclimatic conditions for FAW survivor along Zambezi River basin, the great- est increase in NG is observed in this region. Note that similar response is observed in the north- ern areas of XX and IN weather stations. The smallest increase of NG is concentrated in most of the interior of northern region, around southernmost (MP weather station), coastline of XX and IN weather stations, and over plateau areas of the country (Fig 11B and 11C). As pointed out for GDD from a quarterly point of view, it is noted that the largest (smallest) contribution to summer season NG pattern is observed during OND (JFM). The OND (JFM) months contribute with 1.2 to 1.4 (less than 1, except over northwest of XX and IN weather stations) NG for the summer season as a whole (Fig 11D and 11G). The difference between future and reference scenarios during OND months is predominantly positive, keeping more significant in the central region and around northern areas of XX and IN weather stations. But there is observed negative signal over some areas in northern and southernmost (Fig 11E and 11F). The pattern observed in OND is similar to that observed during the JFM months, but less pronounced for positive response and more pronounced for negative. The latter response occurs mainly on the plateau regions and southernmost of the country (Fig 11H and 11I). It is noted that during the summer season, there is a tripolar pattern of the response for FAW-NG. After the discussion of accumulated GDD in the same period, the increase in FAW-NG was expected because there is an inverse relationship between NG and GDD mani- fested in Eq 5. After all, for the insect complete its life cycle, it needs to accumulate a certain amount of GDD. Hence, there is opposite responses between GDD and NG but both in phase with the regions of highest and lowest heat stresses, simulated by [54]. Even more interesting is that the most pronounced answer is concentrated on Zambezi River basin and surrounding areas, the currently most important maize-belt in Mozambique. However, if it is considers valid the assumption stated by [8], according to which Zambezi River basin may acts as a reservoir for this pest, providing suitable climate for overwintering or making perennial infestation, where it can then recolonize in cooler climates in the adjacent highveld from November to April, the increas- ing of NG will enhance the risk of FAW on agricultural crops in the region, especially on maize. The present findings converge with those found by [27], according to which although sev- eral studies, e.g., Altermatt 2010; Karuppaiah & Sujayanad 2012, quoted in [27] and [57] high- light the expansion of insects geographic ranges in warmer climates, little attention has been given to a possible reduction in or disappearance of pest suitability due to a warmer and drier climate. The findings of [54] shows an increase in air temperature and a decrease in mean pre- cipitation that reduces or nullifies the suitability of many areas for FAW over Mozambique. This may be explained by the fact that when higher air temperatures favor the insect life cycle and developmental rates, the same air temperatures may also affect the FAW survival if this increase exceeds the TB, making the insect facing the heat and dry stress. Another important finding is that the increase in air temperature in cold regions may enhance the insect fitness PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 16 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 11. Same as Fig 9 but for mean number of generations of FAW-NG. Direct link to the base layer of the map is https://www.gadm.org/download_country. html. https://doi.org/10.1371/journal.pclm.0000325.g011 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 17 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Table 2. Mean duration for FAW complete its life cycle during the summer-season under reference, RCP4.5 and RCP8.5 over the study domain. The gray-shaded lines indicate the mean accumulated GDD in the same period in the study domain. FAW (Duration of development cycle) Weather Stations Days (Reference) SGDD Days (RCP4.5) SGDD Days (RCP8.5) SGDD LC 37,7 411,5 27,2 435,0 22,0 400,6 PB 41,3 450,0 25,9 414,1 19,8 359,5 NP 41,3 450,0 25,2 403,5 18,1 329,6 QL 40,7 443,5 20,0 319,6 12,6 229,1 BR 41,3 450,0 23,7 379,7 17,5 319,4 TT 41,3 450,0 22,6 360,9 15,0 272,5 CH 40,9 446,4 26,9 431,1 19,8 360,4 VL 41,3 450,0 26,4 422,7 20,3 370,1 IN 41,3 450,0 24,8 396,9 18,5 336,3 XX 41,3 450,0 25,8 413,3 19,5 354,1 MP 40,6 443,0 27,1 433,3 22,0 400,9 https://doi.org/10.1371/journal.pclm.0000325.t002 and survival, and hence, there is a shift in unsuitable to suitable areas in colder places (plateau areas) and a reduction in or disappearance of suitable conditions (most of low-lying areas) in some regions of Mozambique. Table 2 shows the sum of mean GDD required for FAW to complete its life cycle, computed through Eq 3. The mean time for FAW complete its life cycle is around 40.5, 26.8 and 21.7 days for reference, RCP4.5 and RCP8.5, respectively. Hence, the mean GDD required for each corresponding abovementioned scenario are 431.2, 412.4 and 338.1˚C day. There is a clear fall- ing trend of GDD and shortening the life cycle as radiative forcing become stronger. This result is a consequence of the increase in air temperature in the same direction as the climatic scenarios considered in the present study. In laboratory experiment studies, [39] find out 463 GDD, [33] converge more with the present results, with 391.61±1.42 GDD requirements for egg-to-adult development thermal time. However, Ramirez et al. (1987) quoted by [26] found a significantly higher value of GDD, 599˚C day. Under current climate the estimated number of generations is higher in hottest regions and northern coastal of Mozambique, where the mean global number of FAW generations during the summer season is around 1.07. It means that, it is expected to occur around one generation per month from October to March and the largest contribution came from the central region of the country. Hence, throughout the summer season, the largest number of FAW genera- tions occurs in TT and QL with 7.68 and 7.02, meanwhile the lowest number of FAW NG is observed in LC and CH weather stations with 4.68 and 5.55, respectively (Table 3). There is a little difference between OND and JFM seasons. Table 3. Monthly number of FAW generations during summer-season. Weather Stations LC PB NP QL BR TT CH VL IN XX MP FAW: Number of generations Reference Month OCT NOV DEC JAN FEB MAR Total ONDJFM OND JFM Mean ONDJFM 0,82 0,81 0,77 0,76 0,76 0,76 4,68 2,40 2,28 0,78 1,06 1,10 1,14 1,15 1,12 1,11 6,68 3,30 3,38 1,11 1,07 1,12 1,11 1,09 1,07 1,05 6,51 3,30 3,21 1,08 1,09 1,16 1,20 1,22 1,19 1,16 7,02 3,44 3,58 1,17 1,02 1,08 1,13 1,19 1,17 1,14 6,73 3,23 3,50 1,12 Generations 1,33 1,36 1,27 1,26 1,23 1,23 7,68 3,96 3,72 1,28 0,87 0,93 0,95 0,97 0,93 0,90 5,55 2,75 2,80 0,92 0,99 1,06 1,14 1,18 1,15 1,12 6,65 3,19 3,46 1,11 0,93 0,99 1,07 1,14 1,12 1,09 6,35 2,99 3,36 1,06 0,93 1,00 1,08 1,13 1,09 1,06 6,28 3,01 3,28 1,05 0,87 0,97 1,10 1,15 1,13 1,08 6,31 2,95 3,36 1,05 https://doi.org/10.1371/journal.pclm.0000325.t003 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 18 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Table 4. Monthly number of FAW generations during summer-season, under RCP4.5, and the difference relative to reference period. Weather Stations LC PB NP QL BR TT CH VL IN XX MP FAW: Number of generations RCP45 Month OCT NOV DEC JAN FEB MAR Total ONDJFM OND JFM ONDJFM ONDJFM Mean RCP45-Ref 1,14 1,09 1,07 1,21 1,21 1,14 6,86 3,30 3,56 1,14 2,18 1,44 1,34 1,28 1,17 1,25 1,34 7,81 4,05 3,77 1,30 1,13 1,43 1,34 1,27 1,30 1,34 1,35 8,03 4,04 3,99 1,34 1,52 1,93 2,01 1,93 1,75 1,64 1,58 10,83 5,87 4,96 1,81 3,81 Generations 1,81 1,92 1,73 1,54 1,45 1,41 9,85 5,46 4,40 1,64 2,18 1,26 1,25 1,19 1,17 1,13 1,10 7,09 3,69 3,40 1,18 1,55 1,51 1,47 1,43 1,44 1,42 1,41 8,68 4,41 4,27 1,45 1,95 1,24 1,22 1,26 1,30 1,31 1,30 7,63 3,72 3,91 1,27 0,98 1,37 1,30 1,31 1,34 1,33 1,30 7,95 3,98 3,97 1,33 1,61 1,28 1,21 1,24 1,29 1,28 1,27 7,57 3,74 3,84 1,26 1,29 1,16 1,09 1,13 1,19 1,17 1,15 6,90 3,39 3,51 1,15 0,59 https://doi.org/10.1371/journal.pclm.0000325.t004 Unlike to reference period, the mean monthly number of FAW generations under both future scenarios is above one in all weather stations (Tables 4 and 5). However, the largest and smallest records continue to be recorded around the same weather stations as those of the ref- erence period. From reference period to RCP4.5 and RCP8.5 scenarios, the global mean of FAW generations increase from 1.07 to 1.35 and 1.82, respectively. In addition, the monthly values are slightly higher than one, overestimating in several cases one and half, and in a few others reaching two generations per month. It is important to worth highlighted that the great- est contribution to the total generation in the central region, the most affected region, occurs during OND season, a pattern that is not observed over northern and southern regions of the study domain. When the monthly difference between NG in each future scenarios relative to reference period is made, there is a clear positive response in all weather stations and surrounding areas. But, as noted in previous results, the most significant response occurs for RCP8.5 than for RCP4.5. Moreover, the most positive response still occurring over central region, with 3.81 (9.29) and 2.18 (8.20) NG for QL and TT respectively, under RCP4.5 (RCP8.5). Surprisingly, the order of magnitude of FAW NG response in LC weather station is included in the group of Table 5. Monthly number of FAW generations during summer-season under RCP85, and the difference relative to reference period. Weather Stations LC PB NP QL BR TT CH VL IN XX MP FAW: Number of generations RCP85 Month OCT NOV DEC JAN FEB MAR Total ONDJFM OND JFM ONDJFM ONDJFM Mean RCP85-Ref Generations 1,41 1,42 1,42 1,56 1,52 1,41 8,74 4,25 4,49 1,46 4,06 1,78 1,62 1,51 1,41 1,51 1,62 9,45 4,91 4,54 1,57 2,77 1,96 1,83 1,73 1,76 1,80 1,80 2,98 3,11 3,00 2,60 2,35 2,27 1,98 1,89 1,82 1,84 1,75 1,79 2,73 3,06 2,75 2,39 2,80 2,14 10,87 16,31 11,07 15,88 5,51 5,36 1,97 4,36 9,09 7,22 2,72 9,29 5,69 5,38 1,85 4,34 8,54 7,34 2,65 8,20 1,74 1,75 1,66 1,58 1,52 1,49 9,73 5,15 4,59 1,62 4,19 1,56 1,47 1,46 1,53 1,55 1,55 9,12 4,49 4,63 1,52 2,47 1,75 1,76 1,75 1,77 1,76 1,73 1,72 1,64 1,67 1,68 1,67 1,66 10,52 10,04 5,26 5,26 1,75 4,18 5,03 5,02 1,67 3,76 1,45 1,34 1,37 1,41 1,41 1,41 8,39 4,16 4,23 1,40 2,08 https://doi.org/10.1371/journal.pclm.0000325.t005 PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 19 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 12. Total number of FAW generations for reference period during the summer season, and the difference between RCP4.5 and RCP8.5 relative to reference period in Mozambique. Direct link to the base layer of the map is https://www.gadm.org/download_country.html. https://doi.org/10.1371/journal.pclm.0000325.g012 the second highest, contrary to what was observed in seasonal and global time scale. This unex- pected result becomes from the higher response under both radiative forcing’s. Besides to these results, the lower positive response occurs in MP weather station. Considering the results discussed above and supported on FAW-NG, regions of risk and vul- nerability were bounded and classified according to Fig 12. It can be easily observed that there are two regions marked out as hotspot with more than 6.5 NG throughout south hemisphere summer season for reference period. At the same time, over the regions of humid tropical cli- mate with cold winter (altitude-modified climate in restricted central and northern region) and subtropical climate in the southernmost of the country, the smallest number (less than 5 NG) of FAW generations is observed (Fig 12A). The difference between RCP4.5 and reference period (Fig 12B) is entirely positive, with the huge extension of maximum risk and vulnerability (more than 2.5 NG relative to reference period) occurs along almost the entire Mozambican coastline and around the “narrow strip” of the Zambezi River basin (Fig 12B). Likewise, but most strong response occurs for RCP8.5 scenarios relative to reference period. It is observed from 4 to more than 6.5 NG over the similar hotspot identified in current climate (Fig 12A and 12C). The GDD index represent the delimitation of risk and vulnerability areas for FAW outbreak (Fig 13). Thus, it can be defined risk as the possibility of occurrence (highest or lowest) of FAW in any region of the study domain. As can be seen the minimum level of GDD index occur in restricted plateau areas, meanwhile the low level is noticeable over most of north region and some areas around MP, CH and TT weather stations. Furthermore, medium and medium-high GDD index is predominantly observed over most of central and southern region of Mozambique, with emphasis to the strip of greatest increase of NG (Fig 13). This latest find- ing indicates that where the GDD index is medium-high represent the region of highest risk and extreme vulnerability to outbreak of FAW. Conclusion Our findings indicate that for both RCPs scenarios it is expected a significant reduction of GDD for FAW life cycle along north and central coastline of Mozambique. Besides to this negative GDD response there are two important hotspots, one of them in northern areas of Xai-Xai (XX) PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 20 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Fig 13. GDD index for FAW generations in Mozambique. Direct link to the base layer of the map is https://www.gadm.org/download_ country.html. https://doi.org/10.1371/journal.pclm.0000325.g013 and Inhambane (IN) weather stations, and other throughout the narrow strip and surroundings areas of the Zambezi River basin. Meanwhile, most of the positive GDD response for future cli- mate is observed over the plateau regions, where the air temperatures records are the lowest under present climate in Mozambique. Meaning that the regions with the highest (lowest) GDD accumulation in the present climate have negative response (positive response) in the GDD under RCP4.5 and RCP8.5. The referred response is stronger during OND quarter than JFM, and under RCP8.5 than RCP4.5. The results linked to decrease and increase of GDD are the consequence of significant increase in number of daily air temperatures above TB, and between Tb and Top, respectively. This statement is one of the findings highlighted by [54]. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 21 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique Concerning to NG it is expected that FAW takes out from 37.7 to 41.3 days to complete its life cycle during the summer season in the present climate environment. However, under future climate scenarios, it is observed a reduction in the time to complete the life cycle, as rep- resentative scenarios become stronger relative to reference period. This shortening of time to complete the FAW life cycle is closely related to decreasing in the mean GDD accumulated during summer season, forced by increasing of air temperature under both future scenarios simulated over Mozambique. The immediate consequence of the shortening of the FAW life cycle is the increase in NG, with the largest contribution occurring during OND quarter. Hence, more than 2.5 generations are projected to occur along north and central Mozambique coastline, besides to surroundings of Zambezi River basin, and north areas of XX and IN weather stations. These are the regions of medium-high risk and higher vulnerability to FAW outbreak. From here it is possible to suggest that increasing air temperature allow speed up the insect life cycle leading to a faster increase in FAW populations through the largest NG per year. However, unlike this statement the FAW survival and its development rate may also reduce as air temperature increases up to exceeding its survival thresholds. The year-round availability of maize combined to warm and moist climate over most of the regions mentioned above, may favor a suitable agroclimatic environment for FAW popula- tions to become pandemic in Mozambique. For this reason, through natural migration it should not rule out the possibility of seasonal migration from the maize belt to other regions of Mozambique, just because throughout the life cycle FAW moths use maize growing areas as “stepping-stones” to self-survivor of the species. Author Contributions Conceptualization: Telmo Cosme A. Sumila, Simone E. T. Ferraz, Angelica Durigon. Data curation: Simone E. T. Ferraz, Angelica Durigon. Formal analysis: Telmo Cosme A. Sumila. Funding acquisition: Telmo Cosme A. Sumila, Simone E. T. Ferraz, Angelica Durigon. Investigation: Telmo Cosme A. Sumila. Methodology: Telmo Cosme A. Sumila, Angelica Durigon. Project administration: Simone E. T. Ferraz, Angelica Durigon. Resources: Simone E. T. Ferraz, Angelica Durigon. Software: Simone E. T. Ferraz, Angelica Durigon. Supervision: Simone E. T. Ferraz, Angelica Durigon. Validation: Telmo Cosme A. Sumila, Simone E. T. Ferraz, Angelica Durigon. Visualization: Telmo Cosme A. Sumila. Writing – original draft: Telmo Cosme A. Sumila. Writing – review & editing: Telmo Cosme A. Sumila, Simone E. T. Ferraz, Angelica Durigon. References 1. Baudron F, Zaman-Allanh MA, Chaipa I, Chari N. Understanding the factors influencing fall armyworm (Spodoptera frugiperda J.E. Smith) damage in African smallholder maize fields and quantifying its impact on yield. A case study in Eastern Zimbabwe. Crop Protection, v. 120, p. 141–150, 2019. 2. Jacobs A, Van Vuuren A, Rong IH. Characterisation of the Fall Armyworm (Spodoptera frugiperda J.E. Smith) (Lepidoptera: Noctuidae) from South Africa. African Entomology, v. 26, n. 1, p. 45–49, 2018. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 22 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique 3. Luginbill P. The fall army worm. Nature, v. 121, n. 3054, p. 770–771, 1928. 4. Perkins WD. Laboratory Rearing of the Fall Armyworm Author (s): Published by: Florida Entomological Society. Florida Entomologist, v. 62, n. 2, p. 87–91, 1979. 5. Sparks AN. A Review of the Biology of the Fall ArmywormThe Florida Entomologist, 1979. 6. Westbrook J, Fleischer S, Jairam S, Meagher R, Nagoshi R, et al. Multigenerational migration of fall armyworm, a pest insect. Ecosphere, v. 10, n. 11, 2019. 7. Westbrook JK, Nagoshi RN, Meagher RL, Fleischer SJ, Jairam S. Modeling seasonal migration of fall armyworm moths. International Journal of Biometeorology, v. 60, n. 2, p. 255–267, 2016. https://doi. org/10.1007/s00484-015-1022-x PMID: 26045330 8. Chimweta M, Nyakudya IW, Jimu L. Fall armyworm [Spodoptera frugiperda (J. E. Smith)] damage in maize: management options for flood- recession cropping smallholder farmers. International Journal of Pest Management, v. 0, n. 0, p. 1–13, 2019. 9. Cock MJW, Beseh PK, Buddie AG, Cafa´ G, Crozier J. Molecular methods to detect Spodoptera frugi- perda in Ghana, and implications for monitoring the spread of invasive species in developing countries. Scientific Reports, v. 7, n. 1, p. 1–10, 2017. 10. Day R, Abrahams P, Bateman M, Beale T, Clottey V, Cock M, et al. Fall armyworm: Impacts and impli- cations for Africa. Outlooks on Pest Management, v. 28, n. 5, p. 196–201, 2017a. 11. Day R, Abrahams P, Bateman M, Beale T, Clottey V, Cock M, et al. Fall armyworm: impacts and impli- cations for Africa. Outlooks on pest management. Outlooks on Pest Management, v. 28, n. 5, p. 196– 201, 2017b. 12. Goergen G, Kumar PL, Sankung SB, Togola A, Tamò M. First report of outbreaks of the fall armyworm spodoptera frugiperda (J E Smith) (Lepidoptera, Noctuidae), a new alien invasive pest in West and Cen- tral Africa. PLoS ONE, v. 11, n. 10, p. 1–9, 2016. 13. Early R, Gonza´ lez-Moreno P, Murphy ST, Day R. Forecasting the global extent of invasion of the cereal pest Spodoptera frugiperda, the fall armyworm. NeoBiota, v. 50, n. 40, p. 25–50, 2018. 14. Nagoshi RN, Koffi D, Agboka K, Tounou KA, Banerjee R, Jurat-Fuentes JL, et al. Fall armyworm migra- tion across the lesser antilles and the potential for genetic exchanges between north and south Ameri- can populations. PLoS ONE, v. 12, n. 2, p. 1–18, 2017. 15. Canic¸o A, Mexia A, Santos L. Seasonal dynamics of the alien invasive insect pest spodoptera frugi- perda smith (Lepidoptera: Noctuidae) in manica province, central Mozambique. Insects, v. 11, n. 8, p. 1–12, 2020a. https://doi.org/10.3390/insects11080512 PMID: 32784750 16. Canic¸o A, Mexia A, Santos L. First report of native parasitoids of fall armyworm spodoptera frugiperda smith (Lepidoptera: Noctuidae) in mozambique. Insects, v. 11, n. 9, p. 1–12, 2020b. https://doi.org/10. 3390/insects11090615 PMID: 32911875 17. IAI M. Inque´ rito Integrado Agra´ rio 2020. MADER, Maputo.Mocambique, p. 1–84, 2020. 18. Nagoshi RN, Goergen G, Du Plessis H, van den Berg J, Meagher R Jr. Genetic comparisons of fall armyworm populations from 11 countries spanning sub-Saharan Africa provide insights into strain com- position and migratory behaviors. Scientific Reports, v. 9, n. 1, p. 1–11, 2019. 19. Masson-Delmotte VP, Zhai A, Pirani SL, Connors C, Pe´ an S, Berger N, et al. Climate Change 2021— The Physical Science Basis—Summary for Policymakers. https://doi.org/10.1017/9781009157896.001 20. Pachauri RK, Allen MR, Barros VR, Broome J, Cramer W, Christ R, et al. Climate Change 2014: Syn- thesis Report. Contribution of working groups I, II, III to the Fifth Assessment Report of the Intergovern- mental Panel on Climate Change. IPCC, Geneva, Swizerland, 151 pp, 2014. 21. Bernstein L, Bosh P, Canziani O, Chen Z, Christ R, Davidson O, et al. Climate Change 2007. Synthesis Report. Contribution of Working Groups I, II, III to the Fourth Assessment Report of the Intergovernmen- tal Panel on Climate Change. IPCC, Geneva, Switzerland, 104 pp, 2007. 22. Mavume AF, Banze BE, Macie AO, Queface AJ. Analysis of climate change projections for mozam- bique under the representative concentration pathways. Atmosphere, v. 12, n. 5, 2021. 23. Boggs CL. The fingerprints of global climate change on insect populations. Current Opinion in Insect Science, v. 17, p. 69–73, 2016. https://doi.org/10.1016/j.cois.2016.07.004 PMID: 27720076 24. Shaw AK. Drivers of animal migration and implications in changing environments. Evolutionary Ecology, v. 30, n. 6, p. 991–1007, 2016. 25. Van Dyck H, Bonte D, Puls R, Gotthard K, Maes D, et al. The lost generation hypothesis: Could climate change drive ectotherms into a developmental trap? Oikos, v. 124, n. 1, p. 54–61, 2015. 26. Ramı´rez-Cabral N, Medina-Garcı´a G, Kumar L. Increase of the number of broods of Fall Armyworm (Spodoptera frugiperda) as an indicator of global warming. Revista Chapingo Serie Zonas A´ ridas, v. 19, n. 1, p. 1–16, 2020. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 23 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique 27. Ramirez-Cabral NYZ, Kumar L, Shabani F. Global alterations in areas of suitability for maize production from climate change and using a mechanistic species distribution model (CLIMEX). Scientific Reports, v. 7, n. 1, p. 1–13, 2017a. 28. 29. Timilsena BP, Niassy S, Kimathi E, Abdel-Rahman EM, Seidl-Adams I, Wamalwa M, et al. Potential Distribution of Fall Armyworm in Africa and Beyond, Considering Climate Change and Irrigation Pat- terns. p. 1–24, 2021. Zacarias DA. Global bioclimatic suitability for the fall armyworm, Spodoptera frugiperda (Lepidoptera: Noctuidae), and potential co-occurrence with major host crops under climate change scenarios. Cli- matic Change, v. 161, n. 4, p. 555–566, 2020. 30. Geller L, Zacarias A. Mozambique Fall Armyworm. p. 1–5, 2018. 31. Hogg DB, Pitre HN, Anderson RE. Assessment of Early-Season Phenology of the Fall Armyworm(Lepi- doptera:Noctuidae)in Mississippi 1. Environmental Entomology, v. 11, n. 3, p. 705–710, 1982. 32. Nagoshi RN, Meagher RL. Review of fall armyworm (Lepidoptera: Noctuidae) genetic complexity and migration. Florida Entomologist, v. 91, n. 4, p. 546–554, 2008. 33. Du Plessis H, Schlemmer ML, Berg VDJ. The effect of temperature on the development of Spodoptera frugiperda (Lepidoptera: Noctuidae). Insects, v. 11, n. 4, 2020. https://doi.org/10.3390/ insects11040228 PMID: 32272548 34. Barfield CS, Ashley TR. Effects of Corn Phenology and Temperature on the Life Cycle of the Fall Army- worm, Spodoptera frugiperda (Lepidoptera: Noctuidae). The Florida Entomologist, v. 70, n. 1, p. 110– 116, 1987. 35. Montezano DG, Specht A, Sosa-Go´ mez DR, Roque-Specht VF, Paula-Moraes SV, Peterson JA, et al. Developmental Parameters of Spodoptera frugiperda (Lepidoptera: Noctuidae) Immature Stages Under Controlled and Standardized Conditions. Journal of Agricultural Science, v. 11, n. 8, p. 76, 2019. 36. FAO, CABI. Community-Based Fall Armyworm (Spodoptera frugiperda) monitoring, early warning and management, Training of Trainers Manual. First edition. 112 pp.Licence: CC BY-NC-AS 3.0 IGO, 2019. 37. Maure G, Ndebele-Murisa MR, Pinto I, Muthige M. The southern African climate under 1.5˚c and 2˚c of global warming as simulated by CORDEX regional climate models. Environmental Research Letters, v. 13, n. 6, 2018. 38. Pinto I, Jack C, Hewitson B. Process-based model evaluation and projections over southern Africa from Coordinated Regional Climate Downscaling Experiment and Coupled Model Intercomparison Project Phase 5 models. International Journal of Climatology, v. 38, n. 11, p. 4251–4261, 2018. 39. Busato GR, Grutzmacher AD, Garcia MS, Giolo FP, Zotti MJ, Bandeira JM. Thermal requirements and estimate of the number of generations of biotypes “corn” and “rice” of Spodoptera frugiperda. Pesquisa Agropecuaria Brasileira, v. 40, n. 4, p. 329–335, 2005. 40. Isenhour DJ, Wiseman BR, Widstrom NW. Fall Armyworm (Lepidoptera: Noctuidae) Feeding Responses on Corn Foliage and Foliage/Artificial Diet Medium Mixtures at Different Temperatures1. Journal of Economic Entomology, v. 78, n. 2, p. 328–332, 1985. 41. Simmons AM. Effects of Constant and Fluctuating Temperatures and Humidities on the Survival of Spo- doptera frugiperda Pupae (Lepidoptera: Noctuidae). The Florida Entomologist, v. 76, n. 2, p. 333–340, 1993. 42. Barfield CS, Mitchell ER, Poeb SL. A Temperature-Dependent Model for Fall Armyworm Develop- ment1,2. Annals of the Entomological Society of America, v. 71, n. 1, p. 70–74, 1978. 43. Elderd BD, Reilly JR. Warmer temperatures increase disease transmission and outbreak intensity in a host-pathogen system. Journal of Animal Ecology, v. 83, n. 4, p. 838–849, 2014. https://doi.org/10. 1111/1365-2656.12180 PMID: 24219180 44. Valdez-Torres JB, Soto-Landeros F, Osuna-Enciso T, Ba´ez-Sañudo MA, et al. Phenological prediction models for white corn (Zea mays L.) and fall armyworm (Spodoptera frugiperda J. E. Smith). Agrocien- cia, v. 46, n. 4, p. 399–410, 2012. 45. Wood JR, Poe SL, Leppla NC. Winter survival of fall armyworm pupae in Florida. Environmental Ento- mology, v. 8, n. 6, p. 249–252, 1979. 46. Kumudini S, Andrade F.H, Boote KJ, Brown GA, Dzotsi KA, Edmeades GO, et al. Predicting maize phe- nology: Intercomparison of functions for developmental response to temperature. Agronomy Journal, v. 106, n. 6, p. 2087–2097, 2014. 47. Mcmaster GS, Wilhelm WW. Growing degree-days: One equation, two interpretations. Agricultural and Forest Meteorology, v. 87, n. 4, p. 291–300, 1997. 48. Streck NA, Lago I, Gabriel LF, Samboranha FK. Simulating maize phenology as a function of air tem- perature with a linear and a nonlinear model. Pesquisa Agropecuaria Brasileira, v. 43, n. 4, p. 449–455, 2008. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 24 / 25 PLOS CLIMATEClimate change impact on Spodoptera frugiperda life cycle in Mozambique 49. Wang E, Martre P, Zhao Z, Ewert F, Maiorano A, Rotter RP, et al. The uncertainty of crop yield projec- tions is reduced by improved temperature response functions. Nature Plants, v. 3, n. July, 2017. 50. Wang E, Engel T. Simulation of phenological development of wheat crops. Agricultural Systems, v. 58, n. 1, p. 1–24, 1 set. 1998. 51. Wang N, Wang J, Wang E, Yu Q, Shi Y, He D, et al. Increased uncertainty in simulated maize phenol- ogy with more frequent supra-optimal temperature under climate warming. European Journal of Agron- omy, v. 71, p. 19–33, 2015. 52. Weikai Y, Hunt L. A. An equation for modelling the temperature response of plants using only the cardi- nal temperatures. Annals of Botany, v. 84, n. 5, p. 607–614, 1999. 53. Pereira AR, Angelocci LR, Sentelhas PC. Meteorologia agricola. Agrometeorologia: Fundamentos e Aplicac¸ões Pra´ ticas, v. 306, p. 192, 2007. 54. Sumila TCA, Ferraz SET, Durigon A. Maize phenology as an indicator of climate change simulated by RegCM4 under RCP4.5 and RCP8.5 in Mozambique. African Journal of Agricultural Research. V. 19 (8), p. 774–788, 2023. 55. Correa LRB, Santa-Cecı´lia LVC, Sousa B, Civiadanes FJ, Pedroso EC. Exigências te´rmicas da cocho- nilha-branca Planococcus citri (Risso, 1813) (hemiptera: pseudococcidae) em cafeeiro (coffea arabica) CV. Mundo novo. n. 1988, p. 2–4, 1997. 56. Shabani F, Kumar L, Taylor S. Projecting date palm distribution in Iran under climate change using topography, physicochemical soil properties, soil taxonomy, land use, and climate data. Theoretical and Applied Climatology, v. 118, n. 3, p. 553–567, 2014. 57. Ramirez-Cabral NYZ, Kumar L, Shabani F. Future climate scenarios project a decrease in the risk of fall armyworm outbreaks. Journal of Agricultural Science, v. 155, n. 8, p. 1219–1238, 2017b. PLOS Climate \| https://doi.org/10.1371/journal.pclm.0000325 January 18, 2024 25 / 25 PLOS CLIMATE
10.1371_journal.pbio.3002514	RESEARCH ARTICLE Engineering of Cas12a nuclease variants with enhanced genome-editing specificity Peng Chen1☯, Jin Zhou1☯, Huan Liu1☯, Erchi Zhou2☯, Boxiao He1, Yankang Wu1, Hongjian Wang1, Zaiqiao Sun1, Chonil Paek1,3, Jun Lei1, Yongshun Chen1, Xinghua Zhang2, Lei YinID 1* 1 State Key Laboratory of Virology, Hubei Key Laboratory of Cell Homeostasis, College of Life Sciences, Department of Clinical Oncology, Renmin Hospital of Wuhan University, Wuhan University, Wuhan, China, 2 The Institute for Advanced Studies, College of Life Sciences, State Key Laboratory of Virology, Hubei Key Laboratory of Cell Homeostasis, Wuhan University, Wuhan, China, 3 The Faculty of Life Science, KIM IL SUNG University, Pyongyang, Democratic People’s Republic of Korea ☯ These authors contributed equally to this work. * [email protected] (YC); [email protected] (XZ); [email protected] (LY) Abstract AU : Pleaseconfirmthatallheadinglevelsarerepresentedcorrectly: The clustered regularly interspaced short palindromic repeat (CRISPR)-Cas12a system is a powerful tool in gene editing; however, crRNA-DNA mismatches might induce unwanted cleavage events, especially at the distal end of the PAM. To minimize this limitation, we engi- neered a hyper fidelity AsCas12a variant carrying the mutations S186A/R301A/T315A/ Q1014A/K414A (termed HyperFi-As) by modifying amino acid residues interacting with the tar- get DNA and crRNA strand. HyperFi-As retains on-target activities comparable to wild-type AsCas12a (AsCas12aWT) in human cells. We demonstrated that HyperFi-As has dramatically reduced off-target effects in human cells, and HyperFi-As possessed notably a lower tolerance to mismatch at the position of the PAM-distal region compared with the wild type. Further, a modified single-molecule DNA unzipping assay at proper constant force was applied to evalu- ate the stability and transient stages of the CRISPR/Cas ribonucleoprotein (RNP) complex. Multiple states were sensitively detected during the disassembly of the DNA-Cas12a-crRNA complexes. On off-target DNA substrates, the HyperFi-As-crRNA was harder to maintain the R-loop complex state compared to the AsCas12aWT, which could explain exactly why the HyperFi-As has low off-targeting effects in human cells. Our findings provide a novel version of AsCas12a variant with low off-target effects, especially capable of dealing with the high off-tar- geting in the distal region from the PAM. An insight into how the AsCas12a variant behaves at off-target sites was also revealed at the single-molecule level and the unzipping assay to evalu- ate multiple states of CRISPR/Cas RNP complexes might be greatly helpful for a deep under- standing of how CRISPR/Cas behaves and how to engineer it in future. Introduction The clustered regularly interspaced short palindromic repeat (CRISPR)-associated protein (Cas) system allows for a wide range of applications for gene modification when guided by RNA and in the presence of protospacer-adjacent motif (PAM) sequence [1–8]. However, a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Chen P, Zhou J, Liu H, Zhou E, He B, Wu Y, et al. (2024) Engineering of Cas12a nuclease variants with enhanced genome-editing specificity. PLoS Biol 22(3): e3002514. https://doi.org/ 10.1371/journal.pbio.3002514 Academic Editor: Bon-Kyoung Koo, Center for Genome Engineering, REPUBLIC OF KOREA Received: June 19, 2023 Accepted: January 22, 2024 Published: March 14, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pbio.3002514 Copyright: © 2024 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The GUIDE-seq raw sequencing reads are available at the Gene Expression Omnibus (GEO) under accession GSE229888 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 1 / 24 Funding: This work was supported by the National Key R&D Program of China (2022YFA1303500 to LY), the National Natural Science Foundation of China (32101196 to PC, 32171210 and 31870728 to LY), the Fundamental Research Funds for the Central Universities (2042022kf1189 to LY), the China Postdoctoral Science Foundation (2021TQ0253 and 2022M712468 to PC, 2022M722473 to JZ). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Improved genome editing specificity by HyperFi-As crRNA-DNA mismatches may also induce cleavage activity of Cas nucleases, resulting in undesired cleavage at off-target sites [9–16]. The off-target effects not only confound the reli- ability of research experiments in the lab but also have serious implications for clinical utility. So far, several strategies have been developed to improve the specificity of the Cas nucleases, including the end-point selection after the generation of random mutagenesis by the error- prone PCR [17], direct evolution [18,19], engineering of the sgRNAs (ggX20 sgRNAs and tru- sgRNAs) [20,21], delivery of Cas9-sgRNA ribonucleoprotein (RNP) complexes [22,23], usage of dimeric RNA-guided FokI Nucleases (RFNs) [24], and directional screening in human cells [25]. In particular, structure-guided protein engineering is a very useful method to design high-fidelity Cas-nuclease variants, which has been demonstrated in SpCas9, such as SpCas9-HF1 [15], HypaCas9 [26], eSpCas9 [27], HeFSpCas9 [28], SaCas9-HF [29], etc. CRISPR, clustered regularly Abbreviations: AU : Anabbreviationlisthasbeencompiledforthoseusedinthetext:Pleaseverifythatallentriesarecorrect: interspaced short palindromic repeat; DSB, double- strand break; dsODN, double-stranded oligodeoxynucleotide; EGFP, enhanced GFP; GUIDE-seq, genome-wide unbiased identification of double-stranded breaks enabled by sequencing; PAM, protospacer-adjacent motif; RFLP, restriction-fragment length polymorphism; RFN, RNA-guided FokI Nuclease; RNP, ibonucleoprotein; WT, wild type. In addition to the commonly used type II-A Cas9, the type V-A system Cas12a can also cleave genome DNA efficiently in vivo. Unlike Cas9, Cas12a orthologs (As- (Acidaminococcus sp.) and Lb- (Lachnospiraceae bacterium) Cas12a nucleases) have several distinct features [5– 7,9]. First, Cas12a recognizes the thymine-rich PAM sequence upstream of the target region and triggers the cleavage of target DNA in the PAM distal position, generating staggered ends [5]. Second, Cas12a can process its own CRISPR crRNA array into mature crRNAs to mediate gene editing at different genome sites simultaneously and does not require a trans-activating crRNA [6,7,30]. Moreover, Cas12a has low off-targeting in human cells due to low crRNA-DNA mismatch tolerance [9,31]. Therefore, Cas12a is considered to be suitable for multiplex gene editing and accurate genome modification. Although Cas12a nuclease provides a powerful potential for genome engineering, specific- ity still needs to be improved for clinical application. In the previous studies of genome-wide specificities of CRISPR-AsCas12a nucleases in human cells, several cleavage events happen in the spacer region with a single mismatch base pair [9]. Several AsCas12a variants have been developed to decrease the indels event at the unwanted loci [32–34]. However, mismatches in the PAM-distal region were frequently observed in off-target sites and few strategies were developed to specifically solve the problem. To address the issue, we engineered a hyper fidelity AsCas12a variant with the mutations S186A/R301A/T315A/Q1014A/K414A (termed HyperFi-As) by modifying amino acid residues interacting with the backbone of the target DNA strand and crRNA strand in both the proximal and distal regions to the PAM through structure-guided protein engineering. Using the genome-wide unbiased identification of dou- ble-stranded breaks enabled by sequencing (GUIDE-seq) method, we demonstrated that HyperFi-As dramatically reduced the off-targeting in both the proximal and distal regions to the PAM and displayed better fidelity than AsCas12aWT, the recent versions of Cas12a (As ultra, As plus), and the very low off-target Cas9 variant (SuperFi-SpCas9) reported by Taylor etc. in 2022 which attracts great attentions [35]. Furthermore, the results from the new modi- fied methods of single-molecule magnetic tweezers showed that HyperFi-As reduced the bind- ing capacity and decreased the state of R-loop complex at mismatch sites without compromising on-target binding capacity. All these suggested that HyperFi-As could be a very promising accurate CRISPR-Cas genome-editing tool. The strategy that modifying residues in contact with the backbone of the distal end of the crRNA strand combined with mutating non- specific DNA contacts is very useful for increasing Cas12a specificity. Results Structure-guided protein engineering for high-fidelity AsCas12a Cas12a has been noted as a very prominent tool for gene editing with low off-targeting. Our previous work identified CeCas12a and other variants with stringent PAM recognition to PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 2 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As further lower the off-targeting [36,37]. However, the relatively high off-targeting still existed in the distal end of the PAM. To further investigate the off-targeting in the proximal and distal end of the PAM, AsCas12a was selected since its high-resolution structure has been solved [38]. Firstly, 4 amino acid residues (S186, R301, T315, and Q1014) with direct hydrogen bonds to the phosphate backbone of the target DNA strand within a 3.0-Å distance at the proximal end of PAM were identified (S1 Fig). Among these 4 residues, both R301A and T315A exhib- ited a tendency for reduced cleavage activity at several mismatched sites compared to AsCas12a WT [39]. To study the effects of these mutations systematically, we constructed 4 single amino acid mutants by alanine substitution (S186A/R301A/T315A/Q1014A) and gener- ated all possible double, triple, and quadruple versions by combining these mutations. Then, we employed a previously reported human cell-based enhanced GFP (EGFP) disruption assay to test their off-targeting [36]. The result showed that all 14 variants retained comparable on- target activities to AsCas12aWT when using the fully matched EGFP crRNA (Fig 1A). Next, and characterization of AsCas12a variants bearing substitutions in residues forming nonspecific contacts with target DNA. (A) Fig 1. IdentificationAU : AbbreviationlistshavebeencompiledforthoseusedinFigs1to5:Pleaseverifythatallentriesarecorrect: Characterization of AsCas12a variants containing alanine substitutions in residues that form hydrogen bonds with the target DNA. Assessment of wild-type AsCas12a and variants by using EGFP disruption assay when paired with fully matched crRNA or partially mismatched crRNAs (mean ± SD; n = 3). (B) Average activity of wild-type AsCas12a and mutant variants on targets with optimal PAM TTTV (V = A; G; C) sequence. (C) Activities of wild-type AsCas12a and AsCas12a4m on targets with TTTV PAM sequence across 15 endogenous targets measured by T7 endonuclease I assay (mean ± SD; n = 3). (D) Summary of on-target modifications from (C), with means and 95% confidence intervals shown. The data underlying this figure can be found in S1 Data. EGFP, enhanced GFP; PAM, protospacer-adjacent motif. https://doi.org/10.1371/journal.pbio.3002514.g001 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 3 / 24 ABWild typeS186R301T315AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAQ1014AAsCas12a4m15101520Fully matched crRNAPAM0 20 40 60 80 EGFP disruption（%）1510152015&16 mismatched0 20 40 60 80 1510152017&18 mismatched0 20 40 60 80 1510152019&20 mismatched0 20 40 60 80 1510152021&22 mismatched0 20 40 60 80 1510152016&17 mismatched0 20 40 60 80 1510152018&19 mismatched0 20 40 60 80 1510152020&21 mismatched0 20 40 60 80 1510152022&23 mismatched0 20 40 60 80 1020304050TTTG1TTTG2TTTC1TTTC2TTTA1TTTA2PAMS186R301T315Q1014AAAAAAAAAAAAAAAAAAAAAAAAAAAAPercent modifiedCD020406080Percent modified(%)AAVSCCR5-1CCR5-2DNMT1-1DNMT1-2EMX1FANCF-1FANCF-2HBB-1HBB-2VEGFA-1VEGFA-2B2MCTLA4PD1AsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4m020406080Percent modified(%)PLOS BIOLOGYImproved genome editing specificity by HyperFi-As we sought to evaluate the performance of these variants toward the imperfect crRNA-DNA matching. To do this, we repeated the EGFP disruption assay by combining all variants with EGFP-crRNA expression plasmids that contain 2 substituted bases at positions ranging from 15 to 23 (Fig 1A). We found that the nuclease activity of all triply substituted variants reduced dramatically when 2 adjacent crRNA-DNA base pair mismatches were located at the position from 15 to 19 (reduction ranges from 14.1% to 82.5%) (Fig 1A). Furthermore, the quadruple substitution variant (S186A/R301A/T315A/Q1014A, termed AsCas12a4m) presented the reduced cleavage efficiency ranging from 20.9% to 92.6% compared to AsCas12aWT (Fig 1A). AsCas12a4m retains high on-target activities in human cells Modifying amino acid residues in close contact with the target DNA strand or non-target DNA strand might have various impacts on the cleavage activity at the target site [28]. To sys- tematically determine the activity of AsCas12a4m at endogenous chromosomal sites, we com- pared the activity of AsCas12aWT and all substitution variants (including AsCas12a4m) at 16 endogenous sites using T7E1 (T7 Endonuclease I) assay in 293T cells (Figs 1B, 1C, S1 and S2). We found all variants retained highly comparable activities (90% to 115%) to AsCas12aWT at TTTV (V = A/G/C) PAM sites (Figs 1B and S1), and AsCas12a4m showed at least 90% of the on-target cleavage efficiencies observed with AsCas12aWT at the same sites (Figs 1C, 1D and S2). Based on the results, we selected the AsCas12a4m for the subsequent experiment. crRNA-DNA mismatch tolerance of AsCas12a4m For further investigating crRNA-DNA mismatch tolerance of AsCas12a4m in human cells, we arbitrarily selected 2 endogenous target sites (CFTR and B2M) and constructed a series of plas- mids encoding crRNAs with 1 or 2 mismatches (Fig 2A and 2B). Firstly, we tested the activity of AsCas12aWT and AsCas12a4m guided by crRNAs with single mismatches along the proto- spacer complementarity region. For the CFTR site, AsCas12aWT showed 5.7% to 68.5% cleav- age efficiency at mismatch positions 2, 3, 4, 6, 7, 9, 10, 11, and 13–23 (Figs 2A and S3). For the B2M site, AsCas12aWT presented 6.9% to 38.1% cleavage efficiency at mismatch positions 1, 8, 9, 10, 11, 13–16, and 18–23 (Figs 2B and S4), and did not tolerate single mismatches at posi- tions 2–7 (Figs 2B and S4). Not surprisingly, the results demonstrated AsCas12aWT could mediate DNA cleavage with imperfect crRNA-DNA matching at positions 1 through 23, which were consistent with the previous studies [31]. However, for AsCas12a4m, no tolerance was detected for any single mismatches at positions 1–19 (Fig 2A and 2B). We also tested the editing of AsCas12aWT and AsCas12a4m at CFTR and B2M sites by using crRNAs with 2 adjacent mismatches (Fig 2A and 2B). The analysis showed that AsCas12aWT and AsCa- s12a4m processed nearly undetectable cleavage activities with 2 adjacent mismatches through positions 1–18 for both CFTR and B2M sites (except for AsCas12aWT at B2M 9 and 10 mis- match) (Figs 2A, S3 and S4). By contrast, mismatches (single mismatch or 2 adjacent mis- matches) at positions 20–23 did not substantially reduce the cleavage activities for both AsCas12aWT and AsCas12a4m (Fig 2A and 2B). Overall, AsCas12a4m exhibited high sensi- tivity to crRNA-DNA mismatches in both CFTR and B2M sites especially for the close end region of PAM. The results suggested that AsCas12a4m may enhance the fidelity of gene editing. Genome-wide targeting specificity of AsCas12a4m To evaluate the fidelity of AsCas12a4m at endogenous sites in human cells, we used the GUIDE-seq method to assess the genome-wide specificity of AsCas12aWT and AsCas12a4m with 6 different crRNAs targeted to PD1, B2M, HPRT, DNMT1, CLIC4, and NLRC4. To test PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 4 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Fig 2. Tolerance of AsCas12aWT and AsCas12a4m variant to mismatched crRNAs and genome-wide specificities of AsCas12aWT and AsCas12a4m variant with matched crRNAs targeting endogenous sites. (A, B) Indels induced by AsCas12aWT and AsCas12a4m using crRNAs that contain singly mismatched bases or pairs of mismatched bases toward CFTR (A) and B2M (B). Activity determined by T7 endonuclease I assay. Error bars represent SEM for n = 3. (C) Summary of the total number of off-target sites identified by GUIDE-seq for AsCas12aWT and AsCas12a4m with PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 5 / 24 CFTRMismatched position in spacerTarget strandNon-target strandFully matched1&23&421&22GAGAGAGAGCUGUCCUUGAUACCCUGAGAGAGCUGUCCUUGAUACCGAGAGAGAGCTGTCCTTGATACCTTTACTCTCTCTCGACAGGAACTATGGAAATGACUGAGAGCUGUCCUUGAUACCGAGAGAGAGCUGUCCUUGAUUGCPAM15101520Matched1/23/45/67/89/1011/1213/1415/1617/1819/2021/22Percentmodified(%)020406080020406080CFTRAsCas12aWTAsCas12a4mMismatched position in spacerTarget strandNon-target strandFully matched1223GAGAGAGAGCUGUCCUUGAUACCCAGAGAGAGCUGUCCUUGAUACCGAGAGAGAGCTGTCCTTGATACCTTTACTCTCTCTCGACAGGAACTATGGAAATGUGAGAGAGCUGUCCUUGAUACCGAGAGAGAGCUGUCCUUGAUACGPAM15101520Matched1234567891011121314151617181920212223Percentmodified(%)020406080AsCas12aWTAsCas12a4m020406080AMismatched position in spacer1223Target strandNon-target strandFully matchedCUCACGCUAUCCAGCAGGAAAUGCTCACGCTATCCAGCAGGAAATGTTTAGAGTGCGATAGGTCGTCCTTTACAAATPAM15101520GUCACGCUAUCCAGCAGGAAAUGCACACGCUAUCCAGCAGGAAAUGCUCACGCUAUCCAGCAGGAAAUCPercentmodified(%)020406080Matched1234567891011121314151617181920212223B2MAsCas12aWT020406080AsCas12a4mMismatched position in spacerACUCACGCUAUCCAGCAGGAAAUGCTCACGCTATCCAGCAGGAAATGTTTAGAGTGCGATAGGTCGTCCTTTACAATPAM15101520GACACGCUAUCCAGCAGGAAAUGCUGUCGCUAUCCAGCAGGAAAUGCUCACGCUAUCCAGCAGGAAUAGTarget strandNon-target strandFully matched1&23&421&22Percentmodified(%)020406080020406080Matched1/23/45/67/89/1011/1213/1415/1617/1819/2021/22AsCas12aWTAsCas12a4mB2MBD012345678MismatchesPercentageofreadsNumberofoff-targets0255075100PD1024681000CLIC40255075100140246810HPRT0255075100100246810IL12A0255075100200246810NLRC40255075100240246810DNMT102550751001100246810AsCas12aWTAsCas12a4m0510152520Total numberofoff-targets216AsCas12a4mAA1615AsCas12aWTC0255075100B2M024681000AsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mAsCas12aWTAsCas12a4mPLOS BIOLOGYImproved genome editing specificity by HyperFi-As crRNAs targeting 7 endogenous sites. (D) Percentage of reads detected by GUIDE-seq at the on-target site and off- target sites (ordered by number of mismatches, 0 represented on-target) among total detected reads by AsCas12aWT, AsCas12a4m (top), and numbers of genome-wide off-target sites (bottom). The data underlying this figure can be found in S1 Data. GUIDE-seq, genome-wide unbiased identification of double-stranded breaks enabled by sequencing. https://doi.org/10.1371/journal.pbio.3002514.g002 whether the double-stranded oligodeoxynucleotide (dsODN) tag was integrated into the on- target break site, we performed PCR reactions initiated by 1 primer that specifically annealed to the dsODN and another primer that annealed to the upstream or downstream of on-target break sites (S5 Fig). The dsODN tag was successfully integrated into double-strand breaks (DSBs) induced by AsCas12aWT and AsCas12a4m (S5 Fig). The GUIDE-seq analysis revealed that crRNAs targeting PD1 and B2M generated no unwanted mutations for AsCas12aWT and AsCas12a4m variant. crRNAs targeting HPRT, DNMT1, CLIC4, and NLRC4 generated 1, 10, 4, and 4 off-target sites for AsCas12aWT, respectively, while 0, 1, 1, and 2 off-target sites for AsCas12a4m, respectively (Figs 2C, 2D and S6). GUIDE-seq analysis of AsCas12a4m showed that the number of off-target sites decreased by lower than 3, the on-to off-target read ratio (range 2% to 25%) was improved, and the number of on-target reads still retained, compared to AsCas12aWT (Figs 2D and S6). To validate the off-target sites identified by GUIDE-seq, we confirmed the occurrence of indels using deep sequencing (S7 Fig). All results demonstrated that AsCas12a4m possesses higher fidelity than AsCas12aWT. Improved specificity of AsRVR and enAsHF1 AsRVR and enAsHF1 have been reported to have a broader targeting range than AsCas12aWT [32,39], while altered PAM preferences might induce extra off-targets at the noncanonical PAM sites [36,40]. We sought to graft AsCas12a4m mutations onto AsRVR and enAsHF1 (AsRVR-4m, enAsHF1-4m) and tested whether these alterations would improve targeting specificity. First, we evaluated the editing activities of AsRVR-4m and enAsHF1-4m in endog- enous human genes, and 15 endogenous genomic sites harboring TTTV, TTCV, TATG, TATC, GTTA, CTTA, and GTCA PAMs were selected for T7E1 assay (Figs 3A, 3B and S8). As a result, these mutations did not affect the cleavage activity at all of the tested sites (Fig 3A). Next, we performed crRNA-DNA tolerance assay, according to the previous method (Fig 3C and 3D). Indeed, AsRVR and enAsHF1 exhibited nuclease activity toward crRNA-DNA single mismatch (Figs 3C, 3D and S9), and AsRVR showed more sensitivity to crRNA-DNA mis- match at these sites, compared to enAsHF1 (Figs 3C, 3D and S9). By contrast, AsRVR-4m and enAsHF1-4m caused a significant decrease in the single mismatch cleavage, which was consistent with the above results (Figs 2A, 2B, 3C, 3D and S9). To globally assess the editing specificity of AsRVR-4m and enAsHF1-4m, we performed GUIDE-seq analysis on 3 endogenous sites that were well studied previously [36,40]. As a result, AsRVR-4m has a dramatically lower number of off-target sites at HEK293 site 1, POLQ1, and POLQ2 (9, 0, and 18, respectively), compared with AsRVR (47, 12, and 77, respec- tively). Similarly, the number of off-target sites of enAsHF1-4m at the 3 sites (8, 1, and 22) were lower than those of enAsHF1 (36, 7, and 55, respectively) (Figs 3E and S10). On- to off- target read ratios of AsRVR-4m and enAsHF1-4m at HEK293 site 1, POLQ1, and POLQ2 sig- nificantly increased (57.3%, 100%, and 19.7% for AsRVR-4m and 29.8%, 90.5%, and 19.1% for enAsHF1), compared with AsRVR and enAsHF1(9.6%, 45.6%, 5.8% and 7.5%, 43.6%, 13.3%, respectively) (Fig 3E–3I). We performed deep sequencing to validate the off-targets sites iden- tified by GUIDE-seq (S11 Fig). As expected, the 4m variant has significantly improved the specificity and dramatically reduced the off-target index (0.67 and 0.65 for AsRVR and PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 6 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Fig 3. Refining the specificity of AsRVR and enAsHF1. (A, B) Mean on-target percent modification for wild-type AsRVR and AsRVR-4m, enAsHF1 and enAsHF1-4m with crRNAs across sites encoding TTTV PAMs (A), TTCV PAMs, TATV PAMs, GTTA PAM, CTTA PAM, GTCA PAM (B). (C, D) Indels induced by AsRVR and AsRVR-4m (C), enAsHF1 and enAsHF1-4m (D) using crRNAs that contain singly mismatched bases toward CFTR. Activity determined by T7 endonuclease I assay. Error bars represent SEM for n = 3. (E) Percentage of reads detected by GUIDE-seq at the on-target site and off-target sites (ordered by number of mismatches, 0 represented on-target) among total detected reads by AsRVR and AsRVR-4m, enAsHF1 and enAsHF1-4m (top) and numbers of genome-wide off-target sites (bottom). (F, G) Summary of the total number of off-target sites with TTTN PAMs identified by GUIDE-seq for PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 7 / 24 CFTRAsRVR020406080Matched1234567891011121314151617181920212223AsRVR-4m020406080Percent modified(%)0204060Percent modified(%)PAMTargetgeneVEGFATTTG1TTTG2TTTATTTATTTCCTLA4PD1B2MenAsHF1enAsHF1-4m0204060CFTRTTCCTTCATATG1TATG2TATC1TATC2CTLA4PAMTargetgenePercent modified(%)ABC0204060GTCACFTRTTCCTTCAGTTACTTA2CTTA1DNMT1Percent modified(%)0204060VEGFATTTG1TTTG2TTTATTTATTTCCTLA4PD1B2MPercent modified(%)020406080CFTRenAsHF1Matched1234567891011121314151617181920212223020406080enAsHF1-4mPercent modified(%)DPercentage of readsENumber of off-targetsAsRVRAsRVR-4menAsHF1enAsHF1-4m4710368020406080100POLQ10255075100AsRVRAsRVR-4menAsHF1enAsHF1-4m12017020406080100POLQ20255075100AsRVRAsRVR-4menAsHF1enAsHF1-4m02040608010077185522Number of off-targets02040608010012450HEK293 site 1AsRVRAsRVR-4menAsHF1enAsHF1-4m0204060801004021POLQ1AsRVRAsRVR-4menAsHF1enAsHF1-4m02040608010068144216POLQ2AsRVRAsRVR-4menAsHF1enAsHF1-4mNumber of base substitutions to TTTN PAM sequenceNumber of off-targetsAsRVRAsRVR-4menAsHF1enAsHF1-4m0204060801003563188050020406080100AsRVRAsRVR-4menAsHF1enAsHF1-4m1substitution2substitution3substitution94136020406080100AsRVRAsRVR-4menAsHF1enAsHF1-4m012345678MismatchesHEK293 site 1POLQ1POLQ2Off-target effect indexAsRVRAsRVR-4m0.00.51.010.21enAsHF1enAsHF1-4m0.00.51.010.35Off-target effect indexAsRVRAsRVR-4m0.00.51.00.171enAsHF1enAsHF1-4m0.00.51.010.28FGHIHEK293 site 10255075100AsRVRAsRVR-4mAsRVRAsRVR-4menAsHF1enAsHF1-4menAsHF1enAsHF1-4mPLOS BIOLOGYImproved genome editing specificity by HyperFi-As AsRVR and AsRVR-4m, enAsHF1 and enAsHF1-4m (F) and normalization of off-target effect at TTTN PAMs (value was calculated by the ratio of total off- target sites for AsRVR-4m, enAsHF1-4m to the total off-target sites for AsRVR, enAsHF1 within the detected sites) (G). (H, I) Summary of the total number of off-target sites with base substitutions to the WT-related TTTN PAMs identified by GUIDE-seq for AsRVR and AsRVR-4m, enAsHF1 and enAsHF1-4m (H) and normalization of off-target effect at these sites (I). The data underlying this figure can be found in S1 Data. GUIDE-seq, genome-wide unbiased identification of double-stranded breaks enabled by sequencing; PAM, protospacer-adjacent motif. https://doi.org/10.1371/journal.pbio.3002514.g003 enAsHF1, for at canonical TTTV PAM, 0.83, and 0.72 for AsRVR and enAsHF1 at noncanoni- cal PAM) (Fig 3G and 3I). Effects of residues contacting with the 30 end of crRNA The cleavage activity of AsCas12aWT nuclease was reported to be nearly unaffected by the mismatches in the PAM distal bases between 18 and 23 [9,31]. We also confirmed it by the T7E1 assay (Fig 2A and 2B). To systematically evaluate the effect of the PAM distal bases, we constructed crRNA expression plasmids containing all possible pairs of the mismatched bases between 14 and 23 within the complementarity regions of crRNAs targeted EGFP and repeated EGFP disruption assay. In line with our former experiments (Fig 2A and 2B), the AsCas12a4m variant showed the high sensitivity to mismatched crRNA nucleotides in most places (Fig 4A). However, both AsCas12aWT and AsCas12a4m could mediate enough EGFP disruption when mismatched at the PAM distal positions (Fig 4A). GUIDE-seq analysis for AsCas12aWT and AsCas12a4m were further performed at another well-studied endogenous sites RPL32P3 in 293T cells (Figs 4B and S12). To confirm these GUIDE-seq findings, we used T7E1 assay to assess the frequencies of indel mutations induced by AsCas12aWT and AsCas12a4m at 4 off- target sites (Figs 4C and S12). In detail, at the off-target sites 3 and 4, the dsDNA cleavage events were 15% and 17% for AsCas12aWT, respectively, and nearly not detected for AsCa- s12a4m (Fig 4C). However, both AsCas12aWT and AsCas12a4m showed enough cleavage events at the off-target sites 1 and 2 (23%, 31%, and 20%, 16% cleavage efficiency for AsCa- s12aWT and AsCas12a4m at off-target sites 1 and 2, respectively) (Fig 4C). The results indi- cated that: (1) AsCas12a4m-induced dsDNA breaks with high fidelity; and (2) AsCas12a4m was not sensitive to mismatches positions in the PAM distal regions. To further improve the fidelity in the PAM distal regions, it was initially hypothesized that the unwanted off-targeting with mismatch positions in the PAM distal regions might be mini- mized by decreasing nonspecific interactions with the 30 ends of the guide RNA and the 50 ends of the target DNA. Based on the crystal structure of AsCas12a in complex with crRNA and target DNA [38], we selected 3 residues (K414, Q286, and W382) and 1 residue (S376) that made direct contacts to 30 ends of the guide RNA and the 50 end of the target DNA for ala- nine substitution (Figs 5A and S13A–S13C). Then, the on-target activity of 4 mutants over 6 target sites with TTTV PAM sequence were evaluated. As shown in Fig 5A, single residue sub- stitutions, except for the W382A hardly perturbed the editing efficiencies at target sites. To evaluate the performance of all possible mutant versions, we performed the T7E1 assay. On- target activity and 4 off-targets activities of all mutant versions were evaluated at the RPL32P3 target site in 293T cells. For the on-target site, we observed that the indels caused by AsCa- s12a4m-K414A, AsCas12a4m-S376A, and AsCas12a4m-Q286A mutant were about 32%, 35%, and 31%, respectively, compared to AsCas12aWT, 35% and AsCas12a4m, 33% (Figs 5B and S13D). By contrast, other combinations dramatically decreased the activity (Figs 5B and S13E). In addition, deep sequencing analysis showed that AsCas12a4m-K414A keep highly comparable activities (80% to 100%) to AsCas12aWT (S14 Fig). Indels events did not happen for AsCas12aWT and all As mutants at the off-targeting sites, except AsCas12aWT at the off- target 3 and 4. Furthermore, AsCas12a4m-K414A (termed HyperFi-As) and AsCas12a4m- PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 8 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Fig 4. Performance of AsCas12aWT and AsCas12a4m at PAM-distal mismatches. (A) Cleavage profiles, assessed by EGFP disruption assay for AsCas12aWT and AsCas12a4m at all possible mismatch pairs of bases within the complementarity regions of crRNAs targeted EGFP from 14 to 23. The EGFP disruption (%) are the mean of 3 replicates. (B) The number of potential off-target loci identified by GUIDE-seq analysis for AsCas12aWT and AsCas12a4m at RPL32P3 site. (C) Validation of AsCas12aWT and AsCas12a4m performance at 4 off-targets that identified by GUIDE-seq in panel (B), percent modification of on-target and 4 off-targets of AsCas12aWT and AsCas12a4m assessed by T7E1 assay (from S12 Fig). Error bars represent SEM for n = 3. The data underlying this figure can be found in S1 Data. EGFP, enhanced GFP; GUIDE-seq, genome-wide unbiased identification of double- stranded breaks enabled by sequencing; PAM, protospacer-adjacent motif. https://doi.org/10.1371/journal.pbio.3002514.g004 Q286A were revealed to have the lower off-target rates at the off target 1 and 2 (Figs 5B and S13D). Although further proof of GUIDE-seq analysis indicated that HyperFi-As increased the off- to on-target read ratio at off-target 1, the number of total off-target sites decreased by 13 compared to AsCas12a4m-Q286A (Figs 5C, 5D, S12C, and S12D). We further selected SIPRa, a well-studied target site to perform GUIDE-seq analysis for specificity assessment [41]. The result demonstrated that targeting SIPRa showed off-target activity for HyperFi-As at 3 loci, which was less in number than 49, 9, and 7 for AsCas12aWT, AsCas12a4m, and AsCas12a4m-Q286A, respectively (Fig 5E and 5F). In addition, statistics of the GUIDE-seq reads revealed that HyperFi-As possessed the lowest percentage of off-targets with mismatches in the PAM distal region among As, AsCas12a4m, HyperFi-As, and AsCas12a4m-Q286A (Fig 5F). Systematical comparison of AsCas12a variants and SpCas9 variants To systematically compare the specificity of HyperFi-As with other recent high-fidelity AsCas12a variants and SpCas9 variants, we comprehensively assessed these variants’ capability to reduce genome-wide off-target effects of gRNAs designed against target sites with a large number of homopolymeric sequences in human cells. For the PPP1R13L site, HyperFi-As exhibited a reduced number of off-target sites compared to As variants, and the percentage of on-target was also the highest among variants (Figs 5G and S15). To further confirm it and compare the genome-wide specificities of CRISPR-Cas nucleases including AsCas12aWT, As ultra, As plus, SpCas9, and SuperFi-SpCas9, we repeated GUIDE-seq with 5 gRNAs targeting PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 9 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Fig 5. Generation of high-fidelity derivatives of AsCas12a4m by weakening nonspecific crRNA contacts. (A) Characterization of high-fidelity derivatives of AsCas12a4m containing alanine substitutions in residues that form hydrogen bonds with the PAM-distal region (30 end of the guide RNA and 50 end of the target DNA). Assessment of all variants by using T7E1 assay when paired with 6 crRNAs that targeted DNMT1 and VEGFA loci. (B) Cleavage profiles, assessed by T7E1 assay for As variants at on-target and 4 off-targets. The percent modification (%) are the mean of 3 replicates. (C) The number of potential off-target loci identified by GUIDE-seq analysis for As variants at RPL32P3 site. The statistical results for AsCas12aWT and AsCas12a4m are derived from Fig 4B (see S12 Fig). (D) Specificity ratios of AsCas12aWT, AsCas12a4m, HyperFi-As, and AsCas12a4m+Q, plotted as the ratio of 4 off-target sites GUIDE-seq read counts to on-target. (E) Comparative analysis of AsCas12aWT, AsCas12a4m, HyperFi-As, and AsCas12a4m+Q with crRNA targeting SIPRa loci using GUIDE-seq. (F) Percentage of edited reads detected by GUIDE-seq at on-target site and mismatched sites among total edited reads and the numbers of off-target sites for each AsCas12a. (G) Comparative analysis of AsCas12aWT, HyperFi-As, As ultra, and As plus with crRNA targeting PPP1R13L loci using GUIDE-seq, percentage of reads detected by GUIDE-seq at the on-target site and off-target sites (ordered by number of mismatches, 0 represented on-target) among total detected reads by As variants (top) and PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 10 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As numbers of genome-wide off-target sites (bottom). (H) Genome-wide specificities of AsCas12a and SpCas9 variants. Summary of the total number of genome-wide off-target cleavage sites identified by GUIDE-seq for AsCas12a and SpCas9 variants with gRNAs targeted to nonstandard, repetitive sites. The data underlying this figure can be found in S1 Data. GUIDE-seq, genome-wide unbiased identification of double-stranded breaks enabled by sequencing; PAM, protospacer-adjacent motif. https://doi.org/10.1371/journal.pbio.3002514.g005 sites containing overlapping sequences (HEK293 site 2/3/4/5/6) (Figs 5H and S16–S20). For 5 sites, the high-fidelity variants such as As plus and SuperFi-SpCas9 exhibited improved speci- ficity (Fig 5H), which was consistent with previous reports [33,35,42,43]. However, no on-tar- get gene editing events were detected at the HEK293 site 2/4/6 for SuperFi-SpCas9 (S16, S18 and S20 Figs). Notably, HyperFi-As showed the best specificity among the 6 tested variants of AsCas12a and SpCas9 (Figs 5H and S15–S20). Binding stability and multiple state characteristics of HyperFi-As To explore whether HyperFi-As has better specificity to distinguish the mutant DNA sequence and thus possibly decreases the off-target efficiency, we compared the stability of DNA- Cas12a-crRNA complexes that containing the HyperFi-As and AsCas12aWT on various mutant DNA sequences using single-molecule DNA unzipping experiments, according to the previous study for determining the stability of DNA-Cas12a-crRNA [44]. However, a proper constant force was used instead for sensitive detection. In the end, the life-time of the interme- diate states during the disassembly sensitively detects multiple states by this modified new way. We constructed a series of 34-bp DNA hairpins containing the PAM and various target/off- target sequences (Figs 6A and S21A) as the substrates. We stretched the hairpin between a paramagnetic magnetic bead and a glass slide through 2 dsDNA handles. In the force-increas- ing scan at the loading rate of 1 pN/s, the naked DNA hairpin unzipped at approximately 18 pN, while a high force of approximately 30 pN was required to completely unzip the DNA- Cas12a-crRNA complex (S21B Fig). Note that we used the dead mutants of AsCas12a without cleavage activity and the DNA hairpins were protected. To determine the stability and possible multiple states of each kind of DNA-Cas12a-crRNA complex, we elevated the force from 5 pN to 26 pN and held the force at 26 pN for 120 s (Fig 6B). The measurements were repeated at least 20 times for each molecule, and the results from multiple molecules were yielded. The transient R-loop complex state as the first of 3 states was sensitively detected and 2 steps was illustrated to completely unzip the hairpin that bound crRNA-Cas12a (Fig 6B). Based on the increases in DNA extension, the first step represents the dissociation of the non-target sequence from the R-loop complex, and the second step repre- sents the unzipping of PAM (Fig 6C). The difference between states 1, 2, and 3 is the length of the non-target sequence bound to Cas12a. In state 1, the Cas12a-crRNA-DNA complex forms a tight binding, encompassing both the target sequence and the non-target sequence. State 2 represents a partial dissociation of the Cas12a-crRNA-DNA complex, with the non-target strand upstream of the PAM being released from the R-loop complex. In state 3, the hairpin downstream of the PAM, including the PAM sequence, unfolds under force and is released from the Cas12a-crRNA complex. The transition from state 1 to state 2 indicates the non-tar- get strand upstream of the PAM moving away from the R-loop complex while the transition from state 2 to state 3 indicates the unfolding of the hairpin strand downstream of the PAM including the PAM sequence. The stabilities of the complexes that had the same crRNA but with 3 different DNA sequences (the WT target and 2 real off-target DNA subtracts) were compared. When bound to the on-target, the instability of DNA-Cas12a-crRNA were similar for both HyperFi-As and AsCas12aWT. However, when bound to the off-target, the instability PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 11 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Fig 6. Single-molecule assay for detection of cas12a complex stability. (A) Sketch of the representation of the experimental setup. A DNA hairpin containing target sequence was attached between the functionalized coverslips and a magnetic bead. As the disassociation of crRNA-Cas12a complex from the target DNA, the extension of DNA molecular stretches with the complete unzip of the hairpin. (B) Representative extension trace from the DNA hairpin in the presence of the dCas12a–crRNA complex. States 1, 2, 3 mean the different extension state of the DNA hairpin. (C) Cartoon of Cas12a complex in different states. (D) Histogram of normalized unbinding events probability on truly mismatch DNA target. (E) Histogram of times in different states on truly mismatch DNA target. The data underlying this figure can be found in S1 Data. https://doi.org/10.1371/journal.pbio.3002514.g006 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 12 / 24 FZ1Z2thiolmaleimidestreptavidinbiotinGlassFABCDE020406080state1state2state3***Time (s)AsCas12aWTHyper Fi-AsOff a/b020406080state1state2state3***Time (s)AsCas12aWTHyper Fi-AsOff cstate3crRNAtargetFFstate1PAMcrRNAtargetFFnon-targetstate2PAMcrRNAtargetFFnon-targetTTTGGGTGATCAGACCCAACAGCGTTTGGGTGATCAGACCCAACACCGTTTGGGTGATCAGACCCAACACCGTTTGGGTGATCAGACCCCCCCCCGPAMOff aOff bOff cAGGAGGAGGAGGCCACUAGUCUGGGUUGUCGCUCCcrRNA targeted RPL32P3On targetchrosome 3chrosome 2chrosome 8chrosome 15AsCas12aWTHyper Fi-AsOnOff a/bOff c0.00.20.40.60.8Normalizedpercentageofunbinding*102030360400440050100150200360400440F(pN)HyperFi-AsCas12aAsCas12a WTExtension(nm)state1state2state3Z1Z2state1state2state3Z2Z1Time(s)PLOS BIOLOGYImproved genome editing specificity by HyperFi-As was more affected and increased a lot for HyperFi-As (Fig 6D), which was consistent with the result that HyperFi-As had the similar gene editing efficiency at on-target sites but a lower tol- erance for off-target sites than AsCas12aWT. Detailed analysis of the unzipping assay revealed that HyperFi-As was largely less distributed at the first state when bound to the off-target sites (Fig 6E), which might suggest more dissociations of the non-target sequence from the R-loop complex for HyperFi-As. All these shed important insights on how HyperFi-As could behave in the single-molecule level and lead to low off-targeting. Discussion Off-target effects of CRISPR-Cas gene-editing tools pose a challenge for therapeutic applica- tions. Although AsCas12a has been shown to be an innately highly specific enzyme, it still causes off-target effects in mammalian cells [31,33,36]. Off-target effects can occur at levels 10 or even 100 times lower than the target site, but these small DNA breaks can cause chromo- some breaks and thus chromosomal translocations [37]. Thus, the less off-target we can get, the more safety we can have in terms of therapies. Genome-wide unbiased identification of DSBs by sequencing (GUIDE-seq) is one of the most popular methods to detect real cleavage sites in living cells after genome editing [13]. It is straightforward, has good sensitivity, and reflects the real situations in living cells across the entire genome unbiased. It is therefore well accepted and widely used in many off-targeting studies for Cas9 and Cas12a systems [45–47]. Using GUIDE-seq method, we and other researchers have shown that gene editing in mamma- lian cells using the Cas12a system causes unwanted mutations, although less so compared to the Cas9 system [9,11,12,31,36,48,49]. Using an in vivo EGFP disruption assay, we further clar- ified the off-target effect of AsCas12aWT in the presence of double base pair mismatches. As mentioned above, we observed significant fluorescence reduction when AsCas12aWT was guided by certain mismatched crRNA, particularly when mismatches were located at the distal end of the PAM sequence. To minimize these off-target effects of AsCas12aWT, we engineered a novel AsCas12a variant termed HyperFi-As that shows high gene editing accuracy without compromising on-target efficiency. A strategy of decreasing nonspecific interactions with its target DNA strand has been suc- cessful for high-fidelity Cas9 [15,29]. Since Cas12a is relatively high in fidelity, it was unclear if the strategy could work well for Cas12a in further improving the fidelity. Thus 4 key residues (S186, R301, T315, and Q1014) was identified by analyzing the crystal structure of the AsCas12a-crRNA-target DNA complex. All of them made direct hydrogen bonds to the phos- phate backbone of the target DNA strand within a 3.0-Å distance to provide nonspecific con- tacts. We constructed single amino acid mutant by alanine substitution (S186A/R301A/ T315A/Q1014A) and generated a range of possible combinations and demonstrated that the quadruple substitution variant (termed AsCas12a4m) has the lowest mismatch tolerance by both EGFP disruption assay as well as single mismatch and double mismatch cleavage experi- ments on endogenous genes. As expected, AsCas12a4m showed a decrease in cleavage activity compared to AsCas12aWT at off-target sites, both in terms of the number of off-target sites and in terms of read counts at the sites, as validated by GUIDE-seq analysis across several endogenous loci. Recent studies revealed that the alteration of contacts between PAM proximal DNA and amino acid residues in the PI domain of AsCas12a can expand the range of PAM recognition [32,39]. For instance, AsCas12a variants AsCas12a-RVR and enAsCas12a-HF1 modify DNA efficiently at TATV, and TTYN (TTTN/TTCN), VTTV (ATTV/CTTV/GTTV), TRTV (TATV/TGTV) noncanonical PAMs, respectively. However, our previous report indicated that the noncanonical PAM recognition by Cas12a might induce extra off-target edits [36,37]. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 13 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As And the side effect was also demonstrated by other studies [36,40]. In this work, we success- fully grafted mutations from AsCas12a4m onto AsRVR and enAsCas12a-HF1 and engineered alternative high-fidelity AsRVR-4m and enAsHF1-4m, which preserved the recognition of nonclassical PAMs. EGFP disruption assay and cleavage experiments on endogenous sites suggested that one- base mismatches or two-base mismatches at the position of the PAM-distal region were highly tolerant. And truncation of the 30 ends of the guide RNA could not decrease the undesired mutagenesis at some mismatch sites or at least, so the approach limited, as most off-target sites harbor mismatches in the PAM-distal 30 end [9,50]. Although AsCas12a4m exhibits low toler- ance for mismatches between crRNA and target DNA in other regions, it remains insensitive to mismatches located at the termed promiscuous region (positions 19–23). This observation can be explained by the fact that S186 and Q1014/T315/ R301 form hydrogen bonds with the DNA phosphate backbone of the seed region (positions 1–6) and trunk region (positions 7–18), respectively. We speculated that disruption of nonspecific interactions in the promiscu- ous regions may also contribute to fidelity. Based on the hypothesis, we finally identified that K414A mutation led to an enhancement of specificity in the PAM-distal end and conferred a higher fidelity to AsCas12a4m by altering the contact with the distal 30 end of crRNA (termed HyperFi-As). In addition, through the novel single-molecule DNA unzipping experiments demonstrating the binding of DNA-Cas12a-crRNA complex, HyperFi-As mutant significantly reduced the stability of the complexes and decreased the R-loop state when containing the off- target DNA substrate. The new modified single-molecule unzipping assay capable of detecting different state distributions on the disassembly of DNA-Cas12a-crRNA complexes could be widely used in future various studies and could be very helpful for deep insights of how CRISPR/Cas stepwisely heaved and guiding various engineering. In general, the longer the Cas enzyme and crRNA complex are present in the cell, the greater the likelihood of off-targeting. With virus or plasmid-based delivery, the cellular genome is exposed to the Cas enzyme and crRNA complex for an extended period, which results in an increased risk of off-targeting [51]. Genome editing using Cas12a with RNP or mRNA delivery has been successfully applied to limit the duration of exposure of CRISPR/Cas systems in cells [52,53]. Therefore, introducing the HyperFi-AsCas12a system into cells via RNP or mRNA may further reduce off-target cleavage events. Materials and methods Plasmids construction All plasmids and guide RNAs used in this study can be found in S1 Table. AsCas12a human expression plasmid pY010 was purchased from the nonprofit plasmid repository Addgene (Addgene plasmids #69982). All AsCas12a variants were generated by standard site-directed mutagenesis. In brief, using AsCas12a plasmid pY010 as the template and a pair of primers that one primer carrying site mutation nucleotide to amplify 500 bp fragments by PCR (Phanta MAX Super-Fidelity DNA Polymerase P505, Vazyme). After confirming that the bands were correct by agarose gel electrophoresis, the PCR products were purified. Next, taking 1,000 ng PCR products as the circling mutation primers and 100 ng AsCas12a plasmid pY010 as tem- plate to carry out PCR again. PCR products were purified and use DpnI to digest the original template at 37˚C for 1 h, inactivated at 80˚C for 20 min, take 10 μl digested products for trans- formation. The next day, single colonies were sent for sequencing. Oligonucleotide duplexes corresponding to spacer sequences were PCR amplified and cloned into pU6-As-crRNA plas- mids (Addgene plasmids #78956) for U6 promoter-driven U6 promoter-driven transcription of As crRNAs (ClonExpress II One Step Cloning Kit C112, Vazyme). PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 14 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Cell culture and transfection The HEK293T cells maintained in DMEM medium supplemented with 10% fetal bovine serum and 100 units mL−1 penicillin, 100 μg mL−1 streptomycin sulfate (all cell culture prod- ucts were obtained from Gibco) at 37˚C in 5% CO2. For plasmid-based genome editing, approximately 1.2 × 105 cells were seeded into per well of 24-well plate a day before transfec- tion, 600 ng Cas12a and 300 ng crRNA expression plasmids were transfected into cells using Hieff Trans Liposomal Transfection Reagent (CAT:40802ES03, Yeasen, Shanghai). EGFP disruption For EGFP disruption analysis in HEK293T cells, approximately 2.4 × 104 cells were seeded into per well of a 96-well plate a day before transfection, a total of 100 ng of Cas12a plasmid, 30 ng crRNA expression plasmids, and 30 ng EGFP expression plasmids were transfected into HEK293T cells. A U6 promoter-driven empty plasmid for the substitution of crRNA expres- sion plasmid as a negative control, and 48 h post-transfection, cells were analyzed on the Cyto- FLEX (Beckman Coulter). Assessment of gene editing by T7E1 Approximately 48 h post-transfection, cells were collected by centrifugation and the superna- tants were removed. The 50 μl Lysis buffer and 0.5 μl Proteases were mixed with cells and incu- bated at 55˚C for 30 min, 95˚C for 30 min (Animal Tissue Lysis Component, CAT: 19698ES70, Yeasen, Shanghai). The genomic region flanking the CRISPR target site for each gene was amplified by PCR with Phanta MAX Super-Fidelity DNA Polymerase P505 (Vazyme) using 1 μl cell lysis as template and the primers listed in S1 Table. PCR products were purified and the concentrations were determined; 250 ng of purified PCR products were mixed with 1 μl 10×T7E1 buffer (Vazyme) and ultrapure water to a final volume of 10 μl, and subjected to a re-annealing process to enable heteroduplex formation: 95˚C for 3 min, 95˚C for 30 s, 90˚C for 30 s, 85˚C for 30 s, 80˚C for 30 s, 75˚C for 30 s, 70˚C for 30 s, 65˚C for 30 s, 60˚C for 30 s, 55˚C for 30 s, 50˚C for 30 s, 45˚C for 30 s, 40˚C for 30 s, 35˚C for 30 s, 30˚C for 30 s, and 25˚C for 1 min. After re-annealing, products were treated with T7 Endonuclease I (EN303-01/02, Vazyme) for 15 min at 37˚C. The reaction mixtures were run on 2% agarose gels and imaged with ChemiDoc XRS+ and analyzed according to strip intensities. Indel per- centage was determined by the formula: 100 × (1 –sqrt (b + c)/(a + b + c)), where a is the inte- grated intensity of the undigested PCR product and b and c are the integrated intensities of the cleavage product. GUIDE-seq Briefly, 600 ng Cas12a, 300 ng crRNA expression plasmids, and 10 pmol of the dsODN GUIDE-seq tag were transfected into a 24-well 293T cells (1.2 × 105 cells per well); 48 h after transfection, the genomic DNA was harvested and purified using FastPure Cell/Tissue DNA Isolation Mini Kit DC102 (Vazyme). GUIDE-seq tag integration percentages and on- target modification were assessed by restriction-fragment length polymorphisms (RFLPs) assays and T7E1 assays (as described above), respectively. A total of 1,000 ng genomic DNA was fragmented, end repaired, A-tailing by using Hieff NGS Fast-Pace DNA Fragmentation Reagent (12609ES96, Yeasen, Shanghai). The sequencing library was prepared and sequenced on an Illumina Instrument and data was analyzed using guideseq v1.1 as described previously. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 15 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Western blotting To detect the expression of As variants, cells were collected and lysed after 48 h transfection. Lysates were resolved through SDS-PAGE electrophoresis and transferred onto a polyvinyli- dene fluoride membrane (Millipore, United States of America). Membranes were blocked by blocking buffer (5% non-fat milk in Tris buffered saline with Tween20 (TBST)) for 2 h. Blots were incubated withAU : PleasecheckwhetherthechangesmadeinthesentenceBlotswereincubatedwithHAMouseprimaryantibody:::arecorrect: dilution and β-actin Rabbit primary antibody (CAT#AC026, ABclonal, China) at 1:20,000 dilution for 3 h at room temperature, respectively. After washing steps in TBST, membranes were incubated for 1 h with HRP Goat anti-Mouse IgG (H+L) for HA (CAT#AS003, ABclonal, China) and HRP Goat anti-Rabbit for β-actin (CAT#AS014, ABclonal, China) at 1:50,000 dilu- tion, respectively. HA Mouse primary antibody (CAT#901515, Biolegend, USA) at 1:20,000 Preparation of DNA samples for MT experiments The hairpin region contains the PAM and protospacer in the center, schematic of the single- molecule substrate is available in S21C Fig. All oligonucleotides were purchased from Sangon Biotech. The tethered DNA substrate was a molecule comprised of a 34-bp hairpin region, a 5-nucleotide loop, and 2 dsDNA han- dles, 630 bp and 653 bp, allowing attachment to the glass coverslip and the magnetic bead, respectively. The main step of hairpin construction is showed as follow: 1. We amplified the 653-bp dsDNA3 (6371–7001 bp of λ-DNA) and the 630-bp dsDNA4 (19470–20100 bp of λ-DNA) as hairpin handles through PCR using hairpinF1/hairpi- n_upR3 and hairpin_downF3/hairpin_R1 as primers, respectively. We then amplify the dsDNA1 containing a part of short hairpin sequences by PCR using template1 (6371–7001 bp of lambda DNA amplified with hairpin_F1 and hairpin_upR1 as primers) and the hair- pin_F1, hairpin_upR2 primers. Amplify the dsDNA2 containing the other part of short hairpin sequences by PCR using template2 (19471–20100 bp of lambda DNA amplified with hairpin_downF1 and hairpin_R1 as primers) and the hairpin_downF2, hairpin_R1 primers. 2. The PCR products DNA1 and DNA2 were purified using a Universal DNA Purification Kit (OMEGA Biotech), digested with the restriction enzyme BsaI (NEB), and purified again. 3. Ligation of digested dsDNA1 and dsDNA2 by T4 ligase and form dsDNA5 containing the complete hairpin sequence. 4. Amplify the ssDNA 1 containing the hairpin sequences through OSP using the dsDNA5 as template and hairpin-biotin primers. Amplify the ssDNA 2 through OSP using the dsDNA3 as template and hairpin_upR3 primers. Amplify the ssDNA 3 through OSP using the dsDNA4 as template and hairpin_sh primers. 5. Anneal above 3 ssDNA strands together equimolarly through a process containing a 1-h incubation step at 65˚C followed by a 1-h slow cooling process from 65 to 30˚C. We synthesized all the oligos with the following sequences (Sangon Biotech). hairpin-biotin: bio-ATTTACGCCGGGATATGTCAAGC hairpin_F1: ATTTACGCCGGGATATGTCAAGCCGAAGCATGAAGTG hairpin_upR1: TCTGATGGTCCATACCTGTTACACTGCCTGAATGCAGCCATAGGTGC hairpin_upR2: TGAGGTCTCAAGAAAAACTCACGACGCTTTCTGATGGTCCATACCTGTT PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 16 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As hairpin_upR3: CTGAATGCAGCCATAGGTGC hairpin-sh: SH-AGTCAGTTGCATCAGTCACAAGGG hairpin_R1: AGTCAGTTGCATCAGTCACAAGGG hairpin_downF3: CAGGTATACAGATTAATCCGGC hairpin_downF1: CATACCTGTTACACTGCCTGAATGCCAGGTATACAGATTAATC CGGC hairpin_downF2: TGAGGTCTCATTCTCACGACGCTTTCTGATGGTCCATACCTGTT ACACTGCCTGA MT experiments We used a homemade magnetic tweezers setup to stretch individual DNA molecule, which was described previously [54]. We functionalized the coverslips with (3-aminopropyl) triethoxy silane (APTES, Sigma-Aldrich), which allowed the 50-thiol end of each DNA mole- cule to attach to the amine group of APTES via Sulfo-SMCC crosslinker (Hunan Huteng). We attached streptavidin-coated paramagnetic beads (Dynal M270, Thermo Fisher Scientific) to the 50-biotin end of the DNA molecules. We mixed crRNA and Cas12a at 37˚C for 20 min in advance and added 10 nM Cas12a-crRNA complexes into the flow cell to form DNA-Cas12a- crRNA complexes. Magnetic tweezers measurements were collected at room temperature (21 to 23˚C) in 10 mM Tris-HCl (pH 7.5), 150 mM KCl, and 0.1 mg/ml BSA buffer. Statistics analysis and reproducibility All statistical analyses were performed using GraphPad Prism (v.8.2.1). The exact replication numbers are indicated in the figure legends. The reproducibility was shown by performing at least 2 independent biological replicate experiments. Supporting information S1 Fig. Identification and characterization of AsCas12a engineered variants. (A–E) Struc- tural representations of AsCas12a-crRNA-DNA complex. In structural representations, amino acid residues (S186, R301, T315, and Q1014) that made direct hydrogen bonds to the phos- phate backbone of the target DNA strand within a 3.0-Å distance. Boxes indicate regions shown in detail in B–E. Images generated from PDBID:5B43 (ref. [38]) visualized in PyMOL (v 1.8.6.0). (F) Activities of AsCas12a engineered variants bearing amino acid substitutions when tested against 7 endogenous sites in human cells. Activities assessed by T7E1 assay. (TIF) S2 Fig. Activities of wild-type AsCas12a and AsCas12a4m in human cells. Comparative analysis of wild-type AsCas12a and AsCas12a variant (AsCas12a4m) with 16 sgRNAs targeting 11 genes. Full gel images of Fig 1C. Three independent transfection replicates were done, and activities assessed by T7E1 assay. (TIF) S3 Fig. Tolerance of AsCas12aWT and AsCas12a4m variant to mismatched crRNAs target- ing CFTR. (A) Full gel image for AsCas12aWT activities with single mismatched crRNAs (top) and double mismatched crRNAs (bottom) toward CFTR. (B) Full gel image for AsCa- s12a4m activities with single mismatched crRNAs (top) and double mismatched crRNAs (bot- tom) toward CFTR. Three independent transfection replicates were done, and activities assessed by T7E1 assay. Arrows indicates cleavage products. (EPS) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 17 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As S4 Fig. Tolerance of AsCas12aWT and AsCas12a4m variant to mismatched crRNAs target- ing B2M. (A) Full gel image for AsCas12aWT activities with single mismatched crRNAs (top) and double mismatched crRNAs (bottom) toward B2M. (B) Full gel image for AsCas12a4m activities with single mismatched crRNAs (top) and double mismatched crRNAs (bottom) toward B2M. Three independent transfection replicates were done, and activities assessed by T7E1 assay. Arrows indicates cleavage products. (TIF) S5 Fig. Examining the integration of GUIDE tag with Tag-specific PCR. (A) Schematic of the Tag-PCR and Genome-PCR. Red arrows indicate genome-specific primers; blue arrows indicate tag-specific primers (Tag-F/R). (B) Full gel images of Tag-PCR and Genome-PCR. Red arrow bands indicate the PCR products using the genome-specific primers (B2M-F/R) with tag-specific primers (Tag-F/R). Blue arrow bands indicate the PCR products using the genome-specific primers B2M-F and B2M-R. Con means transfecting without As variants plasmids. (TIF) S6 Fig. Specificity of AsCas12aWT and AsCas12a4m in human cells. (A) Off-target sites for AsCas12aWT and AsCas12a4m with 6 crRNAs targeting 6 endogenous sites (IL12A, B2M, PD1, CLIC4, NLRC4, DNMT1), determined using GUIDE-seq in HEK293 cells. (B) Off-target sites for CeCas12a with 4 crRNAs targeting 4 endogenous sites (B2M, PD1, CLIC4, NLRC4), determined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S7 Fig. Deep sequencing validation the off-target sites identified by GUIDE-seq. Percent modification of on-target and GUIDE-seq detected off-target sites with indel mutations for AsCas12aWT and AsCas12a4m towards CLIC4 (A), DNMT1 (B), and IL12A (C). Mismatched positions within the spacer or PAM are highlighted in red. Indel frequency assessed by deep sequencing. Error bars represent SEM for n = 2. The data underlying this figure can be found in S1 Data. (EPS) S8 Fig. Activities of enAsHF1, AsRVR and enAsHF1-4m, AsRVR-4m in human cells. (A) Activity analysis of enAsHF1, AsRVR and enAsHF1-4m, AsRVR-4m with 5 crRNAs targeting 4 genes (VEGFA, B2M, PD1, CTLA4) at TTTV PAMs. Full gel images of Fig 3A and 3B, activ- ity analysis of enAsHF1, AsRVR and enAsHF1-4m, AsRVR-4m with 10 crRNAs targeting 3 genes (CFTR, CTLA4, DNMT1) at nonclassical PAMs. Full gel images of Fig 3B. Two indepen- dent transfection replicates were done, and activities assessed by T7E1 assay. (TIF) S9 Fig. Tolerance of enAsHF1, AsRVR and enAsHF1-4m, AsRVR-4m variants to mis- matched crRNAs targeting CFTR. (A) Full gel images for enAsHF1 and enAsHF1-4m activi- ties with single mismatched crRNAs toward CFTR. (B) Full gel images for AsRVR AsRVR-4m activities with single mismatched crRNAs toward CFTR. Three independent transfection repli- cates were done for enAsHF1, AsRVR and enAsHF1-4m and 2 independent transfection repli- cates were done for AsRVR-4m, activities assessed by T7E1 assay. Arrows indicates cleavage products. (TIF) S10 Fig. Specificity of AsRVR, AsRVR-4m, enAsHF1, enAsHF1-4m, and AsCas12a4m in human cells. Off-target sites for As variants with crRNAs targeting POLQ target 1, POLQ PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 18 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As target 2, and HEK293 site 1 loci, determined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S11 Fig. Deep sequencing validation the off-target sites identified by GUIDE-seq. (A) On- target cleavages of these variants at 3 endogenous sites evaluated by T7 endonuclease I assay (gel images). (B) Percent modification of GUIDE-seq detected off-target sites with indel muta- tions for enAsHF1, AsRVR, enAsHF1-4m, and AsRVR-4m. Mismatched positions within the spacer or PAM are highlighted in red. Indel frequency assessed by deep sequencing. Error bars represent SEM for n = 2. The data underlying this figure can be found in S1 Data. (TIF) S12 Fig. Specificity of AsCas12aWT, AsCas12a4m, AsCas12a4m+K, AsCas12a4m+Q vari- ants in human cells. (A–D) Off-target sites for As variants with crRNA targeting RPL32P3 loci, determined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (E) Sequences of 4 off-target sites. (F) Full gel images for AsCas12aWT and AsCas12a4m activ- ities toward RPL32P3 on-target site and 4 off-target sites. (TIF) S13 Fig. Identification and characterization of high-fidelity derivatives of AsCas12a4m. (A, B) Structural representations of AsCas12a-crRNA-DNA complex. In structural representa- tions, amino acid residues (K414, S376, and Q286) that made direct hydrogen bonds to the phosphate backbone of the PAM-distal region (30 end of the guide RNA and 50 end of the tar- get DNA). Boxes indicate regions shown in detail in B. Images generated from PDBID:5B43 (ref. [38]) visualized in PyMOL (v 1.8.6.0). (C) Expression of As variants in HEK293T cells. (D, E) Full gel images of AsCas12a4m variants cleavage profiles, 3 independent transfection replicates were done, and assessed by T7E1 assay at on-target and 4 off-targets. (TIF) S14 Fig. Evaluation of the activity of AsCas12aWT, AsCas12a4m, and HyperFi-As. (A) Val- idation of HyperFi-As performance at 24 endogenous target sites with TTTV (V = A, C, or G) PAMs in HEK293T cells. (B) Schematic of the deep sequencing library constructs. Red arrows indicate genome-specific primers with Illumina P5 and P7 adapter sequences. PCR library is about 280 bp. (C) Full gel images of AsCas12aWT, AsCas12a4m, and HyperFi-As deep sequencing libraries. Red box indicated the library band of each endogenous target sites. (D) Comparison of the activity of AsCas12aWT, AsCas12a4m, and HyperFi-As in HEK293T cells. Each dot represents a target site. Indel frequency assessed by deep sequencing. The median and interquartile range are shown. (E) Evaluation of the activity of AsCas12aWT and HyperFi-AsCas12a with different amounts of plasmid. The data underlying this figure can be found in S1 Data. (TIF) S15 Fig. Specificity of AsCas12aWT, As ultra, As plus, and HyperFi-As in human cells. Off-target sites for As variants with crRNA targeting PPP1R13L loci, determined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S16 Fig. Specificity comparison of SpCas9 and AsCas12a variants targeting nonstandard, repetitive sites. Off-target sites for As variants with crRNA targeting HEK293 site 2 loci, PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 19 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As determined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S17 Fig. Specificity comparison of SpCas9 and AsCas12a variants targeting nonstandard, repetitive sites. Off-target sites for As variants with crRNA targeting HEK293 site 3 loci, deter- mined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S18 Fig. Specificity comparison of SpCas9 and AsCas12a variants targeting nonstandard, repetitive sites. Off-target sites for As variants with crRNA targeting HEK293 site 4 loci, deter- mined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S19 Fig. Specificity comparison of SpCas9 and AsCas12a variants targeting nonstandard, repetitive sites. Off-target sites for As variants with crRNA targeting HEK293 site 5 loci, deter- mined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S20 Fig. Specificity comparison of SpCas9 and AsCas12a variants targeting nonstandard, repetitive sites. Off-target sites for As variants with crRNA targeting HEK293 site 6 loci, deter- mined using GUIDE-seq in HEK293 cells. Mismatched positions are highlighted in color, and GUIDE-seq read counts are shown to the right of the on- or off-target sequences. (EPS) S21 Fig. Schematic of the single-molecule assay. (A) The magnetic tweezers setup with the constructed product involving the designed DNA hairpin was tethered between the beads and glass, the DNA hairpin consists a PAM sequence followed by the target sequence of dCas12a- crRNA complex. (B) Stretching FECs from the DNA hairpin in the absence (mid) and pres- ence of the dCas12a–crRNA complex (down). (C) Schematic representation of hairpin DNA construction for single-molecule experiments. (EPS) S1 Data. Individual numerical values of Figs 1–6, S7, S11 and S14. (XLSX) S1 Table. List of sequences used in study. (DOCX) S1 Raw Images. Extended data figures of uncropped gels. (PDF) Acknowledgments We thank all the members of our laboratory for the fruitful discussions and support. Author Contributions Conceptualization: Peng Chen, Jin Zhou, Lei Yin. Funding acquisition: Peng Chen, Jin Zhou. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 20 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Investigation: Peng Chen, Jin Zhou, Huan Liu, Erchi Zhou, Hongjian Wang, Zaiqiao Sun, Chonil Paek, Jun Lei. Methodology: Peng Chen, Jin Zhou, Huan Liu, Erchi Zhou, Yankang Wu, Hongjian Wang, Zaiqiao Sun, Chonil Paek, Jun Lei, Yongshun Chen, Xinghua Zhang. Supervision: Yongshun Chen, Xinghua Zhang, Lei Yin. Validation: Huan Liu, Erchi Zhou, Boxiao He. Visualization: Erchi Zhou, Yankang Wu. Writing – original draft: Peng Chen, Huan Liu. Writing – review & editing: Peng Chen, Jin Zhou, Lei Yin. References 1. Jinek M, Chylinski K, Fonfara I, Hauer M, Doudna JA, Charpentier E. A programmable dual-RNA- guided DNA endonuclease in adaptive bacterial immunity. Science. 2012; 337(6096):816–21. Epub 2012/06/30. https://doi.org/10.1126/science.1225829 PMID: 22745249; PubMed Central PMCID: PMC6286148. 2. Doudna JA, Charpentier E. Genome editing. The new frontier of genome engineering with CRISPR- Cas9. Science. 2014; 346(6213):1258096. Epub 2014/11/29. https://doi.org/10.1126/science.1258096 PMID: 25430774. 3. Hsu PD, Lander ES, Zhang F. Development and applications of CRISPR-Cas9 for genome engineering. Cell. 2014; 157(6):1262–78. Epub 2014/06/07. https://doi.org/10.1016/j.cell.2014.05.010 PMID: 24906146; PubMed Central PMCID: PMC4343198. 4. Clow PA, Du M, Jillette N, Taghbalout A, Zhu JJ, Cheng AW. CRISPR-mediated multiplexed live cell imaging of nonrepetitive genomic loci with one guide RNA per locus. Nat Commun. 2022; 13(1):1871. Epub 2022/04/08. https://doi.org/10.1038/s41467-022-29343-z PMID: 35387989; PubMed Central PMCID: PMC8987088. P.A.C and A.W.C on the inventorship is pending on the Casilio imaging method. The remaining Authors declare no competing interests. 5. 6. 7. Zetsche B, Gootenberg JS, Abudayyeh OO, Slaymaker IM, Makarova KS, Essletzbichler P, et al. Cpf1 is a single RNA-guided endonuclease of a class 2 CRISPR-Cas system. Cell. 2015; 163(3):759–71. Epub 2015/10/01. https://doi.org/10.1016/j.cell.2015.09.038 PMID: 26422227; PubMed Central PMCID: PMC4638220. Zetsche B, Heidenreich M, Mohanraju P, Fedorova I, Kneppers J, DeGennaro EM, et al. Multiplex gene editing by CRISPR-Cpf1 using a single crRNA array. Nat Biotechnol. 2017; 35(1):31–4. Epub 2016/12/ 06. https://doi.org/10.1038/nbt.3737 PMID: 27918548; PubMed Central PMCID: PMC5225075. Fonfara I, Richter H, Bratovič M, Le Rhun A, Charpentier E. The CRISPR-associated DNA-cleaving enzyme Cpf1 also processes precursor CRISPR RNA. Nature. 2016; 532(7600):517–21. Epub 2016/ 04/21. https://doi.org/10.1038/nature17945 PMID: 27096362. 8. Qiu HY, Ji RJ, Zhang Y. Current advances of CRISPR-Cas technology in cell therapy. Cell Insight. 2022; 1(6):100067. Epub 2023/05/17. https://doi.org/10.1016/j.cellin.2022.100067 PMID: 37193354; PubMed Central PMCID: PMC10120314. 9. Kim D, Kim J, Hur JK, Been KW, Yoon SH, Kim JS. Genome-wide analysis reveals specificities of Cpf1 endonucleases in human cells. Nat Biotechnol. 2016; 34(8):863–8. Epub 2016/06/09. https://doi.org/10. 1038/nbt.3609 PMID: 27272384. 10. Fu Y, Foden JA, Khayter C, Maeder ML, Reyon D, Joung JK, et al. High-frequency off-target mutagene- sis induced by CRISPR-Cas nucleases in human cells. Nat Biotechnol. 2013; 31(9):822–6. Epub 2013/ 06/25. https://doi.org/10.1038/nbt.2623 PMID: 23792628; PubMed Central PMCID: PMC3773023. 11. Hsu PD, Scott DA, Weinstein JA, Ran FA, Konermann S, Agarwala V, et al. DNA targeting specificity of RNA-guided Cas9 nucleases. Nat Biotechnol. 2013; 31(9):827–32. Epub 2013/07/23. https://doi.org/ 10.1038/nbt.2647 PMID: 23873081; PubMed Central PMCID: PMC3969858. 12. Pattanayak V, Lin S, Guilinger JP, Ma E, Doudna JA, Liu DR. High-throughput profiling of off-target DNA cleavage reveals RNA-programmed Cas9 nuclease specificity. Nat Biotechnol. 2013; 31(9):839– 43. Epub 2013/08/13. https://doi.org/10.1038/nbt.2673 PMID: 23934178; PubMed Central PMCID: PMC3782611. 13. Tsai SQ, Zheng Z, Nguyen NT, Liebers M, Topkar VV, Thapar V, et al. GUIDE-seq enables genome- wide profiling of off-target cleavage by CRISPR-Cas nucleases. Nat Biotechnol. 2015; 33(2):187–97. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 21 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As Epub 2014/12/17. https://doi.org/10.1038/nbt.3117 PMID: 25513782; PubMed Central PMCID: PMC4320685. 14. Frock RL, Hu J, Meyers RM, Ho YJ, Kii E, Alt FW. Genome-wide detection of DNA double-stranded breaks induced by engineered nucleases. Nat Biotechnol. 2015; 33(2):179–86. Epub 2014/12/17. https://doi.org/10.1038/nbt.3101 PMID: 25503383; PubMed Central PMCID: PMC4320661. 15. Kleinstiver BP, Pattanayak V, Prew MS, Tsai SQ, Nguyen NT, Zheng Z, et al. High-fidelity CRISPR- Cas9 nucleases with no detectable genome-wide off-target effects. Nature. 2016; 529(7587):490–5. Epub 2016/01/07. https://doi.org/10.1038/nature16526 PMID: 26735016; PubMed Central PMCID: PMC4851738. 16. Kim HK, Lee S, Kim Y, Park J, Min S, Choi JW, et al. High-throughput analysis of the activities of xCas9, SpCas9-NG and SpCas9 at matched and mismatched target sequences in human cells. Nat Biomed Eng. 2020; 4(1):111–24. Epub 2020/01/16. https://doi.org/10.1038/s41551-019-0505-1 PMID: 31937939. 17. Casini A, Olivieri M, Petris G, Montagna C, Reginato G, Maule G, et al. A highly specific SpCas9 variant is identified by in vivo screening in yeast. Nat Biotechnol. 2018; 36(3):265–71. Epub 2018/02/13. https:// doi.org/10.1038/nbt.4066 PMID: 29431739; PubMed Central PMCID: PMC6066108. 18. Lee JK, Jeong E, Lee J, Jung M, Shin E, Kim YH, et al. Directed evolution of CRISPR-Cas9 to increase its specificity. Nat Commun. 2018; 9(1):3048. Epub 2018/08/08. https://doi.org/10.1038/s41467-018- 05477-x PMID: 30082838; PubMed Central PMCID: PMC6078992. 19. Hu JH, Miller SM, Geurts MH, Tang W, Chen L, Sun N, et al. Evolved Cas9 variants with broad PAM compatibility and high DNA specificity. Nature. 2018; 556(7699):57–63. Epub 2018/03/08. https://doi. org/10.1038/nature26155 PMID: 29512652; PubMed Central PMCID: PMC5951633. 20. Fu Y, Sander JD, Reyon D, Cascio VM, Joung JK. Improving CRISPR-Cas nuclease specificity using truncated guide RNAs. Nat Biotechnol. 2014; 32(3):279–84. Epub 2014/01/28. https://doi.org/10.1038/ nbt.2808 PMID: 24463574; PubMed Central PMCID: PMC3988262. 21. Kim D, Kim S, Kim S, Park J, Kim JS. Genome-wide target specificities of CRISPR-Cas9 nucleases revealed by multiplex Digenome-seq. Genome Res. 2016; 26(3):406–15. Epub 2016/01/21. https://doi. org/10.1101/gr.199588.115 PMID: 26786045; PubMed Central PMCID: PMC4772022. 22. Vakulskas CA, Dever DP, Rettig GR, Turk R, Jacobi AM, Collingwood MA, et al. A high-fidelity Cas9 mutant delivered as a ribonucleoprotein complex enables efficient gene editing in human hematopoietic stem and progenitor cells. Nat Med. 2018; 24(8):1216–24. Epub 2018/08/08. https://doi.org/10.1038/ s41591-018-0137-0 PMID: 30082871; PubMed Central PMCID: PMC6107069. 23. Kim S, Kim D, Cho SW, Kim J, Kim JS. Highly efficient RNA-guided genome editing in human cells via delivery of purified Cas9 ribonucleoproteins. Genome Res. 2014; 24(6):1012–9. Epub 2014/04/04. https://doi.org/10.1101/gr.171322.113 PMID: 24696461; PubMed Central PMCID: PMC4032847. 24. Tsai SQ, Wyvekens N, Khayter C, Foden JA, Thapar V, Reyon D, et al. Dimeric CRISPR RNA-guided FokI nucleases for highly specific genome editing. Nat Biotechnol. 2014; 32(6):569–76. Epub 2014/04/ 29. https://doi.org/10.1038/nbt.2908 PMID: 24770325; PubMed Central PMCID: PMC4090141. 25. Xie H, Ge X, Yang F, Wang B, Li S, Duan J, et al. High-fidelity SaCas9 identified by directional screening in human cells. PLoS Biol. 2020; 18(7):e3000747. Epub 2020/07/10. https://doi.org/10.1371/journal. pbio.3000747 PMID: 32644995; PubMed Central PMCID: PMC7347106. 26. Chen JS, Dagdas YS, Kleinstiver BP, Welch MM, Sousa AA, Harrington LB, et al. Enhanced proofread- ing governs CRISPR-Cas9 targeting accuracy. Nature. 2017; 550(7676):407–10. Epub 2017/09/21. https://doi.org/10.1038/nature24268 PMID: 28931002; PubMed Central PMCID: PMC5918688. 27. Slaymaker IM, Gao L, Zetsche B, Scott DA, Yan WX, Zhang F. Rationally engineered Cas9 nucleases with improved specificity. Science (New York, NY). 2016; 351(6268):84–8. Epub 2015/12/03. https:// doi.org/10.1126/science.aad5227 PMID: 26628643; PubMed Central PMCID: PMC4714946. 28. Kulcsa´r PI, Ta´ las A, Husza´r K, Ligeti Z, To´ th E, Weinhardt N, et al. Crossing enhanced and high fidelity SpCas9 nucleases to optimize specificity and cleavage. Genome Biol. 2017; 18(1):190. Epub 2017/10/ 08. https://doi.org/10.1186/s13059-017-1318-8 PMID: 28985763; PubMed Central PMCID: PMC6389135. 29. 30. Tan Y, Chu AHY, Bao S, Hoang DA, Kebede FT, Xiong W, et al. Rationally engineered Staphylococcus aureus Cas9 nucleases with high genome-wide specificity. Proc Natl Acad Sci U S A. 2019; 116 (42):20969–76. Epub 2019/10/02. https://doi.org/10.1073/pnas.1906843116 PMID: 31570596; PubMed Central PMCID: PMC6800346. Tak YE, Kleinstiver BP, Nuñez JK, Hsu JY, Horng JE, Gong J, et al. Inducible and multiplex gene regu- lation using CRISPR-Cpf1-based transcription factors. Nat Methods. 2017; 14(12):1163–6. Epub 2017/ 10/31. https://doi.org/10.1038/nmeth.4483 PMID: 29083402; PubMed Central PMCID: PMC5909187. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 22 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As 31. Kleinstiver BP, Tsai SQ, Prew MS, Nguyen NT, Welch MM, Lopez JM, et al. Genome-wide specificities of CRISPR-Cas Cpf1 nucleases in human cells. Nat Biotechnol. 2016; 34(8):869–74. Epub 2016/06/28. https://doi.org/10.1038/nbt.3620 PMID: 27347757; PubMed Central PMCID: PMC4980201. 32. Kleinstiver BP, Sousa AA, Walton RT, Tak YE, Hsu JY, Clement K, et al. Engineered CRISPR-Cas12a variants with increased activities and improved targeting ranges for gene, epigenetic and base editing. Nat Biotechnol. 2019; 37(3):276–82. Epub 2019/02/12. https://doi.org/10.1038/s41587-018-0011-0 PMID: 30742127; PubMed Central PMCID: PMC6401248. 33. Huang H, Huang G, Tan Z, Hu Y, Shan L, Zhou J, et al. Engineered Cas12a-Plus nuclease enables gene editing with enhanced activity and specificity. BMC Biol. 2022; 20(1):91. Epub 2022/04/27. https:// doi.org/10.1186/s12915-022-01296-1 PMID: 35468792; PubMed Central PMCID: PMC9040236. 34. Joung JK, Kleinstiver B, inventorsVariants of CRISPR from Prevotella and Francisella 1 (Cpf1) patent US20220025347(A1). 2021-10-12. 35. Bravo JPK, Liu MS, Hibshman GN, Dangerfield TL, Jung K, McCool RS, et al. Structural basis for mis- match surveillance by CRISPR-Cas9. Nature. 2022; 603(7900):343–7. Epub 2022/03/04. https://doi. org/10.1038/s41586-022-04470-1 PMID: 35236982; PubMed Central PMCID: PMC8907077. Applica- tion based on this research titled “Methods and compositions for improved Cas9 specificity” filed by the Board of Regents, The University of Texas System. The US Patent and Trademark Office (USPTO) has assigned US application no. 63/243,481 to this application, and the filing date of 13 September 2021. K. A.J. is the president of KinTek, which provided the chemical-quench flow instruments and the KinTek Explorer software used in this study. 36. Chen P, Zhou J, Wan Y, Liu H, Li Y, Liu Z, et al. A Cas12a ortholog with stringent PAM recognition fol- lowed by low off-target editing rates for genome editing. Genome Biol. 2020; 21(1):78. Epub 2020/03/ 28. https://doi.org/10.1186/s13059-020-01989-2 PMID: 32213191; PubMed Central PMCID: PMC7093978. 37. Zhou J, Chen P, Wang H, Liu H, Li Y, Zhang Y, et al. Cas12a variants designed for lower genome-wide off-target effect through stringent PAM recognition. Mol Ther. 2022; 30(1):244–55. Epub 2021/10/24. https://doi.org/10.1016/j.ymthe.2021.10.010 PMID: 34687846; PubMed Central PMCID: PMC8753454. 38. Yamano T, Nishimasu H, Zetsche B, Hirano H, Slaymaker IM, Li Y, et al. Crystal Structure of Cpf1 in Complex with Guide RNA and Target DNA. Cell. 2016; 165(4):949–62. Epub 2016/04/27. https://doi. org/10.1016/j.cell.2016.04.003 PMID: 27114038; PubMed Central PMCID: PMC4899970. 39. Gao L, Cox DBT, Yan WX, Manteiga JC, Schneider MW, Yamano T, et al. Engineered Cpf1 variants with altered PAM specificities. Nat Biotechnol. 2017; 35(8):789–92. Epub 2017/06/06. https://doi.org/ 10.1038/nbt.3900 PMID: 28581492; PubMed Central PMCID: PMC5548640. To´ th E, Varga E´ , Kulcsa´ r PI, Kocsis-Jutka V, Krausz SL, Nyeste A, et al. Improved LbCas12a variants with altered PAM specificities further broaden the genome targeting range of Cas12a nucleases. Nucleic Acids Res. 2020; 48(7):3722–33. Epub 2020/02/29. https://doi.org/10.1093/nar/gkaa110 PMID: 32107556; PubMed Central PMCID: PMC7144938. 40. 41. Huang H, Hu Y, Huang G, Ma S, Feng J, Wang D, et al. Tag-seq: a convenient and scalable method for genome-wide specificity assessment of CRISPR/Cas nucleases. Commun Biol. 2021; 4(1):830. Epub 2021/07/04. https://doi.org/10.1038/s42003-021-02351-3 PMID: 34215845; PubMed Central PMCID: PMC8253812. 42. Yang X, Stein KR, Hang HC. Anti-infective bile acids bind and inactivate a Salmonella virulence regula- tor. Nat Chem Biol. 2022. Epub 2022/09/30. https://doi.org/10.1038/s41589-022-01122-3 PMID: 36175659. 43. Kulcsa´r PI, Ta´ las A, Ligeti Z, Krausz SL, Welker E. SuperFi-Cas9 exhibits remarkable fidelity but severely reduced activity yet works effectively with ABE8e. Nat Commun. 2022; 13(1):6858. https://doi. org/10.1038/s41467-022-34527-8 PMID: 36369279; PubMed Central PMCID: PMC9652449. 44. Cui Y, Tang YC, Liang M, Ji QH, Zeng Y, Chen H, et al. Direct observation of the formation of a CRISPR-Cas12a R-loop complex at the single-molecule level. Chem Commun. 2020; 56(14):2123– 2126. https://doi.org/10.1039/c9cc08325a WOS:000516611700007. PMID: 31970368 45. Gao S, Wang Y, Qi T, Wei J, Hu Z, Liu J, et al. Genome editing with natural and engineered CjCas9 orthologs. Mol Ther. 2023; 31(4):1177–87. Epub 2023/02/04. https://doi.org/10.1016/j.ymthe.2023.01. 029 PMID: 36733251; PubMed Central PMCID: PMC10124074. 46. Malinin NL, Lee G, Lazzarotto CR, Li Y, Zheng Z, Nguyen NT, et al. Defining genome-wide CRISPR- Cas genome-editing nuclease activity with GUIDE-seq. Nat Protoc. 2021; 16(12):5592–615. Epub 2021/11/14. https://doi.org/10.1038/s41596-021-00626-x PMID: 34773119; PubMed Central PMCID: PMC9331158. 47. Nobles CL, Reddy S, Salas-McKee J, Liu X, June CH, Melenhorst JJ, et al. iGUIDE: an improved pipe- line for analyzing CRISPR cleavage specificity. Genome Biol. 2019; 20(1):14. Epub 2019/01/19. https:// doi.org/10.1186/s13059-019-1625-3 PMID: 30654827; PubMed Central PMCID: PMC6337799. The PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 23 / 24 PLOS BIOLOGYImproved genome editing specificity by HyperFi-As authors declare that they have no competing interests. PUBLISHER’S NOTE: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. 48. Kuscu C, Arslan S, Singh R, Thorpe J, Adli M. Genome-wide analysis reveals characteristics of off-tar- get sites bound by the Cas9 endonuclease. Nat Biotechnol. 2014; 32(7):677–83. Epub 2014/05/20. https://doi.org/10.1038/nbt.2916 PMID: 24837660. 49. Kim D, Lim K, Kim ST, Yoon SH, Kim K, Ryu SM, et al. Genome-wide target specificities of CRISPR RNA-guided programmable deaminases. Nat Biotechnol. 2017; 35(5):475–80. Epub 2017/04/12. https://doi.org/10.1038/nbt.3852 PMID: 28398345. 50. Kim HK, Song M, Lee J, Menon AV, Jung S, Kang YM, et al. In vivo high-throughput profiling of CRISPR-Cpf1 activity. Nat Methods. 2017; 14(2):153–9. Epub 2016/12/20. https://doi.org/10.1038/ nmeth.4104 PMID: 27992409. 51. Liang X, Potter J, Kumar S, Zou Y, Quintanilla R, Sridharan M, et al. Rapid and highly efficient mamma- lian cell engineering via Cas9 protein transfection. J Biotechnol. 2015; 208:44–53. Epub 2015/05/25. https://doi.org/10.1016/j.jbiotec.2015.04.024 PMID: 26003884. 52. Hur JK, Kim K, Been KW, Baek G, Ye S, Hur JW, et al. Targeted mutagenesis in mice by electroporation of Cpf1 ribonucleoproteins. Nat Biotechnol. 2016; 34(8):807–8. Epub 2016/06/09. https://doi.org/10. 1038/nbt.3596 PMID: 27272385. 53. Dai X, Park JJ, Du Y, Kim HR, Wang G, Errami Y, et al. One-step generation of modular CAR-T cells with AAV-Cpf1. Nat Methods. 2019; 16(3):247–54. Epub 2019/02/26. https://doi.org/10.1038/s41592- 019-0329-7 PMID: 30804551; PubMed Central PMCID: PMC6519746. 54. Yang YJ, Dong HL, Qiang XW, Fu H, Zhou EC, Zhang C, et al. Cytosine Methylation Enhances DNA Condensation Revealed by Equilibrium Measurements Using Magnetic Tweezers. J Am Chem Soc. 2020; 142(20):9203–9209. https://doi.org/10.1021/jacs.9b11957 WOS:000537415600018. PMID: 32330022 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002514 March 14, 2024 24 / 24 PLOS BIOLOGY
10.1126_science.add6162	H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Published as: Science. 2023 July 07; 381(6653): 54–59. Cortical polarity ensures its own asymmetric inheritance in the stomatal lineage to pattern the leaf surface Andrew Muroyama1,2,, Yan Gong1,3, Kensington S. Hartman2, Dominique Bergmann1,4, 1.Department of Biology, Stanford University, Stanford, CA 94305, USA 2.Division of Biological Sciences, Department of Cell and Developmental Biology, University of California San Diego, La Jolla, CA 92093, USA 3.Current Address: Department of Organismic & Evolutionary Biology, Harvard University, Cambridge, MA 02138, USA 4.Howard Hughes Medical Institute, Stanford University, Stanford, CA 94305, USA Abstract Asymmetric cell divisions specify differential cell fates across kingdoms. In metazoans, preferential inheritance of fate determinants into one daughter cell frequently depends on polarity-cytoskeleton interactions. Despite the prevalence of asymmetric divisions throughout plant development, evidence for analogous mechanisms that segregate fate determinants remain elusive. Here, we describe a mechanism in the Arabidopsis leaf epidermis that ensures unequal inheritance of a fate-enforcing polarity domain. By defining a cortical region depleted of stable microtubules, the polarity domain limits possible division orientations. Accordingly, uncoupling the polarity domain from microtubule organization during mitosis leads to aberrant division planes and accompanying cell identity defects. Our data highlight how a common biological module, coupling polarity to fate segregation via the cytoskeleton, can be reconfigured to accommodate unique features of plant development. Stomatal lineage cells break common division orientation rules Asymmetric divisions establish differential cell identities by positioning daughter cells relative to fate-enforcing extrinsic signals or by segregating intrinsic fate determinants (1, 2). The development of many plant organs depends on asymmetric divisions that place daughter This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Co-correspondence to [email protected] and [email protected]. Author Contributions: A.M. and D.C.B. conceived of the study and designed the experiments. Y.G. acquired the time-lapse data for cell tracking in trm678. K.S.H acquired and analyzed the trm678 mutant phenotype data. A.M. performed all other experiments and analyses. A.M. and D.C.B wrote the manuscript with feedback from Y.G. and K.S.H. Competing interests: The authors declare no competing conflicts of interest. Data and materials availability: All newly generated materials in the manuscript are available upon reasonable request. All data are available in the manuscript and associated supplementary materials. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 2 cells in proximity to neighbor-derived signals, such as mobile transcription factors (3, 4), hormones (5-7), or small peptides (8, 9). However, whether plants have mechanisms that ensure preferential inheritance of fate regulators remains unresolved. To interrogate how plant cells couple asymmetric divisions to identity specification, we monitored the cell division and differentiation dynamics of stomatal precursors in the Arabidopsis leaf epidermis (Fig. 1A). In this lineage, flexible asymmetric divisions in morphologically heterogenous early lineage cells create and pattern stomata (cellular valves that mediate plant-atmosphere gas exchange) by priming divergent developmental trajectories in the two daughters. The smaller meristemoid will eventually give rise to the paired guard cells that comprise a stoma, and the larger stomatal lineage ground cell (SLGC) will expand to become a pavement cell. All asymmetric cell divisions within the lineage are preceded by the formation of a plasma membrane-associated polarity complex defined by BREAKING OF ASYMMETRY IN THE STOMATAL LINEAGE (BASL) (10) and BREVIS RADIX family (BRXf) (11) proteins (Fig. 1, A and B). Before mitosis, the BASL/BRXf crescent 1) recruits another polarly localized protein, POLAR (12), which in turn enriches GLYCOGEN SYNTHASE KINASE3 (GSK3)-like kinases at the cortex to promote asymmetric divisions (13), and 2) directs nuclear migration to bias the division site (14). After it is inherited by the SLGC through the asymmetric division, cortical BASL/BRXf enhances MITOGEN ACTIVATED PROTEIN KINASE (MAPK) signaling to suppress meristemoid identity (13, 15). Despite its central role in coordinating asymmetric division entry and daughter-cell identity post- division (16), how BASL/BRXf asymmetric inheritance is regulated is unknown (Fig. 1B). We performed time-lapse imaging of developing cotyledons harboring markers for nuclei (R2D2 (17)), the plasma membrane (ML1p::mCherry-RCI2A) and the polarity crescent (BRXL2p::BRXL2-YFP). In agreement with previous observations, all asymmetric divisions resulted in singular inheritance of the BRXL2 crescent. Close analysis of these cells, however, revealed two asymmetric division subclasses that were defined by their division planes. The majority (73% of divisions) divided along the calculated shortest distance that intersected opposing cell walls at the site of the nucleus (small Δ θ°, see Materials and Methods) (Fig.1, C to F), following the expectations set by the observed division planes in many plant cell types (18, 19). For this class of asymmetric divisions, polarity-directed nuclear migration (14) coupled with minimization of the division plane accurately predicted the final division site. Division planes in the second class of asymmetric divisions (27% of divisions) deviated significantly from the calculated shortest wall (Fig. 1, D, E and G), suggesting that additional inputs control orientation of these early lineage divisions. Stomatal lineage divisions are not unique in breaking the shortest wall rule, but other cases during Arabidopsis development are related to broad, extrinsic influences such as tissue mechanics or hormone signaling (20, 21). Discrete, cell-autonomous mechanisms that tune division orientation have not been described. The morphological heterogeneity of stomatal precursors was well-represented within both asymmetric division subclasses (Fig. 1, F and G), suggesting that unique geometric features do not define asymmetric division subtypes. Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 3 Instead, asymmetrically dividing cells specifically bypassed the calculated shortest division plane (large Δ θ°) when that wall was predicted to intersect the plasma membrane within the cortical polarized site (Fig. 1G). Importantly, if only the non-polarized membrane was considered permissive for division plane placement, these asymmetric divisions continued to follow the shortest wall rule (fig. S1). This correlation suggested that polarized BASL/BRXf may be a cell-intrinsic cue capable of constraining potential orientations to control its own asymmetric inheritance. Next, we tested whether cell polarity is necessary to stratify the two asymmetric division classes by tracking progenitor divisions in basl mutants (basl 35Sp::PIP2A-RFP ML1p::H2B-YFP). Loss of cellular polarity in basl collapsed the two asymmetric division classes into one (Fig. 1, H and I) that varied significantly from the total wild-type asymmetric divisions (Kolmogorov-Smirnov test, p=0.0005). Importantly, basl divisions did not differ significantly from wild-type asymmetric divisions without polarity conflict (p=0.1568), indicating that basl divisions follow the shortest wall rule. Therefore, the polarized BASL/BRXf domain is required to override default division patterns during formative asymmetric divisions. BASL influences preprophase band position To determine the basis of this control, we examined the cortical microtubule structures that play essential roles during division orientation in plant cells (22). TAN1p::CFP-TAN1 foci, which mark the cortical division zone (23), never appear within the BASL/BRXf domain, suggesting that BASL/BRXf operate at an early step during division orientation (fig. S2). We found that the preprophase band of microtubules, which is the first marker of the eventual cortical division site, never formed within the polarity domain (Fig. 1, J to M). An analysis of the Δ θ° between the preprophase band and calculated shortest wall showed a similar bifurcation of asymmetric divisions into two classes: the majority (66%) had preprophase bands that closely aligned to the predicted shortest wall while preprophase bands in the second class (34%) deviated significantly from the shortest distance (fig. S3). In this second class, 1) preprophase bands did not align with the predicted shortest wall when it bisected the polarity domain, and 2) polarity was required for preprophase band realignment away from the shortest wall (fig. S3). Together, these data indicated that the BASL/BRXf polarity crescent might orient divisions by controlling preprophase band placement. To test this hypothesis, we generated lines to monitor BRXL2 inheritance in the trm678 mutant (trm678 BRXL2p::BRXL2-YFP ML1p::mCherry-RCI2A), which does not form preprophase bands (24). In contrast to wild-type asymmetric divisions, where BRXL2 was inherited by a single daughter cell, new cell walls frequently bisected the polarity site in trm678 (32% of divisions) (Fig. 2, A and B), showing that the preprophase band is required to ensure complete inheritance of polarized BASL/BRXf. Next, we tracked cell fate outcomes following incorrect BASL/BRXf inheritance by monitoring progression through the stomatal lineage by tracking MUTE (25), a transcription factor that establishes the identity of the immediate stomatal guard cell precursor. trm678 asymmetric divisions where BRXL2 was correctly inherited by a single daughter cell showed normal lineage progression; MUTE expression was detectable in the smaller cell after the division, and all Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 4 tracked MUTE+ cells in trm678 became paired guard cells (fig. S4). In contrast, trm678 cells where cortical BRXL2 was bisected by the nascent division plane tended to generate daughters that 1) both inherited cortical BRXL2, 2) never transitioned to MUTE+ cells, and 3) failed to become pavement cells (Fig. 2C). In agreement with these tracking data, 7dpg trm678 cotyledons had fewer stomata and a mispatterned epidermis (Fig. 2, D to F). We did not observe any alterations in the relationship between microtubules and BRXL2 in interphase trm678 (fig. S4, D to F). Therefore, we conclude the preprophase band serves as an essential link between the polarity domain and division orientation to regulate stomatal identity. Cortical BASL domains are locally depleted of stable microtubules How does the BASL/BRXf crescent influence preprophase band establishment? By creating a stomatal lineage-specific microtubule reporter line (TMMp::mCherry-TUA5), we could analyze microtubules and the BRXL2 polarity domain along anticlinal walls with high resolution (Fig. 3A). Unexpectedly, anticlinal microtubules were strongly depleted from the plasma membrane within the polarized domain, even in interphase SLGCs (Fig. 3, A to C). We confirmed that the same microtubule depletion zone occurred was observed when using a second polarity reporter, BASL, and in stomatal lineage cells of true leaves (fig. S5). POLAR, which shows overlapping but distinct localization from BASL/BRXf (12), co-localizes with microtubules outside the BASL/BRXL2 domain (fig. S5), indicating that microtubule depletion is correlated specifically with BASL/BRXf and is not a generalized activity of polarized proteins in the stomatal lineage. BASL/BRXf could 1) locally deplete microtubules or 2) opportunistically polarize to already microtubule-poor regions. To distinguish between these possibilities, we performed two analyses. First, we examined cortical microtubule distribution before BASL polarization and found no microtubule-depleted region (fig. S6). Second, we compared microtubule distribution in wild-type and polarity-defective SLGCs and found that microtubule distribution was more homogenous in basl and brx-quad (11), which abrogate polarity, and in lines where addition of a myristoylation signal (BASLp::MYR-BRX-YFP (11)) renders BRX localization largely uniform at the cortex (Fig. 3, D to G, fig. S7). We also followed unmanipulated SLGCs as they lost polarized BASL/BRXf several hours after cell division. Our time course analysis showed that anticlinal microtubules reappeared within previously polarized regions in mature SLGCs (fig. S7D). From these data, we conclude that BASL/ BRXf polarity creates and is required to maintain local microtubule loss at the plasma membrane. Mutual inhibition by opposing plasma membrane-associated domains can drive polarization, as in the conserved PAR networks in animals (26) or a recently described polarity system in the monocot B. distachyon (27). Because our data raised the possibility that microtubules and cortical BASL/BRXf could operate in an analogous manner and inhibit the spread of each other, we tested whether altering microtubule distribution affected the stomatal lineage polarity domain. In agreement with previous results (28, 29), we found that microtubules are not necessary for the formation of a polarized BASL/BRXf domain (fig. S8A). However, quantification revealed a slight but significant spread of the polarity domain along the Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 5 anticlinal wall in the absence of microtubules (fig. S8B). Short plasmolysis treatments, which dramatically disrupt cortical microtubule distribution, similarly altered polarity boundaries and polarity domain size without complete depolarization (fig. S8C). Therefore, microtubules shape BASL/BRXf domain boundaries although they are dispensable for polarity itself. Intriguingly, this interaction shares striking similarities with microtubule- mediated sculpting of ROP GTPase domains in non-dividing cells of the xylem and in trichomes (30-32). Microtubule dynamics are locally altered within polar domains How do cortical BASL/BRXf locally deplete cortical microtubules? Owing to the technical challenges associated with monitoring dynamic microtubule behavior along the anticlinal wall of meristemoids, we created a heterologous system where we could track microtubule dynamics co-incident with the BASL polarity domain. By introgressing a ubiquitous microtubule reporter (35Sp::mCherry-TUA5) into a line expressing a hyperactive version of BASL capable of rescuing the basl phenotype (35Sp::GFP-BASL-IC, hereafter referred to as BASLectopic (10)), we could monitor microtubule organization relative to BASL in the hypocotyl epidermis. BASLectopic locally depleted microtubules along anticlinal walls in the hypocotyl epidermis as in the stomatal lineage (fig. S9), demonstrating that this ectopic system recapitulates the molecular interactions found in the leaf epidermis. BASLectopic domains extended onto the apical surfaces of hypocotyl epidermal cells and locally depleted cortical microtubules (Fig. 4, A and C), allowing us to observe the BASL-mediated effects on microtubules with precision not possible within the stomatal lineage. Increased microtubule severing has been identified as a key reorganizer of cortical microtubule arrays during several developmental transitions (33, 34). However, as severing preferentially occurs at microtubule crossover sites (35, 36) and there were few crossovers in microtubule-depleted BASLectopic regions, severing was largely suppressed within BASLectopic domains (fig. S10, A and B). Tracking of microtubule minus ends within BASLectopic also indicated that local microtubule depletion was not due to decreased minus- end stability (fig. S10C). Instead, we found that BASLectopic had two effects on microtubule plus-ends. First, plus-end polymerization and depolymerization rates were significantly suppressed within BASLectopic (Fig. 4, E and F, fig. S10D). Second, we observed that microtubule plus-ends rapidly underwent catastrophe upon entering the BASLectopic domain (Fig. 4, B and D). Increased catastrophe rates often led to complete loss of the microtubule, reestablishing the microtubule depletion zone. To validate that our BASLectopic findings in the hypocotyl reflect BASL-microtubule interactions within the stomatal lineage, we used two independent approaches. First, we performed time-lapse imaging of stomatal progenitors in BRXL2p::BRXL2-YFP TMMp::mCherry-TUA5 seedlings and observed transient microtubules within the polarity domain that were rapidly depolymerized (fig. S11A). Second, to monitor growing plus ends with higher precision along the anticlinal wall, we generated a stomatal lineage-specific END BINDING PROTEIN 1b (EB1b) reporter (TMMp::EB1b-mCherry) and introgressed it into the BRXL2 reporter line. Fewer EB1b puncta were observed within the native polarity domain in SLGCs than in non-polarized regions of the same cells (fig. S11, B to D). Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 6 Therefore, our analyses of microtubule dynamics in native and heterologous systems reveal that the BASL/BRXf domain destabilizes microtubule plus ends to locally deplete them from the polarized region. Discussion As encoders of spatial information, polarity domains are central regulators of asymmetric cell division in diverse plant tissues (37-40). Here, we provide evidence that the BASL/ BRXf polarity domain robustly serves dual functions to orient asymmetric divisions and specify cell identity by controlling its own inheritance via negative interactions with the microtubule cytoskeleton. Polarity-microtubule interactions now emerge as a common theme to guide asymmetric inheritance of fate regulators during both metazoan and plant asymmetric divisions, albeit through fundamentally different mechanisms (fig. S12). In the canonical asymmetric cell division pathway in animal cells, the cortical polarity domain is responsible for 1) localizing fate determinants to one pole and 2) subsequently directing the division angle to ensure their singular and asymmetric inheritance by exerting pulling forces on astral microtubules (41, 42). The model we advance here differs in several significant ways. First, the proposed mechanism utilizes core plant-specific mitotic structures without the need to invoke a role for astral microtubules, which are absent in plant spindles. Second, rather than ensuring its singular inheritance by precisely specifying the ultimate division plane, BASL/BRXf renders a region of the membrane unavailable as the division site. Third, while this mechanism has an identical outcome—asymmetric inheritance of key fate regulators—it is uniquely suited for a morphologically heterogenous population that, nonetheless, must robustly couple asymmetric division orientation with subsequent daughter cell identities. How does BASL-mediated polarity modulate microtubule dynamics? BASL and BRXf proteins both contain large, disordered regions (43), leading us to favor two general models. In the first, BASL and BRXf scaffold effectors that 1) directly affect plus-end kinetics and 2) potentially bind along the microtubule lattice to impact depolymerization rates. From our analyses in this work, we anticipate that such microtubule-associated effectors would be expressed throughout the cell cycle and in multiple tissues, which has complicated our ongoing efforts to identify them. In the second model, which does not invoke additional downstream factors, polarization via phase-separation could tune the local physical properties of the membrane-adjacent cytoplasm; such a mechanism can modulate microtubule dynamics in yeast (44), and might be hinted at by the dampening of MT dynamics at the polarity zone (fig. S10D). Of the documented, polarity-mediated asymmetric divisions in plants, those that violate the shortest wall rule, such as those in the early Arabidopsis embryo or subsidiary mother cell divisions in Zea mays, may be the closest corollaries to the system presented here. While BASL is both eudicot (43) and stomatal lineage specific, BRX family proteins participate in additional cellular decisions in Arabidopsis and are much more deeply conserved in the green lineage (45), hinting that other tissue-specific regulators could provide context specificity to a common polarity core. Further analysis of polarity will help clarify whether Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 7 this mechanism is shared across plant tissues and species or whether it has evolved for the challenges associated with flexible patterning in the eudicot stomatal lineage. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgments: We thank Dr. David Bouchez for kindly sharing the trm678 seeds and Amy Lanctot for generating POLARp::POLAR-YFP. We thank Dr. Carolyn Rasmussen for kindly sharing the TAN1p::CFP-TAN1 reporter seeds. We thank Dr. Nidhi Sharma for technical help generating the trm678 reporter lines. We thank Dr. Eric Griffis at the Nikon Imaging Center at the University of California, San Diego for assistance with several imaging experiments. We thank Dr. Mark Estelle, Dr. Alexandra Dickinson, and members of the Bergmann lab for feedback on the manuscript. Funding: A.M. was supported by a postdoctoral fellowship from the National Institutes of Health (F32 GM133102-01) and is currently supported by start-up funds from the University of California San Diego. Y.G. was supported by funds from Stanford University and the Howard Hughes Medical Institute. K.S.H. is supported by funds from the University of California, San Diego to A.M. D.C.B. is an investigator of the Howard Hughes Medical Institute. References 1. Sunchu B, Cabernard C, Principles and mechanisms of asymmetric cell division. Development 147, (2020). 2. Venkei ZG, Yamashita YM, Emerging mechanisms of asymmetric stem cell division. J Cell Biol 217, 3785–3795 (2018). [PubMed: 30232100] 3. Nakajima K, Sena G, Nawy T, Benfey PN, Intercellular movement of the putative transcription factor SHR in root patterning. Nature 413, 307–311 (2001). [PubMed: 11565032] 4. Raissig MT et al. , Mobile MUTE specifies subsidiary cells to build physiologically improved grass stomata. Science 355, 1215–1218 (2017). [PubMed: 28302860] 5. Casimiro I et al. , Auxin transport promotes Arabidopsis lateral root initiation. Plant Cell 13, 843– 852 (2001). [PubMed: 11283340] 6. Vatén A, Soyars CL, Tarr PT, Nimchuk ZL, Bergmann DC, Modulation of Asymmetric Division Diversity through Cytokinin and SPEECHLESS Regulatory Interactions in the Arabidopsis Stomatal Lineage. Dev Cell 47, 53–66.e55 (2018). [PubMed: 30197241] 7. Cruz-Ramírez A et al. , A bistable circuit involving SCARECROW-RETINOBLASTOMA integrates cues to inform asymmetric stem cell division. Cell 150, 1002–1015 (2012). [PubMed: 22921914] 8. Hara K, Kajita R, Torii KU, Bergmann DC, Kakimoto T, The secretory peptide gene EPF1 enforces the stomatal one-cell-spacing rule. Genes Dev 21, 1720–1725 (2007). [PubMed: 17639078] 9. Hunt L, Gray JE, The signaling peptide EPF2 controls asymmetric cell divisions during stomatal development. Curr Biol 19, 864–869 (2009). [PubMed: 19398336] 10. Dong J, MacAlister CA, Bergmann DC, BASL controls asymmetric cell division in Arabidopsis. Cell 137, 1320–1330 (2009). [PubMed: 19523675] 11. Rowe MH, Dong J, Weimer AK, Bergmann DC, A Plant-Specific Polarity Module Establishes Cell Fate Asymmetry in the Arabidopsis Stomatal Lineage. bioRxiv, 614636 (2019). 12. Pillitteri LJ, Peterson KM, Horst RJ, Torii KU, Molecular profiling of stomatal meristemoids reveals new component of asymmetric cell division and commonalities among stem cell populations in Arabidopsis. Plant Cell 23, 3260–3275 (2011). [PubMed: 21963668] 13. Houbaert A et al. , POLAR-guided signalling complex assembly and localization drive asymmetric cell division. Nature 563, 574–578 (2018). [PubMed: 30429609] Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 8 14. Muroyama A, Gong Y, Bergmann DC, Opposing, Polarity-Driven Nuclear Migrations Underpin Asymmetric Divisions to Pattern Arabidopsis Stomata. Curr Biol 30, 4467–4475.e4464 (2020). [PubMed: 32946753] 15. Zhang Y, Wang P, Shao W, Zhu JK, Dong J, The BASL polarity protein controls a MAPK signaling feedback loop in asymmetric cell division. Dev Cell 33, 136–149 (2015). [PubMed: 25843888] 16. Gong Y et al. , The Arabidopsis stomatal polarity protein BASL mediates distinct processes before and after cell division to coordinate cell size and fate asymmetries. Development 148, (2021). 17. Liao CY et al. , Reporters for sensitive and quantitative measurement of auxin response. Nat Methods 12, 207–210, 202 p following 210 (2015). [PubMed: 25643149] 18. Besson S, Dumais J, Universal rule for the symmetric division of plant cells. Proc Natl Acad Sci U S A 108, 6294–6299 (2011). [PubMed: 21383128] 19. Shapiro BE, Tobin C, Mjolsness E, Meyerowitz EM, Analysis of cell division patterns in the Arabidopsis shoot apical meristem. Proc Natl Acad Sci U S A 112, 4815–4820 (2015). [PubMed: 25825722] 20. Vaddepalli P et al. , Auxin-dependent control of cytoskeleton and cell shape regulates division orientation in the Arabidopsis embryo. Curr Biol 31, 4946–4955.e4944 (2021). [PubMed: 34610273] 21. Louveaux M, Julien JD, Mirabet V, Boudaoud A, Hamant O, Cell division plane orientation based on tensile stress in Arabidopsis thaliana. Proc Natl Acad Sci U S A 113, E4294–4303 (2016). [PubMed: 27436908] 22. Rasmussen CG, Wright AJ, Müller S, The role of the cytoskeleton and associated proteins in determination of the plant cell division plane. Plant J 75, 258–269 (2013). [PubMed: 23496276] 23. Walker KL, Müller S, Moss D, Ehrhardt DW, Smith LG, Arabidopsis TANGLED identifies the division plane throughout mitosis and cytokinesis. Curr Biol 17, 1827–1836 (2007). [PubMed: 17964159] 24. Schaefer E et al. , The preprophase band of microtubules controls the robustness of division orientation in plants. Science 356, 186–189 (2017). [PubMed: 28408602] 25. Pillitteri LJ, Sloan DB, Bogenschutz NL, Torii KU, Termination of asymmetric cell division and differentiation of stomata. Nature 445, 501–505 (2007). [PubMed: 17183267] 26. Etemad-Moghadam B, Guo S, Kemphues KJ, Asymmetrically distributed PAR-3 protein contributes to cell polarity and spindle alignment in early C. elegans embryos. Cell 83, 743–752 (1995). [PubMed: 8521491] 27. Zhang D et al. , Opposite polarity programs regulate asymmetric subsidiary cell divisions in grasses. Elife 11, (2022). 28. Chan J, Mansfield C, Clouet F, Dorussen D, Coen E, Intrinsic Cell Polarity Coupled to Growth Axis Formation in Tobacco BY-2 Cells. Curr Biol 30, 4999–5006.e4993 (2020). [PubMed: 33035485] 29. Bringmann M, Bergmann DC, Tissue-wide Mechanical Forces Influence the Polarity of Stomatal Stem Cells in Arabidopsis. Curr Biol 27, 877–883 (2017). [PubMed: 28285992] 30. Oda Y, Fukuda H, Initiation of cell wall pattern by a Rho- and microtubule-driven symmetry breaking. Science 337, 1333–1336 (2012). [PubMed: 22984069] 31. Sugiyama Y, Wakazaki M, Toyooka K, Fukuda H, Oda Y, A Novel Plasma Membrane-Anchored Protein Regulates Xylem Cell-Wall Deposition through Microtubule-Dependent Lateral Inhibition of Rho GTPase Domains. Curr Biol 27, 2522–2528.e2524 (2017). [PubMed: 28803875] 32. Yanagisawa M, Alonso JM, Szymanski DB, Microtubule-Dependent Confinement of a Cell Signaling and Actin Polymerization Control Module Regulates Polarized Cell Growth. Curr Biol 28, 2459–2466.e2454 (2018). [PubMed: 30033335] 33. Lindeboom JJ et al. , A mechanism for reorientation of cortical microtubule arrays driven by microtubule severing. Science 342, 1245533 (2013). [PubMed: 24200811] 34. Schneider R et al. , Long-term single-cell imaging and simulations of microtubules reveal principles behind wall patterning during proto-xylem development. Nat Commun 12, 669 (2021). [PubMed: 33510146] 35. Wightman R, Turner SR, Severing at sites of microtubule crossover contributes to microtubule alignment in cortical arrays. Plant J 52, 742–751 (2007). [PubMed: 17877711] Science. Author manuscript; available in PMC 2023 July 07. Muroyama et al. Page 9 36. Zhang Q, Fishel E, Bertroche T, Dixit R, Microtubule severing at crossover sites by katanin generates ordered cortical microtubule arrays in Arabidopsis. Curr Biol 23, 2191–2195 (2013). [PubMed: 24206847] 37. Yoshida S et al. , A SOSEKI-based coordinate system interprets global polarity cues in Arabidopsis. Nat Plants 5, 160–166 (2019). [PubMed: 30737509] 38. Campos R, Goff J, Rodriguez-Furlan C, Van Norman JM, The Arabidopsis Receptor Kinase IRK Is Polarized and Represses Specific Cell Divisions in Roots. Dev Cell 52, 183–195.e184 (2020). [PubMed: 31883775] 39. Cartwright HN, Humphries JA, Smith LG, PAN1: a receptor-like protein that promotes polarization of an asymmetric cell division in maize. Science 323, 649–651 (2009). [PubMed: 19179535] 40. Yoshida S et al. , Genetic control of plant development by overriding a geometric division rule. Dev Cell 29, 75–87 (2014). [PubMed: 24684831] 41. Lechler T, Mapelli M, Spindle positioning and its impact on vertebrate tissue architecture and cell fate. Nat Rev Mol Cell Biol 22, 691–708 (2021). [PubMed: 34158639] 42. Kiyomitsu T, The cortical force-generating machinery: how cortical spindle-pulling forces are generated. Curr Opin Cell Biol 60, 1–8 (2019). [PubMed: 30954860] 43. Nir I et al. , Evolution of polarity protein BASL and the capacity for stomatal lineage asymmetric divisions. Curr Biol 32, 329–337.e325 (2022). [PubMed: 34847354] 44. Molines AT et al. , Physical properties of the cytoplasm modulate the rates of microtubule polymerization and depolymerization. Dev Cell 57, 466–479.e466 (2022). [PubMed: 35231427] 45. Ramalho JJ, Jones VAS, Mutte S, Weijers D, Pole position: How plant cells polarize along the axes. Plant Cell 34, 174–192 (2022). [PubMed: 34338785] 46. Gutierrez R, Lindeboom JJ, Paredez AR, Emons AM, Ehrhardt DW, Arabidopsis cortical microtubules position cellulose synthase delivery to the plasma membrane and interact with cellulose synthase trafficking compartments. Nat Cell Biol 11, 797–806 (2009). [PubMed: 19525940] 47. Roeder AH et al. , Variability in the control of cell division underlies sepal epidermal patterning in Arabidopsis thaliana. PLoS Biol 8, e1000367 (2010). [PubMed: 20485493] 48. Mills AM, Rasmussen CG, Defects in division plane positioning in the root meristematic zone affect cell organization in the differentiation zone. J Cell Sci 135, (2022). 49. Adrian J et al. , Transcriptome dynamics of the stomatal lineage: birth, amplification, and termination of a self-renewing population. Dev Cell 33, 107–118 (2015). [PubMed: 25850675] 50. Nakamura S et al. , Gateway binary vectors with the bialaphos resistance gene, bar, as a selection marker for plant transformation. Biosci Biotechnol Biochem 74, 1315–1319 (2010). [PubMed: 20530878] 51. Davies KA, Bergmann DC, Functional specialization of stomatal bHLHs through modification of DNA-binding and phosphoregulation potential. Proc Natl Acad Sci U S A 111, 15585–15590 (2014). [PubMed: 25304637] H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 10 Figure 1. The Arabidopsis stomatal lineage polarity domain overrides default division rules during asymmetric cell division. A. (Left) Image of the abaxial epidermis from a 3dpg BRXL2p::BRXL2-YFP (magenta) ML1p::mCherry-RCI2A (black) cotyledon. Cells have been pseudo-colored according to the color-scheme below. Scale bar-25μm. (Right) Selected polarized cells highlighting the morphological diversity of cells in this lineage. Scale bar-5μm. B. Stomata (purple) are created through coupled divisions (asymmetric, ACD and symmetric, SCD) and fate transitions. (Top) Illustration of the major fate transitions within the Arabidopsis stomatal lineage; meristemoids in green, and guard mother cell in blue. (Bottom) Illustration of the known functions for the BASL/BRXf domain. Before division (in SPCH+ cells), cortical BASL/BRXf promotes ACD and instructs nuclear migration. Post-division (in stomatal lineage ground cells, SLGCs), inherited BASL/BRXf/YODA phosphorylates MAPK, leading to suppression of SPCH. Mechanisms controlling singular polarity domain inheritance are needed. C. The assay used to quantify Δ θ°, the angle between the calculated shortest wall and the division plane. This assay was used for the data shown in Fig.1, D to I. For additional details, see Materials and Methods. D. Quantification of Δ θ° during wild-type ACDs. n = 95 cells. E. The percent of total wild-type ACDs where polarity did (27%) or did not (73%) conflict with the predicted shortest wall. F-G. (Left) Distribution of Δ θ° in ACDs where the predicted shortest wall did (G) or did not (F) conflict with the polarity domain. (Right) Three examples of ACDs from their respective classes with associated Δ θ°s. The black dotted line marks the calculated shortest wall, and Science. Author manuscript; available in PMC 2023 July 07. Muroyama et al. Page 11 the magenta arrow indicates the polarity domain. Scale bar-5μm. Kolmogorov-Smirnov test comparing the Δ θ°s distributions in the two ACD classes: p < 0.0001. H. Quantification of Δ θ° during progenitor divisions in basl. Kolmogorov-Smirnov test with WT ACDs: p=0.0005. n = 94 cells. I. Examples of three early-lineage divisions in basl with associated Δ θ°s. The black dotted line marks the predicted shortest wall. Scale bar-5μm. J. Representative images of a medial (left) and maximum projection (right) view of preprophase band (PPB) placement relative to the polarity domain in BRXL2p::BRXL2- YFP TMMp::mCherry-TUA5. Black arrows indicate the PPB. Scale bar-5μm. K. Reslice of the cell in (J) showing microtubule and BRXL2 distribution along the anticlinal wall. Black arrows indicate the PPB. Scale bar-5μm. L. Line scan along the cell cortex of the asymmetrically dividing cell shown in (J). The arrows indicate the position of the PPB. M. Microtubule distribution (gray lines) in PPB-forming cells (n=26 cells, average shown in black), aligned to the midpoints of the BRXL2 crescents (magenta line shows the mean ± standard deviation). H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 12 Figure 2. Singular inheritance of the BASL/BRXf polarity domain via preprophase band placement is required for stomatal patterning. A. Stills from representative time-lapse movies showing cell division orientation in Col-0 (top) and trm678 (bottom), relative to the BRXL2-YFP-marked cortical polarity domain. Scale bars-10μm. B. Quantification of Col-0 (n=112) and trm678 (n=116) asymmetric divisions (ACDs) with singular inheritance (BRXL2 inherited by only one daughter) or double inheritance (BRXL2 inherited by both daughter cells). C. Representative time-course images, with associated cartoons, showing an instance of double inheritance and cell fate stalling 62 hours later from a trm678 seedling. Scale bar-25μm. D. Representative images of the epidermis of 7dpg Col-0 and trm678 cotyledons expressing ML1p::mCherry-RCI2A. Stomata are pseudo-colored purple. Scale bars-25μm E. Quantification of stomatal densities (number of stomata per 581.82μm x 581.82μm area) in 7dpg Col-0 and trm678 cotyledons (35 seedlings each). Unpaired t-test — p=0.0002. F. Quantification of the percent of paired stomata per 581.82μm x 581.82μm area in 7dpg Col-0 and trm678 cotyledons (35 seedlings each). Unpaired t-test — p=0.0012. Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 13 Figure 3. The BASL/BRXL2 domain locally depletes cortical microtubules from the plasma membrane. A. (Top) Representative medial section of a polarized stomatal lineage ground cell (SLGC) and associated meristemoid in BRXL2p::BRXL2-YFP TMMp::mCherry-TUA5. (Bottom) Reslice showing microtubule and BRXL2 distribution along the anticlinal face of the same cell. Scale bars-5μm. B. Line scan along the cell periphery of the SLGC shown in (A) showing the normalized BRXL2 and TUA5 fluorescence intensities. C. Microtubule distribution in SLGCs (gray lines) (n=72 cells), aligned to the midpoints of the BRXL2 crescents (magenta line corresponds to the mean signal ± standard deviation). The black line is the average of the plotted microtubule signals. D. (Top) Representative medial section of stomatal lineage cells in basl TMMp::YFP-TUA5. (Bottom) Reslice showing microtubule distribution along the anticlinal face of the same cell. Scale bars-5μm. E. Line scan along the cell cortex of the larger daughter cell in (D) showing the normalized TUA5 fluorescence intensity. F. Microtubule distribution in basl SLGCs (gray lines) (n=58 cells). G. Asymmetric microtubule depletion scores, calculated from integrated fluorescence intensities in polarized and unpolarized cortical domains (see Materials and Methods) in wild-type (n=50), basl (n=40), brx-quad (n=31), and BASLp::MYR-BRX (n=22) SLGCs. One-way ANOVA with Tukey’s post hoc test– **-p < 0.0001. Science. Author manuscript; available in PMC 2023 July 07. H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t H H M I A u t h o r M a n u s c r i p t Muroyama et al. Page 14 Figure 4. Polarized BASL destabilizes microtubule plus-ends to locally bias array organization. A. Representative images of the apical surfaces in epidermal cells of 35Sp::GFP-BASL- IC 35Sp::mCherry-TUA5 hypocotyls. The boxed regions below highlight the microtubule organization within the polarized domains of two cells. Scale bars-10μm. B. Kymographs showing microtubule plus-end dynamics within a non-polarized region (left) and an ectopic polarity domain (right). Asterisks indicate microtubule catastrophes and rescues. Scale bar-1μm. C. Local microtubule depletion within apical BASLectopic domains and comparably sized random regions in control hypocotyls. The local microtubule depletion was derived using 35Sp::mCherry-TUA5 fluorescence (see Materials and Methods). n=25 hypocotyl cells for each. Unpaired t-test — p < 0.0001. D-F. Quantification of microtubule plus-end dynamics within non-polarized (NP) and polarized (P) apical domains. Growth persistence (D), polymerization rate (E) and depolymerization rate (F) were quantified. The numbers above the box-and-whisker plots are mean values ± standard deviation. For all comparisons, unpaired t-tests were used — p < 0.0001. Science. Author manuscript; available in PMC 2023 July 07.
10.1093_nar_gkad331	Published online 1 May 2023 Nucleic Acids Research, 2023, Vol. 51, Web Server issue W419–W426 https://doi.org/10.1093/nar/gkad331 PANGEA: a new gene set enrichment tool for Drosophila and common research organisms Yanhui Hu 1 , 2 ,* , Aram Comjean 1 , 2 , Helen Attrill 3 , Giulia Antonazzo 3 , Jim Thurmond 4 , Weihang Chen 1 , 2 , Fangge Li 2 , Tiffany Chao 2 , Stephanie E. Mohr 1 , 2 , Nicholas H. Brown 3 and Norbert Perrimon 1 , 2 , 5 ,* 1 Department of Genetics, Blavatnik Institute, Harvard Medical School, Harvard University, Boston, MA 02115, USA, 2 Drosophila RNAi Screening Center, Harvard Medical School, 77 Avenue Louis Pasteur, Boston, MA 02115, USA, 3 Department of Physiology, Development and Neuroscience, University of Cambridge, Downing Street, Cambridge CB2 3DY, UK, 4 Department of Biology, Indiana University, Bloomington, IN 47405, USA and 5 Ho w ard Hughes Medical Institute, Boston, MA 02138, USA Received February 21, 2023; Revised March 28, 2023; Editorial Decision April 17, 2023; Accepted April 29, 2023 ABSTRACT GRAPHICAL ABSTRACT Gene set enrichment analysis (GSEA) plays an im- portant role in large-scale data analysis, helping sci- entists discover the underlying biological patterns o ver -represented in a gene list resulting fr om, f or e xample, an ‘omics’ stud y. Gene Ontology (GO) an- notation is the most frequently used classiﬁcation mechanism for gene set deﬁnition. Here we present a new GSEA tool, PANGEA (PAthwa y, Netw ork and Gene-set Enrichment Analysis; https://www.ﬂyrnai. org/ tools/ pangea/ ), developed to allow a more ﬂexi- ble and conﬁgurable approach to data analysis us- ing a variety of classiﬁcation sets. PANGEA allows GO analysis to be performed on different sets of GO annotations, for e xample e xcluding high-throughput studies. Beyond GO, gene sets for pathway anno- tation and protein complex data fr om v arious re- sources as well as expression and disease anno- tation from the Alliance of Genome Resources (Al- liance). In addition, visualizations of results are en- hanced by providing an option to view network of gene set to gene relationships. The tool also al- lows comparison of multiple input gene lists and ac- companying visualisation tools for quick and easy comparison. This new tool will facilitate GSEA for Drosophila and other major model organisms based on high-quality annotated inf ormation av ailable f or these species. INTRODUCTION Modern genetics and genomics owe much to work done us- ing common model organisms. These models continue to make a significant contribution to the understanding of de- velopment, metabolism, neuroscience, behaviour and dis- ease. With the onset of the ‘big data’ era has come a need for analysis platforms that deconvolute complex data from multispecies studies. Model organism databases (MODs) are knowledgebases dedicated to the cur ation, stor age and integration of species-specific data for their r esear ch com- munity. The past decade has seen a number of efforts aimed at pulling together model organism and human data to facilitate a more inter disciplinary approach; e xamples in- clude MARRVEL (Model organism Aggregated Resources for Rare Variant ExpLoration) ( 1 ), Gene2Function ( 2 ) and * To whom correspondence should be addressed. Tel: +1 6174327672; Fax: +1 6174327688; Email: [email protected] Correspondence may also be addressed to Yanhui Hu. Tel: +1 6174327672; Fax: +1 6174327688; Email: claire [email protected] C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. W420 Nucleic Acids Research, 2023, Vol. 51, Web Server issue the Monarch Initiati v e ( 3 ). Furthermore, the Alliance of Genome Resources (Alliance) ( 4 ), a consortium of se v en model organism databases and the Gene Ontology (GO) Consortium (GOC), formed recently with the objecti v e of building an umbrella resource from which users can navi- gate combined data within a single integrated knowledge- base. To help support such a r esour ce, the MODs ar e work- ing to reduce the di v ergence in the way the primary data is curated, stor ed and pr esented to facilita te compara ti v e and translational r esear ch. Although these integrated r esour ces allow sear ch and comparison across certain data classes, large-scale data analysis remains in the domain of stand-alone bioinfor- matic tools such as DAVID ( 5 ), TermMapper ( 6 ), GOrilla ( 7 ), PANTHER Gene List Analysis ( 8 ), WebGestalt ( 9 ) and g:Profiler ( 10 ), with a focus on processing gene list to extract a statistical measure of shared biological features, usually termed gene set enrichment analysis (GSEA). The most frequently used gene set classification in GSEA is GO annotation, which is based on the most widely-used on- tology (a hierarchical controlled vocabulary) in biological r esear ch for the wild-type molecular function(s), biologi- cal process(es) and cellular component(s) associated with a gi v en gene product ( 11 , 12 ). Many GSEA tools also incor- pora te classifica tions from other sources such as Reactome ( 13 ) and KEGG ( 14 ) pathways (e.g. DAVID, WebGestalt and g:Profiler) and, for human studies, there may be addi- tional data sources such as the Human Phenotype Ontol- ogy (HPO) ( 15 ) and Online Mendelian Inheritance in Man (OMIM) ( 16 ) (e.g. WebGestalt). The addition of gene sets beyond the GO allows users to extract more classification information, as well as seek trends and overlap in the en- riched sets. DAVID and g:Profiler are two of the few re- sources that make it possible for users to compare differ- ent sets on the same display; howe v er, the ability to interact with the results post-processing at these r esour ces is rather limited. Despite the abundance of tools, we found that they do not fully meet community needs, primarily because they were overly focused on human gene data and were not us- ing the most up to date data on other species. For exam- ple, Reactome pa thway annota tion is based on computa- tional predictions deri v ed from manually curated human pathways. Ther e ar e a few or ganism-tar geted analysis tools; the prokaryote-centred GSEA FUNAGE-Pro ( 17 ) is one example in which the underlying knowledgebase was as- sembled to cater to the needs of a specific r esear ch commu- nity. A number of useful gene classification r esour ces (e.g. pathways , complexes , gene groups) have been developed by the Drosophila RNAi Screening Center (DRSC) and Fly- Base, the Drosophila knowledgebase ( 18–22 ). Furthermore, in FlyBase and indeed across the MODs, ther e ar e se v eral types of curated data in common, including disease mod- els, phenotypes and gene expression, which could be used for GSEA. In contrast to the annotation of gene function data with GO, which is done in a consistent manner across multiple organisms, other data types ar e r epr esented within the MODs in di v erse ways, reflecting some of the technical differences in the genetics of these organisms. The Alliance was founded to integra te da ta across many MODs ( 4 ) and now provides a source of harmonised data that can also be used for GSEA. To take full advantage of the r esear ch in di v erse model organisms, we describe our creation of a new tool that we name PANGEA. Although our primary focus was on Drosophila genes, we de v eloped PANGEA to also include r at, mouse, zebr afish, nematode worm data, as well as harmonised human data to facilitate translational r esear ch. PANGEA not only incorporates additional gene set classifications from Alliance and MODs, but also have implemented the features that enhance the presentation of enrichment results by allowing the user to select sets and compare them visually to facilitate interpretation as well as making it easy to do parallel GSEA for multiple gene lists. This fle xibility enab les users to adapt the tool to their needs and allow ‘fortuitous’ discovery by widening the pool of knowledge for the purpose of analysis. MATERIALS AND METHODS Building the knowledgebase for PANGEA The gene set classification is a way to group genes based on commonality such as the same biological pathway. We have collected > 300 000 gene sets from various public r esour ces (Tables 1 , 2 ) for fruit fly D. melanogaster , the nematode worm C. elegans , the zebrafish D. rerio , the mouse M. mus- culus , the rat R. norvegicus and human H. sapiens . For an- notations based on a controlled vocabulary arranged in hi- erar chical structur e, such as gene group and phenotype an- notations from FlyBase, gene-to-gene set relationships were assembled after the hierar chical structur e was flattened. An exception was made for GO annotations, which were as- sembled in two ways, with and without being flattened, al- lowing users to choose which output is used in the anal- ysis. GO annotations include evidence codes that indicate the type of evidence supporting the annotation. For exam- ple, ‘IDA’ means that an annotation was supported by a dir ect assay, wher eas ‘ISS’ means that the annotation was inferred from sequence or structural similarity. Using such evidence codes, GO gene sets were built with additional configurations: (i) subsets based on experimental evidence codes, i.e. excluding annotations only based on phyloge- netic, sequence or structural similarity and other computa- tional analyses (IEA, IBA, IBD, IKR, IRD, ISS, ISO, ISA, ISM, IGC, RCA); (ii) subsets excluding annotations only supported by high-throughput (HTP) evidence codes (HTP, HDA, HMP, HGI and HEP); (iii) subsets of GO generic terms (GO slim) provided by the GOC ( http://geneontology. or g/docs/do wnload-ontology/#subsets ); (iv) subsets of very high-le v el GO term classifications used by FlyBase and the Alliance originally generated to support GO summary rib- bon displays. For Drosophila phenotype annotations from FlyBase, we assembled the gene-to-phenotype association using the ‘genotype phenotype data’ file available in the FlyBase Downloads page, in which phenotypes are associ- ated with individual genotypes and controlled vocabulary identifiers are indicated. This allowed us to extract only those genotypes where we could be certain that the phe- notype was associated with the perturbation of a single Drosophila melanogaster gene (i.e. single classical or in- sertional alleles). Because different resources use different gene or protein identifiers, we used an in-house mapping Nucleic Acids Research, 2023, Vol. 51, Web Server issue W421 pr ogram to synchr onise IDs to NCBI Entrez gene IDs, offi- cial gene symbols and gene identifiers of species-specific re- sources, such as MGI and FlyBase (Table 1 ). The PANGEA knowledgebase stores the information of gene set classifica- tion, gene annotation obtained from NCBI as well as the inf ormation f or ID mapping among various r esour ces. Building gene sets of pr eferr ed tissue expression To study the di v ersity and dynamics of the Drosophila transcriptome, the modENCODE consortium sequenced the transcriptome in twenty-nine dissected tissues ( 23 ) and the processed datasets are available at FlyBase ( http: //ftp.flybase.net/r eleases/curr ent/pr ecomputed files/genes/ ). A program was implemented in the Python programming language to identify genes expressed at a substantially higher le v els in one tissue versus any other tissue. This pr ogram first gr oups the RNA-seq datasets based on tissue. For example, all data related to the nervous system are grouped together. It then calculates the average reads per kilobase per million mapped r eads (RPKM) expr ession values for each gene in each tissue group. Genes were identified as pr efer entially expr essed in a gi v en tissue group if their average expression in the tissue group is 3-fold or higher than the av erage e xpression in any other tissue group. Genes with average RPKM value lower than 10 were excluded. Genes defined in this way as ‘tissue-specific’ then get annotated with the relevant tissue to generate the tissues expression classification set. Datasets used for testing Drosophila cell RNAi screen phenotype data was obtained from the DRSC ( https://www.flyrnai.org/ ) via download of a file of all available public screen ‘hits’ (results) ( https: //www.flyrnai.org/RN Ai all hits.txt ). RNAi reagents of op- timal design were selected. The criteria for optimal design were no CAN or CAR repeat, fewer than six predicted OTEs (off-target alignment sites of 19 bp) and a single gene target. CAN and CAR r epeats ar e thr ee base tandem r e- peats such as CAA CAGCA CCAT (CAN repeat, the third position can be A, G, C or T) and CAACA GCA GCAA (CAR repeat, the 3rd position can be A or G). RNAi r eagents wer e mapped to curr ent FlyBase gene identifiers using a DRSC internal mapping tool. Screens focused on major signalling pathways were selected for PANGEA anal- ysis ( 24–29 ). Proteomics data was obtained from Tang et al. ( 30 ) and high-confident prey proteins identified by mass-spec (Supplementary Table S2) were used for analy- sis. Gene set enrichment statistics used at PANGEA Hypergeometric testing is performed to calculate P val- ues for GSEA using the PypeR function in R. Bonfer- roni correction for multiple statistical tests, Benjamini- Hochberg procedure for false discovery rate adjustment, and Benjamini-Yekutieli procedure for false discovery rate adjustment were performed using the p.adjust function in R. Web tool implementation PANGEA is a SaaS (Software as a Service) w e b tool ( https: //www.flyrnai.org/tools/pangea/ ) and is built following a three-tier model, with a w e b-based user-interface at the front end, the knowledgebase at the backend, and the busi- ness logic in the middle tier communicating between the front and back ends by matching input genes with gene sets, doing statistical analysis and building visualization graphs. The front page is written in PHP using the Sym- f on y frame wor k and front-end HTML pages using the Twig template engine. The JQuery JavaScript library is used to facilitate Ajax calls to the back end, with the DataTables plugin f or displa ying tab le vie ws and Cytoscape and Veg- aLite packages for the da ta visualisa tions. The Bootstrap frame wor k and some custom CSS are used on the user in- terface. A mySQL database is used to store the knowledge- base. Both the w e bsite and databases are hosted on the O2 high-performance computing cluster, which is made avail- able by the Research Computing group at Harvard Medical School. RESULTS Pr epar ation of the classified gene sets for GSEA: the PANGEA knowledgebase GSEA relies on high-quality annotation of genes / gene products with information related to their biological func- tions. For PANGEA, we used multiple sources of annota- tion to generate > 300 000 different classes of gene func- tion for fiv e major model organisms ( D. melanogaster, C. elegans, D. rerio, M. musculus, R. norvegicus ) and human. For example, pathway annotations allow users to identify metabolic or signalling pathways that are over-represented in a gene list and help understand causal mechanisms un- derlying the observed phenotype from a scr een. P athway annotations from KEGG, PantherDB, and Reactome, as well as manually curated Drosophila gene sets, such as Fly- Base Signalling Pathways and the DRSC PathON annota- tion ( 18 , 21 ), are included in the PANGEA knowledgebase. The GO annotation set provides the comprehensi v e knowledge on gene functions and we store the gene-to-gene set relationships from GO in two ways. One is the direct gene-to-GO term associations as obtained from the gene association file while the other stores the gene-to-GO term associa tions with considera tion gi v en to child-par ent r ela- tionships. The latter is recommended for use in GSEA as it reflects the intended use of the ontology in curation prac- tice. The direct, gene-to-term set may be useful to under- stand the depth of annotation for each gene. In addition, we also generated two gene annotation subsets using evi- dence codes. The ‘experimental data only’ subset includes only those gene associations that are supported by experi- mental evidence codes. The ‘excluding high-throughput ex- periments’ subset excludes annotations only supported by HTP evidence codes. Excluding HTP data may be impor- tant to avoid bias when analysing similar studies ( 31 ). GO slim subsets are the cut-down versions of GO that give a broad ov ervie w of the ontology content without the detail of the specific, fine-grained terms. The PANGEA knowl- W422 Nucleic Acids Research, 2023, Vol. 51, Web Server issue Table 1. Species coverage by PANGEA and corresponding species-specific databases Species Abbreviation Species specific database URL Example Drosophila melanogaster Homo sapiens Mus musculus Caenorhabditis elegans Danio rerio Rattus norvegicus dm hs mm ce dr rn FlyBase https://flybase.org wg, FBgn0284084 HGNC MGI WormBase ZFIN RDG https://www.genenames.org http://www.informatics.jax.org/ https://www.wormbase.org https://zfin.org https://rgd.mcw.edu/ WNT1, HGNC:12774 Wnt1, MGI:98953 cwn-1, WBGene00000857 wnt1,ZDB-GENE-980526–526 Wnt1, RGD:1597195 Table 2. Knowledgebase of PANGEA built from various gene annotation r esour ces Type Source URL Species covered at PANGEA Gene Ontology GO http://geneontology.org/ hs,mm,rn,dr,dm,ce pathway pathway pathway pathway pathway group group group protein protein KEGG REACTOME PantherDB FlyBase pathway PathON HGNC FlyBase gene group GLAD COMPLEAT EBI protein complex phenotype AGR disease phenotype expression FlyBase phenotype AGR expression https://www.genome.jp/kegg/ https://reactome.org/ http://www.pantherdb.org/ https://flybase.org/ https: //www.flyrnai.org/tools/pathon/ https://www.genenames.org/ https://flybase.org/ https: //www.flyrnai.org/tools/glad/ https: //www.flyrnai.org/compleat/ https: //www.ebi.ac.uk/complexportal https: //www.alliancegenome.org/ https://flybase.org/ https: //www.alliancegenome.org/ hs,mm,rn,dr,dm,ce hs,mm,rn,dr,dm,ce dm dm dm hs dm dm dm hs,mm,rn,dr,dm,ce hs,mm,rn,dr,dm,ce dm mm,rn,dr,dm,ce Source update frequency irregular, usually 1–2 months unknown unknown irregular 2 months irregular unknown 2 months irregular irregular 2 months 3–4 months 2 months 3–4 months edgebase includes two sets of GO slim annotations from differ ent r esour ces. In addition to GO and pa thway annota tions, MODs provide organism-specific curation of important aspects of gene information, such as gene expression and mutant phe- notype, that are not captured in GO. The Alliance is focused on the harmonisation and centralisation of major MODs data ( 4 , 32 ). To take advantage of this effort, we integrated gene-to-tissue expression and gene-to-disease (model) asso- cia tion annota tions from the Alliance into the PANGEA kno wledgebase. As all or ganisms in the Alliance use the Disease Ontology (DO) for annotation, this set is easily comparable across species. The Alliance DO annotation set also includes disease association to model organism genes via an electronic pipeline using orthology with human dis- ease genes which expands the set provide by the MODs. Moreover, for Drosophila genes we assembled an additional gene set from phenotype annota tion a t FlyBase by extract- ing phenotype data associated with a ‘single allele’ genotype (i.e. single classical or insertional alleles), allowing users to perform meaningful enrichment analyses on this data class for the first time. Also included in PANGEA are gene group classifica- tion (eg. kinases and transcription factors) from organism- specific r esour ces (human and fly), protein complex anno- tations for multiple organisms from the EMBL-EBI Com- plex Portal ( 33 ) and COMPLEAT ( 22 ) and bespoke gene sets using Drosophila modENcode RNAseq data to iden- tify genes particularly highly expressed in one tissue (see Materials and Methods). In summary, we have assembled > 300 000 different gene sets that can be used in PANGEA to assess the enrichment of particular biological features in an input gene list. Features of the PANGEA user interface GSEA can be computationally intensi v e because of the number of gene sets being tested and potentially large num- ber of genes entered by users. Ther efor e, the step of pre- processing user’s input by mapping the input gene identi- fiers to the gene identifiers used for gene set annotation, is set up as a standalone ID mapping page (accessed by click- ing ‘Gene Id Mapping’ on the top toolbar) instead of com- bining it with the analysis step. Gene identifiers supported by PANGEA include Entrez Gene IDs, official gene sym- bols and primary gene identifiers from MODs. Users might need to analyse lists of other identifiers such as UniPro- tKB IDs and Ensembl gene IDs. Users can use ‘Gene Id Mapping’ tool and select an organism of choice to map IDs. As gene annotation is an on-going process, the gene identifiers as well as gene symbols might change over time. Even with the same type of gene identifiers such as FlyBase gene ID, the IDs used by users might be from a different FlyBase r elease. Ther efor e, ID-mapping step is an optional Nucleic Acids Research, 2023, Vol. 51, Web Server issue W423 Figure 1. Example of analysing a single gene list using PANGEA. A proteomic interaction dataset was selected from a study of the m6A methyltr ansfer ase complex MTC ( 30 ). The 75 high-confidence interactors of the four subunits of the MTC (METTL3, METTL14, Fl(2)d and Nito) identified by affinity- purified mass spectrometry from Drosophila S2R+ cells were submitted via the ‘Search Single’ option at PANGEA and enrichment analysis was performed over phenotype, GO SLIM2 BP and protein complex annotation from COMPLEAT (literature based). The result was filtered using P value 1 × 10 −5 cut-off and was illustrated as ( A ) a bar graph and ( B ) a network graph of selected gene sets from phenotype annotation and GO SLIM2 BP. Triangle nodes r epr esent gene sets and circle nodes represent genes while the edges r epr esent gene to gene set associations. ( C ) A network graph for selected gene sets from phenotype annotation and COMPLEAT protein complex annotation (literature based). but recommended first step to ensure that the entered IDs are synchronized with the IDs used by PANGEA gene set annotation. Users of FlyBase may also directly export a ‘HitList’ of genes generated in FlyBase to the tool by se- lecting the ‘PANGEA Enrichment Tool (DRSC)’ from the dropdown ‘Export’ menu (Supplementary Figure S1). An option for users to upload a background gene list for the analysis is provided; this may be useful when analysing hits from a focused screen using a kinase sub-library instead of a genome-scale library, for example. PANGEA identi- fies all relevant gene sets and provides enrichment statis- tics such as P -values, adjusted P -values, and fold enrich- ment, as well as the genes shared by the input gene list and gene set members. Users have the option to set differ- ent P -value cut-offs and visualise the results using a bar graph, the height and colour intensity of which can be customised (Figure 1 A). In addition, users can select gene sets of interest to examine the overlap of genes in differ- ent gene sets using the ‘Gene Set Node Graph’ visualisa- tion option. Nodes of different shapes in the network indi- cate genes or gene sets while edges reflect the gene-to-gene set relationship. This type of visualisation can help users identify the most relevant genes in each gene set as well as commonly shared or distinct gene members of the selected gene sets (Figure 1 B, C). An under-appreciated use of GSEA tools is that re- searchers often use them as simple gene classification tools, for example, asking ‘which genes in my list are kinases?’ to help inform further computational or experimental analy- sis. Having di v erse classification sets is important because depending on the type of data / experiment being analysed, different gene sets may be more useful than others. It is often useful to be able to compare similar gene sets from different sources to help evaluate the evidence for support. In addi- tion, PANGEA not only reports genes in an enriched gene set but also reports genes not covered by the gene set cate- gory selected, which may be interesting because of their lack of characterisation. This feature can help user answer ques- tion like ‘which genes in my list are not covered by KEGG annotation?’. W424 Nucleic Acids Research, 2023, Vol. 51, Web Server issue Figure 2. Examples of analysing multiple gene lists using PANGEA. ( A ) The prey proteins of multiple baits from AP mass-spec dataset ( 30 ) were submitted via the ‘Search Multiple’ option at PANGEA. The gene sets of protein complex annotation from COMPLEAT were selected. The comparison of enrichment over annotated protein complexes from the interacting proteins of four different baits was illustrated using a heatma p. ( B ) RN Ai screen data of signalling pathway studies were obtained from DRSC RNAi data repository ( 24 ) and the hits were submitted via the ‘Search Multiple’ option at PANGEA. The gene sets of FlyBase pathway annotation were used. The comparison of the enrichment of signalling pathway components from the screen hits of fiv e studies was illustrated using a heatmap. Often users need to anal yse m ultiple gene lists and com- par e the r esults; howe v er, majority of current w e b-based tools only allow the analysis of a single input list (plus back- ground). Thus, users have to perform comparisons manu- ally or using different tools. To address this need, PANGEA allows users to input multiple gene lists and compare results directly via a heatmap or a dot plot visualisation. For exam- ple, users might input gene hits from different phenotypic scr eens and compar e wha t pa thways, gene groups or biolog- ical processes that are common or different among results (Figure 2 ). Testing the utility of PANGEA To test the utility of PANGEA, we first analysed a pro- teomic interaction dataset from a study of the m6A methyl- tr ansfer ase complex MTC ( 30 ). In this study, using individ- ual pull-downs of the four subunits (METTL3, METTL14, Fl(2)d and Nito) of the MTC complex, high-confidence interactors were identified by mass-spectrometry from Drosophila S2R + cells. Submitting the combined list of 75 interacting proteins from the four baits via the ‘Search Single’ option at PANGEA (accessed by click- ing ‘Search Single’ on the top toolbar) for Fly and per- forming GSEA over phenotype, SLIM2 GO BP as well as protein complex annotation from COMPLEAT (liter- ature based) enrichment, we identified mRNA metabolic pr ocess (GO:0016071), pr otein folding (GO:0006457), ab- nor mal sex-deter mination (FBcv:0000436), abnor mal neu- roanatomy (FBcv:0000435), CCT complex and Spliceo- some complex among the top enriched gene sets with the −5 ) (Figure 1 A). most significant p-values (all < 1 × 10 Next, we visualised SLIM2 GO BP and phenotype annota- tions using a network gra ph. GO mRN A metabolic process hits overlap with both abnormal sex-determination and ab- normal neuroanatomy phenotypes, but GO protein folding hits onl y overla p with the abnormal neuroanatomy pheno- type (Figure 1 B). We also visualised protein complex and phenotype annotations using a different network graph, showing Spliceosome hits overlap with the phenotypes of abnor mal sex-deter mination and abnor mal neuroanatomy while the CCT complex hits only overlap with the abnor- Nucleic Acids Research, 2023, Vol. 51, Web Server issue W425 mal neuroanatomy phenotype (Figure 1 C). Enrichment of se x / reproducti v e phenotypes align with known function of MT C in r egulating the splicing of female-specific Sex lethal (Sxl) and its roles in alternati v e splicing and se xual dimor- phism, as well as the germ stem cell dif ferentia tion in the ovary ( 34 ). These GSEA results are also concordant with the fact that MTC is also known to have a significant role in neuronal mRNA regulation. The benefit of the network visualization is apparent when viewing how the gene set as- signment overlap (Figure 1 B, C), which re v eals that some of the MTC interacting proteins are associated with abnor- mal neuroanatomy phenotype and that the mechanism of the association is through the CCT complex in the pro- cess of protein folding. In contr act, the inter acting pro- teins from Spliceosome have more broad impacts related to both abnormal neuroanatomy phenotype and abnormal sex-determination phenotypes through mRNA metabolic process. We further analysed the protein complexes associated with each individual subunit using the ‘Search multiple’ option at PANGEA (accessed by clicking ‘Search Multi- ple’ on the top toolbar) and inputting the interacting pro- tein lists for each bait, then compared the enrichment re- sult using a heatmap visualisation (Figure 2 A). The re- sults indicated that some complexes, such as spliceosome subunits, are common to all MTC subunits, whereas some ar e mor e specific, such as protein complex es CCT com- plex for METTL14 and METTL13. In addition, we fur- ther analysed phenotype enrichment for proteins associated with each individual subunit using the ‘Search multiple’ op- tion, and comparison of the enrichment results shows many over lapping phenotypes, particular ly with regard to sterility (Supplementary Figure S2). At another use case, we looked at phenotypic cell screen data. Large-scale RN A interference (RN Ai) screening is a powerful method for functional studies in Drosophila . At the DRSC, datasets generated from more than one hundred scr eens ar e pub licly availab le ( 24 ). We selected fiv e screens designed to identify the genes for major signalling pathways and performed a GSEA analysis of the hits using the multi- ple gene list enrichment function of PANGEA. FlyBase sig- nalling pathway gene sets were selected and the results of the fiv e scr eens wer e compar ed side-by-side using a heatmap, w hich clearl y illustrated enrichment of the core components of the corresponding pathways, as well as potential cross- talks between pathways (Figure 2 B). These use cases of PANGEA for phenotype screening data as well as proteomics data demonstrate the value of the tool in validating screen results as well as generating new hypotheses for further study. DISCUSSION GSEA is a computational method used to identify sig- nificantly over-r epr esented gene classes within an input gene list(s) by testing against gene sets assembled based on prior knowledge. Input gene lists are typically from high- throughput screens or analyses. Here we present PANGEA, a newly developed GSEA tool with major model organ- isms as its focus, that includes gene sets that are usually not utilized by other GSEA tools, such as expression and disease annotations from the Alliance, phenotype annota- tions from FlyBase, and GO subsets with different configu- rations. PANGEA is easy to use and has new features such as allowing enrichment analyses for multiple input gene lists and gener ating gr aphical outputs that make comparisons straightf orward f or users. In addition to the use cases pre- sented here, i.e. analysing phenotypic screening and pro- teomic data, we anticipate that the tool will also facilitate analysis of gene lists from other types of data. For exam- ple, analysis of single-cell RNA-seq datasets at PANGEA might help users identify pathways and biological processes that are characteristic of various cell types. Users will also be able to answer questions on classification such as, ‘which genes in this list are kinases?’. PANGEA is designed to ac- commodate a wide range of biological data types and ques- tions, providing users with a w e b-based analysis tool that is easily accessible and user-friendly. We also note that gene classifications are not static, and the generic design of the tool means that it will be easy to update or expand PANGEA for more gene set classification and / or more species. In de v eloping PANGEA, we sought to improv e the effecti v eness of GSEA by (i) providing multiple collections of genes classified by their function in different ways (classified gene sets); (ii) ensuring the data underly- ing the classification of gene function was up to date and (iii) improving the visualization so that results from multi- ple gene sets or multiple gene lists could be compared easily. DA T A A V AILABILITY The online r esour ce is available without restriction at https: //www.flyrnai.org/tools/pangea/ . SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We would like to thank the members of the Perrimon lab- oratory, the FlyBase consortium, the Drosophila RNAi Screening Center (DRSC), and the Transgenic RNAi Project (TRiP) for the discussion and suggestions during the design and implementation of the tool as well as the feed- back during the tool testing. Additional thanks to Gil dos Santos (Harvard, US) and Gillian Millburn (Cambridge, UK) at FlyBase for their genotype-to-phenotype work. [P41 GM132087]; FlyBase FUNDING NIH / NIGMS grant NIH / NHGRI [U41HG000739]; UK Medical Research Council [MR / W024233 / 1]; N.P. is an investigator of Howard Hughes Medical Institute. Funding for open access charge: NIH / NIGMS grant [P41 GM132087]. Conflict of interest statement. None declared. REFRENCES 1. Wang,J., Al-Ouran,R., Hu,Y., Kim,S.Y., Wan,Y.W., Wangler,M.F., Yamamoto,S., Chao,H.T., Comjean,A., Mohr,S.E. et al. (2017) W426 Nucleic Acids Research, 2023, Vol. 51, Web Server issue MARRVEL: integration of Human and Model Organism Genetic Resources to Facilitate Functional Annotation of the Human Genome. Am. J. Hum. Genet. , 100 , 843–853. 2. Hu,Y., Comjean,A., Mohr,S.E., FlyBase,C. and Perrimon,N. (2017) Gene2Function: an Integrated Online Resource for Gene Function Discovery. G3 (Bethesda) , 7 , 2855–2858. 3. Shefchek,K.A., Harris,N.L., Gargano,M., Matentzoglu,N., Unni,D., Brush,M., Keith,D., Conlin,T., Vasilevsky,N., Zhang,X.A. et al. (2020) The Monarch Initiati v e in 2019: an integrati v e data and analytic platform connecting phenotypes to genotypes across species. Nucleic Acids Res. , 48 , D704–D715. 4. Alliance of Genome Resources Consortium (2022) Harmonizing model organism data in the Alliance of Genome Resources. Genetics , 220 , https://doi.org/10.1093/genetics/iyac022 . 5. Sherman,B.T., Hao,M., Qiu,J., Jiao,X., Baseler,M.W., Lane,H.C., Imamichi,T. and Chang,W. (2022) DAVID: a w e b server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res. , 50 , W216–W221. 6. Boyle,E.I., Weng,S., Gollub,J., Jin,H., Botstein,D., Cherry,J.M. and Sherlock,G. (2004) GO::TermFinder–open source software for accessing Gene Ontology information and finding significantly enriched Gene Ontology terms associated with a list of genes. Bioinformatics , 20 , 3710–3715. 7. Eden,E., Navon,R., Steinfeld,I., Lipson,D. and Yakhini,Z. (2009) GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinf. , 10 , 48. 8. Mi,H., Muruganujan,A., Huang,X., Ebert,D., Mills,C., Guo,X. and Thomas,P.D. (2019) Protocol update for large-scale genome and gene function analysis with the PANTHER classification system (v.14.0). Nat. Protoc. , 14 , 703–721. 9. Liao,Y., Wang,J., Jaehnig,E.J., Shi,Z. and Zhang,B. (2019) WebGestalt 2019: gene set analysis toolkit with revamped UIs and APIs. Nucleic Acids Res. , 47 , W199–W205. 10. Raudvere,U., Kolberg,L., Kuzmin,I., Arak,T., Adler,P., Peterson,H. and Vilo,J. (2019) g:profiler: a w e b server for functional enrichment analysis and conversions of gene lists (2019 update). Nucleic Acids Res. , 47 , W191–W198. 11. Gene Ontology, C. (2021) The Gene Ontology r esour ce: enriching a GOld mine. Nucleic Acids Res. , 49 , D325–D334. 12. Ashburner,M., Ball,C.A., Blake,J.A., Botstein,D., Butler,H., Cherry,J.M., Davis,A.P., Dolinski,K., Dwight,S .S ., Eppig,J.T. et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet. , 25 , 25–29. 13. Gillespie,M., Jassal,B., Stephan,R., Milacic,M., Rothfels,K., Senff-Ribeiro,A., Griss,J., Sevilla,C., Matthews,L., Gong,C et al. (2022) The reactome pathway knowledgebase 2022. Nucleic Acids Res. , 50 , D687–D692. 14. Kanehisa,M., Furumichi,M., Sato,Y., Kawashima,M. and Ishiguro-Watanabe,M. (2023) KEGG for tax onom y-based analysis of pathways and genomes. Nucleic Acids Res. , 51 , D587–D592. 15. Kohler,S., Gargano,M., Matentzoglu,N., Carmody,L.C., Le wis-Smith,D., Vasile vsky,N.A., Danis,D., Balagura,G., Baynam,G., Brower,A.M. et al. (2021) The Human Phenotype Ontology in 2021. Nucleic Acids Res. , 49 , D1207–D1217. 16. Amberger,J.S. and Hamosh,A. (2017) Searching Online Mendelian Inheritance in Man (OMIM): a Knowledgebase of Human Genes and Genetic Phenotypes. Curr Protoc Bioinformatics , 58 , 1 2 1–1 2 12. 17. de Jong,A., Kuipers,O.P. and Kok,J. (2022) FUNAGE-Pro: comprehensi v e w e b server for gene set enrichment analysis of prokaryotes. Nucleic Acids Res. , 50 , W330–W336. 18. Gramates,L.S., Agapite,J., Attrill,H., Calvi,B.R., Crosby,M.A., Dos Santos,G., Goodman,J.L., Goutte-Ga tta t,D., Jenkins,V.K., Kaufman,T. et al. (2022) Fly Base: a guided tour of highlighted features. Genetics , 220 , iyac035. 19. Attrill,H., Falls,K., Goodman,J.L., Millburn,G.H., Antonazzo,G., Rey,A.J ., Marygold,S.J . and FlyBase,C. (2016) FlyBase: establishing a Gene Group r esour ce for Drosophila melanogaster. Nucleic Acids Res. , 44 , D786–D792. 20. Hu,Y., Comjean,A., Perkins,L.A., Perrimon,N. and Mohr,S.E. (2015) GLAD: an Online Database of Gene List Annotation for Drosophila. J Genomics , 3 , 75–81. 21. Ding,G., Xiang,X., Hu,Y., Xiao,G., Chen,Y., Binari,R., Comjean,A., Li,J., Rushworth,E., Fu,Z et al. (2021) Coordination of tumor growth and host wasting by tumor-deri v ed Upd3. Cell Rep. , 36 , 109553. 22. Vinayagam,A., Hu,Y., Kulkarni,M., Roesel,C., Sopko,R., Mohr,S.E. and Perrimon,N. (2013) Protein complex-based analysis framework for high-throughput data sets. Sci. Signal , 6 , rs5. 23. Brown,J.B., Boley,N., Eisman,R., May,G.E., Stoiber,M.H., Duff,M.O., Booth,B.W., Wen,J., Park,S., Suzuki,A.M. et al. (2014) Di v ersity and dynamics of the Drosophila transcriptome. Nature , 512 , 393–399. 24. Hu,Y., Comjean,A., Rodiger,J., Liu,Y., Gao,Y., Chung,V., Zirin,J., Perrimon,N. and Mohr,S.E. (2021) Fl yRN Ai.org-the database of the Drosophila RNAi screening center and transgenic RNAi project: 2021 update. Nucleic Acids Res. , 49 , D908–D915. 25. Nybakk en,K., Vok es,S.A., Lin,T.Y., McMahon,A.P. and Perrimon,N. (2005) A genome-wide RNA interference screen in Drosophila melanogaster cells for new components of the Hh signaling pathway. Nat. Genet. , 37 , 1323–1332. 26. DasGupta,R., Kaykas,A., Moon,R.T. and Perrimon,N. (2005) Functional genomic analysis of the Wnt-wingless signaling pathway. Science , 308 , 826–833. 27. Baeg,G.H., Zhou,R. and Perrimon,N. (2005) Genome-wide RNAi analysis of JAK / STAT signaling components in Drosophila. Genes Dev. , 19 , 1861–1870. 28. Kockel,L., Kerr,K.S., Melnick,M., Bruckner,K., Hebrok,M. and Perrimon,N. (2010) Dynamic switch of negati v e feedback regulation in Drosophila Akt-TOR signaling. PLos Genet. , 6 , e1000990. 29. Friedman,A.A., Tucker,G., Singh,R., Yan,D., Vinayagam,A., Hu,Y., Binari,R., Hong,P., Sun,X., Porto,M et al. (2011) Proteomic and functional genomic landscape of receptor tyrosine kinase and ras to extracellular signal-regulated kinase signaling. Sci. Signal , 4 , rs10. 30. Tang,H.W., Weng,J.H., Lee,W.X., Hu,Y., Gu,L., Cho,S., Lee,G., Binari,R., Li,C., Cheng,M.E et al. (2021) mTORC1-chaperonin CCT signaling regulates m(6)A RNA methylation to suppress autophagy. Proc. Natl. Acad. Sci. U.S.A. , 118 , e2021945118. 31. Attrill,H., Gaudet,P., Huntley,R.P., Lovering,R.C., Engel,S.R., Poux,S., Van Auken,K.M., Georghiou,G., Chibucos,M.C., Berardini,T.Z. et al. (2019) Annotation of gene product function from high-throughput studies using the Gene Ontology. Database (Oxford) , 2019 , baz007. 32. Alliance of Genome Resources, C. (2020) Alliance of Genome Resources Portal: unified model organism research platform. Nucleic Acids Res. , 48 , D650–D658. 33. Meldal,B.H.M., Perfetto,L., Combe,C., Lubiana,T., Ferreira Cavalcante,J .V., Bye,A.J .H., Waagmeester,A., Del-Toro,N., Shriv astav a,A., Barrera,E. et al. (2022) Complex Portal 2022: new curation frontiers. Nucleic Acids Res. , 50 , D578–D586. 34. Lence,T., Akhtar,J., Bayer,M., Schmid,K., Spindler,L., Ho,C.H., Kreim,N., Andrade-Navarro,M.A., Poeck,B., Helm,M. et al. (2016) m(6)A modulates neuronal functions and sex determination in Drosophila. Nature , 540 , 242–247. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
10.1093_hmg_ddad091	Human Molecular Genetics, 2023, Vol. 32, 16, 2600–2610 https://doi.org/10.1093/hmg/ddad091 Advance access publication date 1 June 2023 Original Article Continuous, but not intermittent, regimens of hypoxia prevent and reverse ataxia in a murine model of Friedreich’s ataxia Tslil Ast 1,2,3,4,†,‡, Hong Wang1,2,3,4,‡, Eizo Marutani5, Fumiaki Nagashima5, Rajeev Malhotra6, Fumito Ichinose5 and Vamsi K. Mootha1,2,3,4,* 1Broad Institute, Cambridge, MA 02142, USA 2Howard Hughes Medical Institute, Massachusetts General Hospital, Boston, MA 02114, USA 3Department of Molecular Biology, Massachusetts General Hospital, Boston, MA 02114, USA 4Department of Systems Biology, Harvard Medical School, Boston, MA 02115, USA 5Department of Anesthesia, Critical Care, and Pain Medicine, Massachusetts General Hospital, Boston, MA 02114, USA 6Cardiology Division, Department of Medicine, Massachusetts General Hospital, Boston, MA 02114, USA *To whom correspondence should be addressed at: MGH Department of Molecular Biology, 185 Cambridge Street, Boston, MA 02114, USA. Tel: +1 6176439710; Fax: +1 6177265735; Email: [email protected] †Present address: Department of Biomolecular Sciences, The Weizmann Institute of Science, Rehovot 7610001, Israel. ‡These authors contributed equally Abstract Friedreich’s ataxia (FA) is a devastating, multi-systemic neurodegenerative disease affecting thousands of people worldwide. We previously reported that oxygen is a key environmental variable that can modify FA pathogenesis. In particular, we showed that chronic, continuous normobaric hypoxia (11% FIO2) prevents ataxia and neurological disease in a murine model of FA, although it did not improve cardiovascular pathology or lifespan. Here, we report the pre-clinical evaluation of seven ‘hypoxia-inspired’ regimens in the shFxn mouse model of FA, with the long-term goal of designing a safe, practical and effective regimen for clinical translation. We report three chief results. First, a daily, intermittent hypoxia regimen (16 h 11% O2/8 h 21% O2) conferred no benefit and was in fact harmful, resulting in elevated cardiac stress and accelerated mortality. The detrimental effect of this regimen is likely owing to transient tissue hyperoxia that results when daily exposure to 21% O2 combines with chronic polycythemia, as we could blunt this toxicity by pharmacologically inhibiting polycythemia. Second, we report that more mild regimens of chronic hypoxia (17% O2) confer a modest benefit by delaying the onset of ataxia. Third, excitingly, we show that initiating chronic, continuous 11% O2 breathing once advanced neurological disease has already started can rapidly reverse ataxia. Our studies showcase both the promise and limitations of candidate hypoxia-inspired regimens for FA and underscore the need for additional pre-clinical optimization before future translation into humans. Introduction Friedreich’s ataxia (FA) is the most common monogenic mito- chondrial disease and also the most common autosomal recessive ataxia, impacting 1 in 50 000 people worldwide (1–3). FA is a multi-systemic, neurodegenerative disease, characterized primar- ily by progressive spinocerebellar and sensory ataxia that presents between 5 and 20 years of age (1). FA patients also develop other symptoms, including hypertrophic cardiomyopathy, diabetes, and scoliosis. While the neurological symptoms tend to be the most debilitating, cardiac dysfunction is ultimately the leading cause of death in FA, leading to premature mortality at a median age of 37.5 years (4,5). This monogenic disease is caused by depletion in the nuclear-encoded mitochondrial protein, frataxin (FXN) (6). FXN accelerates the mitochondrial biosynthesis of iron–sulfur (Fe–S) clusters (7,8), which are essential and versatile redox cofac- tors. Because Fe–S clusters are embedded in over 60 human pro- teins (9,10), which span diverse pathways, FXN deficiency results in manifold cellular biochemical pathologies and multisystemic disease. Recently, Omaveloxolone received FDA clearance for FA, mak- ing it the first drug approved for any monogenic mitochondrial disease. Omaveloxolone works to pharmacologically restore the reduced antioxidant buffering capacity observed in FA, by stabi- lizing the antioxidant master regulator NRF2 (11,12). However, it should be noted that the effect size of NRF2-targeted therapy is modest. Ongoing ‘precision medicine’ efforts aim to target the root defect of FA with gene therapy, by boosting endogenous FXN transcription or via protein replacement (13), though deliv- ery and toxicity of FXN over-expression remain challenges (14). Other therapies targeting the secondary consequences of FA have been tested in clinical trials without success (13,15). Additional treatment modalities are needed to ensure safe and efficacious therapies that yield meaningful outcomes. We previously have shown that oxygen is a potent environ- mental modulator of FA in several model systems including yeast, worms, human cells, and even mice (16). Specifically, we showed that hypoxia preserves Fe–S cluster levels and boosts their biosynthesis, thereby addressing the primary deficiency in Received: October 6, 2022. Revised: May 8, 2023. Accepted: May 22, 2023 © The Author(s) 2023. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. FA. We also showed that chronic, continuous treatment of shFxn mice (17) with 11% O2 could prevent the onset of ataxia in this murine model of FA. This improvement was present at both early (12 weeks) and later (15 weeks) timepoints of disease progression. While the neurologic phenotype was dramatically prevented by housing the mice in chronic, continuous 11% O2, their cardiac pathology was not improved. shFxn mice housed in hypoxia showed identical survival curves as their normoxic counterparts, consistent with the notion that cardiac pathology determines the lifespan of these mice. Similarly, we have previously shown that in the Ndufs4 KO mouse model of Leigh syndrome, chronic, continuous 11% O2 can prevent the onset of neurological disease, but ultimately, those mice also prematurely succumb to their cardiac pathology (18). While chronic, continuous 11% O2 is extremely effective in preventing the onset of neurological disease in the shFxn mouse model, this regimen is not readily translated into humans, and a pressing question is whether more practical hypoxia regimens may be effective. Moreover, an important question is whether chronic continuous hypoxia initiated in advanced disease can reverse neurological phenotypes. Because hypoxia itself can be dangerous, it is important to rigorously address these questions in a pre-clinical setting. In this work, we explored seven different ‘hypoxia-inspired’ regimens in the murine shFxn model with the goal of identifying regimens that are safe, practical, and effective. Results Here, we sought to build on our foundational observation that chronic, continuous 11% O2 initiated early in disease (Regimen 0) can delay the onset of neurological disease in the shFxn mouse model (16). Specifically, we were interested in identifying alter- native regimens (Regimens 1–7) that might be similarly safe and effective but more clinically practical than our original approach. To this end, we tested seven additional hypoxia-inspired regi- mens in the shFxn mouse model (Table 1), that include: (Regimen 1) intermittent hypoxia (16 h of 11% O2/8 h of 21% O2 daily) to determine if this more practical regimen might be effective; (Regi- men 2) Intermittent hypoxia in combination with the drug PT2399, to blunt the polycythemic response and potentiate the hypoxic effects; (Regimen 3) Chronic, continuous 17% O2 to determine if milder hypoxia is effective; (Regimen 4) Chronic, continuous 17% O2 in combination with PT2399, to enhance the effects of this mild hypoxic regimen; (Regimen 5) Anemia, induced by phlebotomy and an iron-deficient diet, as a gas-free way to create systemic hypoxia; (Regimen 6) Genetic ablation of hepcidin, to mimic sys- temic iron homeostasis changes that occur in hypoxia; (Regimen 7) Chronic, continuous 11% O2 initiated in advanced disease, to determine if hypoxia can be initiated later to reverse advanced ataxia. For each of these regimens, we performed a battery of tests. We monitored body weight (no significant changes) and lifespan. We monitored Hct/Hgb as pharmacodynamic marker of hypoxia therapy. Given our keen interest in neurological status, which previously showed great improvement with hypoxia, we focused on rotarod analysis testing at key timepoints that have been previously established for this mouse model (17). For selected regimens we performed cardiac echocardiography. For all mice we also measured cardiac Gdf15 mRNA, a marker of the integrated stress response (ISR) that we find tracks with disease progres- sion (Supplementary Material, Fig. S1). In line with this result, other laboratories interrogating the shFxn mouse heart have found notable ISR activation (19) and GDF15 has been used as a secreted Human Molecular Genetics, 2023, Vol. 32, No. 16 \| 2601 disease marker in preclinical studies of FXN gene therapy (20). The results from these eight regimens (including Regimen 0, reported previously) are summarized in Table 1. Here we highlight the major findings emerging from this study. Intermittent 11% O2 is harmful to the shFxn mouse, shortening lifespan and worsening cardiac stress We previously reported the therapeutic potential of chronic, con- tinuous 11% O2 (Regimen 0) in preventing onset of ataxia. How- ever, translating this regimen to humans has significant practical challenges as it would require a patient to permanently reside within a low oxygen environment. Intermittent hypoxia exposure (e.g. using sleeping tents enriched with nitrogen), on the other hand, is widely applied in the field of sports training, making it a more accessible and potentially more practical approach. To test the benefits of intermittent hypoxia (Regimen 1), shFxn or control mice were exposed to 11% O2 for only 16 h daily, between 4 p.m. and 8 a.m., while the animals were nocturnally active. This regimen elicited a physiological hypoxic response, as evidenced by the elevated hematocrit present in both WT and shFxn animals (Fig. 1A). During this study, we observed a significantly shortened lifespan in shFxn animals breathing intermittent hypoxia as com- pared with their normoxic counterparts (Fig. 1B). Considering this detrimental effect, we tested whether the onset of ataxia is also accelerated in the intermittent hypoxia regimen, but we did not see any substantial motor-behavioral deficiencies as assessed by accelerating rotarod 6 weeks post-doxycycline induction (Fig. 1C). Moreover, after 12 weeks of doxycycline administration, when ataxia has previously been shown to manifest in the shFxn mice (17), accelerating rotarod analysis revealed similar deficits in shFxn mice breathing normoxia or intermittent hypoxia. Thus, ataxia does not appear to be hastened or worsened by intermittent hypoxic breathing. However, Gdf15 mRNA levels were significantly higher in the hearts of shFxn mice breathing intermittent 11% O2 when compared with continuous 21% O2, pointing to elevated car- diac ISR in this hypoxic regimen. We conclude that intermittent hypoxia is detrimental in the context of FA, specifically to cardiac physiology. Pharmacological blockade of the polycythemic response to intermittent hypoxia blunts cardiac ISR and accelerated mortality We hypothesized that the detrimental effects observed for shFxn mice breathing intermittent hypoxia might be the outcome of unwanted transient tissue hyperoxia, i.e. the combination of a higher hematocrit (Fig. 1A) coupled with daily exposure to 8 h of 21% O2. To test this hypothesis, we utilized an established HIF- 2α antagonist, PT2399, as HIF-2α stabilization is a key driver of hypoxic polycythemia (21,22). We performed twice daily adminis- tration of PT2399 with our intermittent hypoxia regimen and eval- uated the effects of this combination (Regimen 2). Indeed, PT2399 treatment blunted hypoxic polycythemia in all mice (Fig. 2A), returning the hematocrit measures to the range observed for mice breathing 21% O2 (Fig. 1A). Consistent with our hypothesis, PT2399 blunted the accelerated mortality in intermittent hypoxia (P-value: 0.14) when compared with mice treated with the vehi- cle (Fig. 2B). The benefit of combined PT2399 and intermittent hypoxia also extended to improved accelerating rotarod abili- ties 12 weeks post-doxycycline induction (Fig. 2C). However, this enhancement in motor-behavioral capabilities was transient and could not be observed 15 weeks post doxycycline administra- tion (Fig. 2D). Cardiac Gdf15 mRNA levels were also diminished 2602 \| Human Molecular Genetics, 2023, Vol. 32, No. 16 Table 1. Summary of hypoxia regimensapplied to the shFxn mouse model Regimen Number Intervention Protocol Ataxia (versus 21% O2) Lifespan (versus 21% O2) Cardiomyopa- thy (versus 21% O2) Heart GDF15 mRNA (versus 21%O2) 0 1 2 3 4 5 6 7 Chronic 11% O2-Prevention Intermittent 11% O2 Intermittent 11% O2 + PT2399 Mild Hypoxia-17% O2 Mild Hypoxia-17% O2 + PT2399 Anemia Hepcidin-FXN double mutants Chronic 11% O2-Reversal Chronic housing at 11% O2 at time of doxycycline administration Housing in intermittent 11% and 21% O2 (16h on hypoxia/8h off) at time of doxycycline administration Housing in intermittent 11% and 21% O2 (16h on hypoxia/8h off) combined with PT2399 administered twice daily at 100mg/kg Chronic housing at 17% O2 at time of doxycycline administration Chronic housing at 17% O2 at time of doxycycline administration combined with PT2399 administered twice daily at 100 mg/kg Induction of extreme anemia (phlebotomy+low iron diet) prior to doxycycline administration Genetic ablation of hepcidin combined with doxycycline administration Chronic housing at 11% O2 12-weeks post-initiation of doxycycline Improved at all timepoints Unchanged Unchanged Unchanged Worsened Not analyzed Improved at early time point (vs. int. hypoxia) Improved at early time point Improved at early time point Unchanged Not analyzed Unchanged Not analyzed Unchanged Not analyzed Unchanged Unchanged Not analyzed Unchanged Worsened Worsened Improved at all timepoints Unchanged Not analyzed ↑ (cid:2) ↑ ↑ ↑ ↑ (cid:2) ↑ Reference (Ast et al, 2019) This work (Fig. 1) This work (Fig. 2) This work (Fig. 3) This work (Fig. 4) This work (Fig. 5) This work (Fig. 6) This work (Fig. 7) in animals treated with PT2399 in intermittent hypoxia com- pared with their vehicle controls, indicating that cardiac ISR was dampened (likely by blunting cardiac hyperoxia). Owing to these observations, we speculate that secondary polycythemia (owing to intermittent hypoxia) combined with intermittent 21% FIO2 is causing adverse effects in the shFxn mouse, perhaps analogous to daily bouts of reperfusion injury. shFxn mice continuously breathing 17% O2 delays the onset of ataxia We next tested whether chronic breathing of a milder hypoxia regimen would be as beneficial as 11% O2. To this end, we initiated chronic, continuous breathing of 17% O2 (Regimen 3), which would be the equivalent partial pressure of oxygen 1600 m above sea level. This exposure to mild hypoxia resulted in a more modest, but still significant, boost in hematocrit (Fig. 3A), and no adverse effect to lifespan (Fig. 3B). Intriguingly, even mild hypoxia was sufficient to improve the motor-behavioral abilities of shFxn mice 12 weeks post-doxycycline administration (Fig. 3C). However, this beneficial effect in hypoxia did not extend to 15 weeks (Fig. 3D), indicating that 17% O2 acts to delay but not altogether prevent ataxia. Consistent with FA lifespan being primarily driven by cardiac pathology, we found that 17% O2 did not have any effect on cardiac Gdf15 levels for the shFxn mice (Fig. 3E). Collectively, chronic, continuous breathing of 17% O2 could not fully recapit- ulate the more profound benefits of 11% O2 in the shFxn mice, in agreement with our previous work in the Leigh mouse model (18). Given that mild, chronic continuous 17% O2 provided a modest benefit early in disease and elevated the hematocrit, we examined the effects of combining 17% O2 with daily PT2399 administration (Regimen 4) in an attempt to further potentiate hypoxia. While PT2399 could again blunt polycythemia in the context of mild hypoxia (Fig. 4A) and delay the onset of ataxia (Fig. 4C), it did not confer an added benefit in preventing late-stage ataxia (Fig. 4D). Likewise, it did not have any benefit to the lifespan or lower Gdf15 levels (Fig. 4B and E). Anemia is not beneficial to the shFxn mouse while ablation of hepcidin, an iron-regulatory hormone, is detrimental to lifespan We then turned to gas-free approaches that are ‘hypoxia inspired’. First, we tested the effects of anemia (Regimen 5), as this treat- ment would lower tissue oxygen delivery and could be achieved practically through a combination of phlebotomy and a low-iron diet. This approach showed strong efficacy in a mouse model of Leigh syndrome (23). Indeed, we could induce anemia to a similar extent in both WT and shFxn animals through this combined treatment (Fig. 5A). Once the desired hemoglobin concentration of ≤ 5 g/dl was achieved, red blood cell production was blunted by maintaining the animals on a low-iron diet for the remainder of the experiment. Anemia did result in a lower partial pressure of brain oxygen (Fig. 5B). However, anemia did not improve lifespan, ataxia, or Gdf15 levels (Fig. 5C–F). Human Molecular Genetics, 2023, Vol. 32, No. 16 \| 2603 Figure 1. Intermittent hypoxia does not prevent ataxia, shortens lifespan and exacerbates cardiac ISR activation. (A) Hematocrit measurements from WT and shFxn mice housed in 21% O2 or intermittent 11% O2 for 5 weeks. (B) Survival of WT or shFxn mice housed in 21% O2 or intermittent 11% O2. (C, D) Accelerating rotarod analysis for WT or shFxn mice housed in 21% O2 or intermittent 11% O2 at 6 and 12 weeks. Latency to fall measured as mean value of triplicate trials per mouse E. Cardiac Gdf15 mRNA levels at 12 weeks, normalized to Tbp and 21% O2 WT mice. All bar plots show mean ± SD. ∗∗∗∗ Numbers represent group sizes. = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. = P < 0.001, = P < 0.05, = P < 0.01, ∗∗∗ ∗∗ ∗ Figure 2. Blunting the polycythemic response prevents the detrimental effects of intermittent hypoxia. (A) Hematocrit measurements from WT and shFxn mice housed in intermittent 11% O2 with or without daily dosing of PT2399 for 5 weeks. (B) Survival of WT or shFxn mice housed at intermittent 11% O2 with or without PT2399 treatment. (C, D) Accelerating rotarod analysis for WT or shFxn mice housed at intermittent 11% O2 with or without PT2399 treatment at 12 and 15 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (E) Cardiac Gdf15 mRNA levels at 12 weeks, normalized to Tbp and WT mice housed at intermittent 11% O2.All bar plots show mean ± SD. Numbers represent group sizes. = P < 0.01, ∗∗∗ = P < 0.05, ∗∗∗∗ ∗∗ ∗ = P < 0.001, = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. 2604 \| Human Molecular Genetics, 2023, Vol. 32, No. 16 Figure 3. Mild, chronic hypoxia (17% O2) prevents ataxia at early timepoints. (A) Hematocrit measurements from WT and shFxn mice housed in 21% O2 or 17% O2 for 6 weeks. (B) Survival of WT or shFxn mice housed in 21% O2 or 17% O2. (C, D) Accelerating rotarod analysis for WT or shFxn mice housed in 21% O2 or 17% O2 at 12 and 15 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (E) Cardiac Gdf15 mRNA levels at 12 weeks, normalized to Tbp and 21% O2 WT mice. All bar plots show mean ± SD. Numbers represent group sizes. = P < 0.001, ∗∗∗∗ = P < 0.05, = P < 0.01, ∗∗∗ ∗∗ ∗ = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. Figure 4. Blunting the polycythemic response upon mild chronic hypoxia (17% O2) prevents ataxia at early timepoints but does not result in additional benefits. (A) Hemoglobin measurements of shFxn mice housed in 21% O2 or 17% O2 with or without PT2399 treatment for 5 weeks. (B) Survival of shFxn mice housed in 21% O2 or 17% O2 with or without PT2399 treatment. (C, D) Accelerating rotarod analysis for shFxn mice housed in 21% O2 or 17% O2 with or without PT2399 treatment at 12 and 15 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (E) Cardiac Gdf15 mRNA levels at 12 weeks, normalized to Tbp and 21% O2 shFxn mice. All bar plots show mean ± SD. Numbers represent group sizes. = P < 0.01, ∗∗∗ = P < 0.05, ∗∗∗∗ ∗∗ ∗ = P < 0.001, = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. The lack of efficacy we observed with anemia was surprising but in line with our previous cell culture results, where we found that hypoxia is only beneficial to FXN null cells when iron uptake pathways are activated (16). While iron accumulation is a well- established hallmark of FA, it remains unclear whether it is a driver of pathophysiology (24). For example, previous studies have shown that blunting iron uptake via IRP1-knockout is deleterious in a mouse model of hepatic FXN deficiency; in this context, iron accumulation acts to maintain residual Fe–S cluster levels and mitochondrial function (25). We were therefore curious to see what would happen if we increased systemic iron uptake in the shFXN mouse. To this end, we tested the effects of hepcidin knock- out (Hamp−/−) (26) in the shFXN mouse (Regimen 6). Hepcidin is a central regulator of systemic iron homeostasis, whose absence Human Molecular Genetics, 2023, Vol. 32, No. 16 \| 2605 Figure 5. Chronic anemia neither blunts ataxia nor improves lifespan. (A) Hemoglobin measurements of WT or shFxn mice following serial phlebotomy every 2 to 3 days for 21 days, in combination with an Fe-deficient diet. (B) Brain PO2 (vestibular nuclei) of WT or shFxn mice that are untreated or made anemic using phlebotomy, in combination with an Fe-deficient diet. (C) Survival of untreated or anemic WT and shFxn mice. (D, E) Accelerating rotarod and untreated or anemic WT and shFxn mice at 12 and 15 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (F) Cardiac Gdf15 mRNA levels at 13 weeks, normalized to Tbp and untreated WT mice. All bar plots show mean ± SD. Numbers represent group sizes. = P < 0.05, ∗∗ ∗∗∗∗ ∗∗∗ ∗ = P < 0.01, = P < 0.001, = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. triggers multi-systemic iron accumulation (including in the heart and brain) and elevated serum iron content (27). Moreover, hep- cidin levels are known to decline in response to hypoxia (28,29), nominating it as a candidate effector of hypoxia therapy in the shFXN mouse. Hamp−/−/shFxn mice exhibited significantly shorter lifespans, aggravated cardiomyopathy as assessed by echocardio- gram, and higher levels of cardiac Gdf15 mRNA (Fig. 6A and C–E). At early timepoints at which premature mortality was already observed, the Hamp−/−/shFxn did not demonstrate any significant motor-behavioral deficits based on accelerating rotarod (Fig. 6B). Together, these results would indicate that iron dysregulation caused by hepcidin ablation was detrimental to the heart, though we did not see evidence of worsening of the FA neuropathology. Initiating chronic, continuous hypoxia late in disease can rapidly reverse ataxia We next sought to determine whether hypoxia could not only prevent, but also reverse neurological disease in the shFxn mouse model. All of the hypoxia regimens, described before, were preven- tive in nature, i.e. they were initiated at the same time as doxy- cycline treatment and Fxn knockdown thereby acting to delay or prevent disease. However, a clinically relevant question is whether an intervention can reverse FA symptoms following their onset, as most patients are typically diagnosed only after they have presented with early signs and symptoms of ataxia. We previously showed that hypoxia can both prevent and reverse neurological disease in the Leigh syndrome Ndufs4 KO mouse model (18). More- over, it is notable that in the initial description of the doxycycline- inducible shFxn mouse model, it was reported that many of the disease features are reversible (17), i.e. restoring Fxn expression by halting doxycycline administration could reverse many FA pathologies, including ataxia. To determine if hypoxia could also reverse advanced neuro- logical disease, we developed Regimen 7, in which we initiated chronic, continuous 11% O2 breathing after the onset of neuro- logical disease. We first administered doxycycline for 12 weeks to WT or shFxn animals housed in normoxia. At this timepoint, shFxn animals present significant motor-behavioral deficits (17). Then, animals were randomized into two groups: (a) doxycycline administration was halted (leading to the re-expression of Fxn and mimicking potential future gene therapy approaches) and (b) doxycycline treatment maintained while the animals were switched to chronic 11% O2. Cessation of doxycycline led to an extended lifespan, significant improvement in rotarod tests after 6 weeks, and lower cardiac Gdf15 mRNA (Fig. 7B, C and E), consis- tent with the original description of this mouse model (17). Initiat- ing chronic, continuous 11% O2 hypoxia was sufficient to rapidly reverse the motor-behavioral deficit present in shFxn mice, so that already at 3 weeks post-treatment a significant improvement by accelerating rotarod could be observed (Fig. 7C). As with Regimen 0, initiating hypoxia at this late stage did not improve survival or cardiac Gdf15 levels (Fig. 7B and E). It is notable that from a kinetic perspective, the reversal of ataxia by initiating chronic, continuous hypoxia was faster than that of doxycycline cessation, which mimics the effects of systemic gene replacement therapy. Discussion We previously demonstrated in human cell culture, yeast, worm, and mouse models of FA that oxygen is a potent disease-modifier: low oxygen could delay neurological disease, and conversely, high ambient oxygen was detrimental. Indeed, studies in FXN deficient yeast (30–32) and a more recent report using mouse embryonic fibroblasts with the FA-associated FxnG127V mutation (33) have also observed a similar rescue in fitness in hypoxia. Moreover, this hypoxic restoration of bioavailable iron and Fe–S cluster content was recently validated using Mössbauer spectroscopy in a yeast FA model (34). In the shFxn mouse, we previously reported that chronic continuous breathing of 11% O2 could prevent the onset of ataxia, the primary feature of FA. The goal of our present study was to explore seven additional hypoxia-related or inspired ther- apies in the shFxn model—with the long-term goal of designing more practical and effective regimens that harness the power of hypoxia. Given that hypoxia itself can be dangerous, an important 2606 \| Human Molecular Genetics, 2023, Vol. 32, No. 16 Figure 6. Genetic ablation of hepcidin reduces the lifespan and exacerbates the cardiac stress of shFxn mice. (A) Survival of single or double KO mice. (B) Accelerating rotarod analysis for single or double KO mice at 7 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (C) Cardiac Gdf15 mRNA levels at 6–7 weeks, normalized to Tbp and WT mice. (D) Echocardiogram measurement of left ventricular internal diameter at end-systole from single or double KO mice (16). (E) Echocardiogram measurement of cardiac output from single or double KO mice. All bar plots show mean ± SD. Numbers represent group sizes. = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. = P < 0.001, = P < 0.01, = P < 0.05, ∗∗∗∗ ∗∗∗ ∗∗ ∗ Figure 7. Initiating chronic, continuous 11% O2 breathing in late-stage disease can rapidly reverse ataxia. (A) Hematocrit measurements from WT and shFxn mice housed in 21% O2 or 11% O2 for 3 weeks, following 12 weeks of doxycycline treatment. (B) Survival of WT or shFxn mice housed in 21% O2 or 11% O2. (C, D) Accelerating rotarod analysis for WT or shFxn mice upon doxycycline removal or hypoxia treatment at 15 and 18 weeks. Latency to fall measured as mean value of triplicate trials per mouse. (E) Cardiac Gdf15 mRNA levels at 19 weeks, normalized to Tbp and 21% O2 WT mice. All bar plots ∗∗∗ show mean ± SD. Numbers represent group sizes. = P < 0.0001. Two-way ANOVA with Bonferroni’s post-test. = P < 0.001, = P < 0.05, = P < 0.01, ∗∗∗∗ ∗∗ ∗ consideration for any future regimen and an explicit goal of the current study was to establish the safety of all regimens tested. A natural question was whether intermittent hypoxia might represent a safe, efficacious, and practical approach—however, upon housing the shFxn mouse in intermittent hypoxia (Regimen 1), of 16 h 11% O2/8h 21% O2, we observed striking excess mor- tality and an elevation in the cardiac ISR biomarker Gdf15. We hypothesize that these conditions might generate daily hyperoxia exposure for the shFxn mouse, by coupling 21% O2 breathing with polycythemia that would amplify delivery of oxygen in bursts. In support of this hypothesis, by eliminating the hypoxia-driven polycythemic response using the HIF-2 inhibitor PT2399 (Regimen 2), we could blunt the detrimental effects of intermittent hypoxia on lifespan and presumed cardiac stress which was ref lected by an elevated cardiac Gdf15. A similar detrimental effect on lifes- pan and presumed cardiac toxicity (ref lected by elevated cardiac Gdf15 mRNA) was observed upon hepcidin ablation (Regimen 6), which induces iron accumulation in the heart and brain (among other tissues). Iron overload related to hemojuvelin deficiency, a bone morphogenetic protein co-receptor required for hepcidin expression, leads to increased myocardial oxidative stress and is associated with pathologic cardiac hypertrophy and fibrosis (35). Iron loading in the cardiomyocyte results specifically in mitochondrial iron accumulation and dysfunction from oxidative stress (36). While iron is predicted to more safely accumulate in chronic continuous hypoxia (Regimen 0); in Regimen 1, the shFxn animals were exposed to daily bouts of 21% O2, which could in theory lead to oxygen toxicity via the Fenton reaction. One interpretation from these collective observations is that FA cardiac stress might be driven by an elevated vulnerability of this tissue to the interplay of oxygen and iron. The interaction of these two factors is likely giving rise to excess oxidative stress, and perhaps even triggering ferroptosis (37,38), in the cardiac tissue. Conversely, one of our regimens required the use of iron depri- vation to achieve anemia (Regimen 5) as a means to decrease oxygen delivery. While anemia indeed led to tissue deoxygenation (Fig. 5B), it had no benefit to the ataxic phenotype or lifespan of the shFxn mice. This result is in stark contrast with the significant disease rescue that was observed in the Ndufs4 mouse model of Leigh syndrome treated with anemia (23). However, the lack of efficacy of Regimen 5 in the FA model is consistent with our prior human cell culture results (16), where we showed that hypoxia is only beneficial in FXN null cells when coupled with elevated bioavailable iron uptake. Thus, it is possible that the hypoxic conditions generated by an iron deficient diet and phlebotomy were ineffective owing to the inherently iron-depriving nature of this regimen. Given that intermittent hypoxia (Regimen 1) and anemia (Reg- imen 5) were neither safe nor effective in this mouse model, we turned to other potential regimens that might also be more practical than Regimen 0. We tested the effects of chronic 17% O2 breathing (Regimen 3), an oxygen tension found just 1600 m above sea level (e.g. Boulder, CO in the US or Johannesburg in South Africa). Encouragingly, this intervention delayed the onset of ataxia but could not fully recapitulate the longer-term benefits we observed with 11% O2 treatment, and this effect could not be further improved when combined with PT2399 administration (Regimen 4). These findings match our previous cellular work (16), in which we observed that mild hypoxia treatment resulted in a partial rescue of the growth of FXN null human cells, indicating that there is some oxygen threshold that must be met to fully cover for loss of FXN. Perhaps the most encouraging finding of this current study is that chronic, continuous hypoxia initiated after disease onset (Regimen 7) can reverse ataxia in the murine FA model. The original report introducing the doxycycline inducible shFxn knock- down model had already reported the reversibility of many of the disease phenotypes, including ataxia (17). The fact that hypoxia is also able to reverse the neurological phenotype is notable as most patients with FA will already have manifested some form of gait or coordination disturbance before diagnosis is established (39). While it is possible that hypoxia is acting on the neuropathology indirectly, by activating repair pathways that restore damaged tissue, our prior in vitro and cellular work suggest that hypoxia is likely having a more direct effect in preserving and restoring Fe–S clusters, the root biochemical defect in the absence of FXN. These findings hint that the main driver of neurodegeneration in the shFxn animals is the depletion of Fe–S dependent proteins or pathways. The fact that hypoxia reverses neurological disease as rapidly as gene replacement (i.e. cessation of doxycycline) helps to support the notion that hypoxia is acting very proximally in alleviating the root biochemical defect to restore neuronal health. Moreover, these kinetics suggest that hypoxia is not reversing frank neuronal loss, but rather, improving severe neurological dysfunction. It is striking that the cardiomyopathy of FA is not amenable to either chronic 11% O2 or any of the hypoxia-inspired regimens that improve the neuropathology. One reason for this Human Molecular Genetics, 2023, Vol. 32, No. 16 \| 2607 tissue divergence might be that it is easier to achieve tissue hypoxia given the vascular anatomy of the central and peripheral nervous systems than it is in the myocardium, which has no equivalent of a ‘blood brain barrier’ (40). Evaluating this hypoth- esis would require measuring the tissue oxygen tension of the myocardium in our hypoxia regimen. A second reason might that in the brain, hypoxia acts to modulate downstream effectors of frataxin loss or neuronal injury, and that these pathways are not active in the heart. While our findings as to the effects of hypoxia are encouraging, it is important to bear in mind the limitations of the current study. First, the shFxn mouse models an acute and near-total depletion of FXN in a mature animal. Thus, the shFxn mouse does not fully capture any developmental components of FA. However, it is promising that even in the face of the extreme neurolog- ical deficits observed in the shFxn mouse; hypoxia is effective at preventing and reversing ataxia. Second, ataxia was assayed only by rotarod in our newly reported regimens (Regimens 1–7). While the rotarod test tends to be a stringent and comprehensive assessment of neuromuscular function, this test can read out other deficits, including motor deficiencies. Indeed, recent work has demonstrated that the shFxn mouse model displays reduced lean muscle mass (41). As truly safe, practical, and effective regimens are identified it will be worthwhile performing broader neurological testing to establish whether these regimens also improve other neurological features of disease. Considering the urgent clinical need for biomarkers that can quantify FA disease progression and therapeutic response, our data help build upon recent findings pointing to GFD15 as a promising biomarker for FA cardiac pathology. Both in our hands and others, Gdf15 levels rise concurrently with the onset and severity of cardiac disease in murine FA models (20). GDF15 is a cytokine that is transcriptionally activated by the ISR, which is frequently activated in the heart in mitochondrial diseases (42– 45). Moreover, the ISR has been shown to be robustly activated following FXN depletion both in cells and various mouse models (16,19,46). It is quite plausible that the ISR induction observed in these models is driven by the loss of heme synthesis, as this path- way both relies heavily on Fe–S dependent proteins and can serve as a direct ISR trigger, making GDF15 an attractive mechanistic disease biomarker. Collectively, the findings from our original report (16) on the utility of chronic, continuous 11% O2 (Regimen 0), in combination with the lessons gleaned from the current study (Regimens 1–7), lay an important groundwork in eventually translating hypoxia regimens into the clinic for FA patients. While chronic, continuous hypoxia remains a powerful intervention for FA in this pre-clinical model, acting to both prevent and restore ataxia, intermittent hypoxia proved to be particularly harmful to lifespan and car- diomyopathy. The complex interplay between hypoxia, its physio- logical response, and the unique deficits that drive tissue-specific FA pathogenesis is complex and nuanced. Based on these findings we emphasize the need for additional pre-clinical evaluation of hypoxia and hypoxia-inspired regimens before proceeding to test- ing in FA patients. Such studies should be worthwhile, especially as we find that initiating hypoxia at advanced timepoints can pre- vent and even reverse the neurological defects we have assayed. Fortunately, mouse models of FA and other mitochondrial dis- eases are available and can be used for continued evaluation of candidate hypoxia regimens that are safe, practical, and effective in the pre-clinical arena before progressing into clinical trials in patients. 2608 \| Human Molecular Genetics, 2023, Vol. 32, No. 16 Materials and Methods Mice shFxn mice were generously provided by the Geshwind Laboratory at the University of California, Los Angeles. Hamp−/− mice, origi- nally generated by Lesbordes-Brion et al. (26) and backcrossed on a C57BL/6 background (47) were kindly provided by Dr Tomas Ganz (University of California, LA). Pups were weaned and genotyped at ∼25 days after birth. Mouse genotypes from tail biopsies were determined using real time PCR with specific probes designed for each gene (Transnetyx, Cordova, TN). All cages were provided with food and water ad-libitum. Food and water were monitored daily and replenished as needed, and cages were changed weekly. A standard light–dark cycle of ∼12 h light exposure was used. Two to five animals were housed per cage. Body weights were recorded regularly, and mice were humanely euthanized when they had lost 20% of peak body weight, in accordance with the American Veterinary Medical Association guidelines. For all experiments, animals were randomized on a 1:1 basis, balanced by age and sex. All animal studies were approved by the Subcommittee on Research Animal Care and the Institutional Animal Care and Use Committee of Massachusetts General Hospital. Doxycycline knockdown in shFxn mice The average age of the animals at the start of experiments was 2–3 months. Doxycycline treatment followed the established opti- mal dosing protocol (16); 2 mg/ml Doxycycline (Sigma) was added to the drinking water of all animals which was changed weekly. In addition, animals were injected intraperitoneally with doxycy- cline twice a week, starting with 5 mg/kg body weight for 10 weeks followed by 10 mg Dox/kg body at later timepoints. Hypoxia treatments of shFxn mice Wild type and shFxn mice were exposed to chronic hypoxia (11% O2), normoxia (21% O2), intermittent hypoxia (16 h- 11% O2, 8 h- 21% O2), and mild chronic hypoxia (17% O2) at ambient sea-level pressure, using an OxyCycler A84XOV Multi-Chamber Dynamic Oxygen Controller (BioSpherix Ltd, Parish, NY). The CO2 concen- tration in each chamber as well as the temperature and the humidity were monitored continuously. Temperature and humid- ity were maintained at 23–25◦C and 30–70%, respectively. Mice were exposed to gas treatment continuously for 24 h per day, 7 days a week. The chambers were brief ly opened three times a week to weigh the mice, evaluate their neurological status, change the cages, doxycycline injections, and add water and food. Anemia To induce anemia, mice were placed on a low iron diet (Envigo). Concurrently, 150–200 μl of blood were collected every other day until hemoglobin concentration reached the desired values. Within 3 weeks, hemoglobin concentrations were approximately ≤ 5 g/dl and remained stable while the mice remained on a low iron diet for the remainder of the anemia trial. PT2399 administration PT2399 (MedChemExpress) was prepared with 10% ethanol, 30% PEG400, 0.3% methyl cellulose and 0.3% Tween 80. Mice were treated with 100 mg/kg PT2399 or vehicle twice daily by oral gavage. Hemoglobin and hematocrit measurements Fifty microliters of blood were collected by tail snip into a heparinized capillary. Hemoglobin concentration and hematocrit were measured using a blood gas analyzer (ABL800 FLEX, Radiometer, Copenhagen, Denmark). Accelerating rotarod measurements A rotarod machine (Ugo Basile) was used to measure the ability of mice to stay on an accelerating, rotating rod. Mice were accli- mated to the experimental room for at least 30 mins before the start of the measurements. Rotarod parameters were as follows: acceleration of 5 rpm/m and a maximum speed of 40 rpm. On each measurement day, three trials were performed, with individual trials at least 10 m apart to allow mice to recuperate. The median time on rotarod is reported. If mice used their body to grasp the rod (rather than walking on it) for more than 10 s, this time was recorded as time of fall. GDF15 qPCR Animals were sacrificed by CO2 asphyxiation followed by cervical dislocation. Individual hearts were immediately harvested and snap frozen in liquid N2. The tissue was then disrupted with two 5 mm stainless steel beads (Qiagen) using a Qiagen TissueLyser for 2 mins at 25 Hz. RNA was extracted with the RNeasy Tissue Mini Kit (Qiagen) before murine leukemia virus reverse transcription using random primers (Promega). qPCR was performed using the TaqMan technology (Life Technologies), using the following probe Mm00442228_m1 (Gdf15). All data were normalized to Tbp (Mm01277042_m1). Polyacrylamide gel electrophoresis and protein immunoblotting Animals were sacrificed by CO2 asphyxiation followed by cervical dislocation. Tissues were immediately harvested and snap frozen in liquid N2. Whole heart tissue was immersed in approx. 400 μL ice-cold RIPA buffer supplemented with (Thermo Fisher Halt Protease/Phosphatase Inhibitor Cocktail Scientific). The tissue was then lysed with two 5 mm stainless steel beads using a QIAGEN TissueLyser for 2 mins at 25 Hz. The resulting homogenate was centrifuged for 10 mins at maximum speed at 4◦C, and the supernatant was centrifuged a second time to remove residual insoluble material. Protein content of the resulting clarified lysate was determined using the Pierce 660 nm assay (Thermo Fisher Scientific). Appropriate volumes of lysate were boiled for 5 mins in the presence of SDS sample buffer. Electrophoresis was carried out on Novex Tris-Glycine 4–20% gels (Life Technologies) before transfer on a nitrocellulose membrane, 0.45 μm (BioRad). Membranes were blocked for 30 mins with SEA BLOCK Blocking Buffer (Thermo Fisher Scientific) at RT. Membranes were then incubated with primary antibody, diluted in 3%BSA, for 1 h at RT or overnight at 4◦C. Membranes were then washed at RT three times in TBST for 5 mins. The membrane was incubated with goat α-rabbit or α-mouse conjugated to IRDye800 or to IRDye680 (LI-COR Biosciences), diluted in 5% milk, for 1 h at RT. Membranes were washed three times in TBST for 5 mins and were scanned for infrared signal using the Odyssey Imaging System (LI-COR Biosciences). Band intensities were analyzed with Image Studio Lite (LI-COR Biosciences). Echocardiography analysis Cardiac function was evaluated by transthoracic echocardiog- raphy (48). Mice were anesthetized with 3% isof lurane, which was reduced to 1.5% isof lurane during echocardiography. Images were collected using a 14.0-MHz linear probe (Vivid 7; GE Medical System, Milwaukee, WI). Body temperature was maintained at 37◦C during echocardiography. M-mode images were obtained from a parasternal short axis view at the midventricular level with a clear view of the papillary muscle. Left ventricular internal diameters at end-diastole and end-systole were measured (16). LV fractional shortening was calculated on an EchoPAC workstation (GE Healthcare, Wauwatosa, WI). Brain tissue PO2 measurement Mice were anesthetized with isof lurane (induction at 2–4%, main- tenance at 1–1.5%), intubated, and mechanically ventilated with a tidal volume of 8 ml/kg, a respiratory rate of 110 breaths per minute and an inspired fraction of oxygen (FiO2) of 21%. Mice were placed in a prone position and the head was stabilized using a stereotaxic frame (ASI Instruments, MI). After incision and dissection of the skin, an opening in the skull was performed using a micro-drill (MD-1200, Braintree Scientific, MA) and a PO2 probe was inserted at the desired location. The coordinates were ML = −1.25 mm, AP = −6.00 mm, and DV = −3.90 mm from the bregma. Optical PO2 probe (OxyLab, Oxford Optronix, Abingdon, UK) was employed to detect the brain tissue PO2. During the brain PO2 measurement the depth of anesthesia was reduced by lowering the Isof lurane concentration to 0.5–1% to minimize the impact of anesthesia on the brain PO2. Quantification and statistical analysis Data are reported as mean ± SD. Analyses were performed using GraphPad Prism 8.0.1 software. Two-way ANOVA with Bonferroni’s correction was used for multiple comparisons. A log-rank (Man- tel–Cox) test was utilized to compare survival rates. P-value < 0.05 was considered to indicate statistical significance. Supplementary Material Supplementary Material is available at HMG online. Acknowledgements Authors thank Drs. Daniel H. Geschwind (UCLA) and Tomas Ganz (UCLA) for generously providing the shFxn and hepcidin knockout mice, respectively. Authors thank O. Goldberger, S.E. Calvo and M. for their assistance, and all members of the Mootha laboratory for fruitful discussions and feedback. Conf lict of Interest statement. V.K.M. is on the scientific advisory board of and receives equity from 5AM Ventures. V.K.M. is an inventor on patent applications filed by Massachusetts General Hospital on the therapeutic uses of hypoxia for mitochondrial diseases. Funding This work was supported in part by funds from the Friedreich’s Ataxia Research Alliance (FARA), Marriott Family Foundation, and the National Institutes of Health (R01NS124679) to V.K.M. V.K.M is an Investigator of the Howard Hughes Medical Institute. Data Availability Human Molecular Genetics, 2023, Vol. 32, No. 16 \| 2609 References 1. Keita, M., McIntyre, K., Rodden, L.N., Schadt, K. and Lynch, D.R. (2022) Friedreich ataxia: clinical features and new develop- ments. Neurodegener. Dis. Manag, 12, 267–283. 2. Koeppen, A.H. (2011) Friedreich’s ataxia: pathology, pathogene- sis, and molecular genetics. J. Neurol. Sci., 303, 1–12. 3. Pandolfo, M. (2012) Friedreich ataxia. Handb. Clin. Neurol., 103, 275–294. 4. Harding, A.E. (1981) Friedreich’s ataxia: a clinical and genetic study of 90 families with an analysis of early diagnostic crite- ria and intrafamilial clustering of clinical features. Brain, 104, 589–620. 5. Tsou, A.Y., Paulsen, E.K., Lagedrost, S.J., Perlman, S.L., Mathews, K.D., Wilmot, G.R., Ravina, B., Koeppen, A.H. and Lynch, D.R. (2011) Mortality in Friedreich ataxia. J. Neurol. Sci., 307, 46–49. 6. Campuzano, V., Montermini, L., Molto, M.D., Pianese, L., Cossee, M., Cavalcanti, F., Monros, E., Rodius, F., Duclos, F., Monticelli, A. et al. (1979) (1996) Friedreich’s ataxia: autosomal recessive disease caused by an intronic GAA triplet repeat expansion. Science, 271, 1423–1427. 7. Srour, B., Gervason, S., Monfort, B. and D’Autréaux, B. (2020) Mechanism of iron–Sulfur cluster assembly: in the intimacy of iron and Sulfur encounter. Inorganics (Basel), 8(10), 55. 8. Maio, N., Jain, A. and Rouault, T.A. (2020) Mammalian iron-sulfur cluster biogenesis: recent insights into the roles of frataxin, acyl carrier protein and ATPase-mediated transfer to recipient proteins. Curr. Opin. Chem. Biol., 55, 34–44. 9. Andreini, C., Banci, L. and Rosato, A. (2016) Exploiting bacterial operons to illuminate human iron-Sulfur proteins. J. Proteome Res., 15, 1308–1322. 10. Lill, R. and Freibert, S.A. (2020) Mechanisms of mitochondrial iron-sulfur protein biogenesis. Annu. Rev. Biochem., 89, 471–499. 11. Lynch, D.R., Farmer, J., Hauser, L., Blair, I.A., Wang, Q.Q., Mesaros, C., Snyder, N., Boesch, S., Chin, M., Delatycki, M.B. et al. (2019) Safety, pharmacodynamics, and potential benefit of omavelox- olone in Friedreich ataxia. Ann. Clin. Transl. Neurol., 6, 15–26. 12. Lynch, D.R., Chin, M.P., Delatycki, M.B., Subramony, S.H., Corti, M., Hoyle, J.C., Boesch, S., Nachbauer, W., Mariotti, C., Mathews, K.D. et al. (2021) Safety and efficacy of Omaveloxolone in Friedreich ataxia (MOXIe study). Ann. Neurol., 89, 212–225. 13. Zhang, S., Napierala, M. and Napierala, J.S. (2019) Therapeu- tic prospects for Friedreich’s ataxia. Trends Pharmacol. Sci., 40, 229–233. 14. Huichalaf, C., Perfitt, T.L., Kuperman, A., Gooch, R., Kovi, R.C., Brenneman, K.A., Chen, X., Hirenallur-Shanthappa, D., Ma, T., Assaf, B.T. et al. (2022) In vivo overexpression of frataxin causes toxicity mediated by iron-sulfur cluster deficiency. Mol. Ther. Methods. Clin. Dev., 24, 367–378. 15. Strawser, C., Schadt, K., Hauser, L., McCormick, A., Wells, M., Larkindale, J., Lin, H. and Lynch, D.R. (2017) Pharmacological therapeutics in Friedreich ataxia: the present state. Expert. Rev. Neurother., 17, 895–907. 16. Ast, T., Meisel, J.D., Patra, S., Wang, H., Grange, R.M.H., Kim, S.H., Calvo, S.E., Orefice, L.L., Nagashima, F., Ichinose, F. et al. (2019) Hypoxia rescues Frataxin loss by restoring iron sulfur cluster biogenesis. Cell, 177, 1507, e16–1521. 17. Chandran, V., Gao, K., Swarup, V., Versano, R., Dong, H., Jordan, M.C. and Geschwind, D.H. (2017) Inducible and reversible phe- notypes in a novel mouse model of Friedreich’s ataxia. elife, 6, e30054. The data generated for this paper will be available from the corresponding author on reasonable request. 18. Ferrari, M., Jain, I.H., Goldberger, O., Rezoagli, E., Thoonen, R., Cheng, K.H., Sosnovik, D.E., Scherrer-Crosbie, M., Mootha, V.K. 2610 \| Human Molecular Genetics, 2023, Vol. 32, No. 16 and Zapol, W.M. (2017) Hypoxia treatment reverses neurodegen- erative disease in a mouse model of Leigh syndrome. Proc. Natl. Acad. Sci. U. S. A., 114, E4241–E4250. 19. Vasquez-Trincado, C., Patel, M., Sivaramakrishnan, A., Bekeova, C., Anderson-Pullinger, L., Wang, N., Tang, H.Y. and Seifert, E.L. (2021) Adaptation of the heart to Frataxin depletion: evidence that integrated stress response can predominate over mTORC1 activation. Hum. Mol. Genet., ddab216. 20. Belbellaa, B., Reutenauer, L., Monassier, L. and Puccio, H. (2019) Correction of half the cardiomyocytes fully rescue Friedreich ataxia mitochondrial cardiomyopathy through cell- autonomous mechanisms. Hum. Mol. Genet., 28, 1274–1285. 21. Chen, W., Hill, H., Christie, A., Kim, M.S., Holloman, E., Pavia- Jimenez, A., Homayoun, F., Ma, Y., Patel, N., Yell, P. et al. (2016) Targeting renal cell carcinoma with a HIF-2 antagonist. Nature, 539, 112–117. 22. Cho, H., Du, X., Rizzi, J.P., Liberzon, E., Chakraborty, A.A., Gao, W., Carvo, I., Signoretti, S., Bruick, R.K., Josey, J.A. et al. (2016) On- target efficacy of a HIF-2alpha antagonist in preclinical kidney cancer models. Nature, 539, 107–111. Jain, I.H., Zazzeron, L., Goldberger, O., Marutani, E., Wojtkiewicz, G.R., Ast, T., Wang, H., Schleifer, G., Stepanova, A., Brepoels, K. et al. (2019) Leigh syndrome mouse model can be rescued by interventions that normalize brain Hyperoxia, but not HIF activation. Cell Metab., 30, 824, e3–832. 23. 24. Llorens, J.V., Soriano, S., Calap-Quintana, P., Gonzalez-Cabo, P. and Moltó, M.D. (2019) The role of iron in Friedreich’s ataxia: insights from studies in human tissues and cellular and animal models. Front. Neurosci., 13, 75. 25. Martelli, A., Schmucker, S., Reutenauer, L., Mathieu, J.R.R., Peyssonnaux, C., Karim, Z., Puy, H., Galy, B., Hentze, M.W. and Puccio, H. (2015) Iron regulatory protein 1 sustains mitochon- drial iron loading and function in frataxin deficiency. Cell Metab., 21, 311–323. 26. Lesbordes-Brion, J.C., Viatte, L., Bennoun, M., Lou, D.Q., Ramey, G., Houbron, C., Hamard, G., Kahn, A. and Vaulont, S. (2006) Targeted disruption of the hepcidin 1 gene results in severe hemochromatosis. Blood, 108, 1402–1405. 27. Reichert, C.O., da Cunha, J., Levy, D., Maselli, L.M.F., Bydlowski, S.P. and Spada, C. (2017) Hepcidin: homeostasis and diseases related to iron metabolism. Acta Haematol., 137, 220–236. 28. Hintze, K.J. and McClung, J.P. (2011) Hepcidin: a critical regu- lator of iron metabolism during hypoxia. Adv. Hematol., 2011, 510304. 29. Nicolas, G., Chauvet, C., Viatte, L., Danan, J.L., Bigard, X., Devaux, I., Beaumont, C., Kahn, A. and Vaulont, S. (2002) The gene encod- ing the iron regulatory peptide hepcidin is regulated by anemia, hypoxia, and inflammation. J. Clin. Invest., 110, 1037–1044. 30. Bulteau, A.L., Dancis, A., Gareil, M., Montagne, J.J., Camadro, J.M. and Lesuisse, E. (2007) Oxidative stress and protease dysfunction in the yeast model of Friedreich ataxia. Free Radic. Biol. Med., 42, 1561–1570. 31. Snoek, I.S. and Steensma, H.Y. (2006) Why does Kluyveromyces lactis not grow under anaerobic conditions? Comparison of essential anaerobic genes of Saccharomyces cerevisiae with the Kluyveromyces lactis genome. FEMS Yeast Res., 6, 393–403. 32. Zhang, Y., Lyver, E.R., Knight, S.A., Lesuisse, E. and Dancis, A. (2005) Frataxin and mitochondrial carrier proteins, Mrs3p and Mrs4p, cooperate in providing iron for heme synthesis. J. Biol. Chem., 280, 19794–19807. 33. Fil, D., Chacko, B.K., Conley, R., Ouyang, X., Zhang, J., Darley- Usmar, V.M., Zuberi, A.R., Lutz, C.M., Napierala, M. and Napierala, J.S. (2020) Mitochondrial damage and senescence phenotype of cells derived from a novel frataxin G127V point mutation mouse model of Friedreich’s ataxia. Dis. Model. Mech., 13(7), dmm045229. 34. Fernandez, S., Wofford, J.D., Shepherd, R.E., Vali, S.W., Dancis, A. and Lindahl, P.A. (2022) Yeast cells depleted of the frataxin homolog Yfh1 redistribute cellular iron: studies using Moss- bauer spectroscopy and mathematical modeling. J. Biol. Chem., 298, 101921. 35. Das, S.K., Zhabyeyev, P., Basu, R., Patel, V.B., Dyck, J.R.B., Kassiri, Z. and Oudit, G.Y. (2018) Advanced iron-overload cardiomyopathy in a genetic murine model is rescued by resveratrol therapy. Biosci. Rep., 38(1), BSR20171302. 36. Berdoukas, V., Coates, T.D. and Cabantchik, Z.I. (2015) Iron and oxidative stress in cardiomyopathy in thalassemia. Free Radic. Biol. Med., 88, 3–9. 37. Turchi, R., Faraonio, R., Lettieri-Barbato, D. and Aquilano, K. (2020) An overview of the ferroptosis hallmarks in Friedreich’s ataxia. An overview of the ferroptosis hallmarks in Friedreich’s ataxia. Biomol. Ther., 10, 1489. 38. Grazia Cotticelli, M., Xia, S., Lin, D., Lee, T., Terrab, L., Wipf, P., Huryn, D.M. and Wilson, R.B. (2019) Ferroptosis as a novel therapeutic target for Friedreich’s ataxia. J. Pharmacol. Exp. Ther., 369, 47–54. Indelicato, E., Nachbauer, W., Eigentler, A., Amprosi, M., Mat- teucci Gothe, R., Giunti, P., Mariotti, C., Arpa, J., Durr, A., Klop- stock, T. et al. (2020) Onset features and time to diagnosis in Friedreich’s ataxia. Orphanet J. Rare Dis., 15, 198. 39. 40. Keeley, T.P. and Mann, G.E. (2019) Defining physiological nor- moxia for improved translation of cell physiology to animal models and humans. Physiol. Rev., 99, 161–234. 41. Vásquez-Trincado, C., Dunn, J., Han, J.I., Hymms, B., Tamaroff, J., Patel, M., Nguyen, S., Dedio, A., Wade, K., Enigwe, C. et al. (2022) Frataxin deficiency lowers lean mass and triggers the integrated stress response in skeletal muscle. JCI Insight, 7(9), e155201. 42. Emmerson, P.J., Duffin, K.L., Chintharlapalli, S. and Wu, X. (2018) GDF15 and growth control. Front. Physiol., 9, 1712. 43. Bao, X.R., Ong, S.E., Goldberger, O., Peng, J., Sharma, R., Thompson, D.A., Vafai, S.B., Cox, A.G., Marutani, E., Ichinose, F. et al. (2016) Mitochondrial dysfunction remodels one-carbon metabolism in human cells. elife, 5, e10575. 44. Kuhl, I., Miranda, M., Atanassov, I., Kuznetsova, I., Hinze, Y., Mourier, A., Filipovska, A. and Larsson, N.G. (2017) Transcrip- tomic and proteomic landscape of mitochondrial dysfunc- tion reveals secondary coenzyme Q deficiency in mammals. elife, 6. 45. Sharma, R., Reinstadler, B., Engelstad, K., Skinner, O.S., Stack- owitz, E., Haller, R.G., Clish, C.B., Pierce, K., Walker, M.A., Fryer, R. et al. (2021) Circulating markers of NADH-reductive stress cor- relate with mitochondrial disease severity. J. Clin. Invest., 131(2), e136055. 46. Huang, M.L., Sivagurunathan, S., Ting, S., Jansson, P.J., Austin, C.J., Kelly, M., Semsarian, C., Zhang, D. and Richardson, D.R. (2013) Molecular and functional alterations in a mouse cardiac model of Friedreich ataxia: activation of the integrated stress response, eIF2alpha phosphorylation, and the induction of downstream targets. Am. J. Pathol., 183, 745–757. 47. Ramos, E., Ruchala, P., Goodnough, J.B., Kautz, L., Preza, G.C., Nemeth, E. and Ganz, T. (2012) Minihepcidins prevent iron over- load in a hepcidin-deficient mouse model of severe hemochro- matosis. Blood, 120, 3829–3836. Irie, T., Sips, P.Y., Kai, S., Kida, K., Ikeda, K., Hirai, S., Moazzami, K., Jiramongkolchai, P., Bloch, D.B., Doulias, P.T. et al. (2015) S- Nitrosylation of calcium-handling proteins in cardiac adrenergic Signaling and hypertrophy. Circ. Res., 117, 793–803. 48.
10.1126_sciadv.adg1671	S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E D E V E LO P M E N TA L B I O LO G Y Atoh1 drives the heterogeneity of the pontine nuclei neurons and promotes their differentiation Sih-Rong Wu1,2, Jessica C. Butts2,3,4, Matthew S. Caudill1,2, Jean-Pierre Revelli2,3, Ryan S. Dhindsa2,3, Mark A. Durham2,5,6, Huda Y. Zoghbi1,2,3,4,7* Pontine nuclei (PN) neurons mediate the communication between the cerebral cortex andthe cerebellum to refine skilled motor functions. Prior studies showed that PN neurons fall into two subtypes based on their an- atomic location and region-specific connectivity, but the extent of their heterogeneity and its molecular drivers remain unknown. Atoh1 encodes a transcription factor that is expressed in the PN precursors. We previously showed that partial loss of Atoh1 function in mice results in delayed PN development and impaired motor learn- ing. In this study, we performed single-cell RNA sequencing to elucidate the cell state–specific functions of Atoh1 during PN development and found that Atoh1 regulates cell cycle exit, differentiation, migration, and survival of PN neurons. Our data revealed six previously not known PN subtypes that are molecularly and spatially distinct. We found that the PN subtypes exhibit differential vulnerability to partial loss of Atoh1 function, providing in- sights into the prominence of PN phenotypes in patients with ATOH1 missense mutations. Copyright © 2023 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution License 4.0 (CC BY). INTRODUCTION Motor skills such as picking up a cup of coffee or hitting a speedy baseball with a bat require communication between the cerebral cortex and the cerebellum (1). These connections are mediated through the pontine nuclei (PN) that are located in the ventral pons and composed of mostly glutamatergic neurons (2–7). Given their pivotal role in motor functions, several efforts have been made to understand how the PN develop in mammals. PN neurons originate from a group of proliferating neuroepithelial cells residing in the rhombic lip (RL) located in the developing hindbrain. Specifically, Wnt1- and Atoh1-expressing cells within the caudal RL (cRL) give rise to glutamatergic PN neurons (8– 10). PN neurons are born during mouse embryonic day 12.5 (E12.5) to E18.5 and migrate tangentially along the anterior extra- mural stream (AES) until they reach the ventral pons (11, 12). A previous study of PN in rabbit and cat has suggested that PN can be categorized into subpopulations on the basis of their ana- tomical location and connectivity (13). Anatomically, the PN are divided into the basal pontine nucleus (BPN) and the reticuloteg- mental nucleus (RtTg) at the anterior-ventral and posterior-dorsal part of the PN, respectively. Subpopulations of PN neurons have been proposed on the basis of their positions along the rostro- caudal axis, which inherit the expression pattern of Hox2-Hox5 from their progenitors at the cRL (14). Moreover, tracing experi- ments demonstrated that the corticopontine connectivity was estab- lished in a partially region-specific manner (15, 16), suggesting that PN neurons are functionally diverse on the basis of their regional cortical inputs. However, the extent of PN heterogeneity remains 1Department of Neuroscience, Baylor College of Medicine, Houston, TX, USA. 2Jan and Dan Duncan Neurological Research Institute at Texas Children’s Hospital, Houston, TX, USA. 3Department of Molecular and Human Genetics, Baylor College of Medicine, Houston, TX, USA. 4Howard Hughes Medical Institute, Baylor College of Medicine, Houston, TX, USA. 5Program in Developmental Biology, Baylor College of Medicine, Houston, TX, USA. 6Medical Student Scientist Training Program, Baylor College of Medicine, Houston, TX, USA. 7Department of Pediatrics, Baylor College of Medicine, Houston, TX, USA. Corresponding author. Email: [email protected] unclear, and the molecular determinants of PN heterogeneity are not known. Another outstanding question is how a pool of seemingly ho- mogenous Atoh1 + progenitors give rise to diverse progenies in the PN. Atoh1, or Atonal homolog 1, encodes a basic helix-loop- helix transcription factor ATOH1 that is required for the develop- ment of a variety of neurons in the hindbrain (9) and the dorsal spinal cord (17), hair cells in the inner ear (18), Merkel cells in the skin (19), and secretory cells in the gut (20). In the hindbrain, Atoh1-lineage neurons contribute to many key components of the proprioceptive pathway, including the PN (9, 21). A recent report implicated a homozygous missense variant in ATOH1 in two human patients with global developmental delay, motor function deficits, pontocerebellar hypoplasia, and hearing loss (22). It is thus of great interest and clinical relevance to understand how Atoh1 shapes proper PN development. Loss of both copies of Atoh1 in mice results in complete absence of the PN, whereas loss of one copy of Atoh1 in mice has no observ- able change in the PN (23), making it difficult to study the function of Atoh1 during PN development using either the Atoh1 knockout or the heterozygous mouse model. We previously found that substi- tuting the serine at position 193 with an alanine (S193A) resulted in an Atoh1 hypomorphic allele (24). Mice carrying Atoh1S193A over an Atoh1 null allele (Atoh1S193A/−) showed delayed development in the PN neurons at postnatal day 0 (P0) and impaired motor learning as adults. Given the heterogeneity of the PN neurons, it is unclear whether this hypomorphic mutation affects the development of all PN subtypes equally. In other contexts, we learn that different cell types have differential vulnerability to perturbation in genes im- plicated in neurodevelopmental disorders (25, 26). For example, despite being expressed in both neurogenic niches, increased level of the transcription factor Foxg1 compromises the neurogenesis of excitatory neurons but not inhibitory neurons (26). We therefore hypothesized that PN subtypes are molecularly distinct and have differential vulnerability to partial loss of function of Atoh1. In this study, we profiled the single-cell transcriptomes of the PN neurons and characterized the molecular and cellular phenotypes of Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 1 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E the PN in Atoh1S193A/− mice at the single-cell resolution. We found cell state–specific roles of Atoh1 during PN development including regulating cell cycle exit, differentiation, migration, survival, and cellular heterogeneity of the PN neurons. In addition, our single- cell RNA sequencing (scRNA-seq) data revealed six unreported subtypes of the PN neurons at P5 and identified subtype-specific markers. We found that the PN subtypes have differential vulnera- bilities to partial loss of function of Atoh1. Our study provides com- prehensive evidence of how a transcription factor plays multiple roles during neuronal development and demonstrates that neuronal subtypes could have differential sensitivity to perturbation of a gene that is expressed in the precursors of all subtypes. RESULTS Phospho-mutation of Atoh1 at serine-193 leads to PN hypoplasia in mice To test whether partial loss of function of Atoh1 results in PN ab- normalities postnatally, we compared the morphology of the PN in mice with hypomorphic mutation to those in control mice at P0 and P21. We crossed Atoh1S193A/+ mice to Atoh1lacZ/+, whereby the lacZ allele replaced the coding region of Atoh1, creating a null allele. Fol- lowing the tangential migration along the AES (Fig. 1A), most of the PN neurons finished migrating at P0 in control mice (Fig. 1B, left, arrow), with little lacZ signal in the AES (arrowhead). We observed a decreased intensity of the lacZ staining in the PN in Atoh1S193A/lacZ mice (Fig. 1B, right, arrow), indicating potential neuronal loss at P0 in Atoh1S193A/lacZ mice. In addition, consistent with our previous study (24), we found that there were more lacZ+ neurons retained at the AES in Atoh1S193A/lacZ mice (Fig. 1B, right, arrowhead), sug- gesting a developmental delay or a migration deficit in some PN neurons upon partial loss of Atoh1 function. To test whether those neurons retained at the AES ever reach their destination, we examined the PN morphology at a juvenile age, P21. Given that Atoh1 is turned off postnatally in the PN, we permanently labeled the Atoh1-lineage neurons using a Cre-depen- dent lacZ reporter (27) in combination with an Atoh1Cre/+ knock-in mouse in which one copy of Atoh1 is functionally a null allele because the Cre replaced the coding region of Atoh1 (28). This allows us to visualize the PN postnatally (Fig. 1C). We found that the size of the PN was reduced in Atoh1Cre/S193A; Rosalsl-lacZ/+ animals compared to Atoh1Cre/+; Rosalsl-lacZ/+ animals (Fig. 1D). We further confirmed this phenotype quantitatively by performing immunofluorescence staining with osteopontin antibody, a pan-PN neuronal marker (fig. S1). These data suggested that partial loss of function of Atoh1 not only leads to delayed development of PN but also leads to PN hypoplasia in mice reminiscent of the malforma- tion of the PN in patients with ATOH1 missense variant (22). To- gether, these data reinforced the importance of Atoh1 in PN development and demonstrated that Atoh1 hypomorphic allele could provide a useful mouse model to dissect the functions of Atoh1 during PN development. fluorescent Progression of the PN development is impaired in Atoh1S193A/− animals To investigate the mechanisms by which Atoh1 regulates normal PN development, we performed scRNA-seq in developing hindbrains from control mice and mice with Atoh1 hypomorphic reporter mutation. We used a Cre-dependent (Rosalsl-TdTomato) to permanently label Atoh1-lineage neurons with Atoh1Cre/+ mice (Fig. 2, A and B). During development, PN progenitors and migrating PN neurons are spatially dispersed between the cRL to ventral pons. Therefore, to ensure that we capture all PN progenitors and migrating neurons, we collected whole hindbrains from Atoh1Cre/+; Rosalsl-tdTomato/+ (hereafter re- ferred to as control) and Atoh1Cre/S193A; Rosalsl-tdTomato/+ (hereafter referred to as Atoh1S193A/−) embryos. After single-cell dissociation, Atoh1-lineage (TdTom+) neurons were isolated by fluorescence- activated cell sorting (FACS), followed by scRNA-seq with 10X Genomics Chromium platform (Fig. 2C). We collected samples at two time points, E14.5 and E18.5, to examine the phenotypes of Atoh1S193A/− mice in the middle and at the end of neurogenesis of the PN neurons, respectively. We first focused on the E14.5 time point to explore the role of Atoh1 during PN development. After filtering out low-quality cells (see Materials and Methods), a total of 22,191 cells were retained from control hindbrains, and 22,206 cells were retained from Atoh1S193A/− hindbrains (n = 3 animals per genotype). Among the cells in the hindbrain, we identified the PN cells based on the clusters that expressed the established markers of the AES and PN (fig. S2, A to D). This resulted in 2451 control PN cells and 3124 Atoh1S193A/− PN cells for downstream analyses. After unbiased clus- tering of the PN cells, we annotated each cluster by the expression of the known markers (Fig. 2D, fig. S2E, and data S1). Cells at different Fig. 1. Phospho-mutation of Atoh1 at serine-193 leads to PN hypoplasia in mice. (A) Schematic representation of the PN development in mice. PN neurons migrate from cRL to ventral pons through AES during E12.5 to E18.5. Mb, midbrain; Cb, cerebellum. (B) Whole-mount X-galactosidase (X-gal) staining on mouse hindbrain at P0 (ventral view). The arrows and arrowheads denote the neurons at the PN and in the AES, respectively. Scale bars, 1 mm. (C) Strategy to label the Atoh1-lineage neurons. Atoh1Cre/+ knock-in mouse was crossed with mouse carrying either wild-type or hypomorphic Atoh1 (Atoh1S193A) and a Cre-dependent lacZ allele. (D) Whole-mount X-gal staining on mouse hindbrain at P21 (ventral view). The dashed lines outline the area of the PN. Scale bars, 1 mm. Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 2 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 2. Atoh1S193A/− animals exhibited an impaired progression of PN development. (A) Strategy to label the Atoh1-lineage neurons. Atoh1Cre/+ knock-in mouse was crossed with mouse carrying either wild-type or hypomorphic Atoh1 (Atoh1S193A) and a Cre-dependent TdTomato (TdTom) reporter. (B) An image of three-dimensional rendering of E14.5 mouse head using lightsheet microscopy (back view). The TdTom represents the Atoh1-lineage neurons in developing hindbrain. EGL, external granule layer; SC, spinal cord. (C) Workflow for scRNA-seq of Atoh1-lineage neurons from E14.5 and E18.5 mouse hindbrain. Hindbrains were collected from E14.5 and E18.5 control and Atoh1S193A/− embryos (n = 3 per genotype for each time point). After enrichment by sorting, single-cell transcription profiles were captured by 10X Genomics Chromium platform. (D) Expression levels of the selective markers visualized on uniform manifold approximation and projection (UMAP). Ccnd1, Atoh1, Nhlh1, and Mapt are known markers for progenitors, intermediate progenitors, migrating neurons, and differentiated neurons, respectively. (E) UMAP of the five cell states identified in developing PN at E14.5 by scRNA-seq. (F) Proportion of the cells in each cell state. The arrows indicate the direction of the change in Atoh1S193A/− animals. False discovery rate (FDR) < 0.01 and #FDR < 0.05. progenitors cell states during PN development were fully captured in our dataset, including proliferating progenitors (Ccnd1+), postmitotic (Atoh1 high), migrating neurons intermediate (Nhlh1+), and differentiated neurons (Mapt+) in both control and Atoh1S193A/− samples (Fig. 2E). The migrating neurons constituted the largest population and were divided into two clusters. We there- fore annotated the two migrating populations as migrating neurons- 1 and migrating neurons-2 based on their maturity. The migrating neurons-1 are neurons starting to differentiate and migrate away from the cRL (Atoh1 lowNhlh1high), whereas the migrating neurons-2 are neurons expressing not only high level of the marker for migration but also low level of the differentiated neuro- nal marker (Nhlh1highMapt low) (Fig. 2, D and E). Next, we calculated the percentage of the cells in each cell state in control and Atoh1S193A/− animals and performed differential abun- dance test using propeller with speckle R package (29). We found that the proportion of the progenitors increased markedly from 9.3 to 31% with a significant decrease in the percentage of the mi- grating neurons-2 from 29.3 to 19.7% in Atoh1S193A/− animals (Fig. 2F and fig. S3A). The shift in the proportion of the cells in Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 3 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E each state indicates that the progression of the PN development was impaired in Atoh1S193A/− mice. Trajectory analysis with Slingshot (30, 31) showed that in control embryos, more cells progressed further in pseudo-time than in Atoh1S193A/− embryos (fig. S3, B and C). Collectively, these data demonstrate that partial loss of Atoh1 function leads to accumulation of the PN progenitors at the expense of the differentiating PN neurons. Partial loss of function of Atoh1 leads to decreased cell cycle exit in PN progenitors and deficits in differentiation and migration We sought to investigate the mechanisms underlying the altered proportions of the cells in each state in Atoh1S193A/− animals. While we observed an increased proportion of the proliferating pro- genitors in Atoh1S193A/− mice, the proportion of its downstream cell state (i.e., the intermediate progenitors) was not altered (Fig. 2F). Instead, the fraction of the migrating neurons-2 was significantly decreased. Therefore, we proposed two mechanisms underlying the observed phenotypes: Robust function of Atoh1 is required to (i) drive cell cycle exit and (ii) promote differentiation and migra- tion (fig. S3D). To validate our scRNA-seq data, we performed im- munofluorescence staining and histological analyses on the embryonic tissues. First, we performed immunofluorescence stain- ing for MKI67, a marker for proliferating cells, on E14.5 hindbrains of control and Atoh1S193A/− animals. On the basis of the staining of ATOH1 and MKI67 at the cRL in control animals (fig. S4), the MKI67+ proliferating progenitors and ATOH1+ intermediate pro- genitors reside at the ventromedial and dorsolateral cRL, respective- ly (Fig. 3A). In control animals, there were only few TdTom+ cells at the cRL (Fig. 3B, asterisks), suggesting that the proliferating PN pro- genitors become postmitotic and leave the cRL upon the onset of Atoh1 expression. In contrast, there was an increased percentage of the MKI67+ cells that overlapped with TdTom at the cRL in Atoh1S193A/− mice (Fig. 3, B and C), indicating an accumulation of proliferating progenitors at the cRL. The data suggest that robust Atoh1 function is important for the cycling progenitors to become postmitotic. Next, to test whether the differentiating and migrating neurons are reduced in Atoh1S193A/− mice, we used the thymidine analog, 5- chloro-20-deoxyuridine (CldU) to label and quantify the number of PN neurons born within the 24-hour window between E13.5 and E14.5. We injected the pregnant dams with CldU at E13.5 and har- vested the brains from E14.5 embryos. It has been shown that it takes 1 to 2 days for the PN neurons to migrate from cRL to ventral pons (12). Consistent with the literature, we found that most of the CldU-labeled PN neurons (CldU+TdTom+) were located at the AES 24 hours after injection in control animals (Fig. 3, D and E). The number of the CldU+TdTom+ cells at the AES was significantly reduced in Atoh1S193A/− mice (Fig. 3F), sug- gesting that fewer neurons differentiated and migrated away from the cRL in Atoh1S193A/− animals. These data validate our scRNA- seq data in which we found that migrating population was decreased in Atoh1S193A/− mice. Together, our data demonstrate that Atoh1 governs multiple biological processes in addition to specifying PN neurons. Atoh1 is important for both cell cycle exit of the PN pro- genitors and the differentiation and migration of the intermediate progenitors. The cell state–specific dysregulated pathways in Atoh1S193A/− mice To understand how Atoh1 mediates different biological processes during PN development, we sought to identify the genes that were dysregulated in each cell state in Atoh1S193A/− mice compared to control mice. Using differential gene expression analysis, we identified 362 differentially expressed genes (DEGs) [log2 fold change (log2FC) > 0.25 and false discovery rate (FDR) < 0.05] (data S2). We found that intermediate progenitors had the highest number of DEGs, followed by migrating neurons-1 and progenitors (Fig. 4A). We verified that the different number of DEGs per cell state was not simply a result of differences in total cell number in each cell state (fig. S5A). The cell states with the greatest transcrip- tional dysregulation (i.e., progenitors, intermediate progenitors, and migrating neurons-1) also expressed high levels of Atoh1 (Figs. 4A and 2D), which led us to test whether these DEGs were directly regulated by Atoh1. We calculated the percentage of the DEGs that had ATOH1 binding peak(s) identified by ATOH1 chro- matin immunoprecipitation sequencing (32) and performed en- richment analysis for each cell state. We found a significant enrichment of DEGs with ATOH1-binding in progenitors (47%), intermediate progenitors (48%), and migrating neurons-1 (42%) (Fig. 4B), highlighting the direct impact of Atoh1 in these cell states. We found several genes that were direct targets of Atoh1 and have been reported to play a role in PN development (Fig. 4C, shaded box). For instance, Atoh1, which is known to reg- ulate itself (33), was down-regulated in the progenitors. Barhl1, im- portant for migration and survival of the PN neurons (34), was down-regulated in both progenitors and intermediate progenitors upon partial loss of Atoh1 function. Last, Nhlh1, essential for migra- tion of the PN neurons (35), was decreased in the intermediate pro- genitors and the migrating neurons. Moreover, we also identified other DEGs such as Pcp4 and Rab15, whose roles have not been characterized in PN development but have robust changes in ex- pression levels in multiple cell states (Fig. 4C). Pcp4 has been iden- tified as an Atoh1 target in cochlear hair cells (36). Rab15 was also reported as an Atoh1 target in different Atoh1-lineage cells includ- ing developing cerebellar granule neurons (32), Merkel cells (37), cochlear hair cells (36), and neurons in dorsal neural tube (38). Next, we focused on the three most affected cell states and per- formed gene ontology (GO) enrichment analysis to identify the pathways that were dysregulated in Atoh1S193A/− mice beyond the known Atoh1 targets (Fig. 4D and data S3). In line with the finding of increased proliferating cells in Atoh1S193A/− mice, cell proliferation and cell cycle regulation including Notch signaling were among the top enriched GO terms for the up-regulated genes in the progenitor state (Fig. 4D, left). Consistent with the tra- jectory analysis that showed an impaired differentiation process in Atoh1S193A/− mice, down-regulated genes were enriched for neuron differentiation and cytoskeleton organization ontologies across all three cell states (Fig. 4D). These data supported our earlier findings that Atoh1 plays roles in both cell cycle regulation, neuronal differ- entiation, and migration. We found that up-regulated genes in both intermediate progenitors and migrating neurons-1 were enriched for regulators of apoptosis (Fig. 4D, middle and right). For instance, the average level of Hrk, a member of proapoptotic Bcl-2 family, was significantly increased in both intermediate progenitors and mi- grating neurons-1 in Atoh1S193A/− mice (fig. S5B). In addition, other genes involved in programmed cell death such as Pak3 and Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 4 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 3. Partial loss of function of Atoh1 leads to decreased cell cycle exit in PN progenitors and deficits in differentiation and migration. (A) Schematic illustration of mouse coronal section at E14.5. The area of cRL was enlarged on the bottom with locations of three cell states being labeled. Me, medulla. (B) Immunofluorescence staining of MKI67 on E14.5 mouse brain. The cropped view of cRL in control (top) and Atoh1S193A/− (bottom) mice. Atoh1-lineage neurons were labeled with TdTom, and the nuclei were labeled with 40,6-diamidino-2-phenylindole (DAPI). The asterisk denotes the progenitors at cRL. Scale bars, 50 μm. (C) Quantification of the percentage of MKI67 fluorescence overlapping with TdTom in control and Atoh1S193A/− mice. The gray symbols represent the five regions of interest quantified from each animal (n = 3 per genotype). The average percentage for each animal was shown in colored shape. The crossbar denotes the mean per genotype. *Χ2(1) = 53.95, P = 2.06 × 10−13 by mixed-model analysis of variance (ANOVA). (D) Schematic illustration of coronal section at E14.5. The dashed gray box indicates the region of interest shown in (E). V, ventricle; PB, parabrachial nuclei; CN, cerebellar nuclei; VC, ventral cochlear nucleus. (E) Immunofluorescence staining of CldU on E14.5 mouse brain. The areas of AES for control (left) and Atoh1S193A/− (right) animals were shown. Atoh1-lineage neurons were labeled with TdTom, and the nuclei were labeled with DAPI. Scale bars, 50 μm. (F) The average number of the CldU+TdTom+ cells per region of interest (ROI) in control and Atoh1S193A/− mice. The gray symbols represent the 10 regions of interest quan- tified from each animal (n = 3 per genotype). The average number for each animal was shown in colored shape. The crossbar denotes the mean per genotype. Χ2(1) = 39.81, P = 2.80 × 10−10 by mixed-model ANOVA. Dap were also up-regulated in the intermediate progenitor state (fig. S5, C and D). These data indicate that Atoh1 hypomorphic muta- tion might lead to increased cell death. To test this hypothesis, we performed terminal deoxynucleotidyl transferase–mediated deoxy- uridine triphosphate nick end labeling (TUNEL) assay and found that the number of TUNEL-positive cells was significantly increased in Atoh1S193A/− mice at the lateral cRL (Fig. 4, E and F), where the intermediate progenitors (Atoh1high) are located (Fig. 3A and fig. S4). The increased cell apoptosis in Atoh1S193A/− animals may count for the reduced size of PN at older age and suggest that Atoh1 is important for the survival of the intermediate progenitors. In summary, these data highlight the important roles of Atoh1 in progenitors, intermediate progenitors, and migrating neurons by regulating cell cycle, differentiation, migration, and survival of the developing PN neurons. PN neurons are molecularly heterogeneous To determine whether the cellular identities of the differentiated PN neurons are altered in Atoh1S193A/− mice, we performed scRNA-seq in control and Atoh1S193A/− hindbrains at E18.5, when most of the PN neurons have been born and migrated to the ventral pons. We analyzed 1196 control and 1176 Atoh1S193A/− PN cells (n = 3 animals per genotype) (fig. S6, A to D). After unbiased clustering, we annotated the clusters based on the known marker genes (Fig. 5, A and B). As expected, most of the cells at E18.5 are differentiated neurons marked by Mapt expression (Fig. 5A). The differentiated neurons were classified into four subtypes (Fig. 5B), suggesting that they are molecularly heterogeneous. We characterized the marker genes for each subtype based on differential gene expression analysis and named the differentiated PN neuron subtypes as em- bryonic PN1(ePN1) to ePN4 (Fig. 5C, fig. S6E, and data S1). Similar to the E14.5 data, we found a slightly increased proportion of the progenitors in Atoh1S193A/− mice at E18.5 (Fig. 5D). In addition, the percentage of ePN1 cells was significantly reduced from 14.9 to 5.5% (Fig. 5D and fig. S6F), while there was no significant change in other ePN subtypes. These data indicate that ePN sub- types have differential vulnerability to the Atoh1 hypomorphic mutation. To test whether the differential vulnerability of the PN subtypes in Atoh1S193A/− embryos proceeds to the postnatal stage, we first de- termined whether the PN neurons maintained their molecular het- erogeneity at P5. We enriched the PN neurons by dissecting and pooling the PN from control mice at P5 (n = 11 to 13 per replicate) and performed scRNA-seq for TdTom+ cells. Six PN subtypes (PN1 to PN6) were uncovered using unbiased clustering (Fig. 6A), sug- gesting that PN neurons maintained molecularly distinct at P5. In Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 5 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 4. Partial loss of Atoh1 function leads to multiple dysregulated pathways. (A) The number of the up-regulated (left) and down-regulated (right) genes at each cell state in Atoh1S193A/− compared to control. The DEGs were filtered by log2FC > 0.25 and FDR < 0.05. (B) The enrichment of the DEGs with ATOH1-binding in each cell state. The percentage of the DEGs with ATOH1-binding peak was shown in number. The odds ratios were presented with 95% confidence interval performed by Fisher’s exact test. P value was adjusted by Bonferroni. (C) Volcano plot of the DEGs with ATOH1-binding peak grouped by cell state. The shaded box indicates the known Atoh1 targets that have been reported in PN development. (D) Gene ontology (GO) enrichment analysis of the DEGs. The representative biological processes are shown with −log10(FDR). (E) TUNEL staining on cRL in control (top) and Atoh1S193A/− (bottom) at E14.5. The Atoh1-lineage cells were labeled with TdTom, and the nuclei were stained with DAPI. (F) The average number of the TUNEL+ cells per region of interest in control and Atoh1S193A/− mice. The gray symbols represent the five ROIs quantified from each animal (n = 3 per genotype). The average number for each animal was shown in colored shape. The crossbar denotes the mean per genotype. *Χ2(1) = 82.41, P = 2.20 × 10−16 by mixed-model ANOVA. Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 6 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 5. PN neurons at E18.5 are heterogeneous and exhibit differential vulnerability to Atoh1 hypomorphic mutation. (A) Expression levels of the selective markers visualized on UMAP. Ccnd1, Atoh1, Nhlh1, and Mapt are known markers for progenitors, intermediate progenitors, migrating neurons, and differentiated neurons, re- spectively. (B) UMAP of the major cell states and ePN subtypes at E18.5. The dashed line denotes the subtype that was significantly reduced in Atoh1S193A/− mice. (C) Violin plot showing the expression levels of the marker genes. Prox1, Ephb1, Hoxb5, and Ntng1 are the marker genes for ePN1, ePN2, ePN3, and ePN4, respectively. (D) Proportion of the cells in each cell state and PN subtype. The arrow indicates the direction of the change in Atoh1S193A/− animals. FDR < 0.01. Fig. 6. scRNA-seq reveals six PN subtypes in mice at P5. (A) UMAP of the major subtypes of the PN neurons at P5. (B) Violin plot showing the expression levels of the marker genes for each PN subtype. (C) Matching the cell type between E18.5 ePN subtypes and P5 PN subtypes by FR test. addition, we identified the marker genes for each subtype by differ- ential gene expression analysis (Fig. 6B, fig. S7A, and data S1). Notably, several PN subtypes at P5 share similar markers with those at E18.5 (figs. S6E and S7A), indicating that PN subtypes might be conserved between these two time points. Thus, we per- formed Friedman-Rafsky (FR) test to match the cell type in E18.5 and P5 datasets (39). We found concordant signatures between the two time points (Fig. 6C), suggesting that the PN neurons have Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 7 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E partially acquired their molecular signatures by E18.5. Notably, we did not find a matched cell type for PN5 and PN6 in the E18.5 data (Fig. 6C). Given that PN5 and PN6 are two of the smallest popula- tions among all subtypes, those two subtypes could be underrepre- sented in the E18.5 data due to the low total number of the PN neurons being sampled. Together, these data confirm that differen- tiated PN neurons are molecularly heterogeneous at both E18.5 and P5. Moreover, the preferential loss of ePN1 subtype at E18.5 in Atoh1S193A/− embryos spurred our interest in testing whether the six PN subtypes at P5 are affected by Atoh1 hypomorphic mutation equally. The six molecularly defined PN subtypes are spatially segregated We ultimately wanted to test whether the six PN subtypes were dif- ferentially compromised in the Atoh1S193A/− mice. However, exam- ining the cellular phenotypes in PN subtypes is challenging without knowing where those subtypes are located. Unlike the well-charac- terized laminated structure in the cerebral cortex, cerebellum, and retina, the cellular organization of the PN has not been delineated. Thus, we first characterized the anatomic location of the six PN sub- types in control animals and then used the spatial information to identify the cellular phenotypes in Atoh1S193A/− animals. We per- formed fluorescence RNA in situ hybridization (ISH) with RNA- Scope HiPlex assay on serial coronal sections from control mouse brain at P5 (Fig. 7A). We used TdTom probes to label Atoh1-lineage neurons and define the region of interest (i.e., PN). By binning the images, we calculated the average expression of the marker genes within each bin across the PN (Fig. 7B). We found that all the markers exhibit spatial specificity except Cdh8. For example, Hoxb5 is expressed in the caudal part of the PN, which is consistent with the previous study (14). Another marker, Etv1, is highly ex- pressed in a restricted part of the medioventral PN. In contrast, Cdh8 is globally expressed across PN, which is expected based on our P5 scRNA-seq data (Fig. 6B). Notably, we found that Somatos- tatin (Sst), a neuropeptide that is typically expressed in inhibitory neurons, is expressed in a subset of the PN neurons that exhibit a shell-like pattern at the rostral PN (Fig. 7B). To date, there are only few studies showing that Sst is coexpressed in subsets of excitatory neurons in pre-Bötzinger complex, lateral hypothalamus, and dor- solateral periaqueductal gray matter (40–42). Here, we demonstrat- ed that Sst is coexpressed in a subset of Slc17a6 + [Vesicular Glutamate Transporter 2 (VGLUT2+)] PN neurons (fig. S7, B and C). To find the corresponding PN subtypes between ISH and scRNA-seq data, we implemented K-nearest neighbor classification method to annotate the six PN subtypes on the representative regions of interest (Fig. 7C). PN neurons have been categorized into two nuclei, RtTg and BPN, based on their anatomic location. However, there has been no evidence showing that they are molec- ularly different. We found that PN6 is largely restricted to BPN, while PN3 and PN5 mostly reside at RtTg. In contrast, PN1, PN2, and PN4 are distributed across both nuclei. These findings suggest that RtTg and BPN have both shared and distinct neuronal sub- types. Together, our ISH data validate the expression of the marker genes identified by scRNA-seq and establish the spatial map of the PN subtypes that provides a higher resolution of the cy- toarchitecture in the PN. PN subtypes have differential vulnerability to Atoh1 hypomorphic mutation With the spatial map of the PN subtypes being established, we next performed fluorescence ISH to characterize PN subtypes in control and Atoh1S193A/− animals at P5. First, we used TdTom probes to identify PN across serial coronal sections. On the basis of the ana- tomic landmarks other than PN, we aligned the sections from dif- ferent animals to proximity and quantified the size of PN in control and Atoh1S193A/− animals. Consistent with the whole-mount lacZ staining at P21 (Fig. 1D), we observed a reduced size of PN in Atoh1S193A/− animals at P5 (Fig. 8A). Moreover, we found that the caudal PN were most affected with 80% reduction at the most caudal section, while there is no significant difference at the rostral PN (Fig. 8, A and B). Given that different PN subtypes exhibit spatial specificity along the rostral-caudal axis (Fig. 7C), we hypothesized that the PN sub- types are differentially affected in Atoh1S193A/− mice. We thus exam- ined three PN subtypes (PN3, PN4, and PN6) in control and Atoh1S193A/− at P5 by performing dual ISH using TdTom probes and marker genes Cdkn1c, Hoxb5, and Etv1, respectively. Cdkn1c is the marker gene for PN3 subtype, which matches ePN1 subtype at E18.5 (Fig. 6C) and is predicted to be reduced in Atoh1S193A/− animals according to our E18.5 scRNA-seq data (Fig. 5D). More- over, on the basis of the spatial map of the PN (Fig. 7C) and the selective loss of the caudal PN in Atoh1S193A/− mice (Fig. 8, A and B), we predicted that Hoxb5+ (PN4) was compromised by partial loss of function of Atoh1. Last, to test whether there is one subtype that is not sensitive to Atoh1 hypomorphic mutation, we chose Etv1+ subtype (PN6). Given its specific location within the PN (Fig. 7C), changes in PN, if any, should be easily detected. We examined serial sections across the whole PN to exclude the possi- bility that certain subtype might be mislocated. We found fewer Cdkn1c+TdTom+ PN3 neurons in Atoh1S193A/− mice, accompanied by reduced size in the area where the Cdkn1c+ PN3 neurons are normally located in controls (Fig. 8, C and D, top). Hoxb5+TdTom+ PN4 neurons were also reduced in Atoh1S193A/− animals at the caudal sections (Fig. 8, C and D, middle). In contrast, Etv1+TdTom+ PN6 neurons were not affected in Atoh1S193A/− animals (Fig. 8, C and D, bottom). Together, these data demonstrated that although all PN subtypes were derived from Atoh1 + progenitors, certain PN subtypes such as PN3 and PN4 neurons were more vulnerable to partial loss of function of Atoh1, suggesting that the robustness of Atoh1 function is critical for PN subtype cell fate decisions. DISCUSSION In this study, we used a mouse model carrying an Atoh1 hypomor- phic mutation and implemented scRNA-seq technology to eluci- date the roles of Atoh1 in different cell states during PN development. We demonstrate that Atoh1 is involved in multiple biological processes including regulating the cell cycle, differentia- tion, cell survival, migration, and the heterogeneity of the PN neurons. Moreover, we show that Atoh1-lineage PN neurons are classified as six subtypes based on their molecular signatures and provide a list of marker genes for future studies. PN subtypes display differential vulnerability to Atoh1 hypomorphic mutation, which opens up questions such as how cell fate decisions are made and how perturbation of a transcription factor results in subtype-specific phenotypes. Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 8 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 7. The spatial map of the PN subtypes in mice at P5. (A) Workflow of the RNAScope HiPlex Assay on P5 control mouse. (B) Expression patterns of the selective markers for each PN subtype across the rostral to caudal axis of the PN. (C) Schematic summary of the distribution of the PN subtypes across the PN. The dashed line denotes the border between RtTg and BPN. The interplay between Atoh1 and Notch signaling One interesting phenotype that we observed in Atoh1S193A/− mice was the marked increase in proliferating progenitors (Figs. 2F and 3B), suggesting that Atoh1 might promote cell cycle exit. In addi- tion, differential gene expression analysis revealed several dysregu- lated pathways including up-regulated Notch signaling in the proliferating progenitors (Fig. 4D). The interaction between Atoh1 and Notch signaling has been demonstrated in the mamma- lian intestine (43, 44). In the epithelium lining the crypts in the small intestine, the stem cells differentiate into secretory cells when they escape Notch activation by up-regulation of Atoh1. In contrast, the stem cells in which Notch is activated remain as Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 9 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Fig. 8. PN subtypes have different vulnerabilities to Atoh1 hypomorphic mutation. (A) The size of the PN was determined by the areas across seven coronal sections (n = 3 per genotype). Data are presented as means ± SD. P < 0.05; P < 0.01; **P < 0.001 by two-tailed unpaired t test. (B) Representative RNA ISH images for the most rostral (top) and the most caudal (bottom) sections from control (left) and Atoh1S193A/− (right) mice at P5. The PN neurons were labeled with TdTom probes. The nuclei were stained with DAPI. The dashed line encloses the area of the PN. Scale bars, 500 μm. (C) RNA ISH on control (left) and Atoh1S193A/− (right) mice at P5 using Cdkn1c (top), Hoxb5 (middle), and Etv1 (bottom) probes. PN neurons were labeled with TdTom probes. The nuclei were stained with DAPI. The box denotes the area shown in (D). Scale bars, 200 μm. (D) The zoom-in view of PN from the boxes in (C). The expression of Cdkn1c (top), Hoxb5 (middle), and Etv1 (bottom) were shown in green. Scale bars, 20 μm. progenitors at the crypts where Wnt is high. Here, we hypothesize that in cRL, partial loss of function of Atoh1 leads to failure to escape Notch activation, which maintains the cells at the proliferat- ing state. In line with this hypothesis, we observed up-regulation of the Notch receptor Notch1 and its ligand Dll1 as well as genes down- stream from active Notch signaling including Hes1 and Hes5 in the proliferating progenitors of Atoh1S193A/− mice (data S2). Moreover, a recent study in zebrafish showed that inhibition of Notch activity at lower RL led to increased atoh1b+ postmitotic precursors at the expense of atoh1a+ proliferating progenitors (45). This study also supports our hypothesis that Atoh1 and Notch antagonize each other at the cRL in mammals. The molecular heterogeneity of the PN neurons One of the most puzzling questions in neurobiology is how to define a neuronal subtype. In the case of the PN, they have been considered as a group of heterogeneous neurons based on their anatomical lo- cation, origins at the cRL, and functions based on the cortico- ponto-cerebellar connectivity (13–16, 46). However, whether the PN neurons can be further defined by their molecular signature has not been shown. scRNA-seq provides a powerful approach to classify the cell type unbiasedly based on transcription profiles. Here, we uncovered six PN subtypes in P5 mice (Fig. 6, A and B). These six PN subtypes were spatially segregated (Fig. 7), which raises an interesting question regarding whether this spatial map re- flects the topographic connectivity between the cerebral cortex and Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 10 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E the PN. For example, Sst+ PN subtype reside at the rostral dorsal part of the PN (Fig. 7B). This location coincides with where PN receive inputs from brain regions involved in visual pathway includ- ing visual cortex, superior colliculus, inferior colliculus, and pretec- tum (47). Whether there is a subtype-specific connectivity that also reflects the functionality needs further investigation. However, those studies cannot be done without genetic tools to target differ- ent PN subtypes. Our study identified and validated the marker genes for the six PN subtypes. In future studies, one can test whether different PN subtypes preferentially connect with specific groups of neurons in the cerebral cortex and/or cerebellum by in- tersectional labeling and viral tracing approaches with the molecu- lar markers. In addition to the connectivity, the molecular markers that we identified can be used to manipulate PN neurons in a subtype-specific manner for functional characterization. From a pool of Atoh1+ progenitors to diverse PN subtypes How does a pool of seemingly homogeneous progenitors give rise to a diverse population of neurons? In the case of Atoh1-lineage neurons in the hindbrain, Atoh1 + progenitors contribute to neurons in the cerebellum and dozens of brainstem nuclei depend- ing on the rostrocaudal origins at RL and the timing of leaving from RL (9, 10, 48). PN, for example, were derived from rhombomere 6 to 8 where Hox2 to Hox5 are expressed (10). It has not been addressed, however, whether Atoh1 contributes to the diversity of the subpo- pulation within one lineage (i.e., PN in this study). To this end, we tested whether PN subtypes display differential vulnerability to Atoh1 hypomorphic mutation. On the basis of our scRNA-seq data at E18.5 and the histological analysis at P5, partial loss of func- tion of Atoh1 not only reduced the size of the PN but also reduced the diversity of the PN neurons by preferentially affecting PN3 and PN4 subtypes (Fig. 8, C and D). These data suggest that Atoh1 may contribute to the acquisition or maintenance of the heterogeneity of the PN neurons. In summary, this study dissected the functions of Atoh1 during PN differentiation by characterizing the phenotypes in Atoh1 hypo- morphic mutant at single-cell resolution. Our data demonstrate that Atoh1 regulates cell cycle exit, differentiation, migration, and sur- vival during PN development and contributes to the diversity of the PN subtypes. In addition, this study also uncovers the molecular heterogeneity of the PN, which opens new doors for understanding the neural fate decisions, connectivity, and functionality of the PN neurons. MATERIALS AND METHODS Mice The following mouse lines were used in this study: Atoh1lacZ/+ (23), Atoh1S193A/+ (24), Atoh1Cre/+ (28), Rosalsl-lacZ/+ (JAX:02429), and Rosalsl-TdTomato/+ (JAX:007914). All mice were housed in a level 3, American Association for Laboratory Animal Science (AALAS)– certificated facility on a 14-hour light cycle. Husbandry, housing, euthanasia, and experimental guidelines were approved by the In- stitutional Animal Care & Use Committee (IACUC) at Baylor College of Medicine. Whole-mount X-galactosidase staining The brains of P0 pups and P21 mice were dissected out in ice-cold phosphate-buffered saline (PBS). The samples were fixed in 4% paraformaldehyde (PFA) at 4°C for 1 hour (P0) or 2 hours (P21). After brief PBS wash at room temperature (RT), the samples were incubated in equilibration buffer [2 mM MgCl2, 0.05% sodium de- oxycholate, 0.02% NP-40, and 0.1 M sodium phosphate (pH 7.3)] at 4°C for 15 min, followed by 2-hour incubation at 37°C in X-galac- tosidase (X-gal) reaction solution [X-gal (1 mg/ml), 5 mM potassi- um ferrocyanide, and 5 mM potassium ferricyanide in equilibration buffer]. After staining, the samples were washed with PBS three times at RT and fixed again in 4% PFA at 4°C for 1 hour (P0) or 4 hours (P21) before imaging. Sample preparation for scRNA-seq The brains of embryos and P5 pups were dissected out in ice-cold HEBG medium (0.8× B27 and 0.25× GlutaMAX in hibernate E medium). For embryonic studies, hindbrains were collected from Atoh1S193A/− and its littermate control (one embryo per genotype, three independent replicates). For P5 study, PN were microdis- sected out and pooled from 11 to 13 pups (two independent replicates). Tissues were cut into small pieces and transferred to a 1.5-ml microcentrifuge tube using wide-bore pipette tips. The single-cell dissociation protocol was modified from the previous study (49). Briefly, the tissues were incubated with Worthington Papain solution at 37°C for 30 min at 800 rpm. At the end of incu- bation, the samples were transferred to a 15-ml falcon tube, followed by gentle trituration with serologic pipette. The cell pellets were col- lected by centrifuge at 200 rcf for 3 min at 4°C, washed, and resuspended in ice-cold sorting buffer (PBS with 0.05% fetal bovine serum). The single-cell resuspension was loaded to 30-μm cell drainer (CellTrics, SYSMEX 04-004-2326) to remove debris, followed by 40,6-diamidino-2-phenylindole (DAPI) staining at RT for 5 min. TdTom+DAPI− cells (120,000 to 150,000 cells per sample) were sorted into bovine serum albumin–coated 15-ml falcon tube by Sony SH800S cell sorter. The cells were pelleted by centrifuge at 200 rcf for 5 min at 4°C and resuspended in sorting buffer to make the final concentration of 1000 cells/μl. Library construction and sequencing The cDNA libraries were constructed by 10X Genomics 30 v3.1 kit following the user guide. Briefly, ~16,500 cells from one sample were mixed with reversed transcription master mix before loaded into Chromium Chip G. Droplets containing cells, reversed tran- scription reagents, and barcoded gel beads were generated by Chromium Controller. The first strand cDNA was amplified, fragmentated, and ligated with sequencing adaptors and sample indices. The cDNA libraries were sequenced by Illumina NovaSeq 6000. RNAScope HiPlex assay The brain of the P5 pup was flash-frozen and embedded in optimal cutting temperature (OCT) compound. Coronal sections were made by cryostat (Leica) at 20 μm and store in −80°C until use. The RNAScope HiPlex assay (ACD, catalog no. 324419) was per- formed according to the manufacturer’s instructions. Briefly, sec- tions were fixed in 4% PFA for 1 hour at RT, followed by dehydration with 50, 70, and 100% ethanol. The protease treatment was done by incubating the sections with protease III at RT for 30 min. Sections were hybridized with the 12 probes (catalog nos. 487941-T1, 845141-T2, 480301-T3, 557891-T4, 319171-T5, 458331-T6, 485461-T7, 503461-T8, 404631-T9, 503481-T10, Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 11 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E 317041-T11, and 433411-T12) at 40°C for 2 hours, followed by three rounds of amplification. The TdTom reporter was helpful during sectioning to ensure that we covered the whole PN. However, it also introduced hazy background in HiPlex assay. Thus, we quenched the TdTom signal by incubating the sections with 5% for- malin-fixed paraffin-embedded (FFPE) reagent (included in the kit) at RT for 30 min before developing the first round of the targets (T1 to T3). The nuclei were stained with DAPI. The 12 targets were de- tected in four rounds of imaging with three targets per round. Between each round, the fluorophores were cleaved and washed away before the next round of signal development. After image ac- quisition of all targets, the fluorophores were cleaved, and a blank image of each section was taken to serve as background images. Immunofluorescence staining The heads of the E14.5 embryos were dissected in ice-cold PBS. After brief wash, the samples were fixed in 4% PFA at 4°C for 4 hours, followed by PBS wash and incubation in 30% sucrose in PBS at 4°C for 14 to 16 hours. The samples were cryopreserved and stored at −80°C until use. The coronal sections were collected on slides by cryostat (Leica) with 20 or 25 μm in thickness. The slides were rinsed with PBS to remove OCT, followed by permeabi- lization with 0.3% Triton X-100 in PBS for 15 min at RT. Antigen retrieval was performed by heating in antigen retrieval buffer [10 mM sodium citrate and 0.05% Tween-20 (pH 6.0)] at 85°C for 10 min [MKI67/red fluorescent protein (RFP) staining] or 30 min (CldU labeling). After blocking with blocking buffer (5% normal goat serum with 0.3% Triton X-100 in PBS) for 2 hours at RT, the sections were incubated with primary antibodies in blocking buffer at 4°C for 24 hours, followed by PBS wash three times. The sections were incubated with secondary antibodies in blocking buffer at RT for 2 hours. The counterstain was performed by DAPI staining at RT for 10 min. The slides were mounted with ProLong Gold mounting media (Invitrogen). The following antibodies were used in this study with the indicated dilutions: rat anti–5-bromo-20-de- oxyuridine (BrdU)/CldU (1:250; Abcam, ab6326), mouse anti-Ki67 (1:100; R&D Systems, AF7649), rabbit anti-RFP (1:2000; Rockland, 600-401-379), goat anti-rat Alexa Fluor 488 (1:500; Invitrogen, A- 11006), goat anti-rabbit Alexa Fluor 555 (1:500; Invitrogen, A- 21428), goat anti-mouse Alexa Fluor 647 (1:500; Invitrogen, A-21236). CldU labeling CldU was prepared in sterile saline at 4.82 mg/ml and injected into pregnant dams intraperitoneally at E13.5 (85 mg/kg). Twenty-four hours later, the embryos were collected at E14.5, followed by immu- nofluorescence staining using rat anti-BrdU/CldU antibody (1:250; Abcam, ab2326). TUNEL assay The sample preparation follows the same protocol that we used for immunofluorescence staining. Coronal sections were made by cryo- stat (Leica) at 25 μm. TUNEL assay was performed using DeadEnd Fluorometric TUNEL system (Promega) according to the user guide. Briefly, sections were fixed in 4% PFA at RT for 15 min, fol- lowed by proteinase K (20 μg/ml) treatment at RT for 12 min and a second fixation with 4% PFA for 5 min. The sections were equili- brated and incubated with fluorescein-labeled nucleotides and ter- minal deoxynucleotidyl transferase reaction mix at 37°C for 1 hour. After stopping the reaction by 2× saline-sodium citrate (SSC), the nuclei were stained with DAPI. Dual-color fluorescence RNA ISH The brains from P5 pups were embedded in OCT, frozen on dry ice, and stored in −80°C until use. The sections were cut by cryostat (Leica) at 20 μm. We generated a digoxigenin (DIG)–labeled mRNA antisense probes against Slc17a6 and TdTom and fluoresce- in isothiocyanate (FITC)–labeled mRNA against Sst, Etv1, and Hoxb5 using reverse-transcribed mouse cDNA as template and DIG or FITC RNA labeling kits from Roche (Sigma-Aldrich). Primer sequences for Sst, Slc17a6, and TdTom probes are available in Allen Brain Atlas (www.brain-map.org). The following primers were used: 50-ttcagaactcgggtctgctt-30 and 50-gaatcatgcaaaaggtggct-30 for Etv1 probe and 50-gatggatctcagcgtcaacc-30 and 50-tatgagtctggcta- cagccg-30 for Hoxb5 probe. ISH was performed by the RNA In Situ Hybridization Core at Baylor College of Medicine using an auto- mated robotic platform as previously described (50) with modifica- tions of the protocol for double ISH. Briefly, two probes were hybridized to the tissue simultaneously. After the wash and block- ing steps, the DIG-labeled probes were visualized by incubating with tyramide-Cy3 Plus (1:75; PerkinElmer) for 15 min. After washing in TNT (0.05% Tween-20 in 150mM NaCl and 100mM Tris-HCl, pH7.5), the remaining horseradish peroxidase (HRP) ac- tivity was quenched by a 10-min incubation in 0.2 M HCl. The sec- tions were then washed in TNT and blocked in TNB for 15 min, followed by incubation with HRP-labeled sheep anti-FITC antibody (1:500 in TNB; Roche) at RT for 30 min. After washes in TNT, the FITC-labeled probes were visualized by incubating with tyramide- FITC Plus (1:75; PerkinElmer) for 15 min. Following washes in TNT, the slides were stained with DAPI (Invitrogen), washed again, removed from the machine, and mounted with ProLong Diamond (Invitrogen). Image acquisition Images of the whole-mount tissues were obtained by Zeiss Axio Zoom.V16. All fluorescence images were obtained by Nikon Ti2E Inverted Motorized Microscope equipped with CSU-W1 Dual Camera-Dual Disk System with 405/488/561/640-nm lasers with 10×, 20×, or 40× objectives. Data preprocessing The reads were aligned to customized genome that composed of mouse mm10 reference genome and woodchuck hepatitis virus post-transcriptional regulatory element (WPRE) sequence to ensure capturing the TdTom transcripts. The alignment and quan- tification of unique molecular identifier (UMI) were performed on 10× cloud analysis platform by CellRanger pipeline v5.0.1 with default parameters. Individual expression matrix of each sample was filtered by Seurat v4 package (51), followed by doublet removal using DoubletFinder v2 package (52). Briefly, cells with at least 1000 genes expressed and less than 1% of total UMIs that were mitochondrial genes were retained (see the code for detailed criteria depending on the age of the samples.) Doublets with high confidence score identified by DoubletFinder with default parame- ters were removed. Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 12 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E Integration and clustering The data integration and clustering were performed by Seurat package. The filtered datasets of the samples from the same age were combined by integration pipeline. Briefly, the filtered matrices were normalized and scaled by regressing out the percentage of mi- tochondrial genes with SCTransform (53). For E14.5 and E18.5 samples in which different genotypes were compared, a reference- based integration method was used by setting the first replicate as reference to cut down the computational power and time needed. For P5 study, the default integration method was used. Following the integration, principal components analysis (PCA) was per- formed. We selected the top 30 PCAs to generate the K-nearest neighbor graph, which was used to perform clustering with variated resolutions depending on the complexity of the dataset. To annotate the clusters, the top 10 genes that were highly expressed in each cluster were identified by FindAllMarkers function with default parameters. Trajectory analysis The RNA assay of the Seurat object was extracted and transformed into SingleCellExperiment object (sce) by as.SingleCellExperiment function. The sce was used as input to infer the trajectory by sling- shot (30). Kolmogorov-Smirnov test was used to assess whether the distribution of pseudo-time is identical between the genotypes. Differential expression analysis We performed differential expression analysis between control and Atoh1S193A/− samples for each cell state by FindMarkers function from Seurat package. Only genes that were expressed in at least 10% of the cells were included. MAST (54) was used as test method. The P value was corrected by Benjamini-Hochberg FDR method. GO analysis GO analysis was performed using a web-based tool called g:Profiler (55) with custom statistic domain scope. The up- and down-regu- lated DEGs (log2FC > 0.25 and FDR < 0.05) within progenitors, in- termediate progenitors, and migrating neurons-1 were used as inputs independently. GO biological process was selected for the analysis. Image processing for RNAScope HiPlex assay Five confocal z-stacks of images were collected at 100- and 200-μm intervals along the rostral-caudal axis of the pontine nucleus for each of the nine transcripts. In addition, DAPI images were also col- lected to serve as reference images. Individual z-stacks consisted of 10 images separated by 0.9 μm. The z-stacks were first processed by taking the max intensity projection and cropping the resulting tran- script image such that the midline of the pontine aligned with the right-hand border of each image. Visualizing transcript expression Overlaying the transcript images onto the DAPI reference images (56) revealed that some transcripts were not consistently colocalized to the nucleus. To address this ambiguity in our analysis, we did not associate transcript expression with individual cells. Rather, we binned each image into a set of tiles measuring 34 μm × 31 μm and measured the transcript expression in each tile (57). Since tran- scripts may be expressed at very different levels, we normalized the expression for each transcript by the max expression recorded across the five image locations in the pontine. Thus, in Fig. 7B, the expression for each transcript varies from 0 to 1 across the five rostral to caudal sections. To identify the boundaries of the pontine, the max intensity projection across all the transcripts at a given imaging location was smoothed with a Gaussian filter with an SD of 35 μm. This filtered image was converted to a binary image by mean thresholding, and any remaining holes were morphologically closed using a 5-μm × 5-μm structuring element. Last, the pontine boundary was taken to be the edge of the largest single region in the closed binary image. Quantification of the immunostaining and statistics The cell number of the CldU+TdTom+ (Fig. 3F) and TUNEL+ (Fig. 4F) was determined by manual counting on imageJ. The per- centage of MKI67 signal overlapping with TdTom (Fig. 3C) was cal- culated based on Manders’ coefficients using JACoP (58). For the statistical test, we used mixed-model analysis of variance (ANOVA) to compare genotype using lme4 package on R (59) to count for variations among technical and biological replicates. The person who performed the quantification was blinded to the genotypes of the samples. Classification of RNA ISH transcripts The transcript expression levels for each tile in the ISH images can be represented as a nine-dimensional vector, one component for each transcript. To classify these transcript expression vectors as one of the six classes determined from the single-cell RNA-seq anal- ysis, we built a K-nearest neighbor classifier (60). Specifically, for each cell (n = 7029) from the RNA-seq clustering results, we extract- ed the same transcripts as measured in the ISH and the cluster iden- tity. Each of these nine-component vectors was normalized (60) and, along with their respective cluster IDs, was used to train and validate our classifier. The K-nearest neighbor classifier measures the distance between each nine-component vector and a preset number of neighbor vectors to predict class membership. To deter- mine the number of neighbors, we performed sixfold cross-valida- tion on the training dataset and found that 25 neighbors provided an average test accuracy of 80%. We then used this 25 nearest neigh- bor model to predict the cluster identity of each tile in the ISH images. The border of the RtTg and BPN in Fig. 7C was defined using the reference atlas of P6 mouse brain (61). Supplementary Materials This PDF file includes: Figs. S1 to S7 Table S1 Legends for data S1 to S3 References Other Supplementary Material for this manuscript includes the following: Data S1 to S3 View/request a protocol for this paper from Bio-protocol. REFERENCES AND NOTES 1. E. V. Evarts, W. T. Thach, Motor mechanisms of the CNS: Cerebrocerebellar interrelations. Annu. Rev. Physiol. 31, 451–498 (1969). Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 13 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E 2. C. R. Legg, B. Mercier, M. Glickstein, Corticopontine projection in the rat: The distribution of labelled cortical cells after large injections of horseradish peroxidase in the pontine nuclei. J. Comp. Neurol. 286, 427–441 (1989). 27. P. Soriano, Generalized lacZ expression with the ROSA26 Cre reporter strain. Nat. Genet. 21, 70–71 (1999). 28. H. Yang, X. Xie, M. Deng, X. Chen, L. Gan, Generation and characterization of Atoh1-Cre 3. P. Brodal, J. G. Bjaalie, Organization of the pontine nuclei. Neurosci. Res. 13, 83–118 (1992). knock-in mouse line. Genesis 48, 407–413 (2010). 4. C. Schwarz, P. Thier, Binding of signals relevant for action: Towards a hypothesis of the 29. B. Phipson, C. B. Sim, E. R. Porrello, A. W. Hewitt, J. Powell, A. Oshlack, Propeller: Testing for functional role of the pontine nuclei. Trends Neurosci. 22, 443–451 (1999). 5. J. D. Schmahmann, R. Ko, J. MacMore, The human basis pontis: Motor syndromes and differences in cell type proportions in single cell data. Bioinformatics 38, 4720–4726 (2022). topographic organization. Brain 127, 1269–1291 (2004). 6. K. Tziridis, P. W. Dicke, P. Thier, Pontine reference frames for the sensory guidance of movement. Cereb. Cortex 22, 345–362 (2012). 7. G.-Y. Wu, S.-L. Liu, J. Yao, L. Sun, B. Wu, Y. Yang, X. Li, Q.-Q. Sun, H. Feng, J.-F. Sui, Medial prefrontal cortex-pontine nuclei projections modulate suboptimal cue-induced associative motor learning. Cereb. Cortex 28, 880–893 (2018). 8. C. I. Rodriguez, S. M. Dymecki, Origin of the precerebellar system. Neuron 27, 475–486 (2000). 9. V. Y. Wang, M. F. Rose, H. Y. Zoghbi, Math1 expression redefines the rhombic lip derivatives and reveals novel lineages within the brainstem and cerebellum. Neuron 48, 31–43 (2005). 10. A. F. Farago, R. B. Awatramani, S. M. Dymecki, Assembly of the brainstem cochlear nuclear complex is revealed by intersectional and subtractive genetic fate maps. Neuron 50, 205–218 (2006). 11. J. Altman, S. A. Bayer, Development of the precerebellar nuclei in the rat: IV. The anterior precerebellar extramural migratory stream and the nucleus reticularis tegmenti pontis and the basal pontine gray. J. Comp. Neurol. 257, 529–552 (1987). 12. T. Okada, K. Keino-Masu, M. Masu, Migration and nucleogenesis of mouse precerebellar neurons visualized by in utero electroporation of a green fluorescent protein gene. Neu- rosci. Res. 57, 40–49 (2007). 13. A. Brodal, J. Jansen, The ponto-cerebellar projection in the rabbit and cat; experimental investigations. J. Comp. Neurol. 84, 31–118 (1946). 14. T. D. Meglio, C. F. Kratochwil, N. Vilain, A. Loche, A. Vitobello, K. Yonehara, S. M. Hrycaj, B. Roska, A. H. F. M. Peters, A. Eichmann, D. Wellik, S. Ducret, F. M. Rijli, Ezh2 orchestrates topographic migration and connectivity of mouse precerebellar neurons. Science 339, 204–207 (2013). 15. T. B. Leergaard, K. A. Lyngstad, J. H. Thompson, S. Taeymans, B. P. Vos, E. de Schutter, J. M. Bower, J. G. Bjaalie, Rat somatosensory cerebropontocerebellar pathways: Spatial relationships of the somatotopic map of the primary somatosensory cortex are preserved in a three-dimensional clustered pontine map. J. Comp. Neurol. 422, 246–266 (2000). 16. J. U. Henschke, J. M. Pakan, Disynaptic cerebrocerebellar pathways originating from multiple functionally distinct cortical areas. eLife 9, e59148 (2020). 17. N. A. Bermingham, B. A. Hassan, V. Y. Wang, M. Fernandez, S. Banfi, H. J. Bellen, B. Fritzsch, H. Y. Zoghbi, Proprioceptor pathway development is dependent on Math1. Neuron 30, 411–422 (2001). 18. N. A. Bermingham, B. A. Hassan, S. D. Price, M. A. Vollrath, N. Ben-Arie, R. A. Eatock, H. J. Bellen, A. Lysakowski, H. Y. Zoghbi, Math1: An essential gene for the generation of inner ear hair cells. Science 284, 1837–1841 (1999). 19. S. M. Maricich, S. A. Wellnitz, A. M. Nelson, D. R. Lesniak, G. J. Gerling, E. A. Lumpkin, H. Y. Zoghbi, Merkel cells are essential for light-touch responses. Science 324, 1580–1582 (2009). 20. N. F. Shroyer, M. A. Helmrath, V. Y.-C. Wang, B. Antalffy, S. J. Henning, H. Y. Zoghbi, Intestine- specific ablation of mouse atonal homolog 1 (Math1) reveals a role in cellular homeostasis. Gastroenterology 132, 2478–2488 (2007). 21. M. F. Rose, K. A. Ahmad, C. Thaller, H. Y. Zoghbi, Excitatory neurons of the proprioceptive, interoceptive, and arousal hindbrain networks share a developmental requirement for Math1. Proc. Natl. Acad. Sci. U.S.A. 106, 22462–22467 (2009). 22. T. Višnjar, A. Maver, K. Writzl, O. Maloku, G. Bergant, H. Jakliˇ, D. Neubauer, F. Fogolari, N. P. Megliˇ, B. Peterlin, Biallelic ATOH1 gene variant in siblings with pontocerebellar hy- poplasia, developmental delay, and hearing loss. Neurol. Genet. 8, e677 (2022). 23. N. Ben-Arie, B. A. Hassan, N. A. Bermingham, D. M. Malicki, D. Armstrong, M. Matzuk, 30. K. Street, D. Risso, R. B. Fletcher, D. Das, J. Ngai, N. Yosef, E. Purdom, S. Dudoit, Slingshot: Cell lineage and pseudotime inference for single-cell transcriptomics. BMC Genomics 19, 477 (2018). 31. H. R. de Bézieux, K. Van den Berge, K. Street, S. Dudoit, Trajectory inference across multiple conditions with condiments: Differential topology, progression, differentiation, and ex- pression. bioRxiv 2021.03.09.433671 [Preprint]. 10 March 2021. https://doi.org/10.1101/ 2021.03.09.433671. 32. T. J. Klisch, Y. Xi, A. Flora, L. Wang, W. Li, H. Y. Zoghbi, In vivo Atoh1 targetome reveals how a proneural transcription factor regulates cerebellar development. Proc. Natl. Acad. Sci. U.S.A. 108, 3288–3293 (2011). 33. A. W. Helms, A. L. Abney, N. Ben-Arie, H. Y. Zoghbi, J. E. Johnson, Autoregulation and multiple enhancers control Math1 expression in the developing nervous system. Devel- opment 127, 1185–1196 (2000). 34. S. Li, F. Qiu, A. Xu, S. M. Price, M. Xiang, Barhl1 regulates migration and survival of cerebellar granule cells by controlling expression of the neurotrophin-3 gene. J. Neurosci. 24, 3104–3114 (2004). 35. T. Schmid, M. Kruger, T. Braun, NSCL-1 and -2 control the formation of precerebellar nuclei by orchestrating the migration of neuronal precursor cells. J. Neurochem. 102, 2061–2072 (2007). 36. T. Cai, H.-I. Jen, H. Kang, T. J. Klisch, H. Y. Zoghbi, A. K. Groves, Characterization of the transcriptome of nascent hair cells and identification of direct targets of the Atoh1 tran- scription factor. J. Neurosci. 35, 5870–5883 (2015). 37. H. V. Yu, L. Tao, J. Llamas, X. Wang, J. D. Nguyen, T. Trecek, N. Segil, POU4F3 pioneer activity enables ATOH1 to drive diverse mechanoreceptor differentiation through a feed-forward epigenetic mechanism. Proc. Natl. Acad. Sci. U.S.A. 118, e2105137118 (2021). 38. H. C. Lai, T. J. Klisch, R. Roberts, H. Y. Zoghbi, J. E. Johnson, In vivo neuronal subtype-specific targets of Atoh1 (Math1) in dorsal spinal cord. J. Neurosci. 31, 10859–10871 (2011). 39. Y. Zhang, B. Aevermann, R. Gala, R. H. Scheuermann, Cell type matching in single-cell RNA- sequencing data using FR-match. Sci. Rep. 12, 9996 (2022). 40. R. L. Stornetta, D. L. Rosin, H. Wang, C. P. Sevigny, M. C. Weston, P. G. Guyenet, A group of glutamatergic interneurons expressing high levels of both neurokinin-1 receptors and somatostatin identifies the region of the pre-Bötzinger complex. J. Comp. Neurol. 455, 499–512 (2003). 41. L. E. Mickelsen, M. Bolisetty, B. R. Chimileski, A. Fujita, E. J. Beltrami, J. T. Costanzo, J. R. Naparstek, P. Robson, A. C. Jackson, Single-cell transcriptomic analysis of the lateral hypothalamic area reveals molecularly distinct populations of inhibitory and excitatory neurons. Nat. Neurosci. 22, 642–656 (2019). 42. N. Winke, F. Aby, D. Jercog, G. Zoé, D. Girard, M. Landry, L. Castell, E. Valjent, S. Valerio, P. Fossat, C. Herry, Brainstem somatostatin-expressing cells control the emotional regula- tion of pain behavior. bioRxiv 2022.01.20.476899 [Preprint]. 22 January 2022. https://doi. org/10.1101/2022.01.20.476899. 43. Q. Yang, N. A. Bermingham, M. J. Finegold, H. Y. Zoghbi, Requirement of Math1 for se- cretory cell lineage commitment in the mouse intestine. Science 294, 2155–2158 (2001). 44. R. Sancho, C. A. Cremona, A. Behrens, Stem cell and progenitor fate in the mammalian intestine: Notch and lateral inhibition in homeostasis and disease. EMBO Rep. 16, 571–581 (2015). 45. I. Belzunce, C. Belmonte-Mateos, C. Pujades, The interplay of atoh1 genes in the lower rhombic lip during hindbrain morphogenesis. PLOS ONE 15, e0228225 (2020). 46. R. V. Sillitoe, Y. Fu, C. Watson, in The Mouse Nervous System, C. Watson, G. Paxinos, L. Puelles, Eds. (Academic Press, San Diego, 2012), pp. 260–397. H. J. Bellen, H. Y. Zoghbi, Functional conservation of atonal and Math1in the CNS and PNS. Development 127, 1039–1048 (2000). 47. C. F. Kratochwil, U. Maheshwari, F. M. Rijli, The long journey of pontine nuclei neurons: From rhombic lip to cortico-ponto-cerebellar circuitry. Front. Neural Circuits 11, 33 (2017). 24. W. R. Xie, H.-I. Jen, M. L. Seymour, S.-Y. Yeh, F. A. Pereira, A. K. Groves, T. J. Klisch, H. Y. Zoghbi, An Atoh1-S193A phospho-mutant allele causes hearing deficits and motor impairment. J. Neurosci. 37, 8583–8594 (2017). 25. X. Jin, S. K. Simmons, A. Guo, A. S. Shetty, M. Ko, L. Nguyen, V. Jokhi, E. Robinson, P. Oyler, N. Curry, G. Deangeli, S. Lodato, J. Z. Levin, A. Regev, F. Zhang, P. Arlotta, In vivo perturb-seq reveals neuronal and glial abnormalities associated with autism risk genes. Science 370, eaaz6063 (2020). 48. R. Machold, G. Fishell, Math1 is expressed in temporally discrete pools of cerebellar rhombic-lip neural progenitors. Neuron 48, 17–24 (2005). 49. R. Chen, X. Wu, L. Jiang, Y. Zhang, Single-cell RNA-seq reveals hypothalamic cell diversity. Cell Rep. 18, 3227–3241 (2017). 50. M. B. Yaylaoglu, A. Titmus, A. Visel, G. Alvarez-Bolado, C. Thaller, G. Eichele, Comprehensive expression atlas of fibroblast growth factors and their receptors generated by a novel robotic in situ hybridization platform. Dev. Dyn. 234, 371–386 (2005). 26. I. Schaffner, M.-T. Wittmann, T. Vogel, D. C. Lie, Differential vulnerability of adult neurogenic niches to dosage of the neurodevelopmental-disorder linked gene Foxg1. Mol. Psychiatry 28, 497–514 (2023). 51. Y. Hao, S. Hao, E. Andersen-Nissen, W. M. Mauck III, S. Zheng, A. Butler, M. J. Lee, A. J. Wilk, C. Darby, M. Zager, P. Hoffman, M. Stoeckius, E. Papalexi, E. P. Mimitou, J. Jain, A. Srivastava, T. Stuart, L. M. Fleming, B. Yeung, A. J. Rogers, J. M. McElrath, C. A. Blish, R. Gottardo, Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 14 of 15 S C I E N C E A D VA N C E S \| R E S E A R C H A R T I C L E P. Smibert, R. Satija, 3573–3587.e29 (2021). Integrated analysis of multimodal single-cell data. Cell 184, 52. C. S. McGinnis, L. M. Murrow, Z. J. Gartner, DoubletFinder: Doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 8, 329–337.e4 (2019). 53. C. Hafemeister, R. Satija, Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol. 20, 296 (2019). 54. G. Finak, A. McDavid, M. Yajima, J. Deng, V. Gersuk, A. K. Shalek, C. K. Slichter, H. W. Miller, M. J. McElrath, M. Prlic, P. S. Linsley, R. Gottardo, MAST: A flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA se- quencing data. Genome Biol. 16, 278 (2015). 55. U. Raudvere, L. Kolberg, I. Kuzmin, T. Arak, P. Adler, H. Peterson, J. Vilo, G:Profiler: A web server for functional enrichment analysis and conversions of gene lists (2019 update). Nucleic Acids Res. 47, W191–W198 (2019). 56. J. D. Hunter, Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 9, 90–95 (2007). 57. S. van der Walt, J. L. Schönberger, J. Nunez-Iglesias, F. Boulogne, J. D. Warner, N. Yager, E. Gouillart, T. Yu; scikit-image contributors, Scikit-image: Image processing in Python. PeerJ 2, e453 (2014). 58. S. Bolte, F. P. Cordelières, A guided tour into subcellular colocalization analysis in light microscopy. J. Microsc. 224, 213–232 (2006). 59. D. Bates, M. Mächler, B. Bolker, S. Walker, Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015). 60. C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers, P. Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. Kern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. F. Del Río, M. Wiebe, P. Peterson, P. Gérard-Marchant, K. Sheppard, T. Reddy, W. Weckesser, H. Abbasi, C. Gohlke, T. E. Oliphant, Array programming with NumPy. Nature 585, 357–362 (2020). 61. G. Paxinos, Atlas of the Developing Mouse Brain at E17.5, P0 and P6 (Elsevier, Amsterdam; Boston, ed. 1st, 2007), pp. xi, 353 p. 62. M. F. Rose, J. Ren, K. A. Ahmad, H.-T. Chao, T. J. Klisch, A. Flora, J. J. Greer, H. Y. Zoghbi, Math1 is essential for the development of hindbrain neurons critical for perinatal breathing. Neuron 64, 341–354 (2009). Acknowledgments: The content is solely the responsibility of the authors and does not necessarily represent the official views of the Eunice Kennedy Shriver National Institute Of Child Health and Human Development or the National Institutes of Health. We thank the Baylor College of Medicine Center for Comparative Medicine for mouse colony management, D. Yu for assistance with microscopy, C. Ljungberg for assistance with RNA ISH, and C.-W. Logan Hsu for assistance with the tissue clearing and lightsheet microscopy. We thank the members of the Zoghbi lab and M. E. Van Der Heijden for discussions and comments on the manuscript. Funding: This project was supported by Howard Hughes Medical Institute and funding from a Shared Instrumentation grant from the NIH (S10 OD016167) and the NIH IDDRC Grant P50 HD103555 from the Eunice Kennedy Shriver National Institute Of Child Health and Human Development for use of the RNA In Situ Hybridization Core and Neurovisualization Core. J.C.B. was supported by NIH F32 (NS117723). R.S.D. was supported by NIH NINDS F32 (NS127854). This work was supported by Howard Hughes Medical Institute (to H.Y.Z.), National Institutes of Health grant S10 OD016167, National Institutes of Health IDDRC P50 HD103555, National Institutes of Health grant NS117723 (to J.C.B.), and National Institutes of Health grant NS127854 (to R.S.D.). Author contributions: Conceptualization: H.Y.Z., S.-R.W., J.C.B., and M.A.D. Methodology: S.-R.W., J.C.B., and M.A.D. Investigation: S.-R.W., J.C.B., and J.-P.R. Software: S.-R.W., M.S.C., R.S.D., and M.A.D. Visualization: S.-R.W. and M.S.C. Writing—original draft: S.-R.W. Writing —review and editing: H.Y.Z., S.-R.W., J.C.B., M.S.C., and R.S.D. Funding acquisition: H.Y.Z. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data are available in the main text and/or the Supplementary Materials. The raw FASTQ files of the scRNA-seq and the processed files (output from CellRanger) are accessible through GEO (accession number: GSE224031). The code for data analyses is available on figshare (doi: 10.6084/m9.figshare.22490956) and via the link: https:// figshare.com/s/ca568d8f44d020cb6389. Submitted 6 December 2022 Accepted 26 May 2023 Published 30 June 2023 10.1126/sciadv.adg1671 Wu et al., Sci. Adv. 9, eadg1671 (2023) 30 June 2023 15 of 15
10.1126_science.adg6518	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 June 15. Published in final edited form as: Science. 2023 April 21; 380(6642): eadg6518. doi:10.1126/science.adg6518. Base editing rescue of spinal muscular atrophy in cells and in mice Mandana Arbab1,2,3,4,†, Zaneta Matuszek3,4,5,†, Kaitlyn M. Kray6, Ailing Du7, Gregory A. Newby3,4, Anton J. Blatnik6, Aditya Raguram3,4, Michelle F. Richter3,4, Kevin T. Zhao3,4, Jonathan M. Levy3,4, Max W. Shen3,4,8, W. David Arnold9,10, Dan Wang7,11, Jun Xie7, Guangping Gao7,12, Arthur H. M. Burghes6, David R. Liu3,4,13,* 1Department of Neurology, Rosamund Stone Zander Translational Neuroscience Center, Boston Children's Hospital, Boston, MA 02115, USA. 2Department of Neurobiology, Harvard Medical School, Boston, MA 02115, USA. 3Merkin Institute of Transformative Technologies in Healthcare, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA. 4Department of Chemistry and Chemical Biology, Harvard University, Cambridge, MA 02138, USA. 5Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA, 02138, USA. 6Department of Biological Chemistry and Pharmacology, The Ohio State University Wexner Medical Center, 1060 Carmack Road, Columbus, OH 43210, USA. License information: Copyright © 2023 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journal-article- reuse. This article is subject to HHMI’s Open Access to Publications policy. HHMI lab heads have previously granted a nonexclusive CC BY 4.0 license to the public and a sublicensable license to HHMI in their research articles. Pursuant to those licenses, the author- accepted manuscript of this article can be made freely available under a CC BY 4.0 license immediately upon publication.This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding author. [email protected]. †These authors contributed equally to this work Author contributions: Conceptualization: M.A., D.R.L. Methodology: M.A., W.D.A., D.W., A.H.M.B. Software: A.R., M.W.S. Validation: M.A., K.M.K. Formal analysis: M.A., Z.M., G.A.N., A.R., M.W.S. Investigation: M.A., Z.M., K.M.K., A.D., G.A.N., A.J.B. Resources: M.F.R., K.T.Z., J.M.L., J.X., G.G. Writing – original draft: M.A., D.R.L. Visualization: M.A., M.W.S. Supervision: M.A., D.R.L. Project administration: M.A., D.R.L. Funding acquisition: M.A., D.R.L. Competing interests: M.A. and D.R.L. have filed patent applications on this work. D.R.L. is a consultant and equity owner of Beam Therapeutics, Prime Medicine, Pairwise Plants, Chroma Medicine, and Nvelop Therapeutics, companies that use or deliver genome editing or genome engineering agents. A.H.M.B is a consultant for Novartis. WDA has served as a consultant to NMD Pharma, Genentech, Catalyst Pharmaceuticals, Dyne, Avidity Biosciences, Design Therapeutics, Argenx, and Novartis and has received research funding from Novartis, Avidity Biosciences, NMD Pharma, and Biogen. Data and materials availability: The plasmids used in this study are available through AddGene (depositor: David R. Liu). DNA sequencing files can be accessed using the NCBI SRA (SUB# PRJNA871232). All data are available in the main text or the supplementary materials. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 2 7Horae Gene Therapy Center, University of Massachusetts, Medical School, Worcester, MA 01605, USA. 8Computational and Systems Biology Program, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. 9Department of Neurology, The Ohio State University Wexner Medical Center, 1060 Carmack Road, Columbus, OH 43210, USA. 10NextGen Precision Health, University of Missouri, Columbia, MO 65212, USA. 11Horae Gene Therapy Center and RNA Therapeutics Institute, University of Massachusetts, Medical School, Worcester, MA 01605, USA. 12Microbiology and Physiological Systems, University of Massachusetts, Medical School, Worcester, MA 01605, USA. 13Howard Hughes Medical Institute, Harvard University, Cambridge, MA 02138, USA. Abstract Spinal muscular atrophy (SMA), the leading genetic cause of infant mortality, arises from SMN protein insufficiency following SMN1 loss. Approved therapies circumvent endogenous SMN regulation and require repeated dosing or may wane. We describe genome editing of SMN2, an insufficient copy of SMN1 harboring a C6>T mutation, to permanently restore SMN protein levels and rescue SMA phenotypes. We used nucleases or base editors to modify five SMN2 regulatory regions. Base editing converted SMN2 T6>C, restoring SMN protein levels to wild-type. AAV9-mediated base editor delivery in Δ7SMA mice yielded 87% average T6>C conversion, improved motor function, and extended average lifespan, which was enhanced by one-time base editor+nusinersen co-administration (111 versus 17 days untreated). These findings demonstrate the potential of a one-time base editing treatment for SMA. One-Sentence Summary: Base editing enables a one-time treatment for spinal muscular atrophy (SMA) that rescues disease pathology and extends lifespan in mice. SMA is a progressive motor neuron disease and the leading genetic cause of infant mortality(1-3). SMA is caused by homozygous loss or mutation of the essential survival motor neuron 1 (SMN1) gene. One or more copies of the nearly identical (>99.9% sequence identity) SMN2 partially compensates for the loss of SMN1(1, 4, 5). However, SMN1 and SMN2 differ by a silent C•G-to-T•A substitution at nucleotide position 6 of exon 7 (C6T) that results in exon 7 skipping in mRNA transcripts (Fig. 1A)(6, 7). The resulting truncated SMNΔ7 protein is rapidly degraded, causing SMN protein insufficiency that results in loss of motor neurons, paralysis, and death(8-10). Untreated patients with the most common form of SMA (type I) live a median of 6 months(11, 12). Upregulation of SMN protein can rescue motor function and substantially improve the prognosis of SMA patients(13-15). However, endogenous SMN protein is subject to multiple levels of regulation that differs across tissues(16-18). While SMN underexpression can Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 3 fail to rescue SMN phenotypes, SMN overexpression can cause aggregation, toxicity, and tissue pathology(19-21). Three breakthrough therapeutics effectively rescue many SMA phenotypes and improve lifespan by upregulating SMN protein(22). The antisense oligonucleotide (ASO) nusinersen (Spinraza) and the small-molecule risdiplam (Evrysdi) both promote splicing inclusion of exon 7 resulting in ~2-fold upregulation of SMN levels, and have proven highly effective in the clinic(23, 24). However, SMN protein is reduced ~6.5-fold in the spinal cord of untreated SMA patients(25-27). The partial recovery of SMN protein promoted by these therapeutics may be insufficient at early timepoints and in damaged tissues, potentially underlying the limited rescue observed in some patients(28, 29). Moreover, the transient nature of these therapeutics necessitates repeated administration of costly drugs throughout patients’ lifetimes(30, 31). AAV-mediated gene complementation of full-length SMN cDNA by onasemnogene abeparvovec-xioi (Zolgensma) leads to constitutive production of SMN in transduced cells that is not under endogenous control(32-34). In the spinal cord, Zolgensma upregulates SMN transcript levels by ~25%(35), while in other tissues such as the liver and dorsal root ganglia, gene complementation may cause SMN overexpression that under some circumstances can cause long-term toxicity(21). We do not yet know whether SMN overexpression induces toxicity in patients treated with Zolgensma, nor how long AAV- mediated expression will persist in motor neurons in patients (36, 37). As such, a therapeutic modality that restores endogenous gene expression and preserves native SMN regulation by a one-time permanent treatment may address remaining limitations of existing SMA therapies. Genome editing of SMN2, which is present in all SMA patients regardless of the nature of their SMN1 mutation, could enable a one-time treatment for SMA that restores native SMN transcript and protein levels while preserving their endogenous regulatory mechanisms. Results Predictable and precise nuclease editing of SMN2 ISS-N1 increases SMN protein levels SMN protein production from SMN1 and SMN2 genes is constrained by transcriptional, transcriptomic, and post-translational regulatory sequences. We explored using Cas nucleases to create gain-of-function alleles in SMN2 regulatory sequences that upregulate SMN levels. The inclusion of exon 7, which underlies SMN protein stability, is strongly influenced by the downstream intronic splicing silencer ISS-N1 that harbors two heterogeneous nuclear ribonucleoprotein (hnRNPs) A1/A2 binding sites (Fig. 1A)(38). Deletions within, and downstream of the 3’ (3-prime) hnRNP A1/A2 binding domain improve exon 7 inclusion(38-41). We speculated that Cas9 nuclease-mediated disruption of the ISS-N1 genomic locus might increase exon 7 inclusion in SMN2 splicing and thereby increase SMN protein levels (strategy A, Fig. 1B). We used inDelphi, a machine learning model of SpCas9 nuclease editing outcomes, to predict indel outcomes at the ISS-N1 locus that disrupt hnRNP A1/A2 binding and improve full-length SMN splicing of SMN2 (Fig. 1B)(42). InDelphi identified ten spacer sequences predicted to induce ≥4-nt deletions at ISS-N1 and loss of ≥1-nt of the 3’-hnRNP A1/A2 domain (‘predicted % precision’). We estimated editing efficiencies of these strategies based Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 4 on the reported PAM compatibility of these spacer sequences with SpCas9-variant nucleases (‘predicted % PAM efficiency’)(43-46). From 19 possible nuclease editing strategies (A1- A19, defined as different combinations of genome editing agents and guide RNAs) we selected nine (A2, A3, A5, A6, A13, A14, A16, A17, A19) for experimental testing. We co-transfected Δ7SMA mESCs—which lack endogenous Smn1, are homozygous for the full-length human SMN2 gene, carry human SMNΔ7 -cDNA transgenes, and harbor a Mnx1:GFP reporter of motor neurons (SMN2+/+; SMNΔ7; Smn −/−; Mnx1:GFP)(47)— with nuclease expression plasmids that carry a blasticidin-resistance cassette and sgRNA plasmids that carry a hygromycin-resistance cassette. Both plasmids also contain Tol2 transposase sequences to enable stable transposon-mediated genomic integration and antibiotic selection. We achieved 92±5.6% average indel frequencies for the top four strategies targeting the ISS-N1 locus (A2, A3, A5, A6, Fig. 1B). To assess whether nuclease-mediated editing of ISS-N1 improved exon 7 inclusion, we performed reverse-transcription PCR (RT-PCR) of SMN2 from exons 6 to 8, and quantified SMNΔ7 and full-length SMN products (Fig. 1C). We found that all strategies that edited ISS-N1 with high efficiency (≥85%) resulted in a significant increase in exon 7 inclusion averaging 2.2-fold relative to an unrelated sgRNA control (Welch’s two-tailed t-test p=0.01). The increase in exon 7 inclusion caused a substantial increase in SMN protein of 17-fold by A2 and 13-fold by A6 relative to untreated controls (values normalized to histone H3, Welch’s two-tailed t-test p=0.02, Fig. 1D, and fig. S1A). Collectively, these results demonstrate that disruption of the ISS-N1 genomic locus can stably increase full-length SMN splicing and protein phenotypes of SMA. Predictable and precise genome editing of SMN2 exon 8 increases SMN protein levels As an alternative nuclease-mediated approach to upregulate SMN levels, we disrupted post- translational regulatory sequences in SMN2 to increase SMNΔ7 protein stability. The critical difference between full-length SMN and the unstable SMNΔ7 protein is the substitution of 16 amino acids encoded by exon 7 with EMLA, a four-residue degron encoded by exon 8 (Fig. 1A)(8). Extending the coding sequence of exon 8 with five or more heterologous amino acids obscures SMNΔ7 C-terminal degradation signals. These modified SMNΔ7 (SMNΔ7mod) protein variants have increased stability and rescue survival and motor phenotypes of severe SMA mice(48). We designed strategies for Cas nuclease-mediated disruption of exon 8 to generate similar stabilized SMNΔ7mod proteins with therapeutic potential (strategy B1-B16, Supplementary Text, Fig. 1E), and observed up to 7.0-fold increase in SMN protein levels by B11 (Welch’s two-tailed t-test p=0.007, Fig. 1F, and fig. S1B). Some exon 8 editing strategies improved SMN protein stability more than expected based on observed edited genotypes (Fig. 1, E and F). For example, precision edited genotypes were 1.9-fold higher in frequency following B9 editing than B1, yet SMNΔ7mod protein levels were greater in cells edited with B1 (9.1-fold) than B9 (5.7-fold). These data suggest that additional edited genotypes may improve SMN protein stability. Inspection of the non-precisely edited fraction of edited alleles revealed that B1 editing frequently induces Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 5 indels at the exon 8 splice acceptor. Thus, we hypothesized that disrupting splicing of exon 8 improves SMN protein stability(49). To test this hypothesis, we disrupted the canonical AG splice acceptor (SA) motif of exon 8 using either a nuclease or cytosine base editor (C-nuc or C-CBE) in Δ7SMA mESCs (Fig. 1G)(45, 50), and observed 54±2.3% indels from C-nuc and 89±2.3% cytosine base editing from C-CBE. Notably, C-nuc editing resulted in a complex mixture of indel genotypes at the intron-exon junction that resulted in deletion of additional nucleotides beyond the AG motif. Both strategies significantly increased SMN levels in Δ7SMA mESCs, similar to treatment with risdiplam (3.3-fold for C-nuc, 9.5-fold for C-CBE, 9.1-fold for risdiplam relative to untreated, Welch’s two-tailed t-test p<0.05, Fig. 1H, and fig. S1D to G), indicating that alternative splicing at exon 8 improves the stability of SMN2 gene products. We investigated how exon 8 SA disruption affects SMN2 transcripts (Supplementary Text). C-CBE editing induced a minor increase in SMN2 mRNA that only partially explains the 9.5-fold increase in SMN levels (fig. S1H). We also observed a profound shift in SMN2 splice products (Fig. 1I). We investigated whether these alternative splice isoforms improve stability of SMN proteins, and found that transcripts including exon 7 were increased 2-fold by C-CBE (63±2.0%) and 1.6-fold by C-nuc (50±1.1%) relative to untreated cells (24±1.4%). These transcripts often retain intron 7 as in some functional transcript variants of SMN2 (ENST00000511812.5, fig. S1I). Importantly, all transcripts that include exon 7 encode full-length SMN protein and can therefore complement loss of SMN1, Thus, the substantial increase in SMN protein levels following exon 8 SA editing predominantly arises from an increase in normal full-length SMN. Collectively, the tested SMN2 editing strategies permanently increase SMN protein levels up to 17-fold (strategy A2), 9.1-fold (strategy B1), and 9.5-fold (strategy C-CBE). As a 1.5- to 2-fold increase in SMN protein is therapeutic for SMA patients(23, 24), these strategies represent promising approaches for further studies. Efficient and precise base editing of SMN2 splice regulatory elements Several single-nucleotide substitutions in exon 7 strongly regulate splicing of SMN2, including the C-to-T transition at position 6 (C6T) that differentiates SMN1 (C) from SMN2 (T) genes (Fig. 1A), and T44C, G52A, and A54G at the 3’-end of exon 7(51). Using existing and newly developed BE-Hive predictive models of base-editing outcomes (Supplementary Text, fig. S2, A to E), we identified 42 strategies (combinations of base editors and guide RNAs) to modify exon 7 splicing regulatory elements (SREs, Fig. 2, A to C and fig. S2, F and G). We designed 13 spacers targeting C6T using ABE8e (strategy D1-19), or targeting C6T, T44C, G52A, and A54G using ABE8e, ABE7.10 and EA-BE4 deaminases (strategy E1-23). We paired these spacers with 12 compatible SpCas9-variants based on reported PAM preferences (‘predicted % PAM efficiency’)(43, 50, 52). We validated these strategies in Δ7SMA mESCs, and found that the BE-Hive models of SpCas9 base editors predicted edited outcomes of Cas-variant base editors with high accuracy (Cas9-NG(52), NRTH, NRRH, NRCH(44) Pearson’s r=0.810, chimeric SpyMac and iSpMac(45) Pearson’s r=0.910, Supplementary Text, Fig. 2D). Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 6 Base editing of exon 7 SREs was highly efficient. At 3’-SREs, we achieved 69±5.0% T44C editing by E14, 92±4.0% G52A editing by E20, and 95±5.1% A54G editing by E23 (fig. S2, F and G). We achieved nearly complete (94%–99.5%) C6T A•T-to-G•C conversion by strategies targeting C6T at positions P5 (D1, D2), P8 (D10, D11), and P10 (D18, D19) within the protospacer (Fig. 2, A to C). The deaminase in ABE7.10 enabled up to 64±2.5% conversion of T6>C (E7, fig. S2G)(53, 54). The frequency of edited alleles with single-nucleotide T6>C conversion alone (i.e., without any bystander edits or indels) varied substantially between the most efficient C6T editing strategies, ranging from 82±1.9% from D10 to 40±13% from D19 editing (Fig. 2E). Prior studies suggest that the coding sequence at the SMN C-terminus beyond exon 6 does not strongly affect SMN protein function and it is therefore unlikely that single-nucleotide editing precision of C6T is imperative for rescue of SMA(8, 48, 55). Maximizing the sequence similarity of modified SMN2 genes to native SMN1, however, may preserve additional regulatory interactions, including those not yet known. D10, the strategy with the highest precision and efficiency (99±0.7%), did not induce measurable indels and its induced bystander missense nucleotide substitutions (18±2.4%) have previously been shown to benefit inclusion of exon 7 by improved protein binding at the exonic splicing enhancer (fig. S2H)(56, 57). Together, these results establish efficient base editing strategies to convert SMN2 T6>C with high-fidelity and few undesired byproducts. Base editing of SMN2 splice regulatory elements rescues SMN protein levels Next, we asked whether base editing of exon 7 SREs results in functional rescue of cellular SMA phenotypes. The top six ABE8e editing strategies that converted C6T in >97% of alleles increased exon 7 inclusion to 78±10.2% on average, up to 9.7-fold higher than untreated cells (87±1.5% by D10 compared to 9.0±6.6% in untreated, Welch’s two-tailed t-test p<0.002, Fig. 2F). These results are on par with, or exceed, maximum exon 7 inclusion by risdiplam or nusinersen treatment of Δ7SMA mESCs (89±4.3% and 80±0.3%, respectively, Fig. 2F and fig. S1E), and resemble splicing ratios of SMN1 genes (82±7.3% in U2OS cells)(38, 39). Base editing of 3’-SREs in exon 7 also improved inclusion, averaging 60±3.2% following T44C editing by E14, 76±12% following G52A editing by E20, and 50±8.6% following A54G editing by E23 (fig. S2I). These data demonstrate that base editing of various exon 7 SREs can increase full-length SMN splice products. Base editing of 3’-SREs increased SMN protein levels in ways that did not closely mirror observed improvements in exon 7 inclusion. We detected a 3.4-fold increase in SMN protein by E14 base editing of T44C, 23-fold increase by E20 editing of G52A, and 1.6-fold increase by E23 editing of A54G (Welch’s two-tailed t-test p=0.02), despite all three edits inducing comparable improvements in exon 7 inclusion (figs. S2I, and S3, A and B). We hypothesized that unintended bystander edits may underlie this persistent protein instability and found that the T44C and A54G editing strategies frequently ablate the nearby TAA stop codon in exon 7 (fig. S2, F and G). A failure to terminate translation in exon 7 leads to the extension of full-length SMN proteins with the EMLA degron encoded by exon 8 (Fig. 1A). Thus, imprecise editing of T44C or A54G by E14 or E23 results in the translation of unstable full-length SMN-EMLA fusions that prevent upregulation of SMN protein levels. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 7 Editing of G52A by E20 uses the EA-BE4 cytosine deaminase that does not recognize TAA as a substrate and therefore does not induce non-silent bystander changes in 99±0.1% of edited alleles, resulting in a 23-fold improvement in SMN protein levels. Base editing of exon 7 C6T resulted in the greatest upregulation of SMN protein. The top six ABE8e editing strategies that correct C6T in >97% of alleles induced a 41-fold average increase in SMN protein levels compared to untreated controls (normalized to H3, Welch’s two-tailed t-test p<0.0002, Fig. 2G and fig. S3C), indicating complete rescue of normal SMN protein levels in Δ7SMA mESCs, which are ~40-fold reduced relative to wild-type mESCs(47). Despite inducing comparable increase in exon 7 inclusion, base editing of C6T enabled a 4.5-fold and 1.5-fold greater increase in SMN protein levels than risdiplam and nusinersen treatment of Δ7SMA mESCs (9-fold and 17-fold respectively, compared to 41-fold on average across the top six strategy D approaches, Fig. 1H, 2G, and figs. S1, D and F to G, and S3, C and E to F). Normal levels of SMN protein are essential to the function, survival, and long-term health of all species in the animal kingdom(58-61). Restoring wild-type levels of SMN protein as achieved by base editing strategy may thus best maximize the long-term health of SMA patients. Among all genome editing strategies tested, base editing of C6T by D10 induces the greatest increase in exon 7 inclusion (87±1.5%) and best recapitulates native SMN protein levels (95% of wild-type levels, a 38-fold increased versus untreated Δ7SMA mESCs). D10 base editing is highly efficient (99±0.7%) with high on-target precision (82±0.0%). The SMN2 gene arose from a duplication of the chromosomal region containing SMN1, and shares an identical promoter and >99.9% sequence identity with SMN1, including 100% DNA conservation of its protein-coding sequence other than exon 7 C6T(1, 4, 5). We performed RT-qPCR and quantified SMN2 mRNA levels in edited cells, confirming that SMN2 mRNA abundance is not affected by D10 base editing compared to untreated Δ7SMA mESCs or following ABE8e transfection with an unrelated sgRNA (fig. S3G). Together, these data indicate that D10 editing of SMN2 faithfully reproduces the genomic sequence and function of native SMN1 alleles. Therefore, we selected strategy D10 for further study. Off-target editing analysis of ABE8e targeting SMN2 C6T in the human genome Some base editors can induce off-target deamination in cells, including Cas-dependent off- target DNA editing and Cas-independent off-target DNA or RNA editing(62-66). Genomic and transcriptomic off-target deamination by adenine base editors without involvement of the Cas protein component is rare, and deaminase variants that minimize these events have been reported(62, 67). We assessed the Cas-dependent genome specificity of the D10 strategy (ABE8e-SpyMac and P8 sgRNA) characterizing SpyMac Cas9 nuclease with P8 sgRNA using CIRCLE-seq(68), an unbiased and sensitive empirical in vitro off- target detection method. Potential off-target sites nominated by CIRCLE-seq can then be sequenced in-depth in base-edited human cells to provide a sensitive genome-wide analysis of off-target genome editing events induced by the D10 strategy(68, 69). We generated purified D10 strategy ribonucleoprotein (RNP) complexes containing SpyMac nuclease and P8 sgRNA to treat human genomic DNA from HEK293T cells in vitro and analyzed rare off-target genomic cleavage events (fig. S3H). We identified 55 candidate Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 8 SpyMac-dependent DNA off-target loci nominated by the CIRCLE-seq method. Next, we measured on-target and genomic off-target editing at the top 23 CIRCLE-seq-nominated loci in human cells (Supplementary Text, Fig. 2H and fig. S3I). We achieved 49±1.8% C6T on-target base editing at SMN2 in HEK293T cells and observed minimal base editing at SMN1 (0.15±0.07%), which is generally absent in SMA patients. We detected minor levels of D10 base editing at off-target site ranked 19 (0.41±0.14%), which is in an intergenic region of chromosome 15, and no evident base editing (≤0.03% over untreated cells) at the other 21 assayed potential off-target loci. These data indicate high genomic target specificity of the D10 base editing strategy for the on-target locus. Together, these experiments did not detect any coding mutations or sequence changes of anticipated physiological significance in the human genome and support continued preclinical evaluation of the D10 strategy, including assessment of base editor off-target editing measured in various tissues that may accumulate over an extended period of time. We refer to the D10 editing strategy as the ‘ABE strategy’ hereafter. Viral delivery of ABE enables efficient in vivo conversion of SMN2 C6T To enable in vivo SMN2 C6T conversion in an animal model of SMA, we designed an adeno-associated virus (AAV) strategy to package ABE8e-SpyMac and the P8 sgRNA for delivery (v6 AAV-ABE8e, Supplementary Text, Fig. 3A and fig. S3J). The AAV serotype 9 (AAV9) has a well-established tropism for neurons in the CNS of a wide range of organisms, including Δ7SMA mice and human patients( 70-72). In the cortex, AAV9 has been shown to almost exclusively target neurons(72), and intracerebroventricular (ICV) or systemic injection in neonates results in efficient transduction of spinal motor neurons to enable rescue of SMA disease phenotypes and lethality in both mice and humans(13, 32, 73). Thus, we selected AAV9 for delivery of our D10 ABE strategy (‘AAV9-ABE’) to Δ7SMA neonates by ICV injection to correct the SMN2 C6T target in vivo (Fig. 3B). We ICV injected SMA neonates with total 2.7x1013 vg/kg of the dual AAV9-ABE vectors, along with 2.7x1012 vg/kg AAV9-Cbh-eGFP-KASH (Klarsicht/ANC-1/Syne-1 homology domain, hereafter AAV9-GFP)(74) to serve as a viral transduction control. This dose is comparable to doses used for P0 ICV AAV administration of Zolgensma for rescue of Δ7SMA mice, and of other base editor AAVs that enable efficient genome editing in mice(32, 74). We observed typical transduction patterns of AAV9 in the spinal cord (Fig. 3, C to E, Supplementary Text, fig. S4A)(32, 33, 75). We quantified GFP and choline acetyl- transferase (ChAT) double-positive cells in the ventral horn of spinal cords from injected mice and observed a mean transduction efficiency of 43% in spinal motor neurons (Fig. 3F), consistent with transduction efficiencies >20% previously shown to enable significant phenotypic rescue of Δ7SMA mice following ICV injection of self-complementary AAV9- SMN (Zolgensma)(32). Transduction of spinal motor neurons using 2.97x1013 vg/kg AAV9- GFP alone was similar (median 46%) to transduction efficiencies using the ten-fold lower concentration of 2.7x1012 vg/kg, suggesting that the low-dose co-transduction of AAV9- GFP accurately represents the subset of cells transduced by AAV9-ABE. Next, we assessed base editing in transduced cells (Supplementary Text, fig. S4B). We isolated cortical nuclei of treated animals and enriched for AAV9-transduction by sorting Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 9 GFP-positive cells as previously described(74, 76). We observed 87±3.5% conversion of SMN2 C6T among transduced cells (Fig. 3G), a 2.4-fold enrichment over unsorted tissue (37%±4.7%), with high single-nucleotide precision for C6T alone (73±.2.7%) and few indels (<0.4±.1%) or bystander edits, similar to D10 editing in Δ7SMA mESCs (Fig. 2E, fig. S2H, S4C). Collectively, these data confirm that ICV injection of AAV9-ABE in Δ7SMA neonates enables efficient and precise conversion of SMN2 C6T in the CNS of treated animals with minimal undesired byproducts(56, 57, 77). Base editing conversion of C6T effectively converts the native SMN2 gene to SMN1, thereby restoring SMN protein levels to that of wild-type cells. Current SMA drugs induce non-native SMN levels(23, 24, 32-35), and require repeated dosing or may fade over time. The permanent and precise editing of endogenous SMN2 genes that preserves native transcript levels and native regulatory mechanisms governing SMN expression thus may address shortcomings of existing SMA therapies(1, 4, 5, 21, 28, 78). In vitro and in vivo DNA and RNA off-target analysis of ABE8e targeting SMN2 C6T In addition to the off-target analysis in human cells described above, we also assessed the DNA and RNA specificity of the ABE strategy in mouse cells in vitro and in vivo. We performed CIRCLE-seq and validated the top 35 nominated sites in Δ7SMA mESCs (Supplementary Text, fig. S4D). We achieved 95±0.0% on-target editing at the SMN2 transgene and only observed substantial editing at off-target site 5 in an intron of the mucin 16 gene (Muc16, 31±1.9%) that is not expressed in the CNS (fig. S4E)(79). Next, we compared this analysis to off-target editing in vivo following AAV9-ABE ICV injection in Δ7SMA neonates by performing verification of in vivo off targets (VIVO)(80). We observed between 10-27% (average 15±7%) editing at off-target site 5 in intron 54 of Muc16, and between 0.1-0.9% (average 0.5±0.3%) editing at the non-coding off-target site rank 15, compared to 87±3.5% average on-target editing of SMN2 among GFP-positive cells in the CNS across five animals (Fig. 3G and H). These animals ranged from 4 to 18 weeks of age at the time of off-target analysis (26, 36, 42, 80, and 127 days old) and we observe no increase in off-target editing events over time. Thus, off-target editing outcomes observed in cell culture experiments were consistent with those observed in vivo over 18 weeks(80). The ABE strategy did not induce any detected coding mutations in either human or mouse genomes, and off-target editing in vivo was lower than in cell culture (~2-fold lower at Muc16 intron 54), likely due to lower copy number and expression levels in transduced cells in vivo or in vivo gene silencing over time(33, 36, 37). Cas-independent RNA off-target adenine base editing in vivo is typically indistinguishable from background A-to-I conversion due to the low copy-number of ABE-expressing transgenes(33, 81). We investigated RNA off-target editing in Δ7SMA mESCs and differentiated neural lineages including motor neurons, that stably produce ABE8e from low gene copy numbers similar to those resulting from AAV9 transduction (Fig. 3I, Supplementary Text, fig. S4, F to H). Consistent with previous reports(81, 82), whole transcriptome sequencing did not reveal detected accumulation of RNA A-to-I edits over background levels of endogenous A-to-I and A-to-G changes (Fig. 3J, and fig. S4G). Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 10 Collectively, these in vitro and in vivo analyses did not reveal off-target edits of anticipated clinical or physiological significance in human or mouse cells, suggesting high target specificity of the D10 base editing approach. Continued preclinical assessment and minimization of off-target editing is important to ensure the safety of a potential base editing therapeutic for the treatment of SMA in patients. ABE-mediated rescue of SMA pathophysiology in mice The physiology of AAV9-ABE treated Δ7SMA mice was improved compared to untreated animals (Movies S1 and S2). We assessed the rescue of motor phenotypes by electrophysiological measurements in AAV9-ABE treated Δ7SMA mice. We measured compound muscle action potential (CMAP) amplitude and performed motor unit number estimation (MUNE) in the gastrocnemius muscle to assess loss of motor neuron functional integrity, a key feature of SMA and preclinical SMA models(83). We compared outcomes with FDA-approved therapeutics for SMA including ICV injection of Zolgensma, and daily intraperitoneal (IP) injection of risdiplam (Evrysdi) at doses that were previously demonstrated to confer a survival benefit to these mice (2.5x1013 vg/kg Zolgensma and 0.1 mg/kg risdiplam, Fig. 4A)(30, 32). MUNE were reduced by 50% in untreated Δ7SMA animals compared to heterozygous mice at postnatal day (PND) 12, and Zolgensma or 0.1 mg/kg risdiplam showed little to no improvement (50% and 75% relative to heterozygotes respectively, Kruskal-Wallis test p>0.6). In contrast, MUNE in AAV9-ABE treated SMA mice were significantly improved compared to untreated animals (Kruskal-Wallis test p<0.02) and did not significantly differ from heterozygous animals, with values averaging 91% that of heterozygotes. CMAP amplitudes were also higher for AAV9-ABE-treated mice compared to risdiplam-treated or untreated Δ7SMA mice, while CMAP amplitudes did not significantly differ between heterozygotes, Zolgensma-treated mice, and AAV9-ABE-treated animals (Kruskal-Wallis one-way ANOVA p>0.2). Thus, neonatal ICV injection of AAV9- ABE measurably rescues SMA pathophysiology of spinal motor neurons. Next, we assessed survival of ICV AAV9-ABE injected Δ7SMA mice. In SMA type I patients, therapeutic intervention can meaningfully improve disease outcomes if administered in the first several months of life(84-87), however, in Δ7SMA mice survival drops precipitously when animals receive treatment past PND6 (Fig. 4B)(88). This large difference is due in part to the highly accelerated (~150-fold greater) rate of maturation of mice compared to humans in the first month, early perinatal reduction in SMN expression that occurs in mice(89) and humans(28), and the rapid early-onset loss of motor units, which consist of spinal motor neurons and the muscle fibers that they innervate(83, 90). Restoration of SMN protein levels using inducible transgenes demonstrates that high levels of SMN are required by PND4-6 to rescue Δ7SMA mice, and delays of small numbers of days are strongly anti-correlated with survival(32, 88, 89, 91-93). In cells, complete mRNA rescue is not achieved until 7 days post D10 transfection (fig. S5A), and the time to restore SMN protein levels in vivo surpasses the extremely short therapeutic window in Δ7SMA mice. The accumulation of SMN protein following transduction with the dual single-stranded AAV9 ABE8e vectors used in this study requires completion of (1) second-strand synthesis Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 11 of each AAV9-ABE genome(94, 95), (2) transcription and translation of the split-intein ABE protein segments, (3) assembly and trans-splicing of the split ABE protein, (4) RNP assembly and base editing of SMN2, (5) transcription of full-length C6T-modified endogenous SMN2 pre-mRNA driven by its native promoter, and (6) splicing and translation of corrected SMN2 transcripts. Thus, the timing for SMN protein rescue following AAV9- ABE administration is slower than fast-acting splice-switching drugs or constitutive gene complementation from SMN cDNA encoded by a self-complementary AAV9-SMN vector such as Zolgensma(94-96). We recently demonstrated that in vivo base editing impacts protein levels by ~1-3 weeks post-administration(81). Despite the incongruent timeline of base editing-mediated rescue for ideal rescue of Δ7SMA mice, AAV9-ABE increased the lifespan of treated animals by ~33% in two colonies in different institutions (Supplementary Text, Materials and Methods, Fig. 4C, and fig. S5, B to D). Lifespan of treated animals improved from an average of 17 days (median 17 days, maximum 20 days) to 23 days (median 22 days, maximum 33 days, Mantel-Cox test p<0.02,). As anticipated, the lifespan extension resulting from AAV9-ABE treatment is similar to that achieved by scAAV9-SMN gene therapy in post-symptomatic (>PND7) Δ7SMA mice (Fig. 4B)( 32, 73, 88, 93). Collectively, these data demonstrate that postnatal conversion of SMN2 C6T by AAV9-ABE rescues SMA motor phenotypes in mice, including the number (MUNE) and output (CMAP) of functional motor units innervating muscle, and that the prolonged process of AAV9-ABE-mediated SMN restoration results in mostly post-symptomatic rescue in Δ7SMA mice that results in a significant, but limited improvement in animal lifespan. Upregulation of SMN protein levels improves motor function and life expectancy of SMA patients and animal models if achieved prior to onset of neuromuscular pathology and symptoms(13, 32, 86-88, 93), yet even high levels of SMN protein cannot correct neuromuscular junction defects once SMA has progressed to an advanced stage and loss of motor neurons upon cell death is irreversible. We therefore sought to extend the effective therapeutic window for gene editing by transient early administration of an existing approved SMA drug to attenuate disease progression, as has previously been applied to study milder forms of SMA in mice(73, 97, 98). Since SMA patients in a gene editing clinical trial would likely be receiving an SMA drug, repeating the base editing treatment in mice receiving an existing SMA drug would also inform a potential future clinical application of this approach. Combination therapy improves the lifespan of ABE-treated SMA mice Transient SMA drug administration can ameliorate SMA pathology and extend survival of Δ7SMA mice. We hypothesized that attenuating disease progression using nusinersen could extend the unusually short therapeutic window of Δ7SMA mice and allow AAV9-ABE- mediated rescue to begin before extensive irreversible SMA damage occurs. The mechanism of nusinersen (binding to SMN2 pre-mRNA) is orthogonal to base editing of SMN2 genes, and co-transfection of 20 nM nusinersen did not affect base editing outcomes or inclusion of exon 7 in spliced SMN transcripts following D10 in Δ7SMA mESCs (fig. S5, A and E). We assessed whether co-administration of nusinersen can improve phenotypic rescue from Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 12 AAV9-ABE treatment. A single ICV injection of nusinersen at PND0 has been shown to extend survival of Δ7SMA mice by several weeks( 99), thus we co-injected a single low dose (1 μg) of nusinersen together with AAV9-ABE and AAV9-GFP in Δ7SMA neonates (Supplementary Text). As a control, we also treated Δ7SMA neonates with 1 μg nusinersen and AAV9-GFP but no base editor (Fig. 4D). We assessed motor coordination and overall muscle strength at PND7 using the righting reflex test, which measures the time needed for a mouse placed on its back to right itself (Fig. 4E). We observed significant difference between heterozygotes and nusinersen-treated or untreated Δ7SMA mice (Kruskal-Wallis test p≤0.01), but no significant difference between mice treated with combined AAV9- ABE+nusinersen compared to heterozygous littermates (Kruskal-Wallis test p>0.1). Next, we assessed motor strength and coordination of treated and heterozygous mice using an inverted screen test, which measures how long the mice can hang inverted from a screen mesh surface. At PND25, Δ7SMA animals treated with nusinersen alone performed significantly worse than healthy heterozygous mice at inverted screen testing (Kruskal- Wallis test p=0.007, Fig. 4E). In contrast, the AAV9-ABE+nusinersen combination-treated animals showed no significant difference in the inverted screen assay from healthy heterozygous mice. Notably, half of nusinersen-only treated animals were deceased by this timepoint, and age-matched untreated Δ7SMA mice do not survive long enough for this PND25 assay. For a more complete behavioral assessment of treated and heterozygous animals, we performed extensive multiparametric analysis of voluntary movement by open field tracking at PND40 (Fig. 4F, and fig. S5, F to J). Across 33 parameters including traveled distances, velocity, duration, and counts of various activities, the measured behaviors of AAV9-ABE+nusinersen combination-treated animals showed no significant difference with those of heterozygous mice (Mann-Whitney test p>0.5). Neither nusinersen-only treated or untreated age-matched Δ7SMA mice were available as reference for this PND40 assay due to their short lifespan. We also assessed the effect of combination AAV9-ABE and nusinersen treatment on weight and lifespan of Δ7SMA mice. The weight of nusinersen-only and AAV9-ABE+nusinersen combination-treated Δ7SMA mice steadily increased and were indistinguishable for the first week of life, after which weight gain slowed in the nusinersen-only cohort (Fig. 4G). Combination-treated animals maintained on average 61±4.0% the weight of heterozygous animals throughout their lifespans. The nusinersen-only injection improved lifespan of Δ7SMA mice from an average of 17 days (median 17, maximum 20 days, Fig. 4C) to an average 28 days (median 29, maximum 37 days, Mantel-Cox test p=0.0001, Fig. 4H). Importantly, combination treatment of AAV9-ABE with nusinersen improved survival of Δ7SMA mice to on average of 111 days (median 77, Mantel-Cox test p=0.002), with over 60% of animals surviving beyond nusinersen-only controls, and a 10-fold increase in maximum lifespan (37 days maximum with nusinersen only, compared to 360 days maximum with AAV9-ABE). Combination AAV9-ABE+nusinersen-treated SMA mice also exhibited normal behavior and vitality well beyond the lifespan of nusinersen only-injected controls (P40, P96, and P200 in Movies S3 to S5). Collectively, these data indicate that Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 13 transient extension of the very narrow therapeutic window in Δ7SMA mice can greatly improve phenotypic rescue of SMA from base editing of SMN2. While neonatal AAV9-ABE ICV injection alone enables life extension in Δ7SMA mice that resembles >PND7 ICV injection with Zolgensma (Fig. 4B and C)(88), co-administration of 1 μg nusinersen temporarily slows disease progression and broadens the narrow therapeutic window, allowing base editing the opportunity to enable lifespan rescue that more closely resembles that of pre-symptomatic Zolgensma administration at ≤PND3 (Fig. 4H). Moreover, these data demonstrate compatibility of AAV9-ABE with nusinersen as a one-time treatment without evident adverse effects, and with apparent synergy to improve therapeutic outcomes. Such a combination therapy approach may play an important role in future clinical trial designs for one-time SMA treatments that permanently correct a genetic cause of the disease, and for clinical application in patients already receiving treatment. Discussion Current treatment options for SMA have revolutionized care for thousands of patients, effectively extending lifespan, preventing the loss of motor function in pre-symptomatic patients, and delaying progression in symptomatic patients by increasing full-length SMN protein levels(13, 24, 86, 87, 91, 100). However, current therapies do not restore endogenous protein levels and native regulation of SMN, which could result in pathogenic SMN insufficiency in motor neurons or potential long-term toxicity in other tissues(21, 23-28, 35). Furthermore, the transient therapies nusinersen and risdiplam require repeated dosing throughout a patient’s lifetime, and it is unclear whether Zolgensma gene complementation will persist in motor neurons(36, 37). Thus, achieving permanent and endogenously regulated rescue of SMN protein levels is an important goal of a future therapeutic for SMA patients. The optimized D10 ABE strategy developed in this work is a one-time treatment that enables permanent and precise editing of endogenous SMN2 genes while preserving native transcript levels and regulatory mechanisms that govern SMN expression(1, 4, 5, 28, 78, 101). As such, a future base editing therapeutic approach could offer substantial benefits over existing SMA therapies. We compared 79 total nuclease and base editing strategies targeting five regions of SMN2 to induce either post-transcriptional or post-translational regulatory changes in SMN2 that upregulate SMN protein production. BE-Hive and inDelphi machine learning models enabled the design of precise editing strategies that in some cases were not obvious, and pre-selected sgRNAs for genotypic and phenotypic validation of editing outcomes. All SMA patients regardless of their SMN1 mutations must carry the SMN2 gene to complete gestation(7), and thus the genome editing strategies identified in this study have the potential to benefit all SMA patients. While on-target Cas nuclease editing at SMN2 can be precise, DSBs can result in large deletions and chromosomal rearrangements, especially when induced simultaneously at multiple genomic loci(102). Given that SMA patients usually have multiple copies of SMN2, nuclease editing may result in unintended restructuring of the chromosome region (5q13) that harbors SMN genes(103, 104). In contrast, base editors precisely convert Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 14 nucleotides without inducing DSBs(50, 105, 106), and result in greater SMN protein upregulation than the nuclease strategies in this study (up to 50-fold by base editors compared to up to 17-fold by nucleases). We therefore recommend that future gene editing therapeutic strategies for SMA use base editing rather than nucleases. ABE strategy D10 demonstrated high on-target efficiency and specificity, with minimal Cas-dependent or Cas-independent off-target DNA or RNA editing. It is possible that extended base editor expression in cells, as can result from AAV-delivery, could result in a greater accumulation of genomic and transcriptomic off-target events. Therefore, a deeper assessment of genomic and transcriptomic off-targets and efforts to minimize off-target editing risk will be important in the preclinical development of a potential base editing therapeutic for SMA. If needed, Cas-independent editing events can be further minimized by alternative delivery strategies that shorten exposure to base editors(62), and by the use of tailored deaminases such as the V106W variant of TadA-8e(62, 64) or TadA-8.17-m(107). SMA has variable presentation in humans that largely correlates with the copy number of SMN2(108-112). Type I SMA patients have two SMN2 copies and present with symptoms within the first 6 months, type II patients have three copies and present with symptoms by 18 months, while type III patients have 3-4 SMN2 copies with later onset. Early intervention is paramount to achieving the best outcomes for SMA patients. The window to effectively treat type II and III patients is broader than for type I patients, who ideally receive treatment within the first few months of life and up to 18 months(13, 24, 84-87, 100). Indeed, we directly observed the critical role of differences in timing on the order of days in determining the efficacy of an AAV9-ABE treatment in Δ7SMA mice, which have an unusually short (≤6 days) therapeutic window compared to the timescale of base editing (weeks)(88). We show that the FDA-approved ASO drug nusinersen can extend the very short therapeutic window for rescue in Δ7SMA mice, allowing base editing-mediated rescue of SMN protein levels to occur to a greater extent(81). We anticipate that the broader therapeutic window in human SMA patients would provide ample opportunity for AAV9-ABE-mediated restoration of SMN protein levels to take place without the need for co-administration of a transient therapeutic. Nevertheless, our study demonstrates the compatibility of base editing with nusinersen as a combination therapy approach to treat SMA in animals, which may be valuable for future clinical applications. The ICV-injected AAV9-ABE animals in our study exhibited mouse-specific peripheral disease phenotypes that are common in SMA mouse models including necrosis of the extremities(113, 114), while exhibiting otherwise normal behavior and vitality without displays of progressive muscle weakness. However, SMA treatment that is restricted to the CNS also reveals a late onset lethal cardiac abnormality specific to Δ7SMA mice( 32, 115-117), and likely underlies the sudden late-stage fatality observed in ICV AAV9-ABE treated animals in this study. Treating both CNS and peripheral tissues may ameliorate this murine cardiac phenotype to improve lifespan of treated Δ7SMA mice compared to ICV-injected animals(115, 118). Nevertheless, peripheral restoration of SMN protein does not appear to be required to rescue SMA lethality in humans in light of patients successfully treated intrathecally with Spinraza(23, 31, 84, 91, 119). Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 15 As demonstrated in this work, dual-AAV delivery of base editors supports therapeutic levels of editing in mouse models of human disease(120, 121). After these in vivo experiments were completed, our lab developed efficient in vivo base editing using single-AAV9-ABE systems that use size-minimized AAV vector components and one of a suite of small Cas protein domains that are highly active as ABEs(81). Such single-AAV base-editing systems may simplify the development of future base editor therapeutics, and potentially minimize the required dose and potential side effects of AAV in clinical settings(122). Materials and Methods Cell culture Culture of mESCs, HEK293T, and U2OS cells was performed according to previously published protocols(123). mESCs were maintained on 0.2% gelatin-coated plates feeder-free in mESC media composed of Knockout DMEM (Life Technologies) supplemented with 15% defined fetal bovine serum (FBS, HyClone), 0.1 mM nonessential amino acids (NEAA, Life Technologies), Glutamax (GM, Life Technologies), 0.55 mM 2-mercaptoethanol (b- ME, Sigma-Aldrich), 1X ESGRO LIF (Millipore), with the addition of 2i: 5 nM GSK-3 inhibitor XV (Sigma-Aldrich), and 500 nM UO126 (Sigma-Aldrich). Δ7SMA mESCs were a kind gift from Lee L. Rubin. HEK293T cells were purchased from ATCC (CRL-3216) and were maintained in DMEM (Life Technologies) supplemented with 10% fetal bovine serum (ThermoFisher Scientific). U2OS cells were purchased from ATCC (HTB-96) and were maintained in McCoy's 5a medium (Life Technologies) supplemented with 10% fetal bovine serum (ThermoFisher Scientific). All cells were regularly tested for mycoplasma. For genome editing experiments, cells were seeded one day prior to be ~70-80% confluent on the day of transfection and transfected with sgRNA and genome editing plasmids at a 1:1 molar ratio using Lipofectamine 3000 (ThermoFisher Scientific) in accordance with the manufacturer’s protocols. For stable integration of plasmids, cells were co-transfected with Tol2 transposase at an equimolar ratio. Cells that did not undergo antibiotic selection were cultured for 3-5 days before harvesting. For antibiotic selection, Δ7SMA mESCs were treated with 50 μg/mL hygromycin B (Life Technologies) and/or 6.67 μg/mL blasticidin as indicated, starting 24 hours after transfection. For transient selection, antibiotics were removed from the media after 48 hours. Selected cells were allowed to recover and expand prior to harvesting. All sgRNA sequences designed for this study are listed in the supplement. For Δ7SMA mESCs nusinersen experiments, cells were transfected with 20 nM of fully 2′-O-methoxyethyl (MOE)-modified ASO (5’-TCACTTTCATAATGCTGG-3') on a phosphorothioate backbone (TriLink), using Lipofectamine 3000 (ThermoFisher Scientific). After 24 hours media was replaced every other day with fresh mESC+2i media. For splicing rescue by risdiplam, mESC media was supplemented with 0.1–1 μM of risdiplam (RG7916, Selleck Chemicals LLC) in DMSO, as indicated. Cells were harvested at the indicated timepoints. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 16 High-throughput sequencing of genomic DNA Sequencing library preparation was performed according to previously published protocols(50). Primers are listed in the supplement. Briefly, we isolated genomic DNA (gDNA) with the QIAamp DNA mini kit (Qiagen) and used 250-1000 ng of gDNA for individual locus editing experiments and 20 μg of gDNA for comprehensive context library samples. Sequencing libraries were amplified in two steps, first to amplify the locus of interest and second to add full-length Illumina sequencing adapters using the NEBNext Index Primer Sets 1 and 2 (New England Biolabs) or internally ordered primers with equivalent sequences. All PCRs were performed using NEBNext Ultra II Q5 Master Mix. Samples were pooled using Tape Station (Agilent) and quantified using a KAPA Library Quantification Kit (KAPA Biosystems). The pooled samples were sequenced using Illumina NextSeq or MiSeq. Alignment of fastq files and quantification of editing frequency for individual loci was performed using CRISPResso2 in batch mode(67). The editing frequency for each site was calculated as the ratio between the number of modified reads (i.e. containing nucleotide conversions or indels) and the total number of reads. Base editing characterization library analysis was performed as previously described(50). Quantification of SMN splice products We isolated mRNA from Δ7 mESCs with the RNeasy mini kit (Qiagen) and performed reverse transcription using SuperScript IV (ThermoFisher) according to the manufacturer’s protocols. For targeted SMN2 splice product quantitation by qPCR, high-throughput sequencing, or automated electrophoresis we performed reverse transcription with random hexamers. Inclusion of SMN2 exon 7 was quantified by automated electrophoresis using Tape Station (Agilent). For unbiased SMN2 splice product analysis by high-throughput sequencing, we performed reverse transcription using a custom oligo-dT primer with a Read 2 Illumina sequencing stub. The pooled samples were sequenced using Illumina MiSeq. All PCRs were performed using NEBNext Ultra II Q5 Master Mix, with the addition of Sybr Green for qPCR. Primers are listed in Supplementary Table 3. Western Blot Cells harvested for western blot were washed with ice-cold PBS and incubated at 4 °C for 30 min while rocking in RIPA lysis buffer (ThermoFisher) supplemented with 1 mM PMSF (ThermoFisher) and cOmplete EDTA-free protease inhibitor cocktail (Roche). Lysates were clarified by centrifugation at 12,000 rpm at 4 °C for 20 min. Lysates were normalized using BCA (Pierce BCA Protein Assay Kit) and combined with 4x Laemelli buffer (BioRad) and DTT (ThermoFisher) at a final concentration of 1 mM. We loaded 10 μg of reduced protein per gel lane and performed transfer with an iBlot 2 dry blotting system (ThermoFisher) using the following program: 20 V for 1 min, then 23 V for 4 min, then 25 V for 2 min for a total transfer time of 7 minutes. Blocking was performed at room temperature for 60 minutes with block buffer: 1% BSA in TBST (150 mM NaCl, 0.5% Tween-20, 50 mM Tris-Cl, pH 7.5). Membranes were then incubated in primary antibody diluted in block buffer for 2 hours at room temperature. After a washing, secondary antibodies diluted in TBST were added and incubated for 1 hour at room temperature. Membranes were washed again and imaged using a LI-COR Odyssey. Wash steps were 3x 5-minute washes in TBST. Primary Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 17 antibodies used were mouse anti-human SMN (Proteintech 2C6D9), mouse anti-mouse and human SMN (Proteintech 3A8G1) and rabbit anti-histone H3 (Cell Signaling D1H2), secondary antibodies used were LI-COR IRDye 680RD goat anti-rabbit (#926–68071) and goat anti-mouse (#926–68070). Base editor characterization library assay For characterization of the ABE8e-SpCas9 base editor, we used mouse ESCs carrying the comprehensive context library according to previously published protocols(42, 50). Briefly, 15-cm plates with >107 initial cells were transfected with a total of 50 μg of p2T-ABE8e-SpCas9 and 30 μg of Tol2 plasmid to allow for stable genomic integration with Lipofectamine 3000 according to manufacturer protocols, and selected with 10 μg/mL blasticidin starting the day after transfection for 4 days before harvesting. We maintained an average coverage of ~ 300x per library cassette throughout. We collected gDNA from cells 5 days after transfection, after 4 days of antibiotic selection. Cloning Base editor plasmids were constructed by replacing deaminase and Cas-protein domains of the p2T-CMV-ABE7.10-BlastR (Addgene 152989) plasmid by USER cloning (New England Biolabs)(50). Individual sgRNAs were cloned into the SpCas9-hairpin U6 sgRNA expression plasmid (Addgene 71485) using BbsI plasmid digest and Gibson assembly (New England Biolabs). Protospacer sequences and gene-specific primers used for amplification followed by HTS are listed in Supplementary Table 1. Constructs were transformed into Mach1 chemically competent E. coli (ThermoFisher) grown on LB agar plates and liquid cultures were grown in LB broth overnight at 37 °C with 100 μg/mL ampicillin. Individual colonies were validated by Templiphi rolling circle amplification (ThermoFisher) followed by Sanger sequencing. Verified plasmids were prepared by mini, midi, or maxiprep (Qiagen). AAV vectors were cloned by Gibson assembly (NEB) using NEB Stable Competent E. coli (High Efficiency) to insert the sgRNA sequence and C-terminal base editor half of ABE8e- SpyMac into v5 Cbh-AAV-ABE-NpuC+U6-sgRNA (Addgene 137177), and the N-terminal base editor half and a second U6-sgRNA cassette into v5 Cbh-AAV-ABE-NpuN (Addgene 137178)(74). Neural differentiation Differentiation of Δ7SMA mESCs was performed according to established protocols( 124, 125). Briefly, Δ7SMA mESCs maintained on 0.2% gelatin-coated plates feeder-free in mESC media + 2i were plated onto irradiated mouse embryonic fibroblast (iMEF) feeders on 0.2% gelatin-coated plates in mESC media for 7 days to wean cells from 2i factors. Cells were then seeded at 106 in 10-cm tissue culture treated plates for 48 hours for priming and depletion of feeders. Media was replaced with neural differentiation (ND) media composed of 1:1 DMEM:F12 and Neurobasal media (Life Technologies) supplemented with 10% knockout serum-replacement (KOSR, Life Technologies), Glutamax (GM, Life Technologies) and 0.55 mM 2-mercaptoethanol (b-ME, Sigma-Aldrich), for one hour prior to trypsinization and seeding of 2x106 cells in 10-cm non-tissue culture treated dishes for Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 18 24 hours. Single cells and small early embryoid bodies (EBs) in suspension were collected and transferred to 10-cm tissue culture treated plates in fresh ND media for 24 hours. Small EBs that remained in suspension were collected and transferred to 10-cm tissue culture treated plates in fresh ND media with the addition of 1μM retinoic acid (RA, Sigma-Aldrich R2625) for caudal neural differentiation (CND), or with 1μM RA and 0.5 μM smoothened agonist (SmAg, Calbiochem 566660) for motor neuron differentiation (MND) for 72 hours. Large EBs were collected and split into two 10-cm tissue culture treated plates in neural growth (NG) media composed of 1:1 DMEM:F12 and Neurobasal media supplemented with GM, B27 (Life Technologies), and 10ng/mL human recombinant glial cell line-derived neurotrophic factor (GDNF, R&D Systems 212-GD-010) for 48 hours. EBs were monitored for Mnx1:GFP expression to assess motor neuron differentiation efficiency and imaged using a Zeiss inverted fluorescence microscope or collected for downstream whole transcriptome analysis. Whole transcriptome RNA-sequencing Library preparation, sequencing and analysis were performed by SMART-seq2 as previously described(126). Briefly, total RNA was harvested from cells using the RNeasy Mini kit (Qiagen). First, we incubated 20 ng purified total RNA with RNase inhibitor (Clontech Takara 2313B), dNTP mix (Thermo Fisher R0192), and the 3’-RT primer (5’- AAGCAGTGGTATCAACGCAGAGTAC(T30)VN-3’) at 72 °C for 3 min to anneal the RT primer. Next, we performed first-strand synthesis using the template switching oligo (TSO): (5'-AGCAGTGGTATCAACGCAGAGTACrGrG+G-3' Exiqon, Qiagen) together with RNase inhibitor, betaine (Sigma Aldrich B0300-1VL), MgCl2 (Sigma Aldrich 1028) and Maxima RNase H-minus RT (Thermo Fisher EP0751), according to the manufacturer’s protocols. We performed pre-amplification of first-strand libraries with the ISPCR primer: 5'-AAGCAGTGGTATCAACGCAGAGT-3' using KAPA HiFi HotStart (KAPA KK2601) and SYBR green (Thermo Fisher). Whole transcriptome amplification (WTA) product was washed using DNA SPRI beads (Beckman Coulter A63881) and quantified by Agilent Tapestation. We performed Tagmentation and library preparation of 0.25 ng WTA using the Nextera XT kit (Illumina) and Nextera i7 and Nextera i5 barcoding primers. Samples were pooled and washed using washed using DNA SPRI beads and quantified by Agilent Tapestation and the KAPA Universal Library Quantification kit (Roche KK4824). Libraries were run on Illumina NextSeq 550. FASTQs were generated using bcl2fastq v2.20. Trim Galore v0.6.7 in paired-end mode with default parameters to remove low-quality bases, adapter sequences, and unpaired sequences. Trimmed reads were aligned to the GENCODE mouse reference genome M31 (GRCm39) using STAR (v2.7.10a), quantified using kallisto(127), and refined to canonical coding sequences using CCDS release 21(128). For RNA A-to-I off-target analysis, REDItools v1.3 was used to quantify the average frequency of A-to-I editing among all sequenced adenosines in each sample(129), excluding adenosines with read depth <10 or read quality score <30. The transcriptome-wide A-to-I editing frequency was calculated independently for each biological replicate as: (number of reads in which an adenosine was called as a guanosine)/(total number of reads covering all analyzed adenosines). Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 19 Purification of SpyMac Cas nuclease protein SpyMac Cas nuclease protein was cloned into the expression plasmid pD881-SR (Atum, Cat. No. FPB-27E-269). The resulting plasmid was transformed into BL21 Star DE3 competent cells (ThermoFisher, Cat. No. C601003). Colonies were picked for overnight growth in terrific broth (TB)+25 μg/mL kanamycin at 37 °C. The next day, 2 L of pre- warmed TB were inoculated with overnight culture at a starting OD600 of 0.05. Cells were shaken at 37 °C for about 2.5 hours until the OD600 was ~1.5. Cultures were cold shocked in an ice-water slurry for 1 hour, following which L-rhamnose was added to a final concentration of 0.8% to induce. Cultures were then incubated at 18 °C with shaking for 24 hours to produce protein. Following induction, cells were pelleted and flash-frozen in liquid nitrogen and stored at −80 degrees. The next day, cells were resuspended in 30 mL cold lysis buffer (1 M NaCl, 100 mM Tris-HCl pH 7.0, 5 mM TCEP, 20% glycerol, with 5 tablets of cOmplete, EDTA-free protease inhibitor cocktail tablets (Millipore Sigma, Cat. No. 4693132001). Cells were passed three times through a homogenizer (Avestin Emulsiflex-C3) at ~18,000 psi to lyse. Cell debris was pelleted for 20 minutes using a 20,000 g centrifugation at 4 °C. Supernatant was collected and spiked with 40 mM imidazole, followed by a 1-hour incubation at 4 °C with 1 mL of Ni-NTA resin slurry (G Bioscience Cat. No. 786-940, prewashed once with lysis buffer). Protein-bound resin was washed twice with 12 mL of lysis buffer in a gravity column at 4 °C. Protein was eluted in 3 mL of elution buffer (300 mM imidazole, 500 mM NaCl, 100 mM Tris-HCl pH 7.0, 5 mM TCEP, 10% glycerol). Eluted protein was diluted in 40 mL of low-salt buffer (100 mM Tris-HCl, pH 7.0, 1 mM TCEP, 20% glycerol) just before loading into a 50 mL Akta Superloop for ion exchange purification on the Akta Pure25 FPLC. Ion exchange chromatography was conducted on a 5 mL GE Healthcare HiTrap SP HP pre-packed column (Cat. No. 17115201). After washing the column with low-salt buffer, the diluted protein was flowed through the column to bind. The column was then washed in 15 mL of low salt buffer before being subjected to an increasing gradient to a maximum of 80% high salt buffer (1 M NaCl, 100 mM Tris-HCl, pH 7.0, 5 mM TCEP, 20% glycerol) over the course of 50 mL, at a flow rate of 5 mL per minute. 1-mL fractions were collected during this ramp to high-salt buffer. Peaks were assessed by SDS-PAGE to identify fractions containing the desired protein, which were concentrated first using an Amicon Ultra 15-mL centrifugal filter (100-kDa cutoff, Cat. No. UFC910024), followed by a 0.5-mL 100-kDa cutoff Pierce concentrator (Cat. No. 88503). Concentrated protein was quantified using a BCA assay and determined to be 12.6 milligrams per milliliter (ThermoFisher, Cat. No. 23227). CIRCLE-seq off-target editing analysis Off-target analysis using CIRCLE-seq was performed as previously described(68, 130). Briefly, genomic DNA from HEK293T cells or NIH3T3 cells was isolated using Gentra Puregene Kit (Qiagen) according to manufacturer’s instructions. Purified genomic DNA was sheared with a Covaris S2 instrument to an average length of 300 bp. The fragmented DNA was end repaired, poly-A tailed, and ligated to an uracil-containing stem-loop adaptor using the KAPA HTP Library Preparation Kit, PCR Free (KAPA Biosystems). Adaptor ligated DNA was treated with Lambda Exonuclease (NEB) and E. coli Exonuclease I (NEB), then with USER enzyme (NEB) and T4 polynucleotide kinase (NEB). Intramolecular circularization of the DNA was performed with T4 DNA ligase (NEB) and residual linear Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 20 DNA was degraded by Plasmid-Safe ATP-dependent DNase (Lucigen). In vitro cleavage reactions were performed with 250 ng of Plasmid-Safe ATP-dependent DNase-treated circularized DNA, 90 nM of SpyMac Cas9 nuclease protein, Cas9 nuclease buffer (NEB) and 90 nM of synthetic chemically modified sgRNA (Synthego), in 100 μl. Cleaved products were poly-A tailed, ligated with a hairpin adaptor (NEB), treated with USER enzyme (NEB), and amplified by PCR with barcoded universal primers NEBNext Multiplex Oligos for Illumina (NEB), using Kapa HiFi Polymerase (KAPA Biosystems). Libraries were sequenced with 150-bp paired-end reads on an Illumina MiSeq instrument. CIRCLE-seq data analyses were performed using open-source CIRCLE-seq analysis software and default recommended parameters (https://github.com/tsailabSJ/circleseq). Husbandry of Δ7SMA mice All experiments in animals were approved by the Institutional and Animal Care and Use Committee of the Broad Institute of MIT and Harvard and Ohio State University (OSU). Δ7SMA heterozygous mice ( Smn+/−; SMN2+/+; SMNΔ7 +/+) were purchased from the Jackson Laboratory (005025)(55), and maintained in the Broad Institute and OSU vivaria according to recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Pairs of Δ7SMA heterozygotes were crossed to generate Δ7 SMA mice ( Smn−/−; SMN2+/+; SMNΔ7 +/+). On date of birth (PND0), pups were microtattooed on the foot pads (Aramis) with animal-grade permanent ink (Ketchum) using a sterile hypodermic needle (BD) to enable identification of individual pups. Subsequently, biopsies of ~1 mm tissue were taken from the tail using a sterile blade, lysed for genomic DNA extraction, and used for genotyping by PCR. Litter size was controlled to five pups, including 1-3 homozygous mutants, by culling and cross-fostering among same-age mice. Mice of both sexes were included in the study, although sex has been reported to not have a substantial impact on the phenotype of SMA mice (Treat-NMD SOP Code: SMA_M.2.2.003). Electrophysiology experiments were performed at OSU. All other animal studies were performed at the Broad Institute unless indicated otherwise in the text. At the Broad Institute, the mean birthweight of heterozygous animals was 1.7±0.1 grams, and 1.5±0.1 grams for SMA pups, and any animal weighing <1.5 grams at time of birth was excluded from the study. The average weight of SMA neonates at injection on PND0 at the Broad Institute was 1.6±0.2 grams. At OSU, the mean birthweight of heterozygous animals on the day of birth was 1.3±0.1 grams and 1.2±0.1 grams for SMA pups, and any SMA, heterozygous or wild-type pup weighing ≤1.0 grams at time of birth were excluded from the study. The average weight of SMA neonates at injection on PND0 at OSU was 1.3±0.13 grams. By facility, each litter was subjected to the same exclusion criterion (Treat-NMD SOP Code: SMA_M.2.2.003). Cohort sizes were chosen based on prior experience with these animals, known to allow for determination of statistical significance. Animals were monitored daily for morbidity and mortality and weighed every other day from day of birth. Intracerebroventricular injections Neonatal ICV injections were performed as previously described(74, 131). Briefly, glass capillaries (Drummond 5–000-1001-X10) were pulled to a tip diameter of approximately Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 21 100 μm. High-titer qualified AAV was obtained through the Viral Vector Core at UMass Medical School and concentrated using Amicon Ultra-15 centrifugal filter units (Millipore), quantified by qPCR (AAVpro Titration Kit v.2, Clontech), and stored at 4 °C until use. For injection, a small amount of Fast Green was added to the AAV injection solution to assess ventricle targeting. The injection solution was loaded via front-filling using the included Drummond plungers. Δ7SMA pups were anesthetized by placement on ice for 2–3 minutes, until they were immobile and unresponsive to a toe pinch. Up to 4.5 μL of injection mix was injected freehand into each ventricle on PND0-2. Immunofluorescence imaging of spinal cord sections For immunofluorescence staining of transduced spinal motor neurons, Δ7SMA mice were perfused at 25 weeks with ice-cold PBS and ice-cold 4% PFA, the CNS was exposed, and the whole carcass was fixed overnight in 4% PFA. Whole spinal cord was isolated and fixed in 4% PFA overnight, then consecutively transferred to 10%, 20%, and 30% sucrose in three overnight incubations before embedding in OCT for long-term storage at –80°C. Embedded tissue was cryo-sectioned and stained with goat anti-Choline Acetyltransferase (Millipore AB144P), mouse anti-NeuN (EMD Millipore MAB377), mouse anti-GFAP (Sigma-Aldrich MAB3402), rabbit anti-GFP (Thermo scientific A-11122), and Alexa-Fluor secondary antibodies (Life Technologies), and imaged on an SP8 confocal microscope (Leica). Nuclear isolation and sorting of tissues Tissue harvest and nuclear isolation was performed as previously described(74). Briefly, deceased Δ7SMA mice were stored at −80 °C until dissection of the brain and spinal cord tissue. For isolation of the cortex and cerebella were separated from the brain postmortem using surgical scissors. Hemispheres were separated using a scalpel and the cortex was separated from underlying midbrain tissue with a curved spatula. For nuclear isolation, dissected tissue was homogenized using a glass dounce homogenizer (Sigma D8938) (20 strokes with pestle A followed by 20 strokes with pestle B) in 2 mL ice-cold EZ-PREP buffer (Sigma NUC-101). Samples were incubated for 5 minutes with an additional 2 mL EZ-PREP buffer. Nuclei were centrifuged at 500 g for 5 minutes, and the supernatant removed. For spinal cord tissue, wash steps were repeated ten times. Samples were resuspended with gentle pipetting in 4 mL ice-cold Nuclei Suspension Buffer (NSB) consisting of 100 μg/mL BSA and 3.33 μM Vybrant DyeCycle Ruby (ThermoFisher) in PBS and centrifuged at 500 g for 5 minutes. The supernatant was removed and nuclei were resuspended in 1–2 mL NSB, passed through a 35 μm strainer, and sorted into 200 μL Agencourt DNAdvance lysis buffer using a MoFlo Astrios (Beckman Coulter) at the Broad Institute flow cytometry core. All steps were performed on ice or at 4 °C. Genomic DNA was purified according to the DNAdvance (Agencourt) instructions for 200 μL volume. Behavioral assays Righting reflex was recorded on PND7 by placing neonates on their backs and recording the duration to right themselves with a stopwatch up to a maximum of 30 sec. For inverted screen testing, juvenile mice were subjected to the horizontal grid test for mice (Maze Engineers) on PND25 by placing the animals on a wire-mesh screen which the mice are capable of gripping, then inverting the screen over the course of 2 seconds, animal head first, Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 22 over a padded surface made of bedding 4-5 cm high. The time for the animal to fall onto the bedding was recorded with a stopwatch. Each mouse was assessed with three measurements. The procedure is concluded when the animal falls onto the bedding, or if the animal exceeds 120 seconds for the measurement, in which case the screen is reverted so that the mouse is upright, and the mouse is manually removed from the screen. Voluntary movement of adult mice is recorded on PND40 by open field testing (Omnitech Electronics). Mice were brought into the testing room under normal lighting conditions and allowed 30-60 minutes of acclimation. The animals were placed into the locomotor activity chamber with infrared beams crossing the X, Y and Z axes that plot their ambulatory and fine motor movements and rearing behavior. Recordings are analyzed using Fusion 5.1 SuperFlex software. Electrophysiological Measurements Compound muscle action potential (CMAP) and motor unit number estimate (MUNE) measurements were performed as previously described(132). Briefly, at PND12 the right sciatic nerve was stimulated with a pair of insulated 28-gauge monopolar needles (Teca, Oxford Instruments Medical, NY) placed in proximity to the sciatic nerve in the proximal hind limb. Recording electrodes consisted of a pair of fine ring wire electrodes (Alpine Biomed, Skovlunde, Denmark). The active recording electrode (E1) was placed distal to the knee joint over the proximal portion of the triceps surae muscle and the reference electrode (E2) over the metatarsal region of the foot. A disposable strip electrode (Carefusion, Middleton, WI) was placed on the tail to serve as the ground electrode. For CMAP, supramaximal responses were generated maintaining stimulus currents <10 mA and baseline-to-peak amplitude measurements made. For MUNE, an incremental stimulus technique similar to a previously described procedure was used(132). Submaximal stimulation was used to obtain ten incremental responses to calculate the average single motor unit potential (SMUP) amplitude. The first increment was obtained by delivering square wave stimulations at 1 Hz at an intensity between 0.21 mA to 0.70 mA to obtain the minimal all-or-none SMUP response. If the initial response did not occur with stimulus intensity between 0.21 mA and 0.70 mA, the stimulating cathode position was adjusted either closer or farther away from the position of the sciatic nerve in the proximal thigh to decrease or increase the required stimulus intensity, respectively. This first incremental response was accepted if three duplicate responses were observed. To obtain the subsequent incremental responses the stimulation intensity was adjusted in 0.03 mA steps and incremental responses were distinguished visually in real-time to obtain nine additional increments. To be accepted, each increment was required to be: (1) observed for a total of three duplicate responses, (2) visually distinct from the prior increment, and (3) at least 25 μV larger than the prior increment. The peak-to-peak amplitude of each individual incremental response was calculated by subtracting the amplitude of the prior response. The ten incremental values were averaged to estimate average peak-to-peak single motor unit potential (SMUP) amplitude. The maximum CMAP amplitude (peak-to-peak) was divided by the average SMUP amplitude to yield the MUNE. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Statistical Analysis Page 23 Welch’s two-tailed t-tests were used to compare sequencing, splicing, mRNA levels, and immunostaining data. Error bars represent standard deviations of ≥3 independent biological replicates. Root mean squared error (RMSE) and Pearson’s r-correlation were used for correlation analysis of predicted and observed genome editing outcomes, where appropriate. Kruskal-Wallis tests were used to compare physiology measurements and behaviors of mouse cohorts under experimental conditions. Mann-Whitney tests were used to compare multiparametric measurements of voluntary behaviors of mouse cohorts. The logrank Mantel-Cox test Kaplan-Meier survival curves. All statistical tests were calculated by GraphPad Prism 9.4.1 and Microsoft Excel v16.64. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgements: We thank Dr. Anahita Vieira for assistance with editing, Mary O’Reilly for assistance with figures, and Alvin Hsu for assistance with computational analysis. Funding: US National Institutes of Health grant U01 AI142756 (D.R.L.) US National Institutes of Health grant RM1 HG009490 (D.R.L.) US National Institutes of Health grant R01 EB022376 (D.R.L.) US National Institutes of Health grant R35 GM118062 (D.R.L.) US National Institutes of Health grant P01 HL053749 (D.R.L.) Bill and Melinda Gates Foundation (D.R.L.) Howard Hughes Medical Institute (D.R.L.) Friedreich’s Ataxia Accelerator grant (D.R.L.) The Netherlands Organization for Scientific Research Rubicon Fellowship (M.A.) US National Institutes of Health K99 Pathway to Independence Award NS119743-01A1 (M.A.) Helen Hay Whitney Fellowship (G.A.N.) US National Institutes of Health K99 Pathway to Independence Award HL163805 (G.A.N.) National Science Foundation Graduate Research Fellowship (A.R., M.W.S.) Howard Hughes Medical Institute Hanna Gray Fellowship (M.F.R.) Harvard Chemical Biology Training Grant T32 GM095450 (K.T.Z.) UMass Chan Internal Funding (G.G.) US National Institute of Health grant R01HD060586 (A.H.M.B.) Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. References Page 24 1. Lefebvre S, Bürglen L, Reboullet S, Clermont O, Burlet P, Viollet L, Benichou B, Cruaud C, Millasseau P, Zeviani M, Identification and characterization of a spinal muscular atrophy- determining gene. Cell. 80, 155–65 (1995). [PubMed: 7813012] 2. Sugarman EA, Nagan N, Zhu H, Akmaev VR, Zhou Z, Rohlfs EM, Flynn K, Hendrickson BC, Scholl T, Sirko-Osadsa DA, Allitto BA, Pan-ethnic carrier screening and prenatal diagnosis for spinal muscular atrophy: clinical laboratory analysis of > 72 400 specimens. Eur. J. Hum. Genet 20, 27–32 (2012). [PubMed: 21811307] 3. Roberts DF, Chavez J, Court SDM, The genetic component in child mortality. Arch. Dis. Child 45, 33–38 (1970). [PubMed: 4245389] 4. Boda B, Mas C, Giudicelli C, Nepote V, Guimiot F, Levacher B, Zvara A, Santha M, LeGall I, Simonneau M, Survival motor neuron SMN1 and SMN2 gene promoters: Identical sequences and differential expression in neurons and non-neuronal cells. Eur. J. Hum. Genet 12, 729–737 (2004). [PubMed: 15162126] 5. Rochette CF, Gilbert N, Simard LR, SMN gene duplication and the emergence of the SMN2 gene occurred in distinct hominids: SMN2 is unique to Homo sapiens. Hum. Genet 108, 255–266 (2001). [PubMed: 11354640] 6. Lorson CL, Hahnen E, Androphy EJ, Wirth B, A single nucleotide in the SMN gene regulates splicing and is responsible for spinal muscular atrophy. Proc. Natl. Acad. Sci 96, 6307–6311 (1999). [PubMed: 10339583] 7. Monani UR, Lorson CL, Parsons DW, Prior TW, Androphy EJ, Burghes AHM, McPherson JD, A single nucleotide difference that alters splicing patterns distinguishes the SMA gene SMN1 from the copy gene SMN2. Hum. Mol. Genet 8, 1177–1183 (1999). [PubMed: 10369862] 8. Cho S, Dreyfuss G, A degron created by SMN2 exon 7 skipping is a principal contributor to spinal muscular atrophy severity. Genes Dev. 24, 438–42 (2010). [PubMed: 20194437] 9. Vitte J, Fassier C, Tiziano FD, Dalard C, Soave S, Roblot N, Brahe C, Saugier-Veber P, Bonnefont JP, Melki J, Refined characterization of the expression and stability of the SMN gene products. Am. J. Pathol 171, 1269–1280 (2007). [PubMed: 17717146] 10. Burnett BG, Muñoz E, Tandon A, Kwon DY, Sumner CJ, Fischbeck KH, Regulation of SMN Protein Stability. Mol. Cell. Biol 29, 1107–1115 (2009). [PubMed: 19103745] 11. Cobben JM, Lemmink HH, Snoeck I, Barth PA, van der Lee JH, de Visser M, Survival in SMA type I: A prospective analysis of 34 consecutive cases. Neuromuscul. Disord 18, 541–544 (2008). [PubMed: 18579378] 12. Kolb SJ, Coffey CS, Yankey JW, Krosschell K, Arnold WD, Rutkove SB, Swoboda KJ, Reyna SP, Sakonju A, Darras BT, Shell R, Kuntz N, Castro D, Parsons J, Connolly AM, Chiriboga CA, McDonald C, Burnette WB, Werner K, Thangarajh M, Shieh PB, Finanger E, Cudkowicz ME, McGovern MM, McNeil DE, Finkel R, Iannaccone ST, Kaye E, Kingsley A, Renusch SR, McGovern VL, Wang X, Zaworski PG, Prior TW, Burghes AHM, Bartlett A, Kissel JT, Natural history of infantile-onset spinal muscular atrophy. Ann. Neurol 82, 883–891 (2017). [PubMed: 29149772] 13. Mendell JR, Al-Zaidy S, Shell R, Arnold WD, Rodino-Klapac LR, Prior TW, Lowes L, Alfano L, Berry K, Church K, Kissel JT, Nagendran S, L’Italien J, Sproule DM, Wells C, Cardenas JA, Heitzer MD, Kaspar A, Corcoran S, Braun L, Likhite S, Miranda C, Meyer K, Foust KD, Burghes AHM, Kaspar BK, Single-dose gene-replacement therapy for spinal muscular atrophy. N. Engl. J. Med 377, 1713–1722 (2017). [PubMed: 29091557] 14. Ottesen EW, ISS-N1 makes the first FDA-approved drug for spinal muscular atrophy. Transl. Neurosci 8 (2017), pp. 1–6. [PubMed: 28400976] 15. Hoy SM, Onasemnogene Abeparvovec: First Global Approval. 79, 1255–1262 (2019). 16. Kernochan LE, Russo ML, Woodling NS, Huynh TN, Avila AM, Fischbeck KH, Sumner CJ, The role of histone acetylation in SMN gene expression, doi:10.1093/hmg/ddi130. 17. d’Ydewalle C, Ramos DM, Pyles NJ, Ng SY, Gorz M, Pilato CM, Ling K, Kong L, Ward AJ, Rubin LL, Rigo F, Bennett CF, Sumner CJ, The Antisense Transcript SMN-AS1 Regulates SMN Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 25 Expression and Is a Novel Therapeutic Target for Spinal Muscular Atrophy. Neuron. 93, 66–79 (2017). [PubMed: 28017471] 18. Woo CJ, Maier VK, Davey R, Brennan J, Li G, Brothers J, Schwartz B, Gordo S, Kasper A, Okamoto TR, Johansson HE, Mandefro B, Sareen D, Bialek P, Chau BN, Bhat B, Bullough D, Barsoum J, Gene activation of SMN by selective disruption of lncRNA-mediated recruitment of PRC2 for the treatment of spinal muscular atrophy. Proc. Natl. Acad. Sci. U. S. A 114, E1509– E1518 (2017). [PubMed: 28193854] 19. Cucchiarini M, Madry H, Terwilliger EF, Enhanced expression of the central survival of motor neuron (SMN) protein during the pathogenesis of osteoarthritis. J. Cell. Mol. Med 18, 115–124 (2014). [PubMed: 24237934] 20. Blauw HM, Barnes CP, Van Vught PWJ, Van Rheenen W, Verheul M, Cuppen E, Veldink JH, Van Den Berg LH, SMN1 gene duplications are associated with sporadic ALS. Neurology. 78, 776–780 (2012). [PubMed: 22323753] 21. Van Alstyne M, Tattoli I, Delestrée N, Recinos Y, Workman E, Shihabuddin LS, Zhang C, Mentis GZ, Pellizzoni L, Gain of toxic function by long-term AAV9-mediated SMN overexpression in the sensorimotor circuit. Nat. Neurosci, 1–11 (2021). 22. Chaytow H, Faller KME, Huang YT, Gillingwater TH, Spinal muscular atrophy: From approved therapies to future therapeutic targets for personalized medicine. Cell reports. Med 2 (2021), doi:10.1016/J.XCRM.2021.100346. 23. Chiriboga CA, Swoboda KJ, Darras BT, Iannaccone ST, Montes J, De Vivo DC, Norris DA, Bennett CF, Bishop KM, Results from a phase 1 study of nusinersen (ISIS-SMNRx) in children with spinal muscular atrophy. Neurology. 86, 890 (2016). [PubMed: 26865511] 24. Baranello G, Darras BT, Day JW, Deconinck N, Klein A, Masson R, Mercuri E, Rose K, El-Khairi M, Gerber M, Gorni K, Khwaja O, Kletzl H, Scalco RS, Seabrook T, Fontoura P, Servais L, Risdiplam in Type 1 Spinal Muscular Atrophy. N. Engl. J. Med 384, 915–923 (2021). [PubMed: 33626251] 25. Lefebvre S, Burlet P, Liu Q, Bertrandy S, Clermont O, Munnich A, Dreyfuss G, Melki J, Correlation between severity and SMN protein level in spinal muscular atrophy (1997). 26. Coovert DD, Le TT, McAndrew PE, Strasswimmer J, Crawford TO, Mendell JR, Coulson SE, Androphy EJ, Prior TW, Burghes AHM, The Survival Motor Neuron Protein in Spinal Muscular Atrophy. Hum. Mol. Genet 6, 1205–1214 (1997). [PubMed: 9259265] 27. Patrizi AL, Tiziano F, Zappata S, Donati MA, Neri G, Brahe C, SMN protein analysis in fibroblast, amniocyte and CVS cultures from spinal muscular atrophy patients and its relevance for diagnosis (1999). 28. Ramos DM, d’Ydewalle C, Gabbeta V, Dakka A, Klein SK, Norris DA, Matson J, Taylor SJ, Zaworski PG, Prior TW, Snyder PJ, Valdivia D, Hatem CL, Waters I, Gupte N, Swoboda KJ, Rigo F, Frank Bennett C, Naryshkin N, Paushkin S, Crawford TO, Sumner CJ, Age-dependent SMN expression in disease-relevant tissue and implications for SMA treatment. J. Clin. Invest 129, 4817–4831 (2019). [PubMed: 31589162] 29. Kariya S, Obis T, Garone C, Akay T, Sera F, Iwata S, Homma S, Monani UR, Requirement of enhanced Survival Motoneuron protein imposed during neuromuscular junction maturation. J. Clin. Invest 124, 785–800 (2014). [PubMed: 24463453] 30. Ratni H, Ebeling M, Baird J, Bendels S, Bylund J, Chen KS, Denk N, Feng Z, Green L, Guerard M, Jablonski P, Jacobsen B, Khwaja O, Kletzl H, Ko CP, Kustermann S, Marquet A, Metzger F, Mueller B, Naryshkin NA, Paushkin SV, Pinard E, Poirier A, Reutlinger M, Weetall M, Zeller A, Zhao X, Mueller L, Discovery of Risdiplam, a Selective Survival of Motor Neuron-2 (SMN2) Gene Splicing Modifier for the Treatment of Spinal Muscular Atrophy (SMA). J. Med. Chem 61, 6501–6517 (2018). [PubMed: 30044619] 31. Wurster CD, Ludolph AC, Nusinersen for spinal muscular atrophy. Ther. Adv. Neurol. Disord 11 (2018), , doi:10.1177/1756285618754459. 32. Meyer K, Ferraiuolo L, Schmelzer L, Braun L, McGovern V, Likhite S, Michels O, Govoni A, Fitzgerald J, Morales P, Foust KD, Mendell JR, Burghes AHM, Kaspar BK, Improving single injection CSF delivery of AAV9-mediated gene therapy for SMA: A dose-response study in mice and nonhuman primates. Mol. Ther 23, 477–487 (2015). [PubMed: 25358252] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 26 33. Armbruster N, Lattanzi A, Jeavons M, Van Wittenberghe L, Gjata B, Marais T, Martin S, Vignaud A, Voit T, Mavilio F, Barkats M, Buj-Bello A, Efficacy and biodistribution analysis of intracerebroventricular administration of an optimized scAAV9-SMN1 vector in a mouse model of spinal muscular atrophy. Mol. Ther. - Methods Clin. Dev 3, 16060 (2016). [PubMed: 27652289] 34. Bevan AK, Duque S, Foust KD, Morales PR, Braun L, Schmelzer L, Chan CM, McCrate M, Chicoine LG, Coley BD, Porensky PN, Kolb SJ, Mendell JR, Burghes AHM, Kaspar BK, Systemic gene delivery in large species for targeting spinal cord, brain, and peripheral tissues for pediatric disorders. Mol. Ther 19, 1971–1980 (2011). [PubMed: 21811247] 35. Thomsen G, Burghes AHM, Hsieh C, Do J, Chu BTT, Perry S, Barkho B, Kaufmann P, Sproule DM, Feltner DE, Chung WK, McGovern VL, Hevner RF, Conces M, Pierson CR, Scoto M, Muntoni F, Mendell JR, Foust KD, M Burghes AH, Hsieh C, Do J, T Chu BT, Perry S, Barkho B, Kaufmann P, Sproule DM, Feltner DE, Chung WK, McGovern VL, Hevner RF, Conces M, Pierson CR, Scoto M, Muntoni F, Mendell JR, Foust KD, Biodistribution of onasemnogene abeparvovec DNA, mRNA and SMN protein in human tissue. Nat. Med. 2021 2710 27, 1701–1711 (2021). 36. Das A, Vijayan M, Walton EM, Stafford VG, Fiflis DN, Asokan A, Epigenetic Silencing of Recombinant Adeno-associated Virus Genomes by NP220 and the HUSH Complex. J. Virol 96, e0203921 (2022). [PubMed: 34878926] 37. Greig JA, Breton C, Martins1 KM, Zhu Y, He Z, White J, Bell P, Wang L, Wilson JM, Loss of transgene expression limits liver gene therapy in primates. BioRxiv. 8.5.2017 (2022). 38. Singh NK, Singh NN, Androphy EJ, Singh RN, Splicing of a critical exon of human Survival Motor Neuron is regulated by a unique silencer element located in the last intron. Mol. Cell. Biol 26, 1333–1346 (2006). [PubMed: 16449646] 39. Hua Y, Vickers TA, Okunola HL, Bennett CF, Krainer AR, Antisense Masking of an hnRNP A1/A2 Intronic Splicing Silencer Corrects SMN2 Splicing in Transgenic Mice. Am. J. Hum. Genet 82, 834–848 (2008). [PubMed: 18371932] 40. Li JJ, Lin X, Tang C, Lu YQ, Hu X, Zuo E, Li H, Ying W, Sun Y, Lai LL, Chen HZ, Guo XX, Zhang QJ, Wu S, Zhou C, Shen X, Wang Q, Lin MT, Ma LX, Wang N, Krainer AR, Shi L, Yang H, Chen WJ, Disruption of splicing-regulatory elements using CRISPR/Cas9 to rescue spinal muscular atrophy in human iPSCs and mice. Natl. Sci. Rev 7, 92–101 (2020). [PubMed: 34691481] 41. Zhou M, Tang S, Duan N, Xie M, Li Z, Feng M, Wu L, Hu Z, Liang D, Targeted-Deletion of a Tiny Sequence via Prime Editing to Restore SMN Expression. Int. J. Mol. Sci 23 (2022), doi:10.3390/IJMS23147941. 42. Shen* M, Arbab* M, Hsu JY, Worstell D, Culbertson SJ, Krabbe O, Cassa A, Liu DR, Gifford DK, Sherwood RI, Predictable and precise template-free CRISPR editing of pathogenic variants. Nature. 563, 646–651 (2018). [PubMed: 30405244] 43. Walton RT, Christie KA, Whittaker MN, Kleinstiver BP, Unconstrained genome targeting with near-PAMless engineered CRISPR-Cas9 variants. Science (80-. ) 368, 290–296 (2020). 44. Miller SM, Wang T, Randolph PB, Arbab M, Shen MW, Huang TP, Matuszek Z, Newby GA, Rees HA, Liu DR, Continuous evolution of SpCas9 variants compatible with non-G PAMs, doi:10.1038/s41587-020-0412-8. 45. Chatterjee P, Lee J, Nip L, Koseki SRT, Tysinger E, Sontheimer EJ, Jacobson JM, Jakimo N, A Cas9 with PAM recognition for adenine dinucleotides. Nat. Commun 11, 1–6 (2020). [PubMed: 31911652] 46. Nishimasu H, Shi X, Ishiguro S, Gao L, Hirano S, Okazaki S, Noda T, Abudayyeh OO, Gootenberg JS, Mori H, Oura S, Holmes B, Tanaka M, Seki M, Hirano H, Aburatani H, Ishitani R, Ikawa M, Yachie N, Zhang F, Nureki O, Engineered CRISPR-Cas9 nuclease with expanded targeting space. Science (80-. ) 361, 1259–1262 (2018). 47. Rodriguez-Muela N, Litterman NK, Norabuena EM, Mull JL, Galazo MJ, Sun C, Ng SY, Makhortova NR, White A, Lynes MM, Chung WK, Davidow LS, Macklis JD, Rubin LL, Single- Cell Analysis of SMN Reveals Its Broader Role in Neuromuscular Disease. Cell Rep. 18, 1484– 1498 (2017). [PubMed: 28178525] 48. Wolstencroft EC, Mattis V, Bajer AA, Young PJ, Lorson CL, A non-sequence-specific requirement for SMN protein activity: the role of aminoglycosides in inducing elevated SMN protein levels, doi:10.1093/hmg/ddi131. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 27 49. Madocsai C, Lim SR, Geib T, Lam BJ, Hertel KJ, Correction of SMN2 Pre-mRNA splicing by antisense U7 small nuclear RNAs. Mol. Ther 12, 1013–1022 (2005). [PubMed: 16226920] 50. Arbab M, Shen MW, Mok B, Wilson C, Matuszek Ż, Cassa CA, Liu DR, Determinants of Base Editing Outcomes from Target Library Analysis and Machine Learning. Cell. 182, 463–480.e30 (2020). [PubMed: 32533916] 51. Singh NN, Singh RN, Androphy EJ, Modulating role of RNA structure in alternative splicing of a critical exon in the spinal muscular atrophy genes. Nucleic Acids Res. 35, 371–389 (2007). [PubMed: 17170000] 52. Nishimasu H, Shi X, Ishiguro S, Gao L, Hirano S, Okazaki S, Noda T, Abudayyeh OO, Gootenberg JS, Mori H, Oura S, Holmes B, Tanaka M, Seki M, Hirano H, Aburatani H, Ishitani R, Ikawa M, Yachie N, Zhang F, Nureki O, Shi X, Gao L, Abudayyeh OO, Gootenberg JS, Holmes B, Zhang F, Ishiguro S, Mori H, Tanaka M, Seki M, Aburatani H, Noda T, Oura S, Ikawa M, Engineered CRISPR-Cas9 nuclease with expanded targeting space. Science (80-. ) 361, 1259 (2018). 53. Oakes BL, Fellmann C, Rishi H, Taylor KL, Ren SM, Nadler DC, Yokoo R, Arkin AP, Doudna JA, Savage DF, CRISPR-Cas9 Circular Permutants as Programmable Scaffolds for Genome Modification. Cell. 176, 254–267.e16 (2019). [PubMed: 30633905] 54. Huang TP, Zhao KT, Miller SM, Gaudelli NM, Oakes BL, Fellmann C, Savage DF, Liu DR, Circularly permuted and PAM-modified Cas9 variants broaden the targeting scope of base editors. Nat. Biotechnol (2019). 55. Le TT, Pham LT, Butchbach MER, Zhang HL, Monani UR, Coovert DD, Gavrilina TO, Xing L, Bassell GJ, Burghes AHM, SMNΔ7, the major product of the centromeric survival motor neuron (SMN2) gene, extends survival in mice with spinal muscular atrophy and associates with full-length SMN. Hum. Mol. Genet 14, 845–857 (2005). [PubMed: 15703193] 56. Singh NN, Androphy EJ, Singh RN, An extended inhibitory context causes skipping of exon 7 of SMN2 in spinal muscular atrophy. Biochem. Biophys. Res. Commun 315, 381–388 (2004). [PubMed: 14766219] 57. Cartegni L, Hastings ML, Calarco JA, De Stanchina E, Krainer AR, Determinants of Exon 7 Splicing in the Spinal Muscular Atrophy Genes, SMN1 and SMN2. Am. J. Hum. Genet 78, 63 (2006). [PubMed: 16385450] 58. Schrank B, Götz R, Gunnersen JM, Ure JM, V Toyka K, Smith AG, Sendtner M, Gotz R, Gunnersen JM, Ure JM, V Toyka K, Smith AG, Sendtner M, Inactivation of the survival motor neuron gene, a candidate gene for human spinal muscular atrophy, leads to massive cell death in early mouse embryos. Proc. Natl. Acad. Sci. U. S. A 94, 9920–9925 (1997). [PubMed: 9275227] 59. Monani UR, Spinal muscular atrophy: a deficiency in a ubiquitous protein; a motor neuron-specific disease. Neuron. 48, 885–896 (2005). [PubMed: 16364894] 60. Singh RN, Howell MD, Ottesen EW, Singh NN, Diverse role of survival motor neuron protein. Biochim. Biophys. Acta - Gene Regul. Mech 1860 (2017), pp. 299–315. [PubMed: 28095296] 61. Briese M, Esmaeili B, Fraboulet S, Burt EC, Christodoulou S, Towers PR, Davies KE, Sattelle DB, Deletion of smn-1, the Caenorhabditis elegans ortholog of the spinal muscular atrophy gene, results in locomotor dysfunction and reduced lifespan. Hum. Mol. Genet 18, 97–104 (2009). [PubMed: 18829666] 62. Richter MF, Zhao KT, Eton E, Lapinaite A, Newby GA, Thuronyi BW, Wilson C, Koblan LW, Zeng J, Bauer DE, Doudna JA, Liu DR, Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. Nat. Biotechnol (2020), doi:10.1038/ s41587-020-0453-z. 63. Doman JL, Raguram A, Newby GA, Liu DR, Evaluation and minimization of Cas9-independent off-target DNA editing by cytosine base editors. Nat. Biotechnol, doi:10.1038/s41587-020-0414-6. 64. Rees HA, Wilson C, Doman JL, Liu DR, Analysis and minimization of cellular RNA editing by DNA adenine base editors. Sci. Adv 5 (2019), doi:10.1126/sciadv.aax5717. 65. Yu Y, Leete TC, Born DA, Young L, Barrera LA, Lee SJ, Rees HA, Ciaramella G, Gaudelli NM, Cytosine base editors with minimized unguided DNA and RNA off-target events and high on-target activity. Nat. Commun 11, 1–10 (2020). [PubMed: 31911652] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 28 66. Zhou C, Sun Y, Yan R, Liu Y, Zuo E, Gu C, Han L, Wei Y, Hu X, Zeng R, Li Y, Zhou H, Guo F, Yang H, Off-target RNA mutation induced by DNA base editing and its elimination by mutagenesis. Nature. 571, 275–278 (2019). [PubMed: 31181567] 67. Clement K, Rees H, Canver MC, Gehrke JM, Farouni R, Hsu JY, Cole MA, Liu DR, Joung JK, Bauer DE, Pinello L, CRISPResso2 provides accurate and rapid genome editing sequence analysis. Nat. Biotechnol 37 (2019), pp. 224–226. [PubMed: 30809026] 68. Tsai SQ, Nguyen NT, Malagon-Lopez J, V Topkar V, Aryee MJ, Joung JK, CIRCLE-seq: a highly sensitive in vitro screen for genome-wide CRISPR-Cas9 nuclease off-targets. Nat Methods. 14, 607–614 (2017). [PubMed: 28459458] 69. Zhao Y, Dai Z, Liang Y, Yin M, Ma K, He M, Ouyang H, Teng C-B, Sequence-specific inhibition of microRNA via CRISPR/CRISPRi system. Sci. Rep 4, 3943 (2015). 70. Foust KD, Nurre E, Montgomery CL, Hernandez A, Chan CM, Kaspar BK, Curtis M, Kaspar BK, Intravascular AAV9 preferentially targets neonatal-neurons and adult-astrocytes in CNS. Nat. Biotechnol 27, 59–65 (2009). [PubMed: 19098898] 71. Hammond SL, Leek AN, Richman EH, Tjalkens RB, Cellular selectivity of AAV serotypes for gene delivery in neurons and astrocytes by neonatal intracerebroventricular injection. PLoS One. 12 (2017), doi:10.1371/journal.pone.0188830. 72. Mathiesen SN, Lock JL, Schoderboeck L, Abraham WC, Hughes SM, CNS Transduction Benefits of AAV-PHP.eB over AAV9 Are Dependent on Administration Route and Mouse Strain. Mol. Ther. Methods Clin. Dev 19, 447 (2020). [PubMed: 33294493] 73. Arnold WD, McGovern VL, Sanchez B, Li J, Corlett KM, Kolb SJ, Rutkove SB, Burghes AH, The neuromuscular impact of symptomatic SMN restoration in a mouse model of spinal muscular atrophy. Neurobiol. Dis 87, 116–123 (2016). [PubMed: 26733414] 74. Levy JM, Yeh WH, Pendse N, Davis JR, Hennessey E, Butcher R, Koblan LW, Comander J, Liu Q, Liu DR, Cytosine and adenine base editing of the brain, liver, retina, heart and skeletal muscle of mice via adeno-associated viruses. Nat. Biomed. Eng 4, 97–110 (2020). [PubMed: 31937940] 75. Schuster DJ, Dykstra JA, Riedl MS, Kitto KF, Belur LR, Scott McIvor R, Elde RP, Fairbanks CA, Vulchanova L, McIvor RS, Elde RP, Fairbanks CA, Vulchanova L, Biodistribution of adeno- associated virus serotype 9 (AAV9) vector after intrathecal and intravenous delivery in mouse. Front. Neuroanat 8, 42 (2014). [PubMed: 24959122] 76. Swiech L, Heidenreich M, Banerjee A, Habib N, Li Y, Trombetta J, Sur M, Zhang F, In vivo interrogation of gene function in the mammalian brain using CRISPR-Cas9. Nat. Biotechnol 33, 102 (2014). [PubMed: 25326897] 77. Cartegni L, Krainer AR, Disruption of an SF2/ASF-dependent exonic splicing enhancer in SMN2 causes spinal muscular atrophy in the absence of SMN1. Nat. Genet. 2002 304 30, 377–384 (2002). 78. Groen EJN, Perenthaler E, Courtney NL, Jordan CY, Shorrock HK, Van Der Hoorn D, Huang YT, Murray LM, Viero G, Gillingwater TH, Temporal and tissue-specific variability of SMN protein levels in mouse models of spinal muscular atrophy. Hum. Mol. Genet 27, 2851–2862 (2018). [PubMed: 29790918] 79. Karlsson M, Zhang C, Méar L, Zhong W, Digre A, Katona B, Sjöstedt E, Butler L, Odeberg J, Dusart P, Edfors F, Oksvold P, von Feilitzen K, Zwahlen M, Arif M, Altay O, Li X, Ozcan M, Mardonoglu A, Fagerberg L, Mulder J, Luo Y, Ponten F, Uhlén M, Lindskog C, A single–cell type transcriptomics map of human tissues. Sci. Adv 7, eabh2169 (2021). [PubMed: 34321199] 80. Akcakaya P, Bobbin ML, Guo JA, Malagon-Lopez J, Clement K, Garcia SP, Fellows MD, Porritt MJ, Firth MA, Carreras A, Baccega T, Seeliger F, Bjursell M, Tsai SQ, Nguyen NT, Nitsch R, Mayr LM, Pinello L, Bohlooly YM, Aryee MJ, Maresca M, Joung JK, Bohlooly-Y M, Aryee MJ, Maresca M, Joung JK, In vivo CRISPR editing with no detectable genome-wide off-target mutations. Nature. 561, 416–419 (2018). [PubMed: 30209390] 81. Davis JR, Wang X, Witte IP, Huang TP, Levy JM, Raguram A, Banskota S, Seidah NG, Musunuru K, Liu DR, Efficient in vivo base editing via single adeno-associated viruses with size-optimized genomes encoding compact adenine base editors. Nat. Biomed. Eng 6, 1272–1283 (2022). [PubMed: 35902773] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 29 82. Rothgangl T, Dennis MK, Lin PJC, Oka R, Witzigmann D, Villiger L, Qi W, Hruzova M, Kissling L, Lenggenhager D, Borrelli C, Egli S, Frey N, Bakker N, Walker JA, Kadina AP, Victorov DV, Pacesa M, Kreutzer S, Kontarakis Z, Moor A, Jinek M, Weissman D, Stoffel M, van Boxtel R, Holden K, Pardi N, Thöny B, Häberle J, Tam YK, Semple SC, Schwank G, In vivo adenine base editing of PCSK9 in macaques reduces LDL cholesterol levels. Nat. Biotechnol 39, 949–957 (2021). [PubMed: 34012094] 83. David Arnold W, Porensky PN, Mcgovern VL, Iyer CC, Duque S, Li X, Meyer K, Schmelzer L, Kaspar BK, Kolb SJ, Kissel JT, Burghes AHM, Electrophysiological biomarkers in spinal muscular atrophy: proof of concept. Ann. Clin. Transl. Neurol 1, 34 (2014). [PubMed: 24511555] 84. Tscherter A, Rüsch CT, Baumann D, Enzmann C, Hasselmann O, Jacquier D, Jung HH, Kruijshaar ME, Kuehni CE, Neuwirth C, Stettner GM, Klein A, Baumann D, Enzmann C, Hasselmann O, Jacquier D, Jung HH, Klein A, Kruijshaar ME, Kuehni CE, Lötscher N, Neuwirth C, Ramelli GP, Ripellino P, Scheidegger O, Stettner GM, Tscherter A, Wille D-A, Evaluation of real-life outcome data of patients with spinal muscular atrophy treated with nusinersen in Switzerland. Neuromuscul. Disord 0 (2022), doi:10.1016/J.NMD.2022.02.001/ATTACHMENT/ 96B80B10-FB76-49C7-9B7B-571ECC277841/MMC1.DOCX. 85. Dangouloff T, Servais L, Clinical Evidence Supporting Early Treatment Of Patients With Spinal Muscular Atrophy: Current Perspectives. Ther. Clin. Risk Manag 15, 1153 (2019). [PubMed: 31632042] 86. Strauss KA, Farrar MA, Muntoni F, Saito K, Mendell JR, Servais L, McMillan HJ, Finkel RS, Swoboda KJ, Kwon JM, Zaidman CM, Chiriboga CA, Iannaccone ST, Krueger JM, Parsons JA, Shieh PB, Kavanagh S, Tauscher-Wisniewski S, McGill BE, Macek TA, Onasemnogene abeparvovec for presymptomatic infants with two copies of SMN2 at risk for spinal muscular atrophy type 1: the Phase III SPR1NT trial. Nat. Med. 2022 287 28, 1381–1389 (2022). 87. De Vivo DC, Bertini E, Swoboda KJ, Hwu WL, Crawford TO, Finkel RS, Kirschner J, Kuntz NL, Parsons JA, Ryan MM, Butterfield RJ, Topaloglu H, Ben-Omran T, Sansone VA, Jong YJ, Shu F, Staropoli JF, Kerr D, Sandrock AW, Stebbins C, Petrillo M, Braley G, Johnson K, Foster R, Gheuens S, Bhan I, Reyna SP, Fradette S, Farwell W, Nusinersen initiated in infants during the presymptomatic stage of spinal muscular atrophy: Interim efficacy and safety results from the Phase 2 NURTURE study. Neuromuscul. Disord 29, 842–856 (2019). [PubMed: 31704158] 88. Robbins KL, Glascock JJ, Osman EY, Miller MR, Lorson CL, Defining the therapeutic window in a severe animal model of spinal muscular atrophy. Hum. Mol. Genet 23, 4559–4568 (2014). [PubMed: 24722206] 89. Le TT, McGovern VL, Alwine IE, Wang X, Massoni-Laporte A, Rich MM, Burghes AHM, Temporal requirement for high SMN expression in SMA mice. Hum. Mol. Genet 20, 3578–3591 (2011). [PubMed: 21672919] 90. Flurkey K, Currer JM, Harrison DE, Mouse Models in Aging Research. Mouse Biomed. Res 3, 637–672 (2007). 91. Mercuri E, Darras BT, Chiriboga CA, Day JW, Campbell C, Connolly AM, Iannaccone ST, Kirschner J, Kuntz NL, Saito K, Shieh PB, Tulinius M, Mazzone ES, Montes J, Bishop KM, Yang Q, Foster R, Gheuens S, Bennett CF, Farwell W, Schneider E, De Vivo DC, Finkel RS, Nusinersen versus Sham Control in Later-Onset Spinal Muscular Atrophy. N. Engl. J. Med 378, 625–635 (2018). [PubMed: 29443664] 92. Finkel RS, Day JW, Darras BT, Kuntz NL, Connolly AM, Crawford T, Butterfield RJ, Shieh PB, Tennekoon G, Iannaccone ST, Meriggioli M, Tauscher-Wisniewski S, Shoffner J, Ogrinc FG, Kavanagh S, Kernbauer E, Whittle J, Sproule DM, Feltner DE, Mendell JR, One-Time Intrathecal (IT) Administration of AVXS-101 IT Gene-Replacement Therapy for Spinal Muscular Atrophy: Phase 1 Study (STRONG) (2493). Neurology. 94 (2020). 93. Lutz CM, Kariya S, Patruni S, Osborne MA, Liu D, Henderson CE, Li DK, Pellizzoni L, Rojas J, Valenzuela DM, Murphy AJ, Winberg ML, Monani UR, Postsymptomatic restoration of SMN rescues the disease phenotype in a mouse model of severe spinal muscular atrophy. J. Clin. Invest 121, 3029–3041 (2011). [PubMed: 21785219] 94. Wang Z, Ma HI, Li J, Sun L, Zhang J, Xiao X, Rapid and highly efficient transduction by double-stranded adeno-associated virus vectors in vitro and in vivo. Gene Ther. 10, 2105–2111 (2003). [PubMed: 14625564] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 30 95. Hauck B, Zhao W, High K, Xiao W, Intracellular Viral Processing, Not Single-Stranded DNA Accumulation, Is Crucial for Recombinant Adeno-Associated Virus Transduction. J. Virol 78, 13678–13686 (2004). [PubMed: 15564477] 96. McCarty DM, Monahan PE, Samulski RJ, Self-complementary recombinant adeno-associated virus (scAAV) vectors promote efficient transduction independently of DNA synthesis. Gene Ther. 8, 1248–1254 (2001). [PubMed: 11509958] 97. Feng Z, Ling KKY, Zhao X, Zhou C, Karp G, Welch EM, Naryshkin N, Ratni H, Chen KS, Metzger F, Paushkin S, Weetall M, Ko CP, Pharmacologically induced mouse model of adult spinal muscular atrophy to evaluate effectiveness of therapeutics after disease onset. Hum. Mol. Genet 25, 964–975 (2016). [PubMed: 26758873] 98. Naryshkin NA, Weetall M, Dakka A, Narasimhan J, Zhao X, Feng Z, Ling KKY, Karp GM, Qi H, Woll MG, Chen G, Zhang N, Gabbeta V, Vazirani P, Bhattacharyya A, Furia B, Risher N, Sheedy J, Kong R, Ma J, Turpoff A, Lee CS, Zhang X, Moon YC, Trifillis P, Welch EM, Colacino JM, Babiak J, Almstead NG, Peltz SW, Eng LA, Chen KS, Mull JL, Lynes MS, Rubin LL, Fontoura P, Santarelli L, Haehnke D, McCarthy KD, Schmucki R, Ebeling M, Sivaramakrishnan M, Ko CP, Paushkin SV, Ratni H, Gerlach I, Ghosh A, Metzger F, SMN2 splicing modifiers improve motor function and longevity in mice with spinal muscular atrophy. Science (80-. ) 345, 688–693 (2014). 99. Passini MA, Bu J, Richards AM, Kinnecom C, Sardi SP, Stanek LM, Hua Y, Rigo F, Matson J, Hung G, Kaye EM, Shihabuddin LS, Krainer AR, Bennett CF, Cheng SH, Antisense Oligonucleotides Delivered to the Mouse CNS Ameliorate Symptoms of Severe Spinal Muscular Atrophy. Sci. Transl. Med 3, 72ra18–72ra18 (2011). 100. Finkel RS, Mercuri E, Darras BT, Connolly AM, Kuntz NL, Kirschner J, Chiriboga CA, Saito K, Servais L, Tizzano E, Topaloglu H, Tulinius M, Montes J, Glanzman AM, Bishop K, Zhong ZJ, Gheuens S, Bennett CF, Schneider E, Farwell W, De Vivo DC, Nusinersen versus Sham Control in Infantile-Onset Spinal Muscular Atrophy. N. Engl. J. Med 377, 1723–1732 (2017). [PubMed: 29091570] 101. Wadman RI, Stam M, Jansen MD, Der Van Weegen Y, Wijngaarde CA, Harschnitz O, Sodaar P, Braun KPJ, Dooijes D, Lemmink HH, Van Den Berg LH, Van Ludo Der Pol W, A comparative study of SMN protein and mRNA in blood and fibroblasts in patients with spinal muscular atrophy and healthy controls. PLoS One. 11, e0167087–e0167087 (2016). [PubMed: 27893852] 102. Shmakov S, Abudayyeh OO, Makarova KS, Wolf YI, Gootenberg JS, Semenova E, Minakhin L, Joung J, Konermann S, Severinov K, Zhang F, Koonin EV, Discovery and Functional Characterization of Diverse Class 2 CRISPR-Cas Systems. Mol. Cell 60, 385–397 (2015). [PubMed: 26593719] 103. Butchbach MER, Copy number variations in the survival motor neuron genes: Implications for spinal muscular atrophy and other neurodegenerative diseases. Front. Mol. Biosci 3 (2016), , doi:10.3389/fmolb.2016.00007. 104. Kato T, Hara S, Goto Y, Ogawa Y, Okayasu H, Kubota S, Tamano M, Terao M, Takada S, Creation of mutant mice with megabase-sized deletions containing custom-designed breakpoints by means of the CRISPR/Cas9 system. Sci. Rep 7 (2017), doi:10.1038/S41598-017-00140-9. 105. Gaudelli NM, Komor AC, Rees HA, Packer MS, Badran AH, Bryson DI, Liu DR, Programmable base editing of A•T to G•C in genomic DNA without DNA cleavage. Nature. 551, 464–471 (2017). [PubMed: 29160308] 106. Komor AC, Kim YB, Packer MS, Zuris JA, Liu DR, Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature. 533, 420–424 (2016). [PubMed: 27096365] 107. Gaudelli NM, Lam DK, Rees HA, Solá-Esteves NM, Barrera LA, Born DA, Edwards A, Gehrke JM, Lee SJ, Liquori AJ, Murray R, Packer MS, Rinaldi C, Slaymaker IM, Yen J, Young LE, Ciaramella G, Directed evolution of adenine base editors with increased activity and therapeutic application. 38, 892–900 (2020). 108. McAndrew PE, Parsons DW, Simard LR, Rochette C, Ray PN, Mendell JR, Prior TW, Burghes AHM, Identification of Proximal Spinal Muscular Atrophy Carriers and Patients by Analysis of SMNT and SMNC Gene Copy Number. Am. J. Hum. Genet 60, 1411–1422 (1997). [PubMed: 9199562] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 31 109. Mailman MD, Heinz JW, Papp AC, Snyder PJ, Sedra MS, Wirth B, Burghes AHM, Prior TW, Molecular analysis of spinal muscular atrophy and modification of the phenotype by SMN2. Genet. Med 4, 20–26 (2002). [PubMed: 11839954] 110. Ruhno C, McGovern VL, Avenarius MR, Snyder PJ, Prior TW, Nery FC, Muhtaseb A, Roggenbuck JS, Kissel JT, Sansone VA, Siranosian JJ, Johnstone AJ, Nwe PH, Zhang RZ, Swoboda KJ, Burghes AHM, Complete sequencing of the SMN2 gene in SMA patients detects SMN gene deletion junctions and variants in SMN2 that modify the SMA phenotype. Hum. Genet. 2019 1383 138, 241–256 (2019). 111. Calucho M, Bernal S, Alías L, March F, Venceslá A, Rodríguez-Álvarez FJ, Aller E, Fernández RM, Borrego S, Millán JM, Hernández-Chico C, Cuscó I, Fuentes-Prior P, Tizzano EF, Correlation between SMA type and SMN2 copy number revisited: An analysis of 625 unrelated Spanish patients and a compilation of 2834 reported cases. Neuromuscul. Disord 28, 208–215 (2018). [PubMed: 29433793] 112. Prior TW, Krainer AR, Hua Y, Swoboda KJ, Snyder PC, Bridgeman SJ, Burghes AHM, Kissel JT, A Positive Modifier of Spinal Muscular Atrophy in the SMN2 Gene. Am. J. Hum. Genet 85, 408–413 (2009). [PubMed: 19716110] 113. Besse A, Astord S, Marais T, Roda M, Giroux B, Lejeune FX, Relaix F, Smeriglio P, Barkats M, Biferi MG, AAV9-Mediated Expression of SMN Restricted to Neurons Does Not Rescue the Spinal Muscular Atrophy Phenotype in Mice. Mol. Ther 28, 1887–1901 (2020). [PubMed: 32470325] 114. Yu D, Lv M, Chen W, Zhong S, Zhang X, Chen L, Ma T, Tang J, Zhao J, Role of miR-155 in drug resistance of breast cancer. Tumor Biol. 36, 1395–1401 (2015). 115. Bevan AK, Hutchinson KR, Foust KD, Braun L, McGovern VL, Schmelzer L, Ward JG, Petruska JC, Lucchesi PA, Burghes AHMM, Kaspar BK, Early heart failure in the SMNDelta7 model of spinal muscular atrophy and correction by postnatal scAAV9-SMN delivery. Hum. Mol. Genet 19, 3895–3905 (2010). [PubMed: 20639395] 116. Shababi M, Habibi J, Yang HT, Vale SM, Sewell WA, Lorson CL, Cardiac defects contribute to the pathology of spinal muscular atrophy models. Hum. Mol. Genet 19, 4059–4071 (2010). [PubMed: 20696672] 117. McGovern VL, Iyer CC, Arnold WD, Gombash SE, Zaworski PG, Blatnik AJ, Foust KD, Burghes AHMM, David Arnold W, Gombash SE, Zaworski PG, Blatnik AJ, Foust KD, Burghes AHMM, SMN expression is required in motor neurons to rescue electrophysiological deficits in the SMNΔ7 mouse model of SMA. Hum. Mol. Genet 24, 5524–5541 (2015). [PubMed: 26206889] 118. Hua Y, Sahashi K, Rigo F, Hung G, Horev G, Bennett CF, Krainer AR, Peripheral SMN restoration is essential for long-term rescue of a severe spinal muscular atrophy mouse model. Nature. 478, 123–126 (2011). [PubMed: 21979052] 119. Lipnick SL, Agniel DM, Aggarwal R, Makhortova NR, Finlayson SG, Brocato A, Palmer N, Darras BT, Kohane I, Rubin LL, Systemic nature of spinal muscular atrophy revealed by studying insurance claims. PLoS One. 14, e0213680 (2019). [PubMed: 30870495] 120. Villiger L, Grisch-Chan HM, Lindsay H, Ringnalda F, Pogliano CB, Allegri G, Fingerhut R, Haberle J, Matos J, Robinson MD, Thony B, Schwank G, Häberle J, Matos J, Robinson MD, Thöny B, Schwank G, Treatment of a metabolic liver disease by in vivo genome base editing in adult mice. Nat Med. 24, 1519–1525 (2018). [PubMed: 30297904] 121. Yeh WH, Chiang H, Rees HA, Edge ASB, Liu DR, In vivo base editing of post-mitotic sensory cells. Nat. Commun 9 (2018), doi:10.1038/s41467-018-04580-3. 122. Kuzmin DA, Shutova MV, Johnston NR, Smith OP, Fedorin VV, Kukushkin YS, van der Loo JCM, Johnstone EC, The clinical landscape for AAV gene therapies. Nat. Rev. Drug Discov 20, 173–174 (2021). [PubMed: 33495615] 123. Arbab M, Srinivasan S, Hashimoto T, Geijsen N, Sherwood RII, Cloning-free CRISPR. Stem Cell Reports. 5, 908–917 (2015). [PubMed: 26527385] 124. Wichterle H, Lieberam I, Porter JA, Jessell TM, Directed differentiation of embryonic stem cells into motor neurons. Cell. 110, 385–397 (2002). [PubMed: 12176325] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 32 125. Wichterle H, Peljto M, Differentiation of Mouse Embryonic Stem Cells to Spinal Motor Neurons. Curr. Protoc. Stem Cell Biol 5 (2008), doi:10.1002/9780470151808.sc01h01s5. 126. Picelli S, Faridani OR, Björklund ÅK, Winberg G, Sagasser S, Sandberg R, Full-length RNA-seq from single cells using Smart-seq2. Nat. Protoc 9, 171–181 (2014). [PubMed: 24385147] 127. Bray NL, Pimentel H, Melsted P, Pachter L, Near-optimal probabilistic RNA-seq quantification. Nat. Biotechnol 34, 525–527 (2016). [PubMed: 27043002] 128. Pujar S, O’Leary NA, Farrell CM, Loveland JE, Mudge JM, Wallin C, Girón CG, Diekhans M, Barnes I, Bennett R, Berry AE, Cox E, Davidson C, Goldfarb T, Gonzalez JM, Hunt T, Jackson J, Joardar V, Kay MP, Kodali VK, Martin FJ, McAndrews M, McGarvey KM, Murphy M, Rajput B, Rangwala SH, Riddick LD, Seal RL, Suner MM, Webb D, Zhu S, Aken BL, Bruford EA, Bult CJ, Frankish A, Murphy T, Pruitt KD, Consensus coding sequence (CCDS) database: A standardized set of human and mouse protein-coding regions supported by expert curation. Nucleic Acids Res. 46, D221–D228 (2018). [PubMed: 29126148] 129. DIroma MA, Ciaccia L, Pesole G, Picardi E, Elucidating the editome: Bioinformatics approaches for RNA editing detection. Brief. Bioinform 20, 436–447 (2019). [PubMed: 29040360] 130. Lazzarotto CR, Nguyen NT, Tang X, Malagon-Lopez J, Guo JA, Aryee MJ, Joung JK, Tsai SQ, Defining CRISPR–Cas9 genome-wide nuclease activities with CIRCLE-seq. Nat. Protoc 13, 2615–2642 (2018). [PubMed: 30341435] 131. Porensky PN, Mitrpant C, McGovern VL, Bevan AK, Foust KD, Kaspar BK, Wilton SD, Burghes AHM, A single administration of morpholino antisense oligomer rescues spinal muscular atrophy in mouse. Hum. Mol. Genet 21, 1625–1638 (2012). [PubMed: 22186025] 132. Arnold WD, Sheth KA, Wier CG, Kissel JT, Burghes AH, Kolb SJ, Electrophysiological Motor Unit Number Estimation (MUNE) Measuring Compound Muscle Action Potential (CMAP) in Mouse Hindlimb Muscles. J. Vis. Exp, 52899 (2015). [PubMed: 26436455] 133. Cobb MS, Rose FF, Rindt H, Glascock JJ, Shababi M, Miller MR, Osman EY, Yen PF, Garcia ML, Martin BR, Wetz MJ, Mazzasette C, Feng Z, Ko CP, Lorson CL, Development and characterization of an SMN2-based intermediate mouse model of spinal muscular atrophy. Hum. Mol. Genet (2013), doi:10.1093/hmg/ddt037. 134. Farooq F, Balabanian S, Liu X, Holcik M, MacKenzie A, p38 Mitogen-activated protein kinase stabilizes SMN mRNA through RNA binding protein HuR. Hum. Mol. Genet 18, 4035–4045 (2009). [PubMed: 19648294] 135. Singh RN, Singh NN, "Mechanism of splicing regulation of spinal muscular atrophy genes" in Advances in Neurobiology (2018; file:///Users/marbab/Library/Application Support/Mendeley Desktop/Downloaded/Singh, Singh - Unknown - Mechanism of Splicing Regulation of Spinal Muscular Atrophy Genes.pdf), vol. 20, pp. 31–61. [PubMed: 29916015] 136. Yoshimoto S, Harahap NIF, Hamamura Y, Ar Rochmah M, Shima A, Morisada N, Shinohara M, Saito T, Saito K, Lai PS, Matsuo M, Awano H, Morioka I, Iijima K, Nishio H, Alternative splicing of a cryptic exon embedded in intron 6 of SMN1 and SMN2. Hum. Genome Var 3, 1–3 (2016). 137. Seo J, Singh NN, Ottesen EW, Lee BM, Singh RN, A novel human-specific splice isoform alters the critical C-terminus of Survival Motor Neuron protein. Sci. Rep 6, 1–14 (2016). [PubMed: 28442746] 138. Ottesen EW, Luo D, Seo J, Singh NN, Singh RN, Human Survival Motor Neuron genes generate a vast repertoire of circular RNAs. Nucleic Acids Res. 47, 2884–2905 (2019). [PubMed: 30698797] 139. Allen F, Crepaldi L, Alsinet C, Strong AJ, Kleshchevnikov V, De Angeli P, Páleníková P, Khodak A, Kiselev V, Kosicki M, Bassett AR, Harding H, Galanty Y, Muñoz-Martínez F, Metzakopian E, Jackson SP, Parts L, Predicting the mutations generated by repair of Cas9-induced double-strand breaks. Nat. Biotechnol 37, 64–82 (2019). 140. Kim HK, Min S, Song M, Jung S, Choi JW, Kim Y, Lee S, Yoon S, Kim H, Deep learning improves prediction of CRISPR-Cpf1 guide RNA activity. Nat. Biotechnol 36, 239–241 (2018). [PubMed: 29431740] Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 33 141. Kim HK, Kim Y, Lee S, Min S, Bae JY, Choi JW, Park J, Jung D, Yoon S, Kim HH, SpCas9 activity prediction by DeepSpCas9, a deep learning–based model with high generalization performance. Sci. Adv 5, eaax9249–eaax9249 (2019). [PubMed: 31723604] 142. Doench JG, Fusi N, Sullender M, Hegde M, Vaimberg EW, Donovan KF, Smith I, Tothova Z, Wilen C, Orchard R, Virgin HW, Listgarten J, Root DE, Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9. Nat Biotechnol. 34, 184–191 (2016). [PubMed: 26780180] 143. Hu JH, Miller SM, Geurts MH, Tang W, Chen L, Sun N, Zeina CM, Gao X, Rees HA, Lin Z, Liu DR, Evolved Cas9 variants with broad PAM compatibility and high DNA specificity. Nature. 556, 57 (2018). [PubMed: 29512652] 144. Chatterjee P, Lee J, Nip L, Koseki SRTT, Tysinger E, Sontheimer EJ, Jacobson JM, Jakimo N, A Cas9 with PAM recognition for adenine dinucleotides. Nat. Commun 11, 1–6 (2020). [PubMed: 31911652] 145. Lin X, Chen H, Lu YQ, Hong S, Hu X, Gao Y, Lai LL, Li JJ, Wang Z, Ying W, Ma L, Wang N, Zuo E, Yang H, Chen WJ, Base editing-mediated splicing correction therapy for spinal muscular atrophy. Cell Res. 30 (2020), pp. 548–550. [PubMed: 32210360] 146. Lapinaite A, Knott GJ, Palumbo CM, Lin-Shiao E, Richter MF, Zhao KT, Beal PA, Liu DR, Doudna JA, DNA capture by a CRISPR-Cas9-guided adenine base editor. Science (80-. ) 369, 566–571 (2020). 147. Nagaoka SI, Nakaki F, Miyauchi H, Nosaka Y, Ohta H, Yabuta Y, Kurimoto K, Hayashi K, Nakamura T, Yamamoto T, Saitou M, ZGLP1 is a determinant for the oogenic fate in mice. Science (80-. ) 367 (2020), doi:10.1126/science.aay5947. 148. Lee HK, Smith HE, Liu C, Willi M, Hennighausen L, Cytosine base editor 4 but not adenine base editor generates off-target mutations in mouse embryos. Commun. Biol 3, 1–6 (2020). [PubMed: 31925316] 149. Doman JL, Raguram A, Newby GA, Liu DR, Evaluation and minimization of Cas9-independent off-target DNA editing by cytosine base editors. Nat. Biotechnol 38, 620–628 (2020). [PubMed: 32042165] 150. Reichart D, Newby GA., Wakimoto H, Lun M, Gorham JM, Curran JJ, Raguram A, DeLaughter DM, Conner DA, Marsiglia JDC, Kohli S, Chmatal L, Page DC, Zabaleta N, Vandenberghe L, Liu DR, J. G. Seidman1, C. Seidman, Efficient in vivo Genome Editing Prevents Hypertrophic Cardiomyopathy in Mice. Nat. Med in press (2022).in press 151. Huang X, Guo H, Tammana S, Jung YC, Mellgren E, Bassi P, Cao Q, Tu ZJ, Kim YC, Ekker SC, Wu X, Wang SM, Zhou X, Gene transfer efficiency and genome-wide integration profiling of sleeping beauty, Tol2, and PiggyBac transposons in human primary t cells. Mol. Ther 18, 1803–1813 (2010). [PubMed: 20606646] 152. Briggs JA, Li VC, Lee S, Woolf CJ, Klein A, Kirschner MW, Mouse embryonic stem cells can differentiate via multiple paths to the same state. Elife. 6, e26945 (2017). [PubMed: 28990928] 153. Ryu SM, Koo T, Kim K, Lim K, Baek G, Kim ST, Kim HS, Kim DE, Lee H, Chung E, Kim JS, Adenine base editing in mouse embryos and an adult mouse model of Duchenne muscular dystrophy. Nat. Biotechnol 36, 536–539 (2018). [PubMed: 29702637] 154. Zettler J, Schütz V, Mootz HD, The naturally split Npu DnaE intein exhibits an extraordinarily high rate in the protein trans-splicing reaction. FEBS Lett. 583, 909–914 (2009). [PubMed: 19302791] 155. Lim CKW, Gapinske M, Brooks AK, Woods WS, Powell JE, Zeballos C. MA, Winter J, Perez- Pinera P, Gaj T, Treatment of a Mouse Model of ALS by In Vivo Base Editing. Mol. Ther (2020), doi:10.1016/j.ymthe.2020.01.005. 156. Banskota S, Raguram A, Suh S, Du SW, Davis JR, Choi EH, Wang X, Nielsen SC, Newby GA, Randolph PB, Osborn MJ, Musunuru K, Palczewski K, Liu DR, Engineered virus-like particles for efficient in vivo delivery of therapeutic proteins. Cell. 185, 250–265.e16 (2022). [PubMed: 35021064] 157. Lutz CM, Kariya S, Patruni S, Osborne MA, Liu D, Henderson CE, Li DK, Pellizzoni L, Rojas J, Valenzuela DM, Murphy AJ, Winberg ML, Monani UR, Postsymptomatic restoration of SMN Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 34 rescues the disease phenotype in a mouse model of severe spinal muscular atrophy. J. Clin. Invest 121, 3029–3041 (2011). [PubMed: 21785219] 158. Paushkin S, Gubitz AK, Massenet S, Dreyfuss G, The SMN complex, an assemblyosome of ribonucleoproteins. Curr. Opin. Cell Biol 14, 305–312 (2002). [PubMed: 12067652] 159. Gubitz AK, Feng W, Dreyfuss G, The SMN complex. Exp. Cell Res 296, 51–56 (2004). [PubMed: 15120993] 160. Chan YB, Miguel-Aliaga I, Franks C, Thomas N, Trülzsch B, Sattelle DB, Davies KE, van den Heuvel M, Neuromuscular defects in a Drosophila survival motor neuron gene mutant. Hum. Mol. Genet 12, 1367–1376 (2003). [PubMed: 12783845] 161. Szunyogova E, Zhou H, Maxwell GK, Powis RA, Francesco M, Gillingwater TH, Parson SH, Survival Motor Neuron (SMN) protein is required for normal mouse liver development. Sci. Rep 6 (2016), doi:10.1038/srep34635. 162. Chaytow H, Huang YT, Gillingwater TH, Faller KME, The role of survival motor neuron protein (SMN) in protein homeostasis. Cell. Mol. Life Sci 75 (2018), pp. 3877–3894. [PubMed: 29872871] 163. Blatnik AJ, McGovern VL, Le TT, Iyer CC, Kaspar BK, Burghes AHM, Conditional deletion of SMN in cell culture identifies functional SMN alleles. Hum. Mol. Genet 29, 3477–3492 (2020). [PubMed: 33075805] 164. Sivanesan S, Howell MD, Didonato CJ, Singh RN, Antisense oligonucleotide mediated therapy of spinal muscular atrophy. Transl. Neurosci 4 (2013), pp. 1–7. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 35 Fig. 1. Editing SMN2 regulatory regions. (A) Genomic SMN exons 6 to 8, and SMN mRNA and protein products. (B) Nuclease editing strategy and genome editing outcomes of ISS-N1 targeting (strategy A). The table shows combinations of six nucleases, paired with ten sgRNAs complementary to the top (A1-10) or bottom strand (A11-19) identified by arrows that show the DSB site of the sgRNAs relative to the sequence above. (C) Exon 7 inclusion in SMN mRNA after editing, as indicated, measured by automated electrophoresis. (D) SMN protein levels after editing, as indicated, normalized to histone H3. (E) Nuclease editing strategy targeting and genome editing outcomes of targeting the first five codons of exon 8 (strategy B). The table shows combinations of five nucleases, paired with nine sgRNAs complementary to the top (B1-12) or bottom strand (B13-16) identified by arrows that indicate their DSB site, as above. (F) Total SMN protein levels after editing. (G) Nuclease and cytosine base editing strategies and genome editing outcomes of 3’-splice acceptor disruption at exon 8 (strategy C). (H) SMN protein levels following C-nuc and C-CBE editing or treatment with risdiplam, normalized to histone H3. i) Distribution of SMN2 transcript variants after C-nuc and C-CBE editing. Experiments are performed in Δ7SMA mESCs, NT=no treatment, ≤0.05, ≤0.01, *≤0.005. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 36 Fig. 2. Adenine base editing of SMN2 C6T. (A) Adenine base editing of SMN2 C6T (strategy D). (B) Target nucleotide position within the protospacer (P#) for base editing. A typical base editor activity window is illustrated as a heat map. (C) The table shows ABE8e editing strategies with color-coded Cas-variant domains and their corresponding spacers. The protospacer position of the C6T target nucleotide (P#) is indicated. Graph shows genome editing outcomes in Δ7SMA mESCs. (D) Correlation of BE-Hive predicted editing outcomes with observed allele frequencies after base editing with ABE7.10 or ABE8e deaminases fused to different Cas variants. Pearson’s r is shown, 95% CI ranges are 0.9408–0.9998 for SpCas9, 0.5823–0.9201 for SpCas9 engineered and evolved variants, and 0.7557–0.9689 for SpyMac Cas variants. (E) Plot of base editing efficiency and single nucleotide editing precision of C6T by the indicated ABE and spacer combinations. (F) Exon 7 inclusion in SMN mRNA after editing by the indicated strategies, measured by automated electrophoresis. (G) SMN protein levels after editing by the indicated strategies, normalized to histone H3. (H) On-target and off-target base editing of strategy D10 in HEK293T cells. Bars show editing of the most frequently edited nucleotide at each locus, with the P# position shown in parenthesis. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 37 Fig. 3. Adenine base editing in Δ7SMA mice. (A) Dual-AAV vectors encoding split-intein ABE8e-SpyMac and P8 sgRNA cassettes, v6 AAV9-ABE8e. (B) Neonatal ICV injections in Δ7SMA mice with AAV9-ABE, and AAV9- GFP as a transduction control. (C-E) Immunofluorescence images of lumbar spinal cord sections from wild-type Δ7SMA mice at 25 weeks old, ICV injected on PND0-1 with AAV9-ABE, AAV9-GFP, or uninjected as indicated. GFP indicates transduction, ChAT labels spinal motor neurons in the ventral horn, NeuN labels post-mitotic neurons, GFAP labels astrocytes, DAPI stains all nuclei. (F) Quantification of GFP and ChAT double- positive cells within the ventral horn (n=3). (G) Base editing in bulk and GFP+ flow-sorted nuclei of Δ7SMA mice treated with AAV9-ABE+AAV9-GFP ( n=5), AAV9-GFP (n=4), or uninjected (n=3). (H) On-target and off-target editing following VIVO analysis of strategy D10 in Δ7SMA mESCs compared to AAV9-ABE+AAV9-GFP treatment in Δ7SMA mice. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 38 Bars show editing of the most frequently edited nucleotide at each locus, with the P# position shown in parenthesis. (I) Schematic of motor neuron differentiation (MND) and caudal-neural differentiation (CND) of Δ7SMA mESCs. (J) Whole transcriptome A-to-I RNA off-target editing analysis in Δ7SMAmESCs ( n=3), and CND (n=3) and MND (n=3) differentiated cells stably expressing the D10 strategy. Science. Author manuscript; available in PMC 2023 June 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Arbab et al. Page 39 Fig. 4. AAV9-ABE mediated rescue of Δ7SMA mice. (A) (Left) Motor unit number estimation (MUNE) and (Right) compound muscle action potential (CMAP) amplitude at PND12 in heterozygotes (n=11), and Δ7SMA mice treated with Zolgensma (n=5), AAV9-ABE (n=10), risdiplam (n=8), or uninjected (n=7). (B) Kaplan–Meier curve of Δ7SMA neonates ICV injected with Zolgensma from Robbins et al. 2014 (data extracted using PlotDigitizer). Average (av), median (md), and longest (lng) survival in days: untreated (avg-13, med-14, lng-15), PND2 (avg-187, med-204, lng-214), PND3 (avg-102, med-75, lng-182), PND4 (avg-141, med-167, lng-211), PND5 (avg-76, med-37, lng-211), PND6 (avg-73, med-34, lng-211), PND7 (avg-30, med-28, lng-70), and PND8 (avg-18, med-18, lng-22).). (C) Kaplan-Meier curve in AAV9-ABE treated (n=6) and uninjected (n=8) Δ7SMA mice. (D) Neonatal ICV co-injections with AAV9-ABE, AAV9- GFP, and nusinersen. (E) (Left) The time required for Δ7SMA mice to right themselves in the righting reflex assay at PND7. (Right) The hang time of Δ7SMA mice in the inverted screen test at PND25. (F) Analysis of voluntary movement by open field tracking at PND40. (Left) Traveled distance in cm. (Right) Velocity in cm/s. (G, H) Body weight in grams and Kaplan-Meier curve of Δ7SMA mice. Graph line shading represents (G) standard deviation or (H) 95% CI. Animals are treated as indicated. Dots represent individual animals, ≤0.02, ≤0.01, ≤0.005, ***≤0.001. Science. Author manuscript; available in PMC 2023 June 15.
10.1093_nar_gkad501	7438–7450 Nucleic Acids Research, 2023, Vol. 51, No. 14 https://doi.org/10.1093/nar/gkad501 Published online 9 June 2023 Rarely acquired type II-A CRISPR-Cas spacers mediate anti-viral immunity through the targeting of a non-canonical PAM sequence Claire T. K enne y 1 and Luciano A. Marrafﬁni 1 , 2 , * 1 Laboratory of Bacteriology, The Roc k efeller University, 1230 York Ave, New York, NY 10065, USA and 2 Ho w ard Hughes Medical Institute, The Roc k efeller University, 1230 York Ave, New York, NY 10065, USA Received December 23, 2022; Revised May 23, 2023; Editorial Decision May 23, 2023; Accepted June 07, 2023 ABSTRACT GRAPHICAL ABSTRACT The Streptococcus pyogenes type II-A CRISPR-Cas systems pr o vides adaptive imm unity thr ough the ac- quisition of short DNA sequences fr om inv ading vi- ral genomes, called spacer s. Spacer s are transcribed into short RNA guides that match regions of the vi- ral genome f ollowed b y a conserved NGG DNA mo- tif, known as the PAM. These RNA guides, in turn, are used by the Cas9 nuclease to ﬁnd and destr o y complementar y DNA targ ets within the viral g enome. While most of the spacers present in bacterial popu- lations that survive pha ge inf ection target protospac- ers ﬂanked by NGG sequences, there is a small frac- tion that target non-canonical PAMs. Whether these spacers originate through accidental acquisition of phage sequences and / or pr o vide efﬁcient defense is unknown. Here we found that many of them match phag e targ et regions ﬂanked by an NAGG PAM. De- spite being scarcely present in bacterial populations, NAGG spacers pr o vide substantial imm unity in vivo and generate RNA guides that support r ob ust DNA c leav age b y Cas9 in vitr o ; with both activities com- parable to spacers that target sequences followed by the canonical AGG PAM. In contrast, acquisition ex- periments showed that NAGG spacers are acquired at very low frequencies. We therefore conclude that discrimination against these sequences occurs dur- ing immunization of the host. Our results reveal un- expected differences in PAM recognition during the spacer acquisition and targeting stages of the type II-A CRISPR-Cas immune response. INTRODUCTION Cluster ed, r egularly interspaced, short, palindromic r epeats (CRISPR) loci and their associated ( cas ) genes protect bac- teria and archaea from viral (phage) predation through a mechanism that confers adapti v e immunity to these organ- isms ( 1 ). The CRISPR-Cas response to phage infection is divided into two phases. In the immunization phase, Cas integr ases extr act a short fragment of the invading phage DNA, known as a ‘spacer’, and insert it between the re- peats of the CRISPR locus ( 2 ). This allows the bacterial host to collect a memory of the infection that will be used during the second phase of the CRISPR response, known as the targeting phase, to recognize the phage in a subse- quent exposure. This is achieved first through the transcrip- tion and processing of the spacer into a short RNA guide ( 3–5 ), known as the CRISPR RN A (crRN A), w hich forms a ribonucleoprotein complex with crRNA-guided Cas nu- cleases ( 6–9 ). These nucleases use the crRNA to find com- plementary nucleic acids (known as protospacers) of the * To whom correspondence should be addressed. Tel: +1 212 327 7014; Email: [email protected] C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. invading phage and cleave them, thus limiting phage repli- cation and propagation ( 10–12 ). Depending on the cas gene content, CRISPR-Cas sys- tems are classified into six different types ( 13 ) that di- verge in the general molecular mechanism of immunity de- scribed above. Type II CRISPR-Cas systems are defined by the presence of the signature gene cas9 , which encodes a crRNA-guided nuclease that plays key roles in both the im- munization and targeting phases of immunity. Spacers in- corporated into the type II CRISPR locus match phage re- gions that are flanked by a conserved sequence, known as the protospacer-adjacent motif (PAM) ( 14 , 15 ). For one of the most studied of these systems, the type II-A system of Str eptococcus pyog enes SF370, the PAM sequence is NGG ( 4 , 16 ). Spacer acquisition in this system r equir es Cas9, the Cas1 / Cas2 integrase and the DNA-sliding hexameric com- plex formed by Csn2 ( 17 , 18 ). These form a ‘supercomplex’ that selects and integrates DNA sequences from invading phages ( 19 ), preferably from free DNA ends ( 19–22 ). Cas9 plays two different roles in this process. First, it uses its PAM binding-recognition domain to select phage DNA frag- ments that contain NGG motifs ( 17 ). This recognition is achie v ed by two arginine residues, each of which establishes sequence-specific hydrogen-bonding interactions with one of the guanosines of the GG PAM dinucleotide ( 23 ). Sec- ond, Cas9 performs the nucleolytic cleavage of these phage DNA fragments to 30 nucleotides, the length of the spacers in the type II-A CRISPR array ( 19 ). During targeting, S. pyogenes Cas9 uses the crRNA- guide to locate a complementary sequence in the phage DNA and cleave it ( 7 ). Target recognition is first achie v ed by searching for GG dinucleotides on the phage DNA ( 24 ). Cas9 interaction with a potential PAM results in the partial melting of the DNA upstream of the motif , which allo ws an a ttempt a t annealing the crRNA to the bottom strand of the target DNA ( 24 ). First a region of 6–8 nucleotides known as the seed sequence is paired. If this RN A:DN A hybrid is formed, then the rest of the crRNA sequence is paired to the single stranded DN A (ssDN A) target ( 24 ). When a full R-loop structure is formed due to pairing, the two nucle- ase domains of Cas9, HNH and RuvC, each cleave a DNA strand of the target, generating a double strand break in the viral DNA ( 7 ). While this break is genotoxic for the phage and pre v ents its replica tion and propaga tion ( 10 ), the mechanism has weaknesses that can be exploited to cir- cumvent type II-A CRISPR-Cas immunity. Mutant phages present in the population (usually at a very low frequency and known as ‘escapers’) contain nucleotide modifications in either the PAM or seed sequence, and can thereby escape Cas9 cleavage and kill the host cell ( 15 , 22 ). In addition to its role during the targeting phase of type II-A immunity, Cas9’s nuclease activity has been repurposed to cleave ge- nomic DNA of eukaryotic cells for the de v elopment of revo- lutionary, sequence-specific gene editing techniques ( 25 , 26 ). Due to Cas9’s ability to recognize NGG PAM sequences during both the immunization and targeting phases, the vast majority of spacers present in bacterial populations that survi v e phage infection target protospacers flanked by these sequences, and only a very small fraction (0.03%) tar- get non-canonical PAMs ( 17 ). Whether these spacers orig- inate by accidental acquisition of phage sequences and / or Nucleic Acids Research, 2023, Vol. 51, No. 14 7439 provide efficient defense is unknown. Here, we investigated these rare spacers and found that many of them match phage target regions flanked by an NAGG PAM. We deter- mined that, despite being scar cely pr esent in bacterial pop- ulations, NAGG spacers provide substantial immunity in vivo and generate crRNA guides that support robust DNA cleav age b y Cas9 in vitro ; with both acti vities comparab le to spacers that target sequences followed by the canonical AGG PAM. This is also the case for NAGG spacers that are not found in the spacer r epertoir e of surviving bacterial populations. In contrast, acquisition experiments showed that NAGG spacers are acquired at very low frequencies. We ther efor e conclude tha t discrimina tion against these se- quences occurs during immunization of the host. Our re- sults re v eal une xpected differ ences in PAM r ecognition dur- ing the two stages of the type II-A CRISPR-Cas immune response. MATERIALS AND METHODS Plasmid construction The plasmids used in this study are listed in Supple- mentary File 2, as are the sequences of oligonucleotides used in this study. For all experiments, some or all of the S. pyogenes type II-A CRISPR-Cas locus was cloned into staphylococcal vector plasmids, which were expressed within Staphylococcus aureus RN4220 cells. The assays in- vestigating targeting used S.aureus RN220 cells that carried plasmids deri v ed from the parent plasmid, pDB114, which was modified to contain a r epeat-spacer-r epeat CRISPR ar- ray with a defined spacer. To create these plasmids, we used a restriction digest-based cloning approach previously de- scribed ( 27 ). In short, the parent plasmid pDB441, contains a chloramphenicol-resistance cassette, tracr , cas9 , leader se- quence, and two CRISPR repeats flanking a 30 bp sequence housing two bsaI restriction sites. pDB114 was mixed with the BsaI-HFv2 restriction enzyme (NEB) and incubated ◦C for 4–6 h. Two IDT single stranded DNA (ss- at 37 DNA)oligonucleotides (oligos) (a “forward” DNA oligo and a “re v erse” DNA oligo) were phosphorylated with ◦C for 30-60 min. After phosphorylation, PNK (NEB) at 37 the oligos were annealed by adding NaCl and incubating for ◦C, then allowing the reaction to cool to room 5 min at 98 temperature ( ∼2.5 h). The annealed double-stranded DNA (dsDNA) oligos were diluted 1:10 in nuclease-free water and ligated to the digested plasmid in a 20 (cid:2)l reaction at room temperature overnight. Reactions were drop dialyzed for 30 min in ddH 2 O and transformed into electrocompetent S. aureus RN4220 cells. 500 (cid:2)l BHI was added to the transformed cells, which were ◦C for 1–2 h with agita tion, spread onto then incuba ted a t 37 agar plates of BHI and 10 (cid:2)g / ml chloramphenicol, and ◦C . Resulting colonies were con- incuba ted overnight a t 37 firmed for correct plasmid construction by dissolving in 20 (cid:2)l colony lysis buffer (25 mM HEPES, 150 mM KCl, 5% (v / v) glycer ol, 20% sucr ose) ( 28 ), and 1 (cid:2)l l ysosta phin (100 (cid:2)g / mL final, Ambi Products LLC), incubating for 20 min at ◦C , amplifying the CRISPR ar- 37 ray via PCR using one of the IDT ssDNA oligos, and send- ing the samples for Sanger sequencing. ◦C and then 10 min a t 98 7440 Nucleic Acids Research, 2023, Vol. 51, No. 14 Bacterial strains and growth conditions Growth of of S. aureus RN4220 ( 29 ) cultures was carried out in Bacto brain-heart infusion (BHI) broth medium at ◦C with agitation at 220 RPM. Liquid experiments were 37 carried out in 3 ml BHI medium in 15-ml conical tubes un- less otherwise noted. Where v er applicab le, media was sup- plemented with 10 (cid:2)g / ml chloramphenicol to ensure main- tenance of plasmids deri v ed from the chloramphenicol- resistant staphylococcal vector plasmid, pC194 ( 30 ). This includes pDB114, its deri vati v e plasmids, pGG32- (cid:2) trL, and pWJ40 (described below). For RN4220::pKL55-iTET- B1 strains, which contain the kanamycin resistance gene in- tegrated into the RN4220 chromosome ( 31 ), BHI was also supplemented with 25 (cid:2)g / ml kanamycin for chromosomal marker selection. See Supplementary File 2 for a full list of bacterial strains , plasmids , and oligonucleotides used in this study. Bacteriophage propagation The previously constructed, lytic S. aureus phage, F NM4 (cid:3) 4 ( 17 ), and mutant phages deri v ed from F NM4 (cid:3) 4, were used throughout the study. Phage strains were amplified by first launching overnight cultures of S. aureus RN4220 cells. The f ollowing da y, the culture was diluted 1:100 in fresh BHI, supplemented with 5 mM CaCl 2 and the appropriate an- ◦C with agitation to an optical cell tibiotic, and grown at 37 density (OD 600 ) of 0.2–0.6 (about 1 h and 10 min). The cul- ture was rediluted to a normalized OD 600 of 0.2, phage was added at a multiplicity of infection (MOI) of 0.1, and the cultur es wer e grown for an additional 4 h with agitation ◦C . Then, cultur es wer e spun down for 5 min at 4300 a t 37 RPM. The lysates were filtered through 0.45 (cid:2)m syringe fil- ◦C . Plaque forma- ters (Acrodisc) and stored in BHI a t 4 tion assays were conducted to assess the number of infec- tious phage particles in the resulting stocks. In subsequent phage infection assays, an MOI = 1 was used unless oth- erwise noted, and BHI media was supplemented with the appropriate antibiotic and 5 mM CaCl 2 to facilitate phage adsorption. No escapers isolated from NAGG spc strains or AGG spc strains had A2 mutations. To obtain such phage mu- tants, plasmids were constructed, containing a ∼1 kb ho- mology with the phage that included an A to C mutation in the NAGG of interest, and a G to C mutation in the G3 of an upstream, non-essential NGG PAM. WT F NM4 ɣ 4 was passaged on the resulting strain on soft agar. To isolate re- combinant phages, plaques were picked and re-passaged on soft agar to single plaques. Phages from those plaques were then passaged on strains containing plasmids with a spacer targeting only the upstream protospacer with the mutated NCG PAM. The phages isolated from the resulting plaques were amplified and sequenced to confirm the A to C muta- tion in the NAGG of interest. Pr epar ation of electrocompetent S. aureus RN4220 strains Electrocompetent S. aureus RN4220 cells were prepared and transformed with DNA plasmids using a previously described method ( 31 ), with the exception of using BHI medium instead of TSB medium. Plasmid miniprep Unless otherwise indicated, 1–3 ml of an overnight S. au- reus RN4220 cultur e, wer e pelleted and resuspended in 250 (cid:2)l Buffer P1. 10 (cid:2)l Lysostaphin (Ambi Products LLC, 100 (cid:2)g / ml final) was added, and the cells were incubated with ◦C for 30–60 min. Following lysis, plasmids agita tion a t 37 were isolated using the QIAGEN Spin Miniprep kit accord- ing to the manufacturer’s protocol, beginning with addition of P2. DNA was eluted from each column in 40 (cid:2)l Millipore water. Enumeration of colony forming units (CFU) To determine the concentration of bacteria that survi v ed phage infection in liquid culture, ten-fold dilutions of S. au- reus RN4220 strains were spotted on 50% BHI agar plates supplemented with the appropriate antibiotic. The plates ◦C and colon y-f orming units were incubated overnight at 37 (CFU) were enumerated the next day ( 20 ). Phage mutant construction We employed phage recombination techniques using plas- mids to create F NM4 (cid:3) 4 phages with point mutations in different nucleotides of the NAGG PAMs (the A2, G3 or G4 of the N1-A2-G3-G4 sequence) corresponding to spc11, spc14, spc15 and spc17 ( 22 ). See Supplementary File 2 for a full list of plasmids and oligonucleotides used to generate phage mutants in this study. The phages with a mutation in G3 or G4 were isolated as spontaneous escaper plaques following infection of WT F NM4 (cid:3) 4 on a soft agar lawn of S. aureus RN4220 cells carrying plasmids with the NA GG or A GG version of the spacer of interest. Single plaques were isolated and re- passaged to single plaques on lawns of the same strain in a plaque formation assay (see protocol below). Phage DNA ◦C, 10 min), PCR am- was extracted by boiling the phage (98 plified, and Sanger sequenced to identify the plaques with phage harboring a mutation of interest. Enumeration of plaque forming units (PFU) Phage titer assays were performed as previously described ( 31 ). In brief, serial dilutions of the F NM4 (cid:3) 4 or mutant F NM4 (cid:3) 4 stock wer e pr epar ed in triplicate and spotted on fresh 50% BHI top agar lawns containing the appropri- ate antibiotic, 5 mM CaCl 2 and RN4220 cells harboring a type II-A CRISPR plasmid targeting the phage. Plates were ◦C overnight, and resulting plaques were incuba ted a t 37 enumerated the next day to calculate plaque forming units (PFU). Gr owth curv es of bacterial infection Bacterial infection growth was measured as previously de- scribed ( 31 ) with minor alterations. Overnight cultures were launched and the following day, they were diluted 1:100 in fresh BHI, supplemented with 5 mM CaCl 2 and the ap- ◦C with agitation to pr opriate antibiotic, and gr own at 37 OD 600 = 0.2–0.6 (about 1 h and 10 min). The culture was normalized to OD 600 = 0.2, in triplicate. Phage was added at an MOI = 1 (unless noted otherwise), and the cultures were loaded into a 96-well plate (Cellstar). The plate was incu- ◦C with agita tion for 18–24 h, with OD measure- ba ted a t 37 ments taken e v ery ten min (by a Tecan Infinite 200 PRO) to generate growth curves. In vitro cas9 target cleavage assay In vitro cleavage by Cas9 was tested following a previ- ously published protocol ( 22 ) with minor alterations. A 1 kb double stranded DN A (dsDN A) target, containing the spacer and PAM of interest, was generated by mixing 20 (cid:2)l colony lysis buffer with 1 (cid:2)l F NM4 (cid:3) 4 stock and heat- ◦C for 10 min. Samples were spun for ing the mixture to 98 1–2 min at 15000 RPM. Four PCRs were run for each sam- ple, each with 1 (cid:2)l of the supernatant of the spun phage DNA and 2.5 (cid:2)l of each 10 (cid:2)M forward and re v erse oli- gos. The oligos used were oCK124 and oCK125 for spc11 , oCK126 and oCK127 for spc14 , oCK028 and oCK029 for spc15 , and oCK140 and oCK131 for spc17 . For each sample, the four PCRs were combined and run through one DNA Clean & Concentrator TM -5 column in the Zymo Research kit, according to the manufacturer’s protocol. DNA was eluted in 6 (cid:2)l nuclease-free H 2 O and diluted to 100 nM. crRNAs were ordered from IDT to correspond with ei- ther the NAGG or AGG versions of spc11, spc14, spc15 and spc17 (RNA oligos listed in Supplementary File 2). Each was diluted to 100 (cid:2)M in IDTE buffer. 1 (cid:2)l of crRNA was mixed at a 1:1 ratio with 1 (cid:2)l 100 (cid:2)M tracrRNA (IDT), and 8 (cid:2)l DNA duplex buffer (NEB) to form tracrRN A:crRN A ◦C for 5 min RNA duplex es. The mixtur e was heated to 95 ◦C per minute, until and cooled via thermocycler at −0.6 ◦C, amounting to ∼2.5 h). reaching room temperature (20 The RNA duplex es wer e diluted 1:10 in nuclease-free water and used immediately in the cleavage reactions. Cleavage assays were performed much as described pre- viously ( 22 ) at Cas9:crRNA:tr acrRNA concentr ations of 0, 6.25, 12.5, 25, 50, 100 and 200 nM. Briefly, 20 (cid:2)M Cas9 (NEB) was diluted to 1 (cid:2)M by mixing 1.5 (cid:2)l 20 (cid:2)M Cas9, 3 (cid:2)l 10 × NEB Cas9 reaction buffer (NEB), and 25.5 (cid:2)l nuclease-free H 2 O. Then the 1 (cid:2)M Cas9 was mixed 1:1 with 1 (cid:2)M tracrRN A:crRN A (50 (cid:2)l nuclease-free H 2 O, 12.5 (cid:2)l 10 × NEB Cas9 reaction buffer, 25 (cid:2)l 1 (cid:2)M Cas9 and 25 (cid:2)l 1 (cid:2)M tracrRN A:crRN A) and incubated at room tem- perature for 10 min. Six serial, 2-fold dilutions were made by diluting the resulting ribonucleoprotein (RNP) mixture in nuclease free H 2 O. 5 (cid:2)l of the 100 nM DNA substrate was incubated with 45 (cid:2)l of the RNP dilutions for a total reaction volume of 50 (cid:2)l and a final DNA concentration of ◦C for 5 min be- 10 nM. All reactions were incubated at 37 fore being quenched with 1 (cid:2)l proteinase K (NEB). Samples wer e stor ed a t –20 ◦C . For analysis, samples were brought to room temperature and diluted 10-fold in nuclease-free H 2 O. Cleavage prod- ucts were visualized and quantified by automated gel elec- trophoresis and imaging using a Tapestation 4200 bioana- lyzer (Agilent), and following the manufacturer’s protocol for the Agilent D500 ScreenTape Assay (Agilent). Nucleic Acids Research, 2023, Vol. 51, No. 14 7441 NAGG and AGG spacer library generation To generate the NAGG and AGG spacer libraries, a pre- viously published protocol was used ( 32 ), with the fol- lowing modifications. Two libraries of CRISPR plasmids deri v ed from pDB114 were generated, one containing all 787 spacers deri v ed from all NAGG PAMs that are found in F NM4 (cid:3) 4 (NAGG spacer library), and the other con- taining all 787 spacers deri v ed from all AGG PAMs in F NM4 (cid:3) 4 (AGG spacer library). For each NAGG and AGG spacer library, 787 oligonucleotides (85 nucleotides in length), wer e pur chased from Twist Biosciences. Each oligo contained a unique F NM4 (cid:3) 4-ma tching spacer, repea t ho- mology, bsaI sites, and uni v ersal priming sites. The library was made double-stranded by PCR and the product was purified by phenol-chloroform e xtraction. Indi vidual spac- ers from each annealed oligo were introduced into pDB114 via Golden Gate cloning with BsaI-HFv2 (NEB) and T7 DNA ligase (NEB), then electroporated into RN4220 elec- trocompetent cells. To obtain at least 10 × coverage for each spacer in the library, 8000 colonies were pooled, pelleted, resuspended in 1 ml 10% DMSO, aliquoted, and stored at ◦C. Complete coverage of the NA GG and A GG spacer -80 libraries was confirmed via next generation DNA sequenc- ing, as described below. Next generation DNA sequencing Plasmid DNA was extracted from liquid cultures after phage infection experiments (see protocols above) and used as a template for Phusion PCR to amplify the CRISPR ar- ray. After bands of the CRISPR array were extracted and purified from a 2% agarose gel, samples wer e pr epar ed for sequencing with the TrueSeq Nano DNA Library Prep pro- tocol (Illumina) and subject to Illumina MiSeq sequenc- ing. Data analysis was performed in Python, similar to that described previously ( 33 ). In short, the spacer sequences were extracted and the number of reads for each spacer was recorded from the raw Illumina FASTQ files. For each unique spacer sequence matching the F NM4 (cid:3) 4 genome, various characteristics were determined and recorded, in- cluding the spacer’s frequency within the sequencing pop- ulation (number of reads of that spacer compared to total spacer reads sequenced), the position of its protospacer in the F NM4 (cid:3) 4 genome, the strand of the genome on which the protospacer is found, and the PAM sequence flanking the protospacer (see Supplementary File 1). Spacer acquisition from electroporated dsDNA oligonu- cleotides To study the acquisition of spacers from transformed DNA oligos, electrocompetent S. aureus RN4220 cells were made by the protocol abov e, e xcept that the overnight culture was diluted 1:500 (400 (cid:2)l overnight culture, 200 ml BHI, 200 (cid:2)l cm10), resulting in about 2–2.5 h to regrow to OD 600 = 0.4. To further increase the cells’ competence, 0.5 M sucrose was used to perform three washes, instead of two washes in 10% glycerol ( 34 ). The RN4200 strain carried the plasmid, pGG32- (cid:2) trL, that contains a CRISPR array of a single re- peat, and has the long tracr promoter sequence deleted. This deletion results in an increase rate of spacer acquisition ( 27 ), 7442 Nucleic Acids Research, 2023, Vol. 51, No. 14 which is otherwise relati v ely inefficient from transformed dsDNA ( 35 ). To create dsDNA oligos to transform into the pGG32- (cid:2) trL strain, 63 bp amplicons from F NM4 (cid:3) 4 were gener- ated using forward and reverse ssDNA oligos (IDT). They were designed to only contain one spacer and PAM of in- terest. All other NGG sequences outside of the spacer and PAM wer e alter ed to NTG. The oligo pairs used are shown in Supplementary File 2. The pairs were diluted to 1 mM in Duplex Buffer (IDT), and then annealed by first mixing at a 1:1 ratio (20 (cid:2)l 1 mM forward primer, 20 (cid:2)l 1 mM re- verse primer, and 4 (cid:2)l 1M NaCl). The mixture was heated ◦C per ◦C for 5 min and cooled via thermocycler at -0.6 to 95 ◦C, amount- minute until reaching room temperature (20 ing to ∼2.5 h). The annealed DNA samples were dialyzed in ddH 2 O and 100 (cid:2)g of dsDNA was electroporated into competent RN4220 pGG32- (cid:2) trL cells, as described previ- ◦C with ously ( 35 ). Cells were recovered for 2h in BHI at 37 agitation. Plasmids from the transformed staphylococci were iso- lated via the miniprep protocol described above. The CRISPR array was then PCR amplified using primers oN A169 and oN A170 (w hich are highl y sensiti v e for the re- peat sequence), and the PCR reactions were purified via the QIAquick PCR Purification Kit, following its correspond- ing protocol. To maximize yield, the amplified CRISPR ar- ray was isolated by automated gel electrophoresis using a PippinHT machine (model HTG-3010), following the man- ufacture’s PippinHT Quick Guide protocol. For precise DNA size selection, a PippnHT 3% cassette was used with a timed protocol set for sample extraction between 26 and 35 min ( 34 ). The eluates extracted from the PippinHT cassette were then sequenced on an Illumina MiSeq, as described above. Sequence analysis was done in Python to determine the number of spacers acquired from the NNGG, vs the NGG PAMs, on the transformed dsDNA oligos (see above description). Statistics and reproducibility All experiments were independently reproduced three times, unless otherwise sta ted. Sta tistical tests were carried out in GraphPad Prism 9.4.1. No statistical methods were used to predetermine sample size. RESULTS Spacers targeting phage sequences flanked by NAGG PAMs can mediate type II-A CRISPR-cas immunity Pre vious wor k from our lab measured spacer acquisi- tion by a S. pyogenes type II-A CRISPR-Cas system (Supplementary Figure S1A). This locus was cloned into the chloramphenicol-resistant staphylococcal vector pC194 ( 30 ) to generate plasmid pWJ40 ( 31 ) and introduced into S. aureus RN4220 ( 29 ). Bacterial cultures were infected with the lytic staphylococcal phage F NM4 (cid:3) 4 ( 17 ), and next gen- eration sequencing was used to determine the sequence and frequency of the phage DNA inserted into the CRISPR array ( 17 ). In these studies, we detected a low frequency of reads for spacers that match regions of the phage not flanked by the NGG PAM ( 17 ). In order to evaluate whether these sequences correspond to functional spacers, we evalu- ated their ability to provide immunity to the bacterial host. We decided to study sequences that displayed at least 10 reads (Supplementary Table S1). First, we confirmed the ab- sence of a canonical PAM through Sanger sequencing of the F NM4 (cid:3) 4 targets (data not shown). Next, we cloned each spacer into pDB114, a plasmid that lacks the cas1 , cas2 , and csn2 genes involved in spacer acquisition and contains a modified CRISPR array with a single engineered spacer ( 36 ) (Supplementary Figure S1B). Cultures of staphylococci harboring the different plasmids were then infected with F NM4 (cid:3) 4 at a multiplicity of infection (MOI) of 1, and their growth was followed by monitoring optical density at 600 nm (OD 600 ) for 18 h (Figure 1 A). We found that spac- ers 11 and 17 provided full protection (similar to our posi- ti v e control, pRH079, a plasmid deri v ed from pDB114 that harbors a functional, canonical spacer ( 17 )), and spacers 14 and 15 provided partial protection. We also tested the set of spacers for their ability to limit phage propagation by seed- ing F NM4 (cid:3) 4 on lawns of staphylococci harboring pDB114 plasmids programmed with different spacers (Supplemen- tary Figure S1C). The plaque forming units (PFU) were enumerated to calculate the efficiency of plaquing (EOP) for each spacer, relati v e to the PFUs obtained on lawns of bac- teria harboring pDB114, without a targeting spacer (Figure 1 B). Consistent with our observations of bacterial growth, the presence of spacers 11 and 17 dramatically reduced the EOP by fiv e or ders of magnitude, and spacers 14 and 15 by a pproximatel y 3–4 orders of magnitude. Interestingl y, spacer 18, which did not protect bacterial cultures (Figure 1 A), was able to limit phage production by a pproximatel y 4 orders of magnitude in the plaque assay. We looked for sequence similarities in the PAM region of the targets of spacers 11, 14, 15, 17 and 18 and found a con- served NAGG motif (Supplementary Table S1). In order to determine whether this motif is r equir ed for immunity and if so, which of these nucleotides (A2, G3 or G4 of the N1- A2-G3-G4 sequence) is important for target recognition by Cas9, we isolated phages that can escape the immunity me- diated by spacers 11, 14, 15 and 17 from our plaque as- says (Supplementary Figure S1C) and sequenced the target region. We found that escapers harbored mutations in ei- ther the seed or PAM regions of the target that are known to be detrimental for efficient Cas9 cleavage and immunity against F NM4 (cid:3) 4 ( 22 ). Interestingly, in all cases, PAM mu- tations were observed only in G3 (Supplementary File 1). To corroborate this finding, we measured the EOP of se- lected escapers containing a T nucleotide in this position of the motif and, expectedly, we obtained values close to 1; i.e. their respecti v e spacers were not ab le to provide im- munity (Figure 1 C). Since w e w ere unable to isolate phages carrying mutations in A2 or G4, we decided to engineer sub- stitutions of these nucleotides, A2C and G4T. NAGT mo- tifs had a detrimental effect in the targeting of the ‘weaker’ spacers ( 14 ) and ( 15 ), but not on the ‘stronger’ spacers 11 and 17 (Figure 1 C). NCGG motifs flanking the spc15 and spc17 targets, on the other hand, were still able to support Cas9 targeting (Figure 1 C). Unfortunately, w e w ere unable to mutate A2 in the targets of spacers 11 and 14 (most likely the mutant phages are not viable). Altogether, these da ta demonstra te spacers ma tching protospacers followed Nucleic Acids Research, 2023, Vol. 51, No. 14 7443 Cas9 can efficiently cleave DNA targets f ollow ed by a non- canonical NAGG PAM To further investigate the immunity mediated by NAGG spacers, we evaluated the ability of Cas9 to cleave their targets, in vitro . As positi v e controls we used the canoni- cal spacers for these targets; i.e. we shifted the spacer se- quence one nucleotide to include the ‘N’ in the proto- spacer, and thus have a canonical AGG PAM (Figure 2 A– D). We indicated this shift with an asterisk (for example spc11 targets the NAGG PAM; spc11* targets the AGG PAM). These spacers mediated full immunity in vivo (Sup- plementary Figure S2A). We then designed crRNAs corre- sponding to those originated by each of the different spac- ers (see Supplementary File 2 for sequences) and associ- ated them with Cas9 and tracrRNA to obtain ribonucle- oprotein complexes. The target phage DNA corresponding to each NAGG spacer and AGG spacer pair (1 kb, with the protospacer sequence in the middle) was amplified via PCR, purified, and used as a substrate for the Cas9 com- plex. DNA cleavage at increasing concentrations of the nu- clease was quantified using a Ta pestation bioanal yzer (Sup- plementary Figure S2B–E) and plotted (Figure 2 A-D). We found that with the exception of spc14 crRNA, which medi- ated ∼ 50% of the cleavage mediated by the spc14* crRNA, all other crRNAs targeting an NAGG PAM substrate con- veyed similar cleavage as their counterpart guides that di- rect AGG PAM recognition. These results suggest that Cas9 can recognize and cleave targets harboring an NAGG non- canonical PAM. NAGG-spacers mediate an immune response comparable to that of canonical spacers The above results suggest that NAGG-spacers are able to support considerab le le v els of both immunity and Cas9 cleavage. In order to assess their phage targeting efficiency mor e accurately, we dir ectly compar ed them with their canonical counterparts in a competition experiment. To this end, we introduced a kanamycin-resistant cassette into the S. aureus RN4220 genome to use as host for the pDB114 plasmid harboring spacer 11, 14, 15* or 17. The resulting AGG spacer strains and their correspond- ing NAGG spacer strains (kanamycin and chlorampheni- col resistant, and only chloramphenicol resistant, respec- ti v ely) were mixed at a 1:1 ratio and the cultures were infected with F NM4 (cid:3) 4. After 24 h of growth, samples were plated on agar containing either chloramphenicol only or kanamycin and chloramphenicol, to enumerate colony forming units (CFU) and calculate the NA GG / A GG ra- tio for each spacer pair. Mixed cultures that were not infected with phage were used as controls. For all four spacers, we found that that this ratio did not significantly change from 0.5 (Figure 2 E), a result that suggests that the NAGG spacers provide similar immunity as the canonical ones. Non-acquired NAGG spacers provide efficient immunity Gi v en that the four NAGG spacers that we identified in spacer acquisition assays are able to provide substantial Figure 1. Spacers targeting phage sequences flanked by NAGG PAMs can mediate type II-A CRISPR-Cas immunity. ( A ) Cell survival, measured as the OD 600 values after F NM4 (cid:3) 4 infection of cultures carrying plasmids harboring the type II-A CRISPR-Cas locus of S. py ogenes , pro grammed with different spacers that target protospacers flanked by non-canonical PAMs, as well as a non-targeting control (pDB114) and a plasmid with a canonical spacer (pRH079). The av erage curv es of thr ee differ ent r epli- cates are shown, with ± standard deviation (StDev) values shown in lighter colors. ( B ) Propagation ability of (cid:3) NM4 (cid:3) 4 on staphylococci harboring the different plasmids described in (A), measured as efficiency of plaquing (EOP). Mean ± StDev values of three independent experiments are shown. ( C ) Same as in (B) but using cultures harboring spacers 11, 14, 15 or 17, and infecting with mutant phages with substitutions in different nucleotides of the NAGG PAM sequence. by NAGG sequences (her etofor e called ‘NAGG spacers’) can be acquired during the type II-A CRISPR-Cas response and that these NAGG sequences constitute a non-canonical PAM that can support efficient Cas9-mediated phage de- fense, with the central G3 nucleotide being essential for immunity. 7444 Nucleic Acids Research, 2023, Vol. 51, No. 14 Figure 2. Cas9 can efficiently cleave DNA targets followed by a non-canonical NAGG PAM. ( A ) In vitro cleavage assay of a ∼1 kb PCR product containing the spc11 and spc11 target DNA sequences, incubated with increasing concentrations of a 1:1:1 mix of Cas9:tracrRN A:crRN A: 0, 6.25, 12.5, 25, 50, 100 and 200 nM. Substrates and cleavage products were separated and quantified using a Ta pestation bioanal yzer. Mean ± StDev values of three independent experiments are shown. ( B ) Same as (A) but testing spc14 and spc14* target and crRNA sequences. ( C ) Same as (A) but testing spc15 and spc15* target and crRNA sequences. ( D ) Same as (A) but testing spc17 and spc17* target DNA and crRNA sequences. ( E ) Growth competition of staphylococci carrying pDB114 harboring an NAGG spacer or its corresponding canonical AGG spacer and different antibiotic resistance cassettes (chloramphenicol-resistant for NAGG spacers, kanamycin- and chloramphenicol-resistant for AGG spacers), mixed at a 1:1 ratio in the absence of antibiotics and infected with (cid:3) NM4 (cid:3) 4 for 24 h. For each NA GG-A GG spacer pair, the ratio of CFU of the NAGG strain to the AGG strain was determined at 0 and 24 h by plating on selecti v e agar plates. Mean ± StDe v values of three independent e xperiments are shown. P v alues obtained b y a Student’s t -test are shown. immunity, we decided to investigate four NAGG spacers that were not detected in our previous spacer acquisition ex- periments. We chose spacers for which the adjacent canoni- cal spacer (which directs Cas9 to an AGG PAM) displayed either a low (spacers 36 and 25) or a high (spacers 26 and 27) acquisition frequency (Supplementary Table S1). We found that, while spc36 mediated a low, yet significant, le v el of defense (a reduction in EOP of two orders of magni- tude), spacers 25, 26 and 27 provided an immune response as strong as a control spacer targeting a canonical AGG PAM (Figure 3 A and Supplementary Figure S3A). This result suggests that many of the NAGG spacers that are not detected in our spacer acquisition assays can convey efficient immunity. To test this hypothesis thoroughly, we built a spacer library into pDB114 plasmids, targeting all 787 NAGG sequences present in the F NM4 (cid:3) 4 genome. We Nucleic Acids Research, 2023, Vol. 51, No. 14 7445 Figure 3. Non-acquired NAGG spacers provide efficient immunity. ( A ) Propagation ability of (cid:3) NM4 (cid:3) 4 on staphylococci carrying plasmids harboring the type II-A CRISPR-Cas locus of S. py ogenes , pro grammed with dif ferent spacers tha t target protospacers flanked by non-canonical PAMs, as well as a non-targeting control (pC194) and a plasmid with a canonical spacer (pRH079), measured as efficiency of plaquing (EOP). Mean ± StDev values of three independent experiments are shown. ( B ) Enrichment ratios of canonical spacers targeting protospacers flanked by an AGG PAM, calculated using NGS data as the frequency of reads of each spacer sequence after (cid:3) NM4 (cid:3) 4 infection, relati v e to that spacer’s frequency value without phage infection. Each dot r epr esents a differ ent spacer sequence, plotted a t its loca tion within the (cid:3) NM4 (cid:3) 4 genome. The red line and dot r epr esent the aver age enrichment r atio, 1.05. Standar d de viation: ±0.43. ( C ) Same as (B) but for NAGG spacers. Aver age r atio: 1.23; standar d de viation: ±1.57. ( D ) Fitness of NAGG spacers compared to AGG spacers, in a competition assay of the NAGG and AGG spacer libraries, mixed at a 1:1 ratio and infected with F NM4 (cid:3) 4. Fitness is calculated as the frequency of each NAGG spacer relati v e to the total frequency of the NAGG and its cognate AGG spacer combined. Each dot r epr esents the fitness of a different NAGG spacer sequence, plotted at its location within the (cid:3) NM4 (cid:3) 4 genome. The red line and dot r epr esent the average fitness, 0.27. ( E ) Fitness values described in (D) plotted in highest to lowest order. Red lines separate NAGG spacers with fitness values higher than 0.5, between 0.5 and 0.1, and less than 0.1, and the percentage of NAGG spacers falling between each range is noted. also constructed a library of the same 787 spacer sequences shifted by one nucleotide, thus targeting protospacers fol- lowed by AGG PAMs. Plasmids harboring each library were extracted from staphylococcal cultures, the spacer re- gion was amplified via PCR, and the products were sub- jected to next generation sequencing (NGS) to confirm that all spacer sequences were present at similar le v els (Supple- mentary File 1). We then grew each culture for 24 h, in the presence or absence of F NM4 (cid:3) 4 and used NGS to evalu- ate the spacer content within the surviving population. We obtained reads in all four samples for 680 of the 787 spacer sequences, and used the data to calculate an enrichment ra- tio, defined as the read frequency after phage infection rela- ti v e to the frequency obtained for the uninfected culture at the end of the experiment (Supplementary File 1). For the AGG spacers, the ratios were clustered around 1 (average 1.05), and ther efor e the standard deviation was low, ±0.43 (Figure 3 B). On the other hand, NAGG spacers displayed mor e irr egular values. Enrichment r atios r anged from 10 to 0.1 (average 1.23), with an overall higher standard deviation of ±1.57 (Figure 3 C). These results show that while canon- ical spacers are similarly effecti v e in their ability to provide 7446 Nucleic Acids Research, 2023, Vol. 51, No. 14 defense, NAGG spacers mediate more variable targeting efficiency. We also used the NGS data to investigate the impor- tance of different nucleotides in the first position of NAGG PAMs for targeting by Cas9. We compared the mean of the enrichment ratios obtained for AAGG, CAGG, GAGG and TAGG sequences with the overall mean (Supplementary Table S2 and Supplementary File 1). Interestingly, we found a highly significant pr efer ence for A and disfavor for C nu- cleotides in this position. Also, it was noted that in proto- spacers 11, 14, 15 and 17, the last base of the seed sequence is identical to the N of the NAGG PAM (Figure 2 A–D). To investigate how this combination of identical nucleotides contributes to Cas9 immunity, we first looked at the impor- tance of the last nucleotide of the seed sequence using our NGS data, as e xplained abov e. In this case, we did not find a highly significant sta tistical dif ference for the mean tar- geting enrichment values of any last nucleotide of the seed, when compared to the overall mean (Supplementary Table S2 and Supplementary File 1). We then analyzed the data for the dinucleotide sequences composed of the last base of the seed sequence and the N of the NAGG PAM. We found a highly significant statistical difference for the aver- age targeting of AA combinations, but not for CC, GG nor TT (Supplementary Table S2 and Supplementary File 1). Gi v en the substantial pr efer ence of the second of these two A nucleotides (the first A in an AAGG PAM, see above), we belie v e these results reflect this bias but do not indicate that AA in the spc14 target (nor GG in the spc11 and spc15 targets, nor TT in the spc17 target) seed-PAM dinucleotides ar e mor e ef ficient for Cas9 targeting, a t least in our assays. To compare more accurately the immunity provided by canonical and NAGG spacers, we performed a competi- tion experiment in which both libraries were combined at a 1:1 ratio, and the resulting mixed culture was infected with F NM4 (cid:3) 4 phage for 24 h. This experimental set up re- sembles, to some extent, the spacer acquisition process, in which many di v erse spacer sequences are inserted into the CRISPR array of different individual bacteria during phage infection. Using NGS, we first confirmed that the different spacers in this culture were similarly represented. 747 spacer pairs of the 787 total cloned were found in the library, with −4 the great majority present at frequencies ranging from 10 −3 , for both canonical and NAGG spacers (Supple- to 10 mentary Figure S3B-C and Supplementary File 1). We then calculated the enrichment ratio as before, comparing the spacer reads obtained from infected and non-infected cul- tures after 24 h of growth (Supplementary File 1). Of the 747 original NA GG-A GG spacer pairs, 505 spacers appeared in all four experiments. We found that the enrichment ratios for the canonical spacers clustered around values of 10 or 0.1 (Supplementary Figure S3D), suggesting the presence of a subfraction of spacers that provide better immunity in these conditions. The enrichment ratios for the NAGG spacers also seemed to be separated into values either higher or lower than 1, but displayed a more scattered distribu- tion (Supplementary Figure S3E). More importantly, in this experiment the immunity mediated by each NAGG spacer can be directly compared to the internal control provided by its corr esponding canonical, AGG spacer. Ther efor e, we were able to calculate the fitness of each NAGG spacer as its enrichment ratio relati v e to the sum of the AGG and NAGG spacer enrichment ratios (Supplementary File 1). −4 all across the Fitness values ranged between 1 and 10 phage genome, with an average of 0.27 (Figure 3 D). On one hand, 16% of the NAGG spacers displayed a fitness bet- ter than their canonical counterparts ( > 0.5); on the other hand, only 30% showed a fitness value of less than 0.1 (Fig- ur e 3 E). Ther efor e, we conclude that while most AGG spac- ers outperform their corresponding NAGG, they do so by less than an order of magnitude. More generally, the fact that NAGG spacers are not dramatically outcompeted by canonical ones in a mixed population suggests that these spacers, e v en if not acquired by the type II-A CRISPR-Cas system, are able to mount an efficient anti-phage defense. Strong bias against the acquisition of NAGG spacers during the type II-A CRISPR immunization stage Our previous results showed that NAGG spacers can me- diate substantial, if not equal, immunity as spacers match- ing targets flanked by canonical PAMs. In principle, if these spacers ar e acquir ed, they should provide sufficient immu- nity to be maintained within the bacterial population. Since this is not the case, we hypothesized that their relati v e ab- sence from the spacer r epertoir e could be due to a low effi- ciency of acquisition. To test our hypothesis, we looked at the frequency of acquisition of NAGG spacers in the ab- sence of phage infection. This eliminates changes in spacer frequency that are due to the variable ability of each spacer to defend and promote the growth of the host. We looked at spacer acquisition data previously generated in our lab after tr ansformation (via electropor ation) of staphylococci with F NM4 (cid:3) 4 phage DNA, sheared into ∼150 bp fragments by sonication ( 35 ). We compared the frequencies of acquisition of AGG and NAGG spacers and found that, on average, canonical spacers were integrated 2–3 orders of magnitude more than NAGG spacers (Figure 4 A and Supplementary File 1). Very similar results were obtained when we com- pared CGG / NCGG and T GG / NT GG spacers (GGG and NGGG cannot be compared because both sequences are canonical PAMs), not only in the difference between aver- a ges, b ut also in their absolute values. To further test these results, we evaluated the acquisi- tion of double-stranded DNA oligonucleotides containing either the sequence of spc14 and its PAM, or spc17 and its PAM, f ollowing transf orma tion via electropora tion (Fig- ure 4 B). These sequences contain a single GG dinucleotide that can be recognized by Cas9 ( 17 , 23 ) (see Supplemen- tary File 2 for their full sequences). The total number of r eads for acquir ed canonical spacers was a pproximatel y one order of magnitude higher than the number of reads for NAGG spacers. We also made an A to T modification in the oligonucleotide sequence harboring spc17 , to generate a TGG PAM. Although acquisition for the canonical TGG spacer was lower than that of the AGG spacers, acquisition of the NTGG spacer was similarly low to that of the un- modified spc17 oligonucleotide (Figure 4 B). While the trend of these results is consistent with the acquisition of spacers from fra gmented pha ge DNA, the difference in the acquisi- tion of the sequence upstream of the AGG compared to that of the NAGG motif, is much less than what is observed in Nucleic Acids Research, 2023, Vol. 51, No. 14 7447 demonstrate a strong bias against the acquisition of spac- ers deri v ed from non-canonical PAM sequences, e v en when those sequences can support efficient DNA targeting and cleavage. NAGG spacers enable primed spacer acquisition Interestingl y, N AGG spacers represent a form of ‘slipped’ spacer that was previously described in type I CRISPR-Cas systems ( 37 ). In the case of the NAGG PAM, the spacer has a ‘+1 slip’ where the acquired spacer matches a target in which the PAM is not immediately adjacent to the pro- tospacer but instead is one nucleotide away from it ( 37 ). Analysis of these spacers re v ealed that, as opposed to the NAGG spacers, they do not provide efficient immunity. Instead, they can stimulate primed CRISPR adaptation, i.e. the acquisition of additional spacer sequences from in- vaders ( 38 ). Type II-A systems use free DNA ends as sub- strates for spacer acquisition ( 20 , 21 ), ther efor e the pr esence of pre-existing spacers that mediate Cas9 cleavage of the vi- ral DNA enhances the capture of phage sequences from the vicinity of the cut site ( 22 ). As a result of this priming mech- anism, during the e v ents of doub le spacer acquisition, the second spacer integrated into the CRISPR array targets a location of the genome that is close to the region targeted from the first acquired spacer ( 22 , 39 ). To investigate if NAGG spacers are slipped spacers that can mediate priming, we looked for double spacer acqui- sition e v ents in a set of spacers acquired after F NM4 (cid:3) 4 infection ( 40 ). We found 405 unique e v ents in which the first acquired sequence matched a protospacer flanked by an NAGG PAM in the phage genome, and calculated the dis- tance between the first and second acquired spacers, when mapped onto the F NM4 (cid:3) 4 genome (Supplementary File 1). Consistent with previous similar analysis for NGG spac- ers ( 22 , 39 ), the histogram of these distances re v ealed that the majority of the second spacers map within 1 kb of the protospacer targeted by the NAGG spacer (Figure 4 C). This spacer distribution strongly suggests that NAGG spac- ers enable priming and ther efor e r epr esent an example of slipped spacers that can also mediate targeting. DISCUSSION Here, we investigated viral sequences that are rarely ac- quired as spacers during the type II-A CRISPR-Cas re- sponse against phage infection. Many of these sequences result in the recognition of viral targets that contain a non- canonical PAM of the sequence NAGG. We found that these spacers can mediate substantial immunity, e v en those with sequences that are not acquired at all. In contrast to their ability to participate in effecti v e targeting, the acquisi- tion rate of NAGG spacers is very low. Ther efor e, we con- clude that the failure to acquire these sequences, and not their efficiency in promoting phage defense, results in their low abundance within bacterial populations. Our findings highlight a mechanistic separation of the role of Cas9 in PAM recognition during spacer acquisition ( 17 ) and DNA cleavage ( 23 ). This is reminiscent of the findings for type I systems, which display fle xib le r equir ements for PAM recognition during DNA targeting, with many sequences Figur e 4. N AGG spacers ar e rar ely acquir ed during the type II-A CRISPR immune response. ( A ) Scatter plot of frequency of reads for spacers match- ing targets containing different PAM sequences, obtained via NGS anal- ysis of spacer acquisition after transformation of sonicated F NM4 (cid:3) 4 DNA. Mean ± StDev is shown. ( B ) Number of spacer reads, detected by NGS of staphylococcal cultures carrying pGG32- (cid:2) trL, transformed with oligonucleotides harboring PAMs and pre-spacer sequences correspond- ing to different spacers. ( C ) Distance between the targets in the (cid:3) NM4 (cid:3) 4 genome, specified by the first and second spacers acquired after infection of na ¨ıve staphylococci carrying the type II-A CRISPR-Cas system of S. pyo- genes , with the first spacer matching a target with an NAGG PAM. The position of the target of the first NAGG spacer on the (cid:3) NM4 (cid:3) 4 genome is set to 0 kb, and each 1 kb bin contains the number of second spacers acquired from targets that gi v en genomic distance from the first target on the phage DNA. Figure 4 A. We belie v e that this is a consequence of the pres- ence of only two main protospacer sequences in the oligonu- cleotide substrate, in contrast to the 2687 NGG sequences and 2687 NNGG sequences in the (cid:3) NM4 (cid:3) 4 genome. All of these viral sequences are competing for the spacer acqui- sition machinery and it is concei vab le that in such extreme conditions a ∼10 × difference in acquisition efficiency be- tween an NGG spacer and its NNGG counterpart is am- plified to ∼1000 ×. Ther efor e, we conclude tha t these da ta 7448 Nucleic Acids Research, 2023, Vol. 51, No. 14 supporting efficient DNA cleava ge, b ut a much more re- stricted r epertoir e of PAM sequences for acquired spacers ( 41 ). This finding has led to the proposition of using the term Spacer Acquisition Motif (SAM) when referring to the sequence recognized by the acquisition machinery and the term Target Interference Motif (TIM) when referring to the sequence recognized by crRNA-guided nuclease complexes ( 42 ). Pre vious wor k studying the type II-A system of Strep- tococcus thermophilus also identified a minority of spacers carrying imperfect PAMs harboring 1–2 nucleotides that differ from the canonical AGAAW motif ( 43 ). The authors specula ted tha t the existence of non-canonical spacers may reflect the targeting flexibility of Cas9. Our work suggests that, similarly to the NAGG spacers present in the S. pyo- genes type II-A system, the non-canonical spacers captured by the S. thermophilus system may indeed be functional, yet likely under-sampled during the spacer acquisition stage. While high PAM recognition flexibility during DNA cleavage would enable the targeting of escaper phages con- taining mutations in this motif, it is more difficult to envi- sion the evolutionary forces behind stringent PAM recogni- tion during spacer acquisition. Possibly, this is observed as a consequence of the extremely low frequency of spacer in- tegration, combined with the variable targeting efficiency of the NAGG spacers. It is belie v ed that CRISPR systems have e volv ed to have an extremely low rate of spacer acquisition to limit ‘autoimmunity’; i.e. the integration of genomic se- quences into the array that will direct Cas9 to the bacterial chromosome ( 21 , 44 ). Such low frequency implies that the survival of the bacterial population depends on the occur- rence of relati v ely fe w spacer acquisition e v ents. In this con- text, the acquisition of spacers that do not always guarantee efficient targeting, as is the case for NAGG spacers, would ha ve fa v ored the ev olution of a strict selection of NGG spac- ers by the acquisition machinery, in order to ensure that the small population of adapted cells is endowed with a fully competent set of spacers. We also found that NAGG spacers can enhance further spacer acquisition. This result corroborates the ability of these non-canonical spacers to mediate efficient DNA cleav- age, since type II-A systems use free DNA ends as sub- strates for new spacers ( 20–22 ) and ther efor e cleavage of the viral DNA by Cas9 generates these substrates and stim- ulates spacer acquisition from the target site. This phe- nomenon, also known as priming ( 38 ), has similarities in type I CRISPR systems, particularly to the priming me- diated by ‘slipped’ spacers ( 37 ). Slipped spacers originate from imprecise acquisition e v ents and match a protospacer in which the PAM is shifted one nucleotide. Gi v en that the targets of these spacers lack an optimal PAM, they me- dia te inef ficient DNA cleava ge, b ut are able to stimulate primed spacer acquisition in type I CRISPR systems. In this context, NAGG spacers constitute slipped spacers for the type II-A CRISPR system of S. pyog enes tha t enhance fur- ther spacer acquisition but can also media te ef ficient DNA cleavage. Structural studies showed that S. pyogenes Cas9 contains two arginines, each of which interacts with one of the gua- nine residues of the canonical NGG target motif ( 23 ). We currently do not know how these residues are also able to reco gnize N AGG sequences –– w hether they interact with A2 and G3, or with G3 and G4, or whether they adopt a dif- fer ent r ecognition conformation. Howe v er, gi v en that Cas9 participates in spacer acquisition as part of a ‘supercom- plex’ that contains also Cas1, Cas2 and Csn2 ( 17 , 19 ), it is tempting to speculate that PAM recognition is more strin- gent in this potentially confined context, than during target DNA cleavage, which is performed by Cas9 alone ( 7 ). Our r esults ar e in line with, and also expand upon ear- lier wor k, regar ding both spacer acquisition and DNA tar- geting by the S. pyogenes type II-A CRISPR-Cas system. In a previous study that investigated the distribution of spacers within bacterial communities carrying this system, it was determined that the spacer patterns are established early during the CRISPR-Cas immune response and corre- late with spacer acquisition rates, but not with spacer tar- geting efficiency ( 35 ). Here we show that this is also the case for non-canonical N AGG spacers, w hich are poorl y acquired despite their relati v ely high efficiency of targeting, and ther efor e r emain rar e within the host population. With relation to Cas9 targeting, previous comprehensi v e analy- sis of all the possible 5-nucleotide sequences flanking the 3’ end of the protospacer in the S. pyogenes type II-A sys- tem, found that both NAG and NNGG motifs can sup- port a low efficiency of targeting ( ∼50% and 20%, respec- ti v ely) ( 45–47 ). Our evaluation of the effect of different mu- tations of the NAGG sequence showed that changes in G3, which eliminate both the NAG and NNGG motifs, are the most detrimental for the defense mediated by NAGG spac- ers (Figure 1 C). We also found that A2C substitutions have a substantial effect on immunity (Figure 1 C) and that a target flanked by the non-canonical PAM sequence TTGG failed to mediate defense (Figure 1 B-C). These results indi- ca te tha t, w hile N AGG PAMs enable Cas9 target recogni- tion and cleavage, this is not the case for NCGG and NTGG motifs. Finally, the effects of the G4T mutations (NAGT PAM) varied with different protospacer sequences. There- fore, we belie v e that this is the least critical position of the NAGG PAM, a result that correlates with the higher target- ing efficiency previously observed for NAG spacers, com- pared to NNGG spacers (see above). Our study focused on S. pyogenes Cas9, and ther efor e it remains to be determined whether other type II CRISPR- Cas systems can also recognize targets flanked by sequences tha t devia te from canonical PAMs. On the other hand, gi v en that the great majority of CRISPR-based gene editing ap- plications rely on S. pyogenes Cas9 ( 25 , 26 ), our findings have implications for these technologies. In particular with regards to the off-target effects of gene editing ( 48 ), under- standing the flexibility in target recognition by Cas9 could help to identify additional off-target sites and decrease the possibility of collateral cleavage of the human genome. DA T A A V AILABILITY The spacer sequences and accompanying statistics that are discussed in this paper are provided in Supplementary File 1. The original FASTQ files from the NGS experiments have been uploaded to the SRA database [accession # PR- JNA972507]. The raw data from this study are available from the corresponding author upon request. Custom python scripts are deposited at Figshare: https: //figshar e.com/articles/softwar e/kenneyc etal code 2023/ 22970996 . SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We would like to thank Philip Nussenzweig for help with the identification of double spacer acquisition events, Gi- anna Stella for help with Cas9 cleavage assays, and Naama Aviram, Alex Meeske and Andrew Varble for assistance in optimizing protocols. FUNDING C.T.K. is supported by the NIH Medical Scientist Train- ing Program T32 Grant [T32GM07739] and NIH NRSA F31 Individual Predoctoral Fellowship to Promote Di v er- sity [F31GM134665]. L.A.M. is supported by an NIH Di- rector’s Pioneer Award [DP1GM128184]. L.A.M. is an in- vestigator of the Howard Hughes Medical Institute. Fund- ing for open access charge: Institutional funds. Conflict of interest statement. L.A.M. is a cofounder and Scientific Advisory Board member of Intellia Therapeutics, a co-founder of Eligo Biosciences, and member of the Sci- entific Advisory Board of Ancillia Biosciences. REFERENCES 1. Nussenzweig,P.M. and Marraffini,L.A. (2020) Molecular mechanisms of CRISPR-Cas immunity in bacteria. Annu. Rev. Genet. , 54 , 93–120. 2. Barrangou,R., Fremaux,C., De v eau,H., Richar ds,M., Boy av al,P., Moineau,S., Romero,D.A. and Horvath,P. (2007) CRISPR provides acquir ed r esistance against viruses in prokaryotes. Science , 315 , 1709–1712. 3. Brouns,S.J., Jore,M.M., Lundgren,M., Westra,E.R., Slijkhuis,R.J., Snijders,A.P., Dickman,M.J., Makarova,K.S., Koonin,E.V. and van der Oost,J. (2008) Small CRISPR RNAs guide antiviral defense in prokaryotes. Science , 321 , 960–964. 4. Deltcheva,E., Chylinski,K., Sharma,C.M., Gonzales,K., Chao,Y., Pirzada,Z.A., Eckert,M.R., Vogel,J. and Charpentier,E. (2011) CRISPR RNA ma tura tion by trans-encoded small RNA and host factor RNase III. Nature , 471 , 602–607. 5. Hale,C., Kleppe,K., Terns,R.M. and Terns,M.P. (2008) Prokaryotic silencing (psi)RNAs in Pyrococcus furiosus . RNA , 14 , 2572–2579. 6. Gasiunas,G., Barrangou,R., Horvath,P. and Siksnys,V. (2012) Cas9-crRNA ribonucleoprotein complex mediates specific DNA cleavage for adapti v e immunity in bacteria. Proc. Natl. Acad. Sci. U.S.A. , 109 , E2579–E2586. 7. Jinek,M., Chylinski,K., Fonfara,I., Hauer,M., Doudna,J.A. and Charpentier,E. (2012) A programmable dual-RNA-guided DNA endonuclease in adapti v e bacterial immunity. Science , 337 , 816–821. 8. Jor e,M.M., Lundgr en,M., van Duijn,E., Bultema,J.B., Westra,E.R., Waghmare,S.P., Wiedenheft,B., Pul,U., Wurm,R., Wagner,R. et al. (2011) Structural basis for CRISPR RNA-guided DNA recognition by Cascade. Nat. Struct. Mol. Biol. , 18 , 529–536. 9. Kazlauskiene,M., Tamulaitis,G., Kostiuk,G., Venclovas,C. and Siksnys,V. (2016) Spatiotemporal control of Type III-A CRISPR-Cas immunity: coupling DNA degradation with the target RNA recognition. Mol. Cell , 62 , 295–306. 10. Garneau,J.E., Dupuis,M.E., Villion,M., Romero,D.A., Barrangou,R., Boy av al,P., Fremaux,C., Horv ath,P., Magadan,A.H. and Moineau,S. (2010) The CRISPR / Cas bacterial immune system cleaves bacteriophage and plasmid DNA. Nature , 468 , 67–71. Nucleic Acids Research, 2023, Vol. 51, No. 14 7449 11. Samai,P., Pyenson,N., Jiang,W., Goldberg,G.W., Hatoum-Aslan,A. and Marraffini,L.A. (2015) Co-transcriptional DNA and RNA cleavage during Type III CRISPR-Cas immunity. Cell , 161 , 1164–1174. 12. Westra,E.R., van Erp,P.B., Kunne,T., Wong,S.P., Staals,R.H., Seegers,C.L., Bollen,S., Jore,M.M., Semenova,E., Se v erinov,K. et al. (2012) CRISPR immunity relies on the consecuti v e binding and degradation of negati v ely supercoiled invader DNA by Cascade and Cas3. Mol. Cell , 46 , 595–605. 13. Makarova,K.S., Wolf,Y.I., Iranzo,J., Shmakov,S.A., Alkhnbashi,O.S., Brouns,S.J .J ., Charpentier,E., Cheng,D., Haft,D.H., Horvath,P. et al. (2020) Evolutionary classification of CRISPR-Cas systems: a burst of class 2 and deri v ed variants. Nat. Rev. Microbiol. , 18 , 67–83. 14. Bolotin,A., Quinquis,B., Sorokin,A. and Ehrlich,S.D. (2005) Cluster ed r egularly interspaced short palindrome r epeats (CRISPRs) have spacers of extrachromosomal origin. Microbiology , 151 , 2551–2561. 15. De v eau,H., Barrangou,R., Garneau,J.E., Labonte,J., Fremaux,C., Boy av al,P., Romero,D.A., Horvath,P. and Moineau,S. (2008) Phage response to CRISPR-encoded resistance in Streptococcus thermophilus . J. Bacteriol. , 190 , 1390–1400. 16. Bikard,D., Hatoum-Aslan,A., Mucida,D. and Marraffini,L.A. (2012) CRISPR interference can pre v ent natural transformation and virulence acquisition during in vivo bacterial infection. Cell Host Microbe , 12 , 177–186. 17. Heler,R., Samai,P., Modell,J.W., Weiner,C., Goldberg,G.W., Bikard,D. and Marraffini,L.A. (2015) Cas9 specifies functional viral targets during CRISPR-Cas adaptation. Nature , 519 , 199–202. 18. Wei,Y., Terns,R.M. and Terns,M.P. (2015) Cas9 function and host genome sampling in Type II-A CRISPR-Cas adaptation. Genes Dev. , 29 , 356–361. 19. Jakhanwal,S., Cress,B.F., Maguin,P., Lobba,M.J., Marraffini,L.A. and Doudna,J.A. (2021) A CRISPR-Cas9-integrase complex generates precise DNA fragments for genome integration. Nucleic Acids Res. , 49 , 3546–3556. 20. Maguin,P., Varble,A., Modell,J.W. and Marraffini,L.A. (2022) Cleavage of viral DNA by restriction endonucleases stimulates the type II CRISPR-Cas immune response. Mol. Cell , 82 , 907–919. 21. Modell,J.W ., Jiang,W . and Marraffini,L.A. (2017) CRISPR-Cas systems exploit viral DNA injection to establish and maintain adapti v e immunity. Nature , 544 , 101–104. 22. Nussenzweig,P.M., McGinn,J. and Marraffini,L.A. (2019) Cas9 Cleavage of Viral Genomes Primes the Acquisition of New Imm unolo gical Memories. Cell Host Microbe , 26 , 515–526. 23. Anders,C., Niewoehner,O., Duerst,A. and Jinek,M. (2014) Structural basis of PAM-dependent target DNA recognition by the Cas9 endonuclease. Nature , 513 , 569–573. 24. Sternberg,S .H., Redding,S ., Jinek,M., Greene,E.C. and Doudna,J.A. (2014) DNA interrogation by the CRISPR RNA-guided endonuclease Cas9. Nature , 507 , 62–67. 25. Cong,L., Ran,F.A., Cox,D., Lin,S., Barretto,R., Habib,N., Hsu,P.D., Wu,X., Jiang,W., Marraffini,L.A. et al. (2013) Multiplex genome engineering using CRISPR / Cas systems. Science , 339 , 819–823. 26. Mali,P., Yang,L., Esvelt,K.M., Aach,J., Guell,M., Dicarlo,J.E., Norville,J.E. and Church,G.M. (2013) RNA-guided human genome engineering via Cas9. Science , 339 , 823–826. 27. Workman,R.E., Pammi,T., Nguyen,B.T.K., Graeff,L.W., Smith,E., Sebald,S.M., Stoltzfus,M.J., Euler,C.W. and Modell,J.W. (2021) A natural single-guide RNA repurposes Cas9 to autoregulate CRISPR-Cas expression. Cell , 184 , 675–688. 28. Pyenson,N.C., Gayvert,K., Varble,A., Elemento,O. and Marraffini,L.A. (2017) Broad targeting specificity during bacterial Type III CRISPR-Cas immunity constrains viral escape. Cell Host Microbe , 22 , 343–353. 29. Kreiswirth,B.N., Lofdahl,S., Betley,M.J., O’Reilly,M., Schlie v ert,P.M., Bergdoll,M.S. and Novick,R.P. (1983) The toxic shock syndrome ex oto xin structural gene is not detectably transmitted by a prophage. Nature , 305 , 709–712. 30. Horinouchi,S. and Weisblum,B. (1982) Nucleotide sequence and functional map of pC194, a plasmid that specifies inducible chloramphenicol resistance. J. Bacteriol. , 150 , 815–825. 31. Goldberg,G.W ., Jiang,W ., Bikard,D. and Marraffini,L.A. (2014) Conditional tolerance of temperate phages via 7450 Nucleic Acids Research, 2023, Vol. 51, No. 14 transcription-dependent CRISPR-Cas targeting. Nature , 514 , 633–637. 32. Meeske,A.J., Nakandakari-Higa,S. and Marraffini,L.A. (2019) Cas13-induced cellular dormancy pre v ents the rise of CRISPR-resistant bacteriophage. Nature , 570 , 241–245. 33. Heler,R., Wright,A.V., Vucelja,M., Bikard,D., Doudna,J.A. and Marraffini,L.A. (2017) Mutations in Cas9 enhance the rate of acquisition of viral spacer sequences during the CRISPR-Cas immune response. Mol. Cell , 65 , 168–175. 34. Aviram,N ., Thornal,A.N ., Zeevi,D. and Marraffini,L.A. (2022) Different modes of spacer acquisition by the Staphylococcus epidermidis type III-A CRISPR-Cas system. Nucleic Acids Res. , 50 , 1661–1672. 35. Heler,R., Wright,A.V., Vucelja,M., Doudna,J.A. and Marraffini,L.A. (2019) Spacer acquisition rates determine the imm unolo gical di v ersity of the Type II CRISPR-Cas immune response. Cell Host Microbe , 25 , 242–249. 36. Bikard,D., Euler,C.W., Jiang,W., Nussenzweig,P.M., Goldberg,G.W., Duportet,X., Fischetti,V.A. and Marraffini,L.A. (2014) Exploiting CRISPR-Cas nucleases to produce sequence-specific antimicrobials. Nat. Biotechnol. , 32 , 1146–1150. 37. Jackson,S.A., Birkholz,N., Malone,L.M. and Fineran,P.C. (2019) Imprecise spacer acquisition generates CRISPR-Cas immune di v ersity through primed adaptation. Cell Host Microbe , 25 , 250–260. 38. Datsenko,K.A., Pougach,K., Tikhonov,A., Wanner,B.L., Se v erinov,K. and Semenova,E. (2012) Molecular memory of prior infections activates the CRISPR / Cas adaptive bacterial immunity system. Nat. Commun. , 3 , 945. 39. Nicholson,T.J., Jackson,S.A., Croft,B.I., Staals,R.H.J., Fineran,P.C. and Brown,C.M. (2018) Bioinformatic evidence of widespread priming in type I and II CRISPR-Cas systems. RNA Biol. , 16 , 566–576. 40. McGinn,J. and Marraffini,L.A. (2016) CRISPR-Cas systems optimize their immune response by specifying the site of spacer integration. Mol. Cell , 64 , 616–623. 41. Almendros,C., Guzman,N.M., Diez-Villasenor,C., Garcia-Martinez,J. and Mojica,F.J. (2012) Target motifs affecting natural immunity by a constituti v e CRISPR-Cas system in Esc heric hia coli . PLoS One , 7 , e50797. 42. Shah,S .A., Erdmann,S ., Mojica,F.J. and Garrett,R.A. (2013) Protospacer recognition motifs: mixed identities and functional di v ersity. RNA Biol. , 10 , 891–899. 43. Paez-Espino,D., Morovic,W., Sun,C.L., Thomas,B.C., Ueda,K., Stahl,B., Barrangou,R. and Banfield,J.F. (2013) Strong bias in the bacterial CRISPR elements that confer immunity to phage. Nat. Commun. , 4 , 1430. 44. Levy,A., Goren,M.G., Yosef,I., A uster,O ., Manor,M., Amitai,G., Edgar,R., Qimron,U. and Sorek,R. (2015) CRISPR adaptation biases explain pr efer ence for acquisition of foreign DNA. Nature , 520 , 505–510. 45. Collias,D., Leenay,R.T., Slotkowski,R.A., Zuo,Z., Collins,S.P., McGirr,B.A., Liu,J. and Beisel,C.L. (2020) A positi v e, growth-based PAM screen identifies noncanonical motifs recognized by the S. pyogenes Cas9. Sci. Adv. , 6 , eabb4054. 46. Jiang,W., Bikard,D., Cox,D., Zhang,F. and Marraffini,L.A. (2013) RNA-guided editing of bacterial genomes using CRISPR-Cas systems. Nat. Biotechnol. , 31 , 233–239. 47. Leenay,R.T., Maksimchuk,K.R., Slotkowski,R.A., Agrawal,R.N., Gomaa,A.A., Briner,A.E., Barrangou,R. and Beisel,C.L. (2016) Identifying and visualizing functional PAM di v ersity across CRISPR-Cas systems. Mol. Cell , 62 , 137–147. 48. Hsu,P.D ., Scott,D .A., Weinstein,J.A., Ran,F.A., Konermann,S., Agarwala,V., Li,Y., Fine,E.J., Wu,X., Shalem,O. et al. (2013) DNA targeting specificity of RNA-guided Cas9 nucleases. Nat. Biotechnol. , 31 , 827–832. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
10.1126_scitranslmed.adh9917	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Sci Transl Med. Author manuscript; available in PMC 2023 September 14. Published in final edited form as: Sci Transl Med. 2023 July 26; 15(706): eadh9917. doi:10.1126/scitranslmed.adh9917. Signatures of AAV-2 immunity are enriched in children with severe acute hepatitis of unknown etiology Moriah M. Mitchell1,2,3, Yumei Leng1,2, Suresh Boppana4, William J. Britt4, Luz Helena Gutierrez Sanchez5, Stephen J. Elledge1,2,* 1Division of Genetics, Department of Medicine, Howard Hughes Medical Institute, Brigham and Women’s Hospital, Boston, MA 02115, USA 2Department of Genetics, Harvard Medical School, Boston, MA 02115, USA 3Program in Systems, Synthetic, and Quantitative Biology, Harvard University, Boston, MA 02115, USA 4Division of Pediatric Infectious Diseases, Department of Pediatrics, University of Alabama at Birmingham, Birmingham, Alabama, USA 5Division of Gastroenterology, Hepatitis, and Nutrition, Department of Pediatrics, University of Alabama at Birmingham, Birmingham, Alabama, USA Abstract Severe acute hepatitis of unknown etiology in children is under investigation in 35 countries. Although several potential etiologic agents have been investigated, a clear cause for the liver damage observed in these cases remains to be identified. Using VirScan, a high throughput antibody profiling technology, we probed the antibody repertoires of nine cases of severe acute hepatitis of unknown etiology treated at Children’s of Alabama and compared their antibody responses to 38 pediatric and 470 adult controls. We report increased adeno-associated dependoparvovirus A (AAV-A) breadth in cases relative to controls and detailed adeno-associated virus 2 (AAV-2) peptide responses that were conserved in 7 of 9 cases but rarely observed in pediatric and adult control. These findings suggest that AAV-2 is a likely etiologic agent of severe acute hepatitis of unknown etiology. One-sentence summary: AAV-2-reactive antibodies and evidence of AAV-2 helper virus infection are associated with pediatric severe acute hepatitis of unknown etiology. This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding author: [email protected]. Author contributions: M.M.M. and S.J.E. conceptualized the project and wrote the paper. M.M.M. and Y.L. performed the laboratory experiments. M.M.M. analyzed the data. S.B., W.J.B., and H.L.G.S. curated and provided samples and metadata for the AHUE cases. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Introduction Page 2 Recent reports of acute hepatitis of unknown etiology (AHUE) in children have sparked concern globally with over 1000 probable cases identified since October 2021 (1). In the United States, 372 children were under CDC investigation for acute hepatitis of unknown cause as of October 2022 (2) with 22 associated liver transplants and 14 deaths reported (3). Although overall incidence of pediatric hepatitis cases in the United States may not have increased over pre-pandemic incidence (4), spatiotemporal clustering of cases in Alabama, Scotland, and the Netherlands has prompted the search for a shared etiologic agent (1, 3, 5). Hypotheses under investigation include environmental or toxin exposure, pathogen exposure, superantigen or autoimmune reactions to persistent or previous severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2 infection), and altered response to first adenoviral exposure as a result of delayed initial exposure due to infection control behaviors and isolation brought on by the coronavirus disease 2019 (COVID-19) pandemic (3). Human adenoviruses (HAdVs), particularly adenovirus F subtypes 40 and 41, are a leading exposure under investigation as a potential trigger of AHUE (6). In contrast to HAdVs B to E, Adenovirus F is well adapted for gastrointestinal tropism as result of structural differences in the capsid fibers which enable stability at low pH (7) and is a leading cause of gastroenteritis in children (8). Although HAdV infection in general has been implicated in some cases of hepatitis, reports of adenovirus associated hepatitis in immunocompetent patients are scarce (9–11). Of the 9 cases of AHUE identified in the initial Alabama cluster in the United States, 8 (89%) tested positive for human adenovirus infection by whole blood quantitative polymerase chain reaction (qPCR) and all five for which subtyping was possible were found to be Adenovirus 41 (6). Adenovirus infection has been detected less frequently in other case series. In a cluster in Scotland, 5 out of 13 children (38%) tested positive for HAdV by PCR testing of throat swab, blood, or stool (5). Only 45% of all children in the United States and 52% of children in Europe under investigation for AHUE were found to be positive for any HAdV where testing was performed (1, 3). In order to investigate possible viral etiology of AHUE, we employed VirScan, a phage display immunoprecipitation sequencing (PhIP-seq) technology that detects antibody binding to peptides derived from the proteomes of selected pathogens. We studied the anti-viral antibody repertoires of nine patients with AHUE in comparison to pediatric and adult controls. Here, we present evidence of a conserved signature of adeno-associated dependoparvovirus A (AAV-2) immunity in cases of AHUE that was not observed in pediatric or adult controls. These findings point to AAV-2 as a potential etiologic agent for AHUE development. Results Sample characteristics Serum isolated from whole blood of 9 patients with severe acute hepatitis of unknown etiology admitted to Children’s of Alabama between October 1, 2021, and February 28, 2022, were analyzed (Table 1). Serum samples collected from 38 healthy children prior to Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 3 the COVID-19 pandemic were used as pediatric controls. Serum samples obtained from 470 adults with current or prior SARS-CoV-2 infection were used as additional controls. Overall breadth of antiviral antibody responses did not differ between cases and pediatric controls. To explore the hypothesis that AHUE is triggered by atypical immune response due to delayed initial pathogen exposure caused by pandemic infection control measures in early life, we examined the antibody breadth of cases and pediatric control samples collected prior to the COVID-19 pandemic. If delayed initial pathogen exposure played a role in development of AHUE, one would expect to see reduced breadth of antibody response in cases versus controls; however, no difference in overall antibody breadth, measured as the total number of VirScan peptides targeted in a sample, was observed between cases and pediatric controls (p = 0.78 for Welch’s 1-sided t-test). Of the 115,753 peptides in the VirScan library used, samples targeted a mean of 1352 ± 40 peptides. After correcting for multiple testing, the only significant difference in breadth of antibody response at the family level occurred for Parvoviridae (p =1.47 × 10−3 for Welch’s 1-sided t-test). As overall breadth of antibody response was not different between cases and controls, we explored whether differential antibody responses to specific pathogens were observed between cases and controls (Fig. 1A). Although nominally significant differences (Welch’s 1-sided t-test, alpha=0.05) in mean breadth of response between cases and pediatric controls were detected across 5 pathogens (Fig. 1B), only differences in responses to 4 Parvoviridae species (Adeno-associated dependoparvovirus A, Adeno-associated virus, Adeno-associated virus VR-355, and non-human primate Adeno-associated virus) were significant after correcting for multiple hypothesis testing [false discovery rate (FDR) = 0.05]. Of 9 cases, 7 appeared to have strong antibody responses to AAVs. Breadth of pathogen-specific response to other common childhood pathogens in cases fell well within the distributions observed in pediatric controls (Fig. 1B and C). Epitopes Associated with Severe Acute Hepatitis of Unknown Etiology In order to identify peptides targeted at significantly higher frequency in cases than controls, we used Fisher’s one-way Exact Test and adjusted for multiple tests with the Benjamini- Hochberg Procedure. Sixty-nine peptides from 7 viruses were significantly enriched in cases versus pediatric controls (FDR=0.05) (Fig. 2A). Sixty of these peptides were also significantly (FDR=0.05) enriched in cases versus the adult control cohort in which in which samples were collected at a more similar time and location to cases. All peptides targeted at significantly (FDR=0.05) higher relative frequency in cases versus both adult and pediatric controls were derived from AAVs, except one Hepatitis B virus peptide. This peptide is a subsequence of the Hepatitis B virus Large Envelope Protein but shares two motifs with AAV peptides that were targeted at higher relative frequency in cases. The cytomegalovirus (CMV) and influenza peptides were not enriched in cases relative to the adult control group. Many of the AAV peptides targeted in samples were from overlapping tiles from the same protein sequence or homologous regions of several Parvoviridae species (fig. S1 to 3). Targeting of overlapping peptides suggests that a linear epitope is located within the 28 Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 4 amino acid stretch shared by the library peptides. Two distinct regions likely to contain epitopes were identified for the Rep68/78 protein (fig. S1 and 2), whereas a 224 amino acid stretch of 7 overlapping VirScan peptides that likely contains several linear epitopes was observed for the capsid VP1 protein (fig. S3)>. Targeting of homologous regions of other Parvoviridae species is likely evidence of a cross-reaction originating from AAV-2 immunity. Among peptides targeted in all or most cases, these AAV peptides were remarkable in that they were targeted at very low frequency in all control groups (Fig. 2B). Most other peptides targeted at high frequency in cases contained previously identified “public epitopes,” or epitopes targeted frequently in seropositive individuals (12). These peptides were targeted in both the pediatric and adult control cohorts. Cases and controls cluster according to parvovirus reactivity. Complete hierarchical agglomerative clustering according to Parvoviridae reactivity clearly distinguishes the seven cases with any apparent AAV immunity from pediatric controls (Fig. 3A). Cases did not cluster together well when clustered according to Adenoviridae or Herpesviridae reactivity instead (fig. S4). Although adeno-associated dependoparvovirus reactivity was detected in pediatric controls, clustering by Parvoviridae reactivity appeared to be driven by a highly conserved and strong response to a set of specific AAV peptides from the capsid and replication region in cases with AAV immunity. Evidence of AAV-2 Epitope Spreading can be found in cases of AHUE. All positive cases targeted several peptides derived from both the AAV-2 VP1 and AAV-2 Rep68 proteins. The average range of a linear epitope footprint is 4 to 22 amino acids (13– 15). Thus, a single targeted 56-mer VirScan peptide could contain multiple distinct epitopes. When two adjacent peptides are recognized, a minimum of one epitope must exist. If only one epitope exists, it would be located in the overlapping 28-mer shared between the two peptides. If two recognized peptides are separated by a non-scoring peptide, antibodies in the sample bound a minimum of two epitopes. For example, for patient 1, there were a minimum of 7 distinct epitopes in VP1 and 5 in Rep68, demonstrating epitope spreading consistent with a robust antibody response (Fig. 3B). AAV-2 positive cases targeted a mean of 40.4 ± 11.3 Adeno-associated dependoparvovirus A (AAV-A) VirScan peptides per child compared to means of 1.8 ± 0.8 and 1.0 ± 0.2 peptides for pediatric and adult controls, respectively. Homologous regions of other adeno- associated dependoparvoviruses were also targeted at high frequency in cases and at very low frequency or not at all in controls (figs. S1, S2). Strong AAV Antibody Responses were observed in serum from AHUE cases. VirScan Epitope Binding Signal (EBS) correlates with antibody titer and can be used as a quantitative measure of strength of antibody response (16). To identify the strongest antibody responses in each sample, we rank-ordered VirScan peptides by VirScan EBS (from greatest to least) and examined the composition of the top 100 scoring peptides for each individual. Although AAV-A peptides only account for 0.15% of the VirScan library, Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 5 they make up a mean of 6.4% ± 1.8% of the top 100 peptides rank-ordered by EBS in AHUE cases that target AAV-2 (N=7). In contrast, a single AAV-A peptide appears in the top 100 peptides rank-ordered by EBS for only one pediatric control (N=38). Immunity to AAV-2 Helper Viruses was enriched in cases. We detected antibodies to HAdVs in all AAV-2 positive cases, consistent with previous clinical testing (6). We also detect immunity to at least two human herpesviruses (HHVs) in each of these cases (fig. S5A). Some HHVs were observed with greater frequency in cases than pediatric controls (fig. S5B). Discussion In this study, we detected a conserved antibody response to specific regions of AAV-2 proteins in cases of AHUE that is not observed in pediatric or adult controls. Although associations between Adenovirus F viral detection and AHUE have been reported (6), we do not detect any differences in breadth of adenoviral antibody response or response to specific adenovirus peptides in cases versus controls. Because AAV-2 requires a helper virus to replicate and human adenoviruses can fulfill this role, it is possible that human adenovirus exposure is actually a lurking variable and associations between HAdVs and AHUE are spurious. This is supported by evidence that frequency of HAdV detection in the general AHUE population is only 45 to 52% (1, 3) and many of the cases negative for HAdVs tested positive for HHVs (17) which can also act as helper viruses for AAV-2 (18). Furthermore, during the course of the study, additional evidence of AAV-2 exposure in cases of AHUE was independently recorded in cohorts of AHUE patients in the United Kingdom(19, 20). Notably, we have detected increased breadth of antibody response to AAVs in cases versus controls. Each case recognizes multiple regions within the AAV-2 capsid and replicase regions with evidence of epitope spreading. Moreover, these regions were conserved across AHUE cases but very rarely targeted in controls. This is not to say that AAV-2 is not targeted at all in the pediatric control group. Rather, the breadth of the AAV-A antibody response is far greater in AAV-2 positive cases than controls, with the seven AAV positive AHUE cases, each targeting 27 to 60 AAV-A peptides. No pediatric control targets more than 8 such peptides. About 15.4% of pediatric controls under 3 years old and 16.7% of those over three years old target at least 4 AAV-A peptides. These values are only slightly lower than previous AAV-2 seropositivity estimates of 21% in healthy children between 1 and 3 years old and 22% in children 3 to 18 years old (21). Part of this difference is due to the fact that VirScan slightly under detects prevalence for some viruses (e.g. measles) relative to enzyme-linked immunosorbent assays (ELISAs) in individuals with a past history of infection but readily detects recent infections (12, 16). We also report a difference in magnitude of antibody responses to AAV-A in AAV positive AHUE cases relative to controls. Strong AAV-A responses account for a mean of 6.4 ± 1.8% of the top 100 peptides rank-ordered by VirScan EBS in AAV positive cases (N=7). Taken together, the strength and breadth of responses to AAV peptides in cases are consistent with peak antibody concentrations during infections and shortly after (22). This indicates recent infection with AAV-2. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 6 We also detect immunity to HAdVs and HHVs, both of which can act as helper viruses to facilitate AAV-2 replication, in all AAV-2 positive cases. We detect immunity to at least two human herpesviruses in addition to adenovirus immunity in each of these cases. These findings are consistent with previously reported qPCR detection of human adenovirus and human herpesvirus DNA in samples from patients with AHUE (6). Presence of active infection with potential helper viruses and detection of AAV-2 immunity points to likely presence of replicating AAV-2 at the time of AHUE onset. Our data reveals a strong correlation of AHUE with AAV-2 infections exhibiting a high titer and breadth of antibody responses indicative of a recent infection. These data implicate AAV-2 as a likely causative agent of the disease. If so, a central unanswered question is: Why is AAV-2, which infects many people, suddenly more pathogenic in cases of AHUE? Beyond the possibility that this variant is more pathogenic, one possible explanation is that coinfection with multiple helper viruses may act to increase the total number of cells infected with helper virus, and thus, the total number of cells in which AAV-2 replication is possible. This may contribute to increased viral titers, inflammation, and tissue damage leading to severe disease in a subset of AAV-2 infected children, a possibility that needs to be examined in additional cohorts. Consistent with this hypothesis, hepatotoxicity has been reported in high-dose AAV gene therapy trials, even leading to death (23). As a result, immunosuppressants are commonly co-administered with AAV-vectored gene therapy to prevent hepatotoxicity (24). A speculative and complementary hypothesis that may play a role in emergence of AHUE concerns COVID-19 infection control measures and their potential to alter the timing of viral exposures. Years of masking prevents infection with many viruses at normal frequencies while immunity wanes for these viruses. Once masking was reduced, more viruses could have been in simultaneous circulation at higher frequencies in a more vulnerable population, possibly synchronizing infections. Thus, it is possible that as a result of these circumstances children may have been more likely to acquire multiple infections at once due to changes in masking policies and exposures than before the pandemic. This could increase the likelihood of concurrent infection with AAV-2 and one or more replicating helper viruses. Seven patients in our cohort clearly had an immune response to AAV-2, but two did not. We do not believe that rules out AAV-2 as a causative agent of AHUE. Unexplained pediatric hepatitis is not a new phenomenon. Recent and historical cases of pediatric hepatitis with no identified cause are likely to stem from an array of biological mechanisms. It is possible that the two samples without AAV-2 have hepatitis driven by a different cause whereas the remaining seven may be part of a recent outbreak linked to AAV-2. Our study has limitations. It is worth noting that due to sample availability, pediatric controls were not from the same area as cases and were not matched on demographic factors. Comparing cases to well-matched healthy controls from the same area would have been preferable and may have influenced results. Another limitation of this study is sample size. Determining whether the same patterns in AAV-2 antibody response are observed in independent AHUE cohorts would also be useful. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 7 The distinct signatures of AAV-2 immunity detected in this cohort and the cases in Scotland are both elevated over infection frequency in controls. AAV-2 is therefore likely to be in some way responsible for AHUE. How this infection contributes to the development of AHUE mechanistically remains to be elucidated. Further research is needed to characterize the seroprevalence of AAV-2 antibodies in AHUE patients and identify mechanisms by which AAV-2 may be driving pediatric hepatitis onset. Methods Study Design This is an observational case-control study utilizing previously collected serum samples from cases and several control groups as outlined below. The objective of this study was to identify whether there were differences in the antibody responses of cases of AHUE compared to controls with the goal of identifying a viral cause for AHUE. Sample size was determined by sample availability. Each serum sample was run on VirScan with two technical replicates to ensure that results were consistent between replicates. No samples were excluded from analysis. We observed the distributions of antiviral antibodies in cases and controls. Statistical analysis was used to evaluate differences between antibody responses of cases and controls overall and at the family, pathogen, and peptide levels. Sources of Serum VirScan Secondary use of all human samples for the purposes of this work was exempted by the Brigham and Women’s Hospital Institutional Review Board (protocol number 2013P001337). Serum was obtained from nine patients under the age of 18 years admitted to Children’s of Alabama for severe acute hepatitis of unknown cause between October 1, 2021, and February 28, 2022. These samples were derived from patients previously characterized in a published case series (6). Samples from healthy children enrolled in the DIABIMMUNE study were used as pediatric controls. This cohort was described in previous studies (16, 25). The adult cohort was from a previous study designed to measure antibody responses in SARS-CoV-2 patients (26). VirScan was performed using the VirScan 2.0 library (16) following a published protocol (27). In brief, VirScan is a Phage Immunoprecipitation Sequencing (PhIP-seq) technology. The library used in this study displays 115,753 peptides derived from published viral proteome sequences and known Immune Epitope Database (IEDB) epitopes on the surface of T7 bacteriophage. Diluted serum is incubated with the phage library and phage bound to antibodies are immunoprecipitated using Protein A/G coated magnetic beads. The phage insert DNA is then amplified and sequenced to determine which peptides were bound by antibodies in serum and to what degree. Epitope binding signals (EBS), a quantitative measure of antibody binding enrichment to each library peptide, and hits, a binary measure of whether a peptide was targeted or not were computed as previously described (16, 27). In brief, we consider a peptide recognized (“a hit”) if the EBS in both technical replicates is at least 3.5. EBS is presented as mean EBS across both technical replicates. EBS values below 0 have been artificially set to 0 for visualization. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Data analysis Page 8 Overall antibody breadth was calculated as the total number of hits in a sample across the VirScan library. Hits are defined as peptides with epitope binding signals greater than 3.5 in both technical replicates. Pathogen-specific and family-wide breadth were calculated as the total number of hits across peptides derived from published protein sequences for a given pathogen species or family. Multiple sequence alignments were generated using Clustal Omega (28–30). Statistical analysis—Figures were generated using R (version 4.1.2 with packages ggplot2 (31), ggpubr (32), ggmosaic (33), pheatmap (34), and ggplotify (35)), Adobe Illustrator, and UCSF ChimeraX (36). Statistical analysis was performed in R (37) (version 4.1.2). Welch’s T-test, Fisher’s Exact Test, and Benjamini-Hochberg adjusted p-values were computed using the t.test, fisher.test, and p.adjust functions respectively from the stats package (version 3.6.2). To test whether overall breadth of immune response was lower in cases versus controls, we used Welch’s one-sided t-test with an alpha of 0.05. To test whether immune breadth was higher in cases than controls within specific families and pathogens, Welch’s one-sided t-test was used with a Benjamini-Hochberg correction for multiple tests (FDR=0.05). In order to identify peptides targeted at significantly higher frequency in cases than controls, we used Fisher’s one-way Exact Test and adjusted for multiple tests with the Benjamini-Hochberg Procedure (FDR=0.05). Confidence intervals were constructed using 95% confidence intervals based on the t-distribution with the summarySE function in the Rmisc package (version 1.5.1). Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgements: We thank Galit Alter and Ellen Shrock for helpful discussions. Funding: This research was supported by NIH grants 1P01AI165072 to SJE and 1U01CA260462–02 to SB. S.J.E. is an Investigator with the Howard Hughes Medical Institute. Competing Interests: S.J.E. is a founder of TSCAN Therapeutics, ImmuneID MAZE Therapeutics and Mirimus, S.J.E. serves on the scientific advisory board of Homology Medicines, TSCAN Therapeutics, MAZE Therapeutics, none of which impact this work. S.J.E. is an inventor on a patent application filed by the Brigham and Women’s Hospital (US20160320406A) that covers the use of the VirScan library to identify pathogen antibodies in blood. S.B. is a member of GSK CMV Vaccine Advisory Board. S.B. receives research grant funding from Merck and Pfizer on unrelated projects. Bill Britt is a consultant to Moderna’s CMV vaccine program. Data and Materials Availability: All data associated with this study are in the paper or supplementary materials. All reasonable requests for materials to the corresponding author will be fulfilled. The VirScan library is available from S.J.E. under a material transfer agreement with the Brigham and Women’s Hospital. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. References Page 9 1. World Health Organization, in Disease Outbreak News. (2022). 2. Centers for Disease Control and Prevention, “Investigation Updates: Children with Hepatitis of Unknown Cause,” (2022). 3. Technical Report: Acute Hepatitis of Unknown Cause (Centers for Disease Control and Prevention, Atlanta, 2022). 4. Kambhampati AK, Trends in acute hepatitis of unspecified etiology and adenovirus stool testing results in children—United States, 2017–2022. MMWR. Morbidity and Mortality Weekly Report 71, (2022). 5. Marsh K, Tayler R, Pollock L, Roy K, Lakha F, Ho A, Henderson D, Divala T, Currie S, Yirrell D, Investigation into cases of hepatitis of unknown aetiology among young children, Scotland, 1 January 2022 to 12 April 2022. Eurosurveillance 27, 2200318 (2022). [PubMed: 35426362] 6. Gutierrez Sanchez LH, Shiau H, Baker JM, Saaybi S, Buchfellner M, Britt W, Sanchez V, Potter JL, Ingram LA, Kelly D, A case series of children with acute hepatitis and human adenovirus infection. New England Journal of Medicine 387, 620–630 (2022). [PubMed: 35830653] 7. Rafie K, Lenman A, Fuchs J, Rajan A, Arnberg N, Carlson L-A, The structure of enteric human adenovirus 41—A leading cause of diarrhea in children. Science advances 7, eabe0974 (2021). [PubMed: 33523995] 8. Berk AJ, in Fields Virology. (Lippincott Williams & Wilkins, 2013), pp. 1704–1731. 9. Rocholl C, Gerber K, Daly J, Pavia AT, Byington CL, Adenoviral infections in children: the impact of rapid diagnosis. Pediatrics 113, e51–e56 (2004). [PubMed: 14702495] 10. Canan O, Ozçay F, Bilezikçi B, Adenovirus infection as possible cause of acute liver failure in a healthy child: a case report. The Turkish journal of gastroenterology: the official journal of Turkish Society of Gastroenterology 19, 281–283 (2008). [PubMed: 19119490] 11. Schaberg KB, Kambham N, Sibley RK, Higgins JPT, Adenovirus Hepatitis: Clinicopathologic Analysis of 12 Consecutive Cases From a Single Institution. Am J Surg Pathol 41, 810–819 (2017). [PubMed: 28296681] 12. Xu GJ, Kula T, Xu Q, Li MZ, Vernon SD, Ndung’u T, Ruxrungtham K, Sanchez J, Brander C, Chung RT, O’Connor KC, Walker B, Larman HB, Elledge SJ, Comprehensive serological profiling of human populations using a synthetic human virome. Science 348, aaa0698 (2015). [PubMed: 26045439] 13. Gupta S, Ansari HR, Gautam A, Raghava GP, Identification of B-cell epitopes in an antigen for inducing specific class of antibodies. Biology direct 8, 1–15 (2013). [PubMed: 23324625] 14. Singh H, Ansari HR, Raghava GP, Improved method for linear B-cell epitope prediction using antigen’s primary sequence. PloS one 8, e62216 (2013). [PubMed: 23667458] 15. Buus S, Rockberg J, Forsström B, Nilsson P, Uhlen M, Schafer-Nielsen C, High-resolution mapping of linear antibody epitopes using ultrahigh-density peptide microarrays. Molecular & Cellular Proteomics 11, 1790–1800 (2012). [PubMed: 22984286] 16. Mina MJ, Kula T, Leng Y, Li M, de Vries RD, Knip M, Siljander H, Rewers M, Choy DF, Wilson MS, Larman HB, Nelson AN, Griffin DE, de Swart RL, Elledge SJ, Measles virus infection diminishes preexisting antibodies that offer protection from other pathogens. Science 366, 599– 606 (2019). [PubMed: 31672891] 17. Cates J, Baker JM, Almendares O, Kambhampati AK, Burke RM, Balachandran N, Burnett E, Potts CC, Reagan-Steiner S, Kirking HL, Interim analysis of acute hepatitis of unknown etiology in children aged< 10 years—United States, October 2021–June 2022. (2022). 18. Meier AF, Fraefel C, Seyffert M, The interplay between adeno-associated virus and its helper viruses. Viruses 12, 662 (2020). [PubMed: 32575422] 19. Ho A, Orton R, Tayler R, Asamaphan P, Herder V, Davis C, Tong L, Smollett K, Manali M, Allan J, Adeno-associated virus 2 infection in children with non-AE hepatitis. Nature, 1–5 (2023). 20. Morfopoulou S, Buddle S, Montaguth OET, Atkinson L, Guerra-Assunção JA, Marjaneh MM, Chiozzi RZ, Storey N, Campos L, Hutchinson JC, Genomic investigations of unexplained acute hepatitis in children. Nature, 1–2 (2023). Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 10 21. Calcedo R, Morizono H, Wang L, McCarter R, He J, Jones D, Batshaw ML, Wilson JM, Adeno- associated virus antibody profiles in newborns, children, and adolescents. Clinical and Vaccine Immunology 18, 1586–1588 (2011). [PubMed: 21775517] 22. Zuiani A, Wesemann DR, Antibody dynamics and durability in coronavirus disease-19. Clinics in Laboratory Medicine 42, 85–96 (2022). [PubMed: 35153050] 23. Agarwal S, High-dose AAV gene therapy deaths. Nat. Biotechnol 38, 910 (2020). [PubMed: 32760031] 24. Day JW, Mendell JR, Mercuri E, Finkel RS, Strauss KA, Kleyn A, Tauscher-Wisniewski S, Tukov FF, Reyna SP, Chand DH, Clinical trial and postmarketing safety of onasemnogene abeparvovec therapy. Drug Safety 44, 1109–1119 (2021). [PubMed: 34383289] 25. Mustonen N, Siljander H, Peet A, Tillmann V, Härkönen T, Ilonen J, Hyöty H, Knip M, Group DS, Early childhood infections precede development of beta‐cell autoimmunity and type 1 diabetes in children with HLA‐conferred disease risk. Pediatric diabetes 19, 293–299 (2018). [PubMed: 28597957] 26. Shrock E, Fujimura E, Kula T, Timms RT, Lee I-H, Leng Y, Robinson ML, Sie BM, Li MZ, Chen Y, Viral epitope profiling of COVID-19 patients reveals cross-reactivity and correlates of severity. Science 370, (2020). 27. Shrock EL, Shrock CL, Elledge SJ, VirScan: High-throughput Profiling of Antiviral Antibody Epitopes. Bio-protocol 12, e4464–e4464 (2022). [PubMed: 35937932] 28. McWilliam H, Li W, Uludag M, Squizzato S, Park YM, Buso N, Cowley AP, Lopez R, Analysis tool web services from the EMBL-EBI. Nucleic acids research 41, W597–W600 (2013). [PubMed: 23671338] 29. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, Lopez R, McWilliam H, Remmert M, Söding J, Fast, scalable generation of high‐quality protein multiple sequence alignments using Clustal Omega. Molecular systems biology 7, 539 (2011). [PubMed: 21988835] 30. Goujon M, McWilliam H, Li W, Valentin F, Squizzato S, Paern J, Lopez R, A new bioinformatics analysis tools framework at EMBL–EBI. Nucleic acids research 38, W695–W699 (2010). [PubMed: 20439314] 31. Wickham H, Chang W, Henry L, Pedersen TL, Takahashi K, Wilke C, Woo K, Yutani H, Dunnington D, ggplot2: Create Elegant Data Visualisations Using the Grammar of Graphics, version 3.4.2, CRAN (2023); https://CRAN.R-project.org/package=ggplot2. 32. Kassambara A, ggpubr: ‘ggplot2’ Based Publication Ready Plots, version 0.4.0, CRAN (2020); https://CRAN.R-project.org/package=ggpubr. 33. Jeppson H, Hofmann H, Cook D, ggmosaic: Mosaic Plots in the ‘ggplot2’Framework, version 0.2.0, CRAN (2018); https://CRAN.R-project.org/package=ggmosaic. 34. Kolde R, pheatmap: Pretty Heatmaps, version 1.0.12, CRAN (2019); https://CRAN.R-project.org/ package=pheatmap. 35. Yu G, ggplotify: Convert Plot to ‘grob’ or ‘ggplot’ Object, version 0.0.5, CRAN (2020); https:// CRAN.R-project.org/package=ggplotify. 36. Pettersen EF, Goddard TD, Huang CC, Meng EC, Couch GS, Croll TI, Morris JH, Ferrin TE, UCSF ChimeraX: Structure visualization for researchers, educators, and developers. Protein Science 30, 70–82 (2021). [PubMed: 32881101] 37. R. R Core Team, R: A language and environment for statistical computing. (2013). 38. Santosh V, Musayev FN, Jaiswal R, Zárate-Pérez F, Vandewinkel B, Dierckx C, Endicott M, Sharifi K, Dryden K, Henckaerts E, The Cryo-EM structure of AAV2 Rep68 in complex with ssDNA reveals a malleable AAA+ machine that can switch between oligomeric states. Nucleic Acids Research 48, 12983–12999 (2020). [PubMed: 33270897] Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 11 Fig. 1. Breadth of AAV-specific antibody responses differed between cases and controls. (A) Shown is a heatmap of epitope binding signals (EBSs) for all peptides targeted by cases or pediatric controls (n=18,484). Raw data are available in data files S1 to S2. (B) The mean number of species-specific peptides targeted in cases, pediatric controls, and adult controls are shown for pathogens with nominally significant differences (p <0.05) as measured with Welch’s one-sided t-test with 95% confidence interval. Pathogens with statistically significant (FDR=0.05) differences after Benjamini-Hochberg correction are annotated with . (C) Pathogen-specific breadth of antibody response in cases (purple) overlaid on the distribution of breadth in pediatric controls (blue, n=38) is shown for selected common pathogens and AAV’s. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 12 Fig. 2. AAV peptides were targeted at increased frequency in cases versus controls. (A) Shown is the number of peptides targeted at a significantly higher frequency in cases than pediatric controls (Fisher’s one-way Exact test with Benjamini-Hochberg procedure and FDR of 0.05) by viral species. (B) Mosaic plots show reactivity to representative AAV-2 peptides heavily targeted in cases and previously identified herpesvirus, adenovirus, and influenza public epitopes in cases, pediatric controls, and adult controls. The number of individuals that do or do not target each peptide are indicated in boxes on the figure. Peptide-level hit data for all samples and peptides are available in data files S3 to S5. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Mitchell et al. Page 13 Fig. 3. Antibody profiles of patients with AHUE clustered according to parvovirus reactivity and showed evidence of epitope spreading. (A) Shown is a heatmap of epitope binding signals for cases and pediatric controls for Parvoviridae peptides. Each row represents a VirScan library peptide and each column is a sample. Heatmap rows and columns are ordered according to complete agglomerative clustering with corresponding dendrograms shown. The strains and protein regions from which peptide sequences were derived is annotated along the y-axis and case-control status is annotated along the x-axis. (B) The heatmap shows the response to peptides derived from AAV-2 (isolate Srivastava/1982) protein sequences in cases and pediatric controls. Sci Transl Med. Author manuscript; available in PMC 2023 September 14. Mitchell et al. Page 14 Demographic Characteristics of Cases and Controls. Table 1. Cases (N=9) Pediatric controls (N=38) Adult controls (N=470) 1 (11) 5 (56) 3 (33) 14 (37) 24 (63) 2 (22) 7 (78) 3 (33) 6 (67) 20 (53) 18 (47) 38 (100) Age – number (%) < 2 years 2 to 5 years 6 to 10 years 11 to 17 years 18 to 29 years 30 to 39 years 40 to 49 years 50 to 59 years 60 to 69 years 70 to 79 years > 80 years Sex – number (%) Male Female Race/Ethnicity – number (%) Non-Hispanic White Hispanic White Hispanic other Non-Hispanic Black Asian and Pacific islander Non-Hispanic other Hispanic Black Unknown 47 (10) 85 (18) 63 (13) 96 (20) 70 (15) 54 (11) 55 (12) 228 (49) 242 (51) 171 (36) 82 (17) 102 (22) 39 (8) 15 (3) 8 (2) 3 (1) 50 (11) Sci Transl Med. Author manuscript; available in PMC 2023 September 14. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1093_mnras_stab3539	Downloaded from orbit.dtu.dk on: Mar 07, 2025 Rapid build-up of the stellar content in the protocluster core SPT2349−56 at z = 4.3 Chapman, Ryley Hill Scott; Phadke, Kedar A.; Aravena, Manuel; Archipley, Melanie; Ashby, Matthew L N; Béthermin, Matthieu; Canning, Rebecca E A; Gonzalez, Anthony; Greve, Thomas R; Gururajan, Gayathri Total number of authors: 23 Published in: Monthly Notices of the Royal Astronomical Society Link to article, DOI: 10.1093/mnras/stab3539 Publication date: 2022 Document Version Peer reviewed version Link back to DTU Orbit Citation (APA): Chapman, R. H. S., Phadke, K. A., Aravena, M., Archipley, M., Ashby, M. L. N., Béthermin, M., Canning, R. E. A., Gonzalez, A., Greve, T. R., Gururajan, G., Hayward, C. C., Hezaveh, Y., Jarugula, S., MacIntyre, D., Marrone, D. P., Miller, T., Reuter, C., Rotermund, K. M., Scott, D., ... Weiß, A. (2022). Rapid build-up of the stellar content in the protocluster core SPT2349−56 at z = 4.3. Monthly Notices of the Royal Astronomical Society, 512(3), 4352–4377. https://doi.org/10.1093/mnras/stab3539 General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.  Users may download and print one copy of any publication from the public portal for the purpose of private study or research.  You may not further distribute the material or use it for any profit-making activity or commercial gain  You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. The stellar content of SPT2349−56 1 Rapid build-up of the stellar content in the protocluster core SPT2349−56 at z = 4.3 t t p s : / / f r o m h Ryley Hill,1 Scott Chapman,1,2,3 Kedar A. Phadke,4 Manuel Aravena,5 Melanie Archipley,4,7 Matthew L. N. Ashby,6 ID Matthieu Béthermin,8 Rebecca E. A. Canning,9 Anthony Gonzalez,10 Thomas R. Greve,11,12,13 Gayathri Gururajan,8 Christopher C. Hayward,14 Yashar Hezaveh,14,15 Sreevani Jarugula,4 Duncan MacIntyre,1 Daniel P. Marrone,16 Tim Miller,17 Cassie Reuter,4 Kaja M. Rotermund,3 Douglas Scott,1 Justin Spilker,18,19 Joaquin D. Vieira,4,7 George Wang,1 Axel Weiß20 1Department of Physics and Astronomy, University of British Columbia, 6225 Agricultural Road, Vancouver, V6T 1Z1, Canada 2National Research Council, Herzberg Astronomy and Astrophysics, 5071 West Saanich Road, Victoria, V9E 2E7, Canada 3Department of Physics and Atmospheric Science, Dalhousie University, 6310 Coburg Road, B3H 4R2, Halifax, Canada 4Department of Astronomy, University of Illinois, 1002 West Green Street, Urbana, IL 61801, USA 5Núcleo de Astronomía, Facultad de Ingeniería y Ciencias, Universidad Diego Portales, Av. Ejército 441, Santiago, 8320000, Chile 6Center for Astrophysics \| Harvard & Smithsonian, Optical and Infrared Astronomy Division, 60 Garden St., MS-66, Cambridge, MA 02138, USA 7Center for AstroPhysical Surveys, National Center for Supercomputing Applications, Urbana, IL, 61801, USA 8Laboratoire d’Astrophysique de Marseille, 38 rue Frédéric Joliot-Curie, Marseille, 13013, France 9Institute of Cosmology and Gravitation, University of Portsmouth, Dennis Sciama Building, Portsmouth, PO1 3FX, UK 10Department of Astronomy, University of Florida, 211 Bryant Space Science Center, Gainesville, FL 32611-2055, USA 11Cosmic Dawn Center 12DTU Space, Technical University of Denmark, Elektrovej 327, Kgs. Lyngby, DK-2800, Denmark 13Department of Physics and Astronomy, University College London, Gower Street, London, WC1E 6BT, UK 14Center for Computational Astrophysics, Flatiron Institute, 162 Fifth Avenue, New York, NY 10010, USA 15Département de Physique, Université de Montréal, 1375 Avenue Thérèse-Lavoie-Roux, Montréal, H2V 0B3, Canada 16Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, AZ 85721, USA 17Department of Astronomy, Yale University, 52 Hillhouse Avenue, New Haven, CT 06511, USA 18Department of Astronomy, University of Texas at Austin, 2515 Speedway, Stop C1400, Austin, TX 78712, USA 19NHFP Hubble Fellow 20Max-Planck-Institut für Radioastronomie, Auf dem Hügel 69, Bonn, D-53121, Germany a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o i / / . / l i September 2021 ABSTRACT The protocluster SPT2349−56 at z = 4.3 contains one of the most actively star-forming cores known, yet constraints on the total stellar mass of this system are highly uncertain. We have therefore carried out deep optical and infrared observations of this system, probing rest-frame ultraviolet to infrared wavelengths. Using the positions of the spectroscopically-conﬁrmed protocluster members, we identify counterparts and perform detailed source deblending, allowing us to ﬁt spectral energy distributions in order to estimate stellar masses. We show that the galaxies in SPT2349−56 have stellar masses proportional to their high star-formation rates, consistent with other protocluster galaxies and ﬁeld submillimetre galaxies (SMGs) around redshift 4. The galaxies in SPT2349−56 have on average lower molecular gas-to-stellar mass fractions and depletion timescales than ﬁeld SMGs, although with considerable scatter. We construct the stellar-mass function for SPT2349−56 and compare it to the stellar-mass function of z = 1 galaxy clusters, ﬁnding consistent shapes between the two. We measure rest-frame galaxy ultraviolet half-light radii from our HST-F160W imaging, ﬁnding that on average the galaxies in our sample are similar in size to typical star-forming galaxies at these redshifts. However, the brightest HST-detected galaxy in our sample, found near the luminosity-weighted centre of the protocluster core, remains unresolved at this wavelength. Hydrodynamical simulations predict that the core galaxies will quickly merge into a brightest cluster galaxy, thus our observations provide a direct view of the early formation mechanisms of this class of object. Key words: galaxies – formation: galaxies – evolution: submm – galaxies l D o w n o a d e d / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i © 2021 The Author(s) Published by Oxford University Press on behalf of Royal Astronomical SocietyORIGINAL UNEDITED MANUSCRIPT 2 Hill et al. 1 INTRODUCTION In the present day, the large-scale structure of our Universe is made up of ﬁlaments, nodes, and voids, a structure that is often described as the ‘cosmic web’. The nodes of this cosmic web are comprised of galaxy clusters, which are the largest gravitationally bound objects in the Universe. Being such fundamental building blocks, galaxy clusters are well-studied objects; we know that they are seeded by small-amplitude density ﬂuctuations of the sort observed in the cos- mic microwave background (CMB), which then grew and collapsed into the massive structures that we see today (e.g., Wright et al. 1992; Bennett et al. 2003; Springel et al. 2005). While the CMB is very well understood (e.g. Planck Collaboration I 2020), and the details of present-day galaxy clusters are well-described (e.g., Biviano 1998; Giodini et al. 2013; Bykov et al. 2015), the interme- diate phase of evolution, which has become known as the realm of galaxy ‘protoclusters’, still lacks suﬃcient observations to pin down the models (e.g. Overzier 2016). Traditional galaxy cluster searches have made use of the fact that these objects are virialized, allowing the intergalactic gas to heat up and be detected by X-ray facilities (e.g., Rosati et al. 2009; Gobat et al. 2011; Andreon et al. 2014; Wang et al. 2016; Mantz et al. 2018) or at millimetre wavelengths via the Sunyaev- Zeldovich (SZ) eﬀect (e.g., Planck Collaboration XXVII 2016; Bleem et al. 2015; Huang et al. 2020). However, beyond redshifts of around 2, these signatures become too faint for practical de- tection. Protocluster-detection techniques now include searching large optical and infrared sky maps for overdensities of red galax- ies (e.g., Martinache et al. 2018; Greenslade et al. 2018), Lyman- break galaxies (LBGs; e.g., Steidel et al. 2000; Dey et al. 2016), or Lyman-α emitters (LAEs; e.g., Shimasaku et al. 2003; Tamura et al. 2009; Chiang et al. 2015; Dey et al. 2016; Harikane et al. 2019), and searching the area surrounding rare and luminous sources or groups of sources such as radio-loud active galactic nuclei (AGN; e.g., Steidel et al. 2005; Venemans et al. 2007; Wylezalek et al. 2013; Dannerbauer et al. 2014; Noirot et al. 2018) or submillimetre galaxies (SMGs; e.g., Chapman et al. 2009; Umehata et al. 2015; Casey et al. 2015; Hung et al. 2016; Oteo et al. 2018; Lacaille et al. 2019; Long et al. 2020). Another protocluster-selection technique that has recently been gaining attention comes from experiments designed to map the CMB. These experiments typically aim to cover huge areas of the sky at submillimetre and millimetre wavelengths with resolution on the order of a few arcminutes, and in the process ﬁnd some of the brightest and rarest submillimetre and millimetre sources in the sky. After the application of various selection criteria to remove Galac- tic sources and quasars/blazars, follow-up observations with higher- resolution telescopes have subsequently revealed that many of the remaining sources are gravitational lenses (Negrello et al. 2010; Hezaveh et al. 2013; Cañameras et al. 2015; Spilker et al. 2016), or genuine overdensities of luminous star-forming galaxies, repre- senting ideal protocluster candidates (Planck Collaboration XXVII 2015; Flores-Cacho et al. 2016; Miller et al. 2018; Kneissl et al. 2019; Hill et al. 2020; Wang et al. 2021; Koyama et al. 2021). One such source, SPT2349−56, was discovered in the South Pole Telescope (SPT)’s extragalactic mm-wave point-source cata- logue (Vieira et al. 2010; Mocanu et al. 2013; Everett et al. 2020), and is now known to contain dozens of spectroscopically- conﬁrmed star-forming galaxies through Atacama Large Mil- limeter/submillimeter Array (ALMA; Wootten & Thompson 2009) observations (Miller et al. 2018; Hill et al. 2020), several LBGs through Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) and near-infrared wide-ﬁeld imager (FLAMINGOS-2; Eikenberry et al. 2006) observations (Rotermund et al. 2021), and a number of LAEs, alongside a Lyman-α blob (Apostolovski et al. in prep.), through observations with the Very Large Telescope (VLT)’s Multi Unit Spectroscopic Explorer (MUSE; Bacon et al. 2010). This protocluster lies at a redshift of 4.3, and based on observa- tions with the Atacama Pathﬁnder Experiment (APEX) telescope’s Large APEX BOlometer CAmera (LABOCA; Kreysa et al. 2003; Siringo et al. 2009), has an integrated star-formation rate (SFR) of over 10,000 M(cid:12) yr−1 within a diameter of about 500 proper kilo- parsecs, well above what is seen for a single cluster in a wide variety of cosmological simulations (Lim et al. 2021). Hydrody- namical simulations using the known galaxies in SPT2349−56 as the initial conditions predict that most of the galaxies in the centre of this object will merge into a single brightest cluster galaxy (BCG; Rennehan et al. 2020) over a timescale of a few hundred million years. SPT2349−56 is believed to be the core of a Mpc-scale proto- cluster, as evidenced by several infalling subhalos found around it (Hill et al. 2020). In addition to optical observations of SPT2349−56 with GMOS and FLAMINGOS-2, the Spitzer Space Telescope’s In- fraRed Array Camera (IRAC; Fazio et al. 2004) was used to ob- serve this protocluster core in the infrared (Rotermund et al. 2021). These data were used to obtain rest-frame ultraviolet photometry for nine out of the 14 galaxies originally reported by Miller et al. (2018), as well as identifying four LBGs at the same redshift as the structure. While the photometric coverage was sparse, owing to the faintness of the known galaxies in the optical, initial spectral energy distribution (SED) ﬁts suggested that the core has a stellar mass of at least 1012 M(cid:12), and the nine detected galaxies appeared to show signiﬁcant scatter around the z = 4 galaxy main sequence (MS). A search for an overdensity of LBGs out to about 1 Mpc found that the overdensity was too low to meet large-ﬁeld optical survey criteria, meaning that SPT2349−56 would not be picked up by traditional optical surveys searching for distant protoclusters. We have now signiﬁcantly bolstered our optical and infrared coverage of SPT2349−56 using the Hubble Space Telescope (HST), and by increasing our Spitzer-IRAC integration time by a factor of 10. In addition, we have now identiﬁed over 30 galaxy protoclus- ter members. In this paper, we use these new data to analyse the ultraviolet and infrared properties of this much larger sample of protocluster members. In Section 2 we describe these new observations in detail. In Section 3 we outline our data reduction procedure, including deblending, source matching, ﬂux density extraction, SED ﬁtting, and proﬁle ﬁtting. In Section 4 we show our results, in Section 5 we discuss the implications of these observations, and the paper is concluded in Section 6. 2 OBSERVATIONS 2.1 HST HST observed SPT2349−56 with the Wide Field Camera 3 (WFC3) instrument in the F110W and F160W ﬁlters during Cycle 26 (pro- posal ID 15701, PI S. Chapman). Two orbits were carried out for the F110W ﬁlter, totalling 1.6 hours on-source, and three orbits were carried out for the F160W ﬁlter, totalling 2.4 hours on-source. Since the WFC3 detector pixels undersample the PSF (the plate scale is 0.13 arcsec pixel−1, with a full width at half maximum, FWHM, around 1 pixel), the observations utilized a standard sub-pixel dither l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT pattern in order to fully sample the PSF. The ﬁeld-of-view of the WFC3 instrument is about 4.7 arcmin2, suﬃciently large to image all of the protocluster members in the core, as well as the north- ern component. Hill et al. (2020) found CO(4–3) line emission at z = 4.3 from three galaxies in a targeted ALMA Band 3 observation of a red Herschel-SPIRE source (named ‘SPIREc’) located about 1.5 Mpc from the main structure, but these three galaxies are not covered by the HST imaging. The data were calibrated using calwf3, part of the standard HST WFC3 pipeline wfc3tools available in python.1 The indi- vidual exposures were stacked using astrodrizzle; the stacking method used was the median, and sky subtraction was performed. We set the ﬁnal pixel scale to be 0.075 arcsec pixel−1, approximately Nyquist sampling (i.e. sampling by a factor of 2) the beam. The ﬁ- nal rms reached in these images is 0.50 nJy for the F110W ﬁlter (corresponding to a 5σ AB magnitude limit of 30.4), and 0.79 nJy for the F160W ﬁlter (corresponding to a 5σ AB magnitude limit of 29.9). 2 armin × 2 arcmin cutouts of the images, smoothed by a 3-pixel FWHM Gaussian, are shown in Fig. 1. 2.2 IRAC The Spitzer Space Telescope IRAC instrument was used to observe SPT2349−56 at two wavelengths: 3.6 µm and 4.5 µm. A total of four observations have now been carried out since 2009, the ﬁrst two of which were presented in Rotermund et al. (2021), where fur- ther details can be found. The two subsequent observations used in this paper were carried out in January 2018 (proposal ID 13224, PI S. Chapman) and October 2019 (proposal ID 14216, PI S. Chap- man), and uniformly covered 3.6 µm and 4.5 µm with 324 × 100 s dithered exposures (162 exposures per observation for each chan- nel). These new data provide a factor of 10 more exposure time, reducing the instrumental rms by a factor of about 3. Given IRAC’s very large ﬁeld-of-view (about 27 arcmin2), all protocluster mem- bers known to date (including the three SPIREc sources) were cov- ered by these observations. Data from all four observations were combined and set to a pixel scale of 0.6 arcsec pixel−1. The ﬁnal rms levels of the stacked data are 6.9 nJy at 3.6 µm (corresponding to a 5σ AB magnitude limit of 27.5), and 6.5 nJy at 4.5 µm (corresponding to a 5σ AB magnitude limit of 27.6). Cutouts showing the 2 armin × 2 arcmin region around the main structure are shown in Fig. 1. 2.3 Gemini The Gemini observations used in this paper were presented in Rotermund et al. (2021), where details of the imaging, calibration, and data reduction can be found. Here we only provide a brief summary. Since the data were taken before many new protocluster members had been conﬁrmed, we also outline which of these new galaxies are covered by the observations. Data were taken in the g, r, and i bands using the GMOS instrument (Hook et al. 2004), and similarly in the Ks band using the FLAMINGOS-2 instrument (Eikenberry et al. 2004), both of which are part of the Gemini South Observatory. The rms levels reached in these images are 0.2, 0.3, 0.6, and 15.0 nJy, respectively. For the GMOS instrument, the g-, r-, and i-band images have a ﬁeld-of-view of 5.5 arcmin2 and cover all of the core and northern 1 https://github.com/spacetelescope/wfc3tools The stellar content of SPT2349−56 3 sources, but not the three SPIREc galaxies located 1.5 Mpc away from the core. Apostolovski et al. (in prep.) also reports seven LAEs at z = 4.3, four of which are in the central component of the protocluster, while the remaining three are found in the northern component. These seven galaxies are also covered by the GMOS imaging, and will be included in the analysis below. Finally, Rotermund et al. (2021) identiﬁed four LBGs around the core of SPT2349−56; they identify one of the LBGs (LBG1) with a galaxy from Hill et al. (2020) found in the ALMA data (C17), one (LBG4) with an LAE (LAE3), with the remaining two LBGs being unique. However, they note that one of their unique LBGs, LBG2, lies close to galaxy C2 from Hill et al. (2020), and could also be a counterpart. In this paper, we treat LBG2 as the counterpart to C2 as it lies within the 1 arcsec search radius criteria we outline in Section 3.3, and we include the other unique LBG, LBG3, as a separate source in our analysis below. For the FLAMINGOS-2 instrument, the ﬁeld-of-view is circu- lar with a diameter of 6.1 arcmin. While this ﬁeld-of-view is large enough to cover all of the core and northern sources described above, due to a lack of nearby guide stars within the ﬁeld, we were only able to cover the central sources and a few northern sources down to a depth suitable for the detection of the target galaxies in our sample. These include the 23 galaxies from Hill et al. (2020) found through their [Cii] emission, NL1, NL3, N3, ﬁve LAEs, and the LBG. Fig. 1 shows 2 armin × 2 arcmin cutouts for each of these four Gemini ﬁelds, smoothed by a 3-pixel FWHM Gaussian. 3 DATA ANALYSIS 3.1 Gemini and HST ﬂux density measurements We ﬁrst matched our GEMINI GMOS, FLAMINGOS-2, and HST F110W astrometry to our HST-F160W astrometry using the python package astroalign (Beroiz et al. 2020), which identiﬁes bright stars in source and target images and estimates a transformation matrix that aligns the stars in the target image with the same stars in the source image. We note that we are unable to apply this step to our ALMA data because there are no bright sources easily identiﬁable in both our HST-F160W imaging and our ALMA imaging; however, as shown below, after matching sources across both images we do not ﬁnd any signiﬁcant systematic oﬀsets. We ran source-extractor (Bertin & Arnouts 1996) on our GEMINI-GMOS and FLAMINGOS-2 images to extract a catalogue of sources and measure their ﬂux densities. The detection images were smoothed with circular Gaussian kernels having FWHMs of three pixels, equivalent to 0.225 arcsec, as part of the source detection process. Sources were required to consist of at least 3 connected pixels lying 0.8σ above the local background. – in source-extractor this means setting DETECT_MINAREA to 3 and DETECT_THRESH to 0.8. These parameters are similar to the ones used by Rotermund et al. (2021), who set DETECT_MINAREA to 3 and varied DETECT_THRESH between 1.1 and 2.5, depending on the band. The photometry was measured in the unsmoothed images at the locations of the sources identiﬁed in the smoothed detection images using the FLUX_BEST option in source-extractor, which selects either the ﬂux measured in an adaptively-scaled elliptical aperture (FLUX_AUTO) or sums the pixels found above the chosen thresh- old, with a correction for missing ﬂux in the wings of the objects (FLUX_ISOCOR). When crowding becomes an issue, the FLUX_AUTO apertures start double counting the ﬂux densities in pixels. For a l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT 4 Hill et al. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 1. Optical and infrared images of SPT2349−56. ALMA sources are shown as blue squares, the LBG source as a blue circle, LAE sources as blue triangles, and submm-detected sources with no line emission (i.e. likely foreground/background sources) as red squares. The images are smoothed by a 3-pixel FWHM Gaussian, except for the IRAC images, which have not been smoothed. Single-dish submm imaging of SPT2349−56 at 870 µm using the LABOCA instrument are shown as the background contours, starting at 4σ and increasing in steps of 5σ, and deﬁne the core and northern regions of this protocluster ﬁeld (see Hill et al. 2020; Wang et al. 2021). ORIGINAL UNEDITED MANUSCRIPT23:49:4845423936-56:37:003038:0030RADec90kpcgCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpcrCoreNorthALMAsourceLBGsourceLAEsourceNoline23:49:4845423936-56:37:003038:0030RADec90kpciCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpcF110WCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpcF160WCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpcKsCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpc3.6µmCoreNorth23:49:4845423936-56:37:003038:0030RADec90kpc4.5µmCoreNorth given source, if 10 per cent of the FLUX_AUTO measurement comes from other sources, FLUX_ISOCOR is chosen, otherwise FLUX_AUTO is chosen. HST WFC3 source catalogues were produced using source-extractor as well, using the same input parameters described above, except we required nine adjacent pixels to be above 1.2 times the background rms. This setup is similar to the source-extractor parameters used in the Cosmic Assem- bly Near-infrared Deep Extragalactic Legacy Survey (CANDELS; Galametz et al. 2013) in the ‘hot’ mode, which is optimized to detect small and faint galaxies. 3.2 IRAC deblending and ﬂux density measurements Owing to Spitzer’s poor angular resolution compared to Gemini and HST, source blending can become a serious issue, especially in crowded regions like the centre of SPT2349−56. To tackle this is- sue, we use the publicly-available code t-phot (Merlin et al. 2015, 2016), which uses a high-resolution source catalogue as a prior to deblend a low-resolution, blended image. To construct an optimal high-resolution catalogue, we stacked (as a weighted mean) both of our F110W and F160W HST images. We then ran source-extractor with a higher detection threshold (DETECT_MINAREA = 9 and DETECT_THRESH = 2.0), since having too many faint galaxies used as priors with t-phot leads to unre- alistic deblending. We also turned oﬀ source deblending by setting DEBLEND_MINCONT to 1. This means that within a collection of pixels found above the predeﬁned threshold, local maxima will not be classiﬁed as individual sources. This step is necessary because t-phot becomes unreliable when given blended priors. We output a segmentation map from source-extractor, which is used in conjunction with the catalogue in t-phot. The code t-phot requires that the pixel scale of the input prior catalogue is an integer multiple of the low-resolution image, and is also aligned to the same pixel grid. To satisfy these criteria, we used the same astroalign code to align our IRAC data to our high-resolution HST data, and then we used the python function reproject_interp from the module reproject to reproject our IRAC images onto our combined HST image, thus making the IRAC pixels a factor of 8 smaller. t-phot also requires a convolution kernel that can be used with the high-resolution segmentation image to produce the low- resolution template. Assuming that the galaxies in our HST image are resolved, the appropriate kernel would simply be the IRAC PSF. However, owing to the variation in sensitivity across each IRAC pixel, a more complex point response function (PRF) is more appropriate. A set of 5 × 5 PRFs are available from the NASA/IPAC Infrared Science Archive,2 where each PRF is the response from a point source illuminating a position on a pixel divided up into a 5 × 5 grid. For the convolution kernel, we selected the PRF corresponding to a point source illuminating the centre of a pixel. The provided PRFs have a pixel scale of 0.24 arcsec pixel−1, and we used the same reproject function to reproject these pixels to the high-resolution pixel scale of 0.075 arcsec pixel−1. Some sources of interest are not detected in the combined HST image, but are clearly seen in both channels of the IRAC data. For these sources, t-phot enables one to provide a second catalogue of unresolved priors as a list of positions that will be treated as delta The stellar content of SPT2349−56 5 functions before being convolved with the kernel. Five galaxies from Hill et al. (2020) ﬁt this category, and were provided to t-phot as unresolved priors, namely C4, C5, C9, C10, and NL1. With these inputs in hand, we then ran t-phot in two passes. The ﬁrst pass convolves the high-resolution segmentation image with the kernel, and ﬁts an amplitude to each source to best match the low-resolution image. The second pass cross-correlates the model image with the low-resolution image in order to compute small positional shifts for each source, and then ﬁts for the amplitudes a second time. Since our high-resolution HST imaging did not cover the SPIREc sources, their ﬂux densities were measured independently. We performed aperture photometry at the positions of the three SPIREc sources, using an aperture with a radius of 4 pixels (2.4 arcsec), and applied aperture corrections of 1.208 and 1.220 for the 3.6 µm band and the 4.5 µm band, respectively (from the IRAC Instrument Handbook3). SPIREc1 and SPIREc3 are both clearly detected and we were able to measure their ﬂux densities with high signal-to-noise ratio (SNR). SPIREc2 is not a clear detection by eye, and the ﬂux density in the aperture at 3.6 µm is consistent with noise, but at 4.5 µm we were able to measure a > 2σ signal. 3.3 Source matching In order to perform a multiwavelength analysis of the SEDs of these galaxies, we need to match our sources detected in high-resolution (sub)millimetre imaging to their counterparts in our optical and infrared imaging. This becomes complicated due to the fact that the (sub)millimetre imaging is detecting emission from dust, while the ultraviolet to near-infrared imaging is detecting emission from starlight (including extinction by dust). A galaxy’s morphology can be quite complicated in detail, with patches of dust and stars that will not necessarily overlap one another from our line of sight (e.g. Goldader et al. 2002). Additionally, although we have tied the all of the optical and infrared astrometry to the HST F160W frame, we are not able to perform this step to our submm imaging, although the ALMA astrometry relative to that of HST has been measured to be accurate within 0.1 arcsec (e.g., Dunlop et al. 2017; Franco et al. 2018). Theoretically, the 1σ uncertainty of our ALMA-derived posi- tions is given by ∆RA = ∆Dec = 1 √ 2 ln 2 FWHM SNR , (1) where FWHM is the size of the synthesized beam, and SNR is the signal-to-noise ratio of the peak pixel (Ivison et al. 2007). For our ALMA positions, the synthesized beam is about 0.5 arcsec (Hill et al. 2020), so we expect the 1σ angular position uncertainty to be ∆θ ≈ 0.3 arcsec/SNR. Since the probability density of ﬁnd- ing a source at an angle θ from its true position is proportional to θe−θ2/2∆θ2 , one must go out to a distance of 3.42∆θ in order to ﬁnd a correct match with 99.7 per cent (or 3σ) certainty. For our ALMA positions, the lowest (spatial) SNR used to measure a position was about 4, meaning a 3σ search should go out to about 0.3 arcsec. However, this becomes more complicated in practice due to the fact that the (sub)millimetre and optical images could have diﬀerent morphologies, especially for the case of mergers. Previous studies matching ALMA galaxies to HST counterparts have used radial 2 https://irsa.ipac.caltech.edu/data/SPITZER/docs/irac/ calibrationfiles/psfprf/ 3 https://irsa.ipac.caltech.edu/data/SPITZER/docs/irac/ iracinstrumenthandbook/1/ l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT 6 Hill et al. searches around 0.6–1 arcsec (e.g., Dunlop et al. 2017; Long et al. 2020), and here we also adopt 1 arcsec, corresponding to about 6.9 proper kiloparsecs. For comparison, the typical submm size of the galaxies in our sample is about 4 proper kiloparsecs (twice the half-light radius, see Hill et al. 2020). A source is thus considered a match if it lies within 1 arcsec from the ALMA-derived positions provided in Hill et al. (2020), and we allow the possibility of multiple matches; multiple optical counterparts to a submm source are possible if the submm source is a merging galaxy, for example. For cases where a counterpart is less than 1 arcsec from two ALMA galaxies, we assign the match to the closest ALMA galaxy. Despite the comparatively large beamsize of the IRAC imaging, we use the same 1 arcsec matching criteria, since the IRAC positions are nearly identical to the high-resolution HST imaging, which is simply a result of using t-phot for the photometry. Also, for reference, 1 arcsec corresponds to 6.3 pixels in the GMOS images, 13.4 pixels in our WFC3 images, 5.6 pixels in the FLAMINGOS-2 image, and 1.7 pixels in the IRAC images. Next, we take advantage of our extensive wavelength coverage by imposing the additional constraint that a counterpart is not de- tected in the g band – this is simply because at z = 4.3, the g band probes a galaxy’s rest-frame SED at 900 Å, where these galaxies should be much fainter than the sensitivity limit of our Gemini data because of neutral hydrogen’s eﬃciency at absorbing light at this wavelength. In other words, if we ﬁnd a counterpart within 1 arcsec that is also bright in the g band, we assume that it is a line-of-sight interloper, and remove the match from our sample. This criteria removes the matches to C1, C11, C20, C23, and N3. For the LAEs, this criteria removes LAE1. A galaxy 0.4 arcsec east of LAE4 is also detected in the g band, but we see that in the r and i bands an extension appears 0.4 arcsec to the north of this source, which our HST data resolves as a second galaxy. Since the eastern galaxy is detected in the g band but the northern one is not, we only call the northern galaxy a match to LAE4. The Gemini-detected source near C1 (bright in the g band and removed from our sample) was conﬁrmed to be a foreground z = 2.54 galaxy by Rotermund et al. (2021) using spectroscopy with the VLT, validating our submm-matching criteria. However, the ob- served [Oiii] line strength suggests a stellar mass much smaller than what the IRAC ﬂux densities (2.6 and 2.9 µJy at 3.6 and 4.5 µm, respectively) would imply. Quantitatively, Rotermund et al. (2021) place an upper limit on the stellar mass of the foreground galaxy of < 1.6 × 109 M(cid:12) based on an [Oiii] linewidth of 53 km s−1. Zhu et al. (2010) propose a scaling relation between stellar mass and IRAC continuum ﬂux densities using g − r colours as a proxy for star- formation history, which we use to place an upper limit to the con- tribution of the measured IRAC ﬂux density from this foreground galaxy. We measure a colour of g − r = 0.10, corresponding to up- per limits of νLν < 7.2 × 108 L(cid:12) and νLν < 5.0 × 108 L(cid:12) at 3.6 and 4.5 µm, respectively, or S3.6 < 0.06 µJy and S4.5 < 0.05 µJy. Thus, since this foreground galaxy likely contributes less than 2 per cent to the measured IRAC ﬂux density at the position of C1, we assign all of the measured IRAC ﬂux density to C1. Lastly, we look by eye for consistent matches across all the data and make sure sources are not double-matched. C3, C12, C13, C16, C22, and C23 are very close to C6, which is clearly the dominant galaxy in the core of the protocluster at the wavelengths covered by HST. In the IRAC imaging, C6 and these six sources are blended within a single beam, yet we expect that nearly all of the measured ﬂux density can be attributed to only C6, as it is > 25 times brighter than its surrounding galaxies in the HST imaging. We thus assign all of the ﬂux density within a larger 1.5 arcsec region to C6, while providing upper limits for the nearby galaxies. In Appendix A, Fig. A1 (available online), we show the result- ing positional diﬀerences between our matching criteria outlined above and the ALMA positions given in Hill et al. (2020), deﬁned as the ALMA position minus the optical/infrared position. We ﬁnd counterparts in at least one image to 21/29 ALMA-identiﬁed galax- ies, as well as the single LBG, and 4/6 LAEs. Four galaxies are found to have two counterparts in our F110W image, three have two counterparts in our F160W image, and one has three counterparts in our F110W image. In Fig. A1 (available online) we show the mean oﬀset found in each band surrounded by a circle whose semi-major and semi-minor axes are equal to the standard deviations of the oﬀsets in each direction. We ﬁnd a slight oﬀset in the negative x direction of ∆RA ≈ − 0.1 arcsec (consistent across all wavebands, since the astrometry has been tied to a single frame); this could mean that there is a slight mismatch between the HST astrometry and the ALMA astrometry, but an adjustment of this size would have no eﬀect on our results. We investigate the purity of our source matches by running our matching algorithm on random locations within each map, allowing us to calculate a false positive rate by taking the ratio of the matches to the number of random locations tested. Using 1000 random lo- cations, we ﬁnd false positive rates of 0.08, 0.06, 0.23, 0.20, and 0.12 for the r, i, F110W, F160W, and Ks bands, respectively. For reference, our detection rates are 0.32, 0.34, 0.61, 0.47, and 0.25 for the same bands, which are factors of 2–5 higher than the false positive rates. These values are upper limits to the true false positive rates, since we are requiring an optical/infrared-detected galaxy to be less than 1 arcsec from a submm source (i.e. an ALMA-detected galaxy), which is much less common than a match between two optical/infrared-detected galaxies. For cases where we found no counterpart match, we estimate upper limits by calculating the mean ﬂux density measurement un- certainty within a given band. We also add the local background to these non-detection upper limits, calculated within a 3 pixel- diametre circular region centred on the source. This takes into ac- count the fact that C1 is blocked by a foreground galaxy in the GMOS and HST imaging (see Rotermund et al. 2021), so the only upper limit we can place is that it must be fainter than the in- terloping foreground galaxy. This is also necessary in the IRAC imaging where confusion and source blending is a reason for many non-detections around galaxy C6. Lastly, we make a 2σ cut to the photometry measurements of the matched sources. In Table 1 we provide the resulting ﬂux densities and upper limits for each source, and in Appendix B (available online) we show 12 × 12 arcsec cutouts of each source at each wavelength. We note that there are known bad pixels in our HST data near C21 (in both the F110W and F160W ﬁlters as it is a property of the WFC3 detector), but they are outside of the aperture used to measure the ﬂux density of this source. We compare our resulting optical measurements to the ﬂux densities reported in Rotermund et al. (2021), who used slightly diﬀerent apertures (a ﬁxed diametre of 1.6 arcsec for the g, r, and i images, and FLUX_AUTO for the Ks image) and source-extractor parameters. We ﬁnd that we do not detect C2 in the Ks band, nor C6 in the i band, but that we detect sources C2 and C14 in the r image, and C2 and C14 in the i image. In terms of the overlap- ping detections, we ﬁnd good agreement between the ﬂux density measurements. In the IRAC infrared imaging, we identify coun- terparts to all the galaxies in Rotermund et al. (2021), except C3 and C13 that are blended with C6, which is expected because the data used here are signiﬁcantly deeper. The IRAC measurements l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT reported in Rotermund et al. (2021) are systematically larger than those reported here, by average factors of 1.7 and 2.1 at 3.6 and 4.5 µm, respectively. This is also expected, since Rotermund et al. (2021) did not attempt to deblend the IRAC sources while here we did, meaning that the ﬂux inside their ﬁxed apertures should be larger on average due to surrounding source leakage. 3.4 SED ﬁtting To estimate the physical properties of the galaxies belong- ing to the SPT2349−56 protocluster system, we used CIGALE (Burgarella et al. 2005; Noll et al. 2009; Boquien et al. 2019) to ﬁt SEDs to the available photometry. This includes all of the photo- metric data provided in Table 1, as well as the 850 µm, 1.1 mm, and 3.2 mm photometry from Hill et al. (2020). We also input 1σ upper limits to the photometry for non-detections, and include Her- schel-SPIRE 250, 350, and 500 µm constraints from Miller et al. (2018). CIGALE models a galaxy’s rest-frame optical and ultra-violet spectrum using simple stellar population models with variable star- formation histories, including nebular emission lines, and incor- porates a ﬂexible dust attenuation curve that allows the slope and strength of the ultraviolet bump to vary. The thermal dust emission is modeled assuming a power-law dust temperature distribution, and energy balance is imposed such that the energy absorbed by dust, primarily at rest-frame ultraviolet and optical wavelengths, is approximately equal to that re-radiated in the infrared, allowing for discrepancies due to non-isotropy from the ultraviolet and optical emission. In our ﬁts we assumed a delayed star-formation history with a single exponential timescale. The SFR as a function of time is parameterized by SFR ∝ 1 τ2 SFH et/τSFH, (2) where τSFH is the timescale, eﬀectively the time at which the star- formation peaks, and a free parameter of the model. The total current stellar mass, M∗, in this model is found by varying the duration of the star-formation, assuming a Chabrier initial mass function (IMF; Chabrier 2003) with solar metallicity. The assumed star-formation history can have an eﬀect on the resulting stellar mass (up to a factor of 2, see Michałowski et al. 2012), although we did not ﬁnd a large variation in the results after testing several of the available models. Another free parameter is the dust attenuation, given by the amount of extinction present in the V band in magnitudes, AV , which we model using the Calzetti et al. (2000) attenuation curve with a vari- able power-law slope. The dust is modeled following Draine et al. (2014), where the dust is separated into a diﬀuse component heated by the interstellar radiation ﬁeld, and a compact component linked to star-forming regions and heated by a variable radiation ﬁeld. The total dust mass sets the overall normalization. To obtain posterior distributions for the parameters in the ﬁts, CIGALE generates a grid of possible SEDs, and calculates χ2 for each SED. These χ2 values are then translated to a global likeli- hood function, assuming the likelihood is proportional to e−χ2/2. Marginalized posterior distributions are then calculated for the free parameters, and CIGALE returns the mean values and standard devi- ations of these distributions. The resulting stellar masses are given in Table 2, and the best-ﬁt SEDs are shown in Appendix C (available online). Where parameter uncertainties overlap with 0, we provide 1σ upper limits. For galaxies C12, C16, C20, C22, C23, LAE1, and LAE2, only upper limits are available for their photometry across The stellar content of SPT2349−56 7 all wavelengths (both submm and optical/infrared), so we do not attempt to ﬁt SEDs and derived stellar masses. The best-ﬁt CIGALE ttoal stellar masses range from about 1010– 1011 M(cid:12). The Rotermund et al. (2021) stellar masses are larger than those estimated here in proportion to their reported IRAC ﬂux densities (they diﬀer because we use t-phot to improve the IRAC ﬂux density estimates). As a ﬁnal check, from the ﬁts we calculated the SFR averaged over the past 100 Myr, and compared these to the SFRs estimated from ﬁtting modiﬁed blackbody functions to the far-infrared photometry in Hill et al. (2020); the two estimates are in good agreement considering the simple SFH adopted by our SED models, with a median ratio of far-infrared SFR-to-CIGALE SFR of 1.1. 3.4.1 Galaxy N3 In Fig. B1 (available online) we see that the centroid of N3 lies right on top of a bright and fully resolved spiral galaxy. N3 was initially identiﬁed by Hill et al. (2020) using ALMA in Band 3 through the detection of a CO line at 86.502 GHz, consistent with the frequen- cies of the CO lines detected in the other protocluster galaxies, and it was assumed that the CO transition was 4–3, placing the redshift at 4.3. However, the coincident position of the CO emission with a resolved spiral galaxy could mean that the CO emission observed is actually from another transition at lower redshift. While an al- ternative explanation is that there is a genuine protocluster galaxy responsible for the CO emission behind this nearby spiral galaxy, here we investigate the photometric redshift of the spiral galaxy to see if it could in fact emit a CO line at the frequency where a line was observed. To do this, we used all of our available photometry (Table 1 plus the photometry in Hill et al. 2020) to estimate a photometric redshift using CIGALE. When CIGALE is not provided a spectro- scopic redshift, it includes a photometric redshift as an additional free parameter, generating an extra dimension of redshifted SEDs before calculating the χ2 of each model. Photometric redshift un- certainties are computed by converting the χ2 of each model to a likelihood, and computing the corresponding marginalized proba- bility distribution for the redshift. The resulting photometric redshift is found to be 1.7 ± 0.3; the only CO transition consistent with this redshift is the 2–1 transition at 230.538 GHz, which would make the spectroscopic redshift of N3 1.665. Since this redshift matches the photometric redshift quite well, for the remainder of this paper we take this galaxy to be a foreground source at z = 1.665 and remove it from the subsequent analyses. 3.4.2 Submm sources with no line detection Three galaxies were found in our ALMA observations of SPT2349−56 through their continuum only (designated NL in Hill et al. 2020), making them potential line-of-sight interlopers. In order to verify this hypothesis, we ran CIGALE on the complete set of photometry available for these galaxies (Table 1, and the photometry provided in Hill et al. 2020) in order to estimate pho- tometric redshifts. For NL1, we found a photometric redshift of 4.3 ± 0.7; while this is consistent with the redshift of SPT2349−56, we do not include it in any subsequent analyses because the uncer- tainties are still large. Further follow-up will be needed to conﬁrm the membership of this source. For the remaining two galaxies, NL2 has a photometric redshift of 2.7 ± 0.7, and for NL3 the photometric redshift is 2.2 ± 0.5, so these are indeed most likely galaxies in the foreground of the protocluster. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT 8 Hill et al. Table 1. Gemini-GMOS and FLAMINGOS-2, HST-WFC3, and Spitzer-IRAC ﬂux density measurements for all SPT2349−56 galaxies. The names are the same as in Hill et al. (2020), while the names from Miller et al. (2018) are given in brackets for reference. Upper limits are 1σ. Here ellipses indicate where data are not available for a given source. Name C1 (A) C2 (J) C3 (B) C4 (D) C5 (F) C6 (C) C7 (K) C8 (E) C9 (I) C10 (H) C11 (L) C12 C13 (G) C14 (N) C15 C16 C17 (M) C18 C19 C20 C21 C22 C23 NL1 NL3 N1 N2 N3 NL2 SPIREc1 SPIREc2 SPIREc3 LBG3 LAE1 LAE2 LAE3 LAE4 LAE5 LAE6 LAE7 LAE8 <0.10 <0.021 <0.063 <0.27 <0.21 <0.015 <0.010 0.240±0.010 0.223±0.017 <0.034 <0.028 <0.031 Sa 0.78 [µJy] <0.045 Sa 0.63 [µJy] <0.043 Sb 1.5 [µJy] <0.103 Sb 1.1 [µJy] <0.057 <0.013 <0.019 <0.019 <0.014 <0.012 <0.054 <0.060 <0.054 <0.064 <0.055 <0.014 <0.022 <0.017 <0.022 <0.024 <0.006 <0.006 <0.001 <0.019 <0.013 0.071±0.013 0.183±0.020 0.295±0.009 0.381±0.020 Sc 2.1 [µJy] <0.19 <0.25 <0.26 <0.17 <0.22 <0.44 0.51±0.18 <0.31 <0.22 <0.57 <0.006 <0.014 <0.003 0.061±0.011 <0.009 <0.055 <0.052 <0.060 0.362±0.006 1.601±0.012 4.78±0.28 0.082±0.010 0.196±0.019 Sa 0.48 [µJy] <0.032 <0.010 <0.006 <0.002 <0.003 <0.006 <0.006 <0.006 <0.006 <0.005 <0.013 <0.007 <0.007 <0.012 <0.011 <0.013 <0.009 <0.012 <0.006 <0.016 <0.013 <0.011 <0.013 <0.005 0.024±0.008 0.146±0.020 0.261±0.014 0.242±0.015 1.37±0.34 <0.034 0.037±0.004 <0.030 <0.034 0.028±0.004 0.037±0.008 0.051±0.011 0.164±0.013 0.167±0.015 0.45±0.13 0.076±0.021 0.147±0.009 0.213±0.018 <0.034 0.120±0.012 0.317±0.025 0.371±0.009 0.551±0.020 0.99±0.19 Sd Sd 4.5 3.6 [µJy] [µJy] 2.9±0.3 2.6±0.3 3.2±0.3 2.7±0.4 <5.6 <5.4 1.1±0.3 1.1±0.3 1.0±0.3 0.9±0.3 9.8±0.3 9.8±0.4 2.2±0.3 2.1±0.3 4.2±0.3 3.9±0.4 1.3±0.3 0.9±0.3 0.7±0.3 0.7±0.3 <1.3 <1.2 <4.0 <3.8 <4.3 <4.5 1.8±0.3 2.1±0.3 1.1±0.3 1.3±0.3 <2.7 <3.0 1.2±0.4 1.6±0.5 <0.8 <0.7 <1.0 <0.9 <1.8 <1.7 1.1±0.4 1.3±0.4 <1.0 <1.0 <1.3 <1.4 0.9±0.3 1.6±0.3 8.8±0.4 10.4±0.3 3.0±0.3 2.2±0.3 0.7±0.3 1.0±0.3 1.176±0.018 1.561±0.027 1.924±0.039 5.644±0.027 8.289±0.044 10.76±0.46 18.8±0.4 22.3±0.4 1.9±0.4 0.061±0.006 0.072±0.009 0.180±0.016 0.337±0.012 0.442±0.018 1.7±0.3 13.0±0.3 19.4±0.3 . . . 0.7±0.3 6.8±0.3 <1.1 <0.7 <0.9 <0.3 <0.4 <0.3 <0.5 <0.4 <1.2 <0.032 <0.026 0.351±0.016 0.373±0.009 0.343±0.014 <0.030 <0.034 <0.026 <0.088 <0.051 <0.047 <0.033 0.050±0.005 <0.061 0.044±0.006 0.066±0.008 0.53±0.17 <0.8 4.9±0.3 <1.1 <0.8 <1.0 <0.3 <0.5 <0.4 <0.5 <0.5 <1.6 . . . . . . <0.011 <0.031 <0.007 <0.005 <0.021 <0.010 <0.013 <0.008 <0.003 0.115±0.007 0.199±0.017 0.112±0.019 0.094±0.007 0.093±0.014 0.230±0.012 0.415±0.019 0.542±0.033 1.628±0.018 3.308±0.029 5.15±0.36 <0.017 <0.007 <0.010 <0.027 <0.006 <0.016 <0.012 <0.040 <0.005 0.054±0.011 <0.028 0.077±0.011 0.437±0.024 0.251±0.014 0.329±0.020 0.082±0.012 0.139±0.018 0.204±0.010 0.052±0.006 0.107±0.010 0.057±0.011 0.078±0.006 0.089±0.009 <0.17 <0.16 <0.23 <0.32 <0.37 <0.19 <0.30 . . . . . . <0.36 <0.37 <0.31 <0.12 <0.056 <0.026 <0.016 <0.043 . . . . . . . . . <0.17 0.030±0.004 0.052±0.007 <0.011 <0.022 <0.021 <0.055 <0.051 <0.058 <0.012 <0.024 <0.021 <0.057 <0.052 <0.006 <0.006 <0.062 <0.061 <0.010 <0.013 <0.006 <0.003 <0.009 <0.009 <0.015 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . aGemini-GMOS continuum ﬂux densities at 0.48 µm (the g band), 0.63 µm (the r band), and 0.78 µm (the i band). bHST-WFC3 continuum ﬂux densities at 1.1 µm (the F110W ﬁlter) and 1.5 µm (the F160W ﬁlter). cGemini-FLAMINGOS-2 continuum ﬂux densities at 2.1 µm (the Ks band). dSpitzer-IRAC continuum ﬂux densities at 3.6 µm and 4.5 µm. 3.5 Rest-frame ultraviolet proﬁle ﬁtting We next investigate the morphological proﬁles of some of the brighter galaxies detected in our F160W image (observed-frame 1.54 µm, or 290 nm in the rest-frame). We are interested in the char- acteristic sizes of the unobscured stellar emission, compared to their sizes as seen in the submm, where the emission is due to dust and star-formation, thus we choose the longest wavelength covered by our HST data. However, 290 nm is still in the ultraviolet, where most of the ﬂux density is due to young O and B stars. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT Table 2. Best-ﬁt properties of the galaxies in our sample. Here dashes indicate non-detections, while ellipses indicate where data are not available for a given source. The stellar content of SPT2349−56 9 Name C1 (A) C2 (J) C3 (B) C4 (D) C5 (F) C6 (C) C7 (K) C8 (E) C9 (I) C10 (H) C11 (L) C12 C13 (G) C14 (N) C15 C16 C17 (M) C18 C19 C20 C21 C22 C23 NL1 NL3 N1 N2 N3 NL2 SPIREc1 SPIREc2 SPIREc3 LBG3 LAE1 LAE2 LAE3 Ra 1/2,UV [kpc] – 1.26±0.76 – – – 0.47±0.16 – 1.45±0.71 – – – – – – – – 1.76±0.83 – – – – – – – – 1.04±0.87 – – – . . . . . . . . . 1.09±0.62 – – – Rb 1/2,CII [kpc] 2.91±0.02 2.57±0.03 1.33±0.04 1.90±0.01 2.22±0.02 1.30±0.04 2.22±0.03 1.21±0.03 1.42±0.03 1.14±0.03 1.40±0.04 – 1.00±0.09 0.39±0.10 – – 1.59±0.06 – – – – – – – – . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . µd gas 0.34±0.32 0.46±0.21 – – – 0.31±0.10 0.33±0.22 0.21±0.08 – 0.77±0.57 – – 0.11±0.10 0.06±0.04 – – – – – – – – – – – 2.56±0.96 4.84±2.68 – – 0.17±0.15 – 0.20±0.18 – – – – τe dep [Gyr] 0.074±0.018 0.100±0.026 0.046±0.011 0.053±0.013 0.024±0.007 0.058±0.014 0.119±0.033 0.051±0.013 0.049±0.014 0.060±0.016 0.084±0.043 – 0.043±0.011 0.046±0.030 – – – – – – – – – – – 0.075±0.019 0.066±0.022 – – – – – – – – – M c ∗ [1010 M(cid:12)] 22.2±21.3 4.5±1.9 <112.6 <24.0 <21.8 10.9±3.4 3.0±1.9 11.1±4.1 <18.0 1.5±1.1 <6.7 – 7.5±6.8 3.4±1.3 1.2±0.6 – 1.2±0.5 0.6±0.4 <4.8 – 1.0±0.5 – – – – 4.7±1.7 1.0±0.6 – – 39.3±34.0 <4.5 12.3±11.0 0.6±0.4 – – <0.1 0.5±0.4 0.1±0.1 <0.8 0.4±0.3 <0.8 – – – – – – – – – – – – – – – . . . . . . . . . . . . . . . LAE4 LAE5 LAE6 LAE7 LAE8 aBest-ﬁt rest-frame ultraviolet half-light radius, obtained by ﬁtting Sérsic proﬁles to all sources detected in our HST F160W imaging with suﬃcient SNR (see Section 3.5). bBest-ﬁt [Cii] proﬁle half-light radius, obtained by ﬁtting Sérsic proﬁles to [Cii]-detected sources after staking all channels containing line emission (see Hill et al. 2020 and Section 3.6). cBest-ﬁt stellar mass from ﬁtting SEDs using the photometry in Table 1 and from Hill et al. (2020), obtained using CIGALE. dMolecular gas-to-stellar mass fraction, µgas = Mgas / M∗, with Mgas values taken from Hill et al. (2020). eDepletion timescale, τdep = Mgas / SFR, with Mgas and SFR values taken from Hill et al. (2020). l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / / . 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT 10 Hill et al. We created 2 arcsec × 2 arcsec cutouts around each protocluster member galaxy in our sample (except for NL3 and LAE1 where we made 3 arcsec × 3 arcsec cutouts), and ﬁt elliptical 2D Sérsic proﬁles (allowing the Sérsic index to vary) convolved with the HST beam to the sources with optical counterparts containing pixels greater than 5 times the background rms; the relevant sources are C2, C6, C8, C17, N1, and LBG3. While C21 reaches >5 times the background rms as well, we remove it from this analysis due to the nearby bad pixels. We use the HST F160W PSF available from the Space Telescope Science Institute (STScI) instrumentation website,4 taking the PSF corresponding to the response from a point source illuminating the centre of a pixel. The available PSF model supersamples the pixel plate scale of 0.13 arcsec by a factor of 4, and we regrid the PSF model to match the pixel scale of our F160W imaging. This PSF is convolved with each galaxy’s proﬁle to produce our models; this is the same modelling technique used in popular proﬁle-ﬁtting packages such as galfit (Peng et al. 2002, 2010). The resulting beam-deconvolved half-light radii (the length of the semi-major axes of an ellipse containing half the total ﬂux density) are provided in Table 2, and our models are shown in Appendix D (available online). We ﬁnd half-light radii ranging from 1.1 kpc to 1.8 kpc, except for one outlying source, C6, where we ﬁnd a half-light radius of 0.5 kpc. For reference, the corresponding half-light radius of the HST F160W beam is about 0.5 kpc, implying that only galaxy C6 is not resolved. The Sérsic indices range from 0.3 to 2.3. Since most of our HST detections are faint and cannot be ﬁt on an individual basis, we performed a stacking analysis in order to evaluate the average rest-frame ultraviolet proﬁle. As we ultimately want to compare this to the average submm proﬁle, we focused on the subset of our sample with corresponding high-resolution ALMA data at 850 µm (as detailed in the section below). These include all of the galaxies in the core region of SPT2349−56 (C1–C23). We removed C1 from the analysis as the HST detection at its position is a foreground galaxy, and we removed C21 due to the nearby bad pixels. We also excluded C6 since it is clearly an outlier in terms of brightness at this wavelength, and this galaxy will be subject to its own separate analysis. The position at which to centre each cutout for the stack is crucial. In order to obtain an unbiased image of the rest-frame ultraviolet light of submm sources, one should centre the rest-frame ultraviolet images on the positions of the submm sources. However, in practice there are physical oﬀsets between rest-frame ultraviolet light and submm light, so this may not be the best choice. Here we choose to centre the cutouts at the position of the peak HST counterpart for cases where one is detected, and otherwise at the position of the peak pixel in each galaxy’s average [Cii] map, which provides the highest positional accuracy owing to the brightness of the line (see Hill et al. 2020). We also masked pixels above 25σ (after verifying that this did not mask any sample sources) to remove bright nearby objects not associated with the galaxies in the stack. The resulting image is shown in Fig. 2. Following the same steps as above, we ﬁt a Sérsic proﬁle to the stack, and our best-ﬁt model and residual are shown alongside the data. We ﬁnd a half-light radius of 1.24 ± 0.29 kpc and a Sérsic index of 0.75 ± 0.60, consis- tent with the sizes found for the individual galaxies and an expo- nential proﬁle of n = 1. Similarly, we ﬁnd an axis ratio of 1.0 ± 0.3, as expected for stacking random orientation angles. Systematic uncertainties in structural parameters are known to 4 https://www.stsci.edu/HST/instrumentation/wfc3/ data-analysis/psf be important for low SNR sources; for example, van der Wel et al. (2012) found that basic size parameterizations can be determined for galaxies detected in the F160W ﬁlter down to about 24.5 mag. For reference, the galaxies in our sample with a peak pixel SNR > 5 used to obtain size measurements span a range of 23.4–25.7 mag, thus we wish to investigate possible systematic uncertainties and biases. To do this, we simulate F160W maps at the SNRs of the sources where we have ﬁt Sérsic proﬁles. Our simulation covers a grid of Sérsic indices from 0.3 to 2.5 and half-light radii from 0.4 to 2 kpc, and for each Sérsic index and half-light radius we generate three independent maps. Sérsic proﬁles are convolved with the F160W beam, and Gaussian random noise is added to the background such that the peak pixel has the SNR of the given source. For each input half-light radius we calculate the mean and standard deviation of the recovered half-light radii for all values of input Sérsic indices (eﬀectively marginalizing over this parameter). For the wide range of Sérsic indices tested, our algorithm recovers an unbiased estimate of the true half-light radius at all SNRs tested, with scatters of around 0.6 kpc at SNR = 5 and 0.4 kpc at SNR = 10. In Table 2 we include this systematic uncertainty in quadrature with the statistical errors. The uncertainties in our Sérsic indices are large (often overlapping with 0), and a similar test of the recovered Sérsic indices from our simulation indicates systematic uncertainties of order ± 2 at a SNR of 5. For the lowest SNR sources our best-ﬁt Sérsic indices are therefore not likely meaningful, however at the SNRs of our stack and C6 (16 and 112, respectively), the systematic uncertainties are smaller than the statistical uncertainties. 3.6 [Cii] sizes Following the method outlined above and in Hill et al. (2020), we also ﬁt Sérsic proﬁles to cutouts of each galaxy’s extended [Cii] emission. Line emission channels were determined from lower- resolution, deeper ALMA data by ﬁtting Gaussian proﬁles to the spectra, from which we averaged the high-resolution channels from −3σ to 3σ (where σ is the standard deviation of the best-ﬁtting linewidth), or for cases where two Gaussians was a better ﬁt, from −3σL to +3σR, where σL and σR are from the left and right Gaus- sian ﬁts, respectively. 2 arcsec × 2 arcsec cutouts were made around each source, and models were ﬁt to sources with pixels detected above 5 times the background rms by convolving a Sérsic proﬁle with the data’s synthesized beam. We allowed the position, position angle, ellipticity, half-light radius, and Sérsic index to vary in our ﬁts. The results are provided alongside our ultraviolet proﬁle ﬁts in Table 2, and the models are shown in Appendix E (available online). We ﬁnd half-light radii in the range of 1.0 to 2.9 kpc, except for one outlying source, C14, which has a half-light radius of about 0.4 kpc. For reference, the half-light radius of the ALMA synthesized beam is about 0.7 kpc, implying that only C14 is not resolved. The Sérsic indices found range from 0.3 to 2.0. We next determined the average submm-continuum proﬁle fol- lowing the same stacking procedure done for our HST data. High resolution continuum maps were already presented in Hill et al. (2020), obtained by stacking the line-free channels. We selected the same galaxies as in the HST analysis (i.e. each core galaxy except for C1, C6, and C20), and centred the cutouts on the peak of the average [Cii] map. We then ﬁt a Sérsic proﬁle to the stack, and found a half-light radius of 1.18 ± 0.01 kpc with a Sérsic index of 1.22 ± 0.03. Similarly, we ﬁnd an axis ratio of 0.9 ± 0.1, overlap- ping with 1 as expected for stacking random orientation angles. The stack, along with the ﬁt, is shown in Fig. 2. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 11 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l Figure 2. Top row: Stacked F160W image of all the core galaxies in our sample (except C1, C6, and C21; see the text for details). Contours start at 2σ and increase in steps of 4σ. The red dot indicates the position of the best-ﬁt centre, and the red line shows the length and position angle of the best-ﬁt half-light diametre. The middle panel shows the best-ﬁt Sérsic proﬁle, after convolution with the beam and the pixel window function. The residual map is shown at right, with the same contour levels as the left panel. Bottom row: Same as the top row, but stacking our high-resolution 850-µm ALMA continuum images. Contours again start at 2σ and increase in steps of 4σ. 4 RESULTS 4.1 The galaxy main sequence A primary quantity of interest is the galaxy main-sequence (MS), which is the SFR of a galaxy as a function of its stellar mass (e.g. Elbaz et al. 2011), and is commonly used to identify star- bursts and quenched galaxies and therefore can place these pro- tocluster galaxies in the context of galaxy evolution. Recently, Rotermund et al. (2021) estimated stellar masses for 14 member galaxies of SPT2349−56 using SED ﬁts to Gemini and Spitzer pho- tometry, albeit shallower than presented here, and compared these to the z = 4.3 MS. With our improved optical, infrared, and mm- wavelength coverage, as well as a more detailed IRAC deblending algorithm and an expanded catalogue of protocluster galaxies, we are able to expand on this work in much greater detail. In Fig. 3 (top left panel) we show the stellar masses derived in this paper as a function of SFR. The SFRs shown here were derived in Hill et al. (2020) by ﬁtting modiﬁed blackbody distributions to the far-infrared photometry observed by ALMA, since the SFRs produced from our SED ﬁtting could be less reliable due to dust obscuration. In the ﬁts, β was ﬁxed to 2, and the dust tempera- ture, Td, was ﬁxed to 39.6 K (the temperature consistent with the mean ratio of the measured 850 µm ﬂux density to the measured 3.2 mm ﬂux density of the sample). In Hill et al. (2020) the best-ﬁt SEDs were then integrated between 42 and 500 µm to obtain far- infrared luminosities, but here we integrate from 8 to 1000 µm to obtain more complete infrared luminosities in order to be consis- tent with comparison samples; this increases the luminosities by a factor of 1.17. These values were converted to SFRs using a factor of 0.95 × 1010 M(cid:12) yr−1 L−1 (from Kennicutt 1998, modiﬁed for a (cid:12) Chabrier initial mass function; see Chabrier 2003). The core galax- ies of SPT2349−56 (deﬁned as those lying within a 90 kpc-radius of the far-infrared luminosity-weighted centre, where the primary- beam response of the ALMA observation used to ﬁnd protocluster members falls to 0.5, see Hill et al. 2020) are of interest because simulations predict that they will merge into a BCG on a timescale of a few hundred Myr (Rennehan et al. 2020). We have highlighted these galaxies in black in Fig. 3. We also show the LAEs and LBGs of this sample as squares in order to distinguish them from the primary submm-selected galaxies making up the core of this protocluster. For comparison, we also show in Fig. 3 the galaxies found in a similar star-forming protocluster, the Distant Red Core (DRC; Oteo et al. 2018; Long et al. 2020), found at redshift 4. For these protocluster galaxies, the stellar masses were also obtained through optical and infrared SED ﬁtting, while infrared luminosities were derived by scaling the mean template from the ALMA follow- up programme of the LABOCA ECDF-S Submillimetre Survey (ALESS; Simpson et al. 2014) to match their continuum observa- i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT−0.5000.0000.500∆α−0.5000.0000.500∆δstackR1/2=1.24±0.29kpcn=0.75±0.60DataHSTcontinuumOpticalposition−0.5000.0000.500∆α−0.5000.0000.500∆δModel−0.5000.0000.500∆α−0.5000.0000.500∆δResidual−0.5000.0000.500∆α−0.5000.0000.500∆δstackR1/2=1.18±0.01kpcn=1.22±0.03DataALMAcontinuumSubmmposition−0.5000.0000.500∆α−0.5000.0000.500∆δModel−0.5000.0000.500∆α−0.5000.0000.500∆δResidual 12 Hill et al. tions at 2 mm; we applied the same scale factor used in our work to obtain SFRs. To see how these protocluster galaxies compare with ﬁeld galaxies, we focus on samples of high-redshift SMGs, since the SFRs of SMGs typically exceed 100 M(cid:12), which accounts for the majority of our sample. We show the z > 3.5 SMGs from the ALESS survey, originally selected as bright 870-µm point sources in the Extended Chandra Deep Field South (ECDF-S; Simpson et al. 2015), with stellar masses derived by da Cunha et al. (2015) by ﬁt- ting SEDs to optical and infrared photometry, and for the SFRs we converted the infrared luminosities from Swinbank et al. (2014) (obtained by ﬁtting a modiﬁed blackbody SED to the available far-infrared photometry) to SFRs using a conversion factor of 0.95 × 10−10 M(cid:12) yr−1 L−1 . We also show a sample of ﬁeld SMGs (cid:12). around z = 4.4 from Scoville et al. (2016) that were initially selected in a representative fashion from the Cosmic Evolution Survey (COS- MOS) ﬁeld; for these galaxies, stellar masses were also estimated from SED ﬁts to optical and infrared photometry, and the SFRs were derived from rest-frame ultraviolet and infrared continuum measurements, adopting a factor of 2 uncertainty as recommended in the paper. The ﬁnal ﬁeld sample in this comparison comes from a follow-up survey of 707 SMGs detected in the Ultra Deep Survey (UDS) ﬁeld (Dudzeviči¯ut˙e et al. 2020), from which we have taken all galaxies with photometric redshifts between 4 and 5. Here the stellar masses and far-infrared luminosities were obtained by ﬁtting SEDs to photometric data ranging from rest-frame optical to radio wavelengths, and we have converted the far-infrared luminosities to SFRs using the standard conversion factor. While we cannot entirely rule out the possibility that some of these ﬁeld galaxies are in fact in protocluster environments, none of them are in known protoclus- ters, nor are any of them located in environments as overdense in the submm as SPT2349−56 or the DRC. Next, we show galaxies from the ALPINE survey (Le Fèvre et al. 2020; Béthermin et al. 2020; Faisst et al. 2020) be- tween z = 4.4 and z = 4.7; for the ALPINE galaxies, stellar masses were obtained by ﬁtting SEDs to optical and infrared photome- try, and we have taken their infrared luminosities (scaled from 850 µm continuum detections assuming a model template of z ≈ 4 MS galaxies) and converted these to SFRs using the same factor of 0.95 × 1010 M(cid:12) yr−1 L−1 . We then show the best-ﬁt z = 4.5 MS (cid:12) obtained from the ALPINE survey (?); we note that the parameter- ization found by the ALPINE survey is consistent with previously- derived MS parameterizations from e.g. Speagle et al. (2014) at z = 4.3. For reference, we show a scatter of a factor of 2 around the MS, the intrinsic scatter proposed by Schreiber et al. (2015). We see that the galaxies in SPT2349−56 follow the z = 4.5 MS derived by the ALPINE survey, although with considerable scatter, along with the other samples of ﬁeld SMGs and the star- forming galaxies from the ALPINE survey. To investigate this in detail, in Fig. 3 (top right panel) we show the SFRs in SPT2349−56 divided by the SFRs predicted by the ALPINE MS for each galaxy’s measured stellar mass. We include the ﬁeld SMGs from the samples described above, as well as the individual star-forming galaxies from the ALPINE survey. In order to assess the diﬀerence between protocluster galaxies to ﬁeld galaxies around z = 4, we combine our sample of proto- cluster galaxies with those from the DRC, and compare them to the ALESS, COSMOS, and UDS SMGs. Since the SFR is log- normally distributed at a given stellar mass (e.g. Speagle et al. 2014; Schreiber et al. 2015), we compute the weighted mean and weighted standard deviation of the logarithm of the two samples. The weighted mean log SFR / SFRMS for the protocluster galaxies is 0.13, with a weighted standard deviation of 0.46, while for the ﬁeld SMGs the weighted mean log SFR / SFRMS is 0.16, with a weighted standard deviation of 0.43. These values are plotted on Fig. 3 for reference. on We t-test perform an unequal-variance the log SFR / SFRMS values of each sample. Assuming the two samples in question are drawn from Gaussian distributions, the unequal-variance t score is a statistic used to test the hypothesis that the two distributions have equal means and arbitrary variances. It is worth noting that we are only testing the means of the two samples, meaning that even if they have overlapping scatter, we may still reject the null hypothesis that the means are equal. Using this test, we ﬁnd a p-value of 0.72, which can be interpreted as the probability that the means are the same. From this comparison we cannot reject the null hypothesis that the protocluster galaxies in these samples are diﬀerent from ﬁeld galaxies in terms of the MS. However, we emphasize that there could be large systematic errors present, owing from diﬀerences in modelling the SEDs used to ﬁt the available photometry to obtain stellar masses; for example, diﬀerent functional forms for the SFH can be assumed, and there are numerous models of dust extinction available. 4.2 Molecular gas-to-stellar mass fractions Another mass measurement available for the protocluster galaxies in SPT2349−56 is the molecular mass, and a useful evolutionary diagnostic is to see if the molecular gas-to-stellar mass fraction scales with the total stellar mass built-up so far. We deﬁne the molecular gas-to-stellar mass fraction as µgas = Mgas M∗ , (3) where Mgas is the molecular gas mass and M∗ is the stellar mass. We use the molecular gas masses derived in Hill et al. (2020). Brieﬂy, the CO(4–3) transition was observed by ALMA, and line strengths were measured for sources where the line was de- tected. The CO(4–3) line strength was converted to a CO(1–0) line strength using a factor of 0.60 (the mean line strength ratio of the SPT-SMG sample from Spilker et al. 2014), and then this was converted to a molecular gas mass using a conversion factor of αCO = 1 M(cid:12)/(K km s−1 pc2), similar to other studies of SMGs (see e.g., Aravena et al. 2016; Bothwell et al. 2017). The resulting molecular gas-to-stellar mass fractions are given in Table 2; we ﬁnd values ranging from about 0.04 to 5. Interestingly, since our stel- lar masses are lower than the values provided by Rotermund et al. (2021), we estimate larger values of µgas, which is in better agree- ment with the simulations analyzed by Lim et al. (2021). However, these simulations still predict that z ≈ 4 protocluster galaxies have µgas values between 1 and 5, which is only reached by two galaxies (N1 and N2) in our sample. Figure 3 (bottom left panel) shows our molecular gas-to- stellar mass fractions as a function of stellar mass (with the core galaxies again highlighted), compared with the galaxies from the DRC (Oteo et al. 2018; Long et al. 2020), where we have converted their CO(6–5) line intensities to molecular gas masses using an L (cid:48) factor of 0.46 (the mean ratio found for the SPT-SMG (6−5) sample, see Spilker et al. 2014) and an αCO of 1 M(cid:12)/(K km s−1 pc2) (the same scale factor used for our sample). /L (cid:48) (1−0) To compare protocluster galaxies with ﬁeld galaxies around z = 4, we require a sample with molecular gas masses estimated with similar CO transition tracers. A recent CO survey of SMGs selected from the COSMOS ﬁeld, the UDS ﬁeld, and the ALESS l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 13 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 3. Top left: Stellar mass as a function of SFR (i.e. the galaxy main sequence) for all galaxies in the SPT2349−56 protocluster where we could obtain stellar mass estimates from CIGALE SED ﬁtting with uncertainties not overlapping with 0. The SFRs shown here were estimated in Hill et al. (2020). Galaxies highlighted in black are part of the core of the structure, deﬁned as those lying within a radius of 90 kpc of the far-infrared luminosity-weighted centre (see Hill et al. 2020). Also shown are z > 3.5 SMGs from the ALESS survey (da Cunha et al. 2015), a sample of SMGs around z = 4.4 from the COSMOS ﬁeld (Scoville et al. 2016), z = 4–5 SMGs from the UDS ﬁeld (Dudzeviči¯ut˙e et al. 2020), z = 4.4–5.9 galaxies from the ALPINE survey (Béthermin et al. 2020), and a ﬁt of the MS at z = 4.5 from the ALPINE survey (Khusanova et al. 2021), including an intrinsic scatter of ± 0.3 dex proposed by Schreiber et al. (2015). Lastly, we show the galaxies from a similar star-forming z = 4 protocluster known as the DRC (Long et al. 2020). Top right: Measured SFRs divided by the SFR expected for a given stellar mass (SFRMS), assuming the MS relation from the ALPINE survey (Khusanova et al. 2021), shown for the same galaxies in the top-left panel. Since SFR is distributed log-normally for a given stellar mass, the pink horizontal line and shaded region shows the weighted mean and standard deviation of log SFR / SFRMS for the protocluster galaxies, respectively (combining our sample with the galaxies from the DRC), while the grey horizontal line and shaded region shows the weighted mean and standard deviation of log SFR / SFRMS for ﬁeld SMGs, respectively (combining the galaxies from the ALESS and COSMOS surveys). Bottom left: Molecular gas-to-stellar mass fraction, µgas (Eq. 3), as a function of stellar mass, using molecular gas masses from Hill et al. (2020). Our protocluster sample is compared with the ﬁeld SMGs from Birkin et al. (2021), where molecular gas masses have been derived from CO observations similar to our sample, and the DRC. We also include the galaxies from the ALPINE survey by converting their published [Cii] luminosities to molecular gas masses following the prescription outlined in Dessauges-Zavadsky et al. (2020). The weighted mean and standard deviation of log µgas for all the protocluster galaxies is shown as the horizontal pink line and shaded region, respectively, and the weighted mean and standard deviation of log µgas for the ﬁeld galaxies is shown as the horizontal grey line. Bottom right: Depletion timescale, τdep (Eq. 4), as a function of stellar mass, for the same galaxies shown in the top-left panel, along with the same weighted means and standard deviations of the logarithms of the subsamples as horizontal lines and shaded regions, respectively. sample spanning z = 1–5 was carried out by Birkin et al. (2021), and we have selected the galaxies from this survey with z > 3.5 to use as a ﬁeld comparison here; this corresponds to four galaxies from the COSMOS ﬁeld, two galaxies from the UDS ﬁeld, and nine galaxies from the ALESS sample. All of the galaxies in this sample were originally detected as bright submm sources in large single- dish surveys with SCUBA-2 and LABOCA, and have extensive multiwavelength follow-up observations. The detected lines range from CO(5–4) to CO(2–1), and we use the mean line ratios of the sample to convert these line intensities to the CO(1–0) transition, with the CO(2–1)-to-CO(1–0) ratio ﬁxed to 0.9 (for reference, the mean CO(4–3)-to-CO(1–0) ratio was found to be 0.32, compared to 0.60 used for the SPT2349−56 galaxies). We then adopt an αCO of 1 M(cid:12)/(K km s−1 pc2) for this reference sample. Stellar masses are ORIGINAL UNEDITED MANUSCRIPT101010111012M∗[Mﬂ]1101001000SFR[Mﬂyr−1]SPT2349−56DRCFieldSMGs,ALESSFieldSMGs,COSMOSFieldSMGs,UDSFieldgalaxies,ALPINEMainsequence,ALPINE101010111012M∗[Mﬂ]0.1110100SFR/SFRMSSPT2349−56DRCFieldSMGs,ALESSFieldSMGs,COSMOSFieldSMGs,UDSFieldgalaxies,ALPINEMeanprotoclusterSMGsMeanfieldSMGs109101010111012M∗[Mﬂ]0.010.1110µgas=Mgas/M∗SPT2349−56DRCFieldSMGs,Birkinetal.2021Fieldgalaxies,ALPINEMeanprotoclusterSMGsMeanfieldSMGs109101010111012M∗[Mﬂ]0.010.11τdep=Mgas/SFR[Gyr]SPT2349−56DRCFieldSMGs,Birkinetal.2021Fieldgalaxies,ALPINEMeanprotoclusterSMGsMeanfieldSMGs 14 Hill et al. provided for each source, derived by ﬁtting SEDs to the extensive photometry available in these ﬁelds. For comparison with a sample of high-z galaxies that are not SMGs, we include the galaxies from the ALPINE survey here by deriving molecular gas masses from the published [Cii] luminosities, following the prescription outlined in Dessauges-Zavadsky et al. (2020); however, since the [Cii] line is a diﬀerent tracer, we cannot provide any quantitative comparisons with the samples of CO-derived gas masses. We again combine our sample of protocluster galaxies with the DRC, and compare this to the sample from Birkin et al. (2021). Looking at Fig. 3 (bottom left panel) we see that the molecular gas-to-stellar mass also appears log-normally distributed, so we calculate the weighted mean and weighted standard deviation of the logarithm of each sample. The weighted mean log µgas of pro- tocluster galaxies is −0.45, with a weighted standard deviation of 0.54, while for the ﬁeld galaxies we ﬁnd a weighted mean log µgas of −0.05, with a weighted standard deviation of 0.72 (see Fig. 3). A similar unequal-variance t-test on the log µgas values results in a p-value of 0.04. While these results do provide evidence that the mean molecular gas-to-stellar mass fractions of protocluster galax- ies are not the same as those of ﬁeld galaxies (the null hypothesis can be rejected with > 95 per cent conﬁdence), we nonetheless note that there could still be systematic uncertainties unaccounted for in this analysis. 4.3 Depletion timescales A similar quantity of interest is the gas depletion timescale, which is a measure of the amount of time required to convert all of the available mass in gas into mass in stars if the current star-formation rate were to remain constant. This quantity is deﬁned as τdep = Mgas SFR , (4) where this time Mgas is divided by the SFR. In Table 2 we provide estimates of the depletion timescale for each galaxy with a measurement of molecular gas mass (via CO(4–3) line detection) and SFR (via far-infrared photometry). The galaxies in SPT2349−56 have depletion timescales ranging from 0.05 to 0.1 Gyr. In Fig. 3 (bottom right panel) we show our depletion timescales as a function of stellar mass. In order to compare with the same z > 3.5 ﬁeld SMGs from Birkin et al. (2021), we take their pro- vided far-infrared luminosities, obtained by ﬁtting SEDs with the available photometry, and multiply them by the usual factor of 0.95 × 1010 M(cid:12) yr−1 L−1 . We also show the protocluster galaxies (cid:12) from the DRC, and we include the same ALPINE galaxies for ref- erence to a sample of non-SMGs. We ﬁnd that τdep / [Gyr] the depletion timescales appear on average smaller in protocluster galaxies than in ﬁeld SMGs. We take the same log-normal approach to quantify this diﬀerence, ﬁnding that the weighted mean log (cid:16) (cid:17) for the protocluster galax- ies (again combining SPT2349−56 and the DRC) is −1.28, with a weighted standard deviation of 0.18, and the weighted mean log (cid:16) (cid:17) for the ﬁeld galaxies from Birkin et al. (2021) is −0.85, with a weighted standard deviation of 0.26 (see Fig. 3). We therefore ﬁnd that the depletion timescales for protocluster galaxies are smaller than for ﬁeld galaxies, as there is very little overlap be- tween the two distributions. The resulting p-value from an unequal- variance t-test is 5.3 × 10−5, thus we can reject the null hypothesis that the mean values of the two populations are the same. Lastly, τdep / [Gyr] the star-forming galaxies from the ALPINE survey have longer de- pletion timescales than all of the SMGs in this comparison. 4.4 The stellar mass function Our large sample of protocluster galaxies with stellar mass esti- mates allows us to compute the stellar mass function of star-forming galaxies in several diﬀerent ways. Of most interest is what the stellar mass function of the whole protocluster looks like in comparison to lower-z clusters. However, we know that the core galaxies in SPT2349−56 will merge into a single BCG within a few hundred Myr (Rennehan et al. 2020), so we would also like to know what the stellar mass function will look like after the merger by summing their masses. To start, we address the stellar mass completeness of our sam- ple. Most of the galaxies in our sample were selected from line emission surveys in the submm, but others were found through their Ly-α emission or selected as LBGs, making a detailed completeness calculation diﬃcult. Therefore, we begin by simply considering the initial sample of [Cii] and CO(4–3)-selected galaxies (C1–C23, N1, and N2), and include LBG3 as this galaxy also shows signiﬁcant [Cii] emission (Rotermund et al. 2021). We do not include the three SPIREc sources, since they are at a large projected distance where our submm imaging is not complete. Next, in Table 2 we see that 15 of the 26 galaxies in this sub- sample have stellar mass estimates available, thus a lower limit to the completeness of all of our stellar masses is about 60 per cent. We also have stellar mass upper limits available for most of the undetected sources, however they are of the order < 1011 M(cid:12) and are not very constraining. Instead, we can turn to our IRAC 3.6 µm imaging, which is a good tracer of stellar mass, probing rest-frame wavelengths of 680 nm with good resolution where the large pop- ulations of low-mass stars emit. In Fig. 4 we show our derived stellar masses as a function of S3.6 for all of the galaxies in our subsample with both quantities available. We see a tight correlation between the two quantities, as expected, and a best-ﬁt power law of the form M∗ = A (cid:0)S3.6 / S3.6,0(cid:1)γ, with S3.6,0 ﬁxed to 1 µJy, gives A = (8.6 ± 1.4) × 109 M(cid:12) and γ = 1.2 ± 0.1. Using this functional form, we can take the 3.6 µm ﬂux density measurements and upper limits for the 11 remaining galaxies with no stellar mass measure- ments and calculate their expected stellar masses and stellar mass upper limits. We again ﬁnd that all of the galaxies have upper limits of < 1011 M(cid:12), meaning that our stellar mass subsample is 100 per cent complete above 1011 M(cid:12), while ﬁve galaxies have upper limits of < 1010 M(cid:12), so our stellar mass subsample is about 80 per cent complete above 1010 M(cid:12). However, this completeness estimate does not take into account potentially variable dust extinctions between the galaxies, and given the large overall uncertainties in stellar mass estimates, we are not able to provide completeness corrections, so below we present the number counts of our complete subsample and stress the uncertainties. Figure 5 (top panels) shows the diﬀerential and cumulative number counts of the stellar masses of all the galaxies in our re- stricted sample. We have normalized our counts by the volume of a sphere of radius 360 kpc, which is the distance between the core and northern components of SPT2349−56 (see Hill et al. 2020). The true normalization is highly uncertain, but in this analysis we are only interested in the shape of the mass function. We compare this protocluster star-forming stellar mass function to the stellar mass function of a typical z (cid:39) 1 galaxy cluster (van der Burg et al. 2013), obtained by stacking cluster galaxies from a sample of 10 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT clusters (including their BCGs) within 1 Mpc of the cluster cores. (van der Burg et al. 2013) separate star-forming galaxies from ﬁeld galaxies, but we have taken their total number counts as we ex- pect our sample to be more representative of the total counts of SPT2349−56. We set the normalizing volume to be the volume of 10 spheres of radius 1 Mpc, and again emphasize that the absolute normalization here is uncertain but has no eﬀect on the shape of the counts. While this reference sample contains galaxies within a larger proper volume than probed by our ALMA data (1 Mpc versus 360 kpc), previous studies of z < 1 galaxy clusters have found that best-ﬁt parameterizations of stellar mass number counts do not vary considerably when calculated within radii ranging from 0.5 R500 to 2 R500 (where R500 is around 1 Mpc, see van der Burg et al. 2018); the slope, α, ranges from −0.8 to −1.0, while the characteristic mass, M(cid:63), remains constant within the uncertainties. Therefore, we expect our conclusions would remain unchanged if the z (cid:39) 1 com- parison sample were limited to galaxies within 360 kpc, as with our sample. Qualitatively, we see that the shapes of the number counts are in agreement with one another for masses above 1010 M(cid:12). Next, we compute the diﬀerential and cumulative number counts of the molecular gas masses in SPT2349−56 using the values provided in Hill et al. (2020) for the same galaxies in our restricted sample. These functions are shown alongside the stellar mass func- tion in Fig. 5 (top panels). We ﬁnd that the molecular gas mass function eﬀectively tracks the stellar mass function. We then assess the state of the number counts after the merger of the BCG galaxies by summing the masses of all the galaxies within 90 kpc of the far-infrared luminosity-weighted centre (i.e. the region simulated by Rennehan et al. 2020 where the mergers will take place), and treating this as a single point. These are shown alongside our other number counts in Fig. 5 (bottom panels) for comparison. We see that the shapes become linear (in log-log space), and the ﬁnal mass of the merged galaxies is comparable to the masses of the largest galaxies in the z = 1 sample. The amplitude of our counts is also larger than the z = 1 cluster counts at these high- mass bins, although since the normalizations are highly uncertain, we cannot draw any conclusions from this excess. We next ﬁt our protocluster mass functions to functional forms using a maximum-likelihood approach (e.g., Marshall et al. 1983; Wall et al. 2008; Hill et al. 2020), where we minimize the negative log-likelihood of our stellar mass measurements assuming that all sources were selected from data of equal depth: S = −2 ln L = −2 N (cid:213) i=1 ln φ(Mi) + 2V ∫ Mb Ma φ(M)dL + C. (5) In this equation N is the sample size, φ(M) is the model diﬀerential stellar mass number count (in units of M−1 Mpc−3), V is the volume (cid:12) of the survey, Ma and Mb are the mass limits of the sample (which we take to be between the smallest and largest masses in the sample), and C is a constant independent of the model. We investigate two models, a single power-law and a Schechter function. The single power-law has two free parameters, a normalization and a power- law index; explicitly, φ(M) = φ(cid:63) (cid:18) M M(cid:63) (cid:19) α , (6) where α is the slope of the power law, φ(cid:63) is the overall normaliza- tion, and M(cid:63) is ﬁxed to 1010 M(cid:12) and is not a free parameter of the model. For the Schechter function, M(cid:63) is not ﬁxed but treated as a free parameter, describing the point at which the number counts The stellar content of SPT2349−56 15 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / / . 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 4. The stellar mass as a function of S3.6 for all galaxies in our sample with with both measurements available. We ﬁnd a tight correlation between the two quantities, and ﬁt a power law of the form M∗ = A (cid:0)S3.6 / S3.6,0(cid:1)γ , with S3.6,0 ﬁxed to 1 µJy (shown as the solid line). We use this func- tional form to estimate the completeness of our stellar mass sample, ﬁnding 100 per cent completeness above 1011 M(cid:12), 80 per cent completeness above 1010 M(cid:12)., and 60 per cent completeness for the whole sample. transition from a power law to an exponential: φ(M) = φ(cid:63) (cid:19) α (cid:18) M M(cid:63) e−M/M (cid:63) . (7) We use a Markov chain Monte Carlo (MCMC) approach to minimize the log-likelihood, and calculate the odds ratio between a single power-law ﬁt and a Schechter ﬁt (simply LSchechter/LPower−law) to assess which model better-describes the data. We ﬁnd that a Schechter function is more appropriate for the total protocluster number counts, and that a single power-law func- tion is more appropriate for the protocluster containing a BCG. The resulting ﬁt parameters are provided in Table 3 (where the values are the means of the posterior distributions, and the uncertainties are 68 per cent conﬁdence intervals), and in Fig. 5 we show the best-ﬁt functions, where the shaded regions were calculated by varying the best-ﬁt parameters within their 68 per cent conﬁdence intervals and taking the largest diﬀerence. function of to the stellar mass We would now like to quantitatively compare our num- ber counts z = 1 clusters found by van der Burg et al. (2013). To do this, we note that van der Burg et al. (2013) provide a best-ﬁt Schechter function to the stellar mass number counts measured for a z = 1 cluster. They ﬁnd a characteristic stellar mass of (5.2+1.1 −0.2) × 1010 M(cid:12) and a slope of −0.46+0.08 (we ignore the normalization here because it is highly −0.26 uncertain for our sample). For reference, in Fig. 5 we show their best-ﬁt Schechter function as a shaded region encompassing the uncertainties of their best-ﬁt parameters. The characteristic stellar mass found for our total protocluster is (10.6+2.9 −7.1) × 1010 M(cid:12) and the slope found is −0.4+0.4 , which is not statistically diﬀerent from −0.4 the parameters of van der Burg et al. (2013) given the uncertainties, although we emphasize that the uncertainties are large both due to our small sample size and incompleteness and so we cannot make any further conclusions about the evolution of the star-forming stel- lar mass function. ORIGINAL UNEDITED MANUSCRIPT110S3.6[µJy]101010111012M∗[Mﬂ] 16 Hill et al. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 5. Top: Molecular gas mass and stellar mass diﬀerential (left panel) and cumulative (right panel) number counts for SPT2349−56. The stellar mass number counts for z ∼ 1 clusters from van der Burg et al. (2013) is shown in cyan for comparison. Best-ﬁt Schechter functions are shown for our diﬀerential counts, where the shaded regions are calculated by varying the best-ﬁt parameters within their 68 per cent conﬁdence intervals and taking the largest diﬀerence. The best-ﬁt Schechter function from van der Burg et al. (2013) is shown as well, with the shaded region calculated in the same way. Bottom: Since the core galaxies will merge into a single BCG in a few hundred Myr (Rennehan et al. 2020), we show the molecular gas mass and stellar mass diﬀerential (left panel) and cumulative (right panel) number counts after summing the masses of the galaxies within 90 kpc of the far-infrared luminosity-weighted centre and treating this as a single point. We also show best-ﬁt power-law functions as shaded regions, calculated in the same way as above, and the same best-ﬁt Schechter function for z ∼ 1 clusters from van der Burg et al. (2013). Table 3. Best-ﬁt mass function parameters (Eqs. 6 and 7) derived from the stellar mass and molecular gas mass number counts for all of the galaxies in SPT2349−56, and SPT2349−56 after the central galaxies merge into a BCG, estimated by summing the masses of the central galaxies and treating them as a single source. Component Function φ(M) (cid:17) α Protocluster, stellar mass φ(cid:63) (cid:16) M M (cid:63) φ(cid:63) (cid:16) M Protocluster, gas mass M (cid:63) φ(cid:63) (cid:16) M M (cid:63) φ(cid:63) (cid:16) M M (cid:63) BCG, stellar mass BCG, gas mass (cid:17) α e−M /M (cid:63) e−M /M (cid:63) (cid:17) α (cid:17) α aThis parameter was ﬁxed during the ﬁtting. Mpc−3] φ(cid:63) [10−10 M−1 (cid:12) 8+1 −8 22+6 −18 7+1 −6 39+15 −37 M (cid:63) [1010 M(cid:12)] 10.6+2.9 −7.1 4.9+1.1 −3.3 1a 1a α -0.4+0.4 −0.4 -0.1+0.3 −0.5 -1.2+0.5 −0.5 -1.6+0.4 −0.6 ORIGINAL UNEDITED MANUSCRIPT109101010111012Mass[Mﬂ]10−1210−1110−1010−910−8dN/dM[M−1ﬂMpc−3]M∗,SPT2349−56protoclusterSchechterfitMgas,SPT2349−56protoclusterSchechterfitM∗,z=1clusterSchechterfit109101010111012Mass[Mﬂ]10−1100101101N(>M)[Mpc−3]M∗,SPT2349−56protoclusterMgas,SPT2349−56protoclusterM∗,z=1cluster109101010111012Mass[Mﬂ]10−1210−1110−1010−910−8dN/dM[M−1ﬂMpc−3]AfterafewhundredMyrM∗,SPT2349−56mergedPower-lawfitMgas,SPT2349−56mergedPower-lawfitM∗,z=1clusterSchechterfit109101010111012Mass[Mﬂ]10−1100101102N(>M)[Mpc−3]AfterafewhundredMyrM∗,SPT2349−56mergedMgas,SPT2349−56mergedM∗,z=1cluster 4.5 Ultraviolet versus far-infrared sizes Many studies have looked at the physical extent of stellar emis- sion compared to dust emission in SMGs (e.g. Simpson et al. 2015; Lang et al. 2019), ﬁnding that the dust emission (and hence the star formation), probed by rest-frame far-infrared observations, is typically more smooth and compact, while the optical stellar emis- sion is clumpy and extended, likely due to patchy dust attenuation (e.g. Cochrane et al. 2019). Although the HST imaging probes a slightly diﬀerent population of stars in the ultraviolet, we empha- size that HST is still currently the best facility for performing these measurements on objects at such high redshift. The forthcoming James Webb Space Telescope (JWST) will operate at the longer wavelengths needed to resolve the average stellar populations in SPT2349−56, and we will carry out the measurements when the facility becomes available. On the other hand, our 850-µm ALMA observations probe the rest-frame at 160 µm where the dust is ex- pected to be bright. There are also known correlations between a galaxy sizes and quantities such as SFR, stellar mass, and redshift, the most studied likely being the size-mass relation (e.g., Shen et al. 2003; van der Wel et al. 2014; Mowla et al. 2019). All these corre- lations can be examined in a high-redshift protocluster environment using the galaxies in SPT2349−56. Figure 6 (left panel) shows the derived rest-frame ultravio- let galaxy sizes as a function of SFR, compared to a sample of z ≈ 2 ﬁeld SMGs from Swinbank et al. (2010) (with correspond- ing SFR estimates from Chapman et al. 2005), which were ob- served by the ACS instrument onboard HST in the F775W ﬁlter (observed wavelength 770 nm). These SMGs were selected from SCUBA surveys of various other cosmological ﬁelds, and the rest wavelengths probed by the observations range from 170 to 450 nm, comparable to our coverage of 290 nm. We next show a sample of three z ≈ 2.5 ﬁeld dusty star-forming galaxies from Barro et al. (2016) that were observed by the ACS F850LP ﬁlter (910 nm in the observed frame, 260 nm in the rest frame). These galaxies were selected from the CANDELS survey of the Great Observatories Origins Deep Survey-South (GOODS-S) ﬁeld (Grogin et al. 2011) for their compact nature and brightness at far-infrared wavelengths, thus follow-up observations found them to be reasonably bright at submm wavelengths (> 1 mJy at 870 µm) and have large SFRs (> 100 M(cid:12) yr−1), so we simply refer to them as SMGs. In these comparison samples, the Sérsic index was allowed to vary, and the half-light radii are the semi-major axes of an ellipse containing half the total ﬂux density, consistent with our deﬁnition of the half-light radius. We also include star-forming galaxies in the range z = 4–5 from the ALPINE survey with reliable ﬁts to imaging in both HST’s F160W ﬁlter and in ALMA moment-0 maps of [Cii] line emis- sion (Fujimoto et al. 2020); for these galaxies, the rest-wavelength observed in the ultraviolet ranges from 230 to 280 nm. For these measurements, the Sérsic index was ﬁxed to n = 1, but the authors note that ﬁxing n = 0.5 aﬀected their size measurements only at the ≈ 5 per cent level. For comparison, the mean Sérsic index from our ﬁts is 0.84. Fujimoto et al. (2020) only provides circularized sizes, deﬁned with respect to our size measurements as re = r1/2 q, where q is the semi minor-to-semi major axis ratio. In order to statistically convert this sample to semi-major axis sizes, we assume that the galaxies are circular discs with ﬁnite thickness parameterized by the ratio of the scale height to the disc radius, q0, thus the rela- tionship to the observed semi minor-to-semi major axis ratio q is sin2(i) = (1 − q2)/(1 −q2 0) (e.g. Förster Schreiber et al. 2018), where i is the inclination angle. We take q0 = 0.20 (e.g., Genzel et al. 2008; √ The stellar content of SPT2349−56 17 Law et al. 2012; van der Wel et al. 2014), and calculate the average expected √ q assuming an isotropic distribution of galaxy orienta- tions following the method outlined in Appendix A of Law et al. (2009). We ﬁnd (cid:10)√ q(cid:11) = 0.72, so we divide the ALPINE size mea- surements by this value. As a check, we calculated circularized sizes for our sample using the best-ﬁt axis ratios from our Sérsic modelling and compared these directly to the ALPINE circularized size measurements, but did not ﬁnd any systematic diﬀerences. For reference, we show the half-light radius of the HST F160W PSF (approximately half the FWHM of 0.151 arcsec) as a horizontal dashed line. Figure 7 (left panel) shows the rest-frame ultraviolet sizes as a function of stellar mass. Also shown are the same comparison sam- ples, with stellar masses taken from the same studies except for the z ≈ 2 ﬁeld SMGs from Swinbank et al. (2010), where we use stellar masses obtained by Michałowski et al. (2012), and we show the size of our stack (arbitrarily placed at 1011 M(cid:12)) and the HST F160W PSF. Lastly, in Fig. 8 (left panel) we show rest-frame ultraviolet sizes as a function of spectroscopic redshift. Although no trends in SFR or stellar mass are apparent, the Swinbank et al. (2010) SMGs are typically larger than the other comparison samples shown here, although this is to be expected owing to their larger stellar masses (Fig. 7) and the known size-mass relation (e.g. van der Wel et al. 2014). We next turn to our rest-frame far-infrared sizes of galaxies in SPT2349−56, taken from Hill et al. (2020), which were calcu- lated using the same procedure as the [Cii] sizes outlined in Section 3.6, only in this case after averaging the line-free ALMA channels. Fig. 6 (right panel) shows these size measurements as a function of SFR, and compare them to the same ﬁeld galaxies from Barro et al. (2016) (which measure the rest-frame far-infrared at 250 µm). In this comparison we have included the same galaxies from the ALPINE survey with [Cii] size measurements from Fujimoto et al. (2020) (corrected by the mean squareroot of the axis ratio expected from an isotropic distribution, (cid:10)√ q(cid:11) = 0.72), and we have converted these [Cii] size measurements to rest-frame far-infrared size measure- ments using the mean R1/2,C[II]/R1/2,FIR ratio from our sample (described below). Again, the ﬁts to the ALPINE galaxies were done with the Sérsic index ﬁxed to 1, while the mean Sérsic index from our ﬁts is 0.83, and we checked that a direct comparison between our circularized size measurements the the circularized ALPINE size measurements did not yield any systematic diﬀerences. This sam- ple measures the rest-frame far-infrared at 160 µm. We then include galaxies between z = 1.5 and 5.8 with submm size measurements from the UDS ﬁeld measured by ALMA (Gullberg et al. 2019), in this case probing rest wavelengths between 130 and 350 µm, converted to the size of the semi-major axis using the axis ratios provided. We also include the size of the stacked image, arbitrarily placed at 100 M(cid:12) yr−1. For reference, we show the half-light radius √ ab / 2 = 0.2 arcsec, where a and of the ALMA synthesized beam ( b are the major and minor FWHM, respectively) as a horizontal dashed line. In Fig. 7 (right panel) we show the same size measurements, this time as a function of stellar mass, including our stack ar- bitrarily placed at 1011 M(cid:12), and the ALMA PSF. Lastly, Fig. 8 shows the same quantities as a function of redshift. The galaxies in SPT2349−56 span a range of sizes comparable to the literature samples, with no discernible trends in SFR, stellar mass, or redshift. This result is in agreement with the idea that not all SMGs are com- pact starbursts, and instead there is some heterogeneity in the SMG population (e.g., Hayward et al. 2012, 2013). We now turn to the size ratios of our sample, and compare them l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT 18 Hill et al. to the literature. These comparisons should not depend on the choice of circularized size versus semi-major axis length, since we expect that the observed ellipticities of galaxies are equal at ultraviolet and far-infrared lengths. Fig. 9 (right panel) shows the ratio of the rest- frame far-infrared size to the ultraviolet size, R1/2,FIR/R1/2,UV, for the galaxies in SPT2349−56, as well as the size ratio for the stacks. We plot these ratios as a function of SFR as the SFRs of our sample are better-constrained than the stellar masses. All of the galaxies here are core galaxies of SPT2349−56, since our high-resolution submm imaging only covered this region. The ﬁeld SMGs from Barro et al. (2016) and Fujimoto et al. (2020) are shown alongside our sources on this plot, but the ﬁeld SMGs from Swinbank et al. (2010) and (Gullberg et al. 2019) have not been observed at high resolution at both wavelengths and so are omitted from this com- parison. Unfortunately, there are only three sources (C2, C6, and C8) for which we could measure both a far-infrared size and an ultraviolet size. To improve the sample size, we can look at the [Cii] sizes for some of our sources, where the emission is much brighter, and use this as a proxy for the dust sizes. Figure 9 (left panel) plots the ratio R1/2,C[II]/R1/2,FIR as a function of SFR for all sources where both measurements are avail- able. We ﬁnd a relatively consistent ratio across all SFR values, with a mean of 1.3 ± 0.2. Next, in Fig. 9 (right panel) we use the ratio R1/2,C[II]/R1/2,UV divided by the mean ratio above to provide an estimate of the underlying continuum size of sources where such a measurement is not available; this adds one more galaxy (C17) to the sample. The weighted mean of the size ratio for the galaxies in our sam- ple is 1.4, with a standard deviation of 0.6, consistent with the size ratio of 0.9 ± 0.2 from our stacked images, and in agreement with simulations of high-z star-forming galaxies (e.g. Cochrane et al. 2019). Compared to the literature, given the large uncertainties and small sample sizes, we can only conclude that our galaxies show ratios consistent with the ﬁeld. 4.6 Radial surface brightness distributions In Figs. 6–8 we see that C6, the central galaxy of the SPT2349−56 protocluster, has a much smaller rest-frame ultraviolet half-light radius compared to the rest of the protocluster galaxies with an available size measurement. However, most of the galaxies in our sample are not well-enough detected to measure ultraviolet sizes, so to perform a more statistical analysis we can turn to our stacked F160W image (Fig. 2). In Fig. 10 (left panel) we show the surface brightness of the stack as a function of radius, calculated in elliptical annuli with widths of 1 pixel, where the shape of the annuli was set to the best-ﬁt ellipticity and position angle from the Sérsic proﬁle ﬁt. In this plot the points and the error bars are the means and standard deviations of the pixel values within each annulus, respectively. Here we have converted our surface brightness measurements to units of µJy kpc−2 by dividing by the number of kpc2 in a pixel of our image. For the last point our measurement is consistent with zero, so we show the 1σ upper-limit. We also show the best-ﬁt Sérsic function as a shaded region, where the width is obtained by varying the best-ﬁt parameters within their 68 per cent conﬁdence intervals. In Fig. 10 (right panel) we show its surface brightness proﬁle of C6, calculated in the same way as with the stack. In order to assess whether or not we are seeing resolved emis- sion in these surface brightness proﬁles, we compare them to the HST F160W PSF model described in Section 3.5. Figure 10 shows the surface proﬁle of the model PSF for reference, normalized to the value of the central pixel of the stacked submm image (left panel) and galaxy C6 (right panel). Beyond about 2 pixels (or about 1 kpc) our stacked galaxy image shows signiﬁcantly more emission compared to the stacked star, so we are indeed measuring extended emission, yet galaxy C6 is eﬀectively indistinguishable from an un- resolved point source, conﬁrming that C6 is more compact than the average of the other galaxies in SPT2349−56. A similar stacking analysis was performed for the 25 spectroscopically-conﬁrmed z ≈ 2 ﬁeld SMGs of Swinbank et al. (2010) shown in Figs. 6–8, and any diﬀerences in the mean pro- ﬁle of this ﬁeld population compared to the mean proﬁle of our protocluster galaxies could indicate the presence of interesting en- vironmental eﬀects. Upon rescaling and stacking their detections, it was found that a Sérsic index of 2.6 ± 1.0 best described their data, in agreement with our value of 1.72 ± 0.30. In Fig. 10 (left panel) we show their ﬁt as a grey shaded region, where the width corresponds to the uncertainties in their best-ﬁt parameters. We set the scale radius to be 2.7 ± 0.4 kpc, corresponding to the median half-light radius of their sample, and scale the amplitude to have the same integrated ﬂux density as our stack, then convolve the 1-D proﬁle with the HST beam in the F160W ﬁlter. We see that the surface brightness proﬁle of our stack is consistent out to where our data are sensitive, thus we cannot conclude any diﬀerences are seen in the data. We next turn to comparing galaxy C6 with the literature. The three GOODS-S galaxies at z ≈ 2.5 from Barro et al. (2016) ob- served in the F850LP ﬁlter (rest-wavelength 260 nm) were selected for their small rest-frame optical sizes and high stellar-mass den- sities, similar to what we are seeing with galaxy C6. In Fig. 10 (right panel) we show their resulting best-ﬁt Sérsic proﬁles, with each normalization calculated by setting the ratio of the integrated ﬂux density of a given galaxy to the integrated ﬂux density of C6 equal to the ratio of the stellar mass of the same galaxy to the stellar mass of C6. We then convolve each proﬁle with the HST beam in the F160W ﬁlter, converting the units of the beamsize from arcsec to kpc using the redshift of SPT2349−56 in order to provide a direct comparison with our observations. We have highlighted the range in surface brightness they span for clarity. We see that the compact SMGs in this sample would all eﬀectively appear unresolved in our HST imaging, similar to C6. We can also compare the current proﬁle of galaxy C6 to a prediction of its proﬁle from the hydrodynamical simulation pre- sented by Rennehan et al. (2020). Brieﬂy, the simulation evolved the 14 core galaxies initially discovered by Miller et al. (2018) for 1 Gyr in several separate and independent realizations. Each galaxy was initialized with a dark matter halo and a stable gas and stellar disc, with component masses scaled from the available gas mass estimates in Miller et al. (2018) and discs modeled following an exponentially decreasing surface density with a scale size related to the angular momentum (Robertson et al. 2006). In particular, we select four realizations where the stellar masses were computed assuming a molecular gas-to-stellar mass fraction of 2.3 (for ref- erence, the range measured for these galaxies is 0.1 to 4.8), and the halo masses were scaled from the stellar masses by a factor of 100. In each realization the positions were randomly selected to lie within a 65 kpc-radius sphere, and the velocities were drawn from a Gaussian distribution matching the measured line-of-sight velocity dispersion. For each realization, we take the median mass proﬁle between 700 and 800 Myr, then take the mean proﬁle across these realizations. The overall normalization of the proﬁle is highly un- certain in the simulation as it requires radiative transfer models to convert stellar mass into rest-frame ultraviolet ﬂux density, so we simply normalize the proﬁle by the peak pixel in our F160W imag- l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 19 l D o w n o a d e d f r o m h t t p s : / / Figure 6. Left: Ultraviolet half-light radius as a function of SFR for all galaxies in the SPT2349−56 protocluster detected in the F160W band with pixels above 5 times the local rms. Also shown are results for z = 1–3 ﬁeld SMGs from Swinbank et al. (2010) and Barro et al. (2016), and z = 4–5 star-forming galaxies from the ALPINE survey (Fujimoto et al. 2020). For reference, the size measured in our HST F160W stack is shown as the pink square, arbitrarily placed at 100 M(cid:12) yr−1, and the half-light radius of the HST F160W beam is shown as the horizontal dashed line. Right: Far-infrared half-light radius as a function of SFR from Hill et al. (2020), for all galaxies detected by ALMA at 850 µm with pixels above 5 times the local rms, along with our stack shown as the pink square. Also shown are results for the ﬁeld SMGs from Barro et al. (2016), ﬁeld SMGs in the UDS ﬁeld from Gullberg et al. (2019), and [Cii] size measurements from the ALPINE survey (Fujimoto et al. 2020), converted to far-infrared sizes using the mean R1/2,C[II]/R1/2,FIR ratio from our sample (see Fig. 9). The half-light radius of the ALMA synthesized beam is shown as the horizontal dashed line. Figure 7. Left: Ultraviolet half-light radius as a function of stellar mass for the same samples shown in Fig. 6. The stacked size measurement (arbitrarily placed at 1011 M(cid:12)) and HST beam size are also shown for comparison. Right: Far-infrared half-light radius as a function of redshift for the same galaxies shown in Fig. 6, along with the stacked size measurement and the ALMA beam size. ing of C6. The resulting proﬁle is shown in Fig. 10, and we see that after the merger, the BCG will remain quite compact. Nonetheless, we expect the dark matter halo to grow in size after the merger, so in Fig. 10 we also show the resulting shape of the dark matter halo, normalized in the same way for easy comparison with the stellar proﬁle. We see that the shape of the dark matter halo after the merger is much more extended than the stellar mass proﬁle. 5 DISCUSSION 5.1 The stellar properties of SPT2349−56 The protocluster SPT2349−56 is a unique object when ob- served in the submm. We now know that the total SFR exceeds 10,000 M(cid:12) yr−1, and its extreme nature in this regard is due to the fact that it was speciﬁcally selected as one of the brightest unlensed point sources in the SPT mm-wavelength survey. Now that we have studied this protocluster at optical and infrared wavelengths, we have begun to probe the properties of the constituent stars themselves. We have found that the stellar masses derived through SED i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT101001000SFR[Mﬂyr−1]0.1110R1/2,UV[kpc]FieldSMGs,GOODS-SFieldSMGs,variousFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,HSTPSF101001000SFR[Mﬂyr−1]0.1110R1/2,FIR[kpc]FieldSMGs,GOODS-SFieldSMGs,UDSFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,ALMAPSF101010111012M∗[Mﬂ]0.1110R1/2,UV[kpc]FieldSMGs,GOODS-SFieldSMGs,variousFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,HSTPSF101010111012M∗[Mﬂ]0.1110R1/2,FIR[kpc]FieldSMGs,GOODS-SFieldSMGs,UDSFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,ALMAPSF 20 Hill et al. l D o w n o a d e d f r o m h t t p s : / / Figure 8. Left: Ultraviolet half-light radius as a function of redshift for the same samples shown in Fig. 6. The stacked size measurement and HST beam size are also shown for comparison. Right: Far-infrared half-light radius as a function of redshift for the same galaxies shown in Fig. 6, along with the stacked size measurement and the ALMA beam size. Figure 9. Left: Ratio of [Cii] half-light radius to far-infrared half-light radius as a function of SFR for galaxies in SPT2349−56. The mean value is 1.3 (solid line) with a standard deviation of 0.2 (dotted line). Right: Ratio of far-infrared half-light radius to ultraviolet half-light radius; this is shown as the open circles for the four galaxies in SPT2349−56 for which both measurements are available. Since there are only four galaxies with both size measurements available, we show the ratio R1/2,C[II]/R1/2,UV as solid circles and correct it using the mean ratio of [Cii] to far-infrared size from the left panel in order to estimate R1/2,FIR/R1/2,UV. The pink square shows the same ratio for the size measurements of our stacked images. We compare our size ratio measurements to the sample of ﬁeld SMGs from the GOODS-S ﬁeld (Barro et al. 2016) and the star-forming galaxies from the ALPINE survey (Fujimoto et al. 2020). A dashed horizontal line is drawn where R1/2,F = R1/2,UV for clarity. ﬁtting place these galaxies inconspicuously on the MS, as opposed to appearing as outliers above it; given their incredibly high SFRs, this means that they have correspondingly large stellar masses, which are simply hidden by dust. For reference, the total stellar mass of all the known galaxies in SPT2349−56 is (1.5±0.3) × 1012M(cid:12), which is comparable to a large BCG at z < 0.1. Based on our unequal-variance t-test between ﬁeld SMGs and star-forming protocluster galaxies, assuming that the molecular gas- to-stellar mass fractions and depletion timescales are drawn from Gaussian distributions with arbitrary variances, we are able to reject the null hypothesis that the means of the two distributions are equal, although the scatter between the two populations can overlap. Inter- estingly, similar studies of CO lines at moderate redshift (z = 1– 3) have found that star-forming galaxies in cluster/protocluster environments at this epoch have systematically higher molecular gas-to-stellar mass fractions and depletion timescales compared to scaling relations derived from the ﬁeld (e.g., Noble et al. 2017; Hayashi et al. 2018; Tadaki et al. 2019), while at z < 1 galaxy clus- ters have nearly depleted all of their gas and ceased forming stars (e.g., Young et al. 2011; Jablonka et al. 2013; Scott et al. 2013; Boselli et al. 2014; Zabel et al. 2019). To quantify this, in Fig. 11 we show the molecular gas-to-stellar mass fraction (top panel) and depletion timescale (bottom panel) as a function of redshift for the galaxies in SPT2349−56, compared to i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT0123456z0.1110R1/2,UV[kpc]FieldSMGs,GOODS-SFieldSMGs,variousFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,HSTPSF0123456z0.1110R1/2,FIR[kpc]FieldSMGs,GOODS-SFieldSMGs,UDSFieldgalaxies,ALPINESPT2349−56,stackSPT2349−56R1/2,ALMAPSF101001000SFR[Mﬂyr−1]0.51.01.52.0R1/2,CII/R1/2,FIRSPT2349−56MeanStandarddeviation101001000SFR[Mﬂyr−1]01234R1/2,FIR/R1/2,UVSPT2349−56,continuumSPT2349−56,CIIcorrectedSPT2349−56,stackFieldSMGs,GOODS-SFieldgalaxies,ALPINE The stellar content of SPT2349−56 21 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / / . 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 10. Surface brightness as a function of radius for the protocluster galaxies in SPT2349−56, determined within elliptical annuli of 1-pixel width. The points and the error bars are the means and standard deviations of the pixel values within each annulus. Where the standard deviation overlaps with zero, we provide 1σ upper-limit. Left: In magenta we show the surface-brightness proﬁle for our stack of HST-detected galaxies (after removing C6, a bright outlier), with the best-ﬁt Sérsic proﬁle shown as the magenta shaded region, and in cyan we show the HST-F160W PSF. For comparison, we show the best-ﬁt surface-brightness proﬁle from a similar stacking analysis of 25 ﬁeld SMGs (grey shaded region) observed by HST in the F775W ﬁlter by Swinbank et al. (2010), normalized to have the same integrated ﬂux density as our stack. Right: In red we show the surface brightness proﬁle for galaxy C6, and in cyan we show the same HST-F160W PSF. For comparison, in brown we show a set of three best-ﬁt surface-brightness proﬁles of SMGs around z = 2.5 from Barro et al. (2016), normalized by their stellar masses relative to C6. We have then convolved each proﬁle with the HST-F160W PSF, assuming each source is at the same redshift as SPT2349−56, in order to provide a direct comparison with our observations. Here we have highlighted their total range for clarity. We show the expected stellar proﬁle (as a dashed red curve) and dark matter proﬁle (as a dashed black curve) of the BCG complex after 800 Myr, once the central galaxies have merged (Rennehan et al. 2020), also normalized by the peak pixel value of C6. (1−0) /L (cid:48) other clusters and protoclusters with molecular gas mass estimations made through observations of CO, and with suﬃcient multiwave- length coverage to have stellar mass and SFR estimates. At z = 4.0 we show the DRC (Oteo et al. 2018; Long et al. 2020), and we have converted CO(6–5) line intensities to molecular gas masses using an L (cid:48) factor of 0.46 (the mean ratio found for the SPT-SMG (6−5) sample, see Spilker et al. 2014) and an αCO of 1 M(cid:12)/(K km s−1 pc2) (the same scale factor used for our sample). At intermediate redshift (z =1–3) we show results from observations of CO(3–2), CO(2–1), and CO(1–0) in members of three Spitzer Adaptation of the Red- sequence Cluster Survey (SpARCS) clusters (Noble et al. 2017), XMMXCS J2215.9−1738 (Hayashi et al. 2018), three Lyα-selected protoclusters (Tadaki et al. 2019), and two potentially associated overdensities identiﬁed in the COSMOS ﬁeld, CLJ1001 at z = 2.50 (Wang et al. 2016, 2018) and PCL1002 at z = 2.47 (Casey et al. 2015; Champagne et al. 2021). For the COSMOS structure, we show the measurements of the individual galaxies in CLJ1001 from Wang et al. (2018), and the properties of the single galaxy in PCL1002 with a CO(1–0) detection from Champagne et al. (2021); however, Champagne et al. (2021) independently derived the unre- solved properties of CLJ1001, ﬁnding a 25 per cent shorter depletion timescale compared to the average reported by Wang et al. (2018). At low redshift (z <1) we take observations of CO(3–2), CO(2–1), and CO(1–0) in members of the Fornax cluster (Zabel et al. 2019), Abell 2192 and Abell 963 (Cybulski et al. 2016), CL1411.1−1148 (Spérone-Longin et al. 2021), CL0024+16 (Geach et al. 2009, 2011), and MACS J0717.5+3745, Abell 697, 963, 1763, and 2219 (Castignani et al. 2020). These comparison samples con- (1−0) (1−0) (2−1) /L (cid:48) /L (cid:48) sistently used CO conversion factors of L (cid:48) = 0.8 and = 0.5, and αCO ≈ 4 M(cid:12)/(K km s−1 pc2), with small L (cid:48) (3−2) corrections for metallicity made in some cases. This conversion factor is appropriate for normal star-forming galaxies, which were the targets of these literature studies, and a lower conversion fac- tor around 1 M(cid:12)/(K km s−1 pc2) is typically used for SMGs, as with SPT2349−56 and the DRC. Some galaxies in the sample of Castignani et al. (2020) approach the high SFRs in the SMG regime, and they have adopted a smaller conversion factor for those sources. For each cluster in this ﬁgure, we show the mean molecular gas-to- stellar mass fraction and depletion timescale as a square symbol, with error bars representing the standard deviation for clusters with more than two galaxies with suﬃcient data. To represent ﬁeld galaxies in Fig. 11, we use the scaling re- lations for (cid:10)µgas(cid:11) and (cid:10)τdep(cid:11) (where (cid:104)(cid:105) denotes the average of the ﬁeld population) estimated by Tacconi et al. (2018), derived from observations of CO lines and continuum ﬂux densities up to millimetre wavelengths in over 1300 ﬁeld galaxies between z = 0 and 4 (including the sample from Scoville et al. 2016). They pro- vide equations for calculating (cid:10)µgas(cid:11) and (cid:10)τdep(cid:11) as a function of redshift, M∗, SFR, and the eﬀective radius at rest-frame 500 nm; however, since we do not have access to size measurements for most of the galaxies in our comparison sample, we use their ﬁts that do not include this parameter. Figure 11 shows several rep- resentative curves for (cid:10)µgas(cid:11) and (cid:10)τdep(cid:11) given diﬀerent values of M∗ and SFR. In the bottom panels of Fig. 11 we show the ratios µgas / (cid:10)µgas(cid:11) and τdep / (cid:10)τdep(cid:11) (including the expected intrinsic scatter of ± 0.3 dex, see Schreiber et al. 2015), obtained by dividing ORIGINAL UNEDITED MANUSCRIPT110100R[kpc]10−510−410−310−210−1100Iν[µJykpc−2]SPT2349−56stackSPT2349−56fitHST-F160WPSFz=1−3fieldSMGfit110100R[kpc]10−510−410−310−210−1100Iν[µJykpc−2]SPT2349−56C6HST-F160WPSFSPT2349−56BCG,800MyrSPT2349−56DMhalo,800Myrz=2.5fieldSMGs 22 Hill et al. each galaxy’s molecular gas-to-stellar mass fraction and depletion timescale by the prediction from Tacconi et al. (2018), taking into account each galaxy’s redshift, M∗, and SFR. We conﬁrm that at z = 1–3 most cluster environments are gas-rich, as discussed in pre- vious studies, and we see that the galaxies in SPT2349−56 and the DRC continue to fall below the expected molecular gas-to-stellar mass fractions and depletion timescales of ﬁeld galaxies, although the intrinsic scatters of these populations are still large and often overlap. In our comparison of the stellar mass function of SPT2349−56 with the stellar mass function of z = 1 galaxy clusters, we found sim- ilar shapes well-described by Schechter functions. Since we know that the core galaxies will merge over a timescale of a few hundred Myr, we also computed the stellar mass function of SPT2349−56 after summing up the stellar masses of these galaxies and treating them as a single source. In this case we found that the number counts are better-ﬁt by a single power law, indicating that if this structure is to continue along an evolutionary path to become a z = 1 galaxy cluster, the remaining cluster stellar mass will come from lower-mass galaxies (e.g. Naab et al. 2009). However, we have not taken into account the observational biases and incompleteness inherent in our sample, and so it is not clear from the current data whether these galaxies are already within the 1 Mpc environment of SPT2349−56 and have not been detected, or have yet to fall into the galaxy protocluster. It is worth noting that this behaviour is consis- tent with the notion of ‘downsizing’, where the most massive galax- ies formed the earliest times, which has been observed in numerous samples of galaxies (e.g., Cowie et al. 1996; Magliocchetti et al. 2013; Miller et al. 2015; Wilkinson et al. 2017). Lastly, we have noted that the ratio of far-infrared size to ultra- violet size is comparable to star-forming galaxies found around the same redshift, and ﬁeld SMGs at lower redshift (z ≈ 2.5). A similar study at z ≈ 1 also found that the stellar emission in cluster galaxies is more compact than in ﬁeld galaxies (Matharu et al. 2019), which is not what we are seeing here, although the sample sizes we are investigating are small, and there could be systematic diﬀerences in the size measurements. Nonetheless, we might not expect to see many diﬀerences between ﬁeld galaxies and protocluster galaxies at high redshift as there has not been enough time for the clustering environment to shape the residing galaxies. 5.2 The properties of a BCG in formation In our analysis above, we have paid special attention to separating the central galaxies of SPT2349−56 from the wider protocluster. In Rennehan et al. (2020), hydrodynamical simulations using the positions of these central galaxies as the initial conditions pre- dicted a complete merger on timescales of a few hundred Myr, while in Hill et al. (2020) it was found that the velocity distri- bution within this region was consistent with a Gaussian distri- bution, and that the velocity dispersion predicted a central mass of (9 ± 5) × 1012 M(cid:12). Summing up the stellar masses of each of these central galaxies yields a value of (9 ± 2) × 1011 M(cid:12), consis- tent (within the uncertainties) with the mass of (12 ± 3) × 1011 M(cid:12) estimated by Rotermund et al. (2021). Looking at Fig. 3, we see that these central galaxies lie close to the MS found by Khusanova et al. (2021), with no statistically signiﬁcant oﬀsets given the large uncertainties. Similarly, the two galaxies from the northern component of SPT2349−56 are signif- icantly above the MS, but small number statistics again mean that we cannot draw any statistically signiﬁcant conclusions from this observation. The MS distribution of SPT2349−56 overall matches well with what is seen with the DRC, a similar star-forming proto- cluster from the literature, along with other samples of ﬁeld SMGs at high redshift. In Fig. 12 we explore the concept of environmental dependence further by plotting the cumulative mass enclosed within a circular aperture as a function of the area of the aperture, separating out the stellar mass and the molecular gas mass. In this plot the centre of SPT2349−56 is the luminosity-weighted centre, as in Hill et al. (2020). We see that the molecular gas mass and stellar mass track one another across all scales probed by our data, from the region of the forming BCG (90 kpc, or about 0.03 Mpc2) out to the northern component (about 0.5 Mpc away, or at 1 Mpc2). To quantify this, in the bottom panel of Fig. 12 we show the total enclosed molecular gas-to-stellar mass fraction, and we can see that it remains roughly constant at a level of about 0.8 (except for near the centre, but these ﬂuctuations suﬀer from small-number statistics). For comparison, we show the same curve-of-growth for the DRC (Oteo et al. 2018; Long et al. 2020); the behaviour of this protocluster is similar. We then compare the stellar mass proﬁles of these high-z pro- toclusters to the stellar proﬁle of a typical z (cid:39) 1 galaxy cluster from van der Burg et al. (2014), obtained from the stack of the same sam- ple of 10 clusters discussed in Section 5. The authors found a best- ﬁt Navarro-Frenk-White (NFW) proﬁle concentration parameter of 7+1.53 , which we use to plot the mass projected within a cylinder of −0.99 a given area (see e.g. Łokas & Mamon 2001), taking the scale stellar mass to be M200,∗ = 2 × 1012 M(cid:12), the median of their sample. We see that the shapes of the stellar mass curves between the protoclus- ters and the z (cid:39) 1 clusters are similar, although the slope of the pro- toclusters becomes shallower than the slope of the z = 1 clusters at large radii. It is interesting to note that a similar study (Alberts et al. 2021) investigating stacks of galaxy clusters between z = 0.5 and 1.6 in the near-infrared (3–8 µm in the rest-frame) found similarly- concentrated light proﬁles, with NFW concentration parameters around 7; this near-infrared light is expected to trace stellar mass. They also investigated stacks in the far-infrared (250–500 µm in the rest-frame), tracing dust emission and SFR, and found concentra- tion parameters comparable to the near-infrared light. It was found that 20–30 per cent of the integrated cluster far-infrared emission comes from high-mass galaxies, while in Hill et al. (2020) about 50 per cent of the low-resolution single-dish far-infrared emission resolved into massive galaxies. Despite the fact that the galaxies on-track to merge into a BCG are not altogether distinguishable from the rest of the protocluster (or indeed, from most SMGs at these redshifts), galaxy C6 does stand out from our core sample as having the brightest ﬂux density at all optical-through-infrared wavelengths. Our SED modelling found that this galaxy’s stellar mass is (4 ± 1) × 1011 M(cid:12), or roughly half of the total stellar mass expected to make up the BCG after the mergers are complete, and our brightness proﬁle analysis of this galaxy suggests that it is incredibly compact. The region around galaxy C6 is clearly at the centre of this BCG-in-formation. Rotermund et al. (2021) already discussed a plausible evolutionary track for the growth of the stellar mass of C6. Around z = 0.5, BCG stellar masses range from 5–10 × 1011 M(cid:12) (e.g., Hilton et al. 2013), so C6 is already nearly there, and once the merging is complete, it will be at the upper end of BCG masses known. In Fig. 10 we see that C6 is expected to remain compact after the merger, and so if C6 is to continue to grow in stellar mass, we might also expect it to grow by a considerable amount in size. This could happen through dry mergers (mergers between galaxies with little gas, and thus little star formation), which has been found to play an important role in the growth of BCGs (e.g., Liu et al. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 23 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i Figure 11. Top: Molecular gas-to-stellar mass fraction (µgas = Mgas / M∗) as a function of redshift for SPT2349−56 (pink), the DRC (blue), and various clusters and protoclusters with similar CO-derived molecular gas mass data available (grey; references are provided in the text). Points indicate values for individual galaxies, while squares show the mean values for each (proto)cluster, and error bars show the standard deviations for (proto)clusters with more than two sources available. Representative models for the mean molecular gas-to-stellar mass fraction of ﬁeld galaxies ((cid:10)µgas(cid:11)) from Tacconi et al. (2018) are shown as the solid curves. The bottom panel shows the ratio µgas / (cid:10)µgas(cid:11), calculated for each galaxy given its redshift, M∗, and SFR, along with the means and standard deviations for each (proto)cluster. The shaded region indicates the expected intrinsic scatter of ± 0.3 dex (Schreiber et al. 2015). Bottom: Same at the top panel, only showing the depletion timescale (τdep = Mgas / SFR). 2009; Lin et al. 2010; Liu et al. 2015), and simulations predict that this process will increase a BCGs size and make it less compact (e.g., Khochfar & Silk 2006; Oogi & Habe 2012). This would need to occur after C6 merges with the core galaxies in its current vicin- ity, as these mergers will be gas-rich. From our simulation we see that the post-merger dark matter halo will have a large and extended proﬁle, and this could become populated by further accretions and dry mergers if C6 grows following an inside-out scenario, where the slope of the outer proﬁle becomes shallower with increasing mass (e.g., van Dokkum et al. 2010; Bai et al. 2014; Whitney et al. 2019). While we stress that the precise ﬁnal mass and size of a BCG depends strongly on its detailed merger history, something we can- not know from system to system, we see that C6 is at least consistent with the picture that it is the progenitor of one of the most massive ORIGINAL UNEDITED MANUSCRIPT10−210−1100101102µgas=Mgas/M∗M∗=1×1010MﬂSFR=5Mﬂyr−1M∗=5×1010MﬂSFR=100Mﬂyr−1M∗=10×1010MﬂSFR=500Mﬂyr−1<µgas>,Tacconietal.2018SPT2349−56DRCLiterature(proto)clusters012345z10−1100101µgas/<µgas>10−210−1100101τdep=Mgas/SFR[Gyr]M∗=1×1010MﬂSFR=5Mﬂyr−1M∗=5×1010MﬂSFR=100Mﬂyr−1M∗=10×1010MﬂSFR=500Mﬂyr−1<τdep>,Tacconietal.2018SPT2349−56DRCLiterature(proto)clusters012345z10−1100101τdep/<τdep> 24 Hill et al. BCGs seen today, and that it has nearly formed all of its stars at this early epoch of formation. These observations of a BCG in formation provide a direct measure of the ingredients that built up these massive galaxies in the early Universe. There are many studies of the stellar popu- lations of BCGs that try to piece together their formation histo- ries, which broadly point to a fast and early core formation phase (> 10 Gyr ago) followed by a slow and continuous accretion phase (< 10 Gyr ago) fuelled by minor mergers respondible for assem- bling the outer regions (e.g., De Lucia & Blaizot 2007; Collins et al. 2009; Barbosa et al. 2016; Cooke et al. 2019; Edwards et al. 2020). Here we provide a direct observation of this formation in progress, where we have access to information that will likely be lost after the merger of the core galaxies in SPT2349−56 into a BCG. In particular, we see that the ingredients of this BCG are numerous galaxies with very high star-formation rates that, once merged, will already have formed most of the stars that make up a typical BCG. Furthermore, stellar population studies are only able to trace BCG histories back to their early star-forming phase, which in the case of SPT2349−56, will begin after the merger. But by observing these galaxies before they merge into a BCG, we now have access to in- formation about the stellar history of a BCG back to a much earlier time, as traced by the stellar populations of the pre-merger galaxies. 6 CONCLUSION SPT2349−56 was selected as the brightest protocluster candidate from the SPT-SZ 2500 deg2 mm-wavelength survey. This object has since been spectroscopically conﬁrmed to be a true protoclus- ter, containing over 30 submm-bright galaxies and dozens more LBGs and LAEs. In this paper we have described our results from an extensive optical-through-infrared follow-up campaign using ob- servations from Gemini-GMOS and FLAMINGOS-2, HST-F110W and F160W, and Spitzer-IRAC. Owing to the ≈ 2 arcsec spatial resolution of the IRAC images, source blending is an issue that needs to be dealt with. We deblended our IRAC data using t-phot, which used galaxy positions from our HST imaging as a prior and subsequently convolved their proﬁles with the IRAC PRF to develop a catalogue of the underlying source distribution as seen by IRAC. Source catalogues for our Gemini and HST imaging were extracted using standard source-extractor routines. We matched the known protocluster galaxies in SPT2349−56 discovered by ALMA to our optical and infrared catalogues using a simple radial cut of 1 arcsec. We found a match in at least one of the eight optical/infrared ﬁlters for all but six galaxies. In addition, we searched for ALMA counterparts to a small sample of LBGs and LAEs, and found matches for all but one galaxy. Taking the photometry measured in these data and combining it with existing mm-wavelength photometry from Herschel-SPIRE and ALMA, we used CIGALE to ﬁt SEDs to each galaxy, allowing stellar masses, dust extinctions, ages, star-formation timescales, and star-formation rates to vary. We found that the galaxies in SPT2349−56 follow the galaxy main sequence, consistent with other samples of z (cid:39) 4 protocluster galaxies and ﬁeld SMGs. However, we ﬁnd small molecular gas- to-stellar mass fractions and short depletion timescales compared to ﬁeld SMGs at similar redshifts. We perform unequal-variance t- tests, rejecting the null hypothesis that the molecular gas-to-stellar mass fractions and depletion timescales of protocluster galaxies and ﬁeld SMGs have molecular gas-to-stellar mass fractions and depletion timescales drawn from Gaussian distributions with equal means, although the scatter in both populations is large. We ﬁnd the same result using known scaling relations calibrated from large samples of ﬁeld galaxies up to redshift 4. This could mean that protocluster galaxies in SPT2349−56 are at a late stage in their star-formation phase and have already nearly depleted their gas reservoirs as they build up their stars. We computed the stellar-mass function and gas-mass function of SPT2349−56 in two ways: ﬁrst, for the entire sample of galaxies; and second, by collapsing the core galaxies into a single source, reﬂecting the fact that they are expected to merge within a timescale of a few hundred Myr. The stellar- and gas-mass functions track each other well. Comparing the total protocluster stellar mass function to the stellar-mass function of typical z = 1 galaxy clusters, we ﬁnd that the samples are consistent with one another and are well-ﬁt by Schechter functions. The best-ﬁt characteristic mass for the z = 1 −0.2) × 1010 M(cid:12), and the slope is −0.46+0.08 galaxy clusters is (5.2+1.1 , −0.26 compared to SPT2349−56, where we ﬁnd a best-ﬁt characteristic mass of (6.2+0.8 −3.9) × 1010 M(cid:12) and a slope of −0.3+0.3 . On the other −0.3 hand, the protocluster with a merged BCG is better-ﬁt by a single power-law. Thus if SPT2349−56 is to follow a similar trajectory as the z = 1 galaxy clusters, then it must accrete numerous less-massive galaxies, or these less-massive galaxies must already be present but remain undetected in our observations. Due to incompleteness in our sample (about 80 per cent for M∗ > 1010 M(cid:12)), we are unable to distinguish between these two scenarios. We measured the physical sizes of the galaxies in SPT2349−56 in our deep HST-F160W data, which probe rest-frame ultravio- let wavelengths. Upon comparing these measurements with typical star-forming galaxies between redshift 4 and 5, we found that our sample has comparable ultraviolet sizes. We stacked our HST data at the positions of the detected galaxies and compared this with a stack of the same galaxies at submm wavelengths imaged by ALMA, ﬁnd- ing a consistent result. Galaxy C6, the brightest protocluster galaxy in our HST data, is the most compact rest-frame ultraviolet source and remains unresolved by our HST imaging. Hydrodynamical sim- ulations predict that this galaxy is at the centre of a major merger, yet after the merger the emission will still remain compact. Lastly, we investigated the total projected stellar and molecular gas mass of SPT2349−56 as a function of projected area. We found that the molecular gas mass and stellar mass track each other well, with no clear trend in the molecular gas-to-stellar mass fraction as a function of radius (the mean value being about 0.3). The stellar mass distribution of SPT2349−56 also does not appear to be markedly diﬀerent from z = 1 galaxy clusters, although we note that so far we have only probed the central ≈ 1 Mpc of the structure, which is much smaller than the extent of z = 1 clusters. SPT2349−56 is a galaxy protocluster in a remarkable phase of its evolution, reaching a total SFR of over 10,000 M(cid:12) yr−1 within a volume of approximately 0.1 Mpc3. Having been selected specif- ically for its star-forming properties, it is interesting that at optical and infrared wavelengths, the galaxies making up SPT2349−56 are not very luminous and have fairly typical stellar masses. Galaxy C6 is however an exception; this source is at the centre of a massive merger of over 20 galaxies, and is a likely proto-BCG. The transi- tion of galaxy C6 into a BCG is consistent with the ‘downsizing’ scenario of galaxy formation, and provides a direct observation of the constituents and formation mechanism of a BCG in the early Universe. In this case the high-redshift proto-BCG is undergoing a major merger with dozens of galaxies whose properties are similar to the ﬁeld, explaining why present-day BCGs contain old cores that seemingly formed very quickly. l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 25 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l Figure 12. Top: Cumulative mass enclosed within a circular aperture as a function of the area of the aperture, centred on the infrared luminosity-weighted centre of SPT2349−56 (see Hill et al. 2020 for details). In blue we show the molecular gas mass, and in magenta we show the stellar mass. The two masses track each other well, implying that the inner region of the protocluster has not had a chance to diﬀerentiate from the outer regions. Also shown are the stellar and molecular gas masses of the DRC (Oteo et al. 2018; Long et al. 2020), and the best-ﬁt NFW proﬁle to the stellar mass proﬁle of a stack of z = 1 clusters from van der Burg et al. (2014). Bottom: Total molecular gas-to-stellar mass fraction µgas (Eq. 3) enclosed within the same apertures as a function of area. ACKNOWLEDGEMENTS (2020). This paper makes use of The authors wish to thank Dr. Adam Muzzin for a use- ful discussion about the stellar mass function of galaxy clus- ters around redshift 1, and Dr. Douglas Rennehan for pro- viding access to and support with the simulations detailed in Rennehan et al. the following ALMA data: ADS/JAO.ALMA#2017.1.00273.S; and ADS/JAO.ALMA#2018.1.00058.S. ALMA is a partnership of ESO (representing its member states), NSF (USA) and NINS (Japan), to- gether with NRC (Canada), MOST and ASIAA (Taiwan), and KASI (Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, AUI/NRAO, and NAOJ. Herschel is an ESA space observatory with science instruments provided by European-led Principal Investigator con- sortia and with important participation from NASA. This paper is based on observations made with ESO Telescopes at the La Silla Paranal Observatory under programme ID 299.A-5045(A). The Na- tional Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by As- sociated Universities, Inc. The SPT is supported by the National Science Foundation through grant PLR-1248097, with partial sup- port through PHY-1125897, the Kavli Foundation, and the Gordon and Betty Moore Foundation grant GBMF 947. This work is based in part on observations made with the Spitzer Space Telescope, which was operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with NASA. This work was supported by the Natural Sciences and Engineering Research Council of Canada. The Flatiron Institute is supported by the Si- mons Foundation. M.A. has been supported by the grant “CONI- CYT+PCI+REDES 190194”. D.P.M., J.D.V., K.C.L., and K.P. ac- knowledge support from the US NSF under grants AST-1715213 and AST-1716127. K.C.L acknowledges support from the US NSF NRAO under grants SOSPA5-001 and SOSPA4-007, respectively. J.D.V. acknowledges support from an A. P. Sloan Foundation Fel- lowship. M.A. and J.D.V. acknowledge support from the Center for AstroPhysical Surveys at the National Center for Supercomputing Applications in Urbana, IL. S.J. acknowledges support from the US NSF NRAO under grants SOSPA5-001 and SOSPA7-006. i / . / / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT1091010101110121013Totalmass[Mﬂ]M∗,SPT2349−56Mgas,SPT2349−56M∗,DRCMgas,DRCM∗,z=1cluster10−410−310−210−1100101Area[Mpc2]0.010.11.0Totalgasfraction, µgas=Mgas/M∗µgas,SPT2349−56µgas,DRC 26 Hill et al. DATA AVAILABILITY All of the data presented in this paper are publicly available. The HST data can be found under the programme 15701 (PI S. Chap- man), the Spitzer-IRAC data can be found under the four pro- grammes 60194 (PI J. Vieira), 80032 (PI S. Stanford), 13224 (PI S. Chapman), and 14216 (PI S. Chapman), the Gemini-South data can be found under the programme GS-2017B-Q-7 (PI A. Chap- man), and the ALMA data can be found under the two programmes 2017.1.00273.S (PI S. Chapman) and 2018.1.00058.S (PI S. Chap- man). REFERENCES Alberts S., et al., 2021, MNRAS, 501, 1970 Andreon S., Newman A. B., Trinchieri G., Raichoor A., Ellis R. S., Treu T., 2014, A&A, 565, A120 Aravena M., et al., 2016, MNRAS, 457, 4406 Bacon R., et al., 2010, in McLean I. S., Ramsay S. K., Takami H., eds, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series Vol. 7735, Ground-based and Airborne Instrumentation for Astronomy III. p. 773508, doi:10.1117/12.856027 Bai L., et al., 2014, ApJ, 789, 134 Barbosa C. E., Arnaboldi M., Coccato L., Hilker M., Mendes de Oliveira C., Richtler T., 2016, A&A, 589, A139 Barro G., et al., 2016, ApJ, 827, L32 Bennett C. L., et al., 2003, ApJS, 148, 1 Beroiz M., Cabral J. B., Sanchez B., 2020, Astronomy and Computing, 32, 100384 Bertin E., Arnouts S., 1996, A&AS, 117, 393 Béthermin M., et al., 2020, A&A, 643, A2 Birkin J. E., et al., 2021, MNRAS, 501, 3926 Biviano A., 1998, in Mazure A., Casoli F., Durret F., Gerbal D., eds, Untangling Coma Berenices: A New Vision of an Old Cluster. p. 1 (arXiv:astro-ph/9711251) Bleem L. E., et al., 2015, ApJS, 216, 27 Boquien M., Burgarella D., Roehlly Y., Buat V., Ciesla L., Corre D., Inoue Draine B. T., et al., 2014, ApJ, 780, 172 Dudzeviči¯ut˙e U., et al., 2020, MNRAS, 494, 3828 Dunlop J. S., et al., 2017, MNRAS, 466, 861 Edwards L. O. V., et al., 2020, MNRAS, 491, 2617 Eikenberry S. S., et al., 2004, in Moorwood A. F. M., Iye M., eds, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series Vol. 5492, Ground-based Instrumentation for Astronomy. pp 1196– 1207, doi:10.1117/12.549796 Eikenberry S., et al., 2006, in McLean I. S., Iye M., eds, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Se- ries Vol. 6269, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series. p. 626917 (arXiv:astro-ph/0604577), doi:10.1117/12.672095 Elbaz D., et al., 2011, A&A, 533, A119 Everett W. B., et al., 2020, ApJ, 900, 55 Faisst A. L., et al., 2020, ApJS, 247, 61 Fazio G. G., et al., 2004, ApJS, 154, 10 Flores-Cacho I., et al., 2016, A&A, 585, A54 Förster Schreiber N. M., et al., 2018, ApJS, 238, 21 Franco M., et al., 2018, A&A, 620, A152 Fujimoto S., et al., 2020, ApJ, 900, 1 Galametz A., et al., 2013, ApJS, 206, 10 Geach J. E., Smail I., Coppin K., Moran S. M., Edge A. C., Ellis R. S., 2009, MNRAS, 395, L62 Geach J. E., Smail I., Moran S. M., MacArthur L. A., Lagos C. d. P., Edge A. C., 2011, ApJ, 730, L19 Genzel R., et al., 2008, ApJ, 687, 59 Giodini S., Lovisari L., Pointecouteau E., Ettori S., Reiprich T. H., Hoekstra H., 2013, Space Sci. Rev., 177, 247 Gobat R., et al., 2011, A&A, 526, A133 Goldader J. D., Meurer G., Heckman T. M., Seibert M., Sanders D. B., Calzetti D., Steidel C. C., 2002, ApJ, 568, 651 Greenslade J., et al., 2018, MNRAS, 476, 3336 Grogin N. A., et al., 2011, ApJS, 197, 35 Gullberg B., et al., 2019, MNRAS, 490, 4956 Harikane Y., et al., 2019, ApJ, 883, 142 Hayashi M., et al., 2018, ApJ, 856, 118 Hayward C. C., Jonsson P., Kereš D., Magnelli B., Hernquist L., Cox T. J., 2012, MNRAS, 424, 951 A. K., Salas H., 2019, A&A, 622, A103 Hayward C. C., Narayanan D., Kereš D., Jonsson P., Hopkins P. F., Cox T. J., Boselli A., Cortese L., Boquien M., Boissier S., Catinella B., Gavazzi G., Hernquist L., 2013, MNRAS, 428, 2529 Lagos C., Saintonge A., 2014, A&A, 564, A67 Bothwell M. S., et al., 2017, MNRAS, 466, 2825 Burgarella D., Buat V., Iglesias-Páramo J., 2005, MNRAS, 360, 1413 Bykov A. M., Churazov E. M., Ferrari C., Forman W. R., Kaastra J. S., Klein Hezaveh Y. D., et al., 2013, ApJ, 767, 132 Hill R., et al., 2020, MNRAS, 495, 3124 Hilton M., et al., 2013, MNRAS, 435, 3469 Hook I. M., Jørgensen I., Allington-Smith J. R., Davies R. L., Metcalfe N., U., Markevitch M., de Plaa J., 2015, Space Sci. Rev., 188, 141 Murowinski R. G., Crampton D., 2004, PASP, 116, 425 Cañameras R., et al., 2015, A&A, 581, A105 Calzetti D., Armus L., Bohlin R. C., Kinney A. L., Koornneef J., Storchi- Bergmann T., 2000, ApJ, 533, 682 Casey C. M., et al., 2015, ApJ, 808, L33 Castignani G., et al., 2020, A&A, 640, A64 Chabrier G., 2003, PASP, 115, 763 Champagne J. B., et al., 2021, ApJ, 913, 110 Chapman S. C., Blain A. W., Smail I., Ivison R. J., 2005, ApJ, 622, 772 Chapman S. C., Blain A., Ibata R., Ivison R. J., Smail I., Morrison G., 2009, ApJ, 691, 560 Chiang Y.-K., et al., 2015, ApJ, 808, 37 Cochrane R. K., et al., 2019, MNRAS, 488, 1779 Collins C. A., et al., 2009, Nature, 458, 603 Cooke K. C., Kartaltepe J. S., Tyler K. D., Darvish B., Casey C. M., Le Fèvre O., Salvato M., Scoville N., 2019, ApJ, 881, 150 Cowie L. L., Songaila A., Hu E. M., Cohen J. G., 1996, AJ, 112, 839 Cybulski R., et al., 2016, MNRAS, 459, 3287 Dannerbauer H., et al., 2014, A&A, 570, A55 De Lucia G., Blaizot J., 2007, MNRAS, 375, 2 Dessauges-Zavadsky M., et al., 2020, A&A, 643, A5 Dey A., Lee K.-S., Reddy N., Cooper M., Inami H., Hong S., Gonzalez A. H., Jannuzi B. T., 2016, ApJ, 823, 11 Huang N., et al., 2020, AJ, 159, 110 Hung C.-L., et al., 2016, ApJ, 826, 130 Ivison R. J., et al., 2007, MNRAS, 380, 199 Jablonka P., Combes F., Rines K., Finn R., Welch T., 2013, A&A, 557, A103 Kennicutt Jr. R. C., 1998, ARA&A, 36, 189 Khochfar S., Silk J., 2006, ApJ, 648, L21 Khusanova Y., et al., 2021, A&A, 649, A152 Kneissl R., et al., 2019, A&A, 625, A96 Koyama Y., et al., 2021, MNRAS, 503, L1 Kreysa E., et al., 2003, in Phillips T. G., Zmuidzinas J., eds, Society of Photo- Optical Instrumentation Engineers (SPIE) Conference Series Vol. 4855, Proc. SPIE. pp 41–48, doi:10.1117/12.459176 Lacaille K. M., et al., 2019, MNRAS, 488, 1790 Lang P., et al., 2019, ApJ, 879, 54 Law D. R., Steidel C. C., Erb D. K., Larkin J. E., Pettini M., Shapley A. E., Wright S. A., 2009, ApJ, 697, 2057 Law D. R., Steidel C. C., Shapley A. E., Nagy S. R., Reddy N. A., Erb D. K., 2012, ApJ, 745, 85 Le Fèvre O., et al., 2020, A&A, 643, A1 Lim S., Scott D., Babul A., Barnes D. J., Kay S. T., McCarthy I. G., Rennehan D., Vogelsberger M., 2021, MNRAS, 501, 1803 Lin L., et al., 2010, ApJ, 718, 1158 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / . / 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT The stellar content of SPT2349−56 27 Vieira J. D., et al., 2010, ApJ, 719, 763 Wall J. V., Pope A., Scott D., 2008, MNRAS, 383, 435 Wang T., et al., 2016, ApJ, 828, 56 Wang T., et al., 2018, ApJ, 867, L29 Wang G. C. P., et al., 2021, MNRAS, 508, 3754 Whitney A., Conselice C. J., Bhatawdekar R., Duncan K., 2019, ApJ, 887, 113 Wilkinson A., et al., 2017, MNRAS, 464, 1380 Wootten A., Thompson A. R., 2009, IEEE Proceedings, 97, 1463 Wright E. L., et al., 1992, ApJ, 396, L13 Wylezalek D., et al., 2013, ApJ, 769, 79 Young L. M., et al., 2011, MNRAS, 414, 940 Zabel N., et al., 2019, MNRAS, 483, 2251 Zhu H., Research in Astronomy and Astrophysics, 10, 329 Y.-N., Wu H.-N., Cao C., Li 2010, da Cunha E., et al., 2015, ApJ, 806, 110 van Dokkum P. G., et al., 2010, ApJ, 709, 1018 van der Burg R. F. J., et al., 2013, A&A, 557, A15 van der Burg R. F. J., Muzzin A., Hoekstra H., Wilson G., Lidman C., Yee H. K. C., 2014, A&A, 561, A79 van der Burg R. F. J., McGee S., Aussel H., Dahle H., Arnaud M., Pratt G. W., Muzzin A., 2018, A&A, 618, A140 van der Wel A., et al., 2012, ApJS, 203, 24 van der Wel A., et al., 2014, ApJ, 788, 28 This paper has been typeset from a TEX/LATEX ﬁle prepared by the author. Liu F. S., Mao S., Deng Z. G., Xia X. Y., Wen Z. L., 2009, MNRAS, 396, 2003 Liu F. S., Lei F. J., Meng X. M., Jiang D. F., 2015, MNRAS, 447, 1491 Łokas E. L., Mamon G. A., 2001, MNRAS, 321, 155 Long A. S., et al., 2020, ApJ, 898, 133 Magliocchetti M., et al., 2013, MNRAS, 433, 127 Mantz A. B., et al., 2018, A&A, 620, A2 Marshall H. L., Tananbaum H., Avni Y., Zamorani G., 1983, ApJ, 269, 35 Martinache C., et al., 2018, A&A, 620, A198 Matharu J., et al., 2019, MNRAS, 484, 595 Merlin E., et al., 2015, A&A, 582, A15 Merlin E., et al., 2016, A&A, 595, A97 Michałowski M. J., Dunlop J. S., Cirasuolo M., Hjorth J., Hayward C. C., Watson D., 2012, A&A, 541, A85 Miller T. B., Hayward C. C., Chapman S. C., Behroozi P. S., 2015, MNRAS, 452, 878 Miller T. B., et al., 2018, Nature, 556, 469 Mocanu L. M., et al., 2013, ApJ, 779, 61 Mowla L., van der Wel A., van Dokkum P., Miller T. B., 2019, ApJ, 872, L13 Naab T., Johansson P. H., Ostriker J. P., 2009, ApJ, 699, L178 Negrello M., et al., 2010, Science, 330, 800 Noble A. G., et al., 2017, ApJ, 842, L21 Noirot G., et al., 2018, ApJ, 859, 38 Noll S., Burgarella D., Giovannoli E., Buat V., Marcillac D., Muñoz-Mateos J. C., 2009, A&A, 507, 1793 Oogi T., Habe A., 2012, in Umemura M., Omukai K., eds, American Institute of Physics Conference Series Vol. 1480, First Stars IV - from Hayashi to the Future -. pp 406–408 (arXiv:1208.5039), doi:10.1063/1.4754402 Oteo I., et al., 2018, ApJ, 856, 72 Overzier R. A., 2016, A&ARv, 24, 14 Peng C. Y., Ho L. C., Impey C. D., Rix H.-W., 2002, AJ, 124, 266 Peng C. Y., Ho L. C., Impey C. D., Rix H.-W., 2010, AJ, 139, 2097 Planck Collaboration I 2020, A&A, 641, A1 Planck Collaboration XXVII 2015, A&A, 582, A30 Planck Collaboration XXVII 2016, A&A, 594, A27 Rennehan D., Babul A., Hayward C. C., Bottrell C., Hani M. H., Chapman S. C., 2020, MNRAS, 493, 4607 Robertson B., Cox T. J., Hernquist L., Franx M., Hopkins P. F., Martini P., Springel V., 2006, ApJ, 641, 21 Rosati P., et al., 2009, A&A, 508, 583 Rotermund K. M., et al., 2021, MNRAS, 502, 1797 Schreiber C., et al., 2015, A&A, 575, A74 Scott T. C., Usero A., Brinks E., Boselli A., Cortese L., Bravo-Alfaro H., 2013, MNRAS, 429, 221 Scoville N., et al., 2016, ApJ, 820, 83 Shen S., Mo H. J., White S. D. M., Blanton M. R., Kauﬀmann G., Voges W., Brinkmann J., Csabai I., 2003, MNRAS, 343, 978 Shimasaku K., et al., 2003, ApJ, 586, L111 Simpson J. M., et al., 2014, ApJ, 788, 125 Simpson J. M., et al., 2015, ApJ, 799, 81 Siringo G., et al., 2009, A&A, 497, 945 Speagle J. S., Steinhardt C. L., Capak P. L., Silverman J. D., 2014, ApJS, 214, 15 Spérone-Longin D., et al., 2021, A&A, 647, A156 Spilker J. S., et al., 2014, ApJ, 785, 149 Spilker J. S., et al., 2016, ApJ, 826, 112 Springel V., et al., 2005, Nature, 435, 629 Steidel C. C., Adelberger K. L., Shapley A. E., Pettini M., Dickinson M., Giavalisco M., 2000, ApJ, 532, 170 Steidel C. C., Adelberger K. L., Shapley A. E., Erb D. K., Reddy N. A., Pettini M., 2005, ApJ, 626, 44 Swinbank A. M., et al., 2010, MNRAS, 405, 234 Swinbank A. M., et al., 2014, MNRAS, 438, 1267 Tacconi L. J., et al., 2018, ApJ, 853, 179 Tadaki K.-i., et al., 2019, PASJ, 71, 40 Tamura Y., et al., 2009, Nature, 459, 61 Umehata H., et al., 2015, ApJ, 815, L8 Venemans B. P., et al., 2007, A&A, 461, 823 l D o w n o a d e d f r o m h t t p s : / / i / a c a d e m c . o u p . c o m m n r a s / a d v a n c e - a r t i c e d o / l i / / / . 1 0 1 0 9 3 m n r a s / s t a b 3 5 3 9 6 4 4 9 3 9 8 b y D T U L b r a r y u s e r o n 0 7 D e c e m b e r 2 0 2 1 i ORIGINAL UNEDITED MANUSCRIPT
10.1371_journal.pcbi.1011928	RESEARCH ARTICLE HormoneBayes: A novel Bayesian framework for the analysis of pulsatile hormone dynamics Margaritis VoliotisID 1 S. Dhillo2, Krasimira Tsaneva-AtanasovaID 1, Ali Abbara2, Julia K. Prague2,3,4, Johannes D. Veldhuis5, Waljit 1 Department of Mathematics and Living Systems Institute, College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, United Kingdom, 2 Department of Metabolism, Digestion and Reproduction, Imperial College London, Hammersmith Hospital, London, United Kingdom, 3 Department of Diabetes and Endocrinology, MacLeod Diabetes and Endocrine Centre, Royal Devon and Exeter Hospital, Exeter, United Kingdom, 4 College of Medicine and Health, University of Exeter, Exeter, United Kingdom, 5 Emeritus Mayo Clinic, Rochester, Michigan, United States of America [email protected] Abstract The hypothalamus is the central regulator of reproductive hormone secretion. Pulsatile secretion of gonadotropin releasing hormone (GnRH) is fundamental to physiological stimu- lation of the pituitary gland to release luteinizing hormone (LH) and follicle stimulating hor- mone (FSH). Furthermore, GnRH pulsatility is altered in common reproductive disorders such as polycystic ovary syndrome (PCOS) and hypothalamic amenorrhea (HA). LH is mea- sured routinely in clinical practice using an automated chemiluminescent immunoassay method and is the gold standard surrogate marker of GnRH. LH can be measured at fre- quent intervals (e.g., 10 minutely) to assess GnRH/LH pulsatility. However, this is rarely done in clinical practice because it is resource intensive, and there is no open-access, graphical interface software for computational analysis of the LH data available to clinicians. Here we present hormoneBayes, a novel open-access Bayesian framework that can be easily applied to reliably analyze serial LH measurements to assess LH pulsatility. The framework utilizes parsimonious models to simulate hypothalamic signals that drive LH dynamics, together with state-of-the-art (sequential) Monte-Carlo methods to infer key parameters and latent hypothalamic dynamics. We show that this method provides esti- mates for key pulse parameters including inter-pulse interval, secretion and clearance rates and identifies LH pulses in line with the widely used deconvolution method. We show that these parameters can distinguish LH pulsatility in different clinical contexts including in reproductive health and disease in men and women (e.g., healthy men, healthy women before and after menopause, women with HA or PCOS). A further advantage of hormone- Bayes is that our mathematical approach provides a quantified estimation of uncertainty. Our framework will complement methods enabling real-time in-vivo hormone monitoring and therefore has the potential to assist translation of personalized, data-driven, clinical care of patients presenting with conditions of reproductive hormone dysfunction. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Voliotis M, Abbara A, Prague JK, Veldhuis JD, Dhillo WS, Tsaneva-Atanasova K (2024) HormoneBayes: A novel Bayesian framework for the analysis of pulsatile hormone dynamics. PLoS Comput Biol 20(2): e1011928. https://doi.org/ 10.1371/journal.pcbi.1011928 Editor: Marc R. Birtwistle, Clemson University, UNITED STATES Received: June 16, 2023 Accepted: February 19, 2024 Published: February 29, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pcbi.1011928 Copyright: © 2024 Voliotis et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Code can be downloaded from https://git.exeter.ac.uk/mv286/ hormonebayes. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 1 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics Funding: MV and KTA acknowledge the financial support of the EPSRC via grants EP/T017856/1 and EP/N014391/1, and BBSRC via grants BB/ S000550/1 and BB/S001255/1. JKP is supported by a NIHR academic fellowship, MRC (MR/ M024954/1), and Expanding Excellence in England (E3) - Exeter Diabetes Research Unit. AA is supported by an NIHR Clinician Scientist Award (CS-2018-18-ST2-002). WSD is supported by an NIHR Senior Investigator Award. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Author summary Pulsatile hormone secretion is a widespread phenomenon underlying normal physiology and is also disrupted in many common endocrine disorders. To aid assessment and quan- tification of hormonal pulsatility, we developed hormoneBayes, a novel open-access Bayes- ian framework for analyzing hormonal measurements. The framework uses mathematical models to describe pulsatile dynamics, together with Bayesian methods to infer model parameter from data. We demonstrate HormoneBayes utility by analysing pulsatility of luteinising hormone (LH) data in different clinical contexts including in reproductive health and disease. Our framework in combination with real-time in-vivo hormone moni- toring has the potential to assist translation of personalized, data-driven, clinical care of patients presenting endocrine disorders. Introduction Pulsatile hormone dynamics are ubiquitous and play a crucial role in the regulation of many bodily functions related to metabolism, stress, and fertility [1,2]. Hormones are typically secreted in both a basal manner to maintain steady state levels, as well as with superimposed interspersed transient bursts (pulses) [3]. It is now established that the pulsatile nature of hor- monal secretion affects their interaction with receptors and downstream effector action [4–6]. With regards to fertility, the hypothalamus is the central regulator of the reproductive endo- crine axis. Notably, gonadotropin releasing hormone (GnRH) is secreted in a pulsatile man- ner, and seminal studies have demonstrated that this pulsatility is fundamental for its action to stimulate GnRH receptors on pituitary gonadotropes [5]. Moreover, disturbances in GnRH pulsatility are observed in common reproductive disorders including polycystic ovary syn- drome (PCOS) in which GnRH pulsatility is increased [7], and hypothalamic amenorrhea (HA) in which GnRH pulsatility is reduced [8]. However, despite this disparate alteration in GnRH pulsatility, differentiation of these two common reproductive disorders, which may both present similarly with menstrual disturbance, can be challenging [9]. LH is measured rou- tinely in clinical practice using an automated chemiluminescent immunoassay method and is the gold standard surrogate marker of GnRH. Furthermore, LH can be measured at frequent intervals (eg 10minutely) to assess GnRH/LH pulsatility, and accurate assessment of LH pulsa- tility could help facilitate diagnosis and treatment of patients presenting with reproductive endocrine disorders [9]. However, this is rarely done in clinical practice because it is resource- intensive, inconvenient for patients, and there is a lack of software for computational analysis of the LH data readily available to clinicians. Analysis of hormone pulsatility is a challenging computational problem, primarily due the stochastic nature of hormone dynamics and the consequent pulse-to-pulse variability, but also due to extrinsic factors (such as measurement error) obscuring the observed hormone dynam- ics [3]. Several computational methods for the analysis of endocrine data have been proposed in the literature [3,10–15], and deconvolution analysis, is the current gold-standard method for analyzing LH pulsatility in humans [3]. However, all methods lack open-access software implementation, with a user-friendly graphical interface that can be readily used by clinicians. To meet these challenges, we have developed HormoneBayes, a novel, open-access Bayesian framework for the analysis of hormone pulsatility data. HormoneBayes uses a stochastic model, describing hormone levels in the circulation incorporating measurement error, and leverages Bayesian statistics [16] to infer model parameters and latent variable dynamics. We PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 2 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics note that this approach is distinct to the deconvolution-based approach [3,14,15], which employs a single-pulse model to represent the data as a sequence of independent pulses. In this deconvolution-based method, the number of pulses becomes a model parameter that needs to be inferred from the data. In a Bayesian context, this leads to a posterior with unknown dimen- sions, hence posing significant challenges to inference [17,18]. We show that HormoneBayes can be used to accurately identify LH pulses and estimates clinically relevant measures such as inter-pulse intervals and secretion rates. The framework also provides a handle on estimation uncertainty via Bayesian posterior distributions. We showcase how this feature can be used to enable the understanding of alterations in LH pulsatility by analyzing the effect of direct hypo- thalamic stimulation using the neuropeptide kisspeptin on a subject-by-subject basis. We also demonstrate that HormoneBayes can be used to analyze LH pulsatility in different clinical con- texts/reproductive states (including healthy men, women before and after menopause, and women with reproductive disorders such as HA or PCOS). Importantly, the framework comes with an open-access graphical interface that make the core functionality of the framework eas- ily accessible to clinicians and clinical researchers. Results Analyzing pulsatile hormone dynamics using the hormoneBayes framework The hormoneBayes framework allows inference of key physiological parameters describing pulsatile hormone dynamics. The framework utilizes stochastic mathematical models describ- ing circulating hormone levels and state-of-the-art Bayesian machinery to calibrate these mod- els to data of hormone profiles and infer model parameters. Fig 1 presents a parsimonious model (a simple model with great explanatory power) describing circulating LH levels. The model assumes that there are two modes of LH secretion, namely pulsatile and basal. The Fig 1. HormoneBayes: a Bayesian framework for analyzing pulsatile LH dynamics. The framework uses a parsimonious mathematical model to describe LH levels in circulation as the net effect of secretion and clearance. In the model secretion is driven by a basal hypothalamic signal and a pulsatile signal (mimicking the dynamics of the GnRH pulse generator which can be turned ‘on’ or ‘off’). An efficient Markov-Chain Monte-Carlo (MCMC) method performs the Bayesian inference and extracts model parameters and latent hypothalamic dynamics, which are compatible with the observed data. https://doi.org/10.1371/journal.pcbi.1011928.g001 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 3 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics former corresponds to an on/off signal that randomly switches between two states (corre- sponding to a high and a low activity) while the latter corresponds to a continuous noisy signal. Furthermore, the model incorporates LH clearance as a linear first order process leading to an exponential decay of LH levels following a pulse [3]. We note that more detailed clearance models, such as the bi-exponential model [3], could be easily integrated. By considering the processes of LH release and clearance, the model predicts LH circulating dynamics in terms of five key parameters that can be recovered from data: 1. LH clearance rate; 2. maximum LH release rate; 3. strength of the pulsatile signal relative to the basal signal; 4. mean time in the on state; 5. mean time in the off state. Moreover, the model incorporates measurement error as an additional parameter that is determined based on the assay coefficient of variation (CV). HormoneBayes relies on the Bayesian paradigm to extract information from the observed data. Using the Bayes theorem, the method revises our prior beliefs regarding model parame- ters by transforming the parameters’ prior probability density distributions into posterior dis- tributions. Parameter prior distributions enable the user to input context-specific information into the analysis, hence enhancing the flexibility of the method to handle different datasets. For example, when dealing with data from post-menopausal women the user could choose to adjust the parameter priors to acknowledge a higher LH secretion rate and/or more sustained basal secretion. Fig 1 illustrates how the Bayesian machinery allows us to calibrate the model to the data and extract information regarding model parameter and latent hypothalamic sig- nals with an estimate of certainty. As we explain in the sections to follow, this information can be used to identify pulses; summarise the data (e.g., providing mean and standard deviation estimates); and perform statistical tests. Identifying LH pulses Using data to infer the latent hypothalamic signal provides a transparent way to identify pulses based on their likelihood under the model. As explained above, the model assumes LH pulses are partly driven by an on/off hypothalamic signal. This latent variable (i.e., inferred variable) takes two values indicating the ‘on’ (1) and ‘off’ (0) state of the hypothalamic pulse generator, and therefore the expected posterior estimate (inferred from LH profiling data) can be inter- preted as the probability that at any given time the hypothalamic pulse generator is ‘on’. This quantitative measure for accessing the likelihood of a pulse can significantly ease the job a cli- nician trying to decide whether an upstroke in the LH profile represents a pulse or not. Fig 2 Fig 2. Pulse identification using HormoneBayes. (A) Pulses can be identified using the expected value of the pulsatile hypothalamic signal, which can be interpreted as the probability of a pulse at a given timepoint. Here, we mark the onset of a pulse when the pulsatile hypothalamic signal crosses the 0.5 threshold, indicating that at this point a pulse is the most likely event. (B) The majority of the identified pulses (89%, 77/87) are in line with those obtained using the deconvolution method. For the analysis we used LH data obtained from healthy pre-menopausal women in early follicular phase (n = 16). https://doi.org/10.1371/journal.pcbi.1011928.g002 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 4 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics illustrates a representative example of an LH trace with two obvious pulses occurring at times 100min and 300min. These are indeed identified by inspecting the hypothalamic signal profile which peaks at around those times. At time 200min a less pronounced bump in LH could be indicative of an LH pulse, however the inferred pulsatile hypothalamic signal remains well below 0.5, indicating that under the current model there is higher probability that the bump is a measurement artefact rather than a pulse. To validate our pulse identification method, we used a database of LH profiles obtained from healthy pre-menopausal women and compared the identified pulses with those previ- ously obtained by the deconvolution method [19], which uses a maximum likelihood approach to fit a series of pulses to the data [3]. Here, we identify a pulse when the posterior probability that the hypothalamic signal is ‘on’ exceeds a threshold value. We use 0.5 as the threshold value, which signifies there is higher probability that the hypothalamic pulse generator is on (rather than off). This value yields the highest agreement between hormoneBayes and the deconvolution method (see Fig D in S1 Text). We find that hormoneBayes agrees with the deconvolution method in 77 out of 87 identified pulses (89%). Moreover, 13 pulses identified by hormoneBayes were not identified by the deconvolution method. Overall, this suggests a good agreement between the two methods, with hormoneBayes having the added advantage of providing a measure for the likelihood of each pulse that clinicians and researchers can use to inform their clinical decision making. Variation of model parameter within and across groups To test the applicability of hormoneBayes in different contexts we compile a database of LH profiles from four groups with diverse LH dynamics (men, healthy pre-menopausal women with regular menstrual cycles, post-menopausal women, women with HA, and women with PCOS). As illustrated in Fig 3, the model successfully captures the differences in LH dynamics in all four groups. Moreover, the model allows us to summarize LH data through model parameters and assess the variability across and within groups. We find that two model param- eters explain most of the variability between groups, namely the maximum secretion rate and pulsatility strength (Fig 3C). The first parameter describes how much LH can be secreted over time, whilst the second parameter quantifies the strength of the pulsatile signal relative to the basal signal. Based on these two parameters there is a strong distinction between women with PCOS and women with HA, who have lower LH secretion rates and/or diminished LH pulsati- lity strength (i.e., pulsatile signal is weak relative to the basal signal). Furthermore, post-meno- pausal women display higher secretion rates compared to pre-menopausal women but also reduced pulsatility strength (i.e., pulses are less pronounced when higher level of LH are estab- lished post menopause). Interestingly, healthy men and women illustrate a much lower param- eter variability as compared to HA and post-menopausal women, which could be indicative of various degrees of severity of HA/PCOS and tighter LH regulation in healthy individuals of reproductive age. Discussion We have presented HormoneBayes, a novel computational framework for analyzing hormone pulsatility. The framework combines (i) mathematical (mechanistic) models describing hor- mone dynamics with (ii) computational Bayesian machinery for inferring model parameters from data. HormoneBayes, comes with an open-access graphical user interface that make the core functionality of the framework easily accessible, a feature lacking from currently available analysis methods. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 5 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics Fig 3. HormoneBayes handles LH pulsatility analysis in different contexts. (A) Inferred pulsatility strength and maximum secretion rate parameters for different individuals: healthy men (n = 10); healthy post-menopause women (n = 13); healthy pre-menopausal women (n = 4). (B) Inferred parameters for healthy pre-menopausal women (n = 4); women with PCOS (n = 6) and women with HA (n = 5) illustrating how the assessment of LH pulsatility could help facilitate diagnosis of patients presenting with reproductive endocrine disorders. (C) Representative fits of the model are given for one subject in each dataset. https://doi.org/10.1371/journal.pcbi.1011928.g003 Using a parsimonious mathematical (generative) model of LH secretion, we have demon- strated the clinical utility of HormoneBayes in accurately describing LH profiles in various contexts (healthy men, healthy pre-menopausal women, post-menopausal women, women with PCOS and women with HA), and for identifying pulses. A novel feature of hormoneBayes is that it summarizes hormone profiles in terms of model parameters that can be used to pre- dict the underlying clinical conditions or reproductive state. Therefore, in the clinic hormone- Bayes could assist diagnosis based on hormonal profiles by evaluating how well the profile is described by different model/prior configurations, representing distinct physiological states corresponding to different clinical conditions (e.g., PCOS, HA). Ultimately, data analysis using HormoneBayes is as credible as the underlying generative model used to describe hormonal dynamics. Unlike deconvolution methods, where the num- ber of pulses is one of the model parameters to be inferred from the data, our approach relies on a generative model that assumes two modes of LH secretion: pulsatile and basal. This assumption is in par with current physiological understanding of the system and the hypotha- lamic pulse generator hypothesis [20,21]. Furthermore, to model LH circulation levels the model assumes a linear clearance rate. At least one other model used for the analysis of LH pul- satility has used more complex (multiple timescale) clearance dynamics, however we expect this assumption should have a minimal impact for the purpose of assessing LH pulsatility. Nevertheless, the modular design of HormoneBayes allows future extensions of the model with the scope of comparing how well different models capture LH dynamics as well as enabling the analysis of hormone dynamics beyond LH [22,23]. HormoneBayes utilizes the Bayesian paradigm to infer model parameters from the data. Within this paradigm, for each profile the method will output a (posterior) density distribution of model parameters, quantifying how probable parameter values are given the observed pro- file. This is fundamentally different from non-Bayesian methods, which provide point- PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 6 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics estimates of model parameters, as it allows for statistical testing. For instance, inferred poste- rior distributions can be used to evaluate the impact of hormonal interventions on LH secre- tion parameters, and crucially, this statistical evaluation can be conducted not only at the population level but also on an individual basis (see Fig E in S1 Text for an example of this type of personalised analysis). We expect these features of our method regarding personalized analysis will be crucial as measurement technologies mature enabling cheap sampling of hor- mone levels in real time [22,24]. Methods Ethics statement Data included in this manuscript were obtained from five different clinical research studies, involving healthy men [25], healthy pre-menopausal women [19], post-menopausal women [26], women with PCOS and women with HA [8]. Ethical approval for these studies was granted by: the Hammersmith and Queen Charlotte’s and Chelsea Hospitals Research Ethics Committee (registration number 05/Q0406/142) [8,19]; the UK National Research Ethics Committee-Central London (Research Ethics Committee number 14/LO/1098) [25]; and the West London Regional Ethics Committee (15/LO/1481) [26]. Written informed consent was obtained from all subjects. All studies were conducted according to Good Clinical Practice Guidelines. Data collection Participants attended a clinical research facility for 8 hour study visits hat included baseline (vehicle treatment) LH measurements according to the relevant trial protocol as previously described [8,19,25,26]. A cannula was inserted into a peripheral vein under aseptic conditions (time at least -30 minutes), through which all subsequent blood samples were taken every 10 minutes from time 0 until 480 minutes. All participants were ambulatory and could eat and drink freely during the study visit. All blood samples were left to clot for at least 30 minutes prior to centrifugation at 503 rcf for 10 minutes, after which the serum supernatant was extracted and immediately frozen at -20˚C prior to subsequent analysis using an automated chemiluminescent immunoassay method (Abbott Diagnostics, Maidenhead, UK) in batches after study completion. Reference ranges were as follows: LH 4–14 IU/L; respective intra-assay and inter-assay coefficients of variation were 4.1% and 2.7%; analytical sensitivity was 0.5 IU/L. Stochastic model of LH We used a discrete-time, stochastic model to describe pulsatile LH dynamics. The model com- prises of three dynamical variables, Pt, Bt, and LHt, that describe the pulsatile and basal hypo- thalamic signals and the LH concertation in circulation, respectively. The pulsatile hypothalamic signal Pt can take two values: Ht = 0 corresponding to the ON state; and Ht = 1 corresponding to the OFF state. The stochastic dynamics of Ht are governed by the following probability matrix Hs 0 1 https://doi.org/10.1371/journal.pcbi.1011928.t001 Hs+δt 0 1 (cid:0) 1 tOFF � dt 1 tON � dt 1 1 tON 1 (cid:0) � dt 1 tOFF � dt PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 7 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics i.e., parameters τON and τOFF govern the probabilities that the value of H will either flip or remain the same over the time interval (s, s+δt). The evolution of the basal hypothalamic signal, Bt, is described using a discrete time autore- gressive model obeying the following equation Xtþdt ¼ Xt (cid:0) p ffiffiffiffiffi dt � εt; Xt þ dt 2 Bt ¼ 1 1 þ e(cid:0) Xt ; where εt is a normally distributed random variable with zero mean and unit variance. Note that both Bt and Ht are bounded in the interval [0,1]. The two hypothalamic signals drive LH secretion, and along with LH clearance dictate the circulating LH levels, LHt. The equation describing the time evolution of LHt, is LHtþdt ¼ LHt þ ½kðPt � f þ Bt � ð1 (cid:0) f ÞÞ (cid:0) d � LHt� � dt where d denotes the clearance rate, k denotes the maximum secretion rate, and parameter f (termed pulsatility strength) describes the relative strength of the two hypothalamic signals. Finally, the model assumes measurement error in the form: LHobs t ¼ LHtð1 þ ZtÞ where ηt is a normally distributed random variable with zero mean and std. deviation equal to the CV of the assay. Throughout our analysis we have used δt = 1 min, hence, assuming the system dynamics do not change significantly over shorter times. Bayesian inference The hormoneBayes framework uses Bayesian inference to obtain model parameters Θ = (τON, τOFF, k, d, f) and latent variable (Ht, Bt) dynamics from LH profiling data D. In particular, hor- moneBayes solves the inference problem by sampling from the target posterior distribution: ð P Y; Ht; BtjD Þ ¼ PðD; Ht; BtjYÞ � PðYÞ PðDÞ ; where P(Θ) is the prior parameter distribution; PðD; Ht; BtjYÞ ¼ PðDjY; Ht; BtÞ � PðHt; BtÞ is the likelihood of the data given the parameters; and PðDÞ ¼ marginal likelihood or model evidence. PðD; Ht; BtjYÞ � PðYÞ is the R Sampling from the full posterior distribution is performed using a Gibbs sampler, which is an iterative Monte Carlo Markov Chain (MCMC) scheme. The algorithm is initialised with parameter values drawn from the prior distribution, i.e., Θ0~P(Θ) and each subsequent itera- tion, i = 1,. . .,M involves two steps: (1) sampling latent variables (Ht, Bt)i given the data, D, and the current parameter values Θi−1 and (2) sampling new parameter values Θi given D and the latent variables (Ht, Bt)i. The first step is performed using Sequential Monte Carlo (SMC) with ancestral sampling [27]. The second step is further broken down into two parts, first parame- ters (τON, τOFF) are sampled using an adaptive Metropolis-Hastings sampler and then parame- ters (k, d, f) are sampled using the simplified version of the manifold Metropolis adjusted Langevin algorithm (sMMALA) presented in [28]. For the analysis of all datasets in this study we considered the following independent prior distributions: log10tON � Uðlog10ð5Þ; log10ð240ÞÞ and log10tOFF � Uðlog10ð5Þ; log10ð240ÞÞ, based on the sampling rate (10min) and duration (480min) used in the LH profiling studies; log10ðkÞ � N ð0; 5Þ, set as a broad uninformative prior; logð2Þ Þ, based on LH half- ð � N 80; 9:3 d PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 8 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics life data; and log(f)~U(0,1), to accommodate analysis of different LH profiles with high or low pulsatility. Evaluation of the algorithm on synthetic dataset can be found in Fig A-C in S1 Text. HormoneBayes also allows the user to access the effect of pharmacological interventions on LH pulsatility, by fitting in tandem two LH profiling datasets: corresponding to periods before and after the intervention. In this case a composite model is used to allow inference of parame- ters Θc = (τON, τOFF, k, d, f), corresponding to the baseline period (before the intervention), and parameters Θp = (τON,i, τOFF,i, ki, fi) corresponding to the period after the intervention. Here we assume the intervention does not affect the clearance rate d, hence this parameter does not appear in Θp. In mathematical terms the target posterior is now given by � P Yc; YpjDc; Dp � ¼ PðDc; DpjYc; YpÞ � PðYc; YpÞ PðDc; DpÞ ; and sampling from the posterior is performed as described above. An example of this analysis is Fig E of the S1 Text. An open access C++ implementation of HormoneBayes accompanied with a graphical interface implemented in Python and a user manual can be found at https://git.exeter.ac.uk/ mv286/hormonebayes. Supporting information S1 Text. Supplementary figures. Fig A: Testing HormoneBayes on synthetic data. Fig B: Assessing the effect of the prior for the LH clearance rate. Fig C: Tuning HormoneBayes when pulses are not clear by using a more informative prior on parameter f. Fig D: Pulse identifica- tion using HormoneBayes. Fig E: Using HormoneBayes to identify the effect of interventions on LH pulsatility. (PDF) Author Contributions Conceptualization: Margaritis Voliotis. Data curation: Ali Abbara, Julia K. Prague. Formal analysis: Margaritis Voliotis. Investigation: Margaritis Voliotis, Ali Abbara, Julia K. Prague, Waljit S. Dhillo, Krasimira Tsa- neva-Atanasova. Methodology: Margaritis Voliotis, Ali Abbara, Krasimira Tsaneva-Atanasova. Resources: Ali Abbara, Julia K. Prague, Johannes D. Veldhuis, Waljit S. Dhillo. Software: Margaritis Voliotis. Writing – original draft: Margaritis Voliotis. Writing – review & editing: Ali Abbara, Julia K. Prague, Johannes D. Veldhuis, Waljit S. Dhillo, Krasimira Tsaneva-Atanasova. References 1. Zavala E, Wedgwood KCA, Voliotis M, Tabak J, Spiga F, Lightman SL, et al. Mathematical Modelling of Endocrine Systems. Trends Endocrinol Metab. 2019; 30(4):244–57. https://doi.org/10.1016/j.tem.2019. 01.008 PMID: 30799185 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 9 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics 2. Le Tissier P, Fiordelisio Coll T, Mollard P. The Processes of Anterior Pituitary Hormone Pulse Genera- tion. Endocrinology. 2018; 159(10):3524–35. https://doi.org/10.1210/en.2018-00508 PMID: 30020429 3. Keenan DM, Veldhuis JD. Pulsatility of Hypothalamo-Pituitary Hormones: A Challenge in Quantification. Physiology (Bethesda). 2016; 31(1):34–50. https://doi.org/10.1152/physiol.00027.2015 PMID: 26674550 4. Lightman SL, Conway-Campbell BL. The crucial role of pulsatile activity of the HPA axis for continuous dynamic equilibration. Nat Rev Neurosci. 2010; 11(10):710–8. https://doi.org/10.1038/nrn2914 PMID: 20842176 5. Belchetz PE, Plant TM, Nakai Y, Keogh EJ, Knobil E. Hypophysial responses to continuous and inter- mittent delivery of hypopthalamic gonadotropin-releasing hormone. Science. 1978; 202(4368):631–3. https://doi.org/10.1126/science.100883 PMID: 100883 6. McArdle CA, Roberson MS. Gonadotropes and Gonadotropin-Releasing Hormone Signaling. In: Plant TM, Zeleznik AJ, editors. Knobil and Neill’s physiology of reproduction. 4th ed: Academic Press; 2015. 7. Abbara A, Dhillo WS. Targeting Elevated GnRH Pulsatility to Treat Polycystic Ovary Syndrome. J Clin Endocrinol Metab. 2021; 106(10):e4275–e7. https://doi.org/10.1210/clinem/dgab422 PMID: 34117885 8. Jayasena CN, Abbara A, Veldhuis JD, Comninos AN, Ratnasabapathy R, De Silva A, et al. Increasing LH pulsatility in women with hypothalamic amenorrhoea using intravenous infusion of Kisspeptin-54. J Clin Endocrinol Metab. 2014; 99(6):E953–61. https://doi.org/10.1210/jc.2013-1569 PMID: 24517142 9. Phylactou M, Clarke SA, Patel B, Baggaley C, Jayasena CN, Kelsey TW, et al. Clinical and biochemical discriminants between functional hypothalamic amenorrhoea (FHA) and polycystic ovary syndrome (PCOS). Clin Endocrinol (Oxf). 2020. 10. Granqvist E, Hartley M, Morris RJ. BaSAR-A tool in R for frequency detection. Biosystems. 2012; 110 (1):60–3. https://doi.org/10.1016/j.biosystems.2012.07.004 PMID: 22925599 11. Veldhuis JD, Johnson ML. Cluster analysis: a simple, versatile, and robust algorithm for endocrine pulse detection. Am J Physiol. 1986; 250(4 Pt 1):E486–93. https://doi.org/10.1152/ajpendo.1986.250.4. E486 PMID: 3008572 12. Vidal A, Zhang Q, Medigue C, Fabre S, Clement F. DynPeak: an algorithm for pulse detection and fre- quency analysis in hormonal time series. PLoS One. 2012; 7(7):e39001. https://doi.org/10.1371/ journal.pone.0039001 PMID: 22802933 13. Keenan DM, Veldhuis JD, Yang R. Joint recovery of pulsatile and basal hormone secretion by stochas- tic nonlinear random-effects analysis. Am J Physiol. 1998; 275(6):R1939–49. https://doi.org/10.1152/ ajpregu.1998.275.6.R1939 PMID: 9843883 14. 15. Johnson TD. Bayesian deconvolution analysis of pulsatile hormone concentration profiles. Biometrics. 2003; 59(3):650–60. https://doi.org/10.1111/1541-0420.00075 PMID: 14601766 Liu H, Polotsky AJ, Grunwald GK, Carlson NE. Bayesian analysis improves pulse secretion characteri- zation in reproductive hormones. Syst Biol Reprod Med. 2018; 64(1):80–91. https://doi.org/10.1080/ 19396368.2017.1411541 PMID: 29287490 16. Lesaffre E, Lawson A. Bayesian biostatistics. Chichester, West Sussex: Wiley, 2012. 17. Green PJ. Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika. 1995; 82(4):711–32. 18. Stephens M. Bayesian Analysis of Mixture Models with an Unknown Number of Components- An Alter- native to Reversible Jump Methods. The Annals of Statistics. 2000; 28(1):40–74. 19. Narayanaswamy S, Jayasena CN, Ng N, Ratnasabapathy R, Prague JK, Papadopoulou D, et al. Sub- cutaneous infusion of kisspeptin-54 stimulates gonadotrophin release in women and the response cor- relates with basal oestradiol levels. Clin Endocrinol (Oxf). 2016; 84(6):939–45. https://doi.org/10.1111/ cen.12977 PMID: 26572695 20. Voliotis M, Li XF, De Burgh R, Lass G, Lightman SL, O’Byrne KT, et al. The Origin of GnRH Pulse Gen- eration: An Integrative Mathematical-Experimental Approach. J Neurosci. 2019; 39(49):9738–47. https://doi.org/10.1523/JNEUROSCI.0828-19.2019 PMID: 31645462 21. Clarkson J, Han SY, Piet R, McLennan T, Kane GM, Ng J, et al. Definition of the hypothalamic GnRH pulse generator in mice. Proc Natl Acad Sci U S A. 2017; 114(47):E10216–E23. https://doi.org/10. 1073/pnas.1713897114 PMID: 29109258 22. Upton TJ, Zavala E, Methlie P, Kampe O, Tsagarakis S, Oksnes M, et al. High-resolution daily profiles of tissue adrenal steroids by portable automated collection. Sci Transl Med. 2023; 15(701):eadg8464. https://doi.org/10.1126/scitranslmed.adg8464 PMID: 37343084 23. Spiga F, Walker JJ, Terry JR, Lightman SL. HPA axis-rhythms. Compr Physiol. 2014; 4(3):1273–98. https://doi.org/10.1002/cphy.c140003 PMID: 24944037 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 10 / 11 PLOS COMPUTATIONAL BIOLOGYHormoneBayes: a novel Bayesian framework for the analysis of pulsatile hormone dynamics 24. Liang S, Kinghorn AB, Voliotis M, Prague JK, Veldhuis JD, Tsaneva-Atanasova K, et al. Measuring luteinising hormone pulsatility with a robotic aptamer-enabled electrochemical reader. Nat Commun. 2019; 10(1):852. https://doi.org/10.1038/s41467-019-08799-6 PMID: 30787284 25. Narayanaswamy S, Prague JK, Jayasena CN, Papadopoulou DA, Mizamtsidi M, Shah AJ, et al. Investi- gating the KNDy Hypothesis in Humans by Coadministration of Kisspeptin, Neurokinin B, and Naltrex- one in Men. J Clin Endocrinol Metab. 2016; 101(9):3429–36. https://doi.org/10.1210/jc.2016-1911 PMID: 27379743 26. Prague JK, Roberts RE, Comninos AN, Clarke S, Jayasena CN, Nash Z, et al. Neurokinin 3 receptor antagonism as a novel treatment for menopausal hot flushes: a phase 2, randomised, double-blind, pla- cebo-controlled trial. Lancet. 2017; 389(10081):1809–20. https://doi.org/10.1016/S0140-6736(17) 30823-1 PMID: 28385352 27. Lindsten F, Jordan MI, Schon TB. Particle Gibbs with Ancestor Sampling. Journal of Machine Learning Research. 2014; 15:2145–84. 28. Girolami M, Calderhead B. Riemann manifold Langevin and Hamiltonian Monte Carlo methods. J R Statist Soc B. 2011; 73:123–214. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011928 February 29, 2024 11 / 11 PLOS COMPUTATIONAL BIOLOGY
10.1093_nar_gkad439	6528–6539 Nucleic Acids Research, 2023, Vol. 51, No. 13 https://doi.org/10.1093/nar/gkad439 Published online 30 May 2023 NAR Breakthrough Article Enhanced nonenzymatic RNA copying with in-situ activation of short oligonucleotides Dian Ding 1 , 2 , † , Stephanie J. Zhang 1 , 2 , † and Jack W. Szostak 1 , 2 , 3 , 4 ,* 1 Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford Street, Cambridge, MA 02138, USA, 2 Department of Molecular Biology and Center for Computational and Integ r ative Biology, Massachusetts General Hospital, 185 Cambridge Street, Boston, MA 02114, USA, 3 Department of Genetics, Harvard Medical School, 77 Avenue Louis Pasteur, Boston, MA 02115, USA and 4 Ho w ard Hughes Medical Institute, Department of Chemistry, The University of Chicago, Chicago, IL 60637, USA Received April 12, 2023; Revised April 28, 2023; Editorial Decision May 03, 2023; Accepted May 10, 2023 ABSTRACT The nonenzymatic copying of RNA is thought to have been necessary for the transition between pre- biotic chemistry and ribozyme-catalyzed RNA repli- cation in the RNA World. We have previously shown that a potentially prebiotic nucleotide activation path- way based on phospho-Passerini chemistry can lead to the efﬁcient synthesis of 2-aminoimidazole activ ated mononuc leotides when carried out un- der freeze-thaw cycling conditions. Such activated (cid:2) – nucleotides react with each other to form 5 (cid:2) 2-aminoimidazolium bridged dinucleotides, en- 5 abling template-directed primer extension to occur within the same reaction mixture. Ho we ver, mononu- (cid:2) 2- cleotides linked to oligonucleotides by a 5 aminoimidazolium bridge are superior substrates for nonenzymatic primer extension; their higher in- trinsic reactivity and their higher template afﬁnity enable faster template copying at lower substrate concentrations. Here we show that eutectic phase phospho-Passerini chemistry efﬁciently activates short oligonucleotides and promotes the formation of monomer -bridged-oligonuc leotide species during freeze-thaw cycles. We then demonstrate that in-situ g enerated monomer-bridg ed-oligonucleotides lead to efﬁcient nonenzymatic template copying in the same reaction mixture. Our demonstration that mul- tiple steps in the pathway from activation chemistry (cid:2) –5 to RNA copying can occur together in a single com- plex environment simpliﬁes this aspect of the origin of life. GRAPHICAL ABSTRACT INTRODUCTION The origin of life on the early Earth is likely to have involved a series of steps during which ribonucleotides were synthe- sized through abiotic chemical pathways, and subsequently assembled into short RNA oligonucleotides that encoded the genetic information of primordial life ( 1–6 ). A key step in the transition from prebiotic chemistry to the emergence * To whom correspondence should be addressed. Email: [email protected] † The authors wish it to be known that, in their opinion, the first two authors should be regarded as Joint First Authors. Pr esent addr esses: Dian Ding, Chemical Biology and Therapeutics Science Program, Broad Institute of MIT and Harvard, 415 Main St, Cambridge, MA 02142, USA. Stephanie J. Zhang, Department of Pathology, Brigham and Women’s Hospital, 60 Fenwood Rd, Boston, MA 02115, USA. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. of self-replicating RNAs and the advent of Darwinian evo- lution is thought to have been nonenzymatic RNA copying. For decades, template-directed nonenzymatic primer ex- tension reactions have been modeled using intrinsically re- acti v e mononucleotide phosphorimidazolides (Supplemen- (cid:3) - (cid:3) ,5 tary Figure S1A) ( 7–9 ). Recent results suggest that 5 imidazolium-bridged-dinucleotides are the dominant re- acti v e intermediates in facilitating chemical RNA copy- ing in the presence of activated mononucleotides, due to their higher reactivity and stronger binding to the template by two Watson–Crick base pairs (Supplementary Figure S1B) ( 10–13 ). Crystallo gra phic studies have shown that the template-bound bridged dinucleotide intermediate is pre- organized so as to favor the primer extension reaction, such (cid:3) -OH of the primer is properly oriented and in that the 3 close proximity to the adjacent phosphate of the bridged dinucleotide ( 14 , 15 ). While activated mononucleotides can contribute to template copying if present in high concen- trations, the copying fidelity is extremely poor ( 16 ). De- spite these advances, the chemical copying of RNA tem- plates with all four canonical nucleotides remains ineffi- cient. This inefficiency stems in part from the large dif- ferences in affinities and reactivities of the 10 different 2- aminoimidazole(2AI)-bridged dinucleotides, which results in low yields and sequence-dependent biases in template copying ( 13 ). To overcome this problem, monomer-bridged- oligonucleotide intermediates have been proposed to facilitate fast and unbiased copying of templates con- taining all four bases ( 13 , 17 ). Compared to activated mononucleotides and bridged dinucleotides, monomer- bridged-oligonucleotides can bind to the primer-template complex with significantly higher affinity and lead to a 10-fold higher rate of primer extension at template satura- tion (Figure 1Ai) ( 13 ). Thus, short oligonucleotides of only 2 to 4 nucleotides in length can act as sequence-specific ca talysts for ef ficient templa te copying. Such short oligonu- cleotides may be generated prebiotically via untemplated ( 18–20 ) or templa te-directed oligomeriza tion ( 21–23 ). In addition to primer extension, template copying can also be dri v en by the ligation of short oligomeric substrates, which is slower but r equir es fewer reaction steps ( 24 , 25 ). The nonenzyma tic templa ted assembly of functional ribozymes has been shown to be possible using splinted ligation ( 25–27 ). Ther efor e, it is important to explore whether monomer-bridged-oligonucleotide substrates might also enhance nonenzyma tic RNA liga tion for templa te copying (Figure 1 Bi). Gi v en the enhanced reactivity of monomer-bridged- oligonucleotides and the likely prebiotic availability of short oligonucleotides, we hav e inv estigated the in-situ acti va- tion of short oligonucleotides to see if they can be uti- lized as prebiotically available catalysts. Sutherland and cow ork ers have previously reported a nucleotide activa- tion chemistry utilizing methyl isocyanide (MeNC), a com- pound that can be synthesized in a prebiotic ferrocyanide- and nitr oprusside-containing envir onment after ultraviolet (UV) irradiation ( 28 ). By modifying this MeNC-mediated phospho-Passerini activation chemistry, our lab has re- cently identified a potentially prebiotic pathwa y f or the ac- tivation of RNA mononucleotides and the formation of Nucleic Acids Research, 2023, Vol. 51, No. 13 6529 bridged dinucleotides under conditions compatible with nonenzyma tic templa te cop ying ( 29 ). This r eaction is con- ducted in the eutectic phase of a partially frozen reaction mixtur e to incr ease the effecti v e concentrations of reactants, and thus promote bridged dinucleotide formation at a sto- ichiometric 2AI concentration. This pathway allowed us to demonstrate a prebiotically plausible scenario in which the hitherto separated stages of nucleotide activation, bridged dinucleotide forma tion, and nonenzyma tic RNA copying occurr ed concurr entl y under m utuall y compatible condi- tions. Considering the expected heterogeneity of prebiotic mixtures of RNA mono- and oligonucleotides, we have now investigated the activation of short oligonucleotides and ex- plored conditions that allow the subsequent formation of monomer-bridged-oligonucleotides for enhanced nonenzy- matic RNA copying. Here, we show that monomer-bridged-oligonucleotides enhance both nonenzymatic primer extension and lig- ation. We then demonstrate that e v en low concentra- tions of activated oligonucleotides can generate sufficient monomer-bridged-oligonucleotide intermediates to dri v e ef ficient templa te copying. Following the demonstra tion of enhanced RNA copying with pre-activated oligonu- cleotides, we employ the MeNC-mediated chemistry to acti- vate oligonucleotides and dri v e the formation of monomer- bridged-oligonucleotides. Finally, we demonstrate efficient nonenzyma tic templa te copying starting with unactiva ted mono- and oligonucleotides under conditions of in-situ ac- tivation. Our findings contribute to the growing realization that geochemically realistic environments with the right de- gree of chemical complexity must be considered in order to understand the origin of life. MATERIALS AND METHODS Gener al inf ormation Unless otherwise noted, all chemicals were purchased at the highest available purity from Sigma-Aldrich (St. Louis, (cid:3) -monophosphates were obtained MO). All ribonucleoside 5 as free acids from MP Biotechnology (Solon, OH) or Sigma-Aldrich (St. Louis, MO). 2-Aminoimidazole hy- drochloride was obtained from Combi-Blocks (San Diego, CA). Phosphoramidites and reagents for solid-phase RNA synthesis were purchased from ChemGenes (Wilmington, MA) and Glen Research (Ster ling, MA). Deuter ated sol- vents wer e pur chased from Cambridge Isotope Laborato- ries (Tewksbury, MA). The synthesis and storage of methyl isocyanide were carried out as previously described ( 30 ). To pre v ent the acidic hydrolysis of MeNC, which forms methylamine and formic acid, MeNC should be stored at a pH > 8. The ab- solute concentrations of stock solutions were determined by comparing the integrals of 1 H peaks of interest to the (cid:3) -monophosphate, using NMR spec- calibrant, adenosine-5 (cid:3) - troscopy. The exact concentrations of the nucleoside-5 monophosphates were determined by the analysis of se- rial dilutions on a spectrophotometer. When pipetting small volumes of the relati v ely vola tile MeNC , it is common to ex- perience loss of the MeNC inside the pipette tip, likely due to ra pid eva pora tion. To ensure tha t the correct amount of MeNC is transferred to the reaction mixture, we pre-wet 6530 Nucleic Acids Research, 2023, Vol. 51, No. 13 the pipette tips (Sorenson, BioScience, Inc.) by pipetting MeNC up and down at least three times, followed by im- mediate transfer of the pipetted MeNC solution into the re- action mixture and vortexing to ensure complete mixing. Re v erse-phase flash chromato gra phy was performed us- ing prepacked RediSep Rf Gold C18Aq 50 g columns from Teledyne Isco (Lincoln, NE). Preparatory-scale high- performance liquid chromato gra phy (HPLC) was carried out on an Agilent 1290 HPLC system equipped with an Ag- ilent ZORBAX Eclipse-XDB C18 column (21.2 × 250 mm, 7 (cid:2)m particle size) for re v ersed-phase chromato gra phy. An- alytical HPLC was carried out on an Agilent 1100 series with an Agilent Eclipse Plus C18 column (4.6 × 250 mm, 5 ◦C (cid:2)m particle size). 31 P NMR spectra wer e acquir ed at 25 on a Varian Oxford AS-400 NMR spectrometer (162 MHz for 31 P). Oligonucleotide synthesis Primers and templates wer e pur chased from Integrated DN A Technolo gies. The din ucleotide, trin ucleotides, and tetranucleotide (pCG, pCGC, pACG and pCGCA) used for activation chemistry wer e pr epar ed in-house by solid- phase synthesis on a MerMade 6 DN A / RN A synthesizer (Bioautomation, Plano, TX). Authentic samples of the (cid:3) -pyrophosphate di- three riboadenosine dinucleotides, 5 (cid:3) -monophosphate with a adenosine (AppA), di-adenosine 5 (cid:3) -pA), and di-adenosine (cid:3) phosphodiester linkage (pA-2 (cid:3) -5 2 (cid:3) phosphodiester linkage (cid:3) –5 (cid:3) -monophosphate with a 3 5 (cid:3) -pA) wer e pr epar ed by solid-phase synthesis on an (pA-3 Expedite 8909 DN A / RN A synthesiz er. The synthesiz ed oligonucleotides were then deprotected and purified by re v erse-phase chromato gra phy with a 50 g C18Aq column using a gradient elution of 0–10% acetonitrile over 10 col- umn volumes (CVs) in 20 mM triethylammonium bicarbon- ate (TEAB) at pH 7.5. (cid:3) –5 Pr epar ation of activ ated mononucleotides, bridged dinu- cleotides, activated oligonucleotides and mononucleotide- bridged-oligonucleotides All 2-aminoimidazole-activated RNAs (pN n and NppN n , n = 1, 2, 3, 4) were synthesized as previously described ( 13 ). The activated mononucleotide pA was purified by re v erse phase chromato gra phy with a 50 g C18Aq column over 10 CVs of 0–10% acetonitrile in 2 mM TEAB (pH 8). The fractions containing pA were adjusted to pH 9.5–10 with NaOH, aliquoted, and l yophilized. A part from using 20 mM TEAB (pH 7.5), the activated oligonucleotides (pCG, pCGC and pCGCA) were purified in the same manner. The bridged dinucleotide AppC was purified by re v erse- phase HPLC on a C18 column with a gradient of 2–8% acetonitrile over 20 min in 2 mM TEAB (pH 8) at a flow rate of 15 ml / min. Fractions containing AppC were ad- justed to pH 8 with HCl, aliquoted, and lyophilized. The monomer-bridged oligonucleotide intermediates AppCG, A ppCGC and A ppCGCA wer e purified by r e v erse-phase HPLC on a C18 column with a gradient of 2–10% acetoni- trile over 27 min in 20 mM TEAB (pH 7.5) at a flow rate of 15 ml / min. Fractions containing the desired products were adjusted to pH 8 with HCl, aliquoted, and lyophilized. Nonenzymatic primer extension and ligation with pre- synthesized activated RNAs FAM-labeled primer / FAM / A GUGA GUAACUC was used in the nonenzymatic primer extension and ligation reactions with pre-synthesized activated RNAs unless otherwise noted. The primer extension reactions used the (cid:3) -OH- UGCGU GAGUUA CUCA CUAAA. template 5 (cid:3) -OH- Ligation 5 UGCG GAGUUA CUCA CUAAA. Underlined sequences indicate the template region available for substrate binding. reactions template used the Kinetics of primer extension with a single activated RNA sub- strate (pN n or NppN n ). The primer-template complex was first pr epar ed in an annealing solution containing 7.5 (cid:2)M primer, 12.5 (cid:2)M template, 50 mM Tris–HCl (pH 8), 50 mM NaCl and 1 mM ethylenediaminetetraacetic acid ◦C for 30 s and slowly cooling down (EDTA) by hea ting a t 85 ◦C / s. The annealed product was then ◦C at a rate of 0.1 to 25 diluted into the final reaction mixture to a final concentra- tion of 1.5 (cid:2)M primer, 2.5 (cid:2)M template, 200 mM Tris– HCl (pH 8), 100 mM MgCl 2 , and the indicated concentra- tion of the pre-activa ted substra te. The stock solutions of pre-activated species were prepared freshly at various con- centra tions immedia tely befor e being added to the final r e- action mixture to initiate the template copying reaction at room temperature. At each time point, 0.5 (cid:2)l of the reaction mixture was added to 25 (cid:2)l of a quenching buffer contain- ing 25 mM EDTA, 1 × TBE, and 4 (cid:2)M DNA strands com- plementary to the template in formamide. Template copy- ing products were resolved by 20% (19:1) denaturing PAGE, scanned with a Typhoon 9410 scanner, and quantified us- ing the ImageQuant TL software. Reactions containing a blocker oligonucleotide were performed as described above, except using the sequences listed in Supplementary Figure S3D, with final concentrations of 1.5 (cid:2)M primer, 2.5 (cid:2)M template, and 3.5 (cid:2)M blocker. Primer extension with a mixture of activated mono- and oligonucleotides (pA and pN n ). The annealing buffer was pr epar ed as described above with 7.5 (cid:2)M primer and 12.5 (cid:2)M template. The annealing buffer contain- ing the primer-template complex was mixed with MgCl 2 and Tris–HCl (pH 8) at the bottom of the reaction tube, while the fr eshly pr epar ed stock solutions of acti- vated mononucleotide (pA) and activated oligonucleotide (pCG, pCGC or pCGCA) were added separately to the lid or wall of the tube. The reaction tube was immediately centrifuged and vortexed to mix all the materials to initiate the reaction with final concentrations of 1.5 (cid:2)M primer, 2.5 (cid:2)M template, 200 mM Tris–HCl (pH 8), 100 mM MgCl 2 , 5 mM A and 5 / 0.5 / 0.05 mM activated oligonucleotide. Samples were collected at each time point and analyzed as described above. Primer extension with a pre-incubated mixture of activated mono- and oligonucleotides (pA and pCGCA). The an- nealed primer-template complex was pr epar ed as described above and lyophilized to dryness. A mixture of 5 mM A, 0.5 mM CGCA, 200 mM Tris–HCl (pH 8) and 100 mM MgCl 2 was incubated for 1 h at room temperature before being added to the lyophilized primer-template complex, to gi v e final concentrations of 1.5 (cid:2)M primer and 2.5 (cid:2)M tem- plate. Samples were collected at each time point and ana- lyzed as described above. HPLC analysis of spontaneous bridge-formation between ac- tivated mono- and oligonucleotides The activated tetramer pCGCA was used for the kinetic analysis of bridged species due to its strong UV absorbance, which allowed for effecti v e analysis b y HPL C. In primer ex- tension buffer (100 mM MgCl 2 & 200 mM Tris–HCl (pH 8) or 30 mM MgCl 2 & 50 mM Na + -HEPES (pH 8)), 5 mM pA was allowed to react with 0.5 mM pCGCA. After in- cuba tion a t room tempera tur e for the desir ed time, 0.5 M EDTA solution was added in 2.5-fold excess over the MgCl 2 concentration, and the resultant mixture was kept on dry ice before injection or injected immediately into an Agilent ZORBAX Eclipse-X DB C18 column for HPLC analysis. The sample was separated using (A) aqueous 25 mM TEAB buffer (pH 8) and (B) acetonitrile, with a gradient of 3–10% B over 25 min at a flow rate of 15 ml / min. Fractions were collected and analyzed by mass spectroscopy to confirm the identity of the collected species. HPLC analysis of MeNC-mediated activation and bridge- formation Oligonucleotide activation at ambient temperature. 400 mM MeNC, 400 mM 2-meth ylbutyraldeh yde (2MBA) and 200 mM 2AI were added to a solution of 5 mM oligonu- cleotides (2–4 nucleotides in length) in 200 mM Na + - HEPES at pH 8 with 30 mM MgCl 2 . The reaction was al- lowed to sit for 6 hours, the optimal incubation time previ- ously determined ( 30 ). The mixture was then either brought to 10% (v / v) D 2 O for NMR spectroscopy, or separated from MeNC-mediated activation reagents using Sep-Pak ® C18 Cartridges for HPLC analysis. To perform the Sep- Pak ® clean up , the stationary phase of the cartridge was first wetted with acetonitrile and 2 M triethylamine acetate. The reaction sample was then diluted in 1 mL of UltraPure DNase / RNase-free distilled water and slowly loaded onto the cartridge three times. The cartridge was then washed three times with 3 ml of 20 mM TEAB (pH 8) before the oligonucleotides were slowly eluted with 2 ml of 40% ace- tonitrile in 20 mM TEAB (pH 8). The concentration of the total oligonucleotides was determined by a NanoDrop Mi- cr ovolume Spectr ometer. Acetonitrile was removed by leav- ing the eluate under ambient temperature for 1 hour. The sample was then lyophilized, redissolved in UltraPure wa- ter, and analyzed on an Agilent 1100 series HPLC with an Agilent ZORBAX Eclipse-XDB C18 column. The sample was separated using (A) aqueous 25 mM TEAB buffer (pH 8) and (B) acetonitrile, with 2% B for 5 min, then 2% to 13% B over 30 minutes, unless otherwise noted. All peak frac- tions were flash-frozen and ly ophilized bef ore confirming their identity by liquid chromato gra phy–mass spectrometry (LC–MS). Eutectic phase activation of mononucleotides and oligonu- c leotides . A mixture of 5 mM mononucleotide and indi- ca ted concentra tions oligonucleotides was allowed to re- act with stoichiometric 2AI ([2AI] = [pN]+[pN n ]), 30 mM Nucleic Acids Research, 2023, Vol. 51, No. 13 6531 MgCl 2 , 50 mM Na + -HEPES (pH 8), 200 mM 2MBA and 50 mM MeNC under eutectic ice phase conditions. Rapid cooling by liquid nitrogen was used to ensure complete ◦C . Ev- fr eezing befor e the solution was stor ed a t −15 to −13 ery 24 h, the sample was thawed, 50 mM additional MeNC was added, and refrozen to the eutectic ice phase. After 4 days, the products were separated from MeNC-mediated activation reagents and analyzed by analytical HPLC as de- scribed above. the template 5 Nonenzymatic primer extension and ligation with in-situ ac- tivated RNAs All reactions used the thiol-modified primer / 5ThioMC6- D / A GUGA GUAACUC. The primer extension reactions (cid:3) -OH- UGCGU GAGUUACUCA used (cid:3) -OH- CUAAA. Ligation reactions used the template 5 UGCG GAGUUACUCACUAAA. Underlined sequences indicate the template region available for substrate binding. All reactions contained 1 (cid:2)M primer, 1.5 (cid:2)M template, 50 mM Na + -HEPES (pH 8) and 30 mM MgCl 2 . Unacti- vated mononucleotides were supplied at 5 mM each, while unactivated pCGC was supplied at 0.5 mM, except as oth- erwise noted. 2AI was supplied at a stoichiometric concen- tration ([2AI] = [pN]+[pN n ]), except for the cases contain- ing only pA or pCGC, where excess 2AI was supplied at 5.5 mM. The primer and template were first mixed with Na + -HEPES (pH 8). Then stock solutions of unactivated monon ucleotides and / or oligon ucleotides, and 2AI were pr epar ed separately, adjusted to pH 8 with NaOH or HCl, and added to the primer-template mixture, followed by the addition of MgCl 2 . The r eaction mixtur e was brought to 200 mM 2MBA and 50 mM MeNC before being frozen to the eutectic phase ◦C. The freezer temperature a t approxima tely −15 to −13 was monitored, and slight tempera ture fluctua tions were observed occasionall y. Ra pid cooling by liquid nitro gen was used to ensure complete freezing before the sample was in- cuba ted a t the ice-eutectic phase. Every 24 h, the mixture was thawed to room temperature for sample collection, an additional 50 mM MeNC was added, and then subjected to eutectic freezing for one more day. Thawing, sample col- lection, and r efr eezing took less than ten minutes. After f our da ys of eutectic activation, the reaction mixture was brought to room temperature and allowed to react for an- other 24 h. The mixture was then incubated at ambient tem- perature for two more days with the addition of 100 mM 2MBA and 100 mM MeNC per day. Every 24 h, 30 (cid:2)l of the r eaction mixtur e was collected and purified by ZYMO Oligo Clean & Concentrator spin column (ZYMO Research, Irvine, CA). The isola ted ma te- rial was then mixed with 30 (cid:2)l 100 mM Na + -HEPES (pH 8), and any disulfides wer e r educed using a 10-fold molar excess of tris-(2-carboxyethyl) phosphine hydrochloride (TCEP) for 1.5–2 h. Then 0.8 (cid:2)l of 1 mM Alexa 488 C5 maleimide dye dissolved in anhydrous dimethyl sulfoxide was added to the reduced mixture, and the coupling reaction was allowed to proceed in the dark at ambient temperature for 1.5–2 h. The labeled primer-template duplex was separated from free dyes using ZYMO DNA Clean & Concentrator spin columns and concentrated to a volume of 10 (cid:2)l. 1 (cid:2)l of the 6532 Nucleic Acids Research, 2023, Vol. 51, No. 13 dye-labeled mixture was added to 3 (cid:2)l of gel loading buffer containing 8 M Urea, 10 × TBE, and 100 (cid:2)M of an RNA complementary to the template. The sample was heated to ◦C for 30 s and cooled to ambient temperature to ensure 95 the separation of the template from the dye-labeled primer. Template copying products were resolved by 20% (19:1) de- naturing PAGE, scanned with a Typhoon 9410 scanner, and quantified using the ImageQuant TL software. Rapid quenching of nonenzymatic primer extension with in- situ activated pA & pCGC Comparison between EDTA quenched and thawed reac- tions. Parallel experiments were set up at the same time with identical contents in 40- (cid:2)l scale: 5 mM pA, 0.5 mM pCGC, 5.5 mM 2AI, 1 (cid:2)M primer ( / 5ThioMC6- (cid:3) -OH- D / A GUGA GUAACUC), 1.5 (cid:2)M template (5 UGCGU GAGUUA CUCA CUAAA), 50 mM Na + - HEPES (pH 8) and 30 mM MgCl 2 . The reaction mixtures wer e tr eated with 200 mM 2MBA and 50 mM MeNC and then immersed in liquid nitrogen to ensure complete ◦C for fr eezing. Next, the samples wer e kept at −15 to −13 ice-eutectic phase reaction. After 24 h, one of the frozen 40- (cid:2)l samples was added to 10 (cid:2)l of 0.5 M EDTA and vor- texed to make sure all the thawed solution was immediately quenched by EDTA. Another frozen sample was allowed to thaw at ambient temperature. Then the two samples were purified by ZYMO Oligo Clean & Concentrator spin columns, r educed by T CEP and dye-labeled by Alexa 488 C5 maleimide dye, purified again, and resolved by 20% (19:1) denaturing PAGE gel following the same protocols as described earlier. Time course of the EDTA-quenched reaction. Parallel ex- periments were set up as described above in a 40- (cid:2)l scale for each. Each reaction was individually quenched by EDTA at the desired time point before thawing, and analyzed follow- ing the same procedure. Nonenzymatic primer extension with eutectic-phase- activated RNAs A mixture of 5 mM pA, 0.5 mM pCGC, 5.5 mM 2AI, 50 mM Na + -HEPES (pH 8), and 30 mM MgCl 2 was brought to 200 mM 2MBA and 50 mM MeNC. Next, the mixture was immersed in liquid nitrogen to ensur e complete fr eezing and then kept at −15 to ◦C for 24 h. The primer-template mixture was pre- −13 pared and lyophilized into dry po w der. The MeNC- activated mixture in the ice eutectic phase was thawed and immediately added to the primer-template po w der to gi v e a solution containing 1 (cid:2)M primer ( / 5ThioMC6- (cid:3) -OH- D / A GUGA GUAACUC) and 1.5 (cid:2)M template (5 UGCGU GAGUUA CUCA CUAAA). At each time point, 20 (cid:2)l of the reaction mixture was added to 10 (cid:2)l 0.5 M EDTA and purified by ZYMO Oligo Clean & Concen- trator spin columns. The thiolated RNAs were reduced with TCEP and labeled by incubating with Alexa 488 C5 maleimide dye, purified again, and resolved by a 20% (19:1) denaturing PAGE gel as described above. RESULTS Kinetic analysis of RNA copying with pre-synthesized monomer-bridged-oligonucleotides We started by asking whether monomer-bridged- oligonucleotides can significantly improve the efficiency of templa te-directed liga tion as well as nonenzymatic primer e xtension. We hav e pre viously reported kinetic parameters for se v eral nonenzymatic primer extension reactions using monomer-bridged-oligonucleotides ( 13 ). We have repeated these experiments, and compared the kinetics of primer extension with ligation under identical conditions (Figure 1 , Supplementary Figure S2). Our new measurements show almost identical Michaelis constants ( K M ) as proxies for binding affinity, compared to previously reported data. Thus, monomer-bridged-oligonucleotides bind to the primer-template complex much more strongly than activated mononucleotides or bridged dinucleotides. Furthermor e, the maximum r eaction rates ( k obs max ) for primer extension with pA are increased by over 80-fold when pA is bridged to a downstream mononucleotide pC, and by 500- to 700-fold when pA is bridged to an oligonucleotide (Figure 1 A). The enhanced reaction rate a t sa tura tion suggests a grea ter degr ee of pr e-organization into an optimal configuration for reactions ( 15 , 31 ). Gi v en the enhanced nonenzymatic primer extension observed with the monomer-bridged-oligonucleotides, we next asked whether the imidazolium bridging moiety, a str onger electr on withdrawing gr oup than imidazole, would also accelerate ligation in the same conditions (100 mM MgCl 2 , 200 mM Tris–HCl, pH 8) when the bridged mononucleotide component is not bound to the template (Figure 1 B). In this configuration, and under the same re- action conditions, we found that the maximum ligation rate of AppCGC was more than three-fold higher than that of pCGC (Figure 1 B). To further explore the effect of an unbound imidazolium-bridged nucleotide, we employed a sandwiched primer-template-blocker system with a single- nucleotide binding site on the template, and let the primer extend with either pN or NppN (Supplementary Figure S3A). We found that an imidazolium-bridged dinucleotide in w hich onl y one nucleotide is bound to the template ex- hibits a more than a fiv e-fold increase in the k obs max of +1 primer extension reaction (100 mM MgCl 2 , 200 mM Tris–HCl, pH 8). This rate enhancement applies to all four canonical nucleotides as the leaving group (Supplementary Figure S3). These observations suggest that the superior electrophilicity of the imidazolium bridging unit can en- hance both primer extension and liga tion. W hile decreas- ing the reaction pH to protonate the imidazole group of an activated oligonucleotide may also generate a positi v ely charged leaving group just like the imidazolium bridging moiety, deprotonation of the primer 3-OH (as shown in Supplementary Figure S1) would be more challenging at this lower pH, making the nucleophilic attack less favor- able ( 32 ). By employing the imidazolium bridging moiety as the more electrophilic leaving group, we were able to achie v e faster nonenzymatic ligation under the same con- ditions as used for primer extension, while using essen- tially the same reacti v e intermediates. Additionally, this ap- proach could potentially alleviate the inhibition of primer Nucleic Acids Research, 2023, Vol. 51, No. 13 6533 Figure 1. Kinetics of nonenzymatic primer extension and ligation with pure monomer-bridged-oligonucleotides. ( A ) Primer extension. (i) Mechanistic r epr esentation for nonenzymatic primer extension of one nucleotide and displacement of a 2AI-oligonucleotide as the leaving group. (ii) Schematic of nonenzymatic single nucleotide primer extension using a series of activated species. (iii) Michaelis–Menten analysis of primer extension with the illustrated activated species. The k obs values for AC and A are too low to display properly in the figure; see Supplementary Figure S2 for the full Michalis–Menten plot for each substrate. (iv) Kinetic parameters for +1 primer extension. Modified from Figure 7 in ( 14 ) with permission under a Creati v e Commons Attri- bution 4.0 International License. Copyright 2022 Ding et al. ; Published by Oxford University Press on behalf of Nucleic Acids Research. ( B ) Ligation. (i) Mechanistic r epr esenta tion for nonenzyma tic liga tion of an oligonucleotide and displacement of a 2AI-mononucleotide as the leaving group. (ii) Schematic of ligation with either an activated trinucleotide or a monomer-bridged-trinucleotide. (iii) Michaelis–Menten plots for ligation. (iv) Kinetic parameters for ligation. All r eactions wer e performed with 100 mM MgCl 2 and 200 mM Tris–HCl, pH 8 a t ambient tempera ture. Standard errors ( N ≥ 3) are reported at the appropriate significant digit in parentheses. extension caused by the binding of long RNAs down- (cid:3) -end of the blocking stream of a primer. By activating the 5 oligonucleotide with an imidazolium-bridged nucleotide, this blocking strand can take part in faster nonenzymatic ligation. Chemical RNA copying with a mixture of activated mono- and oligonucleotides The significantly improved RNA copying observed with monomer-bridged-oligonucleotide substrates prompted us to consider a more prebiotically plausible scenario in which activated mononucleotides were present together with acti- vated oligonucleotides. To investigate this, we performed ex- periments with a fixed concentration of 5 mM of pA, and lower concentrations of downstream oligonucleotides 2, 3 and 4 nucleotides in length. The pA and activated oligonu- cleotides react with each other in solution to form the monomer-bridged-oligonucleotide substrates for primer ex- tension. We find that as little as 50 (cid:2)M activated trin- ucleotides or tetranucleotides is sufficient to enhance the yield of primer extension, whereas the activated dimer had to be present at a higher concentration (of 5 mM) to achie v e a similar yield (Figure 2 A–C). Howe v er, the primer e xten- sion yield as a function of time exhibits a lag characteristic of a two-step reaction, indicating a requirement for the ini- tial formation of monomer-bridged-oligonucleotides before significant primer extension could occur. Since two sequen- tial reactions, bridge formation and primer extension, hap- pen in this scenario, it was difficult to quantitati v ely assess only the rate of the later primer extension reaction. In order to observe the primer extension reaction with- out the initial delay, we pre-incubated mixtures of activated mono- and oligonucleotides and measured the time 6534 Nucleic Acids Research, 2023, Vol. 51, No. 13 Figure 2. Comparison of primer extension with pre-synthesized monomer-bridged-oligonucleotide vs. mixtures of pre-activated mono- and oligonu- cleotides. ( A ) Comparison of primer extension with AppCG and pA + pCG. ( B ) Comparison of primer extension with AppCGC and pA + pCGC. ( C ) Comparison of primer extension with AppCGCA and pA + pCGCA. ( D ) Comparison between AppCGCA and a mixture of pA + pCGCA preincubated for 1 h. Negati v e controls with only 5 mM pA are shown in A-C. No detectable extension with pA was observed in the first hour. All r eactions wer e performed with 100 mM MgCl 2 and 200 mM Tris–HCl, pH 8. for spontaneous monomer-bridged- r equir ed oligonucleotide formation in solution. A mixture of 5 mM pA and 0.5 mM pCGCA was allowed to react in a primer extension buffer (100 mM MgCl 2 and 200 mM Tris–HCl, pH 8) at room temperature, and we measured the formation of AppCGCA at each time point by HPLC. We found that A ppCGCA spontaneousl y formed from the mixture of pA and pCGCA, with its concentration peaking at 1 h ( ∼53 (cid:2)M, 11% conversion from pCGCA to AppCGCA, Supplementary Figure S4). After 1 h, the concentration of AppCGCA started to decline, likely due to the hydrolysis. To examine primer extension, we pre- incubated a mixture of 5 mM pA and 0.5 mM pCGCA in a primer extension buffer for 1 h before adding that to the primer-template complex. We were pleased to find that primer extension proceeded with no lag, and at a rate very similar to that observed with pre-synthesized AppCGCA (Figure 2 D). Our results suggest that activating e v en a small fraction of the oligonucleotides in a complex mixture may be sufficient to enable efficient nonenzymatic template copying. MeNC-mediated activation and bridge-formation of monomers and oligonucleotides The ef ficient nonenzyma tic RNA copying facilita ted by modest concentrations of activated oligonucleotides prompted us to investigate the activation of oligonu- cleotides under potentially prebiotic conditions. We started by demonstra ting tha t short oligonucleotides can be Nucleic Acids Research, 2023, Vol. 51, No. 13 6535 Figure 3. Ice-eutectic phase MeNC-mediated activation of oligonucleotides (N n ) and subsequent formation of monomer-bridged-oligonucleotides (NN n ). ( A ) The phospho-P asserini r eaction pa thway activa tes mono- and oligonucleotides and promotes the formation of bridged intermediates. ( B ) A scheme illustrating the activation of mono- and oligonucleotides and the formation of bridged species (including NN, NN n and N n N n ) in the ice- eutectic phase. ( C ) A r epr esentati v e HPLC trace of the purified end pr oducts fr om a reaction containing unactivated mono- and dinucleotides (5 mM pA and 2 mM pCG). ( D ) The concentration of each identified peak species from a reaction mixture of mono- and oligonucleotides of indicated length and con- centration (i–iii) underwent eutectic phase activation, in contrast to the spontaneous formation of bridged species from a mixture of (iv) pure pre-activated mono- and tetranucleotides. Only a subset of the species is shown her e for br evity (i–iii), but additional species can be found in Supplementary Table S1. All activation reactions contained 30 mM MgCl 2 , 50 mM Na + -HEPES (pH 8) and stoichiometric 2AI ([2AI] = [pN]+[pN n ]). 200 mM 2MBA and 50 mM MeNC were supplied to initiate the reaction, with subsequent periodic additions of MeNC in three aliquots of 50 mM at the beginning of each day of the freeze-thaw cycle. The corresponding elution profiles are shown in Supplementary Figure S7. The control C (iv) was performed by incubating 5 mM A, 0.5 mM CGCA, 30 mM MgCl 2 , and 50 mM Na + -HEPES (pH 8) for 1 h at room temperature. activated using MeNC, 2MBA and excess 2AI in the same buffer used for nonenzymatic RNA copying (50 mM Na + -HEPES pH 8 and 30 mM MgCl 2 ). When the MgCl 2 concentration was reduced to suppress hydrolysis and side product formation ( 30 ), we found that short oligonucleotides of various lengths can be activated with good efficiency (Supplementary Figure S5). For example, a pproximatel y 89% of dimers (pCA) and 77% of trimers (pCCA) were activated to 2-aminoimidazolides under these conditions. In addition to the ef ficient activa tion of oligonucleotides, our data suggest that the effecti v e concen- trations of reagents required for activation are remar kab ly similar for mononucleotides and oligonucleotides up to tetranucleotides. Gi v en the efficient solution phase activation of oligonu- cleotides with excess 2AI, we ne xt e xplored whether 2AI- bridged mono- and oligonucleotides could form with sto- ichiometric 2AI using MeNC-mediated chemistry (Fig- ure 3 ). To do this, we employed eutectic ice phase con- ditions, which we have previously shown to facilitate the ef ficient activa tion of mononucleotides and the forma tion of bridged dinucleotides. We began by activating 5 mM pA and 2 mM pCG under the eutectic ice phase ( −15 ◦C) with 7 mM 2AI. After four days of activa- to −13 tion, > 60% of the pA and pCG were converted into either activated or bridged species (i.e. bridged dinucleotides or monomer-bridged-dinucleotides) as determined by HPLC analysis (Figure 3 C and Di). Actual le v els of acti vation and bridge formation in the reaction mixture may be higher because of hydrolysis during post-reaction cleanup. Side (cid:3) linked dinucleotide, were min- products, including the 3 imal, as confirmed by the spike-in of their correspond- ing authentic standards (Supplementary Figure S6). We then proceeded to activate longer oligonucleotides at de- creasing concentrations because of their possible lower prebiotic abundance ( 18–23 ) and their ability to enhance (cid:3) –5 6536 Nucleic Acids Research, 2023, Vol. 51, No. 13 nonenzymatic copying at reduced concentrations. We found that 5 mM mononucleotides in combination with 1 mM trinucleotides or 0.5 mM tetranucleotides can also be nearly completely converted to either activated or bridged species (Figure 3 D, Supplementary Table S1). We compared these results with a control experiment where we observed the spontaneous formation of AppCGCA from a mixture of pre-activated pA and pCGCA in the same primer exten- sion buffer for 1 h (this is the optimal incubation period as shown in Supplementary Figure S4). Despite starting with pure activated mono- and oligonucleotides, this control ex- periment produced less monomer-bridged-oligonucleotides than seen in the presence of MeNC activation chemistry (Figure 3 Diii–iv). We then looked into whether a mixture of mononu- cleotides and short oligonucleotides of different lengths could be activated sim ultaneousl y and form bridged species. A mixture containing mono-, di-, and trinucleotides was activa ted ef ficiently with a substantial yield of bridged species (Supplementary Figure S7). We monitored the re- action pr ogress thr oughout the f our da ys of activation and observed that the overall activation yields gradually in- creased while the monomer-bridged-oligonucleotide yield peaked on day two (Supplementary Figure S7). It is en- couraging that we were able to obtain mostly activated and bridged species from a heterogeneous mixture of mono- and oligonucleotides under a prebiotically plausible activation condition that is compatible with chemical RNA copying. In-situ activation of mononucleotides and oligonucleotides for nonenzymatic primer extension and ligation Following the successful formation of monomer-bridged- oligonucleotides under the eutectic ice phase, we sought to a ppl y this process to nonenzymatic RNA copying. We hav e pre viousl y demonstrated RN A copying dri v en by in- situ activation of mononucleotides, but we have not com- pared the process when both mono- and oligonucleotides ar e pr esent, and when both primer extension and ligation can occur. In this study, we used a model system containing 5 mM pA and 0.5 mM pCGC to measure in-situ activation, primer extension, and ligation. To ensure an accurate com- parison, we included control experiments using only the un- activated mononucleotide or only the unactivated oligonu- cleotide, thus pre v enting the formation of the monomer- bridged-oligonucleotide following activation. After the ad- dition of MeNC and 2MBA, the samples were incubated ◦C) and thawed ev- under the eutectic phase ( −15 to −13 ery 24 hours for aliquot removal and the addition of fresh MeNC. After four days of eutectic activation, the samples were brought to room temperature for three more days of further extension. Additional MeNC and 2MBA were sup- plied in the last two days at room temperature to facilitate bridge forma tion, but ef ficient activa tion of nucleotides can- not occur at the low 2AI concentration we used ( 29 , 30 ). With the in-situ activation of pA and pCGC in the eu- tectic phase, we observed about a 70% yield of +1 primer extension within 1 day (Figure 4 A). To determine whether the primer extension occurred during the eutectic ice phase or during the short thawing intervals, we quenched the 1- day reaction sample by adding EDTA before thawing, so that the Mg 2+ would be chelated immediately upon thaw- ing (Supplementary Figure S8). We were surprised to find nearly identical extension yields with and without EDTA quenching prior to thawing. This result suggests that, de- spite the low temperature, the majority of the observed primer extension occurred in the ice eutectic phase. The con- siderab le e xtent of primer e xtension re v eals the efficient ac- tivation of mono- and oligonucleotides, as well as the high reactivity of the monomer-bridged-oligonucleotide even at low temperatures. Since almost complete +1 primer extension was observed in one day with in-situ activated pA and pCGC, we fur- ther analyzed the extent of reaction throughout the first day by EDTA-quenching the partial frozen sample at different time points. We found that significant primer extension did not occur until after 6–8 h of eutectic phase activation (Sup- plementary Figure S9). We interpret this result as indicat- ing the time r equir ed for nucleotide activation and buildup of the monomer-bridged-oligonucleotide, before primer ex- tension can happen. In the case of a mixture of the mononucleotides pA and pC, and in the absence of oligonucleotides, the primer exten- sion proceeded with sigmoidal behavior during the 4 days of eutectic activation, i.e. the reaction pro gressed slowl y ini- tially and accelerated at later times. This is likely because the concentration of AppC peaked during the second day, con- sistent with our previous findings ( 29 ). How ever, w e w ere surprised to see that primer extension almost completely halted during the subsequent incubation at ambient temper- atur e (Figur e 4Aii). We specula te tha t the substantial primer e xtension observ ed during the eutectic phase was due to the continuous generation of AppC and the improved bind- ing of AppC to the template at low temperature. These benefits were lost once the reaction was brought to ambi- ent temperature, leading to minimal subsequent primer ex- tension. Similarly, primer extension when only pA was acti- vated in-situ was very poor because the activation products, pA and AppA, could not bind well to the primer-template complex. These findings provide further support for the im- portance of utilizing oligonucleotides to catalyze template copying. We also examined the ligation of in-situ activated pCGC to the same primer, using a different template. In con- trast to the efficient +1 primer extension observed during the eutectic ice phase, the majority of this ligation did not take place until the room tempera ture incuba tion (Figure 4 B). This difference is most likely explained by the low K M and low k obs max of the ligation reaction (Figure 1 ). Thus, the activated oligonucleotide can bind strongly to the template with multiple Watson-Crick base pairs, but the rate of the ligation reaction is intrinsically slow. As a result, ligation does not require the high effecti v e concen- tration or better substrate binding attained under the ice eutectic phase. Instead, the low temperature of the eutec- tic phase suppresses the reaction. Howe v er, when the reac- tion is brought to room tempera ture, the liga tion reaction can proceed slowly over 2–3 days. A greater extent of liga- tion was observed in conditions that favor the formation of monomer-bridged-oligonucleotides, likely due to the faster liga tion ra te with the monomer-bridged-oligonucleotide (Figure 1 B). Nucleic Acids Research, 2023, Vol. 51, No. 13 6537 Figure 4. Ice-eutectic phase enables in-situ activation of mono- and oligonucleotides for enhanced nonenzymatic primer extension and ligation. ( A ) Primer extension with in-situ activated mono- and / or oligonucleotides. (i) Schematic r epr esentation of +1 primer extension with in-situ activated AppCGC. (ii) Extension yield with: 5 mM pA (black), 5 mM pA & 5 mM pC (green), and 5 mM pA & 0.5 mM pCGC (pink). ( B ) Liga tion with either in-situ activa ted oligonucleotides or a mixture of in-situ activated mono- and oligonucleotides. (i) Schematic r epr esentation of +3 ligation with in-situ activated AppCGC. (ii) Ligation yield with: 0.5 mM pCGC (black) or 5 mM pA & 0.5 mM pCGC (pink). All reactions were initiated with 1 (cid:2)M primer, 1.5 (cid:2)M template, 5.5 mM 2AI, 50 mM Na + -HEPES (pH 8) and 30 mM MgCl 2 . 200 mM 2MBA was added at the beginning of the experiment, while 50 mM MeNC was freshly supplied e v ery da y. After f our da ys of eutectic activation, the r eactions wer e brought to ambient temperatur e for 24 h, then fr esh 100 mM 2MBA and 100 mM MeNC were added e v ery day for two more days. Interestingly, in a typical nonenzymatic +1 primer exten- sion reaction with in-situ activated pA and pCGC, the liga- tion product was also seen after the +A extension (Supple- mentary Figure S10). This primer was first extended by +A using AppCGC under the eutectic phase and then further extended by +CGC ligation at ambient temperature, likely using AppCGC as well. These observations suggest that both primer extension and ligation may occur in an envi- ronment with freeze-thaw cycles followed by warm periods. For example, in-situ activation and primer extension could occur in the eutectic phase, followed by ligation after the temperature rises. Since a prebiotically plausible reaction mixture would likely contain multiple short oligonucleotides, we also in- vestigated in-situ activation and template copying with a mixture of pA, pA CG , and pCGC. The pCGC can form AppCGC to facilitate +1 primer extension, as previously demonstrated, while the pACG can be activated to pACG and AppA CG , which are substrates for a competing lig- ation reaction. We found that the primer extension reac- tion was only slightly slower when the competing pACG was present (Supplementary Figure S11). Almost no liga- tion products of pACG were observ ed, probab ly because most of the primer had already been ra pidl y extended by one nucleotide before +ACG ligation could happen. To sim- ulate prebiotic conditions in which oligonucleotides were present at lower concentrations, we conducted competition experiments with decreasing concentrations of the two trin- ucleotides. We were surprised to find that template copy- ing with only 50 (cid:2)M of each trinucleotide behaved slightly better than the case with 500 (cid:2)M of pACG and pCGC, possibly due to less inhibition from pA CG , pA CG and AppACG a t lower concentra tions and continuous gener- ation of the highly reacti v e AppCGC. Remar kab ly, sub- stantial template copying was still observed even with trin- ucleotide concentrations as low as 5 (cid:2)M each, indicating the strong ca talytic ef fect of short oligonucleotides at minimal concentrations. We have considered a prebiotic hot springs environment in which freeze-thaw cycles might be common. In such a sce- nario , monon ucleotides and short oligonucleotides would be activated in the ice-eutectic phase, then periodically re- leased by melting to flow to another site in which template copying could occur. To mimic this scenario, we first acti- vated a mixture of mono- and oligonucleotides in the ice- eutectic phase as described above, but without the primer and the template. After one day of activation, the mixture was thawed and added to the primer-template complex. We were pleased to observe ∼80% +1 primer extension prod- ucts in just 1 h (Figure 5 ). This nonenzymatic primer ex- tension was performed at a reduced MgCl 2 concentration of 30 mM (compared to 100 mM MgCl 2 in Figures 1 and 2 ) to minimize hydrolysis, yet good template copying was still observed. Our findings suggest that in-situ activation chemistry and short oligonucleotides may have played an essential role in enhancing prebiotic nonenzymatic template copying. DISCUSSION Any prebiotically plausible scenario for the nonenzymatic replication of RN A m ust include an efficient pathwa y f or nucleotide activation, both initially and to counteract the ine vitab le hydrolysis of activated species. In addition, mini- mizing the complexity of the geochemical context is impor- tant, for example, by showing that multiple steps in a path- way can occur together in a single environment. We have no w sho wn that freeze-thaw cy cles can dri v e efficient acti va- tion of both mono- and oligonucleotides through the methyl isocyanide pathway. This efficient activation leads to the spontaneous synthesis of monomers that are imidazolium- bridged to short oligonucleotides, which in turn leads to ef- ficient template-directed primer e xtension. Early wor k on nonenzyma tic templa te cop ying r equir ed very high concen- tra tions of activa ted monomers, typically in the range of 100 mM ( 9 ). In contrast, our work has shown that low mM 6538 Nucleic Acids Research, 2023, Vol. 51, No. 13 Figure 5. Nonenzymatic primer extension with pA and pCGC pre-activated in an ice-eutectic phase for 1 day. ( A ) A scheme illustrating the activation of pA and pCGC by MeNC-mediated chemistry in the ice-eutectic phase and the flow of materials to a new environment containing the primer-template complex upon thawing. ( B ) Primer extension with MeNC-activated mixture demonstrating ∼ 80% yield in 1 hour. (i) A r epr esentati v e PAGE gel analysis. (ii) Primer extension yield vs. time. The eutectic phase activation was performed for 1 day with 5 mM pA, 0.5 mM pCGC, 5.5 mM 2AI, 50 mM Na + -HEPES (pH 8), 30 mM MgCl 2 , 200 mM 2MBA and 50 mM MeNC. After thawing, the mixture was added to the lyophilized primer-template complex to gi v e a final r eaction mixtur e containing 1 (cid:2)M primer and 1.5 (cid:2)M template. concentrations of monomers together with micromolar concentrations of short oligonucleotides can lead to effi- cient copying by primer extension, which seems more likely in a prebiotic scenario. Below we discuss the implications of our results for the synthesis and catalytic roles of monomer- bridged-oligonucleotides in RNA replication, and we con- sider geochemical environments that could potentially host such RN A-catal yzed RN A replication processes. Short oligonucleotides may have been generated on the primordial Earth via untemplated polymerization, e.g. on clay mineral surfaces or as a result of concentration in the ice eutectic phase of partially frozen water ( 18–20 ). Short oligonucleotides can also assemble by the template-directed oligomeriza tion of activa ted monomers or bridged dinu- cleotides ( 21–23 , 33 ). Howe v er, in the conte xt of nonenzy- matic copying, oligonucleotides are typically thought of as potential primers , templates , or ligators , and their potential roles as catalysts have not yet been fully explored. We have previously shown that monomer-bridged-oligonucleotide intermediates can significantly improve the rate of nonen- zymatic primer extension ( 13 ). These substrates can spon- taneously form in mixtures of activated mono- and oligonu- cleotides. We have now shown that isocyanide chemistry can also activate oligonucleotides of varying lengths, and thereby yield monomer-bridged-oligonucleotides under ice- eutectic conditions. Following the in-situ activation of monomers and oligonucleotides, we observed significant primer extension while in the ice-eutectic phase; we also observ ed considerab le ligation but only following thawing. Furthermore, e v en at e xtremely low concentrations of short oligonucleotides (as low as 5 (cid:2)M), nonenzymatic template copying dri v en by in-situ activation can be observed. Fi- nally, we also demonstrated that short oligonucleotides can be activated in one environment and then transferred into a second environment where fast template copying can occur. Thus, short oligonucleotides may have played an important role in the catalysis of template copying on the primordial Earth. We are currently investigating the activation of com- plex mixtures of short oligonucleotides as a potential means of copying arbitrary RNA sequences. We have recently proposed and provided the first exper- imental tests of a model for nonenzymatic RNA replica- tion ( 33 , 34 ). This virtual circular genome model invokes a collection of linear oligonucleotides of varying lengths that map onto both strands of a circular genomic sequence. In this model, all of the component oligonucleotides can play multiple roles, i.e. as primers, templates, and down- stream helpers. Remar kab ly, acti vating all of the oligonu- cleotide components in this system significantly enhanced the observed rate of primer extension ( 34 ), consistent with a role for short, activated oligonucleotides in the catalysis of RNA copying chemistry. The combination of the pre- biotically plausib le acti vation chemistry presented in our current work with the virtual circular genome replication model has the potential to dri v e a nonenzyma tic replica tion system through continuous in-situ activation. It is interesting to consider whether the ice-eutectic me- diated synthesis of bridged substrates could support RNA r eplication within protocells. Fr eeze-thaw cycles appear to be incompatible with vesicle integrity since ice crystals are thought to rupture membrane structures, resulting in the loss of compartmentalized content. We suggest that acti- vation chemistry could happen in a nearby and partially fr ozen envir onment, where periodic thawing allows acti- vated nucleotides to be released and flow into a separate environment in which protocells are replicating. Although monomer-bridged-oligonucleotide substrates are too large to readily diffuse across membranes and into protocells, we have recently described a phenomenon akin to spontaneous endocytosis in model protocells ( 35 ). Substrates internal- ized by endocytosis could then slowly diffuse into the proto- cell ‘cytoplasm’ on a longer time scale. An alternati v e pos- sibility, which we are currently investigating, is that a mod- ification of the isocy anide activ ation chemistry might obvi- ate the need for freeze-thaw cycles, and allow for nucleotide and oligonucleotide activation to proceed within replicating protocells. DA T A A V AILABILITY The data underlying this article are available in the article and in its online supplementary material. SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We thank Prof. Lijun Zhou and Dr Saurja DasGupta for helpful discussions and comments on the manuscript. FUNDING J.W.S. is an investigator of the Howard Hughes Medical In- stitute; Simons Foundation [290363, in part]; National Sci- ence Foundation [2104708 to J.W.S., in part]. Funding for open access charge: Internal funding. Conflict of interest statement. None declared. REFERENCES 1. Joyce,G.F. (1989) RNA evolution and the origins of life. Nature , 338 , 217–224. 2. Orgel,L.E. (2004) Prebiotic chemistry and the origin of the RNA world. Crit. Rev. Biochem. Mol. Biol. , 39 , 99–123. 3. Joyce,G.F. and Szostak,J.W. (2018) Protocells and RNA self-replication. Cold Spring Harb. Perspect. Biol. , 10 , a034801. 4. Wachowius,F., Attwater,J. and Holliger,P. (2017) Nucleic acids: function and potential for abiogenesis. Q. Rev. Biophys. , 50 , e4. 5. Becker,S., Feldmann,J., Wiedemann,S., Okamura,H., Schneider,C., Iwan,K., Crisp,A., Rossa,M., Amatov,T. and Carell,T. (2019) Unified prebiotically plausible synthesis of pyrimidine and purine RNA ribonucleotides. Science , 366 , 76–82. 6. P atel,B.H., Per civalle,C., Ritson,D.J., Duffy,C.D . and Sutherland,J.D . (2015) Common origins of RNA, protein and lipid precursors in a cyanosulfidic protometabolism. Nat. Chem. , 7 , 301–307. 7. Weimann,B.J., Lohrmann,R., Orgel,L.E., Schneider-Bernloehr,H. and Sulston,J.E. (1968) Template-directed synthesis with adenosine-5 (cid:3) -phosphorimidazolide. Science , 161 , 387. 8. Sulston,J., Lohrmann,R., Orgel,L.E. and Miles,H.T. (1968) Nonenzymatic synthesis of oligoadenylates on a polyuridylic acid templa te. Pr oc. Natl. Acad. Sci. U.S.A. , 59 , 726–733. 9. Rembold,H. and Orgel,L.E. (1994) Single-strand regions of poly (G) act as templates for oligo (C) synthesis. J. Mol. Evol. , 38 , 205–210. 10. Walton,T. and Szostak,J.W. (2016) A highly reacti v e imidazolium-bridged dinucleotide intermediate in nonenzymatic RNA primer extension. J. Am. Chem. Soc. , 138 , 11996–12002. 11. Walton,T. and Szostak,J.W. (2017) A kinetic model of nonenzymatic RN A pol ymerization by cytidine-5 (cid:3) -phosphoro-2-aminoimidazolide. Bioc hemistr y , 56 , 5739–5747. 12. Walton,T., Zhang,W., Li,L., Tam,C.P. and Szostak,J.W. (2019) The mechanism of nonenzymatic template copying with imidazole-activated nucleotides. Angew. Chem. Int. Ed. , 58 , 10812–10819. 13. Ding,D., Zhou,L., Giurgiu,C. and Szostak,J.W. (2022) Kinetic explanations for the sequence biases observed in the nonenzymatic copying of RNA templates. Nucleic Acids Res. , 50 , 35–45. Nucleic Acids Research, 2023, Vol. 51, No. 13 6539 14. Zhang,W., Tam,C.P., Walton,T., Fahrenbach,A.C., Birrane,G. and Szostak,J.W. (2017) Insight into the mechanism of nonenzymatic RNA primer extension from the structure of an RNA-GpppG complex. Proc. Natl. Acad. Sci. U.S.A. , 114 , 7659–7664. 15. Zhang,W., Walton,T., Li,L. and Szostak,J.W. (2018) Crystallo gra phic observa tion of nonenzyma tic RNA primer extension. Elife , 7 , e36422. 16. Duzdevich,D., Carr,C.E., Ding,D., Zhang,S.J., Walton,T.S. and Szostak,J.W. (2021) Competition between bridged dinucleotides and activated mononucleotides determines the error frequency of nonenzymatic RNA primer extension. Nucleic Acids Res. , 49 , 3681–3691. 17. Prywes,N., Blain,J.C., del Frate,F. and Szostak,J.W. (2016) Nonenzymatic copying of RNA templates containing all four letters is catalyzed by activated oligonucleotides. Elife , 5 , e17756 . 18. Ferris,J.P. and Ertem,G. (1993) Montmorillonite catalysis of RNA oligomer formation in aqueous solution. A model for the prebiotic formation of RNA. J. Am. Chem. Soc. , 115 , 12270–12275. 19. K anavarioti,A., Monnar d,P.A. and Deamer,D.W. (2001) Eutectic phases in ice facilitate nonenzymatic nucleic acid synthesis. Astrobiology , 1 , 271–281. 20. Monnar d,P.A., K anavarioti,A. and Deamer,D.W. (2003) Eutectic phase polymerization of activated ribonucleotide mixtures yields quasi-equimolar incorporation of purine and pyrimidine nucleobases. J. Am. Chem. Soc. , 125 , 13734–13740. 21. Lohrmann,R., Bridson,P.K. and Orgel,L.E. (1980) Efficient metal-ion catalyzed template-directed oligonucleotide synthesis. Science , 208 , 1464–1465. 22. Inoue,T. and Orgel,L.E. (1982) Oligomerization of (guanosine 5 (cid:3) -phosphor)-2-methylimidazolide on poly(C). An RNA polymerase model. J. Mol. Biol. , 162 , 201–217. 23. Trinks,H., Schr ¨oder,W. and Biebricher,C.K. (2005) Ice and the origin of life. Orig. Life Evol. Biosphys. , 35 , 429–445. 24. Rohatgi,R., Bartel,D.P. and Szostak,J.W. (1996) Kinetic and mechanistic analysis of nonenzymatic, template-directed oligoribonucleotide ligation. J. Am. Chem. Soc. , 118 , 3332–3339. 25. Zhou,L., O’Flaherty,D.K. and Szostak,J.W. (2020) Template-directed copying of RNA by non-enzymatic ligation. Angew. Chem. , 132 , 15812–15817. 26. Zhou,L., O’Flaherty,D.K. and Szostak,J.W. (2020) Assembly of a ribozyme ligase from short oligomers by nonenzymatic ligation. J. Am. Chem. Soc. , 142 , 15961–15965. 27. Wachowius,F. and Holliger,P. (2019) Non-enzymatic assembly of a minimized RN A pol ymerase ribozyme. ChemSystemsChem , 1 , 12–15. 28. Mariani,A., Russell,D.A., Javelle,T. and Sutherland,J.D. (2018) A light-releasable potentially prebiotic nucleotide activating agent. J. Am. Chem. Soc. , 140 , 8657–8661. 29. Zhang,S.J., Duzdevich,D., Ding,D. and Szostak,J.W. (2022) Freeze-thaw cycles enable a prebiotically plausible and continuous pathway from nucleotide activation to nonenzymatic RNA copying. Proc. Natl. Acad. Sci. U.S.A. , 119 , e2116429119. 30. Zhang,S.J., Duzdevich,D. and Szostak,J.W. (2020) Potentially prebiotic activation chemistry compatible with nonenzymatic RNA copying. J. Am. Chem. Soc. , 142 , 14810–14813. 31. Zhang,W., Tam,C.P., Zhou,L., Oh,S .S ., Wang,J. and Szostak,J.W. (2018) Structural rationale for the enhanced catalysis of nonenzymatic RNA primer extension by a downstream oligonucleotide. J. Am. Chem. Soc. , 140 , 2829–2840. 32. Giurgiu,C., Fang,Z., Aitken,H.R., Kim,S.C., Pazienza,L., Mittal,S. and Szostak,J.W. (2021) Structure–Activity relationships in nonenzyma tic templa te-directed RNA synthesis. Ang ew. Chem. Int. Ed. Engl. , 60 , 22925–22932. 33. Zhou,L., Ding,D. and Szostak,J.W. (2021) The virtual circular genome model for primordial RNA replication. RNA , 27 , 1–11. 34. Ding,D., Zhou,L., Mittal,S. and Szostak,J.W. (2023) Experimental tests of the virtual circular genome model for non-enzymatic RNA replication. J. Am. Chem. Soc. , 145 , 7504–7515 35. Zhang,S.J., Lowe,L.A., Anees,P., Krishnan,Y., Fai,T.G., Szostak,J.W. and Wang,A. (2023) Passi v e endocytosis in model protocells. bioRxiv doi: https://doi.org/10.1101/2023.01.07.522792 , 04 May 2023, preprint: not peer re vie wed. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
10.1126_science.adc9498	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 June 16. Published in final edited form as: Science. 2023 April 07; 380(6640): eadc9498. doi:10.1126/science.adc9498. Germline-encoded amino acid-binding motifs drive immunodominant public antibody responses Ellen L. Shrock1,2,3, Richard T. Timms4,†, Tomasz Kula1,2,3,5,†, Elijah L. Mena1,2, Anthony P. West Jr.6, Rui Guo7,8,9, I-Hsiu Lee10, Alexander A. Cohen6, Lindsay G. A. McKay11, Caihong Bi12,13, Keerti12,13, Yumei Leng1,2, Eric Fujimura1,2, Felix Horns14, Mamie Li1,2, Duane R. Wesemann9,12,13,15, Anthony Griffiths11, Benjamin E. Gewurz7,8,9,16, Pamela J. Bjorkman6, Stephen J. Elledge1,2,* 1Department of Genetics, Harvard Medical School, Boston, MA 02115, USA 2Division of Genetics, Department of Medicine, Howard Hughes Medical Institute, Brigham and Women’s Hospital, Boston, MA 02115, USA 3Program in Biological and Biomedical Sciences, Harvard University, Boston, MA 02115, USA 4Cambridge Institute of Therapeutic Immunology and Infectious Disease, Jeffrey Cheah Biomedical Centre, Cambridge Biomedical Campus, University of Cambridge, Cambridge, UK 5Present address: Society of Fellows, Harvard University, Cambridge, MA 02138, USA 6Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA 91125, USA 7Division of Infectious Disease, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA 8Department of Microbiology, Harvard Medical School, Boston, MA 02115, USA 9Broad Institute of Harvard and MIT, Cambridge, MA, 02142, USA 10Center for Systems Biology, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114, USA 11National Emerging Infectious Diseases Laboratories, Boston University School of Medicine, Boston University, Boston, MA 02118, USA 12Division of Allergy and Immunology, Division of Genetics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding author. [email protected]. †These authors contributed equally to this work. Author contributions: Conceptualization: S.J.E., E.L.S., and T.K. Formal analysis: E.L.S., R.T.T., T.K., A.P.W., I.L., A.A.C., L.G.A.M., C.B., and K. Investigation: E.L.S., T.K., R.G., A.A.C., L.G.A.M., C.B., K., Y.L., and M.L. Methodology: E.L.S., R.T.T., T.K., E.L.M., and A.P.W. Resources: E.F. and F.H. Supervision: S.J.E., P.J.B., B.E.G., A.G., and D.R.W. Visualization: E.L.S., T.K., E.L.M., I.L., A.A.C., R.G., A.P.W., L.G.A.M., C.B., and K. Writing – original draft: E.L.S. and S.J.E. Writing – review and editing: E.L.S., S.J.E., R.T.T., T.K., E.L.M., and P.J.B. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 2 13Massachusetts Consortium on Pathogen Readiness, Boston, MA 02115, USA 14Department of Bioengineering, Department of Applied Physics, Chan Zuckerberg Biohub and Stanford University, Stanford, CA 94305, USA 15Ragon Institute of MGH, MIT, and Harvard, Cambridge, MA, 02139 USA 16Graduate Program in Virology, Division of Medical Sciences, Harvard Medical School, Boston, MA 02115, USA Abstract Despite the vast diversity of the antibody repertoire, infected individuals often mount antibody responses to precisely the same epitopes within antigens. The immunological mechanisms underpinning this phenomenon remain unknown. Here, by mapping 376 immunodominant “public epitopes” at high resolution and characterizing several of their cognate antibodies, we conclude that germline-encoded sequences in antibodies drive recurrent recognition. Systematic analysis of antibody–antigen structures uncovered 18 human and 21 partially overlapping mouse germline-encoded amino acid-binding (GRAB) motifs within heavy and light V gene segments which, in case studies, are critical for public epitope recognition. GRAB motifs represent a fundamental component of the immune system’s architecture that ensures antibody recognition of pathogens and promotes species-specific reproducible responses that can exert selective pressure on pathogens. One-sentence summary: Antibody genes harbor GRAB motifs that bind specific amino acids and drive reproducible responses to certain immunodominant epitopes across individuals. Introduction The adaptive immune system relies on an extremely diverse antibody repertoire to mount a response to any pathogen encountered. Antibody diversity is generated by a DNA recombination mechanism occurring in the heavy and light chain genes in which modular VDJ (for heavy) and VJ (for light) gene segments are combinatorially assembled to generate a vast repertoire of variable domain sequences. Immunoglobulin G (IgG) is composed of two heavy and two light chains arranged as a heterodimer with two identical antigen-binding sites, each formed by paired heavy and light chain variable domains. A given antibody has either a kappa or a lambda light chain, which have no known functional difference. Antigen recognition is accomplished primarily by complementarity determining regions (CDRs), which are hypervariable loops within the heavy and light chain variable domains (three in each domain). The heavy and light chain CDR1s and CDR2s are encoded by the V gene segments, whereas the CDR3s span the junctions of the recombined VDJ or VJ gene segments and are thus highly diverse and generally thought to play a dominant role in determining specificity (3.1). The complexity of the antibody repertoire enables the generation of antibodies to virtually any antigen, yet isolated examples of recurrent responses in different individuals to Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 3 particular epitopes have been reported (3.2–3.13). Given the challenge of mapping antibody epitopes at high resolution, it has been unclear how common recurrent antibody responses are and how widely they are shared across human populations. Recently, we developed VirScan, a phage display platform programmed to display peptides spanning the human virome, which enabled the high-throughput identification of antiviral antibody epitopes (3.14–3.18). We used VirScan to profile hundreds of human serum samples (3.14) and found that although many viral peptides recognized by an individual were relatively specific to that person, many other viral peptides—which we termed “public epitopes”—were recognized by a substantial percentage (≤98%) of individuals seropositive for the given virus (3.14). Public epitopes were also observed in VirScan studies of antibody responses to allergens and symbiotic microbiota (3.19–3.21). These findings raised a fundamental question: what mechanisms drive recurrent responses to public epitopes? Results Public epitopes are a general feature of the human antibody response To identify a collection of publicly recognized viral peptides from a VirScan analysis of 569 human sera samples (3.14), we chose the 5 most commonly recognized peptides from all viruses for which there were at least 5 seropositive individuals. This yielded a list of 363 viral peptides, 199 of which were recognized by at least 30% of seropositive individuals (Fig. 3.1A and table S3.1). These peptides were derived from 62 viral species spanning a broad range of viral classes and encompassed both structural and nonstructural proteins. Antibody responses to publicly recognized peptides appeared unrelated to donor age or geographic location and thus appeared to be a general feature of the human antibody response. The publicly recognized viral peptides could harbor either a single epitope recognized by many different individuals or multiple epitopes. To distinguish between these possibilities and map individual epitopes more precisely, we designed an additional VirScan library containing tiled truncations and triple alanine-scan mutations of the 363 publicly recognized 56–amino acid (AA) viral peptides (Fig. 3.1B and table S3.2). We profiled the serum antibody responses of ~70 diverse donors with a wide range of viral exposures with this library and observed that the positions of the epitopes recognized by different individuals within these peptides were often identical (fig. S3.1, A and B). Antibodies recognizing public epitopes have biased light chain isotype usage Next, we examined whether antibodies from different individuals that recognized the same public epitope were structurally similar. We adapted the VirScan assay to separately immunoprecipitated (IP) antibodies with kappa versus lambda light chains (“kappa antibodies” and “lambda antibodies,” respectively) (Fig. 3.1B). If different individuals made structurally similar antibodies against a given public epitope, we expected to detect responses to the epitope mainly in kappa or in lambda IP fractions, but if they made structurally diverse antibodies, we would expect no light chain bias. We reprofiled the ~70 human serum samples with the public epitope truncation and alanine-scan library using the kappa- and lambda-specific IP protocol and found that antibody responses to public epitopes Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 4 were strongly biased in light chain isotype usage. For example, peptides from a 56-AA region of Human Herpesvirus 4 [Epstein-Barr virus (EBV)] were primarily recognized by lambda antibodies across individuals (Fig. 3.1C). Additionally, a 56-AA region from Rhinovirus B contained two distinct public epitopes, one predominantly recognized by lambda and the other by kappa antibodies across individuals (Fig. 3.1C and fig. S3.2). Overall, we observed an inverse distribution in which epitopes tended to be recognized in many donors’ lambda IP samples and few kappa IP samples, or vice versa (Fig. 3.1D). This contrasted with the distribution expected if there were no systematic light chain isotype bias (Fig. 3.1D and fig. S3.3). In the few cases where a peptide was recognized by many kappa and many lambda IP samples, distinct kappa and lambda epitopes were evident from the triple-alanine-scan data. Thus, across human populations, antibodies specific for a given public epitope frequently use the same light chain isotype, suggesting that they may share structural similarity. Moreover, light chains appear to be important for antibody recognition of public epitopes. Antibodies recognizing public epitopes exhibit similar high-resolution footprints Having found that different individuals recognize similar regions within publicly recognized 56-AA peptides, we next sought to map public epitopes at even higher resolution. We designed a VirScan library with 407 short peptide truncations (“minimal peptides”) that captured most of the antibody responses to the original 363 publicly recognized 56-AA peptides (Fig. 3.1C and table S3.3), as well as associated saturating mutants in which each AA was substituted with each of the other 19 possible AAs (3.19, 3.22) (Fig. 3.2A, fig. S3.4, and table S3.4). We profiled the ~70 human serum samples with the saturating mutagenesis public epitope library, using the kappa- and lambda-specific IP protocol, thus generating a set of high-resolution antibody footprints that identified residues critical for antibody recognition. For most minimal peptides, different human serum samples produced high-resolution footprints that were similar (Fig. 3.2B). In many cases, most individuals’ high-resolution footprints for a given minimal peptide were highly correlated, often as much so as technical replicates. In some cases, two or more distinct groups of highly correlated high-resolution footprints for a given minimal peptide were evident. Thus, in many cases, different individuals appear to generate antibodies to precisely the same epitopes. For each minimal peptide, we identified the dominant pattern of critical residues recognized by kappa or, separately, by lambda IP samples. This resulted in a set of 376 consensus viral public epitopes defined at AA resolution: 189 recognized by kappa antibodies (“kappa public epitopes”) and 187 recognized by lambda antibodies (“lambda public epitopes”) (tables S3.5 and S3.6). For 232 of the 376 consensus public epitopes, sera from more than one-third of individuals that recognized the minimal peptide targeted the same consensus pattern of critical residues (Fig. 3.2D and table S3.7). The public epitopes had an average of four critical residues and spanned an average of seven AAs from the first to last critical residue (tables S3.5 and S3.6). Substitution of critical residues with chemically related AAs (3.23, 3.24) were frequently tolerated (e.g., A-S, I-L, I-V, Y-F) (fig. S3.5), with some exceptions: neither K-R nor D-E Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 5 swaps were particularly well tolerated. Some differences in substitution tolerance in kappa versus lambda public epitopes were observed, (e.g., for W-F, W-Y, and A-P), suggesting different modes of binding certain AAs. Critical residues of public epitopes have a distinctive AA composition The large number of consensus public epitopes we mapped allowed us to examine their AA composition, relative to the human viral proteome. Most notably, lysines were significantly enriched among critical residues of lambda (P < 5 x 10−33, binomial test) but not kappa public epitopes. Among other differences, prolines and tryptophans were strongly enriched, whereas serines, threonines, and valines were depleted in all public epitopes. (Fig. 3.3A). Although some of these differences may have resulted from the enrichment of particular residues on protein surfaces, others may have reflected preferential recognition of these AAs by kappa and/or lambda antibodies. The majority of lambda public epitopes have border lysine residues Next, we investigated whether lysine residues were preferentially situated at particular position(s) within lambda public epitopes. We examined the frequency of each AA at border (first or last critical residues) or interior (all other critical residues) positions of public epitopes, relative to their frequency in the human viral proteome (Fig. 3.3, B and C). Several AAs were enriched or depleted at border or interior positions of kappa and/or lambda public epitopes. Most notably, lysine was enriched at border positions of public epitopes (enrichment P < 5 x 10−69). Of all 135 lysines in critical residues of lambda public epitopes, 127 were located at border positions and 61% of all lambda public epitopes featured a border lysine (table S3.6). We hypothesized that some lambda antibodies may harbor specificity for lysine. B cell receptors specific for three public epitopes exhibit conserved gene segment usage but distinct heavy chain CDR3 sequences To explore sequence determinants of specificity for public epitopes, we initially selected two minimal peptides as case studies, both of which elicited highly conserved high- resolution antibody footprints across individuals: a kappa minimal peptide from influenza A hemagglutinin and a lambda minimal peptide from EBV gp350. We then isolated and sequenced B cell receptors (BCRs) specific for these peptides (Fig. 3.4A). We obtained nine BCRs that recognized the influenza A minimal peptide from six donors (Fig. 3.4B; fig. S3.6, A to C; and table S3.8). All had kappa light chains and conserved gene segment usage: IgHV5-51 paired with IgKV4-1. They also featured similar light chain CDR3 sequences. The heavy chain CDR3 sequences were not conserved although they were longer than average (~20 versus ~15 AAs for the overall antibody repertoire) (3.25, 3.26). We profiled each of these antibodies with the saturating mutagenesis public epitope VirScan library and observed similar high-resolution footprints (fig. S3.7). All nine antibodies bound to intact influenza A H3 hemagglutinin trimers, but none were neutralizing (fig. S3.8). We obtained 19 BCRs that recognized the EBV minimal peptide from four donors (Fig. 3.4C; fig. S3.6, A, D, E, and F; and table S3.9). Many shared conserved gene segment Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 6 usage (IgHV1-46, frequently paired with IgLV3-10) but did not share conserved heavy chain CDR3 sequences. The IgHV1-46/IgLV3-10 BCRs from different donors exhibited almost identical high-resolution footprints (fig. S3.9), indicating that different individuals generate BCRs that recognize the EBV minimal peptide in extremely similar ways. A representative subset of the EBV minimal peptide-specific antibodies bound to full-length gp350 (fig. S3.10). As a third case study, we isolated 19 BCRs that bound a publicly recognized SARS-CoV-2 spike peptide that overlaps with the fusion peptide (Fig. 3.4D) (3.18). These BCRs exhibited more diverse V gene segment usage than the flu and EBV BCRs. Nevertheless, 11 BCRs featured IgHV3 genes (IgHV3-30, IgHV3-23, and IgHV3-64D) and diverse heavy chain CDR3 sequences, and these BCRs exhibited very similar high-resolution footprints (fig. S3.11). Thus, IgHV3-30, IgHV3-23, and IgHV3-64D may share common features that enable recognition of the spike fusion peptide epitope. All 19 antibodies bound to the S2 domain of spike though only a few bound to full-length spike, and none out of a representative subset were neutralizing (fig. S3.12). This was consistent with reports of other antibodies that bound to this fusion peptide epitope but only when spike was engaged with ACE2 and thus constrained in the up conformation (3.27–3.29). From these three case studies, the theme of conserved V gene segment usage in the absence of the conserved heavy chain CDR3 sequence suggested that antibodies may recognize public epitopes through germline-encoded sequences within the V gene segments. To examine potential polyspecificity of public epitope-reactive antibodies we profiled each of our influenza A-, SARS-CoV-2-, and a representative subset of our EBV minimal peptide-reactive antibodies against the human virome VirScan library (>100,000 peptides from >200 viral species) (3.15). Almost all the monoclonal antibodies we tested specifically bound peptides containing their cognate public epitope sequences; some cross-reacted with peptides that shared very similar sequences (fig. S3.13). Thus, the phenomenon of public epitopes is due to recurrently generated antibodies specific for these epitopes rather than polyreactive antibodies. A germline-encoded aspartic acid at position 51 of several lambda V gene segments drives specificity for border lysines The border lysine enrichment in lambda public epitopes suggested that lysine might specifically interact with lambda light chains, possibly through pairing with a germline- encoded acidic residue. We searched the Protein Data Bank (PDB) (3.30–3.32) for human lambda (n = 297) and kappa (n = 631) antibody-antigen (Ab-Ag) complexes and found that light chain position 51 directly interacted with lysines in antigens much more frequently in lambda than kappa Ab-Ag complexes (Fig. 3.5A). In these interactions, lambda light chain position 51 was almost always a germline-encoded aspartic acid. We selected two antibodies that bound the same EBV minimal peptide described above but that recognized distinct critical residues: EBV_c186, which recognized a border lysine, and EBV_c40, which did not (Fig. 3.5B and fig. S3.9). We individually mutated each aspartic acid (D) or glutamic acid (E) of their lambda light chains to lysine (K) (for maximal disruption) and assessed impacts on binding by dot blot. The D51K mutation disrupted the Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 7 binding of EBV_c186 but not EBV_c40, suggesting that D51 was important for border lysine recognition (Fig. 3.5B). The D51K mutation also disrupted the binding of three additional antibodies (EBV_c9, EBV_c101, and EBV_c150) whose epitopes within the EBV minimal peptide contained a border lysine, but had no effect on EBV_c124, whose epitope lacked a border lysine (Fig. 3.5C). At least six lambda V gene segments share similar germline-encoded lysine-specific binding motifs Although a single salt bridge may stabilize an interaction, it alone cannot confer lysine specificity. To define additional residues involved, we investigated all Ab-Ag structures from the PDB with light chain position 51-antigen lysine interactions and uncovered a family of six lambda V gene segments (IgLV3-10, IgLV3-25, IgLV6-57, IgLV3-1, IgLV3-21, and IgLV5-37) that shared similar germline-encoded lysine-specific binding motifs. We called these germline-encoded amino acid–binding (GRAB) motifs (Fig. 3.5D and table S3.10). The GRAB motif in IgLV3-1 encompassed germline-encoded residues Y32 from CDR1, D51 from CDR2, and N66 from framework region 3 to specifically bind lysine in the antigen. D51 made a salt bridge with the lysine amine and Y32 made nonpolar interactions with the carbons of the side chain. Five of the six unique IgLV3-1 Ab-Ag complexes in the PDB featured this lysine-GRAB motif interaction (Fig. 3.5D and table S3.11, tab A). IgLV3-10, IgLV3-25, IgLV6-57, and IgLV3-21 harbored nearly identical lysine-specific GRAB motifs to IgLV3-1, whereas the IgLV5-37 GRAB motif differed somewhat, with Y51, D52C, and N32 making cation-pi, salt-bridge, and hydrogen bond interactions with the lysine amine, respectively. Cumulatively, 75% (24 of 32) PDB Ab-Ag structures involving these six V gene segments featured a lysine-GRAB motif interaction (Fig. 3.5D and table S3.11, tab A). Of these 24 structures, 16 involved conformational epitopes, indicating that GRAB motifs are important for recognition of both conformational and linear epitopes. Additionally, the lysine was almost always found at the edge of the epitope in both conformational and linear antigens (fig. S3.14 and table S3.11, tab A), possibly because these GRAB motifs are largely encoded by CDR1 and CDR2, the loops of which are oriented on the “outside” of the variable domain structure relative to the “interior” CDR3 loops. Alignment of Ab-Ag structures with the lysine-GRAB motif interaction showed similar interaction orientations (fig. S3.15A). Thus, IgLV lysine GRAB motifs encode specificity for border lysines and may drive the enrichment of lysine at the borders of lambda public epitopes, profoundly influencing the AA composition of public epitopes. In addition to the six lambda V gene segments with lysine-specific GRAB motifs, four additional lambda V gene segments (IgLV3-902, IgLV3-16, IgLV3-22, and IgLV3-27) had germline-encoded Y/S32, D51, and S/T66 residues that were predicted by AlphaFold2 (3.33, 3.34) to fold into similar structures as the GRAB motifs described above (fig. S3.15B and table S3.11, tab C). However, there are currently no Ab-Ag complexes in the PDB involving these gene segments to confirm their specificity. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 8 Multiple heavy, kappa, and lambda V gene segments harbor GRAB motifs specific for particular AAs The principle of GRAB motifs is not necessarily limited to lambda V gene segments or lysines. Therefore, we expanded our PDB analysis to search all human V gene segments (heavy, kappa, and lambda) for GRAB motifs that recurrently bound any given AA. We identified five additional GRAB motifs (Fig. 3.5E and table S3.11, tab B). For example, IgHV3-21 harbored a GRAB motif encompassing germline-encoded CDR2 residues (S52, S52A, S53, S55, and Y56) to specifically bind aspartate or glutamate in the antigen. A subset of the serine residues hydrogen-bonded with the carboxylate moiety whereas the tyrosine made nonpolar interactions with the carbons in the aspartate/glutamate side chain. Four of eight distinct PDB Ab-Ag complexes involving IgHV3-21 featured this aspartate/ glutamate–GRAB motif interaction. The closely related IgHV3-11 had three alleles (03, 05, and 06) that were not represented in the PDB but had germline-encoded S52, S52A, S53, S55, and Y56 residues predicted to form the same GRAB motif (fig. S3.15C and table S3.11, tab C). IgHV5-51 encoded a GRAB motif comprising germline-encoded CDR2 and framework region 2 residues W33, Y52, D54, D56, and sometimes R58 that specifically interacted with a lysine in the antigen in 8 of the 10 distinct PDB Ab-Ag complexes involving IgHV5-51 (Fig. 3.5, E and F, and table S3.11, tab B). Similar to the IgLV lysine GRAB motif interactions, in the IgHV5-51 GRAB motif, D54 and D56 made salt bridges with the lysine amine whereas W33 and Y52 engaged in nonpolar interactions with the carbons of the side chain. Some GRAB motifs exhibited recognition flexibility. IgKV4-1 harbored a GRAB motif involving Y30A, Y32, and Y92, germline-encoded by the CDR1 and CDR3, which formed nonpolar interactions with antigens. Although primarily recognizing proline (four examples), this GRAB motif could also interact with histidine, valine, arginine, or alanine (one example each), underscoring the chemical utility of tyrosine for protein-protein interactions. Of the 22 distinct PDB IgKV4-1 Ab-Ag complexes, 8 featured interactions between this GRAB motif and the AAs listed above (Fig. 3.5E and table S3.11, tab B). If GRAB motif interactions were important for antigen recognition, they should contribute substantially to the DG of the Ab-Ag complex. Using computational alanine scanning (3.35, 3.36), we predicted the effects on the Ab-Ag complexes of mutating AAs recognized by GRAB motifs. The median predicted DDG was 1.9 kcal/mol and the median predicted fold change in binding affinity (KD) was 21.9, indicating that the GRAB motif interactions were important for Ab-Ag binding (table S3.11, tabs A and B). We also observed five recurrent germline-encoded interactions present in distinct Ab-Ag structures involving the same antigen—evidence of shared antibody responses to public epitopes. For example, recent reports described a public antibody response to the receptor binding domain of SARS-CoV-2 spike that used IgHV3-53 or the closely related IgHV3-66, frequently paired with IgKV1-9 or IgKV3-20 (3.10, 3.11, 3.37). These antibodies had unmutated or nearly unmutated sequences, yet potently neutralized SARS-CoV-2 (3.38). We observed that IgHV3-53/IgHV3-66 engaged in multiple stereotyped interactions between Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 9 germline-encoded residues fromCDR1, framework region 2, CDR2, and framework region 3 and Y473, the backbone near A475, and Y421 of spike (fig. S3.16 and table S3.11, tab D). Additionally, IgKV1-9/IgKV3-20 exhibited stereotyped interactions between germline- encoded residues of framework region 1, CDR1, framework region 3 (for IgKV3-20 only), and CDR3 and Y505 of spike (fig. S3.16 and table S3.11, tab D). These germline-encoded interactions appeared to contribute to the prevalence of the IgHV3-53/IgHV3-66 + IgKV1-9/ IgKV3-20 neutralizing antibody response among individuals exposed to SARS-CoV-2 (3.10, 3.11), much in the way we hypothesize that GRAB motif interactions do. These stereotyped germline-encoded interactions were included in our count of GRAB motifs, with the caveat that their generalizability to other antigens was uncertain. GRAB motifs mediate recognition of an influenza A public epitope If GRAB motifs mediate antibody recognition of public epitopes, mutation of GRAB motif residues should weaken this recognition. As a test case, we used influenza A public epitope- specific antibodies, because these shared conserved IgHV5-51 and IgKV4-1 gene segments, which we knew harbored GRAB motifs specific for lysine and proline, respectively. We individually mutated each AA of the IgHV5-51 and IgKV4-1 GRAB motifs to alanine in flu_c504 and flu_c3 and observed that these mutations often severely reduced binding, whereas mutations of nearby non–GRAB motif or of CDR3 residues often did not affect binding (Fig. 3.6, A and B, and fig. S3.17, A and B). We profiled the mutant versions of flu_c504 and flu_c3 using the saturating mutagenesis public epitope VirScan library. Use of substantial quantities of monoclonal antibody allowed us to obtain high-resolution antibody footprints for most mutants despite weakened binding. Although flu_c504 and flu_c3 originally recognized the critical residues P-GTL-K, mutation of IgHV5-51 GRAB motif residues specifically abolished recognition of the lysine while increasing the dependence on the other critical residues (Fig. 3.6, C and D). Likewise, IgKV4-1 GRAB motif mutations reduced proline recognition (fig. S3.17, C and D). In all cases, mutations outside of GRAB motifs did not affect the high-resolution antibody footprints. Thus, the IgHV5-51 and IgKV4-1 GRAB motifs likely mediate binding to the influenza A public epitope through specific recognition of lysine and proline, respectively, as predicted by the PDB analysis of these V gene segments. Public epitopes are largely species-specific We next asked whether different species recognized the same or distinct public epitopes. We used VirScan to map antibody responses to peptides from SARS-CoV-2 spike in 30 SARS- CoV-2–infected humans (3.18), 9 SARS-CoV-2–infected nonhuman primates (NHPs) (3.39), and 8 C57BL/6 mice vaccinated with adeno-associated virus (AAV) encoding SARS-CoV-2 spike (3.40). The general regions of spike recognized by all three species were similar (Fig. 3.7A), but when we reprofiled these samples using a SARS-CoV-2 public epitope saturating mutagenesis VirScan library (table S3.12), we observed that the precise public epitopes recognized by each species were most often distinct (Fig. 3.7, B to D, and fig. S3.18). Thus, mice and NHPs do not recapitulate the human antibody response to public epitopes. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 10 Different species have partially overlapping sets of GRAB motifs We hypothesized that different species might recognize different public epitopes in part because of distinct sets of GRAB motifs. To investigate, we performed a similar analysis of PDB Ab-Ag complexes as described above, for mouse V gene segments (we analyzed V gene segments from several different mouse strains). This revealed 21 murine GRAB motifs, which partially overlapped with human GRAB motifs (Fig. 3.7, E to I, figs. S3.19 to S3.21, and table S3.13). For example, four mouse V gene segments (denoted with “m”)—mIgHV1-5, mIgHV1-69, mIgHV8-9, and mIgHV8-12—had germline-encoded lysine/arginine-specific GRAB motifs resembling the human IgHV5-51 GRAB motif (Fig. 3.7, E and I, and table S3.13). Although the AAs (W, Y, D, and D) that constituted the human IgHV5-51, mIgHV8-9, and mIgHV8-12 GRAB motifs were equivalent, inmIgHV8-9 and mIgHV8-12, the W was encoded by the CDR2 whereas in human IgHV5-51 it was encoded by the framework region 2, illustrating convergent strategies to form similar GRAB motifs. Several mouse GRAB motifs had no discernable human equivalents (Fig. 3.7, H and I, and table S3.13). These included similar aspartate/glutamate–specific GRAB motifs in mIgHV1-4 and mIgHV1-7, a distinct set of aspartate/glutamate-specific GRAB motifs in mIgHV10-1 and mIgHV10S3, asparagine/glutamine–specific GRAB motifs in mIgLV3 and mIgHV9-2-1, an arginine/lysine–specific GRAB motif in mIgKV5-39, and a tyrosine- specific GRAB motif in IgKV6-17, among others. Furthermore, mice have only three functional lambda V gene segments and these did not share the lysine GRAB motifs present in human lambda V gene segments. In several cases, additional mouse V gene segments shared conserved residues with known mouse GRAB motifs (table S3.13, tab B), but Ab-Ag structures were not present in the PDB to validate their specificity. Thus, only partially overlapping sets of GRAB motifs in mice and humans, potentially coupled with distinct CDR3 sequences and subtle differences in the positions of GRAB motif residues within CDR loops and framework regions, could affect the geometry of antibody binding and hence epitope selection, thereby explaining why human and mice antibodies rarely recognize the same public epitopes. Discussion A fundamental question in immunology is why antibodies recognize particular regions of proteins more frequently than others. Our data support a model in which public epitopes arise in part because they are best aligned for recognition by GRAB motifs. A certain threshold binding energy is required to initiate an antibody response to an epitope. If a GRAB motif within a particular germline V gene segment provides a substantial portion of this binding energy, a larger number of CDR3 sequences would be compatible in the antibody because the CDR3 would need to contribute less binding energy to reach the threshold to progress to affinity maturation. Thus, there would be a relatively abundant precursor population of naïve B cells with adequate affinity for the epitope, and this could lead to an immunodominant, public antibody response (3.41, 3.42). Conversely, if a specific CDR3 sequence were required to provide most of the binding energy for a particular epitope, the antibody response to this epitope would be rarer because of the low precursor Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 11 frequency of naïve B cells with this CDR3 sequence (Fig. 3.8). This model is supported by evidence of precursor frequency and binding affinity affecting antibody selection for VRC01-class HIV neutralizing antibodies from quantitative B cell transfer experiments in studies of immunodominance (3.43). The evolution of antibody genes with germline-encoded sequences that bind AAs commonly found on the surface of proteins is clearly advantageous, allowing the immune system to recognize pathogens quickly and efficiently. However, given their prevalence, shared antibody responses can exert population-wide selective pressures on pathogens. This has been observed clearly for SARS-CoV-2: variants of concern have evolved to evade recognition by the public IgHV3-53/IgHV3-66 + IgKV1-9/IgKV3-20 class of neutralizing antibodies described above, among others (3.44–3.48). Alternatively, if public antibody responses are nonprotective to the host, in principle viruses could exploit this by conserving the cognate epitopes. They could also evolve additional non-neutralizing epitopes easily recognized by host GRAB motifs, thereby eliciting frequent nonprotective antibody responses across the host population and potentially delaying the production of more protective antibodies. Such epitopes could be removed from vaccine formulations. An outstanding question is what selective pressure drove the expansion of lambda V gene segments with lysine-specific GRAB motifs in humans. Although mice do not share these GRAB motifs, 8 rhesus macaque lambda V gene segments share the same residues as known human lysine-specific GRAB motifs (3.49, 3.50) and as many as 39% (12 of 31, including hypothetical GRAB motifs) of the functional human lambda V gene segments have germline-encoded specificity for lysine. The expansion of lysine-specific GRAB motifs in primates suggests adaptation to pathogens, which in turn suggests an advantage for pathogens to be enriched in lysines. Notably, in the SARS-CoV-2 omicron BA.1 variant, 8 of the 30 AA substitutions in spike involve mutation to lysine (3.51). Recent modeling has suggested that positive charges on viruses may recruit heavily sialated mucins to enhance survival in aerosols or aid in interactions with lung surfaces (3.52, 3.53), providing a hypothesis for further study. We have likely discovered only a fraction of all GRAB motifs as our analysis was limited by available PBD Ab-Ag structures (3.54). Indeed, 39% of human V gene segments (50 of 127 annotated by IMGT) were not represented and an additional 15% were only represented in one or two distinct Ab-Ag structures. Additional structural data and enhanced computational approaches will be needed to bridge this gap. Furthermore, GRAB motifs may have specificity for combinations of AAs or more complex topological structures, which were beyond the scope of our current analysis. Nevertheless, the GRAB motifs we identified—18 human, 21 mouse with structural evidence, and an additional 6 human and 27 mouse predicted GRAB motifs based on conservation to known motifs—likely influence the selection and composition of public epitopes, as illustrated by the profound enrichment of border lysines in lambda public epitopes and potentially also the enrichment of proline in kappa public epitopes. This work has several implications: first, it suggests that private rather than public neutralizing antibodies may be superior candidates for inclusion in therapeutic monoclonal Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 12 antibody cocktails, because private antibodies are less likely to exert population-wide selective pressures on pathogens and may thus retain efficacy for future variants. Second, the fact that public, immunodominant antibody responses appear to be largely species- specific may limit our ability to consistently predict how vaccines tested in nonhuman species will perform in humans, especially with respect to cross protection to variants. However, in another context species specificity may prove beneficial: vaccines administered to nonhuman species may elicit neutralizing antibody responses to epitopes that are not publicly recognized by humans. Because human viruses are not under evolutionary pressure to evade these antibodies, they may have therapeutic efficacy for humans against a broad range of variants. Third, the data presented here may enable exploration of the functional consequences of antibody responses to public epitopes, with relevance to vaccine design. Fourth, the set of viral public epitopesmay be useful in diagnostic applications. Fifth, knowledge of the GRAB motifs should aid species-specific B cell epitope prediction algorithms and computational methods to predict and design Ab-Ag interactions. Overall, this study reveals a fundamental structural code inherent in our humoral immune response that shapes epitope selection and composition and drives recurrent antibody responses across individuals and differing epitope selection among species, thus affecting host-pathogen coevolution and human health. Additionally, as T cell receptors are structurally similar to BCRs, it is highly likely that a similar structural code exists within T cell receptor V gene segments that contributes to T cell epitope immunodominance. Materials and Methods Human donor samples Human specimens were collected in accordance with the local protocol governing human research after obtaining informed written consent from the donors. Secondary use of all human samples for the purposes of this work was exempted by the Brigham and Women’s Hospital Institutional Review Board (protocol number 2013P001337). Samples included serum and plasma from donors residing in Peru (n = 24), France (n = 2), and the United States (n = 52) (3.14). The United States cohort included donors with hepatitis C virus (n = 24), donors with human immunodeficiency virus 1 (n = 24), and healthy donors (n = 4). Human serum and plasma samples were stored in aliquots at −80°C until use. Apheresis leukoreduction collars from healthy platelet donors were obtained from the Brigham and Women’s Hospital Specimen Bank under protocol T0276. Access to COVID-19 patient samples was facilitated by the MassCPR. Design and cloning of the public epitope truncation and alanine scanning library We designed peptide sequences of 15, 20, 25, 30, 35,40, and 45AAs in length, tiling through all of the 56-AA publicly recognized peptides with 5-AA overlap. For these shorter peptide truncations, we added random filler AA sequences after the stop codon so that downstream PCR steps would produce amplicons of the same size for all the members of the library. Additionally, we made triple-mutant sequences scanning through the 56-AA peptides. Non-alanine AAs were mutated to alanine, and alanines were mutated to glycine. We reverse translated the peptide sequences into DNA sequences that were (a) codon-optimized for expression in Escherichia coli, (b) lacked restriction sites used in Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 13 VirScan downstream cloning steps (EcoRI and XhoI), and (c) were unique in the 50 nucleotides (nt) at the 5′ end to allow for unambiguous mapping of the sequencing reads. Then we added the adapter sequences AGGAATTCCGCTGCGT to the 5′ end and CAGGGAAGAGCTCGAA to the 3′ end or ATGAATTCGGAGCGGT to the 5′ end and CACTGCACTCGAGACA to the 3′ end to form the 200-nt oligonucleotide sequences, which were synthesized on a releasable DNAmicroarray (Agilent). We PCR-amplified the DNA oligonucleotide library with the primers T7-PFA 5′-AATGATACGGCGGGAATTCCGCTGCGT-3′ and T7-PRA 5′-CAAGCAGAAGACTCGAGCTCTTCCCTG-3′ and, separately, with the primers T7-Pep2-PFb 5′-AATGATACGGCGTGAATTCGGAGCGGT-3′ and T7-Pep2-PRb 5′-CAAGCAGAAGACGTCTCGAGTGCAGTG-3′, digested the products with EcoRI and XhoI, and cloned them into the EcoRI/SalI site of the T7FNS2 vector (3.17). We packaged the resultant library into T7 bacteriophage using the T7 Select Packaging Kit (EMD Millipore) and amplified the library according to the manufacturer’s protocol. We performed VirScan (3.14–3.16), which is based on the phage immunoprecipitation and sequencing (PhIP-Seq) methodology (3.17), as described previously (3.14–3.16, 3.18) or with slight modifications. For the light chain isotype-specific IPs, we substitutedmagnetic protein A and protein G Dynabeads (Invitrogen) with 5 mg of biotinylated goat anti-human kappa (Southern Biotech) or 4 mg of biotinylated goat anti-human lambda (Southern Biotech) antibodies. The day after establishing the phage and serum mixtures, we added these antibodies and incubated the reactions overnight at 4°C. Afterward, we added 20 ml of Pierce streptavidin magnetic beads (Thermo Fisher Scientific), incubated the reactions for 4 hours at room temperature, then continued with the washing steps and the remainder of the protocol, as previously described (3.14–3.16, 3.18). For the isotype-specific depletions, we substituted magnetic protein A and protein G Dynabeads with 15 mg of biotinylated goat anti-human kappa or 10 mg of biotinylated goat anti-human lambda antibodies. The day after establishing the phage and serum mixtures, we added these antibodies to the phage and serum mixtures and let the reactions incubate overnight at 4°C. We then added 60 ml (for kappa depletions) or 40 ml (for lambda depletions) of Pierce streptavidin magnetic beads, incubated the reactions for 4 hours at room temperature, then moved the supernatants into new plates. We added 40 ml of mixed protein A and protein G Dynabeads to the supernatants, incubated the reactions for 4 hours at room temperature, and continued with the IPs and library preparation for multiplexed Illumina sequencing as described previously (3.14–3.16, 3.18). To test whether we could successfully profile antibody responses to the saturating mutagenesis public epitope library, we used the following antibodies against HA tag (which was included in the saturating mutagenesis public epitope library as a control): mouse anti-HA-biotin clone HA-7 (Sigma), rat anti-HA-biotin clone 3F10 (Sigma), and anti-HA magnetic beads clone 2-2.2.14 (Thermo Fisher Scientific) (fig. S3.4). For mouse serum samples, 0.6 ml of mouse serum was used for each VirScan reaction and 40 ml of mixed protein A and protein G Dynabeads were used as the IP reagent. For NHP samples, 0.2 ml of NHP serum was used for each VirScan reaction and 40 ml of mixed Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 14 protein A and protein G Dynabeads were used as the IP reagent. For monoclonal antibodies, ~50 ml of cell culture supernatant was used as input for VirScan reactions involving the saturating mutagenesis public epitope library to generate high-resolution footprints. Unless otherwise specified, 20 ng of purified antibody was used as input for VirScan reactions involving the human virome library to investigate potential polyspecificity. Forty microliters of mixed protein A and protein G Dynabeads were used as the IP reagent. All samples were run in duplicate except for the mouse sera samples profiled with the CoV 56-AA library (however, this library contains duplicate barcoded versions of each peptide, and the measurements for each of the duplicate peptides were averaged). Statistical analysis of VirScan data generated with the public epitope truncation and alanine scanning library We first mapped the sequencing reads to the reference library sequences using Bowtie (3.55) and counted the number of reads corresponding to each peptide in the input library and each sample “output”. For each sample, we normalized the read counts for each peptide by the total read counts for the sample. Then, we divided the normalized read counts of each peptide in the sample by the normalized read counts of each peptide in the input library to obtain an enrichment value. We averaged enrichment values for technical replicates of a sample. For each truncation and alanine scanning mutant peptide, we calculated relative-to-wildtype enrichment values as follows: we first calculated the average enrichment value of the middle 50% of the alanine scanning mutants of a given 56-AA peptide, as we assumed most alanine scanning mutations throughout the peptide would not disrupt the epitope. We found this to be a more robust (i.e., less noisy) representation of the enrichment of the wild-type 56-AA peptide than the enrichment value of the single wild-type 56-AA peptide. Next, we divided the enrichment value of a given peptide truncation or alanine scanning mutant by the average enrichment value of the middle 50% of the alanine scanning mutants to obtain a relative-to-wild-type enrichment value. Finally, for each 56-AA peptide, we generated a heatmap using Python matplotlib to illustrate the relative-to-wild-type enrichment values of the peptide truncations and alanine scanning mutants for all the samples that recognized the wild-type version of the 56-AA peptide (i.e., where the enrichment value of the wildtype 56-AA peptide was >1.5). For the alanine scanning mutants, the values in the heatmap were 1 / (relative-to-wild-type enrichment value), with darker blue colors indicating greater disruption of the epitope. Permutation analysis We limited this analysis to the short peptide truncations (15, 20, 25, and 30-AA in length) as some 56-AA peptides contained more than one distinct public epitope, and we sought to isolate these with the shorter peptides. To perform one permutation, we randomized the kappa and lambda assignments of the pair of IPs for each serum sample. We then counted the number of kappa and lambda IP fractions in which each short peptide truncation was enriched. We performed a total of 1000 permutations. Based on these permutations, we calculated an average distribution of kappa and lambda IP samples in which the short peptide truncations were expected to score, and used this distribution to calculate the fold- Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 15 enrichment of each value in the observed distribution. We also calculated a P-value for each observed value based on how many random permutations had resulted in at least such a high number of peptides scoring in the given numbers of kappa and lambda IP samples. Design and cloning of the saturating mutagenesis public epitope library From the VirScan data generated with the public epitope truncation and alanine scanning library using the kappa and lambda isotype-specific IP protocol, we identified peptide truncations whose relative-to-wild-type enrichment values were at least 0.75 in at least half of the samples in which the wild-type 56-AA peptide scored. We filtered for peptide truncations with read counts of at least 5 in the input library to avoid spurious enrichment values, and peptide truncations for which at least 6 samples recognized the wild-type 56-AA peptide. We first chose 15-AA and 20-AA peptide truncations that met these requirements. This was a relatively stringent set of criteria, so to capture the remaining epitopes within the 56-AA peptides of the public epitope truncation and alanine scanning library, we next set the threshold for the relative-to-wild-type enrichment values to 0.5 and chose the shortest peptide truncation that captured at least half of the samples’ responses to the wild-type 56-AA peptide. With this list of minimal public epitope-containing peptides (“minimal peptides”), we designed saturating mutants such that each AA of the peptide was mutated to the other 19 AAs. As a positive control and to calibrate how antibody responses to saturating mutants would be detected in a VirScan assay, we included HA tag and saturating mutants of this epitope. Because the public epitope peptides were of varying sizes, we added random filler AA sequences after the stop codon so that downstream PCR steps would yield products of the same size for all the members of the library. We reverse-translated the peptide sequences into DNA sequences that were codon-optimized for expression in E. coli, that lacked restriction sites used in downstream cloning steps (EcoRI and XhoI), and that were unique in the 50 nt at the 5′ end to allow for unambiguous mapping of the sequencing reads. Then we added the adapter sequence GGAATTCCGCTGCGT to the 5′ end and CAGG-GAAGAGCTCGA to the 3′ end to form the 198-nt oligonucleotide sequences. These oligonucleotide sequences were synthesized on a releasable DNA microarray (Agilent). We PCR-amplified the DNA oligonucleotide library with the primers T7-PFA 5′-AATGATACGGCGGGAATTCCGCTGCGT-3′ and T7-PRA 5′-CAAGCAGAAGACTCGAGCTCTTCCCTG-3′, digested the product with EcoRI and XhoI, and cloned it into the EcoRI/SalI site of the T7FNS2 vector (3.17).We packaged the resultant library into T7 bacteriophage using the T7 Select Packaging Kit (EMD Millipore) and amplified the library according to the manufacturer’s protocol. Statistical analysis of VirScan data generated with the saturating mutagenesis public epitope library and the SARS-CoV-2 public epitope saturating mutagenesis library We first mapped the sequencing reads to the reference library sequences using Bowtie (3.55) and determined the read counts of each peptide in the input library and each sample “output”. For each sample, we normalized the read counts corresponding to each peptide by the total read counts for the sample. Then, for each peptide, we divided the normalized read Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 16 counts of each peptide in the sample by the normalized read counts of each peptide in the input library in order to obtain the enrichment value. For each minimal peptide and for each sample, we next calculated relative-to-wild-type enrichment values for each AA substitution mutant as follows: we first calculated the average enrichment value of the middle 50%of the alanine substitution mutants, most of which we assumed would be minimally disruptive to the epitope contained in the peptide. We found that the average enrichment value of the middle 50% of the alanine substitution mutants was a more robust representation of the enrichment of the wild-type peptide than the enrichment value of the single wildtype peptide. Next, we divided the enrichment value of a given substitution mutant by the average enrichment value of the middle 50% of the alanine substitution mutants in order to obtain a relative-to-wild-type enrichment value. Finally, for each minimal peptide, if a sample recognized the wild-type version of the peptide (i.e., the enrichment value of the wildtype peptide was >1 and the average enrichment value of the middle 50% of all substitution mutants of the peptide was >1), we then generated an “enrichment matrix” with the relative-to-wild-type enrichment values of all the substitution mutants of the peptide. We also generated a heatmap using Python matplotlib displaying values of 1 / (relative-to-wild-type enrichment value + 0.2) for all the substitution mutants of the peptide. Statistical analysis of VirScan data generated with the human virome library and the CoV 56-AA library VirScan data generated with the human virome library and the CoV 56-AA library were analyzed as previously described (3.15, 3.16, 3.18). Critical residue analysis and definition of kappa and lambda public epitopes For every minimal peptide in the saturating mutagenesis public epitope library, we converted all enrichment matrices from samples that recognized the given peptide into binary matrices: if the relative-to-wild-type enrichment value of a given mutant peptide was <0.5 (i.e., the mutant enriched less than half as well as the middle 50% of all alanine substitution mutants for that peptide), then the mutant was considered to disrupt the epitope and given a value of “1”. Alternatively, the mutant was considered to be permitted and given a value of “0”. Next, we collapsed each binary enrichment matrix into a one-row summary by adding the number of mutants at each position that disrupted the epitope. Then, for each minimal peptide, we converted the one-row summaries into binary one-row summaries: if at least one third of the 19 substitutions at a given position disrupted the epitope, then the position was considered a critical residue and given a value of “1”. Alternatively, the position was given a value of “0”. These data are available on the Harvard Dataverse, doi: 10.7910/DVN/AIXWW2. Next, for each minimal peptide, we counted the number of samples that exhibited the same binary summaries (i.e., that recognized the same pattern of critical residues). The pattern of critical residues shared by the greatest number of samples was considered to be the consensus public epitope, also called the dominant footprint. Thus, we defined the critical residues of the kappa and lambda public epitopes. Samples that recognized the consensus public epitope or a pattern of critical residues that only differed from the consensus public epitope by one position were considered to be part of the dominant footprint group. The Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 17 counts and proportions of kappa or lambda IP samples that recognized each minimal peptide and that were part of the dominant footprint group are provided in table S3.7. We performed this analysis separately for kappa samples and lambda samples and limited the analysis to minimal peptides recognized by at least five kappa samples or at least five lambda samples. Analysis of tolerated AA substitutions For each minimal peptide, we took the average of all the binary matrices for samples that were part of the dominant footprint group. Then, for every critical residue of a given public epitope, we determined which AA substitutions were permitted (i.e., the AA substitutions for which the average of the binary matrices was >0.5). Finally, for each of the 20 AAs, we calculated the frequency at which each of the other 19 AA substitutions were permitted at a critical residue. Public epitope-specific memory B cell isolation and sequencing For most experiments, ~10 fresh (<6 hours from collection) apheresis leukoreduction collars from healthy platelet donors were obtained from the Brigham and Women’s Hospital Specimen Bank under protocol T0276. For the experiment to isolate SARS-CoV-2 public epitope-specific memory B cells, five cryopreserved peripheral blood mononuclear cell (PBMC) samples and one fresh leukopak sample from COVID-19-recovered donors were purchased from Cellero and obtained from the MassCPR COVID-19 Biorepository, respectively. PBMCs were purified on a Ficoll-Paque density gradient. Briefly, 8ml of donor blood was diluted 1:1 with PBS, slowly layered on 16 ml of Ficoll-Paque (Thermo Fisher Scientific) and centrifuged at 400g for 30 min with the brake off. The upper layer containing plasma and platelets was removed and frozen, and the mononuclear cell layer at the interface was extracted and washed four times with PBS at 400g for 10 min with the brake on. PBMCs from the different apheresis collars were counted, pooled together, and switched memory B cells were purified using the Human Switched Memory B cell Kit (Miltenyi) according to the manufacturer’s instructions. We used this kit, which employs a negative-selection protocol, rather than a kit that positively selects for IgG+ memory B cells, to avoid labeling the BCR with an antibody and potentially influencing the ability of the BCR to bind the viral peptide. Purified memory B cells were resuspended in RPMI 1640 (Life Technologies) with 10% (v/v) FBS (Hyclone), 100 U/ml of penicillin, 100 mg/ml of streptomycin, and incubated at 37°C and 5% CO2 for a few hours (≤6 hours, to ensure high cell viability). Memory B cells were stained with biotinylated minimal peptides conjugated to fluoresceinated streptavidin and then fluorescent cells were isolated by FACS using the MoFlo Astrios EQ Cell Sorter (Beckman Coulter) (3.56) (Fig. 3.4A). We typically began with 8x109 PBMCs from ~10 donors and ultimately purified 5x107 switched memory B cells. In cases where we wanted to sort for multiple minimal peptide specificities, we split the switched memory B cells into multiple aliquots, and labeled the different aliquots with distinct CITE-seq barcodes (3.57) customized to be compatible with the Chromium 5′ V(D)J solution (10x Genomics). We sorted ~0.002% of the switched memory B cells for any given minimal peptide specificity, then sequenced their BCRs using the Chromium 5′ V(D)J solution (10x Genomics) according to the manufacturer’s instructions. In addition to amplifying and sequencing the BCR transcripts, we designed custom primers to amplify MHC transcripts and the customized CITE-seq barcodes (3.58, 3.59). SARS-CoV-2 public Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 18 epitope-reactive BCRs were sequenced using the Chromium Next GEM Single Cell 5′ Kit v2 (10x Genomics). MHC transcripts were not sequenced for these cells. Minimal peptides used for staining and sorting memory B cells were designed to have an N-terminal biotin or biotin-Ahx modification followed by a GGGGS linker sequence, and neutral charge. If a minimal peptide sequence was not neutral, the linker was extended with charged AAs to achieve net neutral charge. Biotinylated peptides were ordered from Thermo Fisher Scientific or GenScript, reconstituted in a small amount of DMSO (20–40 ml, or 2 to 4% of the final volume), then diluted to a final concentration of 1 mg/ml in ultrapure water and stored at −20°C in aliquots. The sequences of the biotinylated public epitope peptides were as follows: 146786_neutral_linker_flu: N terminus-Biotin-Ahx-GGGGSVPNGTLVKTITNDQI-C terminus 72153_neutral_linker_EBV: N terminus-Biotin-Ahx-EGEGGGGSPPSTSSKLRPRWTFT-C terminus SARS-CoV-2_S807-832: N terminus-Biotin- GGGGSDPSKPSKRSFIEDLLFNKVTLADAG-C terminus Biotinylated peptides were conjugated to fluorescent streptavidin by combining the reagents in the following ratios: For a 20-AA peptide: 6.3 mg of public epitope peptide: 9.0 mg of streptavidin-APC (Thermo Fisher Scientific) 19 mg of public epitope peptide: 9.5 mg of steptavidin-488 (Thermo Fisher Scientific) 3.9 mg of irrelevant peptide: 10.5 mg of streptavidin-PE (Thermo Fisher Scientific) 3.3 mg of irrelevant peptide: 10 mg of streptavidin-BV421 (Biolegend) For a 25-AA biotinylated peptide: 7.7 mg of public epitope peptide: 9.0 mg of streptavidin-APC 23.4 mg of public epitope peptide: 9.5 mg of streptavidin-488 4.7 mg of irrelevant peptide: 10.5 mg of streptavidin-PE 4.0 mg of irrelevant peptide: 10 mg of streptavidin-BV421 Streptavidin-peptide complexes were incubated at 4°C for ~4 hours on a rotator, then purified using a Bio-Spin® P-30 Gel Column into Tris Buffer (Bio-Rad) according to the manufacturer’s instructions. Immediately prior to staining switched memory B cells, two customized CITE-Seq ADTs per memory B cell aliquot were pooled and cleaned on a 50 kDa cutoff column as previously Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 19 described (3.57). Switched memory B cells were centrifuged at 300g for 10 min, then washed once with 1ml of staining buffer (PBS + 2% BSA + 0.02% Tween 20) and centrifuged at 400g for 4 min. Next, cells were resuspended in 100 ml of staining buffer and 100 ml of cleaned ADT pool (containing ~1–2 mg of each ADT) and incubated with end-over-end mixing for 30 min at 4°C. ADT-labeled switched memory B cells were washed once with staining buffer, then resuspended in purified streptavidin-peptide complexes plus 150 ml of staining buffer and incubated with end-over-end-mixing for 1 hour at 4°C. Then, cells were centrifuged at 400g for 4 min, washed twice in staining buffer, resuspended in 750 ml of staining buffer, and filtered over a 35-mm nylon mesh cell strainer. Cells that were negative for the two fluorophores conjugated to irrelevant peptides and positive for the two fluorophores conjugated to the minimal public epitope peptide were sorted using the MoFlo Astrios EQ Cell Sorter (Beckman Coulter) into 4 ml of RPMI-1640 supplemented with 0.2% BSA, 100 U/ml of penicillin, 100 mg/ml of streptomycin in one well of a 96-well plate. After sorting, cells were immediately used as input for single-cell BCR sequencing using the Chromium 5′ V(D)J solution (10x Genomics) according to the manufacturer’s protocol with slight modifications (described below) to amplify ADT barcodes and MHC transcripts in addition to BCR transcripts. V, D, and J gene segments assigned by Cell Ranger (10x Genomics) were double checked by entering the nucleotide sequence of the BCR variable region as a query sequence in IgBlast (3.60), using the IMGThumanV, D, and J (F + ORF) germline gene databases. Where the IgBlast and Cell Ranger gene segment assignments differed, the IgBlast assignments were used. CITE-Seq ADT customization for Chromium 5′ V(D)J solution (10x Genomics) ADT barcodes were designed as follows to be compatible with the Chromium 5′ V(D)J solution (10x Genomics): 5′ to 3′: 4 nt linker–10xVDJ_ADT_inner primer binding site–Read 2 adaptor sequence– random barcode–13 nt homology to the template switch oligo (10x Genomics) An example sequence is given below: 5′- /5AmMC12/ATCT–GCGTTCGAGCTCTTCCCTG– GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT– ATGGACCTTAAGCGCTACCGGAATGGTTCG–CCCATATAAGAAA −3′ This design allowed for a two-step PCR enrichment, the first step using SI-PCR primer (10x Genomics) and 10xVDJ_ADT_inner, and the second using SI-PCR primer and a Sample Index PCR primer (10x Genomics). After the cDNA amplification step of the Chromium 5′ V(D)J solution (10x Genomics) protocol, amplification products above 400 bp, including MHC and BCR transcripts, were captured on SPRI beads (Beckman Coulter) using 0.6X SPRI. The supernatant containing amplification products under 400 bp, including ADT barcodes, was removed to a separate tube, 1.4X SPRI was added to obtain a final SPRI volume of 2X SPRI, and purified in parallel with the amplification products over 400 bp. ADT Target Enrichment 1 from Amp cDNA was performed using SI-PCR primer and 10xV(D)J_ADT_inner primer. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 20 Amplification products from nine cycles of PCR were purified with 2X SPRI. ADTs were indexed and purified using the conditions given by the Chromium 5′ V(D)J solution (10x Genomics) protocol. The Sample Index PCR served both to index the ADTs and to perform a second step of target enrichment. The sequence of the 10xV(D)J_ADT_inner primer was as follows: 10xV(D)J_ADT_inner: 5′-GCGTTCGAGCTCTTCCCTG-3′ After quantification, libraries were mixed (50% BCR transcripts, 25% MHC I transcripts, 12.5% MHC II transcripts, 12.5% ADT barcodes). Sequencing was performed with a NextSeq 500 (Illumina) per manufacturer’s instructions. Next-generation sequencing reads corresponding to ADTs were separated by cell barcode. For each cell’s ADT reads, the number of times each ADT barcode appeared was counted. The ADT pair with the greatest counts indicated the peptide for which the cell was sorted. MHC enrichment primer design and donor identification strategy Primers to amplify MHC I and DR transcripts were designed by downloading CDS sequences of all HLA-A, HLA-B, HLA-C, DRA, and DRB alleles from the Immuno Polymorphism Database-ImMunoGeneTics information system ® / Human Leukocyte Antigen (IPD-IMGT/HLA) (3.58, 3.59), aligning all alleles of each gene, and designing primers in conserved regions to cover over 95% of alleles and produce reverse transcription products whose lengths would be compatible with the downstream Chromium 5′ V(D)J solution (10x Genomics) protocol. The sequences of the primers designed to amplify the MHC transcripts were as follows: HLA_A_Outer_1_346 bp: 5′-CAGGGCGATGTAATCCTTGC-3′ HLA_A_Outer_2_450: 5′-CAAGGCGATGTAATCCTTGC-3′ HLA_B_Outer_385bp: 5′-TCCTCGTTCAGGGCGATGT-3′ HLA_C_Outer_360bp: 5′-GCGATGTAATCCTTGCCGTC-3′ HLA_A_Inner_1_346bp: 5′-AACCGGCCTCGCTCTGG-3′ HLA_A_Inner_2_450bp: 5′-GAACCGTCCTCGCTCTGGT-3′ HLA_B_Inner_279bp: 5′-TGTGAGACCCGGCCTCG-3′ HLA_C_Inner_250bp: 5′-CTCGCTCTGGTTGTAGTAGC-3′ DRA_outer: 5′-ATGAAACAGATGAGGACG-3′ DRB_outer_1: 5′-CTCGCCGCTGCACTGTG-3′ DRB_outer_2: 5′-CCCCGTAGTTGTGTCTGCA-3′ DRA_inner: 5′-CTCTCTCAGTTCCACAGGGC-3′ Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 21 DRB_inner_1: 5′-CCCAGCTCCGTCACCGC-3′ DRB_inner_2: 5′-GTCCTTCTGGCTGTTCCAG-3′ MHC I Target Enrichment 1 from Amp cDNA was performed using SI-PCR primer and MHC I PCR1 RV “outer” mixture (consisting of HLA-A outer 1, HLA-A outer 2, HLA-B outer, HLA-C outer). Amplification products from 10 cycles of PCR were purified with 0.8X SPRI. MHC II Target Enrichment 1 from Amp cDNA was performed using SI-PCR primer and MHC II PCR1 RV “outer” mixture (consisting of DRA_outer, DRB_outer_1, and DRB_outer_2). Amplification products from 10 cycles of PCR were purified with 0.8X SPRI. MHC I Target Enrichment 2 was performed using 10xV(D)J_PCR2F primer and MHC I inner primer mix 2 (consisting of HLA-A inner 1, HLA-A inner 2, HLA-B inner, HLA-C inner). Amplification products from 10 cycles of PCR were purified with a double-sided size selection using 0.5X SPRI and 0.8X SPRI. MHC II Target Enrichment 2 was performed using 10xV(D)J_PCR2F primer and MHC II inner primer mix 2 (consisting of DRA_inner, DRB_inner_1, and DRB_inner_2). Amplification products from 10 cycles of PCR were purified with a double-sided size selection using 0.5X SPRI and 0.8X SPRI. MHCI and MHCII samples were indexed and purified using the conditions given by the Chromium 5′ V(D)J solution (10x Genomics) protocol. The sequence of the 10xV(D)J_PCR2F is as follows: 10xV(D)J_PCR2F: 5′-AATGATACGGCGACCACCGAGATCT-3′ After quantification, libraries were mixed (50% BCR transcripts, 25% MHC I transcripts, 12.5% MHC II transcripts, 12.5% ADT barcodes). Sequencing was performed with a NextSeq 500 (Illumina) per manufacturer’s instructions. Next-generation sequencing reads corresponding to MHC I and MHC II transcripts were separated by cell barcode. For each cell’s MHC I and MHC II reads, we searched for unique sequences ~90 nt in length that appeared in at least 8% of reads. These unique sequences were considered the MHC alleles for the donor from which the cell originated. Cells with identical MHC alleles were considered to come from the same individual. Identification of influenza A minimal peptide-specific memory B cells We performed two rounds of sorting for memory B cells that bound the influenza A minimal peptide VPNGTLVKTITNDQI, the first using 11 donor blood collars and the second using a new set of 10 donor blood collars. During the first round, we obtained 30 paired heavy and light chain sequences and chose seven to clone and recombinantly express, including four that featured similar gene segment usage. Using dot blot and VirScan with the saturating mutagenesis public epitope library, we validated that only these four antibodies: flu_c326, flu_c357, flu_c504, and flu_c645, originating from three different donors, were specific for the VPNGTLVKTITNDQI peptide (Fig. 3.4B, fig. S3.6A). These four antibodies exhibited highly conserved gene segment usage: they all featured IgHV5-51 in the heavy chain and IgKV4-1 and IgKJ2 in the light chain. These antibodies also shared highly similar light chain CDR3s, but very poorly conserved heavy chain CDR3 sequences. During the second round of the experiment, we obtained 71 Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 22 paired heavy and light chain sequences and chose 12 of these to clone and recombinantly express, including three that contained the IgHV5-51 / IgKV4-1 / IgKJ2 gene segment combination, as well as all antibodies with either IgHV5-51 or IgKV4-1, but not the combination of both. Using dot blot VirScan with the saturating mutagenesis public epitope library, we validated that only the three antibodies with the IgHV5-51 / IgKV4-1 / IgKJ2 combination: flu_c3, flu_c286, and flu_c473, originating from three different individuals, were specific for VPNGTLVKTITNDQI (Fig. 3.4B, fig. S3.6C). None of the antibodies with only IgHV5-51 or only IgKV4-1 bound the influenza A minimal peptide. Of the 101 BCR sequences obtained from VPNGTLVKTITNDQI-sorted memory B cells from the two sorting experiments, the IgHV5-51 / IgKV4-1 / IgKJ2 gene segment combination was only found in the seven antibodies that we ultimately validated as specific for the peptide. The full list of antibodies we tested for specificity for VPNGTLVKTITNDQI is provided in table S3.8. Identification of additional influenza A minimal peptide-specific BCRs from a dataset of BCRs from an individual pre- and post-influenza vaccination To investigate whether we could identify additional BCRs specific for the influenza A minimal peptide VPNGTLVKTITNDQI by simply selecting for the conserved V gene segments and light chain CDR3 characteristics of known binders,we searched a dataset of ~100,000 BCR sequences from an individual pre- and post-influenza vaccination (3.61, 3.62) for those with IgHV5-51, IgKV4-1, IgKJ2, and a light chain CDR3 similar to our example set. We identified five such class-switched BCRs and found that two (flu_c2760 and flu_c4582) bound the influenza A minimal peptide and exhibited similar high-resolution footprints to those of the peptide tetramer-sorted BCRs (Fig. 3.4B, fig. S3.6B, fig. S3.7; table S3.8). Flu_c2760 and flu_c4582were the only BCRs of the set of five with long heavy chain CDR3s (23 and 22 AA, respectively). Flu_c2760 was flu-responsive, expanding in the post-influenza vaccination repertoire (3.62). Identification of EBV minimal peptide-specific memory B cells We also performed two rounds of sorting for the EBV minimal peptide PPSTSSKLRPRWTFT, the first using 11 donor blood collars and the second using a new set of 11 donor blood collars. During the first round, we obtained 54 paired heavy and light chain sequences and chose 6 to clone and recombinantly express. Using dot blot and VirScan with the saturating mutagenesis public epitope library, we validated that one antibody, EBV_c186, was specific for the PPSTSSKLRPRWTFT peptide (Fig. 3.4C and fig. S3.6A). During the second round of sorting, we obtained 94 paired heavy and light chain sequences. We initially chose 10 of these to clone and recombinantly express, including six antibodies from three different donors that featured the same V gene segment usage as EBV_c186, namely, IgHV1-46 / IgLV3-10. These six antibodies: EBV_c9, EBV_c19, EBV_c61, EBV_c149, EBV_c101, and EBV_c150, validated as specific for PPSTSSKLRPRWTFT, as well as three other antibodies: EBV_c40, which featured a IgHV1-46 / IgLV3-1 gene segment combination, and EBV_c3 and EBV_c120, which were from the same B cell precursor and donor and featured a IgHV1-8 / IgLV1-51 combination (Fig. 3.4C, fig. S3.6D). Because most of the antibodies from this initial selection validated as specific for the PPSTSSKLRPRWTFT peptide, we selected 23 Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 23 additional antibodies from the same dataset to clone and recombinantly express, including every remaining antibody that shared the IgHV1-46 / IgLV3-1 or IgHV1-8 / IgLV1-51 combinations, several antibodies with either IgHV1-46 or IgLV3-10 but not both, and several antibodies chosen at random. Using dot blot and VirScan with the saturating mutagenesis public epitope library, we validated that nine of these 23 antibodies were specific for the PPSTSSKLRPRWTFT peptide: three antibodies: EBV_c63, EBV_c83, and EBV_c124, which featured IgHV1-46; four antibodies: EBV_c10, EBV_c77, EBV_c127, and EBV_c138, which were from the same B cell precursor and donor as EBV_c3 and EBV_c120and featured the IgHV1-8 / IgLV1-51 combination; one antibody, EBV_c57, which featured a IgHV1-8 / IgLV1-47 combination; and one additional antibody: EBV_c98, which featured a IgHV3-30 / IgKV2-30 combination (Fig. 3.4C, fig. S3.6E). Of all the cells we sorted and sequenced for the EBV minimal peptide, the IgHV1-46 / IgLV3-10 combination was only found among the antibodies we ultimately validated as specific for the peptide and one other antibody, EBV_c626, which exhibited weak binding to the peptide by dot blot (fig. S3.6F). The full list of antibodies we tested for specificity for the PPSTSSKLRPRWTFT is provided in table S3.9. Recombinant expression of antibodies Heavy and light chain sequences of BCRs of interest were synthesized as gene fragments (IDT, Gene Universal) and cloned into human IgG, IgK, and IgL expression vectors (Human IgG Vector Set, Sigma) according to the manufacturer’s instructions. Recombinant expression of antibodies was performed as previously described using the Expi293 Expression System Kit (Thermo Fisher Scientific) (3.63) or the ExpiCHO Expression System (Thermo Fisher Scientific). Filtered cell culture supernatant was used for dot blot and VirScan characterization. Antibody purification was performed using NAb Protein A Plus Spin Column (Thermo Fisher Scientific) according to the manufacturer’s instructions. Dot blot Nitrocellulose membrane was marked with pencil to indicate the regions where the peptides would be spotted. Four micrograms of recombinant peptide was spotted onto each marked region of the nitrocellulose membrane, then the membrane was cut into strips and blocked in Tris Buffered Saline with Tween 20 (Cell Signaling Technology) and 5% bovine serum albumin (TBST 5% BSA) for 30 min at room temperature. The membrane was next incubated in cell culture supernatant from recombinant antibody expression diluted 1:20 in TBST 5% BSA for 1 hour at room temperature. Nitrocellulose was washed three times with TBST for 5 min per wash, then incubated with anti-human IgG-HRP (Southern Biotech) diluted 1:500 in TBST 5% BSA for 1 hour at room temperature. Nitrocellulose was washed three times with TBST for 5min per wash, then once with TBS for 5 min. Nitrocellulose was incubated with SuperSignal West Femto Maximum Sensitivity Substrate enhanced chemiluminescent substrate (Thermo Fisher Scientific) according to the manufacturer’s instructions and chemiluminescent signals were captured using a CCD camera-based imager. Influenza A hemagglutinin ELISA Expression and purification of hemagglutinin (HA) trimers and antibodies: Cloning, expression, and purification of HA ectodomain trimers were carried out as previously Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 24 described (3.64). Briefly, HAs were expressed as soluble trimers with a C-term foldon trimerization domain and a 6x-His tag (HA1-HA2-glyserthrombin-glyser-foldon- glyser-HHHHHH) (3.65). HA1-HA2 sequences (residue 1-504 H3 numbering) from A/ Aichi/02/1968 Aichi (H3), A/ Shanghai/1/2013 SH13 (H7), A/Jiangxi-Donghu/346/2013 JX346 (H10), A/swine/HuBei/06/2009 HB09 (H4N1), A/California/04/2009 CA09 (H1N1), A/Vietnam/1203/2004 Viet04 (H5N1), A/Japan/305/1957 JP57 (H2N2) and A/guinea fowl/ Hong Kong/1999 WF10 (H9N2) were cloned into a pTT5 expression vector containing the C-term tags. Soluble HA trimers were expressed by transient transfection in the Expi293 Expression System and purified from supernatants by Ni-NTA chromatography followed by size exclusion chromatography (SEC). Purified soluble HA trimers were stored in TBS buffer (20 mM Tris pH 8.0, 150 mM NaCl and 0.02% sodium azide) at 4°C and used for at least one month. FI6 (3.65) antibody was produced as described for hemagglutinin trimers. Briefly, antibody genes were cloned into the pTT5 expression vector, expressed by transient transfection in Expi293T cells, and purified from cell supernatants using Ni-NTA chromatography followed by SEC. Proteins were stored in TBS buffer at 4°C and used for up to 6 months. Influenza A hemagglutinin ELISA: The binding of the recombinant IgG influenza A minimal peptide-specific antibodies to hemagglutinin ectodomain trimers was measured by ELISA as described previously (3.64). Briefly 384-well plates (Sigma) were coated with 10 mg/ml of recombinant HAs in 0.1 M NaHC03 pH 9.8 and stored overnight at 4°C. Plates were then blocked with 3% BSA in TBS-T (TBS with 0.1% Tween 20) for 1 hour at room temperature. Plates were washed with TBS-T after each step. Antibodies were diluted to 10 mg/ml with TBS-T and serially diluted fourfold for a total of eight dilutions and added for 3 hours at room temperature. Secondary HRP-conjugated goat anti-human IgG (Southern Biotech) diluted 1:10,000 was added for 1 hour at room temperature. The SuperSignal ELISA Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific) was used to develop the plates, and the plates were read at 425nm. ELISA binding curves were plotted using Graphpad Prism 8.3. In vitro influenza A neutralization assay Neutralization assays were conducted as described previously (3.64). Briefly, live PB1flankeGFP virus were used for BSL 2 strains A/Aichi/02/1968 (X31; H3N2), A/ California/04/2009 (CA09; H1N1) (3.66), using reagents kindly provided by Dr. J. Bloom (Fred Hutchinson). MDCK cells were used for live neutralization assays. Purified antibodies were diluted to 50 mg/ml except for flu_c3 and flu_c286, which were at 30 mg/ml and 6.7 mg/ml, respectively, and then serially diluted fivefold for a total of eight dilutions. Pseudovirus assays (3.67) were conducted for BSL 3 strains A/Shanghai/1/2013 (SH13; H7), A/Jiangxi-Donghu/346/2013 (JX346;H10) and A/Vietnam/1203/2004 (Viet04; H5). TZM-bl cells were used for the pseudoviral neutralization assays. Antibodies were diluted to 50 mg/ml except for flu_c3 and flu_c286, which were at 30mg/ml and 6.7 mg/ml, respectively, and then serially diluted fourfold for a total of eight dilutions. Percent neutralization was then determined for each concentration of antibody and plotted using Graphpad Prism 8.3. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 25 Live cell immunofluorescence The EBV+ Burkitt lymphoma cell line P3HR-1-ZHT was used for EBV lytic reactivation (3.68). P3HR1-ZHT cells stably express a conditional BZLF1 allele (BZLF1-HT), in which the ligand binding domain of a modified estrogen receptor responsive to 4HT is fused to the BZLF1 C terminus. In the absence of 4HT, BZLF1-HT is retained in the cytosol and is destabilized. 4HT addition stabilizes BZLF1-HT and drives its rapid nuclear translocation which leads to lytic reactivation. P3HR1-ZHT cells (1x106 cells/ml) were reactivated with 4HT (1 mM) for 24 hours. For live cell staining, 1x106 cells were washed twice with live cell staining buffer (PBS with 1 mM EDTA and 0.5% BSA), followed by incubation with primary antibodies at 2 μg/ml for 30 min on ice. Cells were then washed with the staining buffer twice and subsequently stained with Alexa Fluor 488-conjugated anti-human IgG secondary antibody (Jackson Immunoresearch) at 6 μg/ml and Cy5-conjugated anti-gp350 mouse monoclonal antibody 72A1 (gift of Dr. E. Kieff) at 2 μg/ml for 30 min on ice. Labeled cells were washed three more times with FACS buffer and the nucleus was stained by Hoechst 33258 at 1 μg/ml prior to confocal microscopy using the LSM800 system with an Apochromat 63x/1.4 Oil DIC M27 objective lens (Zeiss). Images were analyzed with Adobe Photoshop. SARS-CoV-2 spike ELISA Ninety-six-well maxisorp ELISA plates (Thermo Fisher Scientific) were coated with 2 mg of full-length SARS-CoV-2 (2019-nCoV) spike protein (S, gift of B. Chen, Ragon Institute of MGH, MIT and Harvard) or S2 (purchased from GenScript) in 35 ml of PBS overnight at 4°C. After discarding coating buffer, ELISA plates were blocked with 50 ml of PBS with 3% BSA at room temperature for 2 hours. During the time of incubation, monoclonal antibodies were serially diluted twofold using 1 mg/ml as a starting concentration in 1% BSA prepared in PBS with 0.03% Tween 20. The anti-peanut PT275-H40 monoclonal antibody (gift of Dr. D. R. Wesemann) was used as a negative control. After blocking, the solution was discarded and ELISA plates were washed once in PBS containing 0.05% Tween 20. Monoclonal antibody dilutions were transferred into the plates in duplicates along with standards and incubated overnight at room temperature. Afterward, primary antibody solution was decanted and plates were washed thrice in PBS containing 0.05% Tween 20. Secondary antibody solutions of anti-human IgG alkaline phosphatase (AP) (Southern Biotech) diluted 1:2000 in PBS containing 1% BSA and 0.03% Tween 20were added to each plate at 30 ml per well. Plates were incubated for 90 min at room temperature and then washed thrice. Alkaline phosphatase substrate p nitrophenyl phosphate tablets (Sigma, St. Louis, MO) were dissolved in 0.1 M glycine, pH 10.4, with 1 mM MgCl2 and 1 mM ZnCl2, pH 10.4, to a concentration of 1.6 mg/ml. One hundred microliters of this development/substrate solution was then added to each well in a 96-well plate. Plates were kept in the dark and allowed to develop for 2 hours prior to reading. Absorbance at 405 nm was measured using a microplate reader (Biotek Synergy H1). All samples were run in duplicate wells. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 26 In vitro SARS-CoV-2 neutralization assay Cell culture: NR-596VeroE6 cells (BEI Resources) were maintained in Dulbecco’s modified Eagle medium (DMEM) (Gibco™) with the following additives: 10% heat inactivated fetal bovine serum (Gibco™), GlutaMAX (Gibco™), non-essential amino acids (Gibco™), and sodium pyruvate (Gibco™). The day prior to the assay, 8.0x105 VeroE6 cells were seeded per well of a six-well plate in 2 ml of media. Virus propagation: Authentic SARS-CoV-2 viruses were propagated as previously described (3.46) from passage 4 SARS-CoV-2 USA-WA1/2020 (3.69). Viral plaque reduction neutralization assay: All viral infection quantification assays were performed at biosafety level 4 (BSL-4) at the National Emerging Infectious Disease Laboratories (NEIDL). In brief, an Avicel plaque reduction assay was used to quantify plaques as follows: Antibody samples were serially diluted in Dulbecco’s Phosphate Buffered Saline (DPBS) (Gibco™) using half-log or threefold dilutions. Each dilution was incubated at 37°C and 5% CO2 for 1 hour with 1000 plaque forming units/ml (PFU/ml) of SARS-CoV-2 (isolate USA-WA1/2020 – described above). The maintenance media was then removed from each plate and 200 ml of each inoculum dilution was added to confluent monolayers of NR-596 Vero E6 cells (including 1000 PFU/ml SARS-CoV-2 incubated with DPBS as a positive control and a mock DPBS negative control) in triplicate and incubated for 1 hour at 37°C/5% CO2 with gentle rocking every 10–15 min to prevent monolayer drying. The overlay was prepared by mixing by inversion Avicel 591 overlay (DuPont Nutrition & Biosciences, Wilmington, DE) and 2X Modified Eagle Medium (Temin’s modification, Gibco™) supplemented with 2X antibiotic-antimycotic (Gibco™), 2X GlutaMAX (Gibco™) and 10% fetal bovine serum (Gibco™) in a 1:1 ratio. After 1 hour, 2 ml of overlay was added to each well and the plates were incubated for 48 hours at 37°C/5% CO2. Six-well plates were then fixed using 10% neutral buffered formalin prior to removal from BSL-4 space. The fixed plates were then stained with 0.2% aqueous Gentian Violet (RICCA Chemicals, Arlington, TX) in 10% neutral buffered formalin for 30 min, followed by rinsing and plaque counting. The half maximal inhibitory concentrations (IC50) were calculated using GraphPad Prism 8. PDB analysis to identify GRAB motifs Nucleotide sequences of human and mouse V gene segments were obtained from IMGT V- QUEST (3.49, 3.50) and translated to AA sequences. The mouse V gene segment sequences were from multiple strains of mice. We used the RCSB PDB Search API (Number of Distinct Protein Entities ≥3, Sequence = the AA sequence of the relevant V gene segment with an identity cutoff of 93%) to identify relevant structures of Ab–Ag complexes involving the V gene segment. The following json query was used: “query”: { “type”: “group”, “logical_operator”: “and”, “nodes”: [ Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 27 { “type”: “group”, “logical_operator”: “and”, “nodes”: [ { “type”: “terminal”, “service”: “text”, “parameters”: { “operator”: “greater_or_equal”, “negation”: False, “value”: 3, “attribute”: “rcsb_entry_info.polymer_entity_ count_protein” } } ] }, { “type”: “terminal”, “service”: “sequence”, “parameters”: { “evalue_cutoff”: 0.1, “identity_cutoff”: 0.93, “target”: “pdb_protein_sequence”, “value”: seq } }, ] } Next, for each PDB structure, we loaded the Chothia numbered version of the PDB file from SAbDab (3.70, 3.71) into UCSF Chimera (3.72) and used the command findclash (with parameters VDW overlap −1, hbond 0.0) to detect residues in the antigen chain that interacted with the relevant antibody heavy or light chain. For Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 28 each PDB structure, we generated a rowby-row summary of each antigen residue (in column“Antigen_Residue”) that interacted with the relevant antibody chain, the antibody residues it interacted with (in column “Interacts_With”), and the subset of those antibody residues whose side chain atoms interacted with the antigen AA (in column “Interacts_With_SC”). Next, using SAbPred ANARCI (3.73, 3.74), we annotated the subset of the residues in the column “Interacts_With_SC” thatwere germline encoded (in column “Interacts_With_Germline_SC”). We also included counts of the number of interacting residues (in columns “Num_Interactions”, “Num_SC_Interactions”, “Num_Germline_SC_Interactions”). As a quality control, we entered the relevant antibody heavy or light sequence into IgBlast and annotated the top hit, % alignment, and Evalue (in columns “IgBlast_TopHit”, “Alignment”, and “Evalue”, respectively), and performed a sequence match between the top hit and the original V gene segment query name (in column “Check”) in order to confirm the identity of the V gene segment. We also annotated whether the antigen was a protein or peptide in the column “Antigen_Type”. Then, we collected the output for all the structures with the relevant V gene segment into a file (“_antigenSummary.csv”). We performed a similar analysis for all the antibody residues that interacted with the antigen and collected the output into a file (“_antibody- Summary.csv”). These files are available on the Harvard Dataverse, doi: 10.7910/DVN/ WZCLMB; 10.7910/DVN/DXWJ2Y. Finally, we searched each “_antigen.csv” V gene segment file for recurrent interactions found in multiple unique Ab-Ag structures between certain germline-encoded residues in the antibody and a given AA (or a small set of biochemically similar AAs) in the antigen. These recurrent interactions were considered GRAB motif interactions. V gene segments for which there was a single example of an Ab-Ag interaction that strongly resembled a GRAB motif interaction present in another V gene segment were also considered to harbor GRAB motifs. AlphaFold2 predictions BCR sequences containing IgLV3-16, IgLV3-27, IgLV1-47, and IgHV3-1106 were identified from our single-cell sequencing datasets. If residues of the predicted GRAB motifs were somatically hypermutated, we reverted them to the germline-encoded residues at those positions. We could not find BCRs containing IgLV3-22 or IgLV3-902 in our datasets, so we substituted these germline V gene segments within the sequence of the BCR containing IgLV3-27. A (GGS)x12 linker was inserted between the heavy variable and light variable region of each BCR sequence, and the structures of these “fusion proteins” predicted using AlphaFold2 (3.33, 3.34). Quantification and Statistical Analysis Statistical details of experiments can be found in the figure legends and materials and methods. Data analysis was performed in Python, R, Graphpad Prism, and Excel. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 29 Acknowledgments: We thank L. Vandenberghe, N. Zabaleta, D. Barouch, and A. Chandrashekar for providing sera samples; S. Quake, D. Grishin, J. Bloom, C. O’Leary, C. Glassman, and M. Steffen for helpful discussions; A. Kohlgruber for designing initial versions of the schematics in Fig. 3.1B, Fig. 3.2A, and Fig. 3.4A; C. Araneo and the Immunology Flow Cytometry Facility and A. Ciulla Hurst and the BioPolymers Genomics Core Facility at Harvard Medical School for supporting this work; J. Q. O’Brien for contributing to revision experiments; and R. E. Shrock for reviewing the manuscript. Access to COVID-19 patient samples was facilitated by the MassCPR. We thank the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco, with support from NIH P41-GM103311 for help with molecular graphics. Fig. 3.8 and the print page summary figure were created with BioRender.com. Funding: This research was supported by the SARS-CoV-2 Viral Variants Program and the Value of Vaccine Research Network to S.J.E.; the MassPCR and the NIH 1P01AI165072 to S.J.E., A.G., and D.R.W.; and the NIH 1R01AI129784 to P.J.B. E.L.S. and T.K. were supported by the NSF Graduate Research Fellows Program. R.T.T. is supported by a Pemberton-Trinity Fellowship and a Sir Henry Wellcome Fellowship (201387/Z/ 16/Z). E.L.M. is supported by a Jane Coffin Childs Postdoctoral Fellowship. R.G. is supported by an NIH Pathway to Independence Award (K99/R00) K99DE031016. D.R.W. is also supported by AI139538, AI169619, AI170715, and AI170580. B.E.G. is supported by a Burroughs Wellcome Career Award in Medical Sciences. S.J.E. is an Investigator with the Howard Hughes Medical Institute. REFERENCES 31. Xu JL, Davis MM, Diversity in the CDR3 region of V(H) is sufficient formost antibody specificities. Immunity 13, name 37–45 (2000). doi: 10.1016/S1074-7613(00)00006-6 [PubMed: 10933393] 32. McLean GR et al. Recognition of human cytomegalovirus by human primary immunoglobulins identifies an innate foundation to an adaptive immune response. J. Immunol 174, 4768–4778 (2005). doi: 10.4049/jimmunol.174.8.4768 [PubMed: 15814702] 33. Zhang M, Zharikova D, Mozdzanowska K, Otvos L, Gerhard W, Fine specificity and sequence of antibodies directed against the ectodomain of matrix protein 2 of influenza A virus. Mol. Immunol 43, 2195–2206 (2006). doi: 10.1016/j.molimm.2005.12.015 [PubMed: 16472860] 34. Thomson CA et al. Germline V-genes sculpt the binding site of a family of antibodies neutralizing human cytomegalovirus. EMBO J. 27, 2592–2602 (2008). doi: 10.1038/emboj.2008.179 [PubMed: 18772881] 35. Robert R et al. Restricted V gene usage and VH/VL pairing of mouse humoral response against the N-terminal immunodominant epitope of the amyloid b peptide. Mol. Immunol 48, 59–72 (2010). doi: 10.1016/j.molimm.2010.09.012 [PubMed: 20970857] 36. Gorny MK et al. Human anti-V3 HIV-1 monoclonal antibodies encoded by the VH5–51/VL lambda genes define a conserved antigenic structure. PLOS ONE 6, e27780 (2011). doi: 10.1371/ journal.pone.0027780 [PubMed: 22164215] 37. West AP Jr., Diskin R, Nussenzweig MC, Bjorkman PJ, Structural basis for germ-line gene usage of a potent class of antibodies targeting the CD4-binding site of HIV-1 gp120. Proc. Natl. Acad. Sci. U.S.A 109, E2083–E2090 (2012). doi: 10.1073/pnas.1208984109 [PubMed: 22745174] 38. Robbiani DF et al. Recurrent Potent Human Neutralizing Antibodies to Zika Virus in Brazil and Mexico. Cell 169, 597–609.e11 (2017). doi: 10.1016/j.cell.2017.04.024 [PubMed: 28475892] 39. Chan K-W et al. Structural Comparison of Human Anti-HIV-1 gp120 V3 Monoclonal Antibodies of the Same Gene Usage Induced by Vaccination and Chronic Infection. J. Virol 92, e00641–18 (2018). doi: 10.1128/JVI.00641-18 [PubMed: 29997214] 310. Yuan M et al. Structural basis of a shared antibody response to SARS-CoV-2. Science 369, 1119–1123 (2020). doi: 10.1126/science.abd2321 [PubMed: 32661058] 311. Barnes CO et al. Structures of Human Antibodies Bound to SARS-CoV-2 Spike Reveal Common Epitopes and Recurrent Features of Antibodies. Cell 182, 828–842.e16 (2020). doi: 10.1016/ j.cell.2020.06.025 [PubMed: 32645326] Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 30 312. Voss WN et al. Prevalent, protective, and convergent IgG recognition of SARS-CoV-2 non- RBD spike epitopes. Science 372, 1108–1112 (2021). doi: 10.1126/science.abg5268 [PubMed: 33947773] 313. Mukhamedova M et al. Vaccination with prefusion-stabilized respiratory syncytial virus fusion protein induces genetically and antigenically diverse antibody responses. Immunity 54, 769– 780.e6 (2021). doi: 10.1016/j.immuni.2021.03.004 [PubMed: 33823129] 314. Xu GJ et al. Viral immunology. Comprehensive serological profiling of human populations using a synthetic human virome. Science 348, aaa0698 (2015). doi: 10.1126/science.aaa0698 [PubMed: 26045439] 315. Mina MJ et al. Measles virus infection diminishes preexisting antibodies that offer protection from other pathogens. Science 366, 599–606 (2019). doi: 10.1126/science.aay6485 [PubMed: 31672891] 316. Shrock EL, Shrock CL, Elledge SJ, VirScan: High-throughput Profiling of Antiviral Antibody Epitopes. Biol. Protoc 12, e4464 (2022). doi: 10.1126/science.aay6485 317. Larman HB et al. Autoantigen discovery with a synthetic human peptidome. Nat. Biotechnol 29, 535–541 (2011). doi: 10.1038/nbt.1856 [PubMed: 21602805] 318. Shrock E et al. Viral epitope profiling of COVID-19 patients reveals cross-reactivity and correlates of severity. Science (2020). doi: 10.1126/science.abd4250 319. Chen G et al. High-resolution epitope mapping by AllerScan reveals relationships between IgE and IgG repertoires during peanut oral immunotherapy. Cell Rep. Med 2, 100410 (2021b). doi: 10.1016/j.xcrm.2021.100410 [PubMed: 34755130] 320. Monaco DR et al. Profiling serum antibodies with a pan allergen phage library identifies key wheat allergy epitopes. Nat. Commun 12, 379 (2021). doi: 10.1038/s41467-020-20622-1 [PubMed: 33483508] 321. Vogl T et al. Population-wide diversity and stability of serum antibody epitope repertoires against human microbiota. Nat. Med 27, 1442–1450 (2021). doi: 10.1038/s41591-021-01409-3 [PubMed: 34282338] 322. Barber KW, Shrock E, Elledge SJ, CRISPR-based peptide library display and programmable microarray self-assembly for rapid quantitative protein binding assays. Mol. Cell 81, 3650– 3658.e5 (2021). doi: 10.1016/j.molcel.2021.07.027 [PubMed: 34390675] 323. Henikoff S, Henikoff JG, Amino acid substitution matrices from protein blocks. Proc. Natl. Acad. Sci. U.S.A 89, 10915–10919 (1992). doi: 10.1073/pnas.89.22.10915 [PubMed: 1438297] 324. Gray VE, Hause RJ, Fowler DM, Analysis of Large-Scale Mutagenesis Data To Assess the Impact of Single Amino Acid Substitutions. Genetics 207, 53–61 (2017). doi: 10.1534/ genetics.117.300064 [PubMed: 28751422] 325. Zemlin M et al. Expressed murine and human CDR-H3 intervals of equal length exhibit distinct repertoires that differ in their amino acid composition and predicted range of structures. J. Mol. Biol 334, 733–749 (2003). doi: 10.1016/j.jmb.2003.10.007 [PubMed: 14636599] 326. DeKosky BJ et al. Large-scale sequence and structural comparisons of human naive and antigen- experienced antibody repertoires. Proc. Natl. Acad. Sci. U.S.A 113, E2636–E2645 (2016). doi: 10.1073/pnas.1525510113 [PubMed: 27114511] 327. Low JS et al. ACE2-binding exposes the SARS-CoV-2 fusion peptide to broadly neutralizing coronavirus antibodies. Science 377, eabq2679 (2022). doi: 10.1126/science.abq2679 [PubMed: 35857703] 328. Dacon C et al. Broadly neutralizing antibodies target the coronavirus fusion peptide. Science 377, eabq3773 (2022). doi: 10.1126/science.abq3773 [PubMed: 35857439] 329. Sun X et al. Neutralization mechanism of a human antibody with pan-coronavirus reactivity including SARS-CoV-2. Nat. Microbiol 7, 1063–1074 (2022). doi: 10.1038/s41564-022-01155-3 [PubMed: 35773398] 330. Berman HM et al. The Protein Data Bank. Nucleic Acids Res. 28, 235–242 (2000). doi: 10.1093/nar/28.1.235 [PubMed: 10592235] 331. Berman H, Henrick K, Nakamura H, Announcing the worldwide Protein Data Bank. Nat. Struct. Biol 10, 980 (2003). doi: 10.1038/nsb1203-980 [PubMed: 14634627] Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 31 332. Burley SK et al. RCSB Protein Data Bank: Powerful new tools for exploring 3D structures of biological macromolecules for basic and applied research and education in fundamental biology, biomedicine, biotechnology, bioengineering and energy sciences. Nucleic Acids Res. 49, D437– D451 (2021). doi: 10.1093/nar/gkaa1038 [PubMed: 33211854] 333. Jumper J et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). doi: 10.1038/s41586-021-03819-2 [PubMed: 34265844] 334. Varadi M et al. AlphaFold Protein Structure Database: Massively expanding the structural coverage of proteinsequence space with high-accuracy models. Nucleic Acids Res. 50, D439– D444 (2022). doi: 10.1093/nar/gkab1061 [PubMed: 34791371] 335. Kortemme T, Kim DE, Baker D, Computational alanine scanning of protein-protein interfaces. Sci. STKE 2004, pl2 (2004). doi: 10.1126/stke.2192004pl2 [PubMed: 14872095] 336. Kortemme T, Baker D, A simple physical model for binding energy hot spots in protein-protein complexes. Proc. Natl. Acad. Sci. U.S.A 99, 14116–14121 (2002). doi: 10.1073/pnas.202485799 [PubMed: 12381794] 337. Robbiani DF et al. Convergent antibody responses to SARSCoV-2 in convalescent individuals. Nature 584, 437–442 (2020). doi: 10.1038/s41586-020-2456-9 [PubMed: 32555388] 338. Kreer C et al. Longitudinal Isolation of Potent Near-Germline SARS-CoV-2-Neutralizing Antibodies from COVID-19 Patients. Cell 182, 843–854.e12 (2020). doi: 10.1016/ j.cell.2020.06.044 [PubMed: 32673567] 339. Chandrashekar A et al. SARS-CoV-2 infection protects against rechallenge in rhesus macaques. Science 369, 812–817 (2020). doi: 10.1126/science.abc4776 [PubMed: 32434946] 340. Zabaleta N et al. An AAV-based, room-temperature-stable, single-dose COVID-19 vaccine provides durable immunogenicity and protection in non-human primates. Cell Host Microbe 29, 1437–1453.e8 (2021). doi: 10.1016/j.chom.2021.08.002 [PubMed: 34428428] 341. Angeletti D, Yewdell JW, Understanding and Manipulating Viral Immunity: Antibody Immunodominance Enters Center Stage. Trends Immunol. 39, 549–561 (2018). doi: 10.1016/ j.it.2018.04.008 [PubMed: 29789196] 342. Akram A, Inman RD, Immunodominance: A pivotal principle in host response to viral infections. Clin. Immunol 143, 99–115 (2012). doi: 10.1016/j.clim.2012.01.015 [PubMed: 22391152] 343. Abbott RK et al. Precursor Frequency and Affinity Determine B Cell Competitive Fitness in Germinal Centers, Tested with Germline-Targeting HIV Vaccine Immunogens. Immunity 48, 133–146.e6 (2018). doi: 10.1016/j.immuni.2017.11.023 [PubMed: 29287996] 344. Clark SA et al. SARS-CoV-2 evolution in an immunocompromised host reveals shared neutralization escape mechanisms. Cell 184, 2605–2617.e18 (2021). doi: 10.1016/ j.cell.2021.03.027 [PubMed: 33831372] 345. Zhang Q et al. Potent and protective IGHV3–53/3–66 public antibodies and their shared escape mutant on the spike of SARS-CoV-2. Nat. Commun 12, 4210 (2021). doi: 10.1038/ s41467-021-24514-w [PubMed: 34244522] 346. Tong P et al. Memory B cell repertoire for recognition of evolving SARS-CoV-2 spike. Cell 184, 4969–4980.e15 (2021). doi: 10.1016/j.cell.2021.07.025 [PubMed: 34332650] 347. Li Q et al. The Impact of Mutations in SARS-CoV-2 Spike on Viral Infectivity and Antigenicity. Cell 182, 1284–1294.e9 (2020). doi: 10.1016/j.cell.2020.07.012 [PubMed: 32730807] 348. Cao Y et al. Omicron escapes the majority of existing SARS-CoV-2 neutralizing antibodies. Nature 602, 657–663 (2022). doi: 10.1038/s41586-021-04385-3 [PubMed: 35016194] 349. Brochet X, Lefranc M-P, Giudicelli V, IMGT/V-QUEST: The highly customized and integrated system for IG and TR standardized V-J and V-D-J sequence analysis. Nucleic Acids Res. 36, W503–8 (2008). doi: 10.1093/nar/gkn316 [PubMed: 18503082] 350. Giudicelli V, Brochet X, Lefranc M-P, IMGT/V-QUEST: IMGT standardized analysis of the immunoglobulin (IG) and T cell receptor (TR) nucleotide sequences. Cold Spring Harb. Protoc 2011, pdb.prot5633 (2011). doi: 10.1101/pdb.prot5633 351. Viana R et al. Rapid epidemic expansion of the SARS-CoV-2 Omicron variant in southern Africa. Nature 603, 679–686 (2022). [PubMed: 35042229] Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 32 352. Dommer A et al. #COVIDisAirborne: AI-Enabled Multiscale Computational Microscopy of Delta SARS-CoV-2 in a Respiratory Aerosol. bioRxiv 2021.11.12.468428 [Preprint] (2021); doi: 10.1101/2021.11.12.468428 353. Symmes BA, Stefanski AL, Magin CM, Evans CM, Role of mucins in lung homeostasis: Regulated expression and biosynthesis in health and disease. Biochem. Soc. Trans 46, 707–719 (2018). doi: 10.1042/BST20170455 [PubMed: 29802217] 354. Akbar R et al. A compact vocabulary of paratope-epitope interactions enables predictability of antibody-antigen binding. Cell Rep. 34, 108856 (2021). doi: 10.1016/j.celrep.2021.108856 [PubMed: 33730590] 355. Langmead B, Trapnell C, Pop M, Salzberg SL, Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 10, R25 (2009). doi: 10.1186/ gb-2009-10-3-r25 [PubMed: 19261174] 356. Kodituwakku AP, Jessup C, Zola H, Roberton DM, Isolation of antigen-specific B cells. Immunol. Cell Biol 81, 163–170 (2003). doi: 10.1046/j.1440-1711.2003.01152.x [PubMed: 12752679] 357. Stoeckius M et al. Simultaneous epitope and transcriptome measurement in single cells. Nat. Methods 14, 865–868 (2017). doi: 10.1038/nmeth.4380 [PubMed: 28759029] 358. Robinson J, Malik A, Parham P, Bodmer JG, Marsh SG, IMGT/HLA database—A sequence database for the human major histocompatibility complex. Tissue Antigens 55, 280–287 (2000). doi: 10.1034/j.1399-0039.2000.550314.x [PubMed: 10777106] 359. Robinson J et al. The IPD and IMGT/HLA database: Allele variant databases. Nucleic Acids Res 43, D423–D431 (2015). doi: 10.1093/nar/gku1161 [PubMed: 25414341] 360. Ye J, Ma N, Madden TL, Ostell JM, IgBLAST: an immunoglobulin variable domain sequence analysis tool. Nucleic Acids Res. 41, W34–40 (2013). [PubMed: 23671333] 361. Horns F, Vollmers C, Dekker CL, Quake SR, Signatures of selection in the human antibody repertoire: Selective sweeps, competing subclones, and neutral drift. Proc. Natl. Acad. Sci. U.S.A 116, 1261–1266 (2019). doi: 10.1073/pnas.1814213116 [PubMed: 30622180] 362. Horns F, Dekker CL, Quake SR, Memory B Cell Activation, Broad Anti-influenza Antibodies, and Bystander Activation Revealed by Single-Cell Transcriptomics. Cell Rep. 30, 905–913.e6 (2020). doi: 10.1016/j.celrep.2019.12.063 [PubMed: 31968262] 363. Vazquez-Lombardi R et al. Transient expression of human antibodies in mammalian cells. Nat. Protoc 13, 99–117 (2018). doi: 10.1038/nprot.2017.126 [PubMed: 29240734] 364. Cohen AA et al. Construction, characterization, and immunization of nanoparticles that display a diverse array of influenza HA trimers. PLOS ONE 16, e0247963 (2021). doi: 10.1371/ journal.pone.0247963 [PubMed: 33661993] 365. Ekiert DC et al. A highly conserved neutralizing epitope on group 2 influenza A viruses. Science 333, 843–850 (2011). doi: 10.1126/science.1204839 [PubMed: 21737702] 366. Bloom JD, Gong LI, Baltimore D, Permissive secondary mutations enable the evolution of influenza oseltamivir resistance. Science 328, 1272–1275 (2010). doi: 10.1126/science.1187816 [PubMed: 20522774] 367. Temperton NJ et al. A sensitive retroviral pseudotype assay for influenza H5N1- neutralizing antibodies. Influenza Other Respir. Viruses 1, 105–112 (2007). doi: 10.1111/ j.1750-2659.2007.00016.x [PubMed: 19453415] 368. Guo R et al. MYC Controls the Epstein-Barr Virus Lytic Switch. Mol. Cell 78, 653–669.e8 (2020). doi: 10.1016/j.molcel.2020.03.025 [PubMed: 32315601] 369. Harcourt J et al. Isolation and characterization of SARS-CoV-2 from the first US COVID-19 patient. bioRxiv (2020). doi: 10.1101/2020.03.02.972935 370. Dunbar J et al. SAbDab: The structural antibody database. Nucleic Acids Res 42, D1140–D1146 (2014). doi: 10.1093/nar/gkt1043 [PubMed: 24214988] 371. Schneider C, Raybould MIJ, Deane CM, SAbDab in the age of biotherapeutics: Updates including SAbDab-nano, the nanobody structure tracker. Nucleic Acids Res. 50, D1368–D1372 (2022). doi: 10.1093/nar/gkab1050 [PubMed: 34986602] 372. Pettersen EF et al. UCSF Chimera—A visualization system for exploratory research and analysis. J. Comput. Chem 25, 1605–1612 (2004). doi: 10.1002/jcc.20084 [PubMed: 15264254] Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 33 373. Dunbar J et al. SAbPred: A structure-based antibody prediction server. Nucleic Acids Res 44, W474–8 (2016). doi: 10.1093/nar/gkw361 [PubMed: 27131379] 374. Dunbar J, Deane CM, ANARCI: Antigen receptor numbering and receptor classification. Bioinformatics 32, 298–300 (2016b). doi: 10.1093/bioinformatics/btv552 [PubMed: 26424857] 375. Baum A, Fulton BO, Wloga E, Copin R, Pascal KE, Russo V, Giordano S, Lanza K, Negron N, Ni M, Wei Y, Atwal GS, Murphy AJ, Stahl N, Yancopoulos GD, Kyratsous CA, Antibody cocktail to SARS-CoV-2 spike protein prevents rapid mutational escape seen with individual antibodies. Science 369, 1014–1018 (2020). doi:10.1126/science.abd0831 [PubMed: 32540904] Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 34 Fig. 3.1. Recurrent antibody responses to public epitopes are a general feature of humoral immunity and antibodies that recognize a given public epitope have biased light chain isotype usage. (A) Percentage of individuals seropositive for a given virus who exhibit antibody responses to each publicly recognized 56-AA peptide. For each viral species, publicly recognized 56-AApeptides are arranged in descending order by their value on the x axis. Arrows indicate the peptides detailed in (C). (B) Schematic representation of the VirScan assay using the public epitope truncation and alanine scan library (IP, immunoprecipitation). (C) Antibody responses to public epitopes from EBV envelope glycoprotein gp350(top) and Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 35 rhinovirus B genome polyprotein (bottom) as characterized by VirScan using the kappa and lambda isotype-specific IP protocol. The subset of serum samples exhibiting antibody responses to at least one peptide from the 56-AA region are shown; data are the mean of two technical replicates. AAs are identified by their standard one-letter abbreviations. (D) Number of kappa and lambda IP samples that recognize short truncations (15, 20, 25, and 30 AAs in length) of publicly recognized 56-AA peptides from 62 viral species. Observed data (left) and randomly permuted data (right) are shown. Axes are capped at 40 because most data points fell within this range. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 36 Fig. 3.2. Antibodies recognizing public epitopes often exhibit similar high-resolution footprints. (A) Construction of the saturating mutagenesis public epitope library and VirScan screening procedure. Throughout, AAs are identified by their standard one-letter abbreviations. (B) Representative high-resolution antibody footprints from different individuals for minimal peptides from human cytomegalovirus (left), human herpesvirus 2 (center), and hepatitis C virus (right). The sequence of the minimal peptide is shown on the x axis and the AA substitution on the y axis. The darker the blue color, the greater the disruption of antibody binding. Colored shapes to the left side of each high-resolution footprint are Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 37 coordinated with colored arrows in (C) to indicate the location of the high-resolution footprints and their technical replicates within the clustered heatmaps. (C) Clustered heatmaps illustrating the similarities between all high-resolution antibody footprints for the three example minimal peptides in (B); these were obtained by calculating the pairwise Pearson correlation coefficients between the enrichment matrices (see methods) of each serum sample that recognized the minimal peptide. Colored arrows are coordinated with colored shapes in (B) to indicate the location of the high-resolution footprints from (B) and their technical replicates within the clustered heatmaps. (D) Histogram depicting the fraction of individuals whose kappa or, separately, lambda antibody responses to a given minimal peptide recognized the consensus pattern of critical residues or only differed by one position. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 38 Fig. 3.3. Critical residues of public epitopes have a distinctive AA composition, including profound enrichment of lysine at the borders of lambda public epitopes. (A) AA composition of critical residues of lambda and kappa public epitopes, relative to the entire human virome library. P values (binomial test) are listed below. Those that remain below the significance threshold of 0.05 after correcting for multiple hypothesis testing with the Benjamini–Hochberg false discovery rate method are indicated with asterisks in the bar chart (P < 0.05, P < 10–6, P < 10–9,*********P < 10–30) and colored in red (enriched AAs) or blue (depleted AAs). (B) Diagram indicating the positions of border and interior critical residues in a representative high-resolution antibody footprint. (C) AA composition of critical residues found at border or interior positions of kappa and lambda public epitopes. P values (binomial test) as in (A)are listed at the bottom of the figure. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 39 Fig. 3.4. BCRs specific for example public epitopes share conserved gene segment usage but not heavy chain CDR3 sequences. (A) Schematic representation of the workflow to isolate public epitope-specific B cells. PBMCs were isolated from multiple healthy donors and pooled. Switched(IgG+ and IgA+) memory B cells were then purified by magnetic-activated cell sorting, split into aliquots to be labeled with customized CITE-Seq antibody barcodes, and then stained with fluorescent peptide tetramers. Fluorescent cells were isolated by FACS and analyzed by single-cell BCR sequencing. (B to D) Sequence characteristics of BCRs validated to bind three example Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 40 minimal peptides. Consensus critical residues within the minimal peptides are shown in bold and conserved gene segments and CDR3 sequences are shown in red. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 41 Fig. 3.5. Many antibody V gene segments feature GRAB motifs that recurrently recognize specific AAs in antigens. (A) An aspartic acid germline-encoded at position 51 of several lambda V gene segments drives specificity for border lysines in epitopes. Differences in frequencies (lambda-kappa) with which each position of the light chain contacts lysine residues in the antigen in human lambda (n = 297) and kappa Ab-Ag complexes (n = 631) in the PDB. (B and C) The indicated mutations were introduced into a panel of EBV minimal peptide-specific antibodies and the impacts on antigen binding assayed by dot blot. The epitopes of antibodies shown in red include border lysines, whereas the epitopes of those shown in blue do not. EBV and influenza A minimal peptides (146786 and 72153 in table S3.3, respectively) were spotted on the nitrocellulose membranes; flu_c326 served as a control. (D and E) Representative GRAB motif interactions for the indicated human V gene segments. The antigen residue is shown in magenta and all antibody residues whose sidechains interact Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 42 with the antigen residue are shown in cyan (FR, framework region). Summary tables (right) show the fraction of unique PDB Ab-Ag structures with the relevant V gene segment that feature the GRAB motif interaction. Throughout, PDB structures are visualized with UCSF Chimera (3.72). (F) Images of the other seven unique Ab-Ag structures involving IgHV5-51 that feature the GRAB motif interaction, labeled as in (D) and (E). Throughout, all residue positions follow the Chothia antibody numbering system. References for PDB Ab-Ag structures are provided in table S3.10. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 43 Fig. 3.6. The IgHV5-51 GRAB motif mediates antibody recognition of an influenza A public epitope. (A and B) Mutation of IgHV5-51 GRAB motif residues in influenza A public epitope- specific antibodies severely weakens recognition of the epitope. The indicated mutations were introduced into two influenza A public epitope-specific antibodies, flu_c504 (A) and flu_c3 (B) and the impact on binding assayed by dot blot. Influenza A and EBV minimal peptides (146786 and 72153 in table S3.3, respectively) were spotted onto the nitrocellulose membranes; the EBV_c186 antibody served as a control. (C and D) High-resolution Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 44 footprints for the monoclonal antibodies from (A and B). Heatmaps are labeled as described in Fig. 3.2B. Note that the heatmaps do not depict absolute enrichment values, but rather the relative enrichment of the wild-type peptide compared with the mutant peptide for a given sample. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 45 Fig. 3.7. Public epitopes are largely species-specific, consistent with only partially overlapping sets of GRAB motifs. (A) Antibody responses to peptides fromSARS-CoV-2 spike in SARS-CoV-2-infected humans (n = 30) (3.18), SARS-CoV-2–infected NHPs (n = 9) (3.39), and C57BL/6 mice immunized with an AAV vector carrying stabilized prefusion SARS-CoV-2 spike (n = 8) (3.40). Each row represents a unique individual and each column represents a peptide tile. Darker colors indicate greater enrichment (Z-score) of a peptide in a given sample. Colored arrows are coordinated with colored shapes in (B) to show the positions of select public Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 46 epitopes within spike. (B) Representative high-resolution footprints for minimal public epitope peptides recognized by one or more of the indicated species. Heatmaps are labeled as in Fig. 3.2B. Colored shapes are coordinated with colored arrows in (A)to show the positions of these public epitopes within spike. (C and D) Venn diagrams depicting the number of public epitopes recognized by one or more of the indicated species. (C) shows the number of publicly recognized minimal peptides, whereas (D) shows the number of precise antibody footprints (these were only considered to be shared if different species recognized the same pattern of critical residues). (E to H) Representative GRAB motif interactions for the indicated mouse V gene segments (table S3.13). Mouse GRAB motifs for which we found analogous human GRAB motifs are shown in (E) to (G). Those for which we did not find analogous human GRAB motifs are shown in (H). Images are labeled as in Fig. 3.5,D and E. (I) Summary table showing the fraction of unique PDB Ab-Ag structures with the relevant V gene segment that feature the GRAB motif interaction. Science. Author manuscript; available in PMC 2023 June 16. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Shrock et al. Page 47 Fig. 3.8. Proposed model for development of recurrent antibody responses to public epitopes through recognition by GRAB motifs. The B cell repertoire has a relatively low precursor frequency of BCRs with a specific heavy chain CDR3 (HCDR3) sequence, but a relatively high precursor frequency of BCRs with a certain combination of V gene segments (e.g., IgHV5-51 and IgKV4-1). If GRAB motifs within these V gene segments are sufficient to bind a certain epitope, this can lead to a public antibody response. By contrast, if a specific HCDR3 sequence is required to bind an epitope, this will likely lead to a private antibody response. Science. Author manuscript; available in PMC 2023 June 16.
10.1128_mbio.01039-23	\| Bacteriology \| Research Article Structural disruption of Ntox15 nuclease effector domains by immunity proteins protects against type VI secretion system intoxication in Bacteroidales Dustin E. Bosch,1 Romina Abbasian,1 Bishal Parajuli,1 S. Brook Peterson,2,3 Joseph D. Mougous2,3,4 AUTHOR AFFILIATIONS See affiliation list on p. 13. ABSTRACT Bacteroidales use type VI secretion systems (T6SS) to competitively colonize and persist in the colon. We identify a horizontally transferred T6SS with Ntox15 family nuclease effector (Tde1) that mediates interbacterial antagonism among Bacteroidales, including several derived from a single human donor. Expression of cognate (Tdi1) or orphan immunity proteins in acquired interbacterial defense systems protects against Tde1-dependent attack. We find that immunity protein interaction induces a large effector conformational change in Tde nucleases, disrupting the active site and altering the DNA-binding site. Crystallographic snapshots of isolated Tde1, the Tde1/Tdi1 complex, and homologs from Phocaeicola vulgatus (Tde2/Tdi2) illustrate a conserved mechanism of immunity inserting into the central core of Tde, splitting the nuclease fold into two subdomains. The Tde/Tdi interface and immunity mechanism are distinct from all other polymorphic toxin–immunity interactions of known structure. Bacteroidales abundance has been linked to inflammatory bowel disease activity in prior studies, and we demonstrate that Tde and T6SS structural genes are each enriched in fecal metagenomes from ulcerative colitis subjects. Genetically mobile Tde1-encoding T6SS in Bacteroidales mediate competitive growth and may be involved in inflammatory bowel disease. Broad immunity is conferred by Tdi1 homologs through a fold-disrupting mechanism unique among polymorphic effector–immunity pairs of known structure. IMPORTANCE Bacteroidales are related to inflammatory bowel disease severity and progression. We identify type VI secretion system (T6SS) nuclease effectors (Tde) which are enriched in ulcerative colitis and horizontally transferred on mobile genetic elements. Tde-encoding T6SSs mediate interbacterial competition. Orphan and cognate immunity proteins (Tdi) prevent intoxication by multiple Tde through a new mechanism among polymorphic toxin systems. Tdi inserts into the effector central core, splitting Ntox15 into two subdomains and disrupting the active site. This mechanism may allow for evolution ary diversification of the Tde/Tdi interface as observed in colicin nuclease–immunity interactions, promoting broad neutralization of Tde by orphan Tdi. Tde-dependent T6SS interbacterial antagonism may contribute to Bacteroidales diversity in the context of ulcerative colitis. KEYWORDS microbiome, Bacteroides, type VI secretion system, inflammatory bowel disease, structural biology T he Bacteroidota phylum is a major component of the healthy intestinal microbiome community. Specific taxa within this phylum, and their relative abundances have been linked to diverse diseases including components of the metabolic syndrome (1–3), viral infection (4), and colorectal carcinogenesis (5). Members of the Bacteroidales order may also have a role in severity and progression of inflammatory bowel disease (IBD) (6). Editor Karina B. Xavier, Instituto Gulbenkian de Ciência, Oeiras, Portugal Address correspondence to Dustin E. Bosch, dustin- [email protected]. The authors declare no conflict of interest. See the funding table on p. 14. Received 25 April 2023 Accepted 3 May 2023 Published 22 June 2023 Copyright © 2023 Bosch et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International license. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 1 Research Article mBio IBD includes both Crohn’s disease (CD) and ulcerative colitis (UC), related diseases with distinct pathophysiology. Crohn’s disease is characterized by “full-thickness” inflammation extending through all layers at any location along the gastrointestinal tract. In contrast, the inflammation of UC is confined to the superficial layers of the colon. Active inflammation in Crohn’s disease correlates to Phocaeicola vulgatus abundance, while reduced Bacteroides spp. were observed in UC patients with diarrhea and rectal bleeding (7, 8). Several Bacteroidales were among the taxa with greatest fluctuation over time in a large longitudinal IBD microbiome study, suggesting dynamic re-organization of their niche(s) during disease development (9). In summary, correlative human studies suggest that alterations among specific Bacteroidaceae may contribute IBD progres sion, and patterns differ between Crohn’s disease and UC. The longitudinal changes in Bacteroidales abundance observed in IBD may be influenced by competitive interactions (10, 11). Bacteroidales and other commensals in the intestinal microbiome engage in contact-dependent interbacterial antagonism using toxin secretion systems (11, 12). Type VI secretion system (T6SS) gene clusters encode at least 13 structural proteins that assemble into a needle-like apparatus for delivery of effectors (toxins) into neighboring bacteria (13). A contractile sheath composed of TssB and TssC propels an inner tubu lar structure composed of hexameric hemolysin co-regulatory protein (Hcp) through multi-protein baseplate and membrane complexes. Hcp and the T6SS tip structure are secreted into recipient periplasm and/or cytoplasm, carrying payloads of effectors (toxins) that promote cell death (14). T6SSiii gene clusters found in Bacteroidales are distantly related to model systems in Pseudomonadota (T6SSi), encoding nine shared core structural proteins and five core proteins restricted to Bacteroidales (15). Three prototypic T6SS genetic architectures have been described in Bacteroidales, two found on mobile elements a third largely restricted to Bacteroides fragilis (15). Conjugative transfer of mobile GA1 and GA2 type T6SS among Bacteroidales has been observed within the intestinal microbiome (16, 17). Bacteroides spp. utilize these T6SS to antago nize non-immune Bacteroidales (12, 18) and establish and maintain colonization (11). T6SS effector delivery frequently involves direct interaction with Hcp or the tip structure (19, 20). However, some effector domains are translationally fused to secre ted core components (21, 22). For example, pyocin and colicin type DNAse domain fusions with Hcp mediate T6SS-dependent antagonism in Pseudomonadota (21). T6SS effectors employ a striking array of activities to disrupt essential biologic functions, spanning enzymatic degradation of key small molecules, post-translational modification of essential proteins, and disruption of membrane and peptidoglycan layer barriers (14). Effectors with novel toxin 15 (Ntox15) DNAse domains, also known as toxin_43 domains, degrade recipient genomic DNA (23). Ntox15 effectors in the soil bacterium Agrobacterium tumefaciens, T6SS DNase effectors (Tde1-2), mediate competition in planta. Secretion of A. tumefaciens Tde effectors requires loading onto the C-termini of tip structure proteins with aid of adaptor/chaperone proteins (24, 25). Loading of Tde1/2 onto the tip structure is required for efficient sheath assembly and T6SS secretion (26). Cognate immunity proteins are encoded adjacent to T6SS effectors and neutralize their activity to prevent intoxication of self and kin (27). Immunity proteins usually prevent intoxication by direct occlusion of the effector active site (28, 29). Less common mechanisms include enzymatic antagonism of effector activities, e.g., reversal of toxin-mediated ADP ribosylation (30). Arrays of immunity proteins are also found encoded by gene clusters unassociated with a T6SS apparatus, termed “orphan” immunity proteins. These AIDs are frequently on mobile genetic elements and hori zontally transferred to confer protection from type VI attack, impacting competitive colonization among Bacteroidales (27). Bacteroidales T6SS have been implicated in mouse models of infectious colitis. Commensal B. fragilis strains use T6SS to competitively exclude pathogenic enterotoxin- producing strains and protect against colitis (10). We hypothesize that T6SS effector- mediated competitive colonization underlies associations of Bacteroidales with IBD July/August Volume 14 Issue 4 10.1128/mbio.01039-23 2 Research Article mBio severity and progression (6, 8). In this study, we show that T6SS loci encoding Tde family nuclease effectors are specifically enriched in ulcerative colitis metagenomes compared to Crohn’s disease and healthy controls. We also show that immunity against Tde-medi ated attack occurs by structural disruption of the effector domain, a mechanism unique among polymorphic toxin–immunity pairs of known structure. RESULTS Bacteroidales T6SS, Ntox15 domains, and immunity proteins are enriched in ulcerative colitis fecal metagenomes Prior studies have implicated T6SS and specific effector–immunity pairs in enterotoxi genic B. fragilis colitis (10). Based on these data, we asked whether T6SSiii loci and particular effector types are enriched among bacterial communities of IBD patient fecal samples. We constructed hidden Markov models (HMM) for the conserved Bacteroi dales T6SS structural proteins, as well as ~150 Bacteroidales polymorphic toxin domain families and associated immunity proteins (31). These HMMs were applied to a large collection of publicly available shotgun metagenomic sequencing data from humans with inflammatory bowel disease and healthy controls (32). This Integrative Human Microbiome Project cohort included biweekly stool samples from 67 subjects with Crohn’s disease, 38 with UC, and 27 non-IBD controls (9). Strong correlation of HMM hits among the T6SS structural proteins was observed, as expected, because the correspond ing genes are co-inherited in T6SS loci (Fig. S1A). HMM hit quantities per reads, corrected for relative Bacteroidales abundance, of each T6SS structural protein were similar across metagenomes (Fig. S1C), except for TssH, a AAA family ATPase which was excluded from further analysis due to off-target HMM hits. T6SS structural genes were enriched in fecal metagenomes from UC patients compared to CD (Fig. 1A). The enrichment of T6SS structural gene hits in UC extends to comparison with non-IBD “healthy control” specimens and is not explained by differential relative Bacteroidales abundance (Fig. 1B). Among the ~150 polymorphic toxin domain HMMs, greatest enrichment in UC was for Ntox15 homologs (Fig. 1A). Ntox15 hits, corrected to Bacteroidales abundance were enriched in UC relative to CD and controls, while the associated immunity gene did not differ significantly across groups (Fig. 1C). There was relative enrichment of Ntox15 genes per T6SS structural gene (TssB) in ulcerative colitis samples compared to non-IBD controls, and relative depletion in Crohn’s disease (Fig. S1D; linear fit slope 1.0 [0.9–1.1] for UC, 0.5 [0.4–0.7] for non-IBD, 0.0 [-0.1–0.1] for CD). A subset of the metagenomic data analyzed were time course samples from individual subjects. Multivariate analysis indicated that T6SS hits per Bacteroidales abundance tended to increase over time in subjects with ulcerative colitis (Fig. S1B). We conclude that T6SSiii and Ntox15-encod ing genes are differentially abundant in the intestinal metagenomes of humans with inflammatory bowel disease, and all are enriched in UC. Bacteroidales from a single human intestinal community compete with T6SS encoding Tde nuclease effectors To identify strains for functional studies on Ntox15 effectors, we queried a human intestinal commensal bacteria collection with whole genome sequencing (34). Several Phocaeicola and Bacteroides strains contain nearly identical Ntox15-encoding T6SS of the GA2 type architecture (15). These strains were all isolated from a single human donor, and their T6SS loci are encoded with neighboring mobile genetic element-related genes, highly suggestive of horizontal transfer events. Selection for specific T6SS effector and immunity pairs has importance for competitive colonization and persistence in human gut metagenomes (11). This T6SS encodes several Hcp proteins, a completely conserved (100% amino acid identity) Hcp-effector fusion with C-terminal Ntox15 domain, and an adjacent putative cognate immunity protein (Fig. 1D). This multispecies effector– immunity pair is termed Tde1 and Tdi1 to conform with nomenclature in Agrobacterium (23). Each genome also encodes putative effectors with rearrangement hotspot (RHS) July/August Volume 14 Issue 4 10.1128/mbio.01039-23 3 Research Article mBio FIG 1 Ntox15 domains enriched in IBD metagenomes mediate T6SS-dependent interbacterial antagonism among Bacteroidales. Metagenomic sequencing reads with similarity to Bacteroidales T6SS, Ntox15 domains, and immunity proteins were detected with hidden Markov models [HMMer (33)]. (A) T6SSiii structural genes and Ntox15 domain homologs are enriched in fecal metagenomes from patients with UC compared to CD (32). False discovery rate adjustment for multiple comparisons was with the Benjamini–Hochberg method. (B, C) Aggregated T6SS structural genes and Ntox15 homologs, but not the associated immunity are enriched in UC over CD and non-IBD controls after correction for relative Bacteroidales abundance. P-value reflects Kruskal–Wallis test. (D) A gene structure diagram of a T6SS-encoding locus that is identical in several genetically diverse Bacteroidales isolates from a single human donor. In addition to other T6SS structural genes (gray), there are five hcp genes (blue), including one fused with a C-terminal Ntox15 domain (tde1, green) and an immediately adjacent immunity gene (tdi1, red). An HxxD motif is conserved at the putative active site, predicted to confer nuclease activity. (E) In competitive growth experiments with P. vulgatus ATCC 8482, deletion of tde1 and tdi1 from MSK 16.10 or MSK 16.2 confers reduced relative fitness. Effector/immunity deletion is also a competitive disadvantage relative to the isogenic parental strain. Thymidine kinase (tdk) is deleted to confer resistance to the selection agent floxuridine (FUdR). (F) tde1/tdi1 mediate competition between MSK 16.10 and MSK 16.2, isolates from a single human host. Statistical indicators reflect Student’s t-test: P < 0.01, * P < 0.001. (G) Tde1-dependent antagonism requires structural sheath proteins TssB and TssC. P-values reflect analysis of variance (ANOVA) tests for each recipient. domains adjacent to mobile element genes, which have predicted structural similarity to the Tre23 toxin of Photorhabdus laumondii (35). We hypothesized that Tde1 mediates interbacterial competition among Bacteroidales. Deletion of tde1 in two of these T6SSs, Phocaeicola vulgatus strains MSK 16.2 and MSK 16.10 (34), enhanced competitive survival of a recipient P. vulgatus strain ATCC 8482 that lacks immunity (Fig. 1E). Deletion of tde1 and tdi1 in MSK 16.10 also conferred a competitive disadvantage relative to the isogenic parent strain, indicating that tdi1 likely protects against kin intoxication (Fig. 1E). The competitive disadvantage of tde1 deletion could be explained by requirement of the Hcp domain for T6SS assembly, but the four other hcp-encoding genes may compensate. Horizontal transfer of this mobile T6SS suggested that Tde1 may mediate cell killing among strains from a single host’s microbiome. Indeed, there was tde1-dependent killing of MSK 16.10 by MSK 16.2 when tde1 and tdi1 were removed from the recipient (Fig. 1F). Antagonism of MSK16.10 by MSK 16.2 required assembly of the T6SS apparatus with sheath proteins TssB and TssC (Fig. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 4 Research Article mBio 1G). We conclude that Hcp-Ntox15 effectors mediate T6SS-dependent competition with non-immune Bacteroidales, including strains derived from a single host. Bacteroidales Tde effectors are magnesium dependent DNAses with a distinct α-helical fold To identify mechanisms of Ntox15 effector toxicity, we characterized the structure and enzymatic function of Tde1. The distantly homologous Ntox15 domain-containing effector Tde1 in A. tumefaciens exhibited DNAse activity in vitro and in cells (23). To examine enzymatic activity of Tde1 from P. vulgatus, we co-produced the Ntox15 domain (Tde1tox) in E. coli with Tdi1 to circumvent toxicity, separated it from immunity under denaturing conditions, and refolded it. Tde1tox exhibited DNAse activity on plasmid dsDNA, which was abrogated by mutation of the HxxD active site (H279A) and strongly inhibited by the presence of Tdi1 or chelation of divalent cations using EDTA (Fig. 2A). EDTA-mediated inhibition was reversed by addition of molar excess magnesium salts, but not other divalent cations (Fig. 2A). Slower migration of plasmid DNA in the presence of Tde1tox H279A, excess ZnCl2 or CaCl2 suggest protein binding and/or effects on supercoiling. Consistent with the toxin exhibiting non-specific DNAse activity, catalyt ically inactive Tde1tox H279A/D282A directly interacted with 30-nucleotide single- or double-stranded DNA oligomers of random sequence, with equilibrium binding affinities near 500 nM (Fig. 2B; Fig. S2B). DNA binding affinity may be impacted by the dual point mutations in the active site. FIG 2 The DNAse Tde1 adopts an α-helical predominant fold with HxxD motif active site. (A) Refolded Tde1 Ntox15 domain degraded plasmid dsDNA. Nuclease activity was impaired by mutation of the HxxD motif, addition of molar excess immunity protein, or chelation of divalent cations with EDTA. Tde1tox nuclease activity impairment by EDTA was reversed by addition of molar excess magnesium, but not zinc or calcium. (B) Tde1tox with active site mutations interacted with both double- and single-stranded biotinylated oligonucleotides of random sequence, measured with biolayer interferometry. (C) A crystal structure of catalytically inactive Tde1tox H279A/D282A domain (Table S1) was obtained by molecular replacement using an AlphaFold2 prediction (36). Tde1tox adopts a single domain fold with the predicted DNA binding surface (green). Mutation of key basic residues (green sticks) to alanine or acidic residues decreased DNA binding affinity (Fig. S2F). The active site corresponds to the HxxD motif (red) and contains a modeled sulfate anion, present due to crystallization is high concentrations of ammonium sulfate. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 5 Research Article mBio A structural model of Tde1tox H279A/D282A Ntox15 domain was obtained by X-ray crystallography with diffraction data extending to 2.9 Å resolution (Table S1; Fig. 2C). Although no close homologs of known structure were available, phases were solved by molecular replacement using an AlphaFold2 prediction model (Fig. S2C) (36). The AlphaFold2 prediction model was very similar to the experimental crystal structure; mean Cα r.m.s.d. among the eight monomers in the asymmetric unit was 1.0 Å (Fig. S2D). The Ntox15 domain adopts a globular fold which is predominantly α-helical, forming a short α-sheet between helices 5 and 6 immediately adjacent to the active site (Fig. 2C). A structural similarity search with DALI (37) revealed no close homologs within the PDB, including known nuclease structures (Z score 6.0 and Cα r.m.s.d 4.1 over 77 residues for the top hit, two pore calcium channel PDB id 6NQ1). A cavity adjacent to the mutated HxxD motif marks the active site. A sulfate ion is modeled within the active site, likely an artifact of crystallization in high concentration of ammonium sulfate. However, it may mimic accommodation of negatively charged moieties of the DNA substrate. The DNA binding site predicted with ProNA2020 (38) maps to helices 4, 6, and 7, adjacent to the active site (Fig. 2C). Coulombic surface rendering highlights relative positive surface charge surrounding the active site pocket, consistent with favorable electrostatics for interaction with negatively charged DNA (Fig. S2E). Point mutation of basic residues at the predicted DNA binding surface decreased dsDNA binding affinity (Fig. 2C; Fig. S2F). Charge reversal substitutions had greatest impact on DNA binding, supporting likely importance of electrostatic interactions. We conclude that Ntox15 domains adopt a globular fold, distinct from other nuclease families of known structure, with a structurally well-defined active site that mediates DNAse activity. Orphan tdi are frequent among human intestinal commensal bacteria Ntox15 domain and core T6SS protein-encoding sequences were both enriched in UC metagenomes while immunity-encoding sequences (tdi) were not (Fig. 1C), raising the possibility of widespread tdi genes outside of T6SS loci. We assessed the distribution of T6SS, Ntox15, and immunity protein encoding genes among a large collection of human intestinal commensal genomes (34) using BLAST (39) and the Tde1-related T6SS genes as queries (Fig. 3A). The core structural tssC gene was identified exclusively in Bacteroidota, reflecting substantial sequence-level dissimilarity of the Bacteroidales T6SSiii relative to Pseudomonadota. Ntox15 domain homologs were confidently identified (BLAST E-value < 10−10) in 14 Bacteroidota strains, all with GA2 T6SS architecture. In contrast, 120 strains encoded Tdi1 homologs, including all genetic architectures (Fig. 3A). Nine Bacteroidota shared a similar gene structure with the Tde1-associated system query (Fig. S3), having immediately adjacent Hcp-Ntox15 fusion and immunity proteins within the context of a GA2 T6SS structural gene cluster. More distantly related Ntox15 domain-contain ing proteins were encoded adjacent to Tdi1-like immunity proteins in five Firmicute genomes (Roseburia intestinalis and Tyzzerella nexilis). The genomic context and domain organization (e.g., an LXG domain fusion) suggest association with type VII secretion systems. Notably, most of the Tdi1 homolog encoding genes were found in organisms without a Tde1 homolog, raising consideration of widespread orphan immunity among intestinal commensal bacteria (Fig. 3A) (27). The order-of-magnitude higher frequency of tdi compared to tde is consistent with the higher median frequency of tdi homolog sequen ces in metagenomes (Fig. 1C) and suggests one explanation for lack of correlation between tdi and disease state. Genomic context within 5 kb of these immunity genes frequently contained other putative immunity genes, distinct in sequence and domain structure, as well as genes associated with mobile genetic elements (Fig. S3). These findings suggest that Tdi1 homolog genes are frequently found in arrays of diverse immunity genes associated with mobile genetic elements, compatible with acquired immune defense (AIDs) systems (27). July/August Volume 14 Issue 4 10.1128/mbio.01039-23 6 Research Article mBio FIG 3 Cognate and orphan immunity proteins protect against T6SS-mediated attack by inducing a conformational shift in Tde1 to disrupt the DNA binding and active sites. (A) Query of Tde1, Tdi1, and representative T6SS structural protein (TssC) against a collection of ~1,200 human intestinal commensal genomes (40) with BLAST revealed predominant distribution of homologs within Bacteroidota. TssC homologs from previously described genetic architectures (GA1-3) cluster together (15). Tde1, but not Tdi1 homologs are exclusively in GA2 T6SS. Several Firmicutes harbor tde/tdi pairs not associated with T6SS. Immunity encoding genes were more abundant than tde. (inset) A Venn diagram illustrates that all identified tde1 homologs were accompanied by tdi. tde/tdi pairs were associated with a T6SS apparatus in 9 Bacteroidota and 5 Firmicutes. However, tdi genes were more frequently encountered than tde in both phyla, indicating presence of orphan immunity genes. (B) Tde1tox • Tdi1 exhibited higher thermal stability (melting temperature 67°C) than either component alone (55–55.5°C) in SYBR orange thermal melt experiments. (C and D) Biolayer interferometry demonstrated comparable equilibrium binding affinities of Tde1tox for Tdi1, as well as two homologous orphan immunity proteins (KD 18–24 nM). (E) Expression of Tdi1, as well as two orphan immunity proteins from diverse Bacteroidota protect P. vulgatus ATCC 8482 against tde1-dependent attack by P. vulgatus MSK 16.10. (F) Crystal structures of two homologous Tdetox (blues) and Tdi (gray, tan) complexes demonstrate a splitting of Ntox15 into two subdomains. The subdomains are linked by the DNA binding site and the HxxD motif, which are partially disordered in the crystal structures (dotted lines). The predicted DNA binding site is green, and basic residues required for high affinity DNA interaction represented as sticks. There is high structural similarity among the homologs, indicating a conserved mode of interaction. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 7 Research Article mBio Cognate and orphan immunity proteins promiscuously engage Tde nucleases to protect against killing Frequent occurrence of Tdi homologs in AIDs suggests that orphan immunity toward Tde toxins is an important mechanism of competition among Bacteroidales. Bacteroidales orphan immunity and effector interactions have not been biochemically characterized previously. We first characterized Tde1tox H279A/D282A and Tdi1 binding with multiple biochemical platforms (Fig. 3). Tde1tox interaction with Tdi1 increases thermal stability (melting temperature 67°C versus 55.5°C, Fig. 3B). A Tde1tox/Tdi1 dissociation constant of 18 nM was measured by biolayer interferometry (BLI, Fig. 3C and D). Two putative orphan immunity proteins were selected for further study, based on their presence in several intestinal commensal bacterial genomes, and gene structures compatible with AIDs (Fig. S3). These two proteins, termed Tdi orphan A and B (TdioA and TdioB), share 61–65% sequence identify with Tdi1. Both orphan immunity proteins, recombinantly produced from E. coli, directly interacted with Tde1tox H279A/D282A (Fig. 3D). Affinities of TdioA and TdioB for Tde1tox (24 and 19 nM) were very similar to that of the cognate immunity Tdi1. Orphan immunity proteins co-expressed with the P. vulgatus dnLKV7 homolog Tde2tox in E. coli also formed a stable 1:1 complex, as detected with analytical gel filtration chromatography (Fig. S4). We next examined protective effects of orphan immunity genes in competitive growth experiments. Expression of Tdi1 from a chromosomally inserted transposon (pNBU2) in P. vulgatus ATCC 8482 markedly reduced tde1-dependent killing by MSK 16.10 (Fig. 3E). Similarly, TdioA and TdioB were highly protective. We conclude that orphan immunity proteins directly engage both Tde1 and Tde2 (Fig. 3B through D; Fig. S4). Orphan immunity proteins have high affinity for Tde1 and provide competitive growth advantage in co-culture with the Tde1-encoding strain P. vulgatus MSK 16.10. Immunity proteins disrupt nuclease activity by inserting into the nuclease central core: a new mechanism of polymorphic toxin immunity We next sought a structural explanation for how promiscuous neutralization of Tde effectors by diverse Tdi homologs is achieved. We therefore obtained crystal structures of the Tde1 Ntox15 domain in complex with Tdi1, as well as a homologous complex from P. vulgatus dnLKV7, Tde2tox and Tdi2 (Fig. 3F). The Tdetox/Tdi complex homologs exhibit very similar structure despite 51% sequence identity between the Ntox15 domains, indicating a conserved mode of effector–immunity interaction. Tdi1/2 have structural homology to the Ntox15-associated immunity protein from A. tumefaciens (Atu4351, PDB ID 6ITW), which has been crystallized in isolation (23). Tdi1 and Atu4351 align with a Cα r.m.s.d. of 1.2 Å (Fig. S2E), although Bacteroidales Tdi1 exhibits a slightly more compact overall structure with shortening of several loops (e.g., β8-α5). When bound to immunity proteins, Tde1tox and Tde2tox split into two subdomains (Fig. 3F). Forty percent (17 of 43) of immunity-contacting Tde1/2tox residues in the effector immunity structures form part of the central core in the globular Ntox15 domain alone structure, and many of these are highly conserved (Fig. S5). There is an ~32 amino acid region disordered in the crystal structure, corresponding to β2, α6, and the surrounding loops in the Tde1tox only structure. Notably, this disordered region contains part of the HxxD active site motif and most of the DNA binding site (Fig. S5). Superposition of the Tde1tox alone structure with the Tde1tox/Tdi1 complex indicates a conformational shift characterized by a hinge motion, as well as an ~180° relative rotation of the two Tde1tox subdomains (Fig. 4A). We conclude that Tdi immunity proteins induce a marked conformational shift in Tde effectors, driving a division into two subdomains with disruption of the enzymatic active site and DNA binding motif. Tdi1 and Tdi2 form extensive contacts with the conserved central cores of their Tdetox counterparts (Fig. 3F). Upon immunity interaction, Tde effectors undergo a dramatic conformational shift, highlighted by superposition of the Tde1tox alone and Tde1tox/Tdi1 complex structures (Fig. 4A). The immunity protein does not sterically occlude the active site, but rather splits the effector into subdomains and structurally distorts the active site, July/August Volume 14 Issue 4 10.1128/mbio.01039-23 8 Research Article mBio FIG 4 Effector fold disruption is a new immunity mechanism among polymorphic toxins. (A) The Tde1tox alone structure (red) is superimposed on the Tde1tox/ Tdi1 complex structure. Upon immunity binding, the split subdomains of Tde1tox undergo a relative ~90° hinge motion and ~180° rotation. The DNA binding site (including helix α6) and the active site (HxxD yellow) are disrupted by the conformational shift. (B) Solvation energy gains of effector/immunity interface formation as percentages of monomer solvation energy were calculated with PDBePISA (41). Included structural models with PDB accession and PubMed IDs are listed in Table S2. Tde1tox/orphan immunity calculations are derived from comparative homology models based on the Tde1tox/Tdi1 structure. (C) The “capping” mechanism with non-disruptive steric occlusion of the effector active site is typified by the Pseudomonas aeruginosa T6SS-assocated peptidoglycan hydrolase Tse1/Tsi1. (D) Several T6SS and other polymorphic toxin/immunity interactions involve insertion of the immunity protein into a pre-formed effector active site crevice (“plugging”), typified by P. aeruginosa (P)ppApp synthetase Tas1/immunity. A predicted model of Tas1 alone, supported by an experimental structure of homolog RelQ (not shown, PDB 5DEC), indicates lack of large conformational shift in the effector. (E) A structure of colicin E3 RNAse exhibits engagement of immunity at an “exosite” separate from the enzymatic active site (42). Unlike Tde1tox/Tdi1, large effector conformational shifts are not predicted. which is disordered in the crystal structures. Advances in deep learning have improved prediction accuracy for protein-protein interfaces (43), leading us to ask whether the Tde conformation shift mechanism of Tdi immunity is computationally predictable. However, AlphaFold-Multimer predicted Tde1-2tox/Tdi1-2 complexes inaccurately in the absence of an experimentally derived template structure (Fig. S6). The Tdetoxα4-α5 helices interface with Tdi is approximated by the models, but effector conformational shifts and the secondary immunity interface are not identified. Thus, the Tdi1 immunity mechanism differs from previous structural investigations of T6SS-related effector–immunity pairs July/August Volume 14 Issue 4 10.1128/mbio.01039-23 9 Research Article mBio and cannot be reliably predicted from primary sequences with current deep learning algorithms. To identify similar immunity mechanisms among polymorphic toxins, we compared the Tdetox/Tdi structure to all other polymorphic toxin–immunity pairs in the Protein Data Bank. The hydrophobic nature of Tde’s interactions with Tdi are reflected numerically in solvation energy calculations from the PDBePISA web server (41). Specifically, there is a relatively large solvation energy gain upon complex formation as compared to the Tdetox monomers alone (Fig. 4B). Comparative homology models of Tde1tox in complex with orphan immunity proteins, using the Tde1tox/Tdi1 crystal structure as a template, yielded similar solvation energy changes to the cognate immunity–effector pairs (44). As numeric markers of interface hydrophobicity, solvation energy gains were likewise calculated for each polymorphic toxin–immunity pair in the PDB (Fig. 4B). Most other effector–immunity interfaces cluster with relatively low solvation energy gains for both effector and immunity. Among the T6SS effector–immunity complexes, this pattern corresponds to immunity “capping” for steric occlusion of the effector active site, typified by the T6SS-associated Tse1/Tsi1 interaction in P. aeruginosa (Fig. 4C). Overlay of the Tse1 only structure (PDB 4EQ8) with the Tse1/Tsi1 complex (PDB 4EQA) demonstrates the absence of conformation shifts as found in Tde1 (Fig. 4A through C). A related mecha nism of immunity, “plugging” or insertion of the immunity into a preformed effector active site cleft is illustrated with the Tas1 and immunity complex structure (PDB 6OX6) from P. aeruginosa (Fig. 4D). In contrast with Tdetox/Tdi, interactions of this type uniformly occur at the active site and do not result in large conformational shifts. While a Tas1 only structure is not available, an AlphaFold2 predicted model and structural homolog RelQ from Bacillus subtilis (PDB 5DEC) exhibit similar conformations to the effector in complex with immunity and an open active site crevice (Fig. 4D) (45). Several effector–immunity interactions of this pattern produced relatively high immunity solvation energy gains (Fig. 4B). The E. coli colicin E3 ribonuclease and immunity interfaces (PDB 1E44, 1JCH) showed parallels to Tdetox/Tdi1 in having relatively high effector solvation energy gain calculations (Fig. 4B) and an immunity interface that does not overlap with the effec- tor active site (46) (Fig. 4E). Similar to other colicin nucleases, immunity is conferred by high-affinity interaction at an “exosite” (42, 47). A model of the isolated colicin E3 effector domain, predicted with AlphaFold2, shows a highly similar fold to the immunity complex, except for ~9 residues at the N-terminus. This region is predicted with low confidence in the isolated colicin E3 and assumes a short helix with extensive immunity contacts in the complex crystal structure (Fig. 4E). However, the marked conformational shift and central core interactions observed in Tde1/Tdi1 are lacking. We conclude that Tde conformational shift and active site disruption mediated by Tdi differs from previously described polymorphic toxin–immunity interactions. Immunity contacts with the effector central core are reflected in solvation energy calculations. In contrast to the predominant active site occlusion immunity mechanisms, Tdi inserts into the Tde central core, dividing the effector domain and disrupting the active site structure. DISCUSSION from a single human donor Our finding of essentially identical T6SS apparatus genes and Tde1–Tdi1 within is highly suggestive of diverse Bacteroidales intestinal micro recent horizontal gene transfer, possibly within the donor’s biome. Tde1-dependent competition among these strains implies selective pressure favoring acquisition of T6SS. Consistent with prior literature, we find T6SS gene clusters and acquired immune defense systems frequently associated with mobile genetic elements (16, 27). Active exchange and selection for genetic material relevant to T6SS-mediated attack supports previously described hypotheses that interbacterial competition among the Bacteroidota is an important determinant of the microbial community composition in individual hosts (11, 16). Polymorphic toxins have been implicated in virulence of certain pathogenic bacteria, with mechanisms including toxin delivery to host cells (48, 49). However, disease July/August Volume 14 Issue 4 10.1128/mbio.01039-23 10 Research Article mBio associations with human commensal bacterial polymorphic toxins have been less thoroughly explored. In one prior study, T6SSs of commensal B. fragilis strains were important for competitive exclusion of pathogenic enterotoxin-producing strains (10). In this study, we find enrichment of T6SS structural genes and Ntox15 domains in patients with ulcerative colitis, suggesting positive selection for this effector immunity pair. T6SSs with tde homologs are found in P. vulgatus, and we demonstrate tde-mediated antagonism among three intestinally derived strains. P. vulgatus abundance associates with IBD disease activity (7). Furthermore, colonization with some strains of P. vulgatus modulates inflammation severity in rodent colitis models, although none tested in these model studies are known to encode tde–tdi homologs (50). Bacteroidales T6SSs and Ntox15 effectors might contribute directly to the etiology of UC, or the disease process (inflammation, epithelial disruption, etc.) may favor Bacteroidales with T6SS and tde. The latter hypothesis is supported by significant increases in relative T6SS gene abundance in time course metagenomic data from subjects with UC. Interestingly, UC and Crohn’s disease metagenomes exhibited opposite patterns of Ntox15 gene abundance relative to structural T6SS genes. This pattern raises the possibility that encoding Ntox15 domains may be advantageous to bacteria in UC, but detrimental in Crohn’s disease. Alternatively, there may be differential abundances of Bacteroidales with different T6SSiii genetic architectures in the two disease states, which cannot be quantified with our HMM approach. The Tde–Tdi proteins investigated in our study bear distant homology to T6SS effector–immunity pairs in A. tumefaciens (23). Like A. tumefaciens Tde1, the Bacteroidales Ntox15 domain exhibits magnesium-dependent DNAse activity. These domains are likely toxic due to non-targeted degradation of DNA in recipient cells. Given the enzymatic similarity of the effectors and the structural similarity of the immunity proteins, the immunity mechanism is very likely conserved. Mechanisms of secretion of the Bacteroi dales Tde1/2 fused to Hcp are distinct from the non-covalent tip structure interactions described in Agrobacterium Tde1/2 (25, 26). The adaptor/chaperone proteins Tap-1 and Atu3641 required for Agrobacterium effector delivery are absent in Bacteroidales T6SS (25). Similarly, Bacteroidales Tde lack the N-terminal glycine zipper motif described as important for translocation of Agrobacterium Tde1 into recipient cells (51). Most T6SS immunity proteins of known structure prevent intoxication of self and kin by direct steric occlusion of the effector active site (30, 52, 53), although a subset of immunity proteins also counteract effector-mediated intoxication though enzymatic activity (30). In contrast, Tdi proteins in Bacteroidales induce a large conformational change in cognate effectors, splitting the globular fold into subdomains and structur ally disrupting the substrate binding and active sites. Possible mechanisms include an inherent conformational flexibility in Tde1 with selection of a two-subdomain confor mation for immunity interaction, or an induced fit model of interaction where initial contacts with Tdi promote separation of the two Tdetox subdomains. One possible consequence of the structural rearrangement induced in Tde could be increased efficiency of toxin destruction in the immune recipient cell. For example, Tdi insertion into the central core of Tde may facilitate proteolytic degradation of the effector. Several parallels can be drawn between Tde/Tdi and colicin nuclease and immun ity complexes. For example, colicins E3 and E9 engage immunity proteins at an “exosite” separate from the active site (54). The mechanism of immunity in these scenarios is thought to be steric and electrostatic repulsion of substrates (genomic DNA or the ribosome) (42, 55), in contrast to central core insertion and structural rearrangement of the active site seen in Tde/Tdi. Colicin nuclease immunity proteins are structurally diverse, and a prevailing hypothesis is that exosite interactions allow for evolutionary diversification at the interface, away from the conserved active site (56). Prevalent cross-reactivity of nuclease colicins and immunity proteins (55) also parallels the multi-effector interaction patterns of Tdi immunity proteins. The relatively broad specificity of Tdi immunity interactions with the central core of Tde may have evolved through exosite diversification as posited for colicin nuclease–immunity interactions. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 11 Research Article mBio Promiscuous binding of multiple Tde by a single Tdi may be more advantageous to recipient bacteria than highly specific Tde-directed interaction (i.e., 1:1 correspondence), and may contribute to the high frequency of orphan Tdi in human commensal genome collections. As a class of T6SS effector–immunity pairs important for competition among Bacteroidota, Tde nucleases are neutralized by unique mechanisms, including structural disruption of the active site and substrate binding surface by an immunity-induced large conformational shift. This novel immunity mechanism allows relatively broad neutraliza tion of multiple Ntox15 domains by a single immunity protein. Further study will be required to determine how Tde and Tdi influence Bacteroidales abudance in IBD and the detailed mechanisms by which Tdi insert into the central core of Tde. MATERIALS AND METHODS T6SS gene quantitation in human intestinal metagenomes See supplementary methods for detailed methods. Cloning, plasmids, and Bacteroidales genetics See supplementary methods for detailed methods. Competitive growth Bacteroidales were mixed to a final OD600 reading of 6.0 with 1:1 or 10:1 donor/recipient ratios and plated on BHIS with gentamycin (60 mg/mL) (57) for ~24 h at 37°C in an anerobic chamber (Anaerobe Systems, Morgan Hill, CA, USA). Bacteria were recovered in BHIS liquid media, serially diluted, and quantitatively cultured with and without 5-fluorodeoxyuridine selection. Recipient competitive indices were calculated from colony-forming units as (post-competition recipient/pre-competition recipient)/(post- competition donor/pre-competition donor). For competitive growth experiments with transposon-inserted immunity proteins, expression was induced (or mock in empty transposon controls) with anhydrotetracycline for 3 h prior to co-culture with cell– cell contact inducing conditions as above. All competitive growth experiments were performed with at least biological triplicates and at least two independently replicated experiments. Protein purification, crystallization, and structure determination See supplementary methods for protein purification and crystallization methods. See Table S1 for diffraction data and refinement statistics. Differential scanning fluorimetry Tde1tox H279A/D282A, Tdi1, or the Tde1tox/Tdi1 complex were mixed at 10 µM concentra tion with SYPRO Orange dye at 2× concentration in X1 buffer. Temperature was increased at 0.5°C intervals every 10 s in a CFX real-time PCR detection instrument (BioRAD) with detection of dye fluorescence. Melting temperatures were assigned at the fluorescence curve inflection point. All data shown represent at least triplicate experiments. Biolayer interferometry BLI experiments were conducted on an Octet Red96 instrument (Sartorius). Nucleic acid binding experiments were conducted with 30 base pair biotinylated synthetic oligonucleotides, immobilized on streptavidin biosensors. For Ntox15/immunity binding experiments, hexahistidine immunity proteins (5 mg/mL) were immobilized on NTA biosensors. Equilibrium binding dose–response curves were generated with varying concentrations of Tde1tox H279A/D282A, and additional mutations thereof, in Octet July/August Volume 14 Issue 4 10.1128/mbio.01039-23 12 Research Article mBio kinetics buffer (Sartorius). Association and dissociation intervals were 300 and 600 s, respectively. Affinity constants were determined by one site binding curve fitting of equilibrium binding data in Prism (GraphPad) after subtraction of non-specific binding to an irrelevant surface control (biotin only). All data shown represent at least triplicate experiments. Nuclease activity Plasmid DNA (2 µg of pcDNA3.1) was incubated at 37°C with Tde1tox or H279A mutant (1 µM), immunity protein (10 µM), EDTA (1 mM), and/or divalent cation and chloride salts (10 mM) as indicated in a final volume of 50 µL. Reactions were halted by addition of DNA electrophoresis loading dye, and nucleic acids assessed by 1% agarose electropho resis and ethidium bromide staining. Identification of T6SS, Ntox15, and immunity homologs See supplementary methods for detailed methods. Structural analysis and solvation energy calculations Polymorphic toxin and immunity protein structures were identified in the PDB using keyword searches and protein classification terms. Comparative homology models of Tde1tox with TdioA or TdioB were constructed with SWISS-MODEL using the Tde1tox/Tdi1 crystal structure template (44). All structures were reviewed manually in Chimera (58) to identify effector–immunity interfaces and classify immunity mechanism. Effector and immunity solvation energy gain calculations were performed with PDBePISA (https:// www.ebi.ac.uk/pdbe/pisa/) (41). ACKNOWLEDGMENTS We thank Dr. Eric Pamer and Emily Waligurski at the Duchossois Family Institute for access to an intestinal commensal bacteria strain collection, Dr. Ben Ross for an insightful critique of the manuscript, and the University of Iowa Protein and Crystallography Core for access to BLI instrumentation. This work was supported by the NIH, K08 AI159619 (Bosch DE). This work was supported by the NIH (AI080609 to JDM, etc.). J.D.M. is an HHMI Investigator and is supported by the Lynn M. and Michael D. Garvey Endowed Chair. The Berkeley Center for Structural Biology is supported in part by the Howard Hughes Medical Institute. The Advanced Light Source is a Department of Energy Office of Science User Facility under contract DEAC02-05CH11231. The Pilatus detector on 5.0.1. was funded under NIH grant S10OD021832. The ALS-ENABLE beamlines are supported in part by the National Institutes of Health, National Institute of General Medical Sciences, grant P30 GM124169. D.E.B. – conceptualization, methodology, investigation, resources,data curation, writing, visualization, supervision, funding acquisition; R.A. – investigation, writing; B.P. – investigation, writing; S.B.P. – conceptualization, writing, supervision; J.D.M. – conceptu alization, resources, writing, supervision, funding acquisition. All authors declare no competing interests. AUTHOR AFFILIATIONS 1Department of Pathology, Carver College of Medicine, University of Iowa, Iowa City, Iowa, USA 2Department of Microbiology, University of Washington School of Medicine, Seattle, Washington, USA 3Howard Hughes Medical Institute, University of Washington, Seattle, Washington, USA 4Microbial Interactions and Microbiome Center, University of Washington, Seattle, Washington, USA July/August Volume 14 Issue 4 10.1128/mbio.01039-23 13 mBio Research Article AUTHOR ORCIDs Dustin E. Bosch http://orcid.org/0000-0002-7430-2939 FUNDING Funder HHS \| NIH \| National Institute of Allergy and Infectious Diseases (NIAID) HHS \| NIH \| National Institute of Allergy and Infectious Diseases (NIAID) AUTHOR CONTRIBUTIONS Grant(s) Author(s) K08 AI159619 Dustin E. Bosch AI080609 Joseph D. Mougous Dustin E. Bosch, Conceptualization, Investigation, Supervision, Writing – original draft, Writing – review and editing, Data curation, Methodology, Resources, Visualization \| Romina Abbasian, Investigation, Writing – original draft, Writing – review and editing \| Bishal Parajuli, Investigation, Writing – original draft, Writing – review and editing \| S. Brook Peterson, Conceptualization, Supervision, Writing – original draft, Writing – review and editing \| Joseph D. Mougous, Conceptualization, Funding acquisition, Resources, Supervision, Writing – original draft, Writing – review and editing DIRECT CONTRIBUTION This article is a direct contribution from Joseph Mougous, a Fellow of the American Academy of Microbiology, who arranged for and secured reviews by Arne Rietsch, Case Western Reserve University, and Eric Cascales, Centre national de la recherche scientifi- que, Aix-Marseille Université. DATA AVAILABILITY Crystallographic data have been deposited to the RCSB protein data bank (accessions 8FZY, 8FZZ, and 8G0K). Metagenomic sequencing data were previously published (32) and are publicly available at the NCBI sequence read archive (BioProject PRJNA398089). Plasmids and bacterial strains generated in the study are listed in Table S3 and will be available upon reasonable request to the corresponding author. ADDITIONAL FILES The following material is available online. Supplemental Material Supplemental Material (mBio01039-23-s0001.pdf). Supplemental methods, Figures S1-S6, and Tables S1-S3. REFERENCES 1. 2. 3. Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE, Sogin ML, Jones WJ, Roe BA, Affourtit JP, Egholm M, Henrissat B, Heath AC, Knight R, Gordon JI. 2009. A core gut microbiome in obese and lean twins. Nature 457:480–484. https://doi.org/10.1038/nature07540 Pedersen HK, Gudmundsdottir V, Nielsen HB, Hyotylainen T, Nielsen T, Jensen BAH, Forslund K, Hildebrand F, Prifti E, Falony G, Le Chatelier E, Levenez F, Doré J, Mattila I, Plichta DR, Pöhö P, Hellgren LI, Arumugam M, Sunagawa S, Vieira-Silva S, Jørgensen T, Holm JB, Trošt K, Kristiansen K, Brix S, Raes J, Wang J, Hansen T, Bork P, Brunak S, Oresic M, Ehrlich SD, Pedersen O, MetaHIT Consortium. 2016. Human gut microbes impact host serum metabolome and insulin sensitivity. Nature 535:376–381. https://doi.org/10.1038/nature18646 Vieira-Silva S, Falony G, Belda E, Nielsen T, Aron-Wisnewsky J, Chakaroun R, Forslund SK, Assmann K, Valles-Colomer M, Nguyen TTD, Proost S, 4. 5. Prifti E, Tremaroli V, Pons N, Le Chatelier E, Andreelli F, Bastard J-P, Coelho LP, Galleron N, Hansen TH, Hulot J-S, Lewinter C, Pedersen HK, Quinquis B, Rouault C, Roume H, Salem J-E, Søndertoft NB, Touch S, MetaCardis Consortium, Dumas M-E, Ehrlich SD, Galan P, Gøtze JP, Hansen T, Holst JJ, Køber L, Letunic I, Nielsen J, Oppert J-M, Stumvoll M, Vestergaard H, Zucker J-D, Bork P, Pedersen O, Bäckhed F, Clément K, Raes J. 2020. Statin therapy is associated with lower prevalence of gut Microbiota Dysbiosis. Nature 581:310–315. https://doi.org/10.1038/s41586-020-2269-x Stefan KL, Kim MV, Iwasaki A, Kasper DL. 2020. Commensal microbiota modulation of natural resistance to virus infection. Cell 183:1312–1324. https://doi.org/10.1016/j.cell.2020.10.047 Dejea CM, Fathi P, Craig JM, Boleij A, Taddese R, Geis AL, Wu X, DeStefano Shields CE, Hechenbleikner EM, Huso DL, Anders RA, Giardiello FM, Wick EC, Wang H, Wu S, Pardoll DM, Housseau F, Sears CL. 2018. Patients with familial adenomatous polyposis harbor colonic biofilms containing July/August Volume 14 Issue 4 10.1128/mbio.01039-23 14 Research Article mBio 6. 7. 8. 9. tumorigenic bacteria. Science 359:592–597. https://doi.org/10.1126/ science.aah3648 Delday M, Mulder I, Logan ET, Grant G. 2019. Bacteroides thetaiotaomi cron ameliorates colon inflammation in preclinical models of Crohn's disease. Inflamm Bowel Dis 25:85–96. https://doi.org/10.1093/ibd/izy281 Gonzalez CG, Mills RH, Zhu Q, Sauceda C, Knight R, Dulai PS, Gonzalez DJ. 2022. Location-specific signatures of Crohn's disease at a multi-Omics scale. Microbiome 10:133. https://doi.org/10.1186/s40168-022-01331-x Hellmann J, Ta A, Ollberding NJ, Bezold R, Lake K, Jackson K, Dirksing K, Bonkowski E, Haslam DB, Denson LA. 2023. Patient-reported outcomes independent of correlate with microbial community composition Mucosal inflammation in pediatric inflammatory bowel disease. Inflamm Bowel Dis 29:286–296. https://doi.org/10.1093/ibd/izac175 Lloyd-Price J, Arze C, Ananthakrishnan AN, Schirmer M, Avila-Pacheco J, Poon TW, Andrews E, Ajami NJ, Bonham KS, Brislawn CJ, Casero D, Courtney H, Gonzalez A, Graeber TG, Hall AB, Lake K, Landers CJ, Mallick H, Plichta DR, Prasad M, Rahnavard G, Sauk J, Shungin D, Vázquez-Baeza Y, White RA, IBDMDB Investigators, Braun J, Denson LA, Jansson JK, Knight R, Kugathasan S, McGovern DPB, Petrosino JF, Stappenbeck TS, Winter HS, Clish CB, Franzosa EA, Vlamakis H, Xavier RJ, Huttenhower C. 2019. Multi-Omics of the gut microbial Ecosystem in inflammatory bowel diseases. Nature 569:655–662. https://doi.org/10.1038/s41586- 019-1237-9 12. 11. 10. Hecht AL, Casterline BW, Earley ZM, Goo YA, Goodlett DR, Bubeck Wardenburg J. 2016. Strain competition restricts colonization of an enteric pathogen and prevents colitis. EMBO Rep 17:1281–1291. https:// doi.org/10.15252/embr.201642282 Verster AJ, Ross BD, Radey MC, Bao Y, Goodman AL, Mougous JD, Borenstein E. 2017. The landscape of type VI secretion across human gut microbiomes reveals its role in community composition. Cell Host Microbe 22:411–419. https://doi.org/10.1016/j.chom.2017.08.010 Chatzidaki-Livanis M, Geva-Zatorsky N, Comstock LE. 2016. Bacteroides fragilis type VI secretion systems use novel effector and immunity proteins to antagonize human gut Bacteroidales species. Proc Natl Acad Sci U S A 113:3627–3632. https://doi.org/10.1073/pnas.1522510113 Cherrak Y, Flaugnatti N, Durand E, Journet L, Cascales E. 2019. Structure and activity of the type VI secretion system. Microbiol Spectr 7:1–11. https://doi.org/10.1128/microbiolspec.PSIB-0031-2019 Russell AB, Peterson SB, Mougous JD. 2014. Type VI secretion system effectors: poisons with a purpose. Nat Rev Microbiol 12:137–148. https:// doi.org/10.1038/nrmicro3185 Coyne MJ, Roelofs KG, Comstock LE. 2016. Type VI secretion systems of human gut Bacteroidales segregate into three genetic architectures, two of which are contained on mobile genetic elements. BMC Genomics 17:58. https://doi.org/10.1186/s12864-016-2377-z 13. 14. 15. 16. García-Bayona L, Coyne MJ, Comstock LE. 2021. Mobile type VI secretion system Loci of the gut Bacteroidales display extensive intra-Ecosystem transfer, multi-species spread and geographical clustering. PLoS Genet 17:e1009541. https://doi.org/10.1371/journal.pgen.1009541 Coyne MJ, Zitomersky NL, McGuire AM, Earl AM, Comstock LE. 2014. Evidence of extensive DNA transfer between Bacteroidales species within the human gut. mBio 5:e01305–14. https://doi.org/10.1128/mBio. 01305-14 17. 18. Wexler AG, Bao Y, Whitney JC, Bobay LM, Xavier JB, Schofield WB, Barry NA, Russell AB, Tran BQ, Goo YA, Goodlett DR, Ochman H, Mougous JD, Goodman AL. 2016. Human symbionts inject and neutralize antibacterial toxins to persist in the gut. Proc Natl Acad Sci U S A 113:3639–3644. https://doi.org/10.1073/pnas.1525637113 Silverman JM, Agnello DM, Zheng H, Andrews BT, Li M, Catalano CE, Gonen T, Mougous JD. 2013. Haemolysin coregulated protein is an exported receptor and chaperone of type VI secretion substrates. Mol Cell 51:584–593. https://doi.org/10.1016/j.molcel.2013.07.025 19. 20. Günther P, Quentin D, Ahmad S, Sachar K, Gatsogiannis C, Whitney JC, Raunser S, Satchell KJF. 2022. Structure of a bacterial Rhs effector exported by the type VI secretion system. PLoS Pathog 18:e1010182. https://doi.org/10.1371/journal.ppat.1010182 21. Ma J, Pan Z, Huang J, Sun M, Lu C, Yao H. 2017. The Hcp proteins fused with diverse extended-toxin domains represent a novel pattern of Antibacterial Effectors in type VI secretion systems. Virulence 8:1189– 1202. https://doi.org/10.1080/21505594.2017.1279374 22. Wood TE, Howard SA, Förster A, Nolan LM, Manoli E, Bullen NP, Yau HCL, Hachani A, Hayward RD, Whitney JC, Vollmer W, Freemont PS, Filloux A. 2019. The Pseudomonas aeruginosa T6SS delivers a periplasmic toxin that disrupts bacterial cell morphology. Cell Rep 29:187–201. https://doi. org/10.1016/j.celrep.2019.08.094 23. Ma L-S, Hachani A, Lin J-S, Filloux A, Lai E-M. 2014. Agrobacterium tumefaciens Deploys a Superfamily of type VI secretion Dnase Effectors as weapons for Interbacterial competition in Planta. Cell Host Microbe 16:94–104. https://doi.org/10.1016/j.chom.2014.06.002 24. Wettstadt S, Lai E-M, Filloux A. 2020. Solving the puzzle: Connecting a heterologous Agrobacterium Tumefaciens T6Ss Effector to a Pseudomo nas Aeruginosa spike complex. Front Cell Infect Microbiol 10:291. https:// doi.org/10.3389/fcimb.2020.00291 Bondage DD, Lin J-S, Ma L-S, Kuo C-H, Lai E-M. 2016. Vgrg C terminus confers the type VI Effector transport specificity and is required for binding with PAAR and Adaptor-Effector complex. Proc Natl Acad Sci U S A 113:E3931–40. https://doi.org/10.1073/pnas.1600428113 25. 26. Wu C-F, Lien Y-W, Bondage D, Lin J-S, Pilhofer M, Shih Y-L, Chang JH, Lai E-M. 2020. Effector loading onto the Vgrg carrier activates type VI secretion system assembly. EMBO Rep 21:e47961. https://doi.org/10. 15252/embr.201947961 Ross BD, Verster AJ, Radey MC, Schmidtke DT, Pope CE, Hoffman LR, Hajjar AM, Peterson SB, Borenstein E, Mougous JD. 2019. Human gut bacteria contain acquired interbacterial defence systems. Nature 575:224–228. https://doi.org/10.1038/s41586-019-1708-z 27. 28. Mok BY, de Moraes MH, Zeng J, Bosch DE, Kotrys AV, Raguram A, Hsu F, Radey MC, Peterson SB, Mootha VK, Mougous JD, Liu DR. 2020. A bacterial cytidine deaminase toxin enables CRISPR-free mitochondrial base editing. Nature 583:631–637. https://doi.org/10.1038/s41586-020- 2477-4 32. 30. 31. toxin 29. Hespanhol JT, Sanchez-Limache DE, Nicastro GG, Mead L, Llontop EE, Chagas-Santos G, Farah CS, de Souza RF, Galhardo R da S, Lovering AL, Bayer-Santos E. 2022. Antibacterial T6Ss Effectors with a VRR-Nuc domain are structure-specific Nucleases. Elife 11:e82437. https://doi.org/ 10.7554/eLife.82437 Ting S-Y, Bosch DE, Mangiameli SM, Radey MC, Huang S, Park Y-J, Kelly KA, Filip SK, Goo YA, Eng JK, Allaire M, Veesler D, Wiggins PA, Peterson SB, Mougous JD. 2018. Bifunctional immunity proteins protect bacteria against ftsz-targeting ADP-ribosylating toxins. Cell 175:1380–1392. https://doi.org/10.1016/j.cell.2018.09.037 Zhang D, de Souza RF, Anantharaman V, Iyer LM, Aravind L. 2012. Polymorphic systems: comprehensive characterization of trafficking modes, processing, mechanisms of action, immunity and Ecology using comparative Genomics. Biol Direct 7:18. https://doi.org/ 10.1186/1745-6150-7-18 Integrative H. 2014. The integrative human microbiome project: dynamic analysis of microbiome-host omics profiles during periods of human health and disease. Cell Host Microbe 16:276–289. https://doi. org/10.1016/j.chom.2014.08.014 Potter SC, Luciani A, Eddy SR, Park Y, Lopez R, Finn RD. 2018. HMMER web server: 2018 update. Nucleic Acids Res 46:W200–W204. https://doi.org/ 10.1093/nar/gky448 Sorbara MT, Littmann ER, Fontana E, Moody TU, Kohout CE, Gjonbalaj M, Eaton V, Seok R, Leiner IM, Pamer EG. 2020. Functional and genomic variation between human-derived isolates of Lachnospiraceae reveals inter- and intra-species diversity. Cell Host Microbe 28:134–146. https:// doi.org/10.1016/j.chom.2020.05.005 Jurėnas D, Rosa LT, Rey M, Chamot-Rooke J, Fronzes R, Cascales E. 2021. Mounting, structure and Autocleavage of a type VI secretion-associated Rhs polymorphic toxin. Nat Commun 12:6998. https://doi.org/10.1038/ s41467-021-27388-0 33. 34. 35. 36. Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. 2021. Colabfold-making protein folding accessible to all. biorxiv:1–6. https://doi.org/10.1101/2021.08.15.456425:2021.08.15.456425 37. Holm L. 2020. Using dali for protein structure comparison. Methods Mol Biol 2112:29–42. https://doi.org/10.1007/978-1-0716-0270-6_3 38. Qiu J, Bernhofer M, Heinzinger M, Kemper S, Norambuena T, Melo F, Rost B. 2020. ProNA2020 predicts protein-DNA, protein-RNA, and protein- protein binding proteins and residues from sequence. J Mol Biol 432:2428–2443. https://doi.org/10.1016/j.jmb.2020.02.026 July/August Volume 14 Issue 4 10.1128/mbio.01039-23 15 Research Article mBio 39. 40. 41. 42. 43. Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL. 2009. BLAST+: architecture and applications. BMC Bioinformatics 10:421. https://doi.org/10.1186/1471-2105-10-421 Forster SC, Kumar N, Anonye BO, Almeida A, Viciani E, Stares MD, Dunn M, Mkandawire TT, Zhu A, Shao Y, Pike LJ, Louie T, Browne HP, Mitchell AL, Neville BA, Finn RD, Lawley TD. 2019. A human gut bacterial genome and culture collection for improved metagenomic analyses. Nat Biotechnol 37:186–192. https://doi.org/10.1038/s41587-018-0009-7 Krissinel E, Henrick K. 2007. Inference of macromolecular assemblies from crystalline state. J Mol Biol 372:774–797. https://doi.org/10.1016/j. jmb.2007.05.022 Kleanthous C, Walker D. 2001. Immunity proteins: enzyme inhibitors that avoid the active site. Trends Biochem Sci 26:624–631. https://doi.org/10. 1016/s0968-0004(01)01941-7 Yin R, Feng BY, Varshney A, Pierce BG. 2022. Benchmarking Alphafold for protein complex modeling reveals accuracy determinants. Protein Sci 31:e4379. https://doi.org/10.1002/pro.4379 44. Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, Heer FT, de Beer TAP, Rempfer C, Bordoli L, Lepore R, Schwede T. 2018. SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res 46:W296–W303. https://doi.org/10.1093/ nar/gky427 45. Ahmad S, Wang B, Walker MD, Tran H-KR, Stogios PJ, Savchenko A, Grant RA, McArthur AG, Laub MT, Whitney JC. 2019. An interbacterial toxin inhibits target cell growth by synthesizing (P) ppApp. Nature 575:674– 678. https://doi.org/10.1038/s41586-019-1735-9 47. 46. Walker D, Lancaster L, James R, Kleanthous C. 2004. Identification of the catalytic motif of the microbial ribosome inactivating cytotoxin colicin E3. Protein Sci 13:1603–1611. https://doi.org/10.1110/ps.04658504 Papadakos G, Wojdyla JA, Kleanthous C. 2012. Nuclease colicins and their immunity proteins. Q Rev Biophys 45:57–103. https://doi.org/10. 1017/S0033583511000114 Ray A, Schwartz N, de Souza Santos M, Zhang J, Orth K, Salomon D. 2017. Type VI secretion system MIX-effectors carry both antibacterial and anti- eukaryotic activities. EMBO Rep 18:1978–1990. https://doi.org/10.15252/ embr.201744226 Brodmann M, Schnider ST, Basler M. 2021. Type VI secretion system and its Effectors Pdpc, Pdpd, and Opia contribute to Francisella virulence in 49. 48. 50. Galleria Mellonella larvae. Infect Immun 89:e0057920. https://doi.org/10. 1128/IAI.00579-20 Liu L, Xu M, Lan R, Hu D, Li X, Qiao L, Zhang S, Lin X, Yang J, Ren Z, Xu J. 2022. Bacteroides vulgatus attenuates experimental mice colitis through modulating gut microbiota and immune responses. Front. Immunol 13:1036196. https://doi.org/10.3389/fimmu.2022.1036196 53. 54. 52. 51. Ali J, Yu M, Sung L-K, Cheung Y-W, Lai E-M. 2023. A glycine Zipper motif governs translocation of type VI secretion toxic Effectors across the cytoplasmic membrane of target cells. bioRxiv:1–47. https://doi.org/10. 1101/2022.07.12.499750:2022.07.12.499750 de Moraes MH, Hsu F, Huang D, Bosch DE, Zeng J, Radey MC, Simon N, Ledvina HE, Frick JP, Wiggins PA, Peterson SB, Mougous JD. 2021. An Interbacterial DNA Deaminase toxin directly Mutagenizes surviving target populations. Elife 10:e62967. https://doi.org/10.7554/eLife.62967 Zhang H, Zhang H, Gao Z-Q, Wang W-J, Liu G-F, Xu J-H, Su X-D, Dong Y-H. 2013. Structure of the type VI effector-immunity complex (Tae4-Tai4) provides novel insights into the inhibition mechanism of the effector by its immunity protein. J Biol Chem 288:5928–5939. https://doi.org/10. 1074/jbc.M112.434357 Kühlmann UC, Pommer AJ, Moore GR, James R, Kleanthous C. 2000. Specificity in protein-protein interactions: the structural basis for dual recognition in endonuclease colicin-immunity protein complexes. J Mol Biol 301:1163–1178. https://doi.org/10.1006/jmbi.2000.3945 Cascales E, Buchanan SK, Duché D, Kleanthous C, Lloubès R, Postle K, Riley M, Slatin S, Cavard D. 2007. Colicin biology. Microbiol Mol Biol Rev 71:158–229. https://doi.org/10.1128/MMBR.00036-06 Riley MA. 1998. Molecular mechanisms of bacteriocin evolution. Annu Rev Genet 32:255–278. https://doi.org/10.1146/annurev.genet.32.1.255 Bacic MK, Smith CJ. 2008. Laboratory maintenance and cultivation of Bacteroides species. Curr Protoc Microbiol Chapter 13:Unit 13C.1. https:/ /doi.org/10.1002/9780471729259.mc13c01s9 Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, Ferrin TE. 2004. UCSF Chimera--a visualization system for exploratory research and analysis. J Comput Chem 25:1605–1612. https://doi.org/10. 1002/jcc.20084 55. 56. 57. 58. July/August Volume 14 Issue 4 10.1128/mbio.01039-23 16
10.1128_mbio.00782-23	\| Editor’s Pick \| Virology \| Research Article APOBEC3 degradation is the primary function of HIV-1 Vif determining virion infectivity in the myeloid cell line THP-1 Terumasa Ikeda,1 Ryo Shimizu,1,2 Hesham Nasser,1,3 Michael A. Carpenter,4,5 Adam Z. Cheng,6,7 William L. Brown,6,7 Daniel Sauter,8 Reuben S. Harris4,5 AUTHOR AFFILIATIONS See affiliation list on p. 18. ABSTRACT HIV-1 must overcome multiple innate antiviral mechanisms to replicate in CD4+ T lymphocytes and macrophages. Previous studies have demonstrated that the apolipoprotein B mRNA editing enzyme polypeptide-like 3 (APOBEC3, A3) family of proteins (at least A3D, A3F, A3G, and stable A3H haplotypes) contribute to HIV-1 restriction in CD4+ T lymphocytes. Virus-encoded virion infectivity factor (Vif ) coun teracts this antiviral activity by degrading A3 enzymes allowing HIV-1 replication in infected cells. In addition to A3 proteins, Vif also targets other cellular proteins in CD4+ T lymphocytes, including PPP2R5 proteins. However, whether Vif primarily degrades only A3 proteins during viral replication is currently unknown. Herein, we describe the development and characterization of A3F-, A3F/A3G-, and A3A-to-A3G-null THP-1 cells. In comparison to Vif-proficient HIV-1, Vif-deficient viruses have substantially reduced infectivity in parental and A3F-null THP-1 cells, and a more modest decrease in infectivity in A3F/A3G-null cells. Remarkably, disruption of A3A–A3G protein expression completely restores the infectivity of Vif-deficient viruses in THP-1 cells. These results indicate that the primary function of Vif during infectious HIV-1 production from THP-1 cells is the targeting and degradation of A3 enzymes. IMPORTANCE HIV-1 Vif neutralizes the HIV-1 restriction activity of A3 proteins. However, it is currently unclear whether Vif has additional essential cellular targets. To address this question, we disrupted A3A to A3G genes in the THP-1 myeloid cell line using CRISPR and compared the infectivity of wild-type HIV-1 and Vif mutants with the selective A3 neutralization activities. Our results demonstrate that the infectivity of Vif-deficient HIV-1 and the other Vif mutants is fully restored by ablating the expression of cellular A3A to A3G proteins. These results indicate that A3 proteins are the only essential target of Vif that is required for fully infectious HIV-1 production from THP-1 cells. KEYWORDS HIV-1, APOBEC3, G-to-A mutations, deaminase-dependent mechanism, deaminase-independent mechanism, Vif T he A3 family of proteins comprise seven single-strand DNA cytosine deaminases (A3A–A3D and A3F–A3H) in humans (1–3). A3 enzymes have broad and essential roles in innate antiviral immunity against parasitic DNA-based elements (4–6). Retrovi ruses are sensitive to A3 enzyme activity due to the obligate step of reverse transcription during viral replication that produces single-stranded cDNA intermediates. These viral cDNA intermediates can act as substrates for A3 enzymes, as demonstrated by C-to-U deamination resulting in G-to-A mutations in the genomic strand. To date, the best-char acterized substrate of A3 enzymes is human immunodeficiency virus type 1 (HIV-1). In CD4+ T lymphocytes, four A3 proteins (A3D, A3F, A3G, and stable A3H haplotypes) restrict HIV-1 replication by mutating viral cDNA intermediates and by physically blocking reverse transcription (7–14). A3 enzymes have a preference for specific dinucleotide Invited Editor Stuart J. Neil, King's College London, London, United Kingdom Editor Stephen P. Goff, Columbia University Medical Center, New York, New York, USA Address correspondence to Terumasa Ikeda, [email protected]. Terumasa Ikeda and Ryo Shimizu contributed equally to this article. Author order was determined in order of increasing seniority. The authors declare no conflict of interest. See the funding table on p. 18. Received 28 March 2023 Accepted 22 June 2023 Published 9 August 2023 Copyright © 2023 Ikeda et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International license. Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 1 Research Article mBio motifs (5´-CC for A3G and 5´-TC for other A3 enzymes) at target cytosine bases, which appear as 5´-AG or 5´-AA mutations in the genomic strand (7, 8, 15, 16). Virus-encoded virion infectivity factor (Vif ) functions in disrupting the activity of A3 enzymes. Vif forms an E3 ubiquitin ligase complex that degrades A3 enzymes through a proteasome-mediated pathway (2, 3, 17, 18). The central domain of this complex is a Vif heterodimer with the transcription factor, core binding factor subunit β (CBF-β), which stabilizes Vif during disruption of A3 protein activity (19, 20). Vif also suppresses the transcription of A3 enzymes by hijacking RUNX/CBF-β complex (21). In addition to these Vif-dependent mechanisms, HIV-1 reverse transcriptase and protease have been shown to disrupt the activity of A3 enzymes via Vif-independent mechanisms (22, 23). Recently, functional proteomic analyses have demonstrated that Vif has several target proteins, including the PPP2R5 family of proteins, in CD4+ T cell lines and lymphocytes (24, 25). These findings indicate that Vif may have additional essential target proteins during HIV-1 infection. We previously reported that endogenous A3G protein contributes to HIV-1 restric tion in a deaminase-dependent manner in THP-1 cells (26). Although disruption of the A3G gene nearly eliminates viral G-to-A mutations, Vif-deficient HIV-1 virions have 50% lower infectivity than wild-type HIV-1 or mutants selectively lacking A3G degradation activity (26). These results indicated that Vif-mediated inhibition of A3G and at least one additional A3 proteins is required for efficient infectious HIV-1 production. In the present study, we evaluate the effects of other A3 proteins on HIV-1 infectiv ity by developing and characterizing A3F-, A3F/A3G-, and A3A-to-A3G-null THP-1 cells using HIV-1 Vif mutants with selective A3 neutralization activities. In comparison to wild-type HIV-1, Vif-deficient HIV-1 infectivity is strongly inhibited in A3F-null THP-1 cells and modestly inhibited in A3F/A3G-null THP-1 cells. In contrast, an HIV-1 Vif mutant selectively lacking A3F degradation activity had comparable infectivity to wild-type HIV-1 in A3F-null THP-1 cells and 50% infectivity in parental THP-1 cells, indicating that A3F protein contributes to HIV-1 restriction in THP-1 cells. Furthermore, Vif-deficient HIV-1 infectivity is comparable to wild-type HIV-1 in A3A-to-A3G-null THP-1 cells. These results demonstrate that A3 proteins are the primary target of HIV-1 Vif during infectious virus production from THP-1 cells. RESULTS Endogenous A3H protein is not involved in HIV-1 restriction in THP-1 cells THP-1 cells express significant levels of A3B, A3C, A3F, A3G, and A3H mRNA (26). The results of our previous study indicated that A3G and at least one additional A3 proteins are involved in HIV-1 restriction in THP-1 cells (26). Variations in the amino acid sequence of A3 family proteins are known to influence HIV-1 restriction activity (27), and the A3H gene is the most polymorphic of all human A3 genes (10, 22, 28, 29). The A3H allele is grouped into stable and unstable haplotypes according to the combination of amino acid residues at positions 15, 18, 105, 121, and 178 (10, 22, 28, 29). Stable A3H haplotypes are active against HIV-1, whereas unstable A3H haplotypes have absent or minimal activity as they encode proteins with low stability (9, 10, 22, 29, 30). To determine A3H genotypes, we sequenced A3H cDNA from THP-1 cells. Sequencing data identified an unstable haplotype in the THP-1 genome, termed A3H hapI gene (Fig. 1A). These data suggest that endogenous A3H protein has minimal restriction activity against Vif-deficient HIV-1 in THP-1 cells. The A3H hapI protein results in expression of an unstable protein that has weak anti- HIV-1 activity (28, 29, 31). However, this protein is enzymatically active and has an HIV-1 restriction phenotype similar to the stable A3H haplotype, A3H hapII protein, when both proteins are expressed at the same levels (31). In addition, A3H protein expression levels are upregulated during HIV-1 infection (10, 22), and A3H hapI protein is resistant to Vif- mediated degradation (32). Accordingly, we evaluated whether the expression of A3H hapI protein is associated with HIV-1 restriction in THP-1 cells. To address this question, Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 2 Research Article mBio FIG 1 Endogenous A3H protein does not inhibit HIV-1 in THP-1 cells. (A) A3H haplotypes in THP-1 cells. The indicated positions are key amino acid residues that determine the expression of unstable (hapI) or stable (hapII) A3H protein. (B) Schematic of the susceptibility of HIV-1 Vif mutants to antiviral activity by stable A3H haplotypes. Key amino acid residues of Vif that determine the susceptibility of HIV-1 IIIB to restriction by stable A3H haplotypes. −, full resistance; +, partial resistance; +++, sensitivity. (C) Schematic depiction of the pseudo-single cycle infectivity assay. For details, see the main text and the “Pseudo-single cycle infectivity assays” section in Materials and Methods. (D) Representative infectivity of HIV-1 mutants with hyper- and hypo-functional Vif produced from THP-1 cells. Top panels show the infectivity of hyper-Vif, hypo-Vif, IIIB Vif, and Vif-deficient HIV-1 mutants produced in THP-1 cells. The amounts of produced viruses used to infect TZM-bl cells were normalized to p24 levels. Each bar shows the average of four independent experiments with standard deviation (SD). Data are represented as relative infectivity compared to hyper-Vif HIV-1. Statistical significance was determined using the two-sided paired t test. P < 0.05 compared with the infectivity of hyper-Vif HIV-1. The bottom panels are representative western blots of three independent experiments. The levels of viral and cellular proteins in viral particles and whole-cell lysates are shown. p24 and HSP90 were used as loading controls. (E) G-to-A mutations. Average number of G-to-A mutations in the 564 bp pol gene after infection with hyper-Vif, hypo-Vif, IIIB Vif, or Vif-deficient HIV-1 produced from THP-1. Each bar depicts the average of three independent experiments with SD. (F) G-to-A mutation profile. Dinucleotide sequence contexts of G-to-A mutations in the 564 bp pol gene after infection with the indicated viruses produced from indicated cell lines. Each vertical line indicates the location of the dinucleotide sequence contexts described in the legend within the 564 bp amplicon (horizontal line). we utilized HIV-1 Vif mutants that selectively degrade stable A3H protein (hyper- functional Vif; hyper-Vif ) or lack stable A3H degradation (hypo-functional Vif; hypo-Vif ) (10) (Fig. 1B). IIIB Vif displays an intermediate phenotype (Fig. 1B). Of note, hyper-Vif, hypo-Vif, and IIIB Vif have full neutralization activity against A3D, A3F, and A3G proteins (10). Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 3 Research Article mBio To allow efficient and equivalent delivery of HIV-1 to THP-1 and its derivatives, vesicular stomatitis virus G glycoprotein (VSV-G) pseudotyped full-length HIV-1 and its Vif mutants were produced from 293T cells and titrated with CEM-GXR cells for determination of the multiplicity of infection (MOI) as described (23, 26) (Fig. 1C). Then, these VSV-G pseudotyped infectious viruses were used to infect SupT11 (MOI = 0.05) and THP-1 (MOI = 0.25) cells, thereby creating virus-producing cells (Fig. 1C). The MOI calculations are based on titers determined by CEM-GXR cells and these MOIs therefore may not reflect the true MOI on SupT11 or THP-1 cells. The resultant infectious viru ses were used to measure viral infectivity in TZM-bl cells, evaluate the packaging of A3 proteins by western blotting, and analyze the frequency of G-to-A mutations (Fig. 1C). Hereafter, this assay is referred to as pseudo-single cycle infectivity assay (see the “Pseudo-single cycle infectivity assays” section in the Materials and Methods for details). The susceptibility of hyper-Vif and hypo-Vif HIV-1 to stable A3H hapII protein was validated in SupT11 cell lines (Fig. S1A). In SupT11-vector cells, hyper-Vif, hypo-Vif, IIIB Vif, and Vif-deficient HIV-1 had comparable infectivity with TZM-bl cells (Fig. S1A, top panel). As expected, the infectivity of hypo-Vif and Vif-deficient HIV-1 was restricted in SupT11-A3H hapII cells because these mutants are unable to degrade A3H hapII protein, leading to packaging of A3H hapII protein into viral particles (Fig. S1A, bottom panel). The partial degradation of A3H hapII protein by IIIB Vif resulted in moderate inhibition of IIIB Vif HIV-1 infectivity (Fig. S1A). The infection of hyper-Vif HIV-1 resulted in A3H hapII degradation, where stable A3H hapII protein is undetectable in the viral particles and unable to inhibit the Vif mutant (Fig. S1A). Next, to determine whether G-to-A mutations were introduced into proviral DNA, we recovered proviral DNA from SupT11 cells after infection with each HIV-1 mutant produced from either SupT11-vector or SupT11-A3H hapII cells and sequenced the pol region. Hypo-Vif and Vif-deficient HIV-1 showed G-to-A mutations preferred by A3H protein (GA-to-AA signature motif; hypo-Vif: 2.8 ± 0.7 mutations/kb and ∆Vif: 3.2 ± 0.8 mutations/kb, respectively) in proviral DNA (Fig. S1B and C). These results are consistent with previous reports demonstrating the susceptibility of Vif mutants to A3H hapII protein (10). As shown in Fig. 1D (top panel), hyper-Vif HIV-1, hypo-Vif HIV-1, and IIIB Vif HIV-1 (IIIB) produced in THP-1 cells had similar viral infectivity. While Vif did not degrade A3H protein in THP-1 cells, it was not packaged into viral particles (Fig. 1D, bottom panel). To examine whether G-to-A mutations were introduced into proviral DNA, we sequenced the pol region of the proviruses from SupT11 cells after infection with each HIV-1 mutant produced from THP-1 cells. Sequencing data demonstrated that hyper-Vif HIV-1, hypo-Vif HIV-1, and IIIB Vif HIV-1 had minimal G-to-A mutations preferred by A3H protein (GA-to-AA signature motif ) in proviral DNA (Fig. 1E and F), indicating that endogenous A3H protein expressed in THP-1 cells is not involved in HIV-1 restriction. In contrast, the infectivity of Vif-null HIV-1 was restricted in THP-1 cells and A3G protein was packaged into viral particles, thereby inducing profound G-to-A mutations (10.3 ± 3.5 mutations/kb) (Fig. 1D through F). Most of mutations were in the GG-to-AG signature motif preferred by A3G protein (80 ± 10%) in proviral DNA (Fig. 1E and F). Taken together, these results indicate that A3G and perhaps other A3 proteins, but unlikely A3H protein, contribute to HIV-1 restriction in THP-1 cells. Development of A3F-, A3F/A3G-, and A3A-to-A3G-null THP-1 cells A3F protein has a restrictive effect on HIV-1 among A3 family members and is a target of Vif, in addition to A3G protein, in CD4+ T cell lines and lymphocytes (7, 33–35). To determine whether A3F protein also reduces HIV-1 infectivity in THP-1 cells, we used CRISPR to create A3F and A3F/A3G gene knockout cell lines. Two independent subclones of A3F and A3F/A3G-null THP-1 cells were obtained, as evidenced by the results of genomic DNA sequencing and western blotting (Fig. S2). A3 proteins include single- and double-domain deaminases and are phylogenetically classified into three groups: Z1, Z2, and Z3 domains (3, 4) (Fig. 2A represented in green, orange, and blue, respectively). A3A, A3B carboxy-terminal domain (CTD), and A3G CTD Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 4 Research Article mBio proteins are classified as Z1 domains (Fig. 2A, represented in green). Of note, exon 4 of the A3A gene, exon 7 of the A3B gene, and exon 7 of A3G gene are highly conserved at the nucleotide level (A3A exon 4 and A3B exon 7 have 95% identity; A3A exon 4 and A3G exon 7 have >99% identity; and A3B exon 7 and A3G exon 7 have 95% identity, respectively). Interestingly, each of these exons has an identical sequence (5´-GAG TGG GAG GCT GCG GGC CA). We therefore designed a guide RNA (gRNA) homologous to this sequence and used it to delete the entire 125 kbp interval spanning A3A to A3G genes in THP-1 cells (Fig. 2A, represented in arrows, and Fig. S3). We predicted that successful Cas9-mediated cleavage would cause one of the following three scenarios: (i) fusion of exon 4 of the A3A gene with exon 7 of the A3B gene (30 kbp deletion); (ii) fusion of exon 7 of the A3B gene with exon 7 of the A3G gene (95 kbp deletion); or (iii) fusion of exon 4 of the A3A gene with exon 7 of the A3G gene (125 kbp deletion; Fig. 2A). To obtain THP-1 cells lacking expression of A3A to A3G proteins, a lentiviral vector expressing gRNA against the target sequence was transduced into THP-1 cells. Finally, two independent subclones (THP-1#11-4 and THP-1#11-7) were obtained, with whole-genome sequencing (WGS) analysis demonstrating an extensive deletion between A3A exon 4 and A3G exon 7 at the A3 gene locus (Fig. 2B). In THP-1#11-4, six alleles of the fusion of A3A exon 4 with A3G exon 7 are observed, and each A3A/A3G hybrid exon had six different insertions or deletions (indels) (Fig. S3). THP-1#11-7 harbors three alleles of A3A exon 4 and A3G exon 7 fusions (one may be A3A exon 4) with three different deletions (Fig. S3). Although more than 20 potential off-target sites with two or three nucleotides mismatched with the designed gRNA were predicted, a significant deletion was only found downstream of the predicted A3G pseudogene harboring 2 bp mismatched with the target sequence (Fig. S4; potential target sequence in an orange box and deletions indicated by green dotted lines). In comparison to parental THP-1 cells, these subclones had similar growth capacities under normal cell culture conditions. Reverse transcription-quantitative PCR (RT-qPCR) analyses demonstrated that A3B to A3G mRNAs are not detectable in either clone (Fig. 2C). However, A3A mRNA expression remained detectable in parental THP-1 cells and the two subclones as the A3A promoter remains intact and potentially functional (Fig. 2A through C). A3A mRNA expression is known to be upregulated 100- to 1,000-fold in THP-1 cells treated with type I interferon (IFN) (36). To confirm the expression of A3A mRNA and protein in THP-1 cells, parental THP-1 cell and the respective subclones were cultured in the presence of type I IFN for 6 h, and A3 mRNA and protein expression levels were then analyzed by RT-qPCR and western blotting, respectively. In parental THP-1 cells, A3A, A3B, A3F, and A3G mRNA and protein expression levels were increased following IFN treatment (Fig. 2C and D). In the THP-1#11-4 subclone, A3A mRNA expression is increased following IFN treatment; however, A3A, A3B, A3C, A3F, and A3G proteins are not detectable, even after IFN treatment (Fig. 2C and D). Furthermore, A3A to A3G proteins are not detectable in the THP-1#11-7 subclone under normal cell culture conditions (Fig. 2D). Interestingly, low levels of a protein with comparable size to A3A protein are detected in the THP-1#11-7 subclone after IFN treatment (Fig. 2D). Sanger sequence analyses indicated that this protein was an A3A and A3G hybrid with a 3 bp deletion (Fig. S3). Collectively, these data indicate that the THP-1#11-4 and THP-1#11-7 subclones lack expression of A3A to A3G proteins under normal cell culture conditions and that clone THP-1#11-4 is a clean knockout that fails to express functional versions of any of these proteins. Disruption of A3A to A3G protein expression fully restores the infectivity of Vif-deficient HIV-1 in THP-1 cells We next determined whether endogenous A3F protein is degraded by Vif in addition to A3G protein. HIV-1 Vif mutants with selective A3 neutralization activities were used for pseudo-single cycle infectivity assays as mentioned above. For example, a Vif4A mutant harboring 14AKTK17 substitutions (14DRMR17 in IIIB) is susceptible to HIV-1 restriction activity by A3D and A3F proteins but resistant to the restriction by A3G protein (37–39) (Fig. 3A). We examined the ability of Vif4A to counteract the restriction activity of A3D Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 5 Research Article mBio FIG 2 Disruption of the A3A to A3G genes in THP-1 cells. (A) Schematic of the A3 gene at the A3 locus. The A3 family of genes comprises seven members with one or two Z domains (single- or double-domain deaminases) which belong to three phylogenetically distinct groups shown in green, orange, and blue. Three sites with an identical sequence (5´-GAG TGG GAG GCT GCG GGC CA) in exon 4 of the A3A gene, exon 7 of the A3B gene, and exon 7 of the A3G gene are targeted by gRNA, as indicated by arrows. The three predicted scenarios are shown. Bar represents 15,000 bp. (B) Mapping of WGS data to the A3 locus. Genomic DNAs from parental THP-1, THP-1#11-4, and #11-7 cells were subjected to WGS analysis, with an extensive deletion including the A3A–A3G genes observed in THP-1#11-4 and #11-7 clones. (C) RT-qPCR data. Parental THP-1, THP-1#11-4, and #11-7 cells were treated with 500 units/mL type I IFN. Total RNA was isolated after 6 h. A3 mRNA expression levels were quantified by RT-qPCR and are normalized to TATA-binding protein (TBP) mRNA levels. Each bar represents the average of three independent experiments with SD. Statistical significance was determined using the two-sided paired t test. , P < 0.05 compared to untreated cells. (D) Representative western blots of three independent experiments. Levels of indicated A3 proteins in whole-cell lysates from cells treated with or without type I IFN are shown. HSP90 was used as a loading control. and A3F proteins, although A3D protein could not be detected by western blotting (anti- A3D antibodies are unavailable) and it may be inconsequential because its mRNA expression levels are relatively low in this cell line (26) (Fig. 2C). As our group and others have shown previously (26, 37, 38, 40), HIV-1 with Vif5A containing five alanine substitu tions (40YRHHY44 to 40AAAAA44) is sensitive to the HIV-1 restriction activity of A3G protein but not A3D and A3F proteins (Fig. 3A). HIV-1 harboring Vif4A5A is susceptible to inhibition by A3D, A3F and A3G proteins (37) (Fig. 3A). VSV-G pseudotyped HIV-1 and these Vif mutants were used to infect SupT11 derivatives and engineered A3F-null THP-1 cells. First, the susceptibilities of these Vif mutants to A3F and A3G proteins were validated in SupT11 cell lines (Fig. S5A). In SupT11-vector cells, Vif-proficient HIV-1 and all Vif mutants had comparable infectivity in TZM-bl cells (Fig. S5A). As expected, the infectivity of Vif-deficient HIV-1 and the Vif4A and 4A5A mutants was reduced in SupT11- A3F cells as these mutants are unable to degrade A3F protein, thereby leading to packaging of A3F protein in viral particles (Fig. S5A). Further, infection with Vif-deficient HIV-1 or the Vif5A and Vif4A5A mutants resulted in packaging of A3G protein in viral particles from SupT11-A3G cells in addition to reduced infectivity of these Vif mutants (Fig. S5A). These results are consistent with previous reports demonstrating the suscepti bilities of Vif mutants to A3 proteins (26, 37–40). Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 6 Research Article mBio FIG 3 Pseudo-single cycle infectivity assays for each HIV-1 mutant in A3-null THP-1 cells. (A) Schematic of the susceptibility of HIV-1 Vif mutants to HIV-1 restriction activity by A3F and A3G proteins. Key amino acid residues of Vif that determine the susceptibility of HIV-1 IIIB to restriction by A3F and A3G proteins. −, resistance; +, sensitivity. (B) Representative infectivity of Vif-proficient, Vif-deficient, Vif4A, Vif5A, and Vif4A5A HIV-1 mutants produced from parental or A3-null THP-1 cells. Top panels show the infectivity of indicated HIV-1 mutants produced in parental or A3-null THP-1 cells. The amounts of produced viruses used to infect TZM-bl cells were normalized to p24 levels. Each bar represents the average of four independent experiments with SD. Data are presented as infectivity relative to Vif-proficient HIV-1 (WT). Statistical significance was assessed using the two-sided paired t test. P < 0.05 compared to Vif-proficient HIV-1. Bottom panels are representative western blots of three independent experiments. Levels of indicated viral and cellular proteins in viral particles and whole-cell lysates are shown. p24 and HSP90 were used as loading controls. Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 7 Research Article mBio Pseudo-single cycle infectivity assays were then performed in parental THP-1, A3G- null, and A3F-null cells using these Vif mutants. Vif-proficient HIV-1 degraded A3F and A3G proteins in THP-1 cells, and lower amounts of these A3 proteins were packaged into viral particles (Fig. 3B; THP-1 parent). In contrast, Vif-deficient HIV-1 was unable to degrade A3F and A3G proteins, thereby leading to reduced viral infectivity compared to Vif-proficient HIV-1 (Fig. 3B; THP-1 parent). The infectivity of A3F-susceptible Vif mutants, Vif4A and Vif4A5A, was lower than that of Vif-proficient HIV-1, indicating that endoge nous A3F protein contributes to Vif-deficient HIV-1 restriction in THP-1 cells (Fig. 3B; THP-1 parent). This finding was supported by results in A3G-null THP-1 cells where Vif4A mutants are restricted, as observed in parental THP-1 cells (Fig. 3B; THP-1 ∆A3G). The involvement of endogenous A3G protein in HIV-1 restriction was confirmed in A3G-null THP-1 cells, as reported (26) (Fig. 3B; THP-1 ∆A3G). To determine whether endogenous A3F protein contributes to HIV-1 restriction in THP-1 cells, pseudo-single cycle infectivity assays were performed according to the methods described above in two independent A3F-null THP-1 clones (Fig. 1C; Fig. S2A and B). Vif-deficient HIV-1 and the Vif5A and Vif4A5A mutants had reduced infectivity in A3F-null subclones due to the inhibitory effect of A3G protein (Fig. 3B; THP-1 ∆A3F#1 and #2). However, the infectivity of the Vif4A mutant was restored to near wild-type levels following disruption of A3F gene expression in THP-1 cells. These data demonstrate that endogenous A3F and potentially A3D proteins contribute to Vif-deficient HIV-1 restriction in THP-1 cells, and that Vif degrades A3F protein and thereby prevents packaging and restriction upon target cell infection. A3F and A3G proteins are involved in Vif-deficient HIV-1 restriction in THP-1 cells and are degraded by Vif (26) (Fig. 3B). However, it is unclear whether only these A3 proteins are associated with Vif-deficient HIV-1 restriction in THP-1 cells. To address this issue, we performed pseudo-single cycle infectivity assays in A3F/A3G-null THP-1 cells using separation-of-function Vif mutants. Although Vif-deficient HIV-1 had greater infectivity defects in parental, A3G-null, and A3F-null THP-1 cells compared to wild-type HIV-1 (parent: <10% infectivity, ∆A3G: 30% to 40% infectivity, and ∆A3F: 20% infectivity, respectively), the infectivity of Vif-deficient HIV-1 was 30% lower in A3F/A3G-null THP-1 cells (Fig. 3B; THP-1 parent, ∆A3G, ∆A3F#1 and #2, and ∆A3F/A3G#1 and #2). On the other hand, the Vif4A, Vif5A, and Vif4A5A mutants had similar infectivity to wild-type HIV-1 in A3F/A3G-null THP-1 cells (Fig. 3B; THP-1 ∆A3F/A3G#1 and #2). These data indicate that other A3 proteins, in addition to A3F and A3G proteins, contribute to Vif-deficient HIV-1 restriction in THP-1 cells or that Vif disrupts an additional essential target during infectious virus production from THP-1 cells. The universally recognized primary target of Vif is the A3 family of proteins (2, 3, 17, 18). However, Vif-mediated A3 degradation may mask an additional A3-independ ent Vif function required for fully infectious virus production. To address this issue, we constructed two independent A3A-to-A3G-null THP-1 clones (Fig. 2) and character ized HIV-1 infection using pseudo-single cycle infectivity assays with Vif mutants. As mentioned above, the disruption of A3F and A3G protein expression results in Vif-defi- cient HIV-1 having 70% of wild-type HIV-1 infectivity in THP-1 cells (Fig. 3B; THP-1 ∆A3F/ A3G#1 and #2). Remarkably, Vif-deficient HIV-1 and the other Vif mutants have compara ble infectivity to Vif-proficient HIV-1 lacking expression of A3A to A3G proteins in THP-1 cells (Fig. 3B; THP-1#11-4 and #11-7). These results indicate that A3 degradation is the only function of Vif required for fully infectious virus production from THP-1 cells. A3 proteins restrict HIV-1 infectivity via deaminase-dependent and deami nase-independent mechanisms in THP-1 cells Our previous results indicated that A3G protein is the primary source of A3 mutagenesis in THP-1 cells (26). To further investigate the G-to-A mutation spectra in each A3-null THP-1 subclone, the pol region was cloned and sequenced from the proviruses used in the aforementioned infectivity assays. As expected, GG-to-AG mutations are observed in the proviral DNA of Vif mutants lacking A3G neutralization activity (Vif-deficient HIV-1 and Vif5A and Vif4A5A mutants) produced from SupT11-A3G cells (Fig. S5B and Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 8 Research Article mBio C; SupT11-A3G). Consistent with a previous report (26), THP-1 expresses A3G protein capable of mutating A3G-susceptible Vif mutants, including Vif-deficient HIV-1 and Vif5A and Vif4A5A mutants, as seen in parental THP-1 cells. These GG-to-AG mutations are not observed in A3G-null THP-1 cells (Fig. 4A and B; THP-1 parent and ∆A3G). Similarly, GG-to-AG mutations preferred by A3G protein were seen in the proviruses of the A3G-susceptible Vif mutants produced from two independent A3F-null THP-1 cells, with disruption of A3G protein nearly completely eliminating these mutations in THP-1 cells (Fig. 4A and B; THP-1 ∆A3F#1 and #2, ∆A3F/A3G#1 and #2, #11-4, and #11-7). These data indicate that A3G protein is the primary source of GG-to-AG mutations in HIV-1 proviruses produced by THP-1 cells. Although the Vif mutants lacking A3F neutralization activity (Vif-deficient HIV-1 and Vif4A and Vif4A5A mutants) produced from SupT11-A3F cells have a relatively low number of G-to-A mutations, the observed G-to-A mutations are predominantly within the GA-to-AA sequence motif preferred by A3F protein (Fig. S5B and C; SupT11-A3F). However, A3F-preferred GA-to-AA mutations are not observed in proviruses of A3F- susceptible Vif mutants produced from parental or A3G-null THP-1 cells, in support of prior observations (26) (Fig. 4A and B; THP-1 parent and ∆A3G). In addition, fewer GA-to- AA mutations are observed in THP-1 cells, even after disruption of A3F protein expression (Fig. 4A and B; THP-1 ∆A3F#1 and #2, ∆A3F/A3G#1 and #2, #11-4, and #11-7). Accordingly, these results combine to indicate that A3F protein in THP-1 cells is involved in Vif- deficient HIV-1 restriction via a deaminase-independent mechanism. A3F protein has been shown to inhibit the accumulation of reverse transcription (RT) products (14). To investigate a potential effect on RT, SupT11 cells were infected with viruses from the pseudo-single cycle infectivity assays described above, and late RT (LRT) products were examined by quantitative PCR (qPCR). As expected, all Vif mutants were decreased in LRT products in comparison to wild-type virus when these mutants were produced in parental THP-1 cells and used to infect SupT11 cells (Fig. 4C; THP-1 parent). LRT products of Vif5A and Vif4A mutants were restored to levels comparable to Vif-proficient HIV-1 following the disruption of A3G or A3F protein expression in THP-1 cells (Fig. 4C; THP-1 ∆A3G and ∆A3F#1 and #2), indicat ing that both A3G and A3F proteins inhibit HIV-1 via a deaminase-independent mechanism. However, double knockout of A3G and A3F proteins in THP-1 cells did not increase the LRT products of Vif-deficient HIV-1 compared to those of Vif- proficient virus (Fig. 4C; THP-1 ∆A3F/A3G#1 and #2), indicating that other A3 proteins, in addition to A3F and A3G proteins, may contribute to the restriction of HIV-1 in THP-1 cells via a deaminase-independent mechanism or that a separate protein targeted by Vif blocks the accumulation of RT products. To test this hypothesis, we measured LRT products by infecting SupT11 cells with HIV-1 Vif mutants produced in A3A-to-A3G-null clones. Consistent with the results of the pseudo-single cycle infectivity assays (Fig. 3B), Vif-deficient HIV-1 and other Vif mutants had comparable levels of LRT products to Vif-proficient HIV-1 lacking expression of A3A to A3G proteins in THP-1 cells (Fig. 4C; THP-1#11-4 and #11-7). These data indicate that Vif-mediated A3 degradation is required for fully infectious virus production from THP-1 to counteract deaminase-dependent and -independent HIV-1 restriction by A3 proteins. Transmitted/founder (TF) HIV-1 Vif also only targets A3 family proteins to enable fully infectious virus production from THP-1 cells As an additional experiment, we examined whether the A3-dependent function of Vif was present in TF viruses. To address this issue, Vif-proficient and deficient versions of the CH58 TF virus were produced from parental THP-1 and A3A-to-A3G-null cells, with viral infectivity measured in TZM-bl cells (Fig. 5). Similar to the results observed with IIIB viruses, Vif-deficient CH58 virus was restricted in parental THP-1 cells; however, this restriction is completely abolished by disruption of the A3A to A3G genes (Fig. 5). These Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 9 Research Article mBio FIG 4 A3 proteins inhibit Vif-deficient HIV-1 by both deaminase-dependent and -independent mechanisms in THP-1 cells. (A) G-to-A mutations. Average number of G-to-A mutations in the 564 bp pol gene after infection with hyper-Vif, hypo-Vif, IIIB Vif, or Vif-deficient HIV-1 produced from parental or A3-null THP-1 cells. Each bar depicts the average of three independent experiments with SD. (B) G-to-A mutation profile. Dinucleotide sequence contexts of G-to-A mutations (Continued on next page) Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 10 Research Article FIG 4 (Continued) mBio in the 564 bp pol gene after infection with the indicated viruses produced from indicated cell lines. Each vertical line indicates the location of the dinucleotide sequence contexts described in the legend within the 564 bp amplicon (horizontal line). (C) Representative LRT quantification data for Vif-proficient, Vif-deficient, Vif4A, Vif5A, and Vif4A5A HIV-1 mutants produced from each A3-null THP-1 subclone. Data show LRT products of the indicated HIV-1 mutants produced in parental or indicated A3-null THP-1 cells. The amount of produced viruses used to infect SupT11 cells was normalized to p24 levels. LRT products were measured by qPCR. Each bar represents the average of four independent experiments with SD. LRT products were normalized to the quantity of the CCR5 gene relative to Vif-proficient HIV-1 (WT). Statistical significance was assessed using the two-sided paired t test. P < 0.05 compared to Vif-proficient HIV-1 LRT products. data indicate that TF viruses also utilize a primarily A3-dependent function of Vif during infectious HIV-1 production from THP-1 cells. DISCUSSION Vif-mediated A3 degradation is critical for HIV-1 replication in CD4+ T lymphocytes and myeloid cells (2, 3, 17, 18). In CD4+ T lymphocytes, at least A3D, A3F, A3G, and A3H (only stable haplotypes) proteins are involved in Vif-deficient HIV-1 restriction, and Vif is required to degrade A3 enzymes and allow efficient viral replication (2, 3, 17, 18). However, the degradation of A3 enzymes by Vif during infectious HIV-1 production from myeloid lineage cells has yet to be fully elucidated. We previously reported that A3G protein contributes to Vif-deficient HIV-1 restriction in a deaminase-dependent manner in THP-1 cells (26). Herein, we demonstrate that A3F protein also inhibits Vif-deficient HIV-1 in a largely deaminase-independent manner and that Vif avoids this HIV-1 restriction mechanism by degrading A3F protein (Fig. 3 and 4). Importantly, the results of pseudo-single cycle infectivity assays demonstrate that the disruption of A3A to A3G proteins confers comparable infectivity to wild-type HIV-1 in a Vif-deficient lab-adapted virus (IIIB) and TF virus (CH58) (Fig. 3 to 5). These results indicate that Vif-mediated A3 degradation is the primary function of Vif during infectious HIV-1 production from THP-1 cells. FIG 5 Pseudo-single cycle infectivity assays of TF virus molecular clone in A3A-to-A3G-null THP-1 cells. Infectivity of Vif-proficient and Vif-deficient CH58 viruses. Top panels show the infectivity of Vif-proficient and Vif-deficient HIV-1 produced from parental THP-1, THP-1#11-4, or THP-1#11-7 cells. The amounts of produced viruses used to infect TZM-bl cells were normalized to p24 levels. Each bar represents the average of four independent experiments with SD. Data are represented as relative to Vif-proficient HIV-1 (WT). Statistical significance was assessed using the two-sided paired t test. *P < 0.05 compared to Vif-proficient HIV-1. The bottom panels are representative western blots of three independent experiments. The levels of indicated viral and cellular proteins in viral particles and whole-cell lysates are shown. p24 and HSP90 were used as loading controls. Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 11 Research Article mBio Our results demonstrate that A3F and A3G but not A3H proteins restrict Vif-deficient HIV-1 via deaminase-dependent and -independent mechanisms in THP-1 cells (Fig. 1, 3, and 4). In addition to A3F and A3G proteins, our findings indicate that at least one additional A3 protein is involved in Vif-deficient HIV-1 restriction via a deaminase-inde pendent mechanism (Fig. 3 and 4). Accordingly, the remaining four A3 proteins (A3A, A3B, A3C, and A3D) may contribute to Vif-deficient HIV-1 restriction in a deaminase-inde pendent manner in THP-1 cells (Fig. 4). However, A3A and A3B proteins are highly unlikely to contribute in this manner as A3A mRNA and protein expression levels are very low or undetectable in THP-1 cells without IFN treatment (Fig. 2C and D). Further, both A3A and A3B proteins are resistant to degradation by HIV-1 Vif (7, 34, 41–43). It is therefore plausible that A3C and A3D proteins contribute to Vif-deficient HIV-1 restriction in THP-1 cells. An A3C-isoleucine 188 variant is reportedly more active against HIV-1 than a serine 188 variant (44, 45). To ask which A3C variant is expressed by THP-1 cells, we determined the A3C genotypes of THP-1 cells using cDNA sequencing. These results demonstrated that the amino acid residue of A3C protein at position 188 is serine. This result indicates that A3C protein may have a modest effect on Vif-deficient HIV-1 restriction via a deaminase-independent mechanism in THP-1 cells, as shown in prior studies (44, 45). Similarly, the results of previous studies indicate that A3D protein has a weak effect on Vif-deficient HIV-1 restriction in 293, SupT11, and CEM2n cells (7, 8, 37, 46, 47). Nevertheless, the fact that Vif-deficient HIV-1 has 20% lower infectivity indicates that a synergistic mechanism may enhance the effect of A3 proteins on HIV-1 infectivity (48, 49). Further studies are required to fully elucidate the mechanisms underlying the effect of A3 proteins on HIV-1 infectivity. Similar to CD4+ T lymphocytes, HIV-1 can also target myeloid cells such as mono cytes and macrophages, and these infections are associated with viral dissemination, persistence, and latency (50, 51). Accordingly, it is important to understand the role of restriction factors, including A3 proteins, in myeloid cells. In monocytes, A3A mRNA levels are 10 to 1,000 times higher than other A3 mRNA expression levels, and A3A mRNA expression is reduced by 10- to 100-fold after differentiation into monocyte-derived macrophages (MDMs) (52–54). In contrast, A3G mRNA expression levels are reduced approximately 10-fold lower after differentiation of monocytes into MDMs (52, 53). A3F mRNA expression levels are less variable during the differentiation of monocytes into MDMs (52). Interestingly, suppression of A3A and A3G protein levels by siRNA reportedly leads to a four- to fivefold increase in p24 production by HIV-1-infected monocytes (53). As MDMs are generally more sensitive to HIV-1 infection than monocytes, it is highly likely that A3A and A3G proteins contribute to the lack of susceptibility of monocytes to HIV-1 infection. However, as previous studies have reported that A3A protein is less active against HIV-1 in 293T and SupT11 cell lines (7, 34, 55), further studies are required to determine the contribution of A3A protein to HIV-1 restriction in monocytes. In addition to A3A and A3G proteins, A3F and A3H proteins may be involved in HIV-1 restriction in monocytes. Although A3F mRNA expression levels are essentially unchanged during differentiation from monocytes into MDMs (53), A3F mRNA expres sion levels are comparable to A3G mRNA expression levels (53, 54), indicating that A3F protein likely contributes to HIV-1 restriction in monocytes. It is possible that only stable A3H haplotypes and the A3C-I188 variant are associated with HIV-1 restriction in monocytes. According to previous observations in 293, SupT11, and CEM2n cells (7, 8, 37, 46, 47), A3D protein may modestly contribute to HIV-1 restriction in monocytes. As A3B mRNA expression levels are relatively low, it is unlikely that this A3B protein inhibits HIV-1 in monocytes. However, the contribution of A3 proteins other than A3A and A3G proteins to HIV-1 suppression in monocytes remains unclear, and the antiviral activities of these A3 proteins warrant further investigation. In MDMs, A3A protein appears to be associated with anti-HIV-1 activity as increasing HIV-1 infectivity has been reported following siRNA knockdown of A3A gene (53, 54). In addition, HIV-1 replication assays in MDMs using HIV-1 Vif4A and Vif5A mutants demonstrated that the replication kinetics of both mutants were slower than that Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 12 Research Article mBio of the Vif-proficient HIV-1, indicating that A3D, A3F, and A3G proteins contribute to HIV-1 restriction in MDMs (39). However, the effects of A3D and A3F proteins on HIV-1 replication are donor-dependent, likely due to their respective expression levels (39). As the antiviral activity of A3B, A3C, and A3H proteins has not been reported in MDMs, further studies are required to address these issues. Vif is required for HIV-1 replication in CD4+ T lymphocytes and macrophages (2, 3, 17, 18). In the absence of Vif, HIV-1 is attacked by A3 proteins in CD4+ T lymphocytes, macrophages, monocytes, dendritic cells, and CD4+ T cell lines, and massive G-to-A mutations accumulate in HIV-1 proviral DNA (7, 8, 10, 15, 23, 26, 39, 56, 57). HIV-1 Vif recruits A3 proteins into an E3 ubiquitin ligase complex, thereby avoiding the antiviral activity of these proteins by promoting their degradation through a proteasome-medi ated pathway (2, 3, 17, 18). The primary function of Vif has long been posited to be the suppression of the antiviral activity of A3 proteins. On the other hand, Vif causes G2/M cell cycle arrest (58–60). As the amino acid residues of Vif responsible for G2/M cell cycle arrest do not completely match with the amino acid residues required for Vif-mediated A3 degradation, these functions of Vif may be independent of each other (61–63). In 2016, a functional proteomic analysis identified the PPP2R5 family of proteins, which function as regulators of protein phosphatase 2A, as novel targets of Vif (25). Subse quently, studies revealed that Vif induces G2/M arrest by degrading PPP2R5 proteins (60, 64, 65). Vif-induced G2/M arrest has been observed in many cell types, including 293T, SupT11, CEM-SS, and THP-1 cell lines, and primary CD4+ T lymphocytes (25, 61, 63). However, Vif-mediated G2/M arrest is not required for HIV-1 infection, supporting our findings that A3 family proteins are the sole essential substrate of Vif during infectious virus production from THP-1 cells under normal cell culture conditions (Fig. 3 to 5). It has recently been reported that fragile X mental retardation 1 and diphthamide biosynthesis 7 are degraded by Vif in CD4+ T lymphocytes (24). Further studies are required to determine whether a substrate of Vif other than A3 proteins is required for fully infectious HIV-1 production in vivo. In this study, we revealed an A3-dependent Vif function required for fully infectious HIV-1 production from THP-1 cells using only one lab-adapted virus (IIIB) and one TF virus (CH58) (Fig. 3 to 5). However, it is likely that Vif plays additional roles beyond A3 antagonism in some HIV-1 strains. Furthermore, the results of our pseudo-single cycle infectivity assays do not exclude the possibility that Vif may target non-A3 proteins required for HIV-1 replication. Moreover, we could not exclude the possibility that immunomodulatory effects may induce additional Vif targets other than A3 proteins in HIV-1-infected individuals. In summary, the findings of the studies here demonstrate that the primary target of Vif is the A3 family of proteins during infectious HIV-1 production from THP-1 cells (i.e., A3G, A3F, and potentially A3C and/or A3D proteins; unlikely A3A, A3B, or A3H hapI protein). Whether this observation is applicable to primary CD4+ T lymphocytes and myeloid cells, such as monocytes and macrophages, is important for the development of antiviral therapies targeting the A3-Vif axis. Such studies may contribute to a functional cure for HIV-1 by manipulating A3 mutagenesis. MATERIALS AND METHODS Cell lines and culture conditions 293T (CRL-3216) was obtained from American Type Culture Collection. TZM-bl (#8129) (66) was obtained from the NIH AIDS Reagent Program (NARP). The creation and characterization of the permissive T cell line SupT11 and the SupT11 single clones stably expressing untagged A3 (SupT11-vector, SupT11-A3F, SupT11-A3G, and SupT11- A3H hapII high) have been reported (10, 33). CEM-GXR (CEM-GFP expressing CCR5) was provided by Dr. Todd Allen (Harvard University, USA) (67). THP-1 was provided by Dr. Andrea Cimarelli (INSERM, France) (53). The generation and characterization of Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 13 Research Article mBio THP-1 ΔA3G#1 was reported (26). Adherent cells were cultured in DMEM (Wako, Cat# 044-29765) supplemented with 10% fetal bovine serum (FBS) (Nichirei, Cat#175012) and 1% penicillin/streptomycin (P/S) (Wako, Cat# 168-23191). Suspension cells were maintained in Roswell Park Memorial Institute Medium (RPMI) (Thermo Fisher Scientific, Cat# C11875500BT) with 10% FBS and 1% P/S. Genotyping of A3C and A3H genes Total RNA was isolated from THP-1 by RNA Premium Kit (NIPPON Genetics, Cat# FG-81250). Then, cDNA was synthesized by Transcriptor Reverse Transcriptase (Roche, Cat# 03531287001) and used to amplify A3C or A3H gene with the following primers: [A3C outer primers: (5´-GCG CTT CAG AAA AGA GTG GG) and (5´-GGA GAC AGA CCA TGA GGC); A3C inner primers: (5´-ACA TGA ATC CAC AGA TCA GAA A) and (5´-CCC CTC ACT GGA GAC TCT CC); A3H outer primers: (5´-CCA GAA GCA CAG ATC AGA AAC ACG AT) and (5´-GAC CAG CAG GCT ATG AGG CAA); A3H inner primers: (5´-TGT TAA CAG CCG AAA CAT TCC) and (5´-TCT TGA GTT GCT TCT TGA TAA T)]. The amplified fragments were cloned into the pJET cloning vector (Thermo Fisher Scientific, Cat# K1231). At least 10 independent clones were subjected to Sanger sequencing (Azenta) and sequence data were analyzed by Sequencher v5.4.6 (Gene Codes Corporation). Construction of pLentiCRISPR-Blast The pLentiCRISPR1000 system was previously described (68). pLentiCRISPR1000-Blast was generated by restriction digest with BmtI and MluI to excise the P2A-puromycin cassette. An oligo containing a P2A-blasticidin cassette was purchased from IDT (5´-AGC GGA GCT ACT AAC TTC AGC CTG CTG AAG CAG GCT GGC GAC GTG GAG GAG AAC CCT GGA CCT ACC GGT ATG GCC AAG CCA CTG TCC CAA GAA GAG TCA ACT CTG ATC GAG AGG GCC ACT GCA ACC ATT AAT AGC ATT CCC ATC TCT GAA GAC TAT AGC GTA GCT AGT GCC GCA CTC AGC TCT GAT GGA CGC ATA TTC ACC GGC GTT AAT GTC TAC CAC TTC ACC GGC GGA CCC TGC GCC GAA CTG GTC GTG CTG GGG ACC GCA GCC GCC GCG GCT GCC GGG AAT TTG ACG TGC ATT GTT GCA ATA GGC AAC GAG AAT AGG GGC ATC CTG TCA CCT TGC GGC CGG TGT CGG CAA GTG CTG CTG GAC CTG CAC CCC GGC ATC AAG GCC ATA GTC AAG GAT AGT GAT GGC CAG CCG ACC GCC GTT GGG ATT CGA GAA CTT CTG CCT TCT GGG TAC GTC TGG GAA GGC TAG) and amplified with the primers (5´-CAA GAC TAG TGG AAG CGG AGC TAC TAA CTT CAG CCT GCT GAA GCA GGC TGG CGA CGT GGA GGA and 5´-NNN NAC GCG TCT AGC CTT CCC AGA CGT ACC C) using high-fidelity Phusion polymerase (NEB, Cat# M0530S). The PCR fragment was digested with BmtI and MluI, and ligated into the cut pLentiCRISPR1000, producing pLentiCRISPR1000-Blast. Creation of THP-1 cells disrupting A3 genes An A3F specific guide for exon 3 was designed (Fig. S2A and C) and evaluated man ually for specificity to the A3F target sequence via an alignment with the most related members of the A3 family as described previously (26). Oligos with ends compatible with the Esp3I sites in pLentiCRISPR1000-Blast were purchased from IDT [ΔA3F gRNA: (5´-CAC CGG TAG TAG TAG AGG CGG GCG G) and (5´-CCA TCA TCA TCT CCG CCC GCC CAA G)]. The targeting construct was generated by annealing oligos and cloned by Golden Gate ligation into pLentiCRISPR1000-Blast. A guide with a common sequence among A3A exon 4, A3B exon 7, and A3G exon 7 was designed (Fig. 2A) and oligos with ends compatible with the Esp3I sites in pLentiCRISPR1000 (68) were purchased from IDT [PanZ1 gRNA: (5´-CAC CGT GGC CCG CAG CCT CCC ACT C) and (5´-GAA CGA GTG GGA GGC TGC GGG CCA C)]. The targeting construct was generated by annealing oligos and cloned by Golden Gate ligation into pLentiCRISPR1000 (68). All constructs were confirmed by Sanger sequencing (Azenta) and sequence data were analyzed by Sequencher v5.4.6 (Gene Codes Corporation). For transduction, VSV-G pseudotyped virus was generated by transfecting 2.5 µg of the pLentiCRISPR1000 or pLentiCRISPR1000-Blast targeting construct along with Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 14 Research Article mBio 1.67 µg of pΔ-NRF (HIV-1 gag, pol, rev, tat genes) (69) and 0.83 µg of pMD.G (VSV-G) expression vectors using TransIT-LT1 (Takara, Cat# MIR2306) into 293T cells. At 48 h post-transfection, viral supernatants were harvested, filtered with 0.45 µm filters (Merck, Cat# SLHVR33RB), and concentrated by centrifugation (26,200 × g, 4°C, 2 h). Then, viral pellets were resuspended in 10% FBS/RPMI and incubated with cells for 48 h. Forty- eight hours later, cells were placed under drug selection in 10% FBS/RPMI containing 1 µg/mL puromycin (InvivoGen, Cat# ant-pr) or 6 ng/mL blasticidin (InvivoGen, Cat# ant-bl). Single-cell clones were isolated by the limiting dilution of the drug-resistant cell pool and expanded. The expression levels of A3F protein in THP-1 ΔA3F#1 and #2 and THP-1ΔA3F/A3G#1 and #2 cells were confirmed by western blotting (see “Western blot” section). To confirm indels in the A3F target sequence of the selected clones, genomic DNA was isolated by DNeasy Blood & Tissue Kits (Qiagen, Cat# 69504) and amplified with Choice-Taq DNA polymerase (Denville Scientific, Cat# CB4050-2) using primers (5´-GCT GAA GTC GCC CTT GAA TAA ACA CGC and 5´-TGT CAG TGC TGG CCC CG). The amplified PCR products were cloned into the pJET cloning vector (Thermo Fisher Scientific, Cat# K1231) and subjected to Sanger sequencing (Azenta). To confirm indels in the A3A, A3B, and A3G target sequences of the selected clones (THP-1#11-4 and #11-7), genomic DNA was isolated by DNeasy Blood & Tissue Kits (Qiagen, Cat# 69504) and subjected to whole-genome sequencing (Macrogen). The WGS data were enrolled in the NCBI BioSample database and the respective accession numbers are SAMN35719796 for parental THP-1, SAMN35719797 for THP-1#11-4, and SAMN35719798 for THP-1#11-7. The sequencing data were aligned by Isaac aligner (iSAAC-04.18.11.09). Off-target sites were analyzed by Cas-OFFinder (http://www.rgenome.net/cas-offinder/). For further analysis of indels between A3A and A3G genes, genomic DNAs from THP-1#11-4 and #11-7 were amplified using primers (5´-GGG GCT TTC TGA AAG AAT GAG AAC TGG GC and 5´-CAG CTG GAG ATG GTG GTG AAC AGC C). The amplified PCR products were cloned into the pJET cloning vector (Thermo Fisher Scientific, Cat# K1231) and subjected to Sanger sequencing (Azenta). All sequence data were analyzed by Sequencher v5.4.6 (Gene Codes Corporation). To assess the expression levels of A3 mRNAs and proteins, THP-1 parent, #11-4, and #11-7 were incubated in 10% FBS/RPMI including 500 units/mL IFN (R&D Systems, Cat# 11200-2) for 6 h. Then, cells were harvested and subjected to RT-qPCR (see “RT-qPCR” section) (Fig. 2C) and western blotting (see “Western blot” section) (Fig. 2D). Pseudo-single cycle infectivity assays Vif-proficient and Vif-deficient (X26 and X27) HIV-1 IIIB C200 proviral expression constructs have been reported (70). HIV-1 IIIB C200 mutants with hyper- (H48 and 60EKGE63) and hypo-functional (V39) Vifs have been reported (10). An HIV-1 IIIB C200 Vif 5A mutant (40AAAAA44) has been described (26). HIV-1 IIIB C200 Vif 4A (14AKTK18) and 4A5A (14AKTK18 and 40AAAAA44) mutants were created by digesting pNLCSFV3-4A, and −4A5A proviral DNA construct [(37); kindly provided by Dr. Kei Sato, University of Tokyo, Japan] at SwaI and SalI sites and cloned into pIIIB C200 proviral construct. The proviral expression vector encoding full-length TF virus, CH58 (NARP, #11856) was obtained from the NARP. The creation of Vif-deficient CH58 mutant has been described previously (71). HIV-1 single-cycle assays using VSV-G pseudotyped viruses were performed as described previously (23, 26) (Fig. 1C). 293T cells were cotransfected with 2.4 µg of proviral DNA construct and 0.6 µg of VSV-G expression vector using TransIT-LT1 reagent (Takara, Cat# MIR2306) into 293T cells (3 × 106). Forty-eight hours later, supernatants were harvested, filtered (0.45 µm filters, Merck, Cat# SLHVR33RB), and used to titrate on 2.5 × 104 CEM-GXR reporter cells for MOI determinations. GFP+ cells were measured using a FACS Canto II (BD Biosciences) and the data were analyzed using FlowJo software v10.7.1 (BD Biosciences). One or 5 × 106 target cells were infected with an MOI of 0.05 (for SupT11 derivatives) or 0.25 (for THP-1 derivatives) and washed with phosphate-buffered saline (PBS) twice at 24 h postinfection and then incubated for an additional 24 h. After 24 h, supernatants were collected and filtered. The resulting viral particles were Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 15 Research Article mBio quantified by p24 ELISA (ZeptoMetrix, Cat# 0801008) and used to infect 1 × 104 TZM-bl cells (1 or 2 ng of p24). At 48 h postinfection, the infected cells were lysed with a Bright-Glo luciferase assay system (Promega, Cat# E2650) and the intracellular luciferase activity was measured by a Synergy H1 microplate reader (BioTek) or Centro XS3 LB960 microplate luminometer (Berthold Technologies). Quantification of LRT products Viruses were produced by infecting VSV-G pseudotyped virus into THP-1 cells as described above (see “Pseudo-single cycle infectivity assays” section) and the resulting viral particles were quantified by p24 ELISA (ZeptoMetrix, Cat# 0801008). The viral supernatants including 20 ng of p24 antigen were used for infection into SupT11 cells. At 12 h postinfection, cells were harvested and washed with PBS twice. Then, total DNA was isolated by DNeasy Blood & Tissue Kits (Qiagen, Cat# 69504) and treated with RNase A (Qiagen, Cat# 19101) according to the manufacturer’s instruction. Following DpnI digestion, 50 ng of DNA was used to amplify LRT products and CCR5 gene with the following primers: LRT forward: (5´-CGT CTG TTG TGT GAC TCT GG) and LRT reverse: (5´-TTT TGG CGT ACT CAC CAG TCG); CCR5 forward: (5´-CCA GAA GAG CTG AGA CAT CCG) and CCR5 reverse (5´-GCC AAG CAG CTG AGA GGT TAC T). qPCR was performed using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, Cat# 4367659) and fluorescent signals from resulting PCR products were acquired using a Thermal Cycler Dice Real Time System III (Takara). Finally, each LRT product was represented as values normalized by the quantity of the CCR5 gene (Fig. 4C). RT-qPCR Cells were harvested and washed with PBS twice. Then, total RNA was isolated by RNA Premium Kit (NIPPON Genetics, Cat# FG-81250) and cDNA was synthesized by Transcrip tor Reverse Transcriptase (Roche, Cat# 03531287001) with random hexamer. RT-qPCR was performed using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, Cat# 4367659). Primers for each A3 mRNA have been reported previously (72, 73): A3A forward: (5´-GAG AAG GGA CAA GCA CAT GG) and A3A reverse: (5´-TGG ATC CAT CAA GTG TCT GG); A3B forward: (5´-GAC CCT TTG GTC CTT CGA C) and A3B reverse: (5´-GCA CAG CCC CAG GAG AAG); A3C forward: (5´-AGC GCT TCA GAA AAG AGT GG) and A3C reverse: (5´-AAG TTT CGT TCC GAT CGT TG); A3D forward: (5´-ACC CAA ACG TCA GTC GAA TC) and A3D reverse: (5´-CAC ATT TCT GCG TGG TTC TC); A3F forward: (5´-CCG TTT GGA CGC AAA GAT) and A3F reverse: (5´-CCA GGT GAT CTG GAA ACA CTT); A3G forward: (5´-CCG AGG ACC CGA AGG TTA C) and A3G reverse: (5´-TCC AAC AGT GCT GAA ATT CG); A3H forward: (5´-AGC TGT GGC CAG AAG CAC) and A3H reverse: (5´-CGG AAT GTT TCG GCT GTT); TATA-binding protein (TBP) forward: (5´-CCC ATG ACT CCC ATG ACC) and TBP reverse: (5´-TTT ACA ACC AAG ATT CAC TGT GG). Fluorescent signals from resulting PCR products were acquired using a Thermal Cycler Dice Real Time System III (Takara). Finally, each A3 mRNA expression level was represented as values normalized by TBP mRNA expression levels (Fig. 2C). Hypermutation analyses Hypermutation analyses were performed as previously described (23, 26, 45). Genomic DNAs containing HIV-1 proviruses were recovered by infecting viruses produced in derivatives of THP-1 or SupT11 cells into SupT11 using DNeasy Blood & Tissue Kits (Qiagen, Cat# 69504). Following DpnI digestion, the viral pol region was amplified by nested PCR with outer primers (876 bp) [(5´-TCC ART ATT TRC CAT AAA RAA AAA) and (5´-TTY AGA TTT TTA AAT GGY TYT TGA)] and inner primers (564 bp) [(5´-AAT ATT CCA RTR TAR CAT RAC AAA AAT) and (5´-AAT GGY TYT TGA TAA ATT TGA TAT GT)]. The resulting 564 bp amplicon was subjected to pJET cloning. At least 10 independ ent clones were Sanger sequenced (Azenta) for each condition and analyzed by the Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 16 Research Article mBio HIV sequence database (https://www.hiv.lanl.gov/content/sequence/HYPERMUT/hyper mut.html). Clones with identical mutations were eliminated. Western blot Western blotting for cell and viral lysates was performed as described previously (23, 26, 74). Cells were harvested, washed with PBS twice, and lysed in lysis buffer [25 mM HEPES (pH7.2), 20% glycerol, 125 mM NaCl, 1% Nonidet P40 (NP40) substitute (Nacalai Tesque, Cat# 18558-54)]. After quantification of total protein by protein assay dye (Bio-Rad, Cat# 5000006), lysates were diluted with 2× SDS sample buffer [100 mM Tris-HCl (pH 6.8), 4% SDS, 12% β-mercaptoethanol, 20% glycerol, 0.05% bromophenol blue] and boiled for 10 min. Virions were dissolved in 2× SDS sample buffer and boiled for 10 min after pelleting down using 20% sucrose (26,200 × g, 4°C, 2 h). Then, the quantity of p24 antigen was measured by p24 ELISA (ZeptoMetrix, Cat# 0801008). Proteins in the cell and viral lysates (5 µg of total protein and 10 ng of p24 anti gen) were separated by SDS-PAGE and transferred to polyvinylidene difluoride (PVDF) membranes (Millipore, Cat# IPVH00010). Membranes were blocked with 5% milk in PBS containing 0.1% Tween 20 (0.1% PBST) and incubated in 4% milk/0.1% PBST contain ing primary antibodies: mouse anti-HSP90 (BD Transduction Laboratories, Cat# 610418, 1:5,000); rabbit anti-A3B (5210-87-13, 1:1,000) (75); rabbit anti-A3C (Proteintech, Cat# 105911-1-AP, 1:1,000); rabbit anti-A3F (675, 1:1,000) (76); rabbit anti-A3G (NARP, #10201, 1:2,500); rabbit anti-A3H (Novus Biologicals, NBP1-91682, 1:5,000): mouse anti-Vif (NARP, #6459, 1:2,000); mouse anti-p24 (NARP, #1513, 1:2,000). Subsequently, the membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies: donkey anti-rabbit IgG-HRP (Jackson ImmunoResearch, 711-035-152, 1:5,000); donkey anti-mouse IgG-HRP (Jackson ImmunoResearch, 715-035-150, 1:5,000). SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, Cat# 34095) or Super signal atto (Thermo Fisher Scientific, Cat# A38555) was used for HRP detection. Bands were visualized by the Amersham Imager 600 (Amersham). Statistical analyses Statistical significance was performed using a two-sided paired t test (Fig. 1D, 2C, 3B, 4C, 5, S1A, and S5A). GraphPad Prism software v8.4.3 was used for these statistical tests. ACKNOWLEDGMENTS We would like to thank Haruyo Hasebe, Kimiko Ichihara, Kazuko Kitazato, Otowa Takahashi, and all Ikeda lab members for technical assistance. We also would like to thank Drs. Todd Allen, Andrea Cimarelli, and Kei Sato for sharing reagents. This study was supported in part by AMED Research Program on Emerging and Re-emerging Infectious Diseases (JP21fk0108574, to Hesham Nasser; JP21fk0108494 to Terumasa Ikeda); AMED Research Program on HIV/AIDS (JP22fk0410055, to Terumasa Ikeda); JSPS KAKENHI Grant-in-Aid for Scientific Research C (22K07103, to Terumasa Ikeda); JSPS KAKENHI Grant-in-Aid for Early-Career Scientists (22K16375, to Hesham Nasser); JSPS Leading Initiative for Excellent Young Researchers (LEADER) (to Terumasa Ikeda); Takeda Science Foundation (to Terumasa Ikeda); Mochida Memorial Founda tion for Medical and Pharmaceutical Research (to Terumasa Ikeda); The Naito Founda tion (to Terumasa Ikeda); Shin-Nihon Foundation of Advanced Medical Research (to Terumasa Ikeda); Waksman Foundation of Japan (to Terumasa Ikeda); an intramural grant from Kumamoto University COVID-19 Research Projects (AMABIE) (to Terumasa Ikeda); Intercontinental Research and Educational Platform Aiming for Eradication of HIV/AIDS (to Terumasa Ikeda); International Joint Research Project of the Institute of Medical Science, the University of Tokyo (to Terumasa Ikeda); SPP1923 and the Heisenberg Program of the German Research Foundation (SA 2676/3-1; SA 2676/1-2) (to Daniel Sauter); as well as the Canon Foundation Europe (to Daniel Sauter). Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 17 Research Article mBio Contributions from the Harris lab were supported by NIAID R37-AI064046 and a Recruitment of Established Investigators Award from the Cancer Prevention and Research Institute of Texas (CPRIT RR220053). R.S.H. is an Investigator of the Howard Hughes Medical Institute and the Ewing Halsell President’s Council Distinguished Chair. The authors declare that they have no competing interests. AUTHOR AFFILIATIONS 1Division of Molecular Virology and Genetics, Joint Research Center for Human Retrovirus Infection, Kumamoto University, Kumamoto, Japan 2Graduate School of Medical Sciences, Kumamoto University, Kumamoto, Japan 3Department of Clinical Pathology, Faculty of Medicine, Suez Canal University, Ismailia, Egypt 4Department of Biochemistry and Structural Biology, University of Texas Health San Antonio, San Antonio, Texas, USA 5Howard Hughes Medical Institute, University of Texas Health San Antonio, San Antonio, Texas, USA 6Department of Biochemistry, Molecular Biology and Biophysics, University of Minnesota, Minneapolis, Minnesota, USA 7Institute for Molecular Virology, University of Minnesota, Minneapolis, Minnesota, USA 8Institute for Medical Virology and Epidemiology of Viral Diseases, University Hospital Tübingen, Tübingen, Germany AUTHOR ORCIDs Terumasa Ikeda Adam Z. Cheng Daniel Sauter Reuben S. Harris http://orcid.org/0000-0003-2869-9450 http://orcid.org/0000-0001-6277-9433 http://orcid.org/0000-0001-7665-0040 http://orcid.org/0000-0002-9034-9112 FUNDING Funder Japan Agency for Medical Research and Develop ment (AMED) Grant(s) Author(s) JP21fk0108574 Hesham Nasser Japan Agency for Medical Research and Develop ment (AMED) JP21fk0108494, JP22fk0410055 Terumasa Ikeda MEXT \| Japan Society for the Promotion of Science (JSPS) MEXT \| Japan Society for the Promotion of Science (JSPS) Takeda Science Foundation (TSF) MEXT \| Japan Society for the Promotion of Science (JSPS) Mochida Memorial Foundation for Medical and Pharmaceutical Research (公益財団法人持田記念医学薬学振興財団) Naito Foundation (内藤記念科学振興財団) Shinnihon Foundation of Advanced Medical Treatment Research (公益財団法人新日本先進医療研究財団) Waksman Foundation of Japan Canon Foundation in Europe (CFE) HHS \| NIH \| National Institute of Allergy and Infectious Diseases (NIAID) 22K07103 Terumasa Ikeda 22K16375 Hesham Nasser Terumasa Ikeda Terumasa Ikeda Terumasa Ikeda Terumasa Ikeda Terumasa Ikeda Terumasa Ikeda Daniel Sauter R37-AI064046 Reuben S. Harris Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 18 mBio Research Article AUTHOR CONTRIBUTIONS Terumasa Ikeda, Conceptualization, Investigation, Supervision, Validation, Writing – original draft, Writing – review and editing \| Ryo Shimizu, Conceptualization, Investiga tion, Validation, Writing – review and editing \| Hesham Nasser, Investigation, Validation, Writing – review and editing \| Michael A. Carpenter, Investigation, Validation, Writing – review and editing \| Adam Z. Cheng, Investigation, Validation, Writing – review and editing \| William L. Brown, Investigation, Validation, Writing – review and editing \| Daniel Sauter, Investigation, Validation, Writing – review and editing \| Reuben S. Harris, Investigation, Validation, Writing – review and editing ADDITIONAL FILES The following material is available online. Supplemental Material Figure S1 (mBio00782-23-s0001.tif). Pseudo-single cycle infectivity assays for each HIV-1 mutant in SupT11 cells stably expressing stable A3H haplotype. Supplemental figure legends (mBio00782-23-s0002.docx). Legends to Fig. S1 to S5. Figure S2 (mBio00782-23-s0003.tif). Development of A3F and A3F/A3G-null THP-1 cells. Figure S3 (mBio00782-23-s0004.tif). Sequence analysis of flanking region targeted by gRNA in THP-1#11-4 and #11-7. Figure S4 (mBio00782-23-s0005.tif). Deletions around predicted A3G pseudogene. Figure S5 (mBio00782-23-s0006.tif). Pseudo-single cycle infectivity assays for each HIV-1 mutant in SupT11 cells stably expressing A3. REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. Ikeda T, Yue Y, Shimizu R, Nasser H. 2021. Potential utilization of APOBEC3-mediated mutagenesis for an HIV-1 functional cure. Front Microbiol 12:686357. https://doi.org/10.3389/fmicb.2021.686357 Desimmie BA, Delviks-Frankenberrry KA, Burdick RC, Qi D, Izumi T, Pathak VK. 2014. Multiple APOBEC3 restriction factors for HIV-1 and one Vif to rule them all. J Mol Biol 426:1220–1245. https://doi.org/10.1016/j. jmb.2013.10.033 Harris RS, Dudley JP. 2015. APOBECs and virus restriction. Virology 479– 480:131–145. https://doi.org/10.1016/j.virol.2015.03.012 Koito A, Ikeda T. 2012. Apolipoprotein B mRNA-editing, catalytic polypeptide cytidine deaminases and retroviral restriction. Wiley Interdiscip Rev RNA 3:529–541. https://doi.org/10.1002/wrna.1117 Cheng AZ, Moraes SN, Shaban NM, Fanunza E, Bierle CJ, Southern PJ, Bresnahan WA, Rice SA, Harris RS. 2021. APOBECs and herpesviruses. Viruses 13:390. https://doi.org/10.3390/v13030390 Holmes RK, Malim MH, Bishop KN. 2007. APOBEC-mediated viral restriction: not simply editing? Trends Biochem Sci 32:118–128. https:// doi.org/10.1016/j.tibs.2007.01.004 Hultquist JF, Lengyel JA, Refsland EW, LaRue RS, Lackey L, Brown WL, Harris RS. 2011. Human and rhesus APOBEC3D, APOBEC3F, APOBEC3G, and APOBEC3H demonstrate a conserved capacity to restrict Vif- deficient HIV-1. J Virol 85:11220–11234. https://doi.org/10.1128/JVI. 05238-11 Refsland EW, Hultquist JF, Harris RS. 2012. Endogenous origins of HIV-1 G-to-A hypermutation and restriction in the nonpermissive T cell line CEM2n. PLoS Pathog 8:e1002800. https://doi.org/10.1371/journal.ppat. 1002800 Ooms M, Brayton B, Letko M, Maio SM, Pilcher CD, Hecht FM, Barbour JD, Simon V. 2013. HIV-1 Vif adaptation to human APOBEC3H haplotypes. Cell Host Microbe 14:411–421. https://doi.org/10.1016/j.chom.2013.09. 006 Refsland EW, Hultquist JF, Luengas EM, Ikeda T, Shaban NM, Law EK, Brown WL, Reilly C, Emerman M, Harris RS. 2014. Natural polymorphisms in human APOBEC3H and HIV-1 Vif combine in primary T lymphocytes to infectivity. PLoS Genet affect viral G-to-A mutation 10:e1004761. https://doi.org/10.1371/journal.pgen.1004761 levels and 11. Newman ENC, Holmes RK, Craig HM, Klein KC, Lingappa JR, Malim MH, Sheehy AM. 2005. Antiviral function of APOBEC3G can be dissociated from cytidine deaminase activity. Curr Biol 15:166–170. https://doi.org/ 10.1016/j.cub.2004.12.068 12. Wang X, Ao Z, Chen L, Kobinger G, Peng J, Yao X. 2012. The cellular antiviral protein APOBEC3G interacts with HIV-1 reverse transcriptase and inhibits its function during viral replication. J Virol 86:3777–3786. https://doi.org/10.1128/JVI.06594-11 Pollpeter D, Parsons M, Sobala AE, Coxhead S, Lang RD, Bruns AM, Papaioannou S, McDonnell JM, Apolonia L, Chowdhury JA, Horvath CM, Malim MH. 2018. Deep sequencing of HIV-1 reverse transcripts reveals the multifaceted antiviral functions of APOBEC3G. Nat Microbiol 3:220– 233. https://doi.org/10.1038/s41564-017-0063-9 13. 14. Holmes RK, Koning FA, Bishop KN, Malim MH. 2007. APOBEC3F can inhibit the accumulation of HIV-1 reverse transcription products in the absence of hypermutation. comparisons with APOBEC3G. J Biol Chem 282:2587–2595. https://doi.org/10.1074/jbc.M607298200 16. 15. Harris RS, Bishop KN, Sheehy AM, Craig HM, Petersen-Mahrt SK, Watt IN, Neuberger MS, Malim MH. 2003. DNA deamination mediates innate immunity to retroviral infection. Cell 113:803–809. https://doi.org/10. 1016/s0092-8674(03)00423-9 Rathore A, Carpenter MA, Demir Ö, Ikeda T, Li M, Shaban NM, Law EK, Anokhin D, Brown WL, Amaro RE, Harris RS. 2013. The local dinucleotide preference of APOBEC3G can be altered from 5'-CC to 5'-TC by a single amino acid substitution. J Mol Biol 425:4442–4454. https://doi.org/10. 1016/j.jmb.2013.07.040 Salamango DJ, Harris RS. 2020. Dual functionality of HIV-1 Vif in APOBEC3 counteraction and cell cycle arrest. Front Microbiol 11:622012. https://doi.org/10.3389/fmicb.2020.622012 Takaori-Kondo A, Shindo K. 2013. HIV-1 Vif: a guardian of the virus that opens up a new era in the research field of restriction factors. Front Microbiol 4:34. https://doi.org/10.3389/fmicb.2013.00034 Jäger S, Kim DY, Hultquist JF, Shindo K, LaRue RS, Kwon E, Li M, Anderson BD, Yen L, Stanley D, Mahon C, Kane J, Franks-Skiba K, Cimermancic P, Burlingame A, Sali A, Craik CS, Harris RS, Gross JD, Krogan NJ. 2011. Vif 18. 17. 19. Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 19 Research Article mBio 20. hijacks CBF-β to degrade APOBEC3G and promote HIV-1 infection. Nature 481:371–375. https://doi.org/10.1038/nature10693 Zhang W, Du J, Evans SL, Yu Y, Yu XF. 2011. T-cell differentiation factor CBF-β regulates HIV-1 Vif-mediated evasion of host restriction. Nature 481:376–379. https://doi.org/10.1038/nature10718 22. 21. Anderson BD, Harris RS. 2015. Transcriptional regulation of APOBEC3 antiviral immunity through the CBF-β/RUNX axis. Sci Adv 1:e1500296. https://doi.org/10.1126/sciadv.1500296 Ebrahimi D, Richards CM, Carpenter MA, Wang J, Ikeda T, Becker JT, Cheng AZ, McCann JL, Shaban NM, Salamango DJ, Starrett GJ, Lingappa JR, Yong J, Brown WL, Harris RS. 2018. Genetic and mechanistic basis for APOBEC3H alternative splicing, retrovirus restriction, and counteraction by HIV-1 protease. Nat Commun 9:4137. https://doi.org/10.1038/s41467- 018-06594-3 Ikeda T, Symeonides M, Albin JS, Li M, Thali M, Harris RS, Silvestri G. 2018. HIV-1 adaptation studies reveal a novel Env-mediated homeostasis mechanism for evading lethal hypermutation by APOBEC3G. PLoS Pathog 14:e1007010. https://doi.org/10.1371/journal.ppat.1007010 24. Naamati A, Williamson JC, Greenwood EJ, Marelli S, Lehner PJ, Matheson NJ. 2019. Functional proteomic atlas of HIV infection in primary human CD4+ T cells. Elife 8:e41431. https://doi.org/10.7554/eLife.41431 23. 26. 25. Greenwood EJ, Matheson NJ, Wals K, van den Boomen DJ, Antrobus R, Williamson JC, Lehner PJ. 2016. Temporal proteomic analysis of HIV infection reveals remodelling of the host phosphoproteome by lentiviral Vif variants. Elife 5:e18296. https://doi.org/10.7554/eLife.18296 Ikeda T, Molan AM, Jarvis MC, Carpenter MA, Salamango DJ, Brown WL, Harris RS. 2019. HIV-1 restriction by endogenous APOBEC3G in the myeloid cell line THP-1. J Gen Virol 100:1140–1152. https://doi.org/10. 1099/jgv.0.001276 Sadeghpour S, Khodaee S, Rahnama M, Rahimi H, Ebrahimi D. 2021. Human APOBEC3 variations and viral infection. Viruses 13:1366. https:// doi.org/10.3390/v13071366 27. 28. OhAinle M, Kerns JA, Li MMH, Malik HS, Emerman M. 2008. Antiretroele ment activity of APOBEC3H was lost twice in recent human evolution. Cell Host Microbe 4:249–259. https://doi.org/10.1016/j.chom.2008.07. 005 29. Wang X, Abudu A, Son S, Dang Y, Venta PJ, Zheng Y-H. 2011. Analysis of human APOBEC3H haplotypes and anti-human immunodeficiency virus type 1 activity. J Virol 85:3142–3152. https://doi.org/10.1128/JVI.02049- 10 31. 30. Nakano Y, Misawa N, Juarez-Fernandez G, Moriwaki M, Nakaoka S, Funo T, Yamada E, Soper A, Yoshikawa R, Ebrahimi D, Tachiki Y, Iwami S, Harris RS, Koyanagi Y, Sato K, Cullen BR. 2017. HIV-1 competition experiments in humanized mice show that APOBEC3H imposes selective pressure and promotes virus adaptation. PLoS Pathog 13:e1006606. https://doi. org/10.1371/journal.ppat.1006606 Starrett GJ, Luengas EM, McCann JL, Ebrahimi D, Temiz NA, Love RP, Feng Y, Adolph MB, Chelico L, Law EK, Carpenter MA, Harris RS. 2016. The DNA cytosine deaminase APOBEC3H haplotype I likely contributes to breast and lung cancer mutagenesis. Nat Commun 7:12918. https://doi.org/10. 1038/ncomms12918 Zhen A, Wang T, Zhao K, Xiong Y, Yu X-F. 2010. A single amino acid difference in human APOBEC3H variants determines HIV-1 Vif sensitivity. J Virol 84:1902–1911. https://doi.org/10.1128/JVI.01509-09 32. 34. 33. Albin JS, Brown WL, Harris RS. 2014. Catalytic activity of APOBEC3F is required for efficient restriction of Vif-deficient human immunodefi- ciency virus. Virology 450–451:49–54. https://doi.org/10.1016/j.virol. 2013.11.041 Bishop KN, Holmes RK, Sheehy AM, Davidson NO, Cho S-J, Malim MH. 2004. Cytidine deamination of retroviral DNA by diverse APOBEC proteins. Curr Biol 14:1392–1396. https://doi.org/10.1016/j.cub.2004.06. 057 Zheng Y-H, Irwin D, Kurosu T, Tokunaga K, Sata T, Peterlin BM. 2004. Human APOBEC3F is another host factor that blocks human immunode ficiency virus type 1 replication. J Virol 78:6073–6076. https://doi.org/10. 1128/JVI.78.11.6073-6076.2004 Land AM, Law EK, Carpenter MA, Lackey L, Brown WL, Harris RS. 2013. Endogenous APOBEC3A DNA cytosine deaminase is cytoplasmic and nongenotoxic. J Biol Chem 288:17253–17260. https://doi.org/10.1074/ jbc.M113.458661 35. 36. 37. 38. 39. 40. 41. Sato K, Takeuchi JS, Misawa N, Izumi T, Kobayashi T, Kimura Y, Iwami S, Takaori-Kondo A, Hu W-S, Aihara K, Ito M, An DS, Pathak VK, Koyanagi Y. 2014. APOBEC3D and APOBEC3F potently promote HIV-1 diversification and evolution in humanized mouse model. PLoS Pathog 10:e1004453. https://doi.org/10.1371/journal.ppat.1004453 Russell RA, Smith J, Barr R, Bhattacharyya D, Pathak VK. 2009. Distinct domains within APOBEC3G and APOBEC3F interact with separate regions of human immunodeficiency virus type 1 Vif. J Virol 83:1992– 2003. https://doi.org/10.1128/JVI.01621-08 Chaipan C, Smith JL, Hu W-S, Pathak VK. 2013. APOBEC3G restricts HIV-1 to a greater extent than APOBEC3F and APOBEC3De in human primary CD4+ T cells and macrophages. J Virol 87:444–453. https://doi.org/10. 1128/JVI.00676-12 Smith JL, Pathak VK. 2010. Identification of specific determinants of human APOBEC3F, APOBEC3C, and APOBEC3De and African green monkey APOBEC3F that interact with HIV-1 Vif. J Virol 84:12599–12608. https://doi.org/10.1128/JVI.01437-10 Yu Q, Chen D, König R, Mariani R, Unutmaz D, Landau NR. 2004. APOBEC3B and APOBEC3C are potent inhibitors of simian immunodefi- ciency virus replication. J Biol Chem 279:53379–53386. https://doi.org/ 10.1074/jbc.M408802200 42. Doehle BP, Schäfer A, Cullen BR. 2005. Human APOBEC3B is a potent inhibitor of HIV-1 infectivity and is resistant to HIV-1 Vif. Virology 339:281–288. https://doi.org/10.1016/j.virol.2005.06.005 43. Goila-Gaur R, Khan MA, Miyagi E, Kao S, Strebel K. 2007. Targeting APOBEC3A to the viral nucleoprotein complex confers antiviral activity. Retrovirology 4:61. https://doi.org/10.1186/1742-4690-4-61 44. Wittkopp CJ, Adolph MB, Wu LI, Chelico L, Emerman M. 2016. A single nucleotide polymorphism in human APOBEC3C enhances restriction of lentiviruses. PLoS Pathog 12:e1005865. https://doi.org/10.1371/journal. ppat.1005865 47. 45. Anderson BD, Ikeda T, Moghadasi SA, Martin AS, Brown WL, Harris RS. 2018. Natural APOBEC3C variants can elicit differential HIV-1 restriction activity. Retrovirology 15:78. https://doi.org/10.1186/s12977-018-0459-5 46. Dang Y, Wang X, Esselman WJ, Zheng Y-H. 2006. Identification of APOBEC3De as another antiretroviral factor from the human APOBEC family. J Virol 80:10522–10533. https://doi.org/10.1128/JVI.01123-06 Takei H, Fukuda H, Pan G, Yamazaki H, Matsumoto T, Kazuma Y, Fujii M, Nakayama S, Kobayashi IS, Shindo K, Yamashita R, Shirakawa K, Takaori- Kondo A, Kobayashi SS. 2020. Alternative splicing of APOBEC3D generates functional diversity and its role as a DNA mutator. Int J Hematol 112:395–408. https://doi.org/10.1007/s12185-020-02904-y 48. Desimmie BA, Burdick RC, Izumi T, Doi H, Shao W, Alvord WG, Sato K, Koyanagi Y, Jones S, Wilson E, Hill S, Maldarelli F, Hu W-S, Pathak VK. 2016. APOBEC3 proteins can copackage and comutate HIV-1 genomes. Nucleic Acids Res 44:7848–7865. https://doi.org/10.1093/nar/gkw653 49. Ara A, Love RP, Follack TB, Ahmed KA, Adolph MB, Chelico L. 2017. Mechanism of enhanced HIV restriction by virion coencapsidated cytidine deaminases APOBEC3F and APOBEC3G. J Virol 91:e02230-16. https://doi.org/10.1128/JVI.02230-16 50. Hendricks CM, Cordeiro T, Gomes AP, Stevenson M. 2021. The interplay in viral persistence. Front Microbiol of HIV-1 and macrophages 12:646447. https://doi.org/10.3389/fmicb.2021.646447 52. 51. Herskovitz J, Gendelman HE. 2019. HIV and the macrophage: from cell reservoirs to drug delivery to viral eradication. J Neuroimmune Pharmacol 14:52–67. https://doi.org/10.1007/s11481-018-9785-6 Peng G, Lei KJ, Jin W, Greenwell-Wild T, Wahl SM. 2006. Induction of APOBEC3 family proteins, a defensive maneuver underlying interferon- induced anti-HIV-1 activity. J Exp Med 203:41–46. https://doi.org/10. 1084/jem.20051512 Berger G, Durand S, Fargier G, Nguyen X-N, Cordeil S, Bouaziz S, Muriaux D, Darlix J-L, Cimarelli A, Emerman M. 2011. APOBEC3A is a specific inhibitor of the early phases of HIV-1 infection in myeloid cells. PLoS Pathog 7:e1002221. https://doi.org/10.1371/journal.ppat.1002221 Koning FA, Newman ENC, Kim E-Y, Kunstman KJ, Wolinsky SM, Malim MH. 2009. Defining APOBEC3 expression patterns in human tissues and hematopoietic cell subsets. J Virol 83:9474–9485. https://doi.org/10. 1128/JVI.01089-09 53. 54. 55. Aguiar RS, Lovsin N, Tanuri A, Peterlin BM. 2008. Vpr.A3A chimera inhibits HIV replication. J Biol Chem 283:2518–2525. https://doi.org/10.1074/jbc. M706436200 Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 20 Research Article mBio 56. 57. 58. Koning FA, Goujon C, Bauby H, Malim MH. 2011. Target cell-mediated editing of HIV-1 cDNA by APOBEC3 proteins in human macrophages. J Virol 85:13448–13452. https://doi.org/10.1128/JVI.00775-11 Pion M, Granelli-Piperno A, Mangeat B, Stalder R, Correa R, Steinman RM, Piguet V. 2006. APOBEC3G/3F mediates intrinsic resistance of monocyte- derived dendritic cells to HIV-1 infection. J Exp Med 203:2887–2893. https://doi.org/10.1084/jem.20061519 Sakai K, Dimas J, Lenardo MJ. 2006. The Vif and Vpr accessory proteins independently cause HIV-1-induced T cell cytopathicity and cell cycle arrest. Proc Natl Acad Sci U S A 103:3369–3374. https://doi.org/10.1073/ pnas.0509417103 59. Wang J, Shackelford JM, Casella CR, Shivers DK, Rapaport EL, Liu B, Yu X- F, Finkel TH. 2007. The Vif accessory protein alters the cell cycle of human immunodeficiency virus type 1 infected cells. Virology 359:243–252. https://doi.org/10.1016/j.virol.2006.09.026 Salamango DJ, Ikeda T, Moghadasi SA, Wang J, McCann JL, Serebrenik AA, Ebrahimi D, Jarvis MC, Brown WL, Harris RS. 2019. HIV-1 Vif triggers cell cycle arrest by degrading cellular PPP2R5 phospho-regulators. Cell Rep 29:1057–1065. https://doi.org/10.1016/j.celrep.2019.09.057 60. 61. DeHart JL, Bosque A, Harris RS, Planelles V. 2008. Human immunodefi- ciency virus type 1 Vif induces cell cycle delay via recruitment of the same E3 ubiquitin ligase complex that targets APOBEC3 proteins for degradation. J Virol 82:9265–9272. https://doi.org/10.1128/JVI.00377-08 Zhao K, Du J, Rui Y, Zheng W, Kang J, Hou J, Wang K, Zhang W, Simon VA, Yu X-F. 2015. Evolutionarily conserved pressure for the existence of distinct G2/M cell cycle arrest and A3H inactivation functions in HIV-1 Vif. Cell Cycle 14:838–847. https://doi.org/10.1080/15384101.2014. 1000212 62. 63. Du J, Rui Y, Zheng W, Li P, Kang J, Zhao K, Sun T, Yu X-F. 2019. Vif-CBFβ interaction is essential for Vif-induced cell cycle arrest. Biochem Biophys Res Commun 511:910–915. https://doi.org/10.1016/j.bbrc.2019.02.136 64. Marelli S, Williamson JC, Protasio AV, Naamati A, Greenwood EJ, Deane JE, Lehner PJ, Matheson NJ. 2020. Antagonism of PP2A is an independ ent and conserved function of HIV-1 Vif and causes cell cycle arrest. Elife 9:e53036. https://doi.org/10.7554/eLife.53036 65. Nagata K, Shindo K, Matsui Y, Shirakawa K, Takaori-Kondo A. 2020. Critical role of PP2A-B56 family protein degradation in HIV-1 Vif mediated G2 cell cycle arrest. Biochem Biophys Res Commun 527:257– 263. https://doi.org/10.1016/j.bbrc.2020.04.123 66. Wei X, Decker JM, Liu H, Zhang Z, Arani RB, Kilby JM, Saag MS, Wu X, Shaw GM, Kappes JC. 2002. Emergence of resistant human immunodefi- ciency virus type 1 inhibitor (T-20) monotherapy. Antimicrob Agents Chemother 46:1896–1905. https://doi. org/10.1128/AAC.46.6.1896-1905.2002 in patients receiving fusion 67. 68. Brockman MA, Tanzi GO, Walker BD, Allen TM. 2006. Use of a novel GFP reporter cell line to examine replication capacity of CXCR4- and CCR5- tropic HIV-1 by flow cytometry. J Virol Methods 131:134–142. https://doi. org/10.1016/j.jviromet.2005.08.003 Carpenter MA, Law EK, Serebrenik A, Brown WL, Harris RS. 2019. A lentivirus-based system for Cas9/gRNA expression and subsequent removal by cre-mediated recombination. Methods 156:79–84. https:// doi.org/10.1016/j.ymeth.2018.12.006 69. Naldini L, Blömer U, Gallay P, Ory D, Mulligan R, Gage FH, Verma IM, Trono D. 1996. In vivo gene delivery and stable transduction of nondividing cells by a lentiviral vector. Science 272:263–267. https://doi. org/10.1126/science.272.5259.263 70. Haché G, Shindo K, Albin JS, Harris RS. 2008. Evolution of HIV-1 isolates that use a novel Vif-independent mechanism to resist restriction by human APOBEC23G. Curr Biol 18:819–824. https://doi.org/10.1016/j.cub. 2008.04.073 72. 71. Hopfensperger K, Richard J, Stürzel CM, Bibollet-Ruche F, Apps R, Leoz M, Plantier J-C, Hahn BH, Finzi A, Kirchhoff F, Sauter D, Brockman MA, Smithgall TE. 2020. Convergent evolution of HLA-C downmodulation in HIV-1 and HIV-2. mBio 11. https://doi.org/10.1128/mBio.00782-20 Refsland EW, Stenglein MD, Shindo K, Albin JS, Brown WL, Harris RS. 2010. Quantitative profiling of the full APOBEC3 mRNA repertoire in lymphocytes and tissues: implications for HIV-1 restriction. Nucleic Acids Res 38:4274–4284. https://doi.org/10.1093/nar/gkq174 Burns MB, Lackey L, Carpenter MA, Rathore A, Land AM, Leonard B, Refsland EW, Kotandeniya D, Tretyakova N, Nikas JB, Yee D, Temiz NA, Donohue DE, McDougle RM, Brown WL, Law EK, Harris RS. 2013. APOBEC3B is an enzymatic source of mutation in breast cancer. Nature 494:366–370. https://doi.org/10.1038/nature11881 73. 74. Nasser H, Shimizu R, Ito J, Genotype to Phenotype Japan (G2P-Japan) Consortium, Saito A, Sato K, Ikeda T. 2022. Monitoring fusion kinetics of viral and target cell membranes in living cells using a SARS-CoV-2 spike- protein-mediated membrane fusion assay. STAR Protoc 3:101773. https:/ /doi.org/10.1016/j.xpro.2022.101773 Brown WL, Law EK, Argyris PP, Carpenter MA, Levin-Klein R, Ranum AN, Molan AM, Forster CL, Anderson BD, Lackey L, Harris RS. 2019. A rabbit monoclonal antibody against the antiviral and cancer genomic DNA mutating enzyme APOBEC3B. Antibodies (Basel) 8:47. https://doi.org/10. 3390/antib8030047 75. 76. Wang J, Shaban NM, Land AM, Brown WL, Harris RS. 2018. Simian immunodeficiency virus Vif and human APOBEC3B interactions resemble those between HIV-1 Vif and human APOBEC3G. J Virol 92:e00447-18. https://doi.org/10.1128/JVI.00447-18 Month XXXX Volume 0 Issue 0 10.1128/mbio.00782-23 21
10.1093_ijrl_eeac041	Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of epistemic injustice epistemic injustice Nuno Ferreira Publication date Publication date 01-02-2023 Licence Licence This work is made available under the CC BY 4.0 licence and should only be used in accordance with that licence. For more information on the specific terms, consult the repository record for this item. Document Version Document Version Published version Citation for this work (American Psychological Association 7th edition) Citation for this work (American Psychological Association 7th edition) Ferreira, N. (2023). Utterly unbelievable: the discourse of ‘fake’ SOGI asylum claims as a form of epistemic injustice (Version 1). University of Sussex. https://hdl.handle.net/10779/uos.23484497.v1 Published in Published in International Journal of Refugee Law Link to external publisher version Link to external publisher version https://doi.org/10.1093/ijrl/eeac041 Copyright and reuse: Copyright and reuse: This work was downloaded from Sussex Research Open (SRO). This document is made available in line with publisher policy and may differ from the published version. Please cite the published version where possible. Copyright and all moral rights to the version of the paper presented here belong to the individual author(s) and/or other copyright owners unless otherwise stated. For more information on this work, SRO or to report an issue, you can contact the repository administrators at [email protected]. Discover more of the University’s research at https://sussex.figshare.com/ International Journal of Refugee Law, 2022, Vol XX, No XX, 1–30 https://doi.org/10.1093/ijrl/eeac041 A RT I C L E S Utterly Unbelievable: The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice Nuno Ferreira* A B ST R A CT Media and political debates on refugees and migration are dominated by a discourse of ‘fake’ and ‘bogus’ asylum claims. This article explores how this discourse affects in acute ways those people claiming asylum on grounds of sexual orientation or gender identity (SOGI). In particular, the article shows how such a discourse of ‘fakeness’ goes far beyond the well-documented and often inadequate credibility assessments carried out by asylum authorities. By framing the analysis within the context of the scholarship on epistemic injustice, and by drawing on a large body of primary and secondary data, this article reveals how the discourse of ‘fake’ SOGI claims permeates the conduct not only of asylum adjudicators, but also of all other actors in the asylum system, including non-governmental organizations, support groups, legal representatives, and even asylum claimants and refugees themselves. Following from this theoretically informed exploration of primary data, the article concludes with the impossibility of determining the ‘truth’ in SOGI asylum cases, while also offering some guidance on means that can be employed to alleviate the epistemic injustice produced by the asylum system against SOGI asylum claimants and refugees. * Professor of Law, University of Sussex, United Kingdom. This contribution has been produced in the context of the ‘Sexual Orientation and Gender Identity Claims of Asylum: A European Human Rights Challenge’ (SOGICA Project) (<https://www.sogica.org>). The Project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No 677693). The author wishes to thank Carmelo Danisi, Moira Dustin, Nina Held, Charlotte Skeet, Bal Sokhi-Bulley, and Christina Miliou Theocharaki, as well as the anonymous journal reviewers, for their constructive feedback on earlier drafts. © The Author(s) 2023. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. • 1 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 2 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice 1. I N T RO D U C T I O N According to the United Nations High Commissioner for Refugees (UNHCR), at the end of 2021, there were 31.7 million refugees and people seeking asylum in the world.1 These individuals face numerous social and legal obstacles to obtaining international protection, including having to demonstrate the credibility of their asylum claim during the adjudica- tion process. It is the nature of refugee status determination procedures that claimants must establish their entitlement to international protection, and that authorities must scrutinize the evidence available. The credibility of asylum claims may be called into question either because different elements of the testimony are not consistent with each other (internal credibility), or the testimony is not consistent with information gathered by the asylum authorities (external credibility).2 While the need for such credibility assessment is not in itself problematic, even before the legal adjudication process starts, claimants are often al- ready labelled as ‘bogus’, their claims are presumed to be ‘fake’, and asylum authorities and the broader public alike adopt a sceptical – even a cynical – mindset.3 People are perceived as ‘potential fraudsters’ as soon as they file their asylum claims and, by assuming that their claims are ‘false’, States maintain control over their borders (for example, to reduce levels of immigration and feed into xenophobic and populist political discourses) without having to question the system of international protection or a State’s democratic credentials within the international community.4 Some researchers argue that decision makers in countries such as Spain, the United States of America (USA), and the United Kingdom (UK) seem to be trained to disbelieve5 and carry out their functions according to an ‘unwritten (meta) message of mistrust’.6 Existing scholarship has thus identified strong elements of willing- ness and consciousness in discrediting asylum claims independently of their merits.7 Discussions about ‘fake’ asylum claims are fuelled by, and contribute towards, broader anti-refugee and anti-migration rhetoric in the media and political debates.8 1 UNHCR, ‘Figures at a Glance’ <https://www.unhcr.org/en-au/figures-at-a-glance.html> ac- cessed 12 September 2022. 2 Gábor Gyulai and others, Credibility Assessment in Asylum Procedures: A Multidisciplinary Training 3 4 5 Manual, vol 1 (Hungarian Helsinki Committee 2013) 31. Jessica Anderson and others, ‘The Culture of Disbelief: An Ethnographic Approach to Understanding an Under-Theorised Concept in the UK Asylum System’ (2014) Refugee Studies Centre Working Paper Series No 102; James Souter, ‘A Culture of Disbelief or Denial? Critiquing Refugee Status Determination in the United Kingdom’ (2011) 1 Oxford Monitor of Forced Migration 48. Cécile Rousseau and Patricia Foxen, ‘Le Mythe du Réfugie Menteur: Un Mensonge Indispensable?’ [The Myth of the Lying Refugee: An Essential Lie?] (2006) 71 L’Evolution Psychiatrique 505, 506–07. Carol Bohmer and Amy Shuman, ‘Producing Epistemologies of Ignorance in the Political Asylum Application Process’ (2007) 14 Identities 603, 615. 6 Olga Jubany, ‘Constructing Truths in a Culture of Disbelief: Understanding Asylum Screening from Within’ (2011) 26 International Sociology 74, 81. Rousseau and Foxen (n 4) 510. 7 8 Gillian McFadyen, ‘The Language of Labelling and the Politics of Hospitality in the British Asylum System’ (2016) 18 British Journal of Politics and International Relations 599, 611–12; Giuseppe Salvaggiulo, ‘La Sentenza della Cassazione: “I Racconti dei Richiedenti Asilo sono Stereotipati e Troppo Simili Tra Loro”’ [The Supreme Court’s Decision: The Testimonies of Asylum Claimants Are Stereotyped and Too Similar to Each Other] La Stampa (16 January 2020) l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 3 of 30 It is also clear that the ‘genuine refugee is discursively constructed in a particular legal, political, and cultural context’.9 This affects in critical ways those asylum claims based on sexual orientation or gender identity (SOGI). SOGI claims require a discrete ana- lysis in this context on account of the particular issues they raise in relation to different aspects of asylum adjudication, especially the need for claimants to prove their SOGI identity, the role of private actors in persecution, the intense social prejudice against SOGI claimants, the role of legislation – namely criminalization – in the country of origin in sanctioning that prejudice, and the particular psychosocial challenges that these claimants face in terms of personal identity and community integration in the host State.10 SOGI claimants are often accused of ‘fabricating’ their stories,11 including in media pieces that build on the assumption that pretending a certain sexual orientation or gender identity is easy for the claimant and a sure-fire way of obtaining international protection.12 This is especially the case where there is evidence of persecution against sexual and gender minorities in particular countries of origin. However, there is no guarantee of ‘automatic protection’ under such circumstances. Claimants must still go through the refugee status determination procedure, and authorities often place particular emphasis on the credibility assessment of SOGI claims. Such an assessment may depend mostly on the claimant’s own testimony – checked against the available country of origin information (COI) – owing to the limited documentary or witness evidence generally available in such cases. Furthermore, the ‘genuineness of a LGBT refugee is prone to constant negotiation and renegotiation dependent on ongoing developments occurring within the wider cultural politics of immigration and global sexual politics’.13 As already explored by several authors, this cynical mindset in relation to SOGI claimants creates a damaging ‘culture of disbelief ’ in asylum authorities in several <https://www.lastampa.it/cronaca/2020/01/16/news/sentenza-choc-della-cassazione-i- racconti-dei-richiedenti-asilo-sono-stereotipati-e-troppo-simili-tra-loro-1.38339774/> accessed 12 September 2022; Mehta Suketu, ‘The Asylum Seeker’ (The New Yorker, 25 July 2011) <https:// www.newyorker.com/magazine/2011/08/01/the-asylum-seeker> accessed 12 September 2022. 9 Deniz Akin, ‘Discursive Construction of Genuine LGBT Refugees’ (2018) 23 Lambda Nordica 21, 23. 10 Nuno Ferreira, ‘Sexuality and Citizenship in Europe: Sociolegal and Human Rights Perspectives’ (2018) 27 Social and Legal Studies 253, 254. 11 Rousseau and Foxen (n 4). 12 Dan Bilefsky, ‘Gays Seeking Asylum in US Encounter a New Hurdle’ The New York Times (29 January 2011) <https://www.nytimes.com/2011/01/29/nyregion/29asylum.html> accessed 12 September 2022; Francesca Ronchin, ‘Permessi di Soggiorno per i Migranti, L’Escamotage dell’Orientamento Sessuale’ [Residence Permits for Migrants, the Deception of Sexual Orientation] Corriere della Sera (23 October 2019) <https://www.corriere.it/video- articoli/2019/10/23/permessi-soggiorno-migranti-l-escamotage-dell-orientamento-sessuale/ ece27a72-e52c-11e9-b924-6943fd13a6fb.shtml> accessed 12 September 2022. Most of the primary data and secondary sources explored in this article refer more explicitly to sexual orientation but also hold relevance in relation to gender identity, hence the scope of the article encompassing both. 13 Akin (n 9) 36. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 4 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice countries.14 In 2001, when deciding a SOGI asylum claim, a UK judge raised the possibility of ‘encouraging a flood of fraudulent Zimbabwean (and no doubt other) asylum-seekers posing as sodomites’.15 Although we have come a long way since then, an ingrained concern persists that SOGI asylum claimants may be lying about their stories. Although SOGI claims may not be statistically more prone to being used in a deceptive way,16 and while acknowledging that they may indeed be used in a deceptive way, SOGI claimants are deeply affected by the scepticism that accompanies their asylum claims. Despite this culture of disbelief being well known to scholars, policymakers, and refugees, there is limited research on what makes SOGI claims – or claimants – so un- believable as to render them ‘fake’ in the eyes of decision makers, especially in light of the thorough, objective, individualized, and sensitive process that is required to assess their claims.17 It is crucial to explore in an in-depth manner the mechanisms behind such presumptions of ‘fakeness’. This article does so through a novel, theoretically and empirically informed analysis that examines all actors in the asylum system. The analysis reveals that the discourse of ‘fake’ SOGI claims not only strongly influences asylum au- thorities (often under political pressure to refuse claims, or hardened by listening to so many terrible stories) and the wider public (influenced by populist, racist, and homo/ transphobic social trends), but also affects the most unlikely stakeholders: on the one hand, non-governmental organizations (NGOs), support groups, and legal representa- tives take it upon themselves to filter out ‘fake’ claims from the asylum system, and, on the other hand, other SOGI claimants and refugees consider it necessary to themselves identify ‘fake’ claimants in order to contribute to the groups that support them and to protect the chances of future ‘genuine’ SOGI asylum claimants obtaining international protection. This article offers a theoretically informed analysis of these dynamics by engaging with this subject matter from the perspective of the body of literature on epistemic in- justice. The analysis is also empirically informed, drawing extensively on primary data collected through fieldwork carried out in several locations in Europe between 2017 14 Carmelo Danisi and others, Queering Asylum in Europe: Legal and Social Experiences of Seeking International Protection on Grounds of Sexual Orientation and Gender Identity (Springer 2021) ch 7; Agathe Fauchier, ‘Kosovo: What Does the Future Hold for LGBT People?’ (2013) 42 Forced Migration Review 36, 38; Theo Gavrielides and others, ‘Supporting and Including LGBTI Migrants: Needs, Experiences and Good Practices (Epsilon Project)’ (IARS International Institute 2017); Jenni Millbank, ‘From Discretion to Disbelief: Recent Trends in Refugee Determinations on the Basis of Sexual Orientation in Australia and the United Kingdom’ (2009) 13 International Journal of Human Rights 391. 15 Z v Secretary of State for the Home Department [2001] UKIAT 01TH02634, para 4. 16 John Vine, ‘An Investigation into the Home Office’s Handling of Asylum Claims Made on the Grounds of Sexual Orientation: March–June 2014’ (Independent Chief Inspector of Borders and Immigration 2014) para 5.21. 17 UNHCR, ‘Guidelines on International Protection No 9: Claims to Refugee Status Based on Sexual Orientation and/or Gender Identity within the Context of Article 1A(2) of the 1951 Convention and/or Its 1967 Protocol relating to the Status of Refugees’, HCR/GIP/12/09 (23 October 2012) para 62. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 5 of 30 and 2019.18 This fieldwork – carried out in the context of the ‘Sexual Orientation and Gender Identity Claims of Asylum’ (SOGICA Project) – concentrated on Council of Europe and European Union (EU) institutions, and the countries of Germany, Italy, and the UK. It included: 143 semi-structured interviews with SOGI asylum claimants and refugees, NGOs, policymakers, decision makers, members of the judiciary, legal repre- sentatives, and other professionals; 16 focus groups with SOGI asylum claimants and refugees; 24 non-participant contextual observations of court hearings; two online sur- veys of SOGI asylum claimants and refugees and professionals working with them; and freedom of information requests relating to case studies lodged in all three countries. In order to ensure anonymity, respect participants’ agency, and distinguish between the sources, the article uses sources in the following ways: individuals are referred to either by their first name or by a pseudonym (according to their stated preference); references note the capacity in which participants were interviewed and the country in which they were based (if no capacity is specified, then the participant was an asylum claimant or a legally recognized refugee); focus groups are identified by their number and location; court hearings are identified by the level of the court, its broad geographical location, and the year in which the hearing took place; and survey respondents are referred to by a letter (S for ‘supporter’ and C for ‘claimant’) and a numerical identifier.19 The article begins with a discussion of the theoretical framework on which the sub- sequent analysis relies, with an emphasis on the relevance of the scholarship on epi- stemic injustice for asylum law and policy (part 2). In part 3, the analysis of the primary data begins by exploring how epistemic injustice operates during the asylum adjudica- tion process, and how epistemic injustice is produced by asylum decision makers. In part 4, the focus shifts to the roles of NGOs, support groups, and legal representatives, as well as asylum claimants and refugees themselves, who are often ignored in such de- bates but are undoubtedly also key actors in the discourse of ‘fake’ claims, as evidenced by the primary data. Part 5 explores key means to address the epistemic injustice pro- duced by the actors discussed in parts 3 and 4, even though achieving the ‘truth’ is ultimately impossible. Finally, part 6 reiterates the need to accept the impossibility of determining the ‘truth’ in SOGI asylum claims and to alleviate the epistemic injustice of the asylum system for SOGI claimants. ‘Fake’ and ‘truth’ are used with quotation marks throughout the article to high- light the impossibility of determining the veracity of claims. Even when a claimant may acknowledge not having a genuine SOGI claim, their sexual orientation or gender identity may, in fact, be relevant to their need for international protection, although the claimant may choose to deny this, owing to emotional, social, or cul- tural factors. 18 Ethics approval was obtained from the University of Sussex (certificate of approval for Ethical Review ER/NH285/1). Written and informed consent was obtained from all the participants. The project – including the collection of empirical data – was carried out by all the team mem- bers: Carmelo Danisi, Moira Dustin, Nuno Ferreira, and Nina Held. For full details of the methodology, see Danisi and others (n 14) ch 2; SOGICA, ‘Fieldwork’ <https://www.sogica.org/en/fieldwork/> accessed 12 September 2022. 19 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 6 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice 2. C R E AT I N G E P I ST E M I C I N J U ST I C E I N T H E Q U E ST F O R ‘ T R U T H ’ As Foucault’s work so thoroughly explores, the quest for producing ‘truth’ has been central to the production of knowledge in the West – including in relation to sexuality – and is deeply embedded in subjective relationships of power.20 More specifically: Truth is a thing of this world: it is produced only by virtue of multiple forms of constraint. And it induces regular effects of power. Each society has its regime of truth, its ‘general politics’ of truth: that is, the types of discourse which it accepts and makes function as true; the mechanisms and instances which enable one to distinguish true and false statements, the means by which each is sanctioned; the techniques and procedures accorded value in the acquisition of truth; the status of those who are charged with saying what counts as true.21 Similarly, Bourdieu suggests that constructing a discourse as ‘true’ or ‘false’ essentially depends on the power dynamics that underpin social and institutional relationships.22 As explained by Spivak, there are a range of historical and ideological factors that pre- vent those inhabiting the ‘periphery’ – surely including asylum claimants and refugees – from being heard.23 All these scholarly contributions point to the fact that interper- sonal and institutional ‘power’ is a factor that shapes how we produce ‘truths’ and ‘lies’. Moreover, ‘truths’ and ‘lies’ are not produced according to what is ‘true’ or ‘false’ (if it were ever possible to determine this), but according to what is convenient, to order events around conformity and deviance.24 Consequently, epistemic injustice – under- stood here as injustice in the context of the production of knowledge – is rife in any system of ‘truth production’. In other words, no matter how a society produces know- ledge, there is bound to be unfairness as to who decides what is true or not, and how this is done. In the context of asylum law and policy, this includes two main forms of injustice: testimonial injustice and contributory injustice. On the one hand, testimonial injustice occurs when ‘prejudice causes a hearer to give a deflated level of credibility to a speaker’s word’,25 with such prejudice operating in relation to all different spheres of life that may affect a person’s social identity in the mind of the hearer. This entails a symbolic degradation, namely the listener undermines the other’s humanity,26 and oppresses the other by diminishing their self-confidence and thwarting their development.27 On the other hand, building on Pohlhaus’s work on 20 Michel Foucault, The History of Sexuality, vol 1 (Penguin Books 1990). 21 Michel Foucault, The Foucault Reader (Penguin Books 1991) 72–73. 22 Pierre Bourdieu, Language and Symbolic Power (Polity Press 1993). 23 Rosalind Morris (ed), Can the Subaltern Speak? Reflections on the History of an Idea (Columbia University Press 2010); Gayatri Chakravorty Spivak, ‘Can the Subaltern Speak?’ in Cary Nelson (ed), Marxism and the Interpretation of Culture (Macmillan Education 1988). 24 Michel de Certeau, Histoire et Psychanalyse Entre Science et Fiction (Gallimard 1987). 25 Miranda Fricker, Epistemic Injustice: Power and the Ethics of Knowing (Oxford University Press 2007) 1. ibid 44. ibid 58. 26 27 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 7 of 30 ‘willful hermeneutical ignorance’,28 Dotson sees contributory injustice ‘as the circum- stance where an epistemic agent’s willful hermeneutical ignorance in maintaining and utilizing structurally prejudiced hermeneutical resources thwarts a knower’s ability to contribute to shared epistemic resources within a given epistemic community by com- promising her epistemic agency’.29 Nonetheless, epistemic injustice (also) derives from the fact that ‘institutions struc- ture interactions according to cultural norms that impede parity of participation’.30 As Doan explains, this prevents ‘people from testifying and being heard, asking relevant questions, contesting claims and standards of evidence, and otherwise participating in everyday epistemic practices as peers’31 – something that is directly relevant to the asylum system. Consequently, Doan submits that ‘epistemic injustice ought to be under- stood as rooted in the oppressive and dysfunctional epistemic norms undergirding ac- tual communities and institutions’.32 As such, struggles for epistemic recognition require changes not only at the individual level but also at the social and institutional levels. The responsibility and the initiative for undoing epistemic injustice rest not only with single individuals but with all actors in the system, without ‘occluding the agency and resistance of victims’.33 This is of direct relevance for present purposes, since all actors in the asylum system contribute to epistemic injustice which, in turn, affects SOGI asylum claimants and refugees. In fact, a transformative strategy that is able to ‘correct unjust outcomes precisely by restructuring the underlying generative framework’ may be required.34 Asylum systems are textbook examples of how the State can devise and operation- alize repressive and flawed epistemic norms. States deploy political technologies to govern the movement and conduct of refugees, namely by determining which ones are ‘bogus refugees’ and which ones are ‘persons in real need of protection’.35 Looking at asylum systems through a Foucauldian and Fanonian lens, Lorenzini and Tazzioli adopt poststructural and decolonial prisms to highlight how: the question of (the production of) truth is at the core of the mechanisms of sub- jection and subjectivation which are at stake in the processing of asylum claims. Asylum seekers are usually seen as suspect subjects who have to demonstrate that 28 Gaile Pohlhaus, ‘Relational Knowing and Epistemic Injustice: Toward a Theory of Willful Hermeneutical Ignorance’ (2012) 27 Hypatia 715. 29 Kristie Dotson, ‘A Cautionary Tale: On Limiting Epistemic Oppression’ (2012) 33 Frontiers: A Journal of Women Studies 24, 32. 30 Nancy Fraser, ‘Social Justice in the Age of Identity Politics: Redistribution, Recognition, and Participation’ in Nancy Fraser and Axel Honneth (eds), Redistribution or Recognition? A Political- Philosophical Exchange (Verso Books 2003) 29. 31 Michael Doan, ‘Resisting Structural Epistemic Injustice’ (2018) 4(4) Feminist Philosophy Quarterly 13. ibid 15. ibid 8 (emphasis in original). Fraser (n 30) 74. 32 33 34 35 Daniele Lorenzini and Martina Tazzioli, ‘Confessional Subjects and Conducts of Non-Truth: Foucault, Fanon, and the Making of the Subject’ (2018) 35 Theory, Culture & Society 71, 72. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 8 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice they really are in need of protection; yet, at the same time, they are considered as subjects incapable of telling the truth.36 In this process, more than ‘truth’, we are in the presence of the ‘production of ignor- ance’,37 showing that both ‘truth’ and ‘fakeness’ are discursively constructed. During the production of knowledge in the asylum system, there is a clear ‘struggle over truth’.38 By default, asylum systems privilege the epistemic resources of decision makers over claimants, thus legitimizing the former’s prerogative to ‘arbitrarily and am- biguously misinterpret asylum applicants’ experiences, cultures, and countries’ – the so-called ‘institutional comfort’ enjoyed by decision makers.39 In the asylum context, this institutional comfort translates into testimonial injustice in the form of denying ap- plicants’ experiences, ignoring available information, and deciding which information or criteria to use. Simultaneously, the asylum system is characterized by contributory injustice in the form of knowingly and voluntarily employing prejudiced hermeneut- ical resources to undermine the epistemic agency of the claimants.40 Testimonial and contributory injustice combined produce a powerful version of epistemic injustice in asylum systems. In the midst of such an epistemologically unfair system, asylum claimants may find themselves both dehumanized and ignored. Doubting the truth of the claimant is a vio- lence perpetrated against them, which produces and increases their (narrative) vulner- ability, and constitutes a form of epistemological and symbolic violence.41 At the same time, decision makers may see their personal experiences as universal and therefore suitable to be used as the basis for judging the veracity of claimants’ testimonies.42 As Jubany concluded from her research in Spain and the UK, based on decision makers’ ‘professional knowledge’, Chinese claimants are held to be untrustworthy, African claimants are perceived as liars, those from the Indian subcontinent are accused of being incoherent and using artificial stories, and those from Turkey are judged as cun- ning and exaggerated.43 ‘Intuition’, having a ‘feeling’, ‘just knowing’, or a certain ‘look’ are seen as legitimate means to determine the truthfulness of a claimant’s story and are used as justification for denying international protection.44 Even worse, the use of accelerated procedures (often coupled with the contested notion of ‘safe country’)45 36 ibid 72 (citations omitted). 37 Bohmer and Shuman (n 5). 38 Lorenzini and Tazzioli (n 35) 82. 39 Ezgi Sertler, ‘The Institution of Gender-Based Asylum and Epistemic Injustice: A Structural Limit’ (2018) 4(3) Feminist Philosophy Quarterly 3. ibid 2, 16. 40 41 Massimo Prearo, ‘The Moral Politics of LGBTI Asylum: How the State Deals with the SOGI Framework’ (2020) 34 Journal of Refugee Studies 1454. 42 Rousseau and Foxen (n 4) 511. 43 Jubany (n 6) 83–84. ibid 86–87; Rousseau and Foxen (n 4) 516. 44 45 Danisi and others (n 14) ch 6.7. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 9 of 30 has rendered claimants’ speech ‘increasingly irrelevant’,46 depriving them of the oppor- tunity to fully articulate their experiences and fears of persecution.47 Reaching an ‘objective truth’ is not achievable, just as proving that a claim is ‘fake’ is not possible.48 In other words, ‘the pretense of judgment based on evidence obscures the real problem of the unavailability of necessary information’.49 Barsky notes that ‘we cannot employ the tools of discourse analysis, no matter how sophisticated, to distin- guish between truthful and untruthful statements in refugee hearings, except at a very superficial level’.50 ‘Fake’ claims are thus discursively produced: it is the discourse cre- ated by all the actors involved that labels claims as ‘fake’ and forms the subject position of the ‘fake’ claimant. This is true for SOGI claims as well: it is not possible to reach an ‘objective truth’ about them but, in the face of the ‘practical decisionism’ that asylum authorities face, ‘the various organizations and persons that claim that it is impossible to evaluate legitimately the truths of LGBT-ness are unsuccessful’.51 Historically, mem- bers of SOGI minorities had to hide their true identity and desires – and so society was full of ‘fake heterosexuals’ – but now, in a sort of inversion of the ‘politics of truth’, the fear is one of ‘fake homosexuals’.52 In this tangled web of the ‘politics of truth’, decision makers and other actors may overlook the fact that both sexual orientation and gender identity are socially constructed, culturally heterogeneous, fluid, complex, performed, and negotiated categories.53 A greater awareness of the nature of sexual orientation and gender identity would facilitate asylum decisions that more sensitively and accurately engage with SOGI claims, in ways that are also more socially and culturally appropriate. In a Foucauldian sense, the ‘fake’ SOGI claim and ‘fake’ SOGI claimant’s subject position are (also) discursively produced, thus constituting a sub-category of ‘fake’ claims. As a consequence, ‘only those whose sexual and gender practices are intelligible according to hegemonic gender and sexuality norms can become eligible for permitted border-crossing’, thus further entrenching the fixed, homonormative sexual ontologies 46 Lorenzini and Tazzioli (n 35) 82. 47 An infamous version of this phenomenon can be seen in the UK’s Detained Fast Track system for detained individuals, whereby people were deported without being given the opportunity to appeal against negative Home Office decisions. The system was declared unlawful by the High Court in Detention Action v First-tier Tribunal (Immigration and Asylum Chamber) [2015] EWHC 1689 (Admin). The negative practical consequences of such systems are illustrated in the case of PN, a Ugandan lesbian claimant: see PN (Uganda) v Secretary of State for the Home Department [2020] EWCA Civ 1213. 48 Rousseau and Foxen (n 4) 518. 49 Bohmer and Shuman (n 5) 622. 50 Robert F Barsky, Arguing and Justifying: Assessing the Convention Refugees’ Choice of Moment, Motive and Host Country (Ashgate Publishing 2000) 14. 51 Maja Hertoghs and Willem Schinkel, ‘The State’s Sexual Desires: The Performance of Sexuality in the Dutch Asylum Procedure’ (2018) 47 Theory and Society 691, 697. 52 Eric Fassin and Manuela Cordero Salcedo, ‘Becoming Gay? Immigration Policies and the Truth of Sexual Identity’ (2015) 44 Archives of Sexual Behavior 1117, 1121. ibid 1121–24. 53 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 10 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice that underlie the asylum system.54 The asylum system adopts a ‘privileged configur- ation of sexual orientation [that] reflects a particular historical configuration of gen- dered, raced and classed interests and experiences’.55 Moreover, while not every denial of international protection to a SOGI claimant is an instance of epistemic injustice, the asylum adjudication process becomes a ‘test of sexual veracity by means of a truthful performance’, on the basis of the ‘facticity of sexuality’, thereby legitimizing and sanc- tioning certain gender and sexuality performances but not others.56 The following parts of the article explore how all actors in the asylum system play a role in the ‘politics of truth’ of SOGI claims. 3. T H E ‘ U N T R U T H ’ O F T H E A S Y LU M A D J U D I C AT I O N P RO C E D U R E The evidence examined for this article revealed that at both an administrative and ju- dicial level, there is significant institutional comfort relating to SOGI-based asylum claims (see part 2). Decision makers may not only be sceptical about such claims, but may deny that there is any ‘truth’ to them. Through their disbelief, decision makers exer- cise their power to produce testimonial injustice and reduce the humanity of claimants. As Victor – a SOGI asylum claimant participant in the UK – put it, decision makers: wouldn’t want to listen to you. … If you try to explain something [to] the person, it is like you are offending them for you being there to, you know, to understand for them, you are already offending them [and] everything you are saying is not true.57 Decision makers’ role in the production of epistemic injustice is also apparent in their inclination to believe that a SOGI claim is ‘fake’ when there is simply an increase in the number of such claims.58 For example, Titti, a decision maker in Italy, spoke of ‘huge peaks’ in SOGI claims, of having heard about 15 such claims in one month in an Italian region, which prompted her to examine them more carefully. Bilal, a UK Home Office presenting officer, also expressed scepticism after an increase in SOGI claims: ‘I think I have had some cynicism … the gay Pakistani cases, because there seemed suddenly to be a huge raft and they all had very similar narratives’. Similarly, in Germany, an NGO participant reported that even gay decision makers were ‘extremely suspicious’ about a rise in SOGI asylum claims, thus leading to an increase in the number of rejections59 and demonstrating decision makers’ power to deny the ‘truth’ of claimants’ testimonies. 54 Mariska Jung, ‘Logics of Citizenship and Violence of Rights: The Queer Migrant Body and the Asylum System’ (2015) 3 Birkbeck Law Review 311, 324. 55 David AB Murray, ‘Real Queer: “Authentic” LGBT Refugee Claimants and Homonationalism in the Canadian Refugee System’ (2014) 56 Anthropologica 21, 22. 56 Hertoghs and Schinkel (n 51) 691, 693. 57 Focus Group No 2, Glasgow, UK. 58 Celeste, social worker, Italy. Several participants described them as ‘fashionable stories’. 59 Thomas, NGO volunteer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 11 of 30 Another recurrent theme in the discourse of ‘fake’ claims is the degree of similarity between different claimants’ testimonies. Maria Grazia, a decision maker in Italy, be- came aware of this soon after assuming her role: I realised how much the SOGI element is exploited. It is not a perception induced by a particularly backward policy from a certain political field. This element is really used to get protection … Yes, when I realised that the stories are all similar. … [T]he first time I made an appeal before a Court, I had an asylum claimant who had brought me a page from a newspaper in Nigeria where there was a photo of a man on the ground full of blood and a photo of the applicant, wanted for homosexuality. And I thought ‘Damn, how will the judge not believe this story? It is also in the newspaper’. And in the commission they told me ‘Look, these are photomontages and in reality the story they bring is always this: relationship with the partner, partner killed because of being homosexual, escape …’ And the grim, particularly violent element is always added in. Similarly, a German judge, Oscar, said that: the more you have listened to asylum claimants from a country, the sooner you will notice whether this really happens [claimants using fake stories] or if that is more likely. These are stories that are passed on from asylum claimant to asylum claimant and which they always try to use here [in court]. So, typical stories. A similarity between stories can, however, also be due to legal representatives some- times promoting ‘pre-prepared’ stories to their clients,60 which can lead to more rejec- tions by the authorities. In any case, it is clear that such similarities prompt decision makers to use their power to undervalue testimonies and interpret evidence in a way that undermines it, thus producing testimonial and contributory injustice. Interestingly, unique stories are also often seen as questionable, as they do not fit the scenarios familiar to decision makers.61 For example, Sofia and Emma, NGO workers in Germany, explained that asylum authorities may reject the ‘truth’ of a claimant’s tes- timony simply because it is different from other asylum claims: one [woman] who has experienced forced prostitution in China, so from Uganda to China, then she had different [experiences], then fled to other African coun- tries, where she was raped, and then [fled] again to Germany, where she has been almost forcibly prostituted. And … she is also lesbian, and with her partner, so to speak, and different things … escaped, and so, for the Federal Office, this is so blatant that it cannot be credible. The perfect fit of testimonies with publicly known events or common perceptions of SOGI minorities is also a reason for decision makers to label a story as ‘fake’ and deny 60 This has been observed in the Canadian context, for instance. See Rousseau and Foxen (n 4) 513. 61 ibid. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 12 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice the claimant’s epistemic agency. Italian decision maker Roberto explained how, at a training session, [w]e projected images of public facts, of things of this type [described in a case study] that happened in Nigeria, to show these things may happen … the classic fake claim is produced like this, put together subsequently. That is, people know that these things happen in their country, they tell you with extreme precision what they read in a newspaper or they heard in their communities, but there is no … ‘And have you had problems with your family? How did you live it?’, ‘Well’, ‘Are you in touch with your father or your mother?’, ‘Yes’, ‘And what do they tell you?’, ‘Nothing’ … Everything is missing. The experience is an individual experi- ence, unique, not repeatable but cannot be devoid of any form of perception. In the UK, there are also concerns that SOGI claims may be ‘fake’ when claimants cor- respond ‘too neatly’ to SOGI stereotypes: I think it is possibly the case that the people who see an advantage in making a claim based on [sexual] orientation will not really understand what [sexual] orientation is about, and will … go in for a stereotypical presentation. Doesn’t mean to say that what could be perceived as stereotypical may not actually be someone’s choice, they may wish to advertise themselves in some way, but that is one type of thing, I think, which would tend to indicate … a claim that didn’t have any sort of substance to it.62 It is a clear illustration of the discursive production of sexual orientation and gender identity that claimants are expected to fit Western stereotypes of what being an ‘out and proud’ LGBTIQ+ person means.63 At the same time, however, they must not fit those stereotypes too neatly or they will be accused of ‘faking’ their stories.64 Claimants from certain countries of origin seem to be regarded with particular scep- ticism by decision makers, who may use their institutional comfort to deny the ‘truth’ of those claimants’ testimonies. For example, Barbara, a lawyer in Germany, asserted that decision makers have basic prejudices against some countries of origin and as- sume that claims from those countries are always fabricated. These countries include Cameroon, Eritrea, Ethiopia, Nigeria, and The Gambia.65 As Daniele, a decision maker in Italy, explained: I believe so, that there is an X number of [fake] claims, more or less significant depending on the country [of origin], because there are countries – and this is known informally – or nationalities in relation to which the simple fact of 62 Adrian, judge, UK. 63 The acronym LGBTIQ+ stands for lesbian, gay, bisexual, trans, intersex, queer, and others. 64 Danisi and others (n 14) ch 7.5. 65 Barbara, lawyer, Germany; Chiara, NGO worker, Italy; Celeste and Susanna, social workers, Italy; Damiano, lawyer, Italy; Diego, Giulia, Giulio, Jonathan, and Riccardo, LGBTIQ+ group volunteers, Italy; Emilia, judge, Germany; Nelo, Italy; Roberto, decision maker, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 13 of 30 presenting this [SOGI] claim could be a disgrace, so it is very difficult for one to do it falsely. That is, it is very difficult for a Malian to present a claim based on sexual orientation falsely, if he is not homosexual. Because in this environment, from a cultural point of view, the origin, etc, it’s really a heavy thing. Instead, there are [countries of] origin for which the problem is minor. For Nigerians, for ex- ample, this type of claim is made with greater ease, even motivating one’s sexual orientation in a somewhat extravagant way … in sum, I must tell you the truth. l D o w n o a d e d f r o m h t t Daniele acknowledged that asylum claimants from some African countries are unlikely to ‘fake’ a SOGI claim, as there is enormous stigma associated with being a member of a SOGI minority, potentially even leading to exclusion from the diaspora community. Yet, the discourse of ‘fake’ claims persists in relation to some countries. Italian decision maker Roberto shared his scepticism about Nigerian claimants claiming to be gay: since I’m here, I have only heard a Turkish national claiming asylum for being transgender … a Somali national for being homosexual, no one from Eritrea. It’s clear that the great weight [in SOGI asylum] of some nationalities [like Nigerians] makes you be more doubtful. Similarly, Filippo, a senior judge in Italy, commented that some colleagues do not wish to listen to asylum claimants because they sell each other ‘absurd stories’, especially when they arrive from particular countries, such as Nigeria. This inclination to sus- pect the ‘fakeness’ of SOGI claims relating to certain countries of origin can worsen when decision makers are mainly, or only, allocated claims from certain geographical areas,66 and has a clear gendered dimension, as illustrated by this example relating to Nigerian women: If you come from Nigeria or come from Benin City, you are 100 per cent a victim of trafficking. So whatever you say about why you ran away, the commission will use the lens of trafficking. And therefore it [the claim] is considered untrue, be- cause you are a victim of trafficking.67 Julian, a SOGI asylum claimant in Germany, also spoke about the bias German de- cision makers frequently show towards female claimants from Uganda: ‘My interviewer was really biased. I entered and he said “Oh, you’re from Uganda, I guess you’re now going to tell me that lesbian story”. Before I could even start’. Such outright denial of claimants’ truthfulness on the basis of their country of origin evidences both testimo- nial and contributory injustice. Epistemic injustice is increased by the fact that, in practice, the discourse of ‘fake’ claims also seems to raise the standard of proof, as decision makers appear to require further evidence to ensure the claimant is not fabricating their story.68 Bilal, a UK Home Office presenting officer, expressed such concern: 66 Rousseau and Foxen (n 4) 517. 67 Celeste, social worker, Italy. 68 Silver, Italy. p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 14 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice some people are exploiting the lack of evidence because you don’t need to produce any, so you can pretend to be, say, a gay man or woman, and be suc- cessful because you don’t need to produce any evidence. So there is, there is an avenue for … you know, because of that hole in the system being exploited. Although asylum claimants (SOGI or otherwise) do need to produce evidence to sup- port their claims,69 the perception that it is easy to succeed in (unsubstantiated) SOGI claims seems to be in the mind of this official. In a more extreme example of testimonial injustice and abuse of institutional com- fort, a judge during a 2018 court observation in Hesse, Germany, asserted at the begin- ning of the hearing that he did not believe the claimant, and intimidated a supporting witness by telling him that he could receive a 12-month prison sentence if he provided false information. For the judge, the claimant’s story was not credible: ‘This story is so deceitful, it’s unbelievable! He has five children and tells me that he is gay all the way! That is unbelievable!’ The assumption that a gay man could not biologically father chil- dren dominated the judge’s thinking, reflecting a stereotypical view that pervaded the appeal hearing with a presumption of ‘fakeness’. The concern that witnesses may contribute to ‘fake’ claims was also highlighted by judges during the fieldwork, rendering witnesses victims of testimonial injustice as well. For example, a judge in the UK stated: One issue we have had is witnesses who’ve given evidence in other cases … this can mean they are active in their own community but can lead to witnesses for hire. We had a situation [a couple of years ago] of claimants from Pakistan and [the] same witnesses came along … Then another issue is social media conver- sations … usually the other person isn’t called as witness, usually they say they don’t know where the person is, but this is evidence that I had a relationship with X. The problem is that falls foul of [the] view that we decide on the basis of oral evidence and if you can’t cross-examine, how much weight can you put on it?70 The emphasis on oral evidence, despite the availability of other (written) evidence, is detrimental to SOGI claimants, as many potential supporting witnesses may not wish to offer oral evidence for fear of ‘coming out’ and being exposed to harm, stigma, or discrimination. It is a form of contributory injustice that becomes even more worrying when the skin colour of witnesses influences judges’ assessments of the genuineness of the claims. As an NGO volunteer in the UK observed: ‘If you take lots of witnesses to court, if they are white and middle class, they are believed’. Conversely, in a case relating to two Pakistani claimants, the judge said that a Pakistani couple were not ‘worth much’ as witnesses.71 69 In the UK, for example, claimants are expected to submit evidence to support a sexual orienta- tion claim, even if just in the form of an oral testimony: UK Home Office, ‘Asylum Interviews’ (Version 7.0, 2019) 31–32. 70 Ernest, judge, UK. 71 Joseph, NGO volunteer, UK. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 15 of 30 Overall, there are clear signs that judges often believe that SOGI claims are fabri- cated, rendering the judges key actors in the epistemic injustice that entraps SOGI asylum claimants: ‘evidence doesn’t seem to persuade some judges at all’.72 Yet, there are also positive examples of judges who refuse to reproduce prejudices against such claim- ants or to contribute to the discourse of ‘fake’ claims. For instance, during an appeal hearing observed in the UK, a judge reassured the appellant that ‘the fact that you’ve had a son doesn’t mean you’re not a lesbian’.73 Silvana, a judge in Italy, suggested that the polemics of ‘fake’ claims are exaggerated and stereotypical, fuelled by the media. As she put it, we should be more concerned about the persecution and discrimination experi- enced by SOGI minorities around the world: It is absolutely normal that you go to a country where homosexuality is not a crime from a country in which it is a crime. Instead, the question that should be asked is how come so many countries still criminalise homosexuality. If there were not so many countries criminalising homosexuality, there would be far fewer requests for protection, I believe. The experiences shared by participants reflect serious degrees of testimonial and con- tributory injustice in the refugee status determination process. However, as the next part of the article shows, decision makers are not the only actors in the asylum system who determine which SOGI claims are seen as ‘true’ and which are seen as ‘fake’. 4. ‘FA K E’ C L A I M S D I S CO U R S E A M O N G ST C I V I L S O C I ET Y A CTO R S Civil society actors – understood here as the range of non-governmental actors active in the field of asylum,74 including NGOs, support groups, and legal representatives, as well as claimants and refugees themselves – also play a role in the power dynamics that shape the discursive construction of what is ‘true’ or ‘fake’ in SOGI asylum claims. While activists ‘contest the sexual and territorial borders’, they also ‘unwillingly con- tribute to their re-inscription’, thus becoming ‘border performers’ and reinforcing State formations.75 McGuirk similarly asserts that NGOs, while ‘ostensibly resisting these constructions, paradoxically create new ones, embedded in wider homonationalist dis- courses that promote a clear victim/savior binary’, mainly owing to the need to attract donations and media attention.76 NGOs working in this field thus dedicate much time and energy to grappling with ‘popular imaginaries’ concerning ‘people pretending to 74 72 Bilal, UK Home Office presenting officer. 73 First-tier Tribunal, London, 2018. Simone Chambers and Jeffrey Kopstein, ‘Civil Society and the State’ in John S Dryzek, Bonnie Honig, and Anne Phillips (eds), The Oxford Handbook of Political Theory (Oxford University Press 2006) 363. Jung (n 54) 333–34. Siobhán McGuirk, ‘Neoliberalism and LGBT Asylum: A Play in Five Acts’ in Siobhán McGuirk and Adrienne Pine (eds), Asylum for Sale: Profit and Protest in the Migration Industry (PM Press 2020) 269. On homonationalism, see Jasbir K Puar, Terrorist Assemblages: Homonationalism in Queer Times (Duke University Press 2007). 76 75 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 16 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice be gay to get asylum’.77 Martorano notes that NGOs in the field of migration face ten- sions between their humanitarian and ethical values, on the one hand, and the bureau- cratic demands of institutions, on the other, eventually replicating the asylum system’s selective policies of assistance for material and moral reasons.78 As this part explores, non-governmental actors often find themselves trapped in the ‘politics of truth’ of the asylum system and are pushed to contribute to the harmful discourse of ‘fake’ claims, even if unwittingly or reluctantly. Some are tolerant of this role; others resist it, refusing to judge someone else’s ‘truth’. Some NGOs and support groups tend to adopt a relatively ‘hands-off ’ approach in relation to determining the veracity of SOGI claims, showing understanding for pos- sible contradictions and changes of narrative: sometimes, even knowing that the story was false, we know of people who have had it [international protection], sorry if … but on the other hand, people about whom we had no doubts and instead have not [been granted international pro- tection] … because they contradicted themselves, because when they arrived in Italy they said something else … because they are stunned by the journey, be- cause they are afraid, they don’t know what to expect, they don’t know that it [sexual orientation and gender identity] is a [ground for asylum request].79 Others are more ‘hands on’, identifying claims they perceive to be ‘fake’ and thus using their relative power to become actors in the discursive production of SOGI and epi- stemic injustice. In line with scholarly work that has identified this phenomenon in the Italian context,80 the fieldwork conducted for the present project found this dynamic operating in support groups: Let’s say that if they come into contact with us, we filter them out first, so we try not to pursue cases in which we don’t believe, but I would say that if I esti- mate the requests for assistance and those we decided to pursue, it’s more or less fifty-fifty.81 Social workers employed in NGO contexts also shared these concerns: I think in relation to The Gambia maybe [we have fake claims]. Because there was an absurd boom in 2014 in requests for reasons of sexual orientation, in the sense … obviously also connected with the question that there is more infor- mation. I believe that many [claimants] before didn’t know that they had this 77 McGuirk (n 76) 271. 78 Noemi Martorano, ‘I Gruppi di Supporto Alle e Ai Richiedenti Asilo LGBTI in Italia: Modelli Organizzativi e Tensioni Associative’ [Support Groups for LGBTI Asylum Claimants in Italy: Organizational Models and Associative Tensions] in Massimo Prearo and Noemi Martorano (eds), Migranti LGBT: Pratiche, Politiche Contesti di Accoglienza (Edizioni ETS 2020) 149–51. 79 Anna, LGBTIQ+ group volunteer, Italy. 80 Martorano (n 78) 149–80. 81 Giulia, LGBTIQ+ group volunteer, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 17 of 30 possibility, but also because they saw that other compatriots have had [inter- national protection].82 Without in any way belittling the essential work carried out by so many NGOs working with migrants and refugees, and while fully understanding NGOs’ need to prioritize limited resources, this approach translates – even if unconsciously – into a methodo- logical homonormativity, in line with a tendency by solidarity movements to construct the ‘ideal subject of solidarity’.83 By doing this, ‘activists respond to and reconstruct the dominant rhetoric, a rhetoric on the basis of which queer and migrant people are ex- cluded and their presence [is] made illegitimate’.84 Even though they may wish to resist the logics of normativity and unleash the power of queer politics, some NGO staff and volunteers – by acting as ‘preliminary judges’ and refusing assistance to those claimants whose testimonies are not believed to be ‘true’85 – mimic the culture of disbelief of decision makers and thus reinforce State-sponsored policies of subjection and assimila- tion.86 In the process, they deprive claimants of their epistemic agency. Amongst these civil society actors are legal practitioners, who play a key role in guiding (or, sometimes, misguiding) claimants through their asylum journey, thereby co-producing the epistemic injustice that entraps them. Legal practitioners are often the first to be wary of ‘standard’ and ‘cyclical’ stories when approached by new clients.87 In Germany, for instance, one lawyer stated that: It’s true that there are … refugees faking [sexual orientation or gender identity]. Probably more women than men, because for men, male homophobia is much bigger, so, I mean, that is certainly a bigger challenge for men … it happened to me that I was sent a woman by the lesbian counselling centre and then she came again a half year later and was pregnant and then told me ‘well, what should I have said, then?’ … That is surely very aggravating. But it happens – I think the figures are not that big.88 Similarly, in Italy, Mara, a lawyer working for an LGBTIQ+ NGO, said that: [W]e do make them follow a process and it is a psychological process, a journey with the mediator, with the operator, we try to make them participate in some ac- tivities that can also be language courses, to try to understand if there is a genuine interest … or whether it is only functional to obtaining the [NGO membership] 82 Susanna, social worker, Italy. 83 Anna Carastathis and Myrto Tsilimpounidi, ‘Methodological Heteronormativity and the 84 “Refugee Crisis”’ (2018) 18 Feminist Media Studies 1120, 1121. Jung (n 54) 315. 85 Martorano (n 78) 168. 86 Jung (n 54) 316–17. On ‘queer politics’ more generally, see eg Michael Warner (ed), Fear of a Queer Planet: Queer Politics and Social Theory (University of Minnesota Press 1993); James Penney, After Queer Theory: The Limits of Sexual Politics (Pluto Press 2015). 87 Bohmer and Shuman (n 5) 614. 88 Janina, lawyer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 18 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice card. … Yes, yes [we do a screening]. Some [claimants] already arrive after the [interview with the] commission, with the rejection and when they have to do the appeal, then we become even more suspicious … It is obvious that we can never be completely sure, but, in short, we try to work on it. In the UK, a volunteer with an LGBTIQ+ support group said that ‘[s]ome solicitors just don’t believe their LGBTQ clients, some feel very uncomfortable around the issue of sexuality as reason for protection’.89 It is unclear whether this was on account of homophobia or for another reason, but such accounts reflect the role legal represen- tatives play in the discursive production of knowledge about asylum claimants’ sexual orientation or gender identity and the epistemic injustice that results. There is a sense that it is possible to ‘know the fake ones from the real ones’,90 des- pite the fact that determining the objective ‘truth’ about someone’s sexual orientation or gender identity is impossible, given the socially and culturally constructed nature of these notions. Both the scholarly literature and asylum policy largely ignore that claim- ants and refugees are themselves key actors in this ‘politics of truth’. As such, they are co-opted by the asylum system to perpetuate the epistemic injustice that underpins the system, and on which the system depends in order to achieve its aims. Some claimants who volunteer with NGOs and support groups are indeed keen on ‘sifting out’ those who do not seem to have ‘genuine’ claims: So when somebody say, is he gay? First of all making intention clear, we send our missionaries on ground, we monitor the person, we know if he’s really a gay. And when we are satisfied … then we give him our membership card.91 they [claimants] are the first ones not to want within the group people who are not really homosexuals, they do not want us to use up our reputation as an asso- ciation for people who are not homosexuals, because they say ‘then, if we help everyone, the commission does not believe us anymore and therefore we cannot help more people’.92 The need to preserve the reputation of NGOs and support groups in order to retain their capacity to support SOGI claimants thus leads to assessments of the genuineness of new claimants, sometimes rendering claimants themselves part of the epistemic in- justice inflicted on one another. An NGO’s reputation cannot be sacrificed by ‘fake’ claims – something observed by Giametta in the French context and Martorano in the Italian context.93 In particular, fellow nationals of potential SOGI claimants function as subjective and powerful ‘filters’, acting as unofficial assessors of the ‘truth’ of their claims: 89 Survey respondent S110. 90 Alain, Italy. 91 Kennedy, Italy. 92 Giulia, LGBTIQ+ group volunteer, Italy. 93 Calogero Giametta, ‘New Asylum Protection Categories and Elusive Filtering Devices: The Case of “Queer Asylum” in France and the UK’ (2020) 46 Journal of Ethnic and Migration Studies 142, 148; Martorano (n 78) 152–53. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 19 of 30 I can’t tell you that we realise it immediately but only after a few questions, also be- cause they [our group members] come from those same countries, etc, when a new one arrives and says ‘I’m from The Gambia, I’m gay’, we have ten from The Gambia who listen to him and, when they tell their story, they are able to contradict him or to notice inconsistencies and then we decide fairly quickly which cases to pursue.94 now we have started to support the new guys with guys of the same nationality who were already with [our group] for many months, so that they are aware of the social and cultural dynamics of the country in question … and that they know how the society of the country in question reacts to homosexuals … who then speaks to us privately and tells us ‘Look, things in Nigeria don’t work that way. Society would never have reacted this way, so he’s lying’.95 ‘Genuine’ SOGI claimants not only take part in this discursive construction of ‘fake’ claims, but also express great frustration about such claimants: I see a lot of straight men come here and say that they’re gay and they’re not gay and they got acceptance. And it kind of makes you feel, you know? Some type of way. Because you’re from Jamaica and you know these men are not gay.96 It is not always clear how some claimants ‘know’ that other claimants are not members of a SOGI minority. Some members of focus groups in Germany felt particularly upset by the injustice of ‘fake’ claims and where this left ‘genuine’ claimants: I had trouble with this Jamaican from the camp, and we know he is not gay be- cause he told us, and even one time he caught an infliction because I was like saying to him ‘You say you’re not gay, then why do you come to Germany to seek refuge as gay? You are just mashing up Germany for people like us who really want to seek refugee status. You’re not gay, so why are you here?’ … Even a guy at our camp is not a gay and he got through. And his friend that is truly gay didn’t get through. He got turned down like us.97 But the straight guys who come here and seek asylum, they just come to make money and they know after two, three years they can go back home because they have saved enough money. And the thing that they don’t understand, they come here and they spoil the opportunity that we as gay people get to come here and seek asylum.98 SOGI asylum claimants thus fear that ‘fake’ claims (or perceptions thereof) will hurt their chances of obtaining a positive decision, especially as decision makers may be- come suspicious about the increasing numbers of SOGI claims. This has also been 94 Giulia, LGBTIQ+ group volunteer, Italy. 95 Nicola, LGBTIQ+ group volunteer, Italy. 96 Angel, Germany. 97 Trudy Ann, Germany. 98 Emroy, Focus Group No 1, Hesse, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 20 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice observed in the context of resettlement work carried out by UNHCR in Turkey, where SOGI claimants become self-appointed screening officers to determine the ‘inauthen- ticity’ and ‘un-deservingness’ of fellow claimants.99 While these concerns are under- standable, it is important to acknowledge that, by adopting such a ‘filtering logic’, civil society actors find themselves unexpectedly co-opted into carrying out ‘perverse prac- tices of policing’, border control, and surveillance,100 simultaneously becoming actors in the epistemic injustice underlying the asylum system. The knowledge contributed and produced by NGOs, legal representatives, asylum claimants, and refugees may play an important and legitimate role in building reliable and up-to-date COI. Yet, such knowledge is not devoid of stereotypes and generaliza- tions, and it can be used to the detriment of SOGI claimants with genuine claims. The irony of supporters and refugees undermining the ‘truth’ of other asylum claim- ants did not escape some of the participants interviewed in this project, whose role in the system can be described as one of ‘counter-conduct’ and resistance against the epi- stemic injustice and dehumanization experienced by SOGI claimants. Seth, an NGO worker in the UK, articulated his frustration at these dynamics in striking terms: ‘“[A]s chief puff I decree that, you know, he is a member of my tribe, so therefore, you know … you know, grant him asylum”. You know, it is ridiculous. … And who am I to sit in judgement’. Responses to potentially ‘fake’ claims in host countries should thus be more sophisticated and socially aware: there is an exaggerated alarmism in relation to this specific subject, because it is true that we know … [of] an increase in the number of [SOGI] claims, which is understandably coherent with an increase in flows and consequently consistent with the greater awareness that now exists, and of the training that once did not exist. … The answers that were given to interpret or manage [an increase in ‘fake’ claims] were not of a social nature. They [the answers] have been from a perspec- tive of demonisation, derision, denunciation, criticism.101 Some NGOs adopt a more constructive approach that avoids the traps of epistemic injustice, for example by offering support to any claimant who requires assistance, even if their claims may seem dubious.102 The use of the limited resources can still be ra- tionalized by imposing some requirements. For example, Joseph, a volunteer with an LGBTIQ+ group in the UK, referred to requiring a minimum period of interaction with the NGO before a supporting statement is produced: ‘[Avoiding “fake” claims] is one of the reasons we said that we wouldn’t, we would not write support letters until people had been coming [to the group] for six months’. Other participants pointed out the risk of generalizing from individual experiences, and the difficulty of ‘faking’ claims: 99 Mert Koçak, ‘Who Is “Queerer” and Deserves Resettlement? Queer Asylum Seekers and Their Deservingness of Refugee Status in Turkey’ (2020) 29 Middle East Critique 29. 100 Giametta (n 93) 147; Martorano (n 78) 172. 101 Vincenzo, LGBTIQ+ group volunteer, Italy. 102 Sara Cesaro, ‘The (Micro-)Politics of Support for LGBT Asylum Seekers in France’ in Richard CM Mole (ed), Queer Migration and Asylum in Europe (University College London Press 2021) 228–29. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 21 of 30 I haven’t met anybody here that I don’t believe is gay. Because I also believe that it is an extreme hurdle for people from this cultural centre to apply for asylum on this [SOGI] basis anyway, if he is not really gay (if his family is here, that’s even out of the question). … Well, I think that’s difficult to do, culturally, since people would have to be good actors.103 I cannot rule that out [‘fake’ claims], but for most people who reveal their sexu- ality or their sexual identity, I think that, … they do that very authentically … there are also very many risks that come with it and therefore it is also a particu- larly vulnerable status that one then has [as a SOGI asylum claimant]. And vol- untarily exposing oneself to that, I do not know, I find that rather unrealistic.104 It was also clear to some participants that attempts to assert the genuineness of SOGI claims replicated the injustice of some asylum authorities’ practices and prerogatives, which NGOs should not emulate: ‘How do I know if the person is really lesbian or gay?’ And that totally upsets me, because I think, when you grow up as a queer or lesbian person and face so many prejudices and somehow so much discrimination … Who would volun- tarily choose this kind of ‘identity’ as an identity? … And I think, these are really rare cases where people would lie about this. … these are mostly people from the [decision-making] institutions that ask such questions and possibly … ‘so, they are not gay, lesbian, trans’ and that … they do not know … do not understand the complexity of living a queer lifestyle. And yes, the stigma associated with it in society, in the family, in the psyche of that person … And whether that is a lie or truth, that’s very … I do not know … absurd.105 While not being able to completely rule out the possibility of claimants fabricating SOGI claims, these participants found it highly unlikely, considering the socio-cultural environment that asylum claimants have to navigate. Ashley, a psychotherapist in the UK, noted that ‘if you haven’t lived with the experience of clandestine sexuality, you won’t be able to fake or feign the language and methods and devices that you use to get through it’. Damiano, a lawyer in Italy, also emphasized how much more difficult life in reception centres could be once it was known that a claimant had a SOGI-based claim. Moreover, it is important to recognize the desperate circumstances that may lead someone not to be entirely honest about their claim, as well as to understand the subjectively, socially, culturally constructed, and fluid nature of sexual orientation and gender identity: But of course, there are cases of people – and one cannot blame them individu- ally – who have experienced or have heard that being gay is a good reason to be recognised [as a refugee] and then try this. It is a way out of the delays in their individual situation.106 103 Thomas, NGO volunteer, Germany. 104 Louis, LGBTIQ+ group volunteer, Germany. 105 Mariya, NGO worker, Germany (emphasis in original). 106 Thomas, NGO volunteer, Germany. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 22 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice for somebody to come repeatedly month in, month out, to a LGBT support group and stuff, then if they are not LGBT, then maybe there is questions at the back of their mind [about their sexual orientation or gender identity] or maybe there is some, you know … and even if they are not [LGBTIQ+], it is ok.107 Above all, some NGO workers make a conscious choice not to assume the role of a decision maker or to follow the way the authorities exercise power: ‘We do not want to play BAMF 2 here’.108 Crucially, and as the scholarship on epistemic injustice highlights, they demonstrate awareness of the fact that there is no verifiable ‘truth’ in respect of a person’s sexual orientation or gender identity: you really can never know that. I’m not [able to], anyway, I could never tell if anyone is gay, lesbian, trans, bi, intersex, such a declaration can only be made about oneself, and even that is flexible, yes … that’s why I always take it as it comes.109 my job is not to make that decision [whether someone is telling the ‘truth’ or not] and I find that if you let your mind go into that, you make that decision about whether or not somebody is telling the truth, I think that makes you a bad lawyer, because who am I to make that decision? … I don’t go there. … That is not my job.110 Although such an approach may impose a higher workload on these NGOs, it seems to be accepted as a way for relevant NGO staff or volunteers to avoid having to make judgements. Given the impossibility of determining what is actually ‘true’, it is impera- tive to identify the key means to address the toxic effects of the ‘politics of truth’ and the vigilante approach that various actors in the asylum system – whether public author- ities or civil society – may have towards SOGI claimants. 5. U S I N G R E F U G E E L AW A N D P O L I C Y TO V I N D I C AT E S O G I R E F U G E E S ’ O W N ‘ T R U T H S ’ The analysis so far has made it clear that: (1) it is not possible to determine the ‘truth’ about someone’s sexual orientation or gender identity, and (2) SOGI claimants see their epistemic agency seriously and continuously damaged by the asylum system (even if they reclaim agency in a variety of other ways).111 While bearing in mind that ‘truth’ is not achievable, we also need to accept that – at least for the foreseeable fu- ture – asylum systems will continue to pursue some sort of objectivity. That being the case, this part attempts to discuss some policy-oriented means to alleviate the epistemic injustice experienced by SOGI asylum claimants. The proposals below fall 107 Seth, NGO worker, UK. 108 Thomas, NGO volunteer, Germany. ‘BAMF’ stands for ‘Bundesamt für Migration und Flüchtlinge’, the German Federal Office for Migration and Refugees. 109 Matthias, social worker, Germany. 110 Deirdre, lawyer, UK. 111 In the words of Bohmer and Shuman, the ‘process deprives the asylum applicants of the right to determine what counts in their own stories’. Bohmer and Shuman (n 5) 624. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 23 of 30 short of a transformative strategy, as suggested by Fraser (see part 2 above), and do not include all the measures that would be necessary at an individual, social, and institu- tional level in order to eliminate epistemic injustice completely in the asylum system. Nevertheless, they offer a pragmatic and realistic approach to mitigating the problems identified above, even within a generally hostile, populistic, and xenophobic political environment.112 The ‘fake’ claims debate should include an honest acknowledgment of the possibility that some claims may not be entirely genuine, but the discussion cannot stop there: I think there is an element of truth [in the ‘fake’ claims debate]. I mean, I think any system in the world, regardless of what, will be abused by some people, for some purposes. I think that is not something we can deny, I don’t think it is so much of a problem necessarily as it is often made out to be. I think there is a lot of fear around that. I also don’t think that the fact that there are some bad apples should prevent genuine cases from receiving the consideration that they actually deserve.113 Many participants acknowledged the desperation felt by asylum claimants to escape persecution and obtain international protection. Desperation can ‘legitimately’ make claimants lose ‘perspective’ and present stories that are not their own in the hope of increasing their chances of being granted international protection (for instance, if they know someone else was successful with that story).114 The lack of available information about SOGI asylum (that sexual orientation or gender identity can be the basis of a claim) when claimants arrive in Europe and/or lodge a claim – combined with claim- ants’ frequent lack of knowledge about the way SOGI minorities are treated in host countries, fear of discrimination by the host community and by their own diaspora, and internalized homo/transphobia – can also understandably lead claimants to em- bellish their fear of persecution.115 Additionally, Chiara, an NGO worker in Italy, made the point that, even if a claimant is not entirely honest in their testimony, this is not necessarily an ‘abuse of the system’, in the sense that the claimant may nevertheless be deserving of international protection. The focus should instead be on those who profit economically from ‘selling stories’ (such as smugglers),116 and from training claimants in how to use those stories: We have also had reports that there are organisations that even train people on how to present themselves as being gay in asylum procedures, because even if the person is not necessarily gay themselves, because it will help your process. And that there are again, apparently, some organisations that charge for such services. … I think that is a more pressing issue. I feel this is maybe a bit contentious, 112 Danisi and others (n 14) ch 4.1. 113 Jules, staff member, ILGA-Europe. 114 Giuseppe, lawyer, Italy; Sofia and Emma, NGO workers, Germany; Terry, member of the European Parliament. 115 Damiano, lawyer, and Valentina, social worker, Italy. 116 Helena, staff member, European Asylum Support Office (EASO), now European Union Agency for Asylum (EUAA). l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 24 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice I feel that anyone who is seeking asylum and who goes through all the effort and hassle and trouble of coming here and seeking asylum whether or not they are gay, whatever their sexual orientation or their gender identity, clearly there was a reason strong enough to motivate them to come, and they should be given a fair chance. So, I don’t necessarily have very strong qualms about people trying to maximise their chances … so long of course that it doesn’t count to a scale that it actually affects those who generally need this particular means. My problem would then become more with those who start to profit off it.117 Media reports from the Netherlands and the USA, for example, affirm concerns that there are people who exploit asylum claimants by selling them stories of successful SOGI claims.118 The focus should thus be on those exploiting SOGI claimants rather than on the risk that some SOGI claimants may not be entirely ‘truthful’. In this context, the ‘filtering’ role played by civil society actors is unwelcome, and these actors’ doubts about whether a claim is genuine are often perceived as judgmental. Consequently, claimants affected by this exercise of power by civil society actors have expressed sad- ness and frustration at being dehumanized and deprived of their ‘truth’ by those from whom they seek support: Because when we come to the groups, we need comfort. We need comfort. We need counselling, we need help. Not to be judged, not to be judged. There is no point why you judge me, when I come to the group, you wait to the Home Office to decide for me, why do you judge me? … You wait for the Home Office and you decide, yes. You don’t need to upset people.119 Some participants also referred to the excessive ‘craving for truth’,120 and, more gener- ally, how this pressure reflects the prejudice and arrogance of civil society actors: I don’t like it either that an association says ‘Ah, but for me he is not gay’. But how can you say that? Again, the LGBTI community also has prejudices … And then, how can you pretend to have the right to judge that a person who comes from a country totally different from yours, does not speak your language, has a totally different mind-set, you say ‘for me he is not gay’. But on what basis do you say that? Even in that case, you have to … put aside certain prejudices that some LGBTI volunteers have, and think that in any case those who have to take a decision are the commission … and that the decision should not be made in the sense that the person must prove irrefutably that they are LGBTI, but that they can offer a story that is more or less coherent.121 117 Jules, staff member, ILGA-Europe. 118 See examples in Bilefsky (n 12); Marion MacGregor, ‘Dutch Government Cracks Down on Ugandan Asylum Seekers after “Fake” LGBT Claims’ (InfoMigrants, 10 November 2020) <https://www.infomigrants.net/en/post/28401/dutch-government-cracks-down-on- ugandan-asylum-seekers-after-fake-lgbt-claims> accessed 12 September 2022. 119 Miria, UK. 120 Giulia, LGBTIQ+ group volunteer, Italy. 121 Cristina, UNHCR official, Italy. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 25 of 30 The ‘truth’ about SOGI asylum claims is unachievable, since both ‘truth’ and ‘fakeness’ about a person’s sexual orientation or gender identity are discursively produced by all actors in the asylum system. Nonetheless, from a pragmatic and policy perspective, it is important to use all the tools available to make the asylum system fairer for SOGI claimants and to enhance its epistemic justice. Five are identified below. First, claimants should be provided with comprehensive information about key aspects of the asylum system when they first lodge their claim, including that sexual orientation and gender identity can be the basis for an asylum claim.122 The fact that this does not happen currently renders it more difficult for States to fulfil their obliga- tion to identify claimants’ special procedural needs.123 Secondly, the right to free legal assistance and representation should be expanded beyond appeal procedures,124 as well as funded more substantially by domestic authorities, to ensure quality representation at all stages of the asylum procedure. This would allow SOGI claimants to lodge better developed initial claims, supported by evidence and informed by sound legal advice, which is not currently the situation in Europe.125 Thirdly, asylum procedures need to be informed by greater respect for claimants’ rights and dignity, as well as a stronger spirit of empathy. This study’s fieldwork showed that this does not happen at present.126 It is essential to ensure that SOGI claimants have enough time to prepare adequately and present their cases effectively. More care needs to be invested in the choice of locations for asylum interviews, training in inter- view techniques, and the quality of interpreting services, as well as ensuring that claim- ants have an opportunity to clarify any apparent contradictions. Overall, a relationship of trust between the claimant and the decision maker needs to be fostered.127 Fourthly, should a decision maker retain doubts after the interview, it is important to apply the principle of the benefit of the doubt whenever possible. It is clear that this principle is not currently applied with the consistency and scope it warrants, either at a domestic or an international level.128 This is compounded by the fact that the claimant’s self-identification in terms of sexual orientation or gender identity is not afforded suf- ficient value: it may not be the end of the matter,129 but it is undoubtedly the starting 122 Nuno Ferreira, ‘Reforming the Common European Asylum System: Enough Rainbow for Queer Asylum Seekers?’ [2018] Rivista di Studi Giuridici sull’Orientamento Sessuale e l’Identità di Genere 25, 33. 123 European Council on Refugees and Exiles, ‘The Concept of Vulnerability in European Asylum Procedures’ (2017) 21. 124 Currently, and within the Common European Asylum System, this right refers only to appeal procedures: see Directive 2013/32/EU of 26 June 2013 on common procedures for granting and withdrawing international protection (recast) [2013] OJ L180/60, art 20. 125 Danisi and others (n 14) ch 6.2.2. 126 127 128 129 ibid ch 6. ibid ch 11.3.2. ibid ch 7.4.1; Nuno Ferreira, ‘An Exercise in Detachment: The Strasbourg Court and Sexual Minority Refugees’ in Mole (ed), Queer Migration and Asylum in Europe (n 102). Joined Cases C–148/13, C–149/13, and C–150/13, A, B and C v Staatssecretaris van Veiligheid en Justitie [2014] ECLI:EU:C:2014:2406, para 49. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 26 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice point, and decision makers need to take that seriously.130 Rather, some decision makers go as far as reversing the presumption of truth reflected in the principle of the benefit of the doubt and believing that, in case of doubt, a story should be assumed to be false.131 Yet, not only is respect for the principle of the benefit of the doubt a legal requirement, according to UNHCR,132 it is also advisable from a policy perspective: Because, you can even make an argument, I think, that if somebody is so des- perate to stay, that they are actually willing to lie about their sexuality and tell you that they’re, they are gay or whatever, you know, where they know that within their own society this is something which is not seen as acceptable, which does put them at risk … You have got to be pretty desperate to lie about it, so you know … I belong to the group that tends to do benefit of the doubt.133 Some decision makers do seem to be conscious of the need to adopt a lower standard of proof and apply the benefit of the doubt whenever possible: it was bollocks [a ‘fake’ claim], really. And you do get cases like that, yes, of course, you do, yes, and it makes judges battle weary and cynical, of course. And you have got to put that on one side all the time. … But, you know, you remind yourself all the time, it is a lower standard, lower standard [of proof]. It is not a balance of probabilities, it is the lower standard, and if in doubt you must give, you must give the benefit of the doubt.134 In conjunction with a lower standard of proof and the benefit of the doubt, emphasis should shift from the claimant’s ‘true’ sexual orientation or gender identity to the risk of persecution, conditions in the country of origin, and the quality of COI.135 This would better balance decision makers’ determination of the ‘truth’ of SOGI claimants’ mem- bership of a particular social group with an analysis of the risks facing claimants if they are returned to their countries of origin. Fifthly, decision makers would benefit from better training and working conditions, to avoid lack of preparation, burnout, and desensitization. This fatigue and loss of em- pathy over the years have been documented, for example, in Canada.136 Mandatory and regular training on general SOGI matters and SOGI-related COI – including the so- cial and cultural nature and variations of SOGI – would equip decision makers with more appropriate knowledge and skills to deal with SOGI claims in a non-stereotypical 130 Moira Dustin and Nuno Ferreira, ‘Improving SOGI Asylum Adjudication: Putting Persecution Ahead of Identity’ (2021) 40(3) Refugee Survey Quarterly 315. Jubany (n 6) 87. 131 132 UNHCR, Handbook on Procedures and Criteria for Determining Refugee Status and Guidelines on International Protection under the 1951 Convention and the 1967 Protocol relating to the Status of Refugees, HCR/1P/ENG/REV.4 (1979, reissued 2019) paras 203–04. Jean, member of the European Parliament. 133 134 Harry, senior judge, UK. 135 Dustin and Ferreira (n 130). 136 Rousseau and Foxen (n 4) 517. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 27 of 30 or uncynical manner. Moreover, as Helena, a staff member in the European Asylum Support Office (EASO, now European Union Agency for Asylum, EUAA), argued, de- cision makers are invariably affected by the stories of war, rape, and torture to which they listen on a daily basis. According to Jubany, the fact that (in Spain and the UK) there are fewer female than male decision makers also means that the female decision makers frequently listen to stories of rape and sexual violence, thus contributing to greater scepticism and desensitization.137 Finding that these stories must be to some extent ‘fake’ becomes a natural protective mechanism.138 States thus need to improve the training and working conditions of decision makers by providing mandatory and regular training, flexible working conditions, career breaks, and appropriate forms of staff support, including counselling and training in vicarious trauma and self-care,139 as well as abstaining from putting decision makers under any form of pressure to reject asylum claims. None of these suggestions will help determine the ‘truth’ in SOGI claims; such an endeavour is doomed to fail. Nevertheless, the five broad recommendations delineated here can assist in increasing the epistemic justice of the asylum system for SOGI claim- ants – as well as potentially for all asylum claimants – as they have the potential to help claimants have a greater say (both quantitatively and qualitatively) in the discursive construction of the ‘truth’ of their claims. By pursuing greater respect for the right to information, investing in legal aid, improving asylum procedures, applying the prin- ciple of the benefit of the doubt, and improving the training and working conditions of decision makers, we could further reduce the already negligible risk of ‘fake’ SOGI claims. By setting the example and operating an asylum adjudication system that re- spects claimants’ ‘truths’ and does not indiscriminately label their stories as ‘fake’, civil society actors would, in turn, gradually discard their roles as ‘filters’ of ‘fakeness’. NGOs’ institutional reputations would not be damaged if they occasionally offered support to a claimant not being, or not having undergone, exactly what their testimony states, since what matters is to respect claimants’ rights and to treat them with impartiality and humanity. The principle of the benefit of the doubt, in particular, would support all actors in the asylum system in a journey towards greater empathy, belief, and re- spect, better fulfilling the aims of the international protection system. Crucially, this would support refugees’ struggles for epistemic recognition and, at the same time, give them more power to define their own identities and prevent asylum authorities from dictating the terms. 6. CO N C LU S I O N SOGI asylum claimants face the impossible task of proving they are queer enough but not too queer, proving they come from a country where SOGI minorities face enough risk of persecution but where there is not a generalized risk of violence, and, above all, proving the ‘truth’ of their claim where decision makers commonly have a mindset im- bued with scepticism, cynicism, and prejudice. It is all too easy to consider a claim to 137 Jubany (n 6) 84. 138 Deirdre, lawyer, UK. 139 Danisi and others (n 14) ch 11.3.1. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 28 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice be ‘fake’, which renders the asylum system deeply unjust from an epistemic perspective. By adopting the Foucauldian-inspired body of literature on epistemic injustice as a the- oretical framework, this article has identified the crucial ways in which SOGI claimants are deprived of epistemic agency, not only by asylum authorities, but also by NGOs, support groups, legal representatives, and other SOGI claimants and refugees. While relying mostly on empirical data collected in Germany, Italy, and the UK, and notwith- standing any country-level disparities, both the study participants and the documen- tary sources confirmed that all actors in the asylum system to some extent contribute to discourses on ‘fake’ claims. This justifies the concern expressed in this article that asylum systems across Europe and elsewhere are designed in a way that seeks to estab- lish a ‘truth’ that cannot be established, and to deny SOGI claimants their ‘truth’. The topic of ‘fake’ claims is most often used by anti-migration and anti-refugee politicians as part of a xenophobic and racist rhetoric. This applies to asylum claims in general, and SOGI ones in particular, thus often also reflecting homophobia and transphobia. That may explain why discussing ‘fake’ claims seems taboo in academic circles and grey literature. Instead, this article has faced this issue without subterfuge: there may be SOGI claims that lack complete veracity, but then again, ‘truth’ in relation to a person’s sexual orientation or gender identity is illusory, since it is largely subject- ively, socially, and culturally constructed. The theoretically informed and empirically grounded approach employed here may usefully be replicated in relation to other categories of asylum claims, such as those based on religious grounds or gender-based violence, which are also severely affected by discourses of ‘fakeness’ and difficulties with standards and burdens of proof.140 If ‘fake’ claims exist, they are undoubtedly few – ‘exceedingly rare’, in the words of Neilson and Adams.141 More importantly, nobody can claim the role of final arbiter of the ‘truth’, as any system of production of ‘knowledge’ and ‘truth’ is discursively con- structed, shaped by power relations, and characterized by epistemic injustice. The ‘fake’ claimant – especially if thought of as a pervasive and dangerous phenomenon – is thus a myth: a convenient myth to help society make sense of a challenging situation, and design a solution for it.142 As Jean, a member of the European Parliament, said: I think it [fake claims] is another part of the mythology. I would be very inter- ested to see what the figures are on that, because I am willing to bet that most Member States don’t have them. … [I]t is one of those claims that … I think is invented for a purpose. … lot of countries work with the culture of disbelief, the idea that somehow, you know, this [sexual orientation or gender identity] almost 140 See eg Uwe Berlit, Harald Doerig, and Hugo Storey, ‘Credibility Assessment in Claims based on Persecution for Reasons of Religious Conversion and Homosexuality: A Practitioners Approach’ (2015) 27 International Journal of Refugee Law 649; Isabella Mighetto, ‘The Contingency of Credibility: Gender-Related Persecution, Traumatic Memory and Home Office Interviews’ (2016) 3 SOAS Law Journal 1. 141 Victoria Neilson and Lori Adams, ‘Gay Asylum Seekers’ The New York Times (7 February accessed <https://www.nytimes.com/2011/02/07/opinion/lweb07gay.html> 2011) 12 September 2022. 142 Rousseau and Foxen (n 4) 507. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice • Page 29 of 30 is a sort of privileged grounds for claim. … I cannot think of anything that I have seen in terms of evidence, that would back that statement, at all. Another member of the European Parliament, Terry, had a similar view: the step to say ‘I am a gay man’, or the step to say ‘I am a trans woman’, without being it, just, you know, to get asylum, and to have it easier … is so high [large] that the number of people who would actually do that and then can tell a cred- ible story about how they were suffering from this, and how it made their life different, very difficult … that the attention that is given to this in the media is completely over the top. In other words, if there is an ‘abuse’, it is an ‘abuse’ committed by States that construct ‘bogus asylum seekers’ and ‘irregular migrants’.143 Our response should thus be at a policy and social level, to facilitate legal and documented migration paths. This would help prevent people providing embellished accounts instead of their own stories be- cause they are desperate. There may only be discursively constructed ‘truth’ and ‘fakeness’ rather than ob- jective ones. But to the extent that one is obliged to try to ‘prove’ something – as asylum claimants are – then systems and processes should facilitate epistemic justice as much as possible. Telling one’s story – even when including experiences of violence – can be empowering,144 but that is frustrated if the listener denies the experiences being recounted and thus dehumanizes the speaker. In fact, denying the claimant’s testi- mony can be even more traumatizing and distressing for the claimant than the original trauma.145 Yet, the need to safeguard the ‘integrity of the system’ is used as an excuse to search for models of decision making that can expunge ‘false’ SOGI claims.146 SOGI claims are thus a powerful example of the disturbing epistemic injustice that asylum systems produce. Decision makers involved with SOGI claims enjoy a clear ‘institutional comfort’ that is used to facilitate testimonial and contributory injustice.147 This not only results in excessive and inappropriate use of discretion by decision makers,148 but also feeds into a toxic discourse of ‘fakeness’. While it may not be possible to completely domesticate such discretion and eradicate the discourse of ‘fake’ claims, it is realistic to combat and reduce the current testimonial and contributory injustices in SOGI claims. As explored above, the focus should be on ensuring respect for the right to information, investing 143 Valentina, social worker, Italy. 144 Amanda Burgess-Proctor, ‘Methodological and Ethical Issues in Feminist Research with Abused Women: Reflections on Participants’ Vulnerability and Empowerment’ (2015) 48 Women’s Studies International Forum 124. 145 Rousseau and Foxen (n 4) 519. 146 M Yanick Saila-Ngita, ‘Sex, Lies, and Videotape: Considering the ABC Case and Adopting the DSSH Method for the Protection of the Rights of LGBTI Asylum Seekers’ (2018) 24 Southwestern Journal of International Law 275, 298. 147 Sertler (n 39). 148 Danisi and others (n 14) ch 7. l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 Page 30 of 30 • The Discourse of ‘Fake’ SOGI Asylum Claims as a Form of Epistemic Injustice in legal aid, improving asylum procedures, applying the principle of the benefit of the doubt, and improving decision makers’ training and working conditions. A more trans- formative strategy – one that completely eliminates epistemic injustice in asylum sys- tems – should be the long-term aim. Indeed, it is a moral obligation, and ‘to be human is to be moral, and you cannot have a day off when it suits you’.149 l D o w n o a d e d f r o m h t t p s : / / i a c a d e m c . o u p . c o m / i j r l / a d v a n c e - a r t i c e d o / l i / . 1 0 1 0 9 3 / i j r l / i / e e a c 0 4 1 7 0 1 9 6 2 3 b y S u s s e x U n v e r s i t y u s e r o n 0 8 F e b r u a r y 2 0 2 3 149 Lloyd Jones, Mister Pip ( John Murray 2008) 180.
10.1126_science.adg7883	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 September 13. Published in final edited form as: Science. 2023 April 21; 380(6642): 301–308. doi:10.1126/science.adg7883. Structure of the R2 non-LTR retrotransposon initiating target- primed reverse transcription Max E. Wilkinson1,2,3,4,5, Chris J. Frangieh1,2,3,4,5,6, Rhiannon K. Macrae1,2,3,4,5, Feng Zhang1,2,3,4,5,* 1Howard Hughes Medical Institute; Cambridge, MA 02139, USA. 2Broad Institute of MIT and Harvard; Cambridge, MA 02142, USA. 3McGovern Institute for Brain Research, Massachusetts Institute of Technology; Cambridge, MA 02139, USA. 4Department of Brain and Cognitive Science, Massachusetts Institute of Technology; Cambridge, MA 02139, USA. 5Department of Biological Engineering, Massachusetts Institute of Technology; Cambridge, MA 02139, USA. 6Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology; Cambridge, MA 02139, USA. Abstract Non-LTR retrotransposons, or Long Interspersed Nuclear Elements (LINEs), are an abundant class of eukaryotic transposons that insert into genomes by target-primed reverse transcription (TPRT). During TPRT, a target DNA sequence is nicked and primes reverse transcription of the retrotransposon RNA. Here, we report the cryo-electron microscopy structure of the Bombyx mori R2 non-LTR retrotransposon initiating TPRT at its ribosomal DNA target. The target DNA sequence is unwound at the insertion site and recognized by an upstream motif. An extension of the reverse transcriptase (RT) domain recognizes the retrotransposon RNA and guides the 3′ end into the RT active site to template reverse transcription. We used Cas9 to retarget R2 in vitro to non-native sequences, suggesting future use as a reprogrammable RNA-based gene-insertion tool. One-Sentence Summary: A retrotransposon structure shows the principles of target DNA selection and self RNA recognition. This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. *Corresponding author. [email protected]. Author contributions: M.E.W. and F.Z. conceived the project. M.E.W. designed and performed experiments and solved the cryo-EM structure. C.J.F. generated and analyzed sequencing data. F.Z. supervised the research and experimental design with support from R.K.M. M.E.W. wrote the manuscript with input from all authors. Competing interests: F.Z. is a scientific advisor and cofounder of Editas Medicine, Beam Therapeutics, Pairwise Plants, Arbor Biotechnologies, and Aera Therapeutics. F.Z. is a scientific advisor for Octant. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 2 Non-long terminal repeat (non-LTR) retrotransposons are the most abundant class of mobile genetic element (MGE) in the human genome, mostly represented by the LINE-1 and SINE (or Alu) elements (1). Despite their prevalence and contribution to genetic diversity and dysregulation through mutagenicity and recombination (1–3) and their prospective use as gene insertion tools, there is much left to understand about their mobility mechanisms (4). Pioneering research on the Bombyx mori (silk moth) R2 element (R2Bm), which selectively inserts into the 28S rRNA gene, has contributed significantly to our understanding of this type of MGE (5). R2, like all non-LTR retrotransposons, encodes an open reading frame (ORF) with DNA-binding, endonuclease, and reverse transcriptase activities (Fig. 1A). The endonuclease domain (restriction-like endonuclease, RLE) nicks the target DNA, and the reverse transcriptase domain uses the exposed 3’ end from the nick to prime reverse transcription of the R2 RNA, resulting in a new genomic copy of the R2 element (Fig. 1B) (6, 7). This process is called target-primed reverse transcription (TPRT), and is characteristic of non-LTR retrotransposons and their group II intron ancestors (8, 9). The nicked strand that primes reverse transcription is referred to as the bottom strand. Complementarity between the bottom strand and the 3′ end of the R2 RNA (3′ homology) is not required to initiate reverse transcription (10) Non-LTR retrotransposons are specific for reverse transcribing their own RNA; for R2, this specificity requires an element in the 3′UTR but the precise motif has not been located (11). It is also unclear how R2 specifically recognizes the 28S rRNA target gene, or how DNA nicking is coupled to reverse transcription within the same protein. To address these questions, we solved a cryo-EM structure of the Bombyx mori R2 protein (R2Bm) initiating TPRT at the 28S rRNA gene using its own 3′UTR. The structure reveals an extensive interface with the target DNA, a small core region of the 3′UTR required for TPRT, and shows that R2Bm can be engineered to reprogram its insertion site. Reconstitution and cryo-EM structure of an R2 TPRT complex We overexpressed R2Bm in Escherichia coli and purified it to apparent homogeneity (fig. S1). The purified protein was active in vitro, reproducing previously found biochemical activities, including RNA-stimulated nicking of the target DNA bottom strand, site-specific TPRT when supplied with in vitro transcribed 3′UTR RNA, and low levels of template jumping (Fig. 1C) (6, 12). It is unclear if 3′ homology is required for TPRT in vivo; however, consistent with previous findings, we found that downstream sequences up to 10 nt do not inhibit activity in vitro (Fig. 1C) (10). Sequencing of TPRT junctions confirmed that homology-mediated TPRT is more likely to initiate reverse transcription at the 3′ end of the 3′ UTR rather than skipping bases or inserting untemplated nucleotides (fig. S2). (10). To assemble a complex stalled during initiation of TPRT, we incubated R2Bm with target DNA, 3′UTR RNA, and the chain-terminator nucleotide 2′,3′-dideoxythymidine (ddT), which mimics the first nucleotide incorporated in the TPRT reaction (dT) but does not allow further elongation. Purified TPRT complexes contained stoichiometric amounts of R2Bm, 3′UTR RNA, and target DNA with > 99% of the bottom strand nicked (fig. S1). Initial attempts at cryo-EM imaging failed due to the preferred orientation and flexibility of the complex. To overcome these issues, we used a carbon support on the cryo-EM grid and added 5 nt of downstream 28S RNA sequence to the 3′ end of the 3′UTR RNA to stabilize the complex by forming a primer-template duplex with the target DNA bottom strand. With Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 3 these modifications, we obtained a cryo-EM reconstruction of the R2 TPRT complex at 3.1 Å resolution (Fig. 1D, fig. S3–4, table S1). The core of the R2Bm protein is a reverse-transcriptase domain (RT) similar to group II intron RTs (13), followed by a C-terminal ɑ-helical thumb domain and preceded by a characteristic N-terminal extension domain (NTE0) implicated in template switching (14), but the R2Bm RT includes a further N-terminal extension (NTE-1) that binds the 3′UTR RNA (Fig. 1E, F) (15). Preceding the NTE-1 element are two DNA binding domains: the N-terminal C2H2 zinc finger domain (N-ZnF) and a Myb domain. C-terminal to the thumb domain lies an ɑ-helical linker domain that packs against the thumb, followed by a CCHC zinc-finger domain (ZnF) conserved in many LINE ORFs (4). The ZnF then links to the C-terminal RLE domain, which cleaves the target DNA. This domain arrangement closely resembles Prp8 (13, 16, 17), the core protein of the spliceosome, underscoring the close relationship between pre-mRNA splicing and retrotransposons. There are several key interactions between the R2Bm protein, 3′UTR RNA, and target DNA (Fig. 1E, F). The two strands of the target DNA separate around the ZnF domain, with the bottom strand feeding into the RLE active site where the scissile phosphate remains bound, while the top strand snakes along the opposing surface of the RLE. The RT active site contains a heteroduplex formed by the nicked bottom strand of the target DNA (5′ to the cleavage site) and the 5 nt of 28S RNA homology extension beyond the 3′UTR RNA (Fig. 1G). This target heteroduplex is surrounded by residues important for RT activity (18), and the cryo-EM density shows incorporation of the ddT chain terminator nucleotide into the bottom strand (Fig. 1H). The 5′ end of the bottom strand remains base-paired to the top strand as it leaves the RLE, and this downstream DNA region has weak cryo-EM density, suggesting it is not tightly bound by R2Bm. The 248-nt 3′UTR RNA is mostly not resolved in the cryo-EM density except for a core 40-nt region, which wraps around the NTE-1 ɑ helix of R2Bm and the 3′ end of which is guided into the RT active site via the NTE0 domain. R2Bm recognizes a sequence motif upstream of the cleavage site The target 28S DNA sequence has extensive interactions with R2Bm (summarized in Fig. 2A). Upstream bases from –38 to –7 and downstream bases from +6 to +21 are respectively paired, whereas the 11 base pairs from –6 to +5 are melted around the RLE domain (bases are numbered relative to the bottom strand cleavage site). The upstream DNA has a 40° bend and binds along the surface of the RT, linker, and thumb domains in a manner similar to the DNA in a recent group IIC intron maturase structure (Fig. 2B, fig. S5, S6) (19). Many of the contacts between R2Bm and the DNA are via the phosphate backbone, suggesting that they are not sequence-specific. Based on the structure, however, we predicted that two regions are key for sequence-specific DNA recognition by R2Bm: a 13-bp upstream motif from –34 to –22, which is bound by the N-terminal N-ZnF and Myb domains, and the 7 bp from –6 to +1, which are bound by the RLE (Fig. 2A). We term these regions the Retrotransposon Upstream Motif (RUM) and Retrotransposon-Associated INsertion site (RASIN), respectively. Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 4 Consistent with the importance of the RUM region for R2 activity, mutating the entire upstream sequence between –38 to –7 eliminated bottom strand cleavage, whereas mutating the downstream sequences between +6 and + 37 preserved wild-type levels of bottom strand cleavage and TPRT (Fig. 2C) (20). Adding just the 13-bp RUM region to the upstream mutant at positions –34 to –22 restored near-wild-type activity, whereas a point mutant RUM (G–27 to C) did not rescue activity (Fig. 2C). This region of the target was strongly protected in a previous DNase footprinting assay (21). To systematically determine the importance of each base within the RUM, we performed an R2 cleavage assay on a DNA target with the upstream region (–38 to –7) mutated and the RUM (–34 to –22) replaced with a 13N library (Fig. 2D). Sequencing of cleaved targets revealed a consensus RUM sequence A–31WWWGCNNNA–22, where W is A/T and N is any nucleotide, with minor preferences in other positions (Fig. 2E). This consensus is a close match to the wild-type 28S sequence A–31ACGGCGGGA–22 , with the differences underlined. The RUM is recognized by three domains: N-ZnF, Myb, and an R2-specific insertion ‘6a’ in the RT domain between motifs 6 and 7 (Fig. 2B, fig. S7). The N-ZnF has the classical C2H2 fold with a zinc ion coordinated between an ɑ-helix and a β-hairpin, but unusually the ɑ-helix binds in the widened minor groove of the DNA from bases –18 to –23 instead of the typical major groove (Fig. 2F, fig. S6) (22). The preference for A at base –22 in the RUM is likely due to N-ZnF Arg125, which hydrogen bonds with the minor-groove–facing side of the A–T base pair (Fig. 2F). The Myb domain forms a typical three-helix bundle, with the third helix bound in the major groove from bases –31 to –34 (22) while its linker to N-ZnF engages with base –30 (Fig. 2G). This is reminiscent of other Myb–DNA structures, including telomere-interacting protein Rap1 (23). The Myb domain recognizes the A at base –31 via hydrogen bonds with Lys149 (Fig. 2G). Although Arg198 contacts bases at positions –33 and –34, these contacts appear not to be sequence specific, as the RUM screen showed only weak sequence preferences in this region (Fig. 2E, G). Deletion of the N-ZnF and Myb domains together (ΔN mutant) completely inhibits target DNA nicking and subsequent TPRT (Fig. 2C) (20). The central GC of the RUM is recognized by His673 and Lys675 of the loop 6a of the RT domain (Fig. 2H). Structural predictions suggest that this loop is unique among non-LTR RT domains to R2 proteins (fig. S7). We found that deletion of the 6a loop inhibits target DNA nicking (Fig. 2C). Finally, we found that the distance between the RUM and the bottom strand cleavage site (the RASIN) is important: increasing the distance by one base was tolerated, but further increase or any decrease to the distance inhibited target cleavage (Fig. 2I). Target DNA interactions at the cleavage and integration site The second key region for DNA target recognition by R2Bm is the target site for nicking by the RLE domain and R2 insertion, which we term the RASIN. In our structure, the 11 base pairs of the RASIN from –6 to +5 are melted around the RLE domain. The ZnF appears to act as the “zip,” stacking on the last upstream pair C–G(–7) with Arg922 and Arg924 and holding unzipped strands apart (Fig. 3A). Strand melting may be enhanced by the 40° bend in target DNA around the RUM (Fig. 1F). Bases –6 to –1 on the bottom strand then follow a cleft between the ZnF and the RLE, which adopts a canonical PD-(D/E)xK-family nuclease fold, but with the characteristic Lys1026 on an ɑ helix instead of the usual β strand Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 5 (Fig. 3B) (24). This lysine, along with catalytic residues Asp996 and Asp1009, are 4 – 6 Å from the scissile phosphate of C(–1), suggesting C(–1) may be close to its position during catalysis of bottom strand cleavage. On the top strand, bases –6 to +2 all make extensive contacts along a cleft between the RLE and linker domains, except for A(–4), which flips out and contacts C126 of the 3′UTR (Fig. 3C). To determine the relative importance of the bases in the RASIN, we mutated each of the 11 bp individually and tested the effect on bottom strand cleavage. Mutating T(+1) to A abolished cleavage entirely, and mutating T(–6), T(–5), and A(–3) severely decreased activity, whereas other changes were tolerated (Fig. 3D). This suggests the following RASIN motif for cleavage, given in top strand sense: T–6TNANNT+1. Because only the bottom strand of the RASIN enters the RLE active site, we tested the activity of R2Bm on a single-stranded DNA with the bottom strand sequence and found that it was cut, albeit weakly (Fig. 3E). Endonuclease activity was strongly stimulated by providing a 60-nt top strand spanning the RASIN and upstream and downstream sequences, but was similarly stimulated by a 32-nt top strand complementary only to the upstream region containing the RUM. A 17-nt top strand complementary to the downstream sequence did not stimulate activity (Fig. 3E). This suggests that the RUM in a double-stranded state is important for recruiting the R2Bm RLE to the RASIN bottom strand, and that the top strand of the RASIN, despite its extensive interaction with R2Bm, is dispensable for specific bottom strand cleavage. However, when we added deoxynucleotides to these reactions, TPRT activity was eliminated in the absence of the top strand from the RASIN downstream but was partially rescued if the 3′UTR RNA contained 3′ homology to the target site (Fig. 3E). The top strand RASIN bases A(–4), A(–3), and G(–2) are grasped by Arg901 and Asp902 of the R2Bm linker (Fig. 3C). We mutated these two residues to alanine and tested TPRT activity on a fully double-stranded substrate, and found that TPRT activity was reduced and partially rescued by 3′ homology (Fig. 3E). These results suggest two important factors for initiating TPRT when the 3′UTR RNA lacks 3′ homology. One: presence of a top strand downstream of the RASIN, which may help retain the nicked bottom strand, and two: contacts between R2Bm and the top strand RASIN, which help the nicked bottom strand “pivot” into the RT active site. R2Bm binds a small core region of the 3′UTR R2Bm can only initiate TPRT on RNAs containing the R2 3′UTR (self-specificity), but the molecular basis for this is not known (25). Multiple models have been proposed for the secondary structure of the R2 3′UTR, and the divergent sequences of R2 RNAs have hindered identification of key bases (26, 27). A model for the R2 3′UTR secondary structure based on chemical probing is shown in Fig. 4A and has at least 11 stems (26). In our cryo-EM map, we resolved density for two stems and their flanking single-stranded regions (Fig. 4B). Based on nomenclature commonly used for structured RNAs, we name these stems P1 (nucleotides 33 – 38 and 120 – 135) and P2 (nucleotides 131 – 137 and 236 – 242), and term the single-stranded junction between P1 and P2 as J1/2 and the single-stranded region preceding P1 as J0/1. The rest of the 3′UTR may occupy a diffuse cloud of cryo-EM density next to these core regions (Fig. 4C). Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 6 P1 and J1/2 are mainly recognized by an ɑ helix from the R2Bm NTE-1 domain, which packs into the major groove of P1 and is wrapped by J1/2 (Fig. 4B). Arg307 recognizes the Hoogsteen edge of P1 G33, and the interaction is secured by Arg310 and Arg311. Consistently, these residues were previously shown to be essential for RNA binding (15), and the first 45 bases of the 3′UTR are essential for TPRT activity (11). J1/2 makes numerous sequence-specific contacts (Fig. 4D): A127 forms a sugar-edge pair with the Watson-Crick face of J0/1 A32 , A128 hydrogen bonds to Leu732 and Lys733 of the R2Bm thumb domain and stacks on NTE-1 Tyr314, U129 hydrogen bonds to Glu319 and Lys322 of NTE-1, and C126 stacks on and hydrogen bonds with A(−4) from the top strand of the DNA target (Fig. 4B, D). To test if regions of the R2 3′UTR not clearly visible in the cryo-EM density are required for TPRT activity, we designed a 43-nt minimal 3′UTR – “R2 tag” – that contains only the sequences visible in the cryo-EM density, linked by tetraloops (Fig. 4E). The R2 tag was reverse transcribed as efficiently as the full 248-nt 3′UTR in a TPRT reaction. We tested the importance of the J1/2 linker by making single base transversions and found that A127U reduced activity and A128U almost completely abolished TPRT activity (Fig. 4F). Mutating G33 to C to disrupt base pairing at the bottom of stem P1 also reduced activity but could be rescued by the compensatory C125G mutation (Fig. 4F). Mutation of J0/1 A32 to G reduced activity, but mutations to C or U were tolerated. Equivalents to P1, P2, J0/1, and J1/2 can be identified in the secondary structures of diverse R2 elements (26) (fig. S7). The P1 and P2 stems have different sizes and base compositions, but positions 2 and 3 of J1/2, corresponding to A127 and A128, are conserved as adenosines, consistent with their importance for TPRT. Because the R2 tag alone is efficiently integrated in a TPRT reaction, we tested if adding the R2 tag to the 3′ end of a “cargo” RNA would allow its integration at the 28S target site. We added the R2 tag to the 3′ end of a 239-nt CMV promoter RNA. This tagged RNA was used as efficiently as wild-type R2 3′UTR in a TPRT reaction, whereas an untagged RNA was not used, nor was an RNA tagged with R2-tag A128U mutant (Fig. 4G). A larger RNA containing the 720-nt coding sequence for GFP and a 3′ R2 tag was also reverse transcribed in a TPRT reaction (Fig. 4G). R2Bm can be retargeted with CRISPR-Cas9 Our structural and biochemical observations suggest a multi-step model for initiation of TPRT: the R2Bm N-terminal domains first detect a RUM sequence, followed by cleavage of the bottom strand at the RASIN site, possible pivoting of the nick around the top strand into the RT active site, annealing of any 3′ homology to the nicked bottom strand, and finally initiation of reverse transcription (Fig. 5A). This model implies that R2Bm could prime reverse transcription off an exogenously nicked bottom strand close to the R2Bm binding site (Fig. 5B). To test this, we replaced the RASIN and downstream sequences of the 28S DNA target with an unrelated sequence containing an efficient SpCas9 target sequence, but kept the RUM sequence to anchor R2Bm (Fig. 5B). This substrate could not be cleaved by R2Bm, but was nicked efficiently by a SpCas9 H840A nickase mutant (Fig. 5C). When SpCas9 and R2Bm were added together with a single-guide RNA (sgRNA) and an R2 Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 7 3′UTR RNA with 5 nt of 3′ homology to the sgRNA nick site, we detected low amounts of TPRT activity. This activity was enhanced when the R2Bm and SpCas9 proteins were fused with a 33XTEN flexible linker (Fig. 5C). The RUM was not required for Cas9-directed TPRT, as mutating the RUM did not reduce activity (Fig. 5C). This suggests Cas9 might be able to direct R2Bm to perform TPRT at loci other than the 28S target. We mixed the R2Bm-Cas9(H840A) fusion protein with a 192-bp target sequence from Drosophila virilis, various sgRNAs, and R2 3′UTRs with 10 nt of 3′ homology to the nick site dictated by the sgRNA (Fig. 5D). We found TPRT activity at all Cas9 nick sites, with one sgRNA (guide 2) giving efficient activity (Fig. 5E). Adding R2Bm and SpCas9(H840A) as separate polypeptides also yielded efficient TPRT with guide 2, but was less robust with other guides (fig. S9). The 239-nt CMV promoter RNA with a 3′ R2 tag and 10 nt of homology to the guide 2 nick site was also reverse transcribed efficiently; this activity required the R2 tag and was reduced in the absence of 3′ homology or with the R2 tag A128U mutation (Fig. 5E). Larger RNAs like GFP could also be reverse transcribed at the guide 2 nick site (fig. S9). In summary, R2Bm can be retargeted by Cas9 to perform TPRT at unrelated loci, and the R2 tag can direct incorporation of cargo RNAs at these sites. Discussion Here we show the structure of a non-LTR retrotransposon during transposition, and we dissect the principles of target DNA and self-RNA recognition. Our structure suggests that R2Bm uses its N-ZnF and Myb domains to locate the endonuclease target sequence, a model that contrasts with the model for other non-LTR retrotransposons where the endonuclease domain is the only determinant of target site selection (28, 29). We identified two essential target site motifs - the RUM and RASIN - that are recognized by R2Bm, but we note that searching the B. mori genome with a RUM-RASIN consensus motif yields many potential off-target sites outside of the ribosomal DNA arrays (fig. S10). We examined the sequence of a previously identified B. mori non-28S insertion in (30) and found the target site had limited similarity with 28S but had a TTAAcG\|T RASIN motif (‘\|’ indicates insertion site, lower-case is deviation from 28S) and a GCTACTTGCGCAT RUM the correct distance upstream of the RASIN (fig. S10). Non-28S insertions however are rare, and so it is likely other factors are important in regulating R2Bm transposition, including chromatin accessibility, other sequence motifs, or the ability of the target DNA to bend and melt. Non-LTR retrotransposons form a diverse family, and even within the R2 superclade there are notable differences between elements. R2Bm is a representative of the R2-D clade of elements, which have a single C2H2 N-terminal ZnF domain, but R2-A clade elements have three tandem N-terminal ZnF domains (31) that may create a more extensive DNA- binding interface with greater stringency in target site selection. More broadly, non-LTR retrotransposons can be divided into two types based on their endonuclease domains: those that like R2Bm use a C-terminal restriction enzyme-like (RLE) domain, and those that, like human LINE-1, use an unrelated N-terminal apurinic/apyrimidinic endonuclease (APE) domain (32, 33). Structure prediction using AlphaFold (34) suggests that, in these retrotransposons, the APE domain has a distinct position to the RLE domain in R2Bm, suggesting there may be mechanistic differences in how target cleavage is coupled to reverse transcription (fig. S5) (35). Nonetheless, the similarity between the DNA interface on the Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 8 R2Bm thumb domain and the corresponding interface in the group IIC intron (fig. S5) suggests this interface might be conserved amongst most non-LTR retrotransposons (19). Indeed, the upstream DNA from R2Bm was easily modeled into an AlphaFold model of human LINE-1 ORF2, including the thumb interactions but also strand separation by the CCHC ZnF domain, which in LINE-1 ORF2 corresponds to the C-terminal cysteine-rich domain (fig. S5). Overall, the results of this work advance our understanding of transposition by non-LTR retrotransposons and suggest avenues for engineering these transposons for targeted gene insertions. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgments: We thank S. Zhu and M. Walsh for valuable discussions; E. Brignole and C. Borsa for the smooth running of the MIT.nano cryo-EM facility, established in part with financial support from the Arnold and Mabel Beckman Foundation; S. Lövestam for a critical reading of the manuscript; and the entire Zhang lab for support and advice. We thank T. H. Eickbush and colleagues for their inspiring and pioneering work on R2 elements. Funding: Helen Hay Whitney Foundation Postdoctoral Fellowship (MEW) Howard Hughes Medical Institute (MEW, FZ) National Institutes of Health grant 2R01HG009761-05 (FZ) Hock E. Tan and K. Lisa Yang Center for Autism Research at MIT (FZ) Yang-Tan Molecular Therapeutics Center at McGovern (FZ) BT Charitable Foundation (FZ) The Phillips family (FZ) J. and P. Poitras (FZ) Data and materials availability: The cryo-EM map has been deposited in the Electron Microscopy Data Bank with accession code EMD-40033. The coordinates for the atomic model have been deposited in the Protein Data Bank with accession code 8GH6. The raw cryo-EM data have been deposited in EMPIAR with accession code EMPIAR-11458. References and Notes 1. Kazazian HH Jr, Moran JV, Mobile DNA in Health and Disease. N. Engl. J. Med 377, 361–370 (2017). [PubMed: 28745987] 2. Priest SJ, Yadav V, Roth C, Dahlmann TA, Kück U, Magwene PM, Heitman J, Uncontrolled transposition following RNAi loss causes hypermutation and antifungal drug resistance in clinical isolates of Cryptococcus neoformans. Nat Microbiol. 7, 1239–1251 (2022). [PubMed: 35918426] Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 9 3. Richardson SR, Doucet AJ, Kopera HC, Moldovan JB, Garcia-Perez JL, Moran JV, The Influence of LINE-1 and SINE Retrotransposons on Mammalian Genomes. Microbiol Spectr. 3, MDNA3–0061– 2014 (2015). 4. Fujiwara H, Site-specific non-LTR retrotransposons. Microbiol Spectr. 3, MDNA3–0001–2014 (2015). 5. Eickbush TH, Eickbush DG, Integration, Regulation, and Long-Term Stability of R2 Retrotransposons. Microbiol Spectr. 3, MDNA3–0011–2014 (2015). 6. Luan DD, Korman MH, Jakubczak JL, Eickbush TH, Reverse transcription of R2Bm RNA is primed by a nick at the chromosomal target site: a mechanism for non-LTR retrotransposition. Cell. 72, 595–605 (1993). [PubMed: 7679954] 7. Yang J, Malik HS, Eickbush TH, Identification of the endonuclease domain encoded by R2 and other site-specific, non-long terminal repeat retrotransposable elements. Proc. Natl. Acad. Sci. U. S. A 96, 7847–7852 (1999). [PubMed: 10393910] 8. Lambowitz AM, Zimmerly S, Group II introns: mobile ribozymes that invade DNA. Cold Spring Harb. Perspect. Biol 3, a003616 (2011). 9. Zimmerly S, Guo H, Perlman PS, Lambowitz AM, Group II intron mobility occurs by target DNA-primed reverse transcription. Cell. 82, 545–554 (1995). [PubMed: 7664334] 10. Luan DD, Eickbush TH, Downstream 28S gene sequences on the RNA template affect the choice of primer and the accuracy of initiation by the R2 reverse transcriptase. Mol. Cell. Biol 16, 4726– 4734 (1996). [PubMed: 8756630] 11. Luan DD, Eickbush TH, RNA template requirements for target DNA-primed reverse transcription by the R2 retrotransposable element. Mol. Cell. Biol 15, 3882–3891 (1995). [PubMed: 7540721] 12. Bibiłło A, Eickbush TH, The reverse transcriptase of the R2 non-LTR retrotransposon: continuous synthesis of cDNA on non-continuous RNA templates. J. Mol. Biol 316, 459–473 (2002). [PubMed: 11866511] 13. Zhao C, Pyle AM, Crystal structures of a group II intron maturase reveal a missing link in spliceosome evolution. Nat. Struct. Mol. Biol 23, 558–565 (2016). [PubMed: 27136328] 14. Lentzsch AM, Stamos JL, Yao J, Russell R, Lambowitz AM, Structural basis for template switching by a group II intron-encoded non-LTR-retroelement reverse transcriptase. J. Biol. Chem 297, 100971 (2021). 15. Jamburuthugoda VK, Eickbush TH, Identification of RNA binding motifs in the R2 retrotransposon-encoded reverse transcriptase. Nucleic Acids Res. 42, 8405–8415 (2014). [PubMed: 24957604] 16. Galej WP, Oubridge C, Newman AJ, Nagai K, Crystal structure of Prp8 reveals active site cavity of the spliceosome. Nature. 493, 638–643 (2013). [PubMed: 23354046] 17. Mahbub MM, Chowdhury SM, Christensen SM, Globular domain structure and function of restriction-like-endonuclease LINEs: similarities to eukaryotic splicing factor Prp8. Mob. DNA 8, 16 (2017). [PubMed: 29151899] 18. Pimentel SC, Upton HE, Collins K, Separable structural requirements for cDNA synthesis, nontemplated extension, and template jumping by a non-LTR retroelement reverse transcriptase. J. Biol. Chem 298, 101624 (2022). 19. Chung K, Xu L, Chai P, Peng J, Devarkar SC, Pyle AM, Structures of a mobile intron retroelement poised to attack its structured DNA target. Science. 378, 627–634 (2022). [PubMed: 36356138] 20. Christensen SM, Bibillo A, Eickbush TH, Role of the Bombyx mori R2 element N-terminal domain in the target-primed reverse transcription (TPRT) reaction. Nucleic Acids Res. 33, 6461– 6468 (2005). [PubMed: 16284201] 21. Christensen S, Eickbush TH, Footprint of the retrotransposon R2Bm protein on its target site before and after cleavage. J. Mol. Biol 336, 1035–1045 (2004). [PubMed: 15037067] 22. Klug A, The discovery of zinc fingers and their applications in gene regulation and genome manipulation. Annu. Rev. Biochem 79, 213–231 (2010). [PubMed: 20192761] 23. Konig P, Giraldo R, Chapman L, Rhodes D, The crystal structure of the DNA-binding domain of yeast RAP1 in complex with telomeric DNA. Cell. 85, 125–136 (1996). [PubMed: 8620531] Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 10 24. Govindaraju A, Cortez JD, Reveal B, Christensen SM, Endonuclease domain of non-LTR retrotransposons: loss-of-function mutants and modeling of the R2Bm endonuclease. Nucleic Acids Res. 44, 3276–3287 (2016). [PubMed: 26961309] 25. Osanai M, Takahashi H, Kojima KK, Hamada M, Fujiwara H, Essential motifs in the 3’ untranslated region required for retrotransposition and the precise start of reverse transcription in non-long-terminal-repeat retrotransposon SART1. Mol. Cell. Biol 24, 7902–7913 (2004). [PubMed: 15340053] 26. Ruschak AM, Mathews DH, Bibillo A, Spinelli SL, Childs JL, Eickbush TH, Turner DH, Secondary structure models of the 3’ untranslated regions of diverse R2 RNAs. RNA. 10, 978–987 (2004). [PubMed: 15146081] 27. Mathews DH, Banerjee AR, Luan DD, Eickbush TH, Turner DH, Secondary structure model of the RNA recognized by the reverse transcriptase from the R2 retrotransposable element. RNA. 3, 1–16 (1997). [PubMed: 8990394] 28. Takahashi H, Fujiwara H, Transplantation of target site specificity by swapping the endonuclease domains of two LINEs. EMBO J. 21, 408–417 (2002). [PubMed: 11823433] 29. Feng Q, Moran JV, Kazazian HH Jr, Boeke JD, Human L1 retrotransposon encodes a conserved endonuclease required for retrotransposition. Cell. 87, 905–916 (1996). [PubMed: 8945517] 30. Xiong Y, Burke WD, Jakubczak JL, Eickbush TH, Ribosomal DNA insertion elements R1Bm and R2Bm can transpose in a sequence specific manner to locations outside the 28S genes. Nucleic Acids Res. 16, 10561–10573 (1988). [PubMed: 2849750] 31. Luchetti A, Mantovani B, Non-LTR R2 element evolutionary patterns: phylogenetic incongruences, rapid radiation and the maintenance of multiple lineages. PLoS One. 8, e57076 (2013). [PubMed: 23451148] 32. Malik HS, Burke WD, Eickbush TH, The age and evolution of non-LTR retrotransposable elements. Mol. Biol. Evol 16, 793–805 (1999). [PubMed: 10368957] 33. Arkhipova IR, Using bioinformatic and phylogenetic approaches to classify transposable elements and understand their complex evolutionary histories. Mob. DNA 8, 1–14 (2017). [PubMed: 28096902] 34. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, Bridgland A, Meyer C, Kohl SAA, Ballard AJ, Cowie A, Romera-Paredes B, Nikolov S, Jain R, Adler J, Back T, Petersen S, Reiman D, Clancy E, Zielinski M, Steinegger M, Pacholska M, Berghammer T, Bodenstein S, Silver D, Vinyals O, Senior AW, Kavukcuoglu K, Kohli P, Hassabis D, Highly accurate protein structure prediction with AlphaFold. Nature. 596, 583–589 (2021). [PubMed: 34265844] 35. Miller I, Totrov M, Korotchkina L, Kazyulkin DN, Gudkov AV, Korolev S, Structural dissection of sequence recognition and catalytic mechanism of human LINE-1 endonuclease. Nucleic Acids Res. 49, 11350–11366 (2021). [PubMed: 34554261] 36. Schmid-Burgk JL, Gao L, Li D, Gardner Z, Strecker J, Lash B, Zhang F, Highly Parallel Profiling of Cas9 Variant Specificity. Mol. Cell 78, 794–800.e8 (2020). [PubMed: 32187529] 37. Crooks GE, Hon G, Chandonia J-M, Brenner SE, WebLogo: a sequence logo generator. Genome Res. 14, 1188–1190 (2004). [PubMed: 15173120] 38. Kimanius D, Dong L, Sharov G, Nakane T, Scheres SHW, New tools for automated cryo-EM single-particle analysis in RELION-4.0. Biochem. J 478, 4169–4185 (2021). [PubMed: 34783343] 39. Bepler T, Morin A, Rapp M, Brasch J, Shapiro L, Noble AJ, Berger B, Positive-unlabeled convolutional neural networks for particle picking in cryo-electron micrographs. Nat. Methods 16, 1153–1160 (2019). [PubMed: 31591578] 40. Jamali K, Kimanius D, Scheres SHW, A graph neural network approach to automated model building in cryo-EM maps (2022), doi:10.48550/arXiv.2210.00006. 41. Casañal A, Lohkamp B, Emsley P, Current developments in Coot for macromolecular model building of Electron Cryo-microscopy and Crystallographic Data. Protein Sci. 29, 1069–1078 (2020). [PubMed: 31730249] 42. Croll TI, ISOLDE: a physically realistic environment for model building into low-resolution electron-density maps. Acta Crystallographica Section D: Structural Biology. 74, 519–530 (2018). [PubMed: 29872003] Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 11 43. Liebschner D, Afonine PV, Baker ML, Bunkóczi G, Chen VB, Croll TI, Hintze B, Hung LW, Jain S, McCoy AJ, Moriarty NW, Oeffner RD, Poon BK, Prisant MG, Read RJ, Richardson JS, Richardson DC, Sammito MD, Sobolev OV, Stockwell DH, Terwilliger TC, Urzhumtsev AG, Videau LL, Williams CJ, Adams PD, Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. Acta Crystallogr D Struct Biol. 75, 861– 877 (2019). [PubMed: 31588918] 44. Pettersen EF, Goddard TD, Huang CC, Meng EC, Couch GS, Croll TI, Morris JH, Ferrin TE, UCSF ChimeraX: Structure visualization for researchers, educators, and developers. Protein Sci. 30 (2021), doi:10.1002/pro.3943. 45. Li S, Olson WK, Lu X-J, Web 3DNA 2.0 for the analysis, visualization, and modeling of 3D nucleic acid structures. Nucleic Acids Res. 47, W26–W34 (2019). [PubMed: 31114927] 46. Grant CE, Bailey TL, Noble WS, FIMO: scanning for occurrences of a given motif. Bioinformatics. 27, 1017–1018 (2011). [PubMed: 21330290] 47. Eickbush DG, Burke WD, Eickbush TH, Evolution of the R2 retrotransposon ribozyme and its self-cleavage site. PLoS One. 8, e66441 (2013). [PubMed: 24066021] Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 12 Fig. 1. Cryo-EM structure of the R2Bm retrotransposon. (A) Domains of the R2Bm retrotransposon. ZnF, zinc finger; NTE, N-terminal extension; RT, reverse transcriptase; RLE, restriction-like endonuclease. (B) Schematic of target- primed reverse transcription (TPRT). (C) Denaturing gel of in vitro TPRT reactions on a labeled 211-bp 28S DNA target. The same gel was visualized by Cy5 fluorescence and toluidine blue staining. (D) Cryo-EM density of the R2Bm TPRT complex. (E) Cartoon of the cryo-EM structure. Stars represent active sites. (F) Atomic model for the R2Bm Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 13 TPRT complex. (G) Reverse transcriptase domain and template/primer duplex. (H) Reverse transcriptase active site. Cryo-EM density is shown as a gray transparent surface. Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 14 Fig. 2. Target DNA recognition upstream of the R2 cleavage site. (A) Schematic of interactions with the target DNA. Bases are numbered relative to the bottom strand cleavage site. Positions of protein domains are shown by shaded rectangles. (B) Structure of R2Bm around the upstream DNA sequences. (C) Effect of upstream DNA mutations on target cleavage. The schematic shows the sequences of five DNA sequences tested in top-strand sense; dots represent bases identical to wildtype. Red triangle, bottom strand cleavage site. Denaturing gels show in vitro TPRT reactions on labeled 211-bp 28S DNA targets. ΔN, deletion of N-terminal N-ZnF and Myb domains. ΔRT6a, deletion of Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 15 residues 672 – 677 (DGHRKK) of the RT6a loop. (D) Screen for identifying active RUM sequences. Nicking sites of R2Bm and the restriction endonuclease Nt.BbvCI are shown by triangles. (E) Sequence logo for sequences enriched in the RUM screen. (F, G, H) Details of interactions between the target DNA and the N-ZnF, Myb, and RT6a loop. (I) Effect of altering the distance between the RUM and RASIN motifs. Denaturing gel shows in vitro TPRT reactions on labeled 211-bp 28S DNA targets. Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 16 Fig. 3. Target DNA recognition at the R2 cleavage site. (A) Interactions of the top and bottom strands of the target DNA with the ZnF domain of R2Bm. Star, RLE active site. (B) Interactions of the DNA bottom strand with the RLE domain. (C) Interactions of the DNA top strand with the RLE domain. Residues mutated in the RD>AA mutant are highlighted. (D) RASIN sequence requirements for bottom strand cleavage. The labeled 211-bp 28S DNA targets were incubated with R2Bm and 3′ UTR RNA in the absence of dNTPs. The reactions were analyzed with a denaturing gel. Mutations are notated in top-strand sense, but both strands were mutated. (E) Denaturing gel showing R2Bm cleavage and TPRT activity on partially-stranded substrates. Reactions contained a fluorescein-labeled 76-nt bottom strand. Reactions as indicated also contained 17 nt of downstream top strand sequence (17d), 32 nt of upstream top strand sequence (32u), or 60 nt of top strand sequence fully complementary to the bottom strand spanning the upstream and downstream regions. RD>AA; R2Bm R901A D902A. Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 17 Fig. 4. Interactions of R2Bm with the 3′ UTR RNA. (A) Secondary structure diagram of the 3′ UTR RNA, based on (26). Thicker strokes represent nucleotides visible in the cryo-EM density. Nucleotides are numbered from the first base of the 3′ UTR (the base following the stop codon). (B) Structure of the 3′ UTR RNA core and the R2Bm NTE-1 domain. Dotted lines, hydrogen bonds. (C) Low-pass filtered cryo-EM map. (D) Interactions between 3′ UTR bases. Dotted lines, hydrogen bonds. (E) Secondary structure of the R2 tag RNA. Unshaded bases are not in the full-length 3′ UTR. (F) Denaturing gel of in vitro TPRT reactions on a labeled 211-bp 28S DNA target using various R2 RNAs. Highlighted mutants are in the J1/2 region. The same gel was visualized by Cy5 fluorescence and toluidine blue staining. (G) The R2-tag allows TPRT of cargo RNAs. Denaturing gel shows TPRT reactions with equimolar amounts of the indicated RNAs and a labeled 211-bp 28S DNA target. R2 tag (43 nt) was added to the 3′ end of a 239-nt RNA encoding the CMV promoter or a 764-nt RNA encoding GFP. Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 18 Fig. 5. Mechanism and retargeting of first strand synthesis by R2Bm. (A) Model for the initial stages of target site cleavage and first strand synthesis. (B) Design of R2Bm + Cas9 experiments. (C) Complementation of DNA target site mutants by Cas9 cleavage in trans and cis. The denaturing gel shows in vitro TPRT reactions on a labeled 211-bp target corresponding to the wild-type 28S target, or two 235-bp targets: one where the RASIN TAAGGTA is replaced by 31 bp of unrelated sequence, and another where the 13-bp RUM is additionally scrambled. R2Bm and SpCas9(H840A) were added in trans, or in cis connected by a 33XTEN linker ( fusion indicated by a shaded box). The sgRNA is complementary to the inserted sequence and nicks 40 nt from the last RUM base. The R2 RNA is the 3′ UTR with 5 nt of 3′ homology to the nick site. (D) Sequences used for retargeting R2Bm to an unrelated locus from the Drosophila virilis genome. (E) Denaturing Science. Author manuscript; available in PMC 2023 September 13. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wilkinson et al. Page 19 gel of in vitro TPRT reactions on the labeled 192-bp Drosophila virilis target. sgRNAs are numbered as in (D); all R2 RNAs or R2-tagged RNAs have 10 nt of 3′ homology to the nick site of the sgRNA. Science. Author manuscript; available in PMC 2023 September 13.
10.1371_journal.pbio.3002483	RESEARCH ARTICLE GABAergic regulation of striatal spiny projection neurons depends upon their activity state Michelle Day1, Marziyeh Belal1, William C. Surmeier1, Alexandria Melendez2, David Wokosin1, Tatiana Tkatch1,3, Vernon R. J. ClarkeID 1,3, D. James SurmeierID 1,3* a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Day M, Belal M, Surmeier WC, Melendez A, Wokosin D, Tkatch T, et al. (2024) GABAergic regulation of striatal spiny projection neurons depends upon their activity state. PLoS Biol 22(1): e3002483. https://doi.org/10.1371/journal. pbio.3002483 Academic Editor: Alberto Bacci, ICM - Institut du Cerveau et de la Moelle e´pinière Hoˆpital Pitie´- Salpêtrière47, bd de l’Hoˆpital, FRANCE Received: March 17, 2023 Accepted: December 26, 2023 Published: January 31, 2024 Copyright: © 2024 Day et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All datasets are publicly available: Figs 1–8: dx.doi.org/10.5281/ zenodo.10386854. S1–S4 Figs: dx.doi.org/10. 5281/zenodo.10387118. R code for graphical outputs and statistical analyses of Figs 1–4: dx.doi. org/10.5281/zenodo.10386496. Modelling code and R code to recreate all modelling figures dx.doi. org/10.5281/zenodo.10162264. Funding: This work was supported by grants to DJS from the Cure Huntington’s Disease Initiative 1 Department of Neuroscience, Feinberg School of Medicine, Northwestern University, Chicago, Illinois, United States of America, 2 Department of Neurology, Baylor College of Medicine, Houston, Texas, United States of America, 3 Aligning Science Across Parkinson’s (ASAP) Collaborative Research Network, Chevy Chase, Maryland, United States of America * [email protected] Abstract Synaptic transmission mediated by GABAA receptors (GABAARs) in adult, principal striatal spiny projection neurons (SPNs) can suppress ongoing spiking, but its effect on synaptic integration at subthreshold membrane potentials is less well characterized, particularly those near the resting down-state. To fill this gap, a combination of molecular, optogenetic, optical, and electrophysiological approaches were used to study SPNs in mouse ex vivo brain slices, and computational tools were used to model somatodendritic synaptic integra- tion. In perforated patch recordings, activation of GABAARs, either by uncaging of GABA or by optogenetic stimulation of GABAergic synapses, evoked currents with a reversal poten- tial near −60 mV in both juvenile and adult SPNs. Transcriptomic analysis and pharmacolog- ical work suggested that this relatively positive GABAAR reversal potential was not attributable to NKCC1 expression, but rather to HCO3- permeability. Regardless, from down-state potentials, optogenetic activation of dendritic GABAergic synapses depolarized SPNs. This GABAAR-mediated depolarization summed with trailing ionotropic glutamate receptor (iGluR) stimulation, promoting dendritic spikes and increasing somatic depolariza- tion. Simulations revealed that a diffuse dendritic GABAergic input to SPNs effectively enhanced the response to dendritic iGluR signaling and promoted dendritic spikes. Taken together, our results demonstrate that GABAARs can work in concert with iGluRs to excite adult SPNs when they are in the resting down-state, suggesting that their inhibitory role is limited to brief periods near spike threshold. This state-dependence calls for a reformulation for the role of intrastriatal GABAergic circuits. Introduction The striatum is the largest component of the basal ganglia circuitry regulating goal-directed actions and habits [1,2]. The principal neurons of the striatum are GABAergic spiny projection neurons (SPNs). SPNs integrate information arising from extrastriatal glutamatergic neurons, PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 1 / 26 Foundation, Inc. (A5071), 350 7th Ave, Suite 200, NY 10001 New York, United States, https:// chdifoundation.org/; the JPB Foundation (GR- 2021-2960), 875 Third Avenue 29th Floor New York, NY 10022, https://www.jpbfoundation.org; the National Institute of Neurological Disorders and Stroke (R37 NS034696), P.O. Box 5801. Bethesda, MD 20824; https://www.ninds.nih.gov; and Aligning Science Across Parkinson’s (ASAP- 020551) through the Michael J. Fox Foundation for Parkinson’s Research (MJFF); Aligning Science Across Parkinson’s Collaborative Research Network, Chevy Chase, MD, 20815, https:// parkinsonsroadmap.org. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abbreviations: 2PLSM, two-photon laser scanning microscopy; AAV, adeno-associated virus; CA, carbonic anhydrase; ChAT, choline acetyltransferase; ChI, cholinergic interneuron; FSI, fast-spiking GABAergic interneuron; iGluR, ionotropic glutamate receptor; LTSI, low-threshold spike GABAergic interneuron; nAChR, nicotinic acetylcholine receptor; NGFI, neurogliaform interneuron; NMDAR, N-methyl-D-aspartate receptor; PBS, phosphate-buffered saline; PFA, paraformaldehyde; PSP, postsynaptic potential; qPCR, quantitative polymerase chain reaction; SPN, spiny projection neuron; THI, tyrosine hydroxylase interneuron. GABAergic regulation of striatal spiny projection neurons depends upon their activity state intrastriatal GABAergic interneurons, and collaterals of neighboring GABAergic SPNs. These intrastriatal GABAergic synapses, which constitute about 20% of all SPN synapses [3], and the postsynaptic GABAARs transducing the effects of synaptically released GABA, are widely viewed as inhibitory, working in opposition to dendritic excitatory glutamatergic input to sup- press SPN spiking [4]. Although the ability of SPN GABAARs to suppress spiking is clearcut, characterizing them as categorically inhibitory fails to consider 2 salient features of adult SPNs. First, the reversal potential of GABAARs of SPNs appears to be relatively depolarized [5–7]. In particular, perfo- rated patch recordings from relatively immature SPNs place the GABAAR reversal potential near −60 mV [6]. Indirect estimates of the GABAAR reversal potential have yielded similar val- ues in more mature neurons [7,8]. Another key feature of SPN physiology that is relevant to the functional impact of GABAARs is their resting membrane potential. In the absence of syn- aptic input, the physiology of SPNs is dominated by somatodendritic Kir2 K+ channels, which drives the membrane potential to near the K+ equilibrium potential of roughly −80 mV [9]. In vivo, this so-called “down-state” in adult SPNs is interrupted by synaptically driven periods of depolarization when the somatic membrane transitions to potentials near −60 mV, close to spike threshold [10]. Although the synaptic determinants of these “up-states” in vivo are not well defined, ex vivo preparations have revealed that the depolarization arising from clustered glutamatergic synaptic activity on distal dendrites can drive regenerative, plateau potentials that mimic up-states [11–13]. A critical trigger of SPN dendritic plateau potentials is the engagement of N-methyl-D-aspartate receptors (NMDARs), which depends not only upon glutamate but also membrane depolarization and the displacement of pore-blocking Mg2+. This unblocking requires that the membrane depolarize to around −60 mV. Although recent work has shown that dendritic spikes in SPNs can be abbreviated by trailing GABAergic input at the site of glutamatergic stimulation [13], the roles of dendritic location, timing, and synap- tic strength in determining the interaction between GABAARs and ionotropic glutamate receptors (iGluRs) have not been systematically explored in SPNs. To better understand the role of GABAAR signaling in adult SPNs, a combination of experi- mental and computational approaches was employed. These studies revealed that juvenile and adult SPNs do not express significant levels of mRNA coding for NKCC1 but do express mRNA -/Cl- transporters. Perforated patch recordings from SPNs in ex vivo brain for KCC2 and HCO3 slices from mice over a broad range of ages (1 to 9 months) revealed that the GABAAR reversal potential was stable and near −60 mV. Thus, engagement of either synaptic or extra-synaptic GABAARs excited SPNs in the down-state, pushing them toward spike threshold. Furthermore, both experimental and modeling work demonstrated that leading dendritic GABAAR postsyn- aptic potentials (PSPs) effectively summed with trailing, spatially coincident iGluR-mediated depolarization. Moreover, computational studies revealed that when the GABAergic input was electronically remote from iGluRs, the 2 inputs effectively worked together to drive membrane depolarization regardless of timing. Given that SPNs in vivo appear to reside primarily at mem- brane potentials well below the reversal potential for GABAARs, these results suggest that physi- ological consequences of SPN GABAergic synapses should not be considered as simply inhibitory and that in a wide range of situations GABAARs work in concert with iGluRs to move SPNs closer to spike threshold, promoting their participation in network function. Results SPNs robustly expressed mRNA for KCC2, but not NKCC1 It is commonly thought that the reversal potential of GABAARs [14,15] is governed in large part by the balance between the plasma membrane cation/Cl- co-transporters–NKCC1 and PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 2 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 1. KCC2 mRNA is robustly expressed in SPNs, but not NKCC1. (A) The RiboTag construct AAV5-DOI-hSyn- RpI22I1-3Xflag-2A-eGFP was injected into the striatum of Adora2a-cre mice at P18 or at 6 months of age. (B) The coronal slice images demonstrate both the coverage and restriction to the striatum of the stereotaxically injected AAV carrying the RiboTag and eGFP genes (stereotaxic injection coordinates: ML = −1.85, AP = +0.74, DV = −3.50). Scale bars = 1 mm and 20 μm. Ten days later, the infected tissue (green fluorescence) was dissected out with the aid of fluorescence microscopy and qPCR was performed. (C) mRNA abundance (ΔCT) levels for the chloride cotransporters NKCC1 (SLC12A1) and KCC2 (SLC12A5) were determined by qPCR in striata from Adora2a-cre mice 4 weeks and 6 months of age. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/ zenodo.10386854. AAV, adeno-associated virus; qPCR, quantitative polymerase chain reaction; SPN, spiny projection neuron. https://doi.org/10.1371/journal.pbio.3002483.g001 KCC2 [8,16]. To determine whether SPNs expressed NKCC1 and KCC2, the striata of 1- and 6-month-old Adora2-Cre mice were stereotaxically injected with an adeno-associated virus (AAV) carrying a DIO-RiboTag expression construct [17] (Fig 1A). Four weeks later, mice were killed; total striatal mRNA and RiboTag-associated mRNA were harvested for quantita- tive polymerase chain reaction (qPCR) and RNASeq analyses (Fig 1B). These experiments revealed that iSPNs robustly expressed mRNA coding for KCC2 (Slc12a5), but not NKCC1 (Slc12a2) (Fig 1C). The relative expression of these transcripts did not change within the time window examined (Fig 1C). The expression of RiboTag harvested, iSPN-specific transcripts was like the mRNA harvested from the entire striatum (Fig 1C). The GABAAR reversal potential was near −60 mV in both young and adult SPNs To determine the reversal potential of GABAARs in SPNs, ex vivo brain slices were prepared from young adult (6 to 7 months old) mice and then gramicidin perforated patch recordings were made from identified SPNs. Gramicidin is selectively permeable to monovalent cations, leaving the intracellular Cl- concentration ([Cl-]i) unperturbed. To visualize dendrites, SPNs were sparsely labeled using an AAV carrying a SuperClomeleon expression plasmid [18] (Fig 2A). To activate GABAARs, RuBi-GABA was uncaged on the soma and dendrites using a blue laser spot (Fig 2A). The somatic membrane potential was clamped at membrane poten- tials between −50 and −70 mV prior to uncaging GABA and the resulting currents monitored (Fig 2B). The amplitude and polarity of uncaging evoked currents were then plotted as a func- tion of somatic membrane potential. The estimated reversal potential for somatic GABAARs PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 3 / 26 AAge (months) 1234567Harvest tissueisolate mRNA Harvest tissueisolate mRNA RiboTag-4-20246NKCC1 (SLC12A1)(cid:2)-CT (HPRT)KCC2 (SLC12A5)P30P180P30P180increasing abundanceCstriatumiSPNstriatumiSPNstriatumiSPNstriatumiSPNRiboTag-eGFPstriatumcerebral cortexRiboTag-eGFPBPLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 2. The GABAAR reversal potential is near −60 mV in both young and adult SPNs. (A) Clomeleon-expressing iSPNs allowed visual identification of dendrites in gramicidin perforated-patch recording conditions where cells cannot be loaded with dyes via internal-pipette solution (920 nm laser maximum projection image, scale bar = 40 μm). When low (<100 MO) access-resistance was achieved in voltage-clamp mode, RuBi-GABA (10 μM) was uncaged with a 473 nm laser spot (approximately 1 μm diameter, 1 ms) in the presence of the synaptic blockers: TTX (1 μM), AP5 (50 μM), NBQX (5 μM), CGP-55845 (1 μM). The laser was targeted to the somatic region or to distal dendrites (blue spots, projection image). (B) Representative voltage traces showing GABA responses, recorded in serial, from the soma (top traces, scale bars = 20 pA/2 s) or the dendrite (lower traces, scale bars = 10 pA/2 s) as the membrane was manually stepped from −70 mV to −50 mV. (C) Plot of the current/voltage relationship between somatic and dendritic activation. The data, represented by medians with interquartile ranges, did not differ significantly between the soma and dendritic compartments (n = 5 each; soma, dendrite; slope = 0.96, 0.71; x-intercept = −55.9, −59.6 mV; R2 = 0.77, 0.85, respectively). Current measurements were rounded to the nearest 0.5 pA. (D) Current-clamp experiments in gramicidin PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 4 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state perforated-patch mode were performed to examine age-dependent shifts in reversal potential. Here, Adora2a-eGFP positive iSPNs could be visually identified and patched. When low (<100 MO) access-resistance was achieved in current-clamp mode, the resting membrane potential along with series of hyperpolarizing and depolarizing steps were used to examine cell health (traces, scale bars = 20 mV/200 ms). (E) RuBi-GABA (10 μM) was uncaged over the full-field (3 ms duration, 60× lens) with a 473 nm LED in the presence of the synaptic blockers: AP5 (50 μM), NBQX (5 μM), CGP-55845 (1 μM). Representative current traces showing GABA responses as the membrane was manually stepped from −80 mV to −50 mV, scale bars = 5 mV/200 ms. (F) Plot of the change in PSP amplitude at P30, P90, and P270. The data, represented by medians with interquartile ranges, did not differ significantly between the 3 ages tested (n = 5 each P30, P90, P270; slope = 0.75, 0.70, 0.75; x-intercept = −61.6, −61.5, −61.5 mV; R2 = 0.93, 0.91, 0.94, respectively). Values were calculated to the nearest 0.5 mV for the ΔV/V measurements. The data shows that the reversal for GABA-induced current is a full 20 mV+ above the resting membrane potential for SPNs, typically, −80 to −85 mV. (G) The addition of bumetanide (NKCC1 blocker, 10 μM) did not change the GABAAR reversal potential significantly (n = 5, p = 0.4076). (H) Perforated patch recordings were obtained from Adora2a-eGFP iSPNs and then the reversal potential of GABAARs determined before and after inhibition of CA with acetazolamide (aceta, 10 μM). (I) Representative traces recorded from a visually identified iSPN from an Adora2a-eGFP mouse in gramicidin perforated patch in current-clamp mode in the synaptic blockers: AP5 (50 μM), NBQX (5 μM), CGP-55845 (1 μM), MPEP (1 μM), and CPCCOEt (50 μM). RuBi-GABA (15 μM) was uncaged using a single LED pulse (470 nm, 25 ms). The pulse was applied at an interval of 30 s while manually stepping the cell to different potentials from −80 to −50 mV, scale bars = 10 mV/100 ms. (J) Summary data shows that application of acetazolamide shifted the reversal of the GABA-induced current to more negative potentials (p = 0.03125, n = 6). Membrane potentials were adjusted to correct for the estimated liquid junction potential and then binned into 5 mV increments (−70, −65, −60, −55 and −50 mV). The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. CA, carbonic anhydrase; PSP, postsynaptic potential; SPN, spiny projection neuron. https://doi.org/10.1371/journal.pbio.3002483.g002 was near −55 mV (Fig 2C). Dendritic uncaging of GABA evoked currents which also reversed in polarity near −60 mV (Fig 2C). To determine if there was any shift in the GABAAR reversal potential after weaning, grami- cidin perforated patch recordings were made from SPNs in ex vivo brain slices taken from mice at 3 ages: young (approximately 1 month old), young adult (6 to 7 months old), and adult (approximately 9 months old) mice. SPNs recorded in this mode displayed the characteristic inward rectification, delayed time to the first spike at rheobase, and sustained repetitive spiking with suprathreshold current injection (Fig 2D). The membrane potential changes evoked in SPNs by RuBi-GABA uncaging on the peri-somatic membrane reversed near −60 mV at all ages (Fig 2E and 2F). Why is the reversal potential of the GABAARs relatively depolarized? The striatal circuitry is largely quiescent in the ex vivo brain slice, making it highly unlikely that ongoing GABAer- gic signaling was loading neurons with Cl- and pushing the reversal potential in a depolarized direction. Despite the absence of detectable levels of its mRNA, the functional contribution of NKCC1 to the reversal potential of GABAAR was tested by bath application of the NKCC1-se- lective antagonist bumetanide (10 μM); bumetanide did not change the GABAAR reversal -, the other potential (Fig 2G). Since GABAARs exhibit a significant permeability to HCO3 - equilibrium potential [8]. To determinant of the GABAAR reversal potential is the HCO3 - in determining the GABAAR reversal potential, perfo- assess the role of intracellular HCO3 rated patch recordings were obtained from SPNs in ex vivo brain slices (as described above) and then the reversal potential of GABAARs determined before and after inhibition of car- bonic anhydrase (CA) with acetazolamide (10 μM). In vivo, CA catalyzes the conversion of cytosolic CO2 to H+ and HCO3 - (Fig 2H) [8]. RiboTag/RNASeq analysis revealed that iSPNs expressed 2 cytosolic CA subtypes with intracellular catalytic domains (Car2>Car7; RNASeq read ratio = 4.3)—in agreement with previous work [19]. Nonspecific inhibition of these CAs with acetazolamide led to a significant negative shift in the reversal potential of GABAARs (Fig 2I and 2J), consistent with the inference that the relatively depolarized GABAAR reversal - flux. potential in SPNs was attributable to HCO3 GABAAR engagement depolarized SPNs in the down-state To study the role of synaptic GABA release, a mixed population of striatal GABAergic inter- neurons were activated by optogenetic stimulation of cholinergic interneurons (ChIs) [20]. Working through nicotinic acetylcholine receptors (nAChRs), ChIs can activate both neuro- gliaform interneurons (NGFIs) and tyrosine hydroxylase interneurons (THIs), giving rise to PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 5 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 3. Optogenetic stimulation of ChIs or NPY-expressing interneurons evokes robust GABAAR-mediated PSPs in both iSPNs and dSPNs. (A) AAV9-hSyn-chronos-flex-eGFP was stereotaxically injected into the striatum of two-month-old ChAT-Cre X D1tdTomato mice (Stereotaxic coordinate injection: ML = −1.2, AP = −0.7, DV = −3.4). The coronal confocal slice image shows the expression of Chronos (green cells) in a ChAT-cre neuron (cholinergic interneurons) along with dSPNs expressing tdTomato (red cells, scale bar = 40 μm). The tissue was dissected and recorded from 21 days postinjection. (B) The mean (± SEM) of ChI-evoked EPSP responses recorded from visually identified SPNs in gramicidin perforated patch in current-clamp mode in the presence of synaptic blockers: NBQX (5 μM), AP5 (50 μM), CGP-55845 (1 μM), MPEP (1 μM), and CPCCOEt (50 μM). The LED pulse (470 nm, 5 ms) was applied at an interval of 60 s. The traces recorded before and after the addition of gabazine (10 μM). Scale bars = 1 mV/100 ms. (C) Box plots of data from dSPNs (n = 8) and iSPNs (n = 6). (D) NPY-Cre X D1tdTomato mice were injected as described in (A). Confocal image showing NPY-Cre neurons expressing Chronos (green) and dSPNs expressing tdTomato (red, scale bar = 40 μm). (E) Mean (+ SEM) of NPY-Cre-evoked EPSP responses recorded from visually identified dSPNs in gramicidin perforated patch in current-clamp mode in the presence of blockers as described in (B) before and after the addition of Gabazine (10 μM). Traces from dSPN recorded in NPY (n = 4). Scale bars = 1 mV/100 ms. (F) Summary data for dSPNs (n = 4) and for iSPNs (n = 4). The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. ChAT, choline acetyltransferase; ChI, cholinergic interneuron; PSP, postsynaptic potential; SPN, spiny projection neuron. https://doi.org/10.1371/journal.pbio.3002483.g003 GABAAR-mediated currents in SPNs [21]. To monitor evoked responses in SPNs, perforated patch recordings were made from identified iSPNs or dSPNs using the approach described above. To optogenetically activate ChIs, an AAV carrying a Cre recombinase-dependent expression construct for Chronos was injected into the striatum of transgenic mice expressing Cre recombinase under the control of the choline acetyltransferase (ChAT) promoter PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 6 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state (Fig 3A). In the ex vivo brain slice, SPNs are quiescent and reside in the down-state near −80 mV [10]. As predicted from the GABA uncaging studies above, optical stimulation of ChIs in the presence of iGluR antagonists evoked depolarizing, PSPs in SPNs that were blocked by the GABAAR antagonist gabazine (Fig 3B and 3C). To simplify the afferent circuitry engaged in these experiments, mice expressing Cre recom- binase under the control of the NPY promoter (NPY-Cre) were injected with the same AAV vector used in the ChAT-Cre mice (Fig 3D). NPY is expressed by NGFis and low-threshold spike GABAergic interneurons (LTSIs) [4]—both of which make GABAergic synapses primar- ily on SPN dendrites [22]. Optogenetic activation of NPY-expressing interneurons alone pro- duced depolarizing PSPs that were kinetically similar to those evoked by optogenetic stimulation of ChIs (Fig 3E and 3F). GABAAR activation enhanced the depolarization produced by iGluRs As shown previously, in both SPNs and pyramidal neurons [5,6,23], a depolarizing GABAAR input can boost the response to a trailing intrasomatic current injection and enhance the prob- ability of spiking. However, GABAAR activation also can suppress spike generation by mem- brane shunting and pushing the membrane potential below spike threshold, which is typically between −45 and −50 mV [23]. How might the interaction between GABAARs and iGluRs play out in dendrites? A key fea- ture of SPN dendrites beyond about the first major branch point (approximately 80 μm from the soma) is the ability to generate dendritic spikes or plateau potentials that can last for 50 to 200 ms [11–13,24]. These dendritic spikes require the temporal convergence of 10 to 15 glutamater- gic inputs over a relatively short stretch (approximately 20 μm) of dendrite, which produces enough of a local depolarization to engage NMDARs and voltage-dependent Ca2+ channels. Pre- vious experimental and modeling work has shown that opening GABAARs near the site of gluta- matergic stimulation after spike initiation can truncate them, much like somatic situation described above. Indeed, as modeling suggests that the dendritic membrane potential during these spikes rises close to 0 mV, GABAAR opening should hyperpolarize the dendrites [13]. But, what if the GABAAR activation precedes the glutamatergic input to dendrites? A priori, one might predict that the dendritic depolarization produced by GABAAR opening would enhance the response to trailing glutamatergic input, much like the situation described at the soma. To test this hypothesis, 2 sets of experiments were performed. Identified iSPNs or dSPNs were recorded from in whole-cell mode to allow them to be filled with a dye (Alexa 568) and imaged using two-photon laser scanning microscopy (2PLSM) [25]. The [Cl-] in the pipette was adjusted to yield a GABAAR reversal potential near −60 mV. Next, a region of par- focal dendrite was identified to allow two-photon uncaging of DNI-glutamate at visualized spine heads [11,26,27]. In the first set of experiments, dendritic GABAARs were activated by optogenetic stimulation of ChIs as described above. Because of their large axonal field and those of the NGFIs/THIs they activate [4,28], optogenetic stimulation of ChIs should produce a diffuse GABAergic input to the dendrites of the recorded SPN (Fig 4A and 4B). As shown above, optogenetic stimulation of ChIs alone evoked a consistent but modest somatic depolari- zation (Fig 4C and 4D). Dendritic uncaging of glutamate alone also evoked a somatic depolar- ization. The number of axospinous sites stimulated was adjusted to be subthreshold for dendritic spike generation (assessed by the decay of membrane potential after termination of uncaging) (Fig 4C and 4D). When this uncaging event was preceded by ChI-evoked GABAAR depolarization, the resulting magnitude and duration of the somatic depolarization was signifi- cantly increased in both types of SPN (Fig 4E). A scatter plot of the algebraic sum of the ampli- tudes of the GABAergic and glutamatergic PSPs in isolation against the amplitude of the PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 7 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 4. ChI-evoked stimulation of NGF interneurons and synaptic GABA release enhances glutamate-evoked state transitions. (A) Maximum projection image of a visually identified dSPN from a D1R-tdTomato x ChAT-cre mouse with a high magnification image of a distal dendrite where 720 nm 2PLSM spot uncaging of DNI-Glu (2PLU, 5 mM) was conducted (red dots). Tomato+ dSPNs were patched in whole-cell mode and the cells were loaded with Alexa 568 for clear identification of dendrites and spines. Scale bars = 40 μm cell, 5 μm dendrite. (B) Scheme for interrogating endogenous GABA release from NGFIs onto SPNs via optogenetic stimulation of ChAT-cre mice expressing Chronos. (C, D) Throughout the dendrites, glutamate uPSPs in dSPNs and iSPNs can be evoked by uncaging DNI-Glu (5 mM, 1 × 15 spines, 1 ms pulses at 500 Hz, red traces, 720 nm laser) while stimulating GABA release from NGFIs with the blue laser (1 × 3 ms pulse, blue traces, 473 nm, within approximately 20 μm of the dendrite). From the quiescent PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 8 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state down-state, GABAAR activation is depolarizing and pushes SPNs toward enhanced dendritic integration in both dSPN and iSPN dendrites (Glu-2PLU + GABAA opto = black trace, scale bars = 5 mV/200 ms). (E) Summary data showing the enhancement in amplitude and duration of the plateaus at ½ the maximum amplitude (1/2max) in iSPNs and dSPNs combined (n = 11 total: 3 iSPNs + 8 dSPNS; p < 0.001 for both amplitude and 1/2max duration, respectively). (F) Scatter plot of duration at ½ maximum amplitude vs. amplitude for clustered glutamate alone (red) and following GABAAR activation (black). Median effects (open circles) and the median absolute difference as capped lines are also illustrated. All experiments are conducted in the appropriate cocktail of synaptic blockers: CGP-55845 (1 μM), MPEP (1 μM), and CPCCOEt (50 μM). The data underlying the graphs shown in the figure can be found in dx.doi.org/10. 5281/zenodo.10386854. 2PLSM, two-photon laser scanning microscopy; ChAT, choline acetyltransferase; ChI, cholinergic interneuron; SPN, spiny projection neuron. https://doi.org/10.1371/journal.pbio.3002483.g004 response to the combined stimulation revealed that the 2 inputs almost invariably summed lin- early or supra-linearly (S1 Fig). A scatter plot of the amplitude and duration of iGluR-medi- ated responses demonstrated that prior engagement of GABAergic interneurons (by ChI stimulation) enhanced the iGluR-mediated responses (Fig 4F). Thus, transiently opening den- dritic GABAARs produced a dendritic membrane potential change that enhanced the ability of subsequent dendritic glutamatergic input to push SPNs toward the local spike threshold. Computational modeling of dendritic integration in SPNs Although intriguing, the experimental results presented are limited by the inability to control the timing and location of GABAergic input to dendrites in a rapid precise manner. Under- standing how the timing and dendritic location of GABAAR activation modulates the response to clustered excitatory input could provide insight into the role of GABAergic interneurons in striatal computation. To help achieve a better grasp of the mechanisms underlying this interac- tion, a modified NEURON model of a dSPN [13,29–31] was used to assess the impact of tim- ing and location of GABAergic input on the response to clustered glutamatergic synaptic input to a stretch of distal dendrite. As observed experimentally, clustered glutamatergic input was able to generate NMDAR-dependent, dendritic spikes or plateau potentials when deliv- ered to distal dendrites of a quiescent neuron (Fig 5A–5C). When GABAergic synapses were activated near glutamatergic synapses, the model behaved as previously described by Du and colleagues. That is, GABAergic input at almost any point during the dendritic spike (when the local membrane potential was near −30 mV) led to inhi- bition of both the dendritic and somatic membrane potential (Fig 5D–5F). However, the impact of GABAergic input was very different when the site of stimulation was at some dis- tance from that of glutamatergic stimulation. For example, if the GABAergic input was distrib- uted at distal locations across the dendritic tree, as predicted to happen following ChI or NGFI/THI activation, the effect was consistently excitatory. To illustrate this point, the distrib- uted GABAergic input was followed by a subthreshold dendritic glutamatergic input. In this scenario, the combination of GABAergic and glutamatergic input led to an NMDAR-depen- dent dendritic spike (Figs 5G–5I and S4A–S4C)—just as seen experimentally using optoge- netic stimulation of GABAergic interneurons and 2P uncaging of glutamate. The default value for cytoplasmic resistivity (Ra) in these computational simulations was 200 O cm. However, Ra is a difficult parameter to measure experimentally and there is little overall consensus as to its true value and estimates vary at least 5-fold (70 to 350 O cm) [32]. For this reason, simulations were repeated with Ra set to 100 O cm. The results were qualita- tively similar to those described above (S2 Fig), suggesting that Ra is not a major factor in our simulations when varied within the proposed physiological range. In many of the previous studies examining the impact of GABAARs on dendritic integra- tion of glutamatergic input, the focus has been on the role of timing and location dependent GABAAR -mediated shunting of iGluR-evoked EPSPs on the same dendrite [33–35]. As PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 9 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 5. Computational modeling of dSPN dendritic activity. (A) Morphology of reconstructed dSPN with cartoon to illustrate the direction of stimulation along a dendrite. (B) Synaptic potentials recorded at the soma in response to clustered spine activation (18 neighboring glutamatergic synapses stimulated sequentially at 1 ms intervals) in 2 separate dendrites. A dendritic spike is generated to the distal (orange trace) but not proximal dendritic input (blue trace). The quarterdrop duration is defined as the time interval between the last stimulation and the time for the membrane voltage to drop by one quarter of its peak value. (C) Quarterdrop interval (ms) plotted as a function of path distance from the center of that dendrite to the soma (μm) for every dendrite with spines (50) of the reconstructed dSPN. Dendritic spikes were only reliably observed in distal dendrites (>100 μm from cell soma). (D) Onsite phasic GABAergic activation and glutamatergic activation delivered to the same distal dendrite. Synaptic potentials recorded at the dendrite (E) and soma (F). GABA synaptic activation comprised 3 simultaneous stimulations delivered 5 times at an interval of 1 ms to the midpoint of the dendrite. The timing of this phasic input was varied relative to a fixed clustered supra-threshold glutamatergic input (delivered to 18 spines; 1 ms interval as before) in intervals of 10 ms from −10 (blue) to 80 ms (orange). For comparison, the effect of glutamatergic activation alone is illustrated by a thick gray line. Onsite GABAergic activation causes a dramatic cessation of dendritic and somatic potentials in a manner consistent with the relative timing of the 2 inputs. (G) Offsite phasic GABAergic activation delivered to 4 distal dendritic locations. A clustered glutamatergic input (15 spines; 1 ms interval) delivered to the same dendrite as before resulted in a subthreshold synaptic potential (red) at the dendritic site of delivery (H) and soma (I). Similarly, the effect of only activating GABAergic synapses at 1 ms intervals at each of the 4 offsite dendritic locations simultaneously (3 per dendrite; 12 in total) resulted in a moderate postsynaptic potential (blue trace). When delivered sequentially, with GABAergic activation preceding glutamatergic by 10 ms, the previously subthreshold glutamatergic input (red traces) resulted in the generation of a spike (black trace; H and I). The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. SPN, spiny projection neuron. https://doi.org/10.1371/journal.pbio.3002483.g005 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 10 / 26 glutamate uncagingdSPNproximal dendritedistal dendriteA‘quarterdrop’ duration (ms) 75502502575125175225275distance from soma (µm) 50 ms‘quarterdrop’ durationsomaproximal 2P glutamateuncagingBCDGDGsupra-threshold glutamatergic synaptic stimulationsupra-threshold glutamatergic synaptic stimulation-60 mV-85 mV-20 mV-85 mVEsomatic voltagedendritic voltageF50 ms-60 mV-85 mVsomatic voltagephasic GABA synaptic activationphasic GABA synaptic activationsub-threshold glutamatergic synaptic stimulationsub-threshold glutamatergic synaptic stimulation-60 mV-85 mV-20 mV-85 mVHsomatic voltagedendritic voltageI50 msphasic GABA synaptic activationphasic GABA synaptic activationglutamatergic inputGABAergic inputGABAergic inputglutamatergic inputGABAergic input aloneglutamatergic input alonetogetherGABAergic input aloneglutamatergic input alonetogetherdistal 2P glutamateuncagingPLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state shown above, our results are largely consistent with this literature. Of particular interest is the timing dependence of the interaction. To explore this relationship in SPNs, NEURON simula- tions were run with on-site glutamatergic and GABAergic input to the same distal dendrite (Fig 6A). As expected, there was a strong timing dependence on the interaction between syn- aptic events. As experimentally shown by others [6,23], when the glutamatergic EPSP preceded a neighboring GABAergic input, the effect of opening GABAARs on the input impedance (i.e., shunting) was clearly evident (Fig 6B and 6C). However, when the glutamatergic input arrived later, there was synaptic summation, albeit sublinear at both dendritic (Fig 6B) and somatic locations (Fig 6C). To better illustrate the quantitative interaction between the 2 inputs at the dendritic site of stimulation, 2 plots were generated. In one, relative amplitude of glutamater- gic EPSP (P2) with a concomitant GABAergic input was divided by the amplitude of the gluta- matergic EPSP alone (P1) and then plotted as a function of the relative timing of the 2 inputs (Fig 6D). This ratio (P2/P1) fell when the glutamatergic input preceded the GABAergic input and then rose when it trailed the GABAergic input. Similarly, if the ratio of the peak amplitude of the aggregate potential (P3) was divided by the peak amplitude of the isolated glutamatergic EPSP (P1) and plotted as a function of the relative timing of the 2 inputs, the ratio fell when the glutamatergic input preceded the GABAergic input, but then rose above 1 when the glutama- tergic trailed the GABAergic input (Fig 6E). Of greater interest, given the architecture of striatal circuits, was how a diffuse GABAergic input (mimicking conditions produced by ChI activation) would affect the interaction between synaptic events. To explore this interaction, the temporal relationship between a focal glutamater- gic input to a distal dendrite and a GABAergic input to 4 neighboring dendrites was examined (Fig 6F). Not surprisingly, in this situation there was no shunting and the 2 inputs summed at both the dendritic (Fig 6G) and somatic (Fig 6H) locations—regardless of relative timing. In fact, GABAergic input enhanced the ability of glutamatergic synapses to trigger a dendritic spike (Fig 6G and 6H). To probe the dendritic interaction, the relative amplitude of the mixed PSP (mea- sured relative to the underlying GABAergic PSP—P2) was divided by the amplitude of the isolated glutamatergic EPSP (P1) and then plotted as a function of the relative timing of the 2 inputs. At all intervals, the ratio was greater than or equal to 1 (Fig 6I). A qualitatively similar plot was obtained by computing the ratio of the peak amplitude of the mixed PSP (P3) divided by the glutamatergic EPSP amplitude (P1) as a function of relative timing of the 2 inputs (Fig 6J). In these simulations, 3 GABAergic synapses were activated in sequence (1 ms interval) at 4 distal, off-path dendrites (a total of 12 GABAergic synapses activated). Increasing the number of dendrites stimulated to 12, each receiving 1 GABAergic synapse (i.e., to maintain a total of 12 synapses activated) produced qualitatively similar results (S4D–S4H Fig). The widening of the temporal window for supralinear summation presumably resulted from an increase in amplitude of the GABAergic PSP at the site of glutamatergic input. In this simulation, 9 (or more) GABAergic synapses were needed for supralinear summation (S4I and S4H Fig). When the location of GABAergic synapses was shifted to a more proximal dendritic location, shunting decreased (S3A–S3C Fig) and the temporal window for supralinear summation broadened (S3D and S3E Fig). Thus, only GABAergic synapses near the site of glutamatergic input prevented supralinear summation. However, if a dendritic spike was generated, proximal GABAergic input had little effect on its propagation to the soma. SPNs exhibit tonic GABAAR-mediated currents [36]. To assess its effect on dendritic inte- gration, tonic GABAAR current was modeled as a uniformly distributed conductance (Fig 7A). As expected, increasing tonic GABAAR conductance density progressively depolarized the membrane potential, approaching the reversal potential for EGABA (−60 mV) (Fig 7B and 7C). A modest elevation in the tonic GABAAR current led to a dendritic spike in response to a previously subthreshold, clustered excitatory input. These spikes, like those described PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 11 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 6. The interaction between glutamatergic and GABAergic synaptic activity. (A) On-site phasic GABAergic activity was delivered to the same dendrite as glutamatergic synaptic input. Synaptic potentials are illustrated at (B) dendritic site of glutamatergic activity and (C) cell soma. The black trace represents the effect of glutamate-mediated excitation alone (15 neighboring spines activated at 1 ms intervals along the chosen dendrite). The light gray trace illustrates the PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 12 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state effect of phasic GABAergic stimulation (3 simultaneous synaptic stimulations delivered 4 times at an interval of 1 ms to the same dendritic site giving a total of 12 GABA synapses activated). Color traces represent the effect of varying glutamatergic stimulation at temporal intervals relative to the fixed GABAergic input described (blue to orange illustrate 5 traces with Δt = tGLUT−tGABA in the range of −10 to 30 ms, respectively, at 10 ms intervals). Inset illustrates the measurement of P1, P2, and P3. The absolute amplitude of the synaptic potential in the dendrite was measured in the absence of GABAergic activity (P1) or either relative to the amplitude of the underlying depolarizing phasic GABAergic potential at the peak of the postsynaptic response (P2) or relative to the underlying baseline (P3). (D) and (E) show the sublinear effect of varying glutamatergic spine activation relative to a fixed onsite GABA synaptic input on P2 and P3 normalized to P1, respectively. (F) Off-site phasic GABAergic activity was delivered to 4 distal dendrites distinct from the dendrite receiving clustered spine excitation. As before, (G) and (H) show simulated synaptic potentials recorded at dendrite and soma. The black trace represents the same glutamatergic input as above (i.e., 15 spines activated at 1 ms intervals). The light gray trace represents the effect of phasic GABAergic activity delivered to the 4 distal dendrites at the same time (as 3 synaptic simulations at 1 ms intervals per dendrite giving a total of 12 GABA synapses activated). Again, as before, color traces show the effect of altering the timing of clustered spine activation relative to GABA activity. In contrast to on-site activity, Δt = 0, 10, and 20 ms results in the generation of a dendritic spike. Note that the peak of the EPSPGLUT is still clearly visible before the trailing spike manifests. (I) and (J) illustrate the supralinear effect of varying glutamatergic spine activation relative to a fixed off-site GABA synaptic input on P2 and P3 normalized to P1, respectively. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. https://doi.org/10.1371/journal.pbio.3002483.g006 previously (S4A–S4C Fig), were dependent upon NMDARs (Fig 7E). The excitatory effect of tonic GABAergic currents on dendritic excitability was evident over a broad range of conduc- tance values, with shunting becoming significant only at large values (Fig 7D). Thus, the “dose-response” relationship between tonic GABAAR current and “boosting” of the dendritic response to glutamatergic input had an inverted “U” shape. To better illustrate the role of location in dictating the shunting effect of a GABAergic syn- apse, the dendritic voltage and input impedance at the distal dendritic site was computed for a range of GABAergic synapses/dendrite (0–24). Near the site of GABAergic input, the evoked dendritic depolarization progressively increased with the number of GABAergic synapses, but the input impedance fell in parallel (Fig 8A–8C). In contrast, on neighboring dendrites, the depolarization grew with the number of synapses, but there was little local change in input impedance (Fig 8D–8F). The inference to be drawn from these simulations is that the interaction between dendritic glutamatergic and GABAergic synapses depends upon both location and timing. When co- localized, the timing of the 2 inputs dictates their interaction, as previously described [6,23]. However, when the GABAergic input is more diffuse, as predicted to occur with activation of ChIs or NGFIs, the relative timing of GABAergic and glutamatergic inputs becomes less important, and the 2 depolarizing inputs sum. This additivity allows GABAergic and glutama- tergic synapses to work together to trigger dendritic spikes—as observed experimentally. Discussion There are 2 main conclusions that can be drawn from the data presented. First, the GABAAR reversal potential in mature striatal SPNs is near −60 mV. This relatively depolarized reversal - permeability of GABAARs [8,15]. Second, in rest- potential was largely attributable the HCO3 ing SPNs residing near the K+ equilibrium potential (approximately −80 mV), engagement of striatal GABAergic interneurons produced a depolarizing PSP much like that generated by iGluRs—in contrast to the inhibitory effect on SPNs near spike threshold. Both experimental and computational studies of resting SPNs demonstrated that GABAergic synapses could pro- mote dendritic spike generation in response to glutamatergic input in a variety of biologically relevant situations. Taken together, these results argue that SPN GABAAR signaling should be considered as state-dependent and not strictly inhibitory or excitatory [8,15,36,37]. GABAAR signaling in resting SPNs was excitatory Based upon perforated patch recordings, the SPN GABAAR reversal potential remained rela- tively depolarized from the earliest time point examined (post-weaning) to adulthood. Based PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 13 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 7. The interaction between tonic GABAAR signaling and glutamatergic activity. (A) Tonic GABAergic signaling was modeled as a conductance that was evenly distributed throughout the soma and dendritic tree. Glutamatergic synaptic input was delivered to the dendrite illustrated as clustered spine excitation as before. Synaptic potentials at the dendritic site of glutamatergic input (B) and at soma (C) are illustrated. The black trace represents the effect of synaptically released glutamate alone (15 neighboring synapses excited at 1 ms intervals). Color traces (blue to red) illustrate the effect of increasing tonic GABA conductance (from 10−6 to 3 × 10−2 S/cm2) on the glutamate response and resting membrane potential. Inset of (D) illustrates the measurement of P3 relative to P1 with absolute amplitudes measured relative to its underlying resting membrane potential. (D) Increasing tonic GABA conductance density caused an incremental depolarization (ΔV in red) that approached the reversal potential for EGABA (−60 mV) at >10−2 S/cm2. Increasing depolarization (2–3 mV to 10−5 S/cm2 density) was accompanied by a dendritic spike in response to a glutamatergic input that was subthreshold in the absence of tonic GABA activation. This supralinear summation persisted for greater than an order of magnitude increase in tonic GABA activation (3 × 10−4 S/cm2); larger values led to local shunting. (E) Illustrates the requirement for glutamate-mediated synaptic activation of NMDAR for tonic GABA-mediated spike generation. The simulation was identical to that illustrated in (B–D) except that the NMDAR conductance at glutamatergic synapses was zero. In the absence of NMDARs, a combination of onsite shunt and reduced AMPAR driving force arising from depolarization presumably underlies the GABA conductance density-dependent decrease in response. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. NMDAR, N-methyl-D-aspartate receptor. https://doi.org/10.1371/journal.pbio.3002483.g007 upon the analysis of mRNA harvested from iSPNs (using the RiboTag method), it can be con- cluded that mature SPNs robustly expressed KCC2, but not NKCC1, after weaning. The low level of NKCC1 expression in iSPNs was not due to our detection methodology, as at the tissue level, NKCC1 mRNA was readily seen. Moreover, NKCC1-selective antagonist bumetanide had no effect on the GABAAR reversal potential. Although it is possible that SPNs expressed NKCC1 at embryonic or pre-weaning stages of development, NKCC1 was not a determinant of SPN GABAAR function in mice post-weaning. What then is the explanation for the depolarized reversal potential of GABAARs in SPNs? SPNs expressed mRNA for 2 isoforms of CA with cytosolic catalytic sites capable of converting intracellular CO2 to H+ and HCO3 -. Inhibiting CAs with acetazolamide led to a significant PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 14 / 26 -60 mV-85 mV-30 mV-85 mVBsomatic voltagedendritic voltageD50 msglutamatergic inputAP1P3EGABA = -60 mVACP3 / P101.02.00.51.50510152025(cid:2)V (mV)1e-061e-051e-041e-031e-02tonic GABA conductance density (S/cm2)EP3 / P101.02.00.51.50510152025(cid:2)V (mV)1e-061e-051e-041e-031e-02tonic GABA conductance dendity (S/cm2)EGABA = -60 mVAincreasing tonic GABAAR conductanceincreasing tonic GABAAR conductancewithout NMDARs gtonic GABAglutamatergic synaptic stimulationPLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Fig 8. On vs. off-site GABAergic activation and local input impedance. (A) On-site phasic GABAergic activity was delivered and recorded in the same dendrite. (B) Color traces (blue to orange; above) represent simulated GABAergic PSPs and local dendritic impedance (to 10 Hz; below) to increasing synaptic activation (1, 3, 6, 12, and 24 GABA synapses per dendrite). (C) Increasing PSP amplitude (blue) to GABAergic synaptic stimulation reduced local impedance (red). (D) Off-site phasic GABAergic activity was delivered to 4 distal dendrites and recorded in the same dendrite as in (A). (E) and (F) as above. Note the lack of effect of increasing off-site GABAergic activity on local impedance. The dotted lines on panels (C) and (F) illustrate the effect of activating 12 GABAergic synapses in total, which represents the common scenario setting for these 2 conditions throughout the study. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10386854. PSP, postsynaptic potential. https://doi.org/10.1371/journal.pbio.3002483.g008 negative shift in the GABAAR reversal potential in perforated patch recordings. Thus, CA mediated regulation of intracellular pH and HCO3 GABAAR reversal potential [8,38]. - contributed to the depolarized SPN One of the distinctive physiological features of SPNs is that at rest, constitutively active, inwardly rectifying Kir2 K+ channels control the transmembrane potential, pulling SPNs close to the K+ equilibrium potential near −80 mV—the so-called “down-state” of SPNs [10,39]. As these channels are distributed throughout the somatodendritic membrane, they also are a major determinant of local input resistance and dendritic electrotonic structure [3]. With depolarization intracellular Mg2+ ions and polyamines are swept into the channel pore block- ing it [40]. Thus, in dendritic regions of SPNs in the down-state, GABAAR activation will depolarize the membrane and, in so doing, shut off Kir2 channels. As the depolarization pro- duced by transient opening of synaptically activated GABAARs outlasts the change in input impedance, dendritic GABAAR activation should not only depolarize dendrites but increase PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 15 / 26 ABCDFEdendriticsiteGABAergic inputGABAergic inputdendriticsiteGABAergic synaptic activationGABA synapses/dendrite051015202502010PSP amplitude (mV)0200100impedance (MΩ)25 ms-85 mV-60 mVdendritic voltage50 MΩ300 MΩdendritic impedanceGABAergic synaptic activationGABA synapses/dendrite051015202502010PSP amplitude (mV)0200100impedance (MΩ)25 ms-85 mV-60 mVdendritic voltage50 MΩ300 MΩdendritic impedancePLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state their input impedance. Even in the case of tonic GABA signaling, GABAAR opening and Kir2 K+ channel block will counteract one another producing less of a change in dendritic input impedance than would have been predicted otherwise. Although in principle, GABAAR signaling in SPN dendrites should enhance the ability of glutamatergic synapses to promote spike generation, this has never been directly tested with synaptic stimulation. Previous work on this interaction in ex vivo brain slices used intraso- matic current injection to demonstrate the interaction [6]. To fill this gap, optogenetic tools were used in conjunction with two-photon uncaging of glutamate at visualized dendritic spines. Optogenetic stimulation of ChIs was used to engage intrastriatal GABAergic interneu- rons that synapse on SPN dendrites [20]. Indeed, optogenetic stimulation of ChIs evoked a robust gabazine-sensitive PSP in SPNs recorded in perforated patch mode, as did direct opto- genetic stimulation of NPY-positive GABAergic interneurons. In both cases, the GABAergic PSP was considerably delayed and slower than those evoked by SPN collaterals or fast-spiking interneurons [4,22]; most likely, this reflects the large axonal arbor of ChIs and NGFIs and the resulting diffuse release of GABA over the SPN dendritic tree. To probe the interaction between this diffuse GABAergic input and focal activation of glu- tamatergic synapses, two-photon laser scanning uncaging of glutamate along a parfocal stretch of distal dendrite was used in conjunction with optogenetic stimulation of ChIs. Spatiotempo- rally convergent glutamatergic input to distal dendrites of SPNs are capable of triggering local spikes [11–13,24], as described in pyramidal neurons [41,42]. Importantly, when a ChI-evoked GABAergic PSP was followed a few milliseconds later by dendritic uncaging of glutamate, the SPN response to glutamate was enhanced and often reached the threshold for a local dendritic spike. Thus, from the down-state, GABAergic and glutamatergic synapses worked in concert to drive dendritic depolarization of SPNs. Simulations of SPN dendritic integration using a biologically accurate NEURON model [13] reproduced this experimental observation, underscoring the importance of dendritic loca- tion in determining the interaction of GABAergic and glutamatergic synapses. These studies revealed that the timing of GABAergic inputs became less important to their interaction with glutamatergic synapses as the 2 became more electrotonically remote from one another. More- over, these simulations suggested that a spatially diffuse, tonic GABAAR conductance effec- tively boosted the response to glutamatergic synaptic signaling over a broad range of values. Functional implications for intrastriatal circuitry The striatum is composed almost entirely of GABAergic neurons, the exception being ChIs. Although there has been a great deal of speculation about the function of the intrastriatal cir- cuitry, its precise role in goal-directed behavior and habit execution remains obscure. In part, this lack of clarity may stem from thinking about GABAergic signaling as being exclusively inhibitory. Our results, in alignment with several previous reports, argue that this narrow view should be broadened. Consider for a moment the role of the intrastriatal GABAergic circuitry controlled by ChIs through fast nAChRs. ChIs have been implicated in several basal ganglia functions, including the response to salient events, set-shifting, and movement sequencing [43]. On the face of it, the robust and diffuse coupling of ChIs to SPNs through “inhibitory” GABAergic interneurons makes no sense in any of these contexts. However, the recognition that ChI-driven GABAergic input to quiescent iSPNs and dSPNs works in concert with gluta- matergic signals to promote dendritic depolarization and pull SPNs into an “up-state” creates a much more rational framework. Thus, ChIs serve to rapidly bring SPNs “online” and ready to respond to cortical and thalamic signals directing movement. In this context, it is worth not- ing that in vivo, SPNs reside well above the K+ equilibrium potential [44–46]; this “resting” PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 16 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state state appears to be dynamically controlled by synaptic input, which very well could be created by the GABAergic input arising from spontaneously active GABAergic interneurons and those driven by tonically active ChIs [4]. It is also worth considering the potential role of fast-spiking GABAergic interneurons (FSIs). FSIs preferentially target the perisomatic region of SPNs [47] and are widely considered to be part of a fast, feedforward inhibitory circuit linking the striatum with motor cortices [48]. While there is no doubt that FSI input to a spiking SPN is inhibitory, in a quiescent SPN, FSI input should act in precisely the same way as dendritic input and push the membrane potential toward −60 mV. Acting in this way, phasic FSI input to SPNs should facilitate—not inhibit—the response to trailing glutamatergic input from cortical pyramidal neurons. Thus, the timing of signals becomes a critical determinant of whether they should be considered “excitatory” or “inhibitory”; i.e., GABAergic input to SPNs should not be blanketly considered inhibitory. Materials and methods Animals All animal experiments were performed according to the NIH Guide for the Care and Use of Laboratory Animals and approved by the Northwestern University Animal Care and Use Committee (approval numbers: IS00019822, IS00016344, IS00010979, and IS00015064 for ASAP, CHDI, JPB, and NIH/NINDS, respectively). Northwestern University has an Animal Welfare Assurance on file with the Office of Laboratory Animal Welfare (A3283-01). The fol- lowing transgenic male and female mice were used: Adora2a-eGFP (C57BL/6J), RRID: MMRC_010541-UCD; Drd1-tdTomato (FVB), RRID:MMRRC_030512-UNC; Chat-cre (C57BL/6J), RRID:MMRRC_017269-UCD; NPY-cre, (C57BL/6J), RRID: MMRRC_034810-UCD; and Adora2a-cre (C57BL/6J), RRID:MMRRC_034744-UCD. Chat- cre and NPY-cre mice were backcrossed to Adora2a-eGFP and DRD1-tdTomato reporter lines in house. Mice were group-housed with food and water ad libitum on a 12-h light/dark cycle with temperatures of 65˚ to 75˚F and 40% to 60% humidity. Stereotaxic surgery An isoflurane precision vaporizer (Smiths Medical PM) was used to anesthetize mice. Mice were then placed on a stereotaxic frame (David Kopf Instruments), with a Cunningham adap- tor (Harvard Apparatus) to maintain anesthesia delivery during surgery. The skull was exposed, and a small hole was drilled at the desired injection site. The following stereotaxic coordinates were used: Striatum, AP = +0.74, ML = −1.85, DV = −3.50. The Allen Mouse Brain Atlas, online version 1, 2008 (RRID:SCR_002978; http://mouse.brain-map.org/static/ atlas) was used as a reference for the coordinates and generating diagrams. For each mouse, the distance between bregma and lambda was calculated and used to adjust the coordinates. For AAV injections, approximately 500 nl of viral vector was delivered using a glass micropi- pette (Drummond Scientific) pulled with a P-97 glass puller (Sutter Instruments). Surgeries for electrophysiology experiments utilizing Chronos were performed unilaterally while surger- ies for RiboTag tissue collection were executed bilaterally. Electrophysiology experiments using Chronos were performed after at least 21 postoperative days; tissue collection for Ribo- Tag was performed 10 days after injection (dx.doi.org/10.17504/protocols.io.81wgby191vpk/ v1). PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 17 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state RiboTag profiling AAVs for expression of RiboTag under a cre-dependent promoter (AAV5-hsyn-DIO-Rpl22l1- 3Flag-2A-eGFP-WPRE RRID:Addgene_214265, titers 2.24 × 1013 viral genomes/ml) were ste- reotaxically injected into the striatum in Adora2a-cre mice at P18 or p170, as described above. Ten days after injection, mice were killed and the striatal tissue expressing RiboTag was dis- sected out using fluorescence microscopy and then frozen at −80˚C. RiboTag immunoprecipi- tation was carried out as previously described [49]. Briefly, tissue was homogenized in cold homogenization buffer [50 mM tris (pH 7.4), 100 mM KCl, 10 mM MgCl2, 1 mM dithiothrei- tol, cycloheximide (100 μg/ml), protease inhibitors, recombinant ribonuclease (RNase) inhibi- tors, and 1% NP-40]. Homogenates were centrifuged at 10,000g for 10 min, and the supernatant was collected and precleared with protein G magnetic beads (Thermo Fisher Sci- entific) for 1 h at 4˚C, under constant rotation. Immunoprecipitations were carried out with anti-Flag magnetic beads (Sigma-Aldrich) at 4˚C overnight with constant rotation, followed by 4 washes in high-salt buffer [50 mM tris (pH = 7.4), 350 mM KCl, 10 mM MgCl2, 1% NP- 40, 1 mM dithiothreitol, and cycloheximide (100 μg/ml)]. RNA was extracted using RNeasy Micro RNA extraction kit (QIAGEN) according to the manufacturer’s instructions (dx.doi. org/10.17504/protocols.io.261gedwyyv47/v1). Quantitative real-time PCR RNA was extracted from the dissected striatal tissue using RNeasy mini kit (QIAGEN). cDNA was synthetized by using the SuperScript IV VILO Master Mix (Applied Biosystems) and pre- amplified for 10 cycles using TaqMan PreAmp Master Mix and pool of TaqMan Gene Expres- sion Assays (Applied Biosystems). The resulting product was diluted and then used for PCR with the corresponding TaqMan Gene Expression Assay and TaqMan Fast Advanced Master Mix. Data were normalized to Hprt by the comparative CT (2-DDCT) method. TaqMan probes were used for PCR amplification of Hprt, Mm03024075_m1, Slc12a2 (NKCC1), Mm01265955_m1, Slc12a5 (KCC2), Mm00803929_m1, Slc4a3 (AE3) Mm00436654_g1, and Slc4a10 (NCBE) Mm00473827_m1. Experimental Ct values were normalized to hprt values using the following formula: ΔCt = Ct (gene of interest) − Ct (hprt). The final expression levels were shown as ΔCt values (dx.doi.org/10.17504/protocols.io.e6nvwd4o2lmk/v1). Ex vivo slice preparation Coronal or parasagittal slices (275 μm thickness) were obtained from mice ranging in age from 4 weeks to 9 months. Mice were acutely anesthetized with a mixture of ketamine (50 mg/kg) and xylazine (4.5 mg/kg) and perfused transcardially with oxygenated ice-cold saline (4˚C) containing in mM: 125 NaCl, 3 KCl, 2.5 MgCl2, 0.5 CaCl2, 25 NaHCO3, 1.25 NaH2PO4, and 10 glucose (satu- rated with 95% O2-5% CO2; pH 7.4; 300 mOsm/l). After perfusion, mice were decapitated, and the brains were rapidly removed. Slices were obtained in oxygenated ice-cold saline using a vibra- tome (VT1000S, Leica Microsystems). Slices were transferred to an ACSF-filled holding chamber containing in mM: 125 NaCl, 3 KCl, 1 MgCl2, 2 CaCl2, 25 NaHCO3, 1.25 NaH2PO4 and 10 glu- cose (saturated with 95% O2-5% CO2; pH 7.4; 300 mOsm/l) and held there for approximately 30 min at 34˚ before being allowed to come to room temperature (21 to 25˚C) where they remained until recording (dx.doi.org/10.17504/protocols.io.dm6gp328jvzp/v2). Electrophysiological recordings For electrophysiological recordings slices were transferred to a submersion-style recording chamber mounted on an Olympus BX51 upright microscope (60×/1.0 NA objective) equipped PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 18 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state with infrared differential interference contrast. Whole-cell and perforated patch clamp electrophysiological recordings were performed with Multiclamp 700B amplifier. Signals were filtered at 1 KHz. Stimulation and display of electrophysiological recordings were obtained with custom-written freeware WinFluor (John Dempster, Strathclyde University, Glasgow, United Kingdom; http://spider.science.strath.ac.uk/sipbs/software_winfluor.htm) that syn- chronizes two-photon imaging and electrophysiology. Targeted electrophysiological record- ings were obtained from visually identified iSPNs or dSPNs. Patch pipettes (3 to 5 MO) were prepared with a Sutter Instruments horizontal puller using borosilicate glass with filament and filled with (in mM): 120 potassium-D-gluconate, 13 KCl, 10 HEPES, 0.05 EGTA, 4 ATP-Mg, 0.5 GTP-Na, 10 phosphocreatine-di (tris); pH was adjusted to 7.25 with KOH and osmolarity to 275 to 280 mOsm (dx.doi.org/10.17504/protocols.io.rm7vzx1w2gx1/v2). In perforated- patch experiments, 10 μM gramicidin was added to the internal recording solution to induce chloride-impermeable pore formation along with 25 μM Alexa Fluor 568 hydrazide Na+ salt (Invitrogen) to visualize any potential pore rupture. All perforated-patch recordings were cor- rected for liquid junction potential. Electrophysiological characterization of neurons was made in current clamp configuration. The amplifier bridge circuit was adjusted to compensate for electrode resistance. Access resistances were continuously monitored, and experiments were discarded if changes >20% were observed (dx.doi.org/10.17504/protocols.io.36wgq3y9olk5/ v1). Digitized data were imported for analysis with commercial software (IGOR Pro 6.0, WaveMetrics, Oregon RRID:SCR_000325). Optogenetic stimulation Simultaneous electrophysiological and Chronos optogenetic photo-stimulation or RuBi-GABA uncaging were performed with a targeted focal spot blue laser (473 nm Aurora laser launch, Prai- rie Technologies) system using the Photostimulus Editor in WinFluor. The Point Photo-activation module (Prairie Technologies) allows 2 different stimulation areas, and intensities, with sub-μm (small spot) critical illumination or an additional lens to stimulate approximately 8 μm (large spot) diameter photo-stimulation in the sample focal plane with the 60×/1.0 objective. To control the release of synaptic GABA, Chat-cre and NPY-cre mice were injected with Chronos (AAV-/ hsyn-flex-chronos-GFP or AAV-/hsyn-flex-chronos-tdTomato; UNC GTC Vector Core) as described in the Stereotaxic Surgery section of these methods. To activate NPY or Chat Chronos containing axons in the striatum, the targeted 473 nm spots were positioned adjacent to individual dendritic spines to photo-stimulate presynaptic terminals impinging on iSPNs or dSPNs. The laser power was calibrated to evoke a somatic postsynaptic potential of 2 to 5 mV. Although the laser was aimed peri-dendritically, the blue excitation laser light will travel in a focusing hourglass (small spot) or column (larger spot) through the slice with likely activation of the large dendritic fields of the striatal interneurons, above and possibly below the sample focal plane, resulting in dif- fuse synaptic GABAAR activation of postsynaptic receptors. For simultaneous stimulation of 5 to 10 spines and the reversal potential experiments, the larger blue laser spot was used. Additionally, whole-field photo-stimulation through the 60× objective (26.5FN with approximately 440 μm diameter exposure) was coordinated with an epi-fluorescence-based LED (475/30 nm, pE-100, CoolLED) reflected through an eGFP filter cube and controlled with the Stimulus Editor in Win- Fluor. The time synchronized results were displayed in the WinFluor main Record Images and Sig- nals window (dx.doi.org/10.17504/protocols.io.rm7vzx2nrgx1/v2). Two-photon excitation uncaging of DNI-Glutamate Simultaneous two-photon laser uncaging (720 nm) and optogenetic stimulation of synaptic GABA (473 nm) were performed using a laser scanning microscope system (Ultima, Bruker PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 19 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Technologies; formerly Prairie Technologies) with a tunable imaging laser (Chameleon- Ultra1, Coherent Laser Group, Santa Clara, California, United States of America) and Olym- pus BX-51WI upright microscope with 60×/1.0NA water-dipping objective lens was used to locate and acquire a whole-cell patch clamp; 810 nm from the imaging 2P laser excited Alexa Fluor 568 (580 to 630 nm; R3896 PMT, Hamamatsu) to visualize dendrites of the patched soma and distal (>100 μm) dendritic spines from planar sections (approximately 20 μm) of the same dendrite with image zoom 4 and 50 μm FOV. Custom written software (WinFluor, John Dempster and its PhotoStimulusEditor module, Nicholas Schwarz; features now available in PrairieView 5.x) was used to direct, control, test, synchronize, and display electrophysiologi- cal recordings combined with laser imaging and photo-stimulation. Simultaneous 2PLU (720 nm, Coherent Chameleon) and single-photon optogenetic stimulation of synaptic GABA (473 nm, Prairie Aurora Laser Launch) were provided by a second, separate, independently con- trolled galvanometer mirror pair in the Ultima system. The 3 laser beams were optically com- bined (760DCLPXR, Chroma Technologies) in the scan head and aligned to the microscope optical path. DNI-glutamate (5 mM, Femtonics, Budapest, Hungary) was perfused in the recorded area and then excited by the 720 nm 2P laser. Pulses of 1 ms duration (approximately 10 mW sample power) were delivered to single spines located in the same focal plane where the laser average power or spot location was calibrated to evoke a somatic excitatory PSP of 1 to 2 mV for each spine. During synchronized acquisitions, the blue laser GABA photo-stimula- tion (1 pulse, 3-ms duration) preceded the glutamate uncaging of approximately 15 spines with 1-ms duration and 1-ms inter-stimulation interval. These experiments were all conducted in the appropriate cocktail of synaptic blockers: CGP-55845 (1 μM), MPEP (1 μM), and CPCCOEt (50 μM) (dx.doi.org/10.17504/protocols.io.rm7vzx1w2gx1/v2). Pharmacological reagents Stock solutions were prepared before experiments and added to the perfusion solution or focally applied with pressure ejection in the final concentration indicated. Two-photon laser uncaging and optogenetic experiments were performed in the presence of AP5 (50 μM), NBQX (5 μM), CGP 55845 (1 μM), MPEP (1 μM), and CPCCOEt (50 μM). Bumetanide (10 μM) and acetazolamide (10 μM) were used to probe for roles of NKCC1 and carbonic anhydrase, respectively. All drugs were obtained from Hello Bio, Sigma-Aldrich or Tocris. Confocal imaging Fixed tissue was prepared by transcardially perfusing terminally anesthetized mice with phos- phate-buffered saline (PBS; Sigma-Aldrich) immediately followed by 4% paraformaldehyde (PFA; diluted in PBS from a 16% stock solution; Electron Microscopy Sciences). The brain was then removed and transferred into PFA solution overnight before being thoroughly rinsed and stored in PBS at 4˚C. Fixed brains were then sectioned into 50-μm thick coronal slices on a Leica VT1200S vibratome and collected in PBS. The sections were positioned on microscopy slides (VWR), allowed to dry and mounted with ProLong Diamond (Thermo Fisher Scientific) and #1.5 glass coverslips (VWR). Mounted sections were stored at 4˚C until imaged with an Olympus FV10i-DUC confocal laser scanning microscope, using 10×/0.4 (air) or 60×/1.35 (oil) objective. FIJI (NIH, RRID:SCR_002285) was used to adjust images for brightness, con- trast, and pseudo-coloring (dx.doi.org/10.17504/protocols.io.kxygx3nrkg8j/v1). Modeling The NEURON (Neuron 8.2; RRID:SCR_005393) [31] + Python (Python Programming Lan- guage RRID:SCR_008394) model of a morphologically reconstructed SPN was integrated into PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 20 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state a previously established model [13,29,30]. Cytoplasmic resistivity (Ra) was set to 200 O cm and specific capacitance was 1 μF cm-2. The original compartmentalized model (https://senselab.med. yale.edu/ModelDB/ShowModel?model=266775&file=/lib/params_dMSN.json#tabs-2) is biophy- sically detailed and comprised a total of 700 segments with the following active and passive con- ductances: transient fast inactivating Na+ (Naf), persistent Na+ (Nap), fast A-type K+ (Kaf), slowly inactivating K+ (Kas), inwardly rectifying K+ (Kir), delayed rectifier K+ (Kdr), small con- ductance Ca2+-activated K+ (SK), large conductance Ca2+-activated K+ (BK), L-type Ca2+ (Cav 1.2 and 1.3), N-type Ca2+ (Cav 2.2), R-type Ca2+ (Cav 2.3) T-type Ca2+ (Cav 3.2 and 3.3). Channel distributions over cellular compartments were as previously described and are presented in Table A and Table B in S1 Text (based on Table 2 from Lindroos and colleagues). Synaptic spines were added to all dendritic locations further than 30 μm from the cell soma. Spines were added at a density of 1.711 per μm to give a total of circa 5,500 spines for the reconstructed dSPN. The spines comprised a cylindrical head with a diameter of 0.5 μm connected to dendrites via a neck 1-μm long with diameter of 0.1 μm. The morphologically reconstructed model dSPN had a rest- ing membrane potential of −84 mV; a modest hyperpolarizing current step of 200 pA gave a “rec- tified range” input resistance of approximately 85 MO and membrane time constant of 10.5 ms. The dSPN had an estimated whole-cell capacitance of 180 pF. Candidate spines were selected as separate nearest neighbors along a dendrite at a start point of approximately two-thirds the length. NMDA and AMPA conductances were inserted into spines to be activated. Synaptic cur- rents were modeled using a two-state kinetic model where the normalized peak conductance is determined by rise and decay time constants T1 and T2 (T2 > T1) (as per Du and colleagues, Lin- droos and colleagues, Lindroos and Kotaleski). The maximal conductances of AMPA and NMDA responses were 350 and 752.5 pS, respectively. The reversal potentials for AMPA and NMDA was 0 mV. Spines were activated at 1 ms intervals in succession with stimulation moving away from the soma as per electrophysiological activation. The threshold for generating an up- state in the absence of any GABAergic activation was 15 glutamatergic inputs. GABA synapses were inserted directly onto the same dendritic location (the midpoint of the chosen dendrite). The maximal conductance was 1,000 pS and reversal potential set to −60 mV. GABAergic responses were generated by activating these synapses simultaneously in up to groups of 3 at an interval of 1 ms. For most simulations, onsite activation comprised a total of 12 activated GABA synapses. For offsite activation, 4 distal locations were selected and activated simultaneously mul- tiple times at an interval of 1 ms. Again, for most simulations each dendrite received a burst of 3 activated GABA synapses at an interval of 1 ms making a total of 12 activated GABA synapses across 4 dendritic locations. All code for the simulations is publicly available (https://github.com/ vernonclarke/SPNfinal/tree/v1.0; dx.doi.org/10.5281/zenodo.10162265). Statistical analysis Data was graphically presented using nonparametric box and whisker plots. In these plots, the center line is the median, the edges of the box mark the interquartiles of the distribution and the lines extend to the limiting values of the sample distribution; outliers (defined as values fur- ther away from the median than 1.5 × interquartile range) are marked as asterisks. Statistical significance was determined using nonparametric tests (Wilcoxon signed rank test using either the exact method or with continuity correction and Mann–Whitney U test, as appropriate) using R Statistical Software (v4.2.3; R Core Team 2023, RRID:SCR_001905) [50]. R was used to perform linear regression analysis of data used to estimate reversal potentials. The regres- sion analysis was performed with mean values for data at each voltage. Means were rounded to the nearest whole number for the current traces (Fig 2B and 2C) and to the nearest 0.5 mV for the ΔV/V traces (Fig 2E and 2F) prior to running the regression. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 21 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Supporting information S1 Fig. The summation of GABAergic and glutamatergic PSPs was typically linear or supralinear. (A) Diagram showing how the amplitudes of the individual ChI-evoked GABAergic and 2P uncaging-evoked glutamatergic PSPs were measured, along with the amplitude of the response to combined stimulation (see Fig 4). (B) Scatter plot of response amplitude to the combined stimulation against the arithmetic sum of the individual GABAer- gic and glutamatergic PSPs. Most of the data points derived from individual SPNs fell on the diagonal or above, demonstrating linearity or supralinear behavior. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10387118. (EPS) S2 Fig. Reducing cytoplasmic resistivity (Ra) has no effect on the fundamental outcomes of computational modeling of dSPN dendritic activity. The simulations in this figure are iden- tical to those of Fig 5, except cytoplasmic resistivity (Ra) is reduced from 200 to 100 O cm and the underlying conductances of synaptically activated glutamatergic AMPA and NMDARs are increased by 60% (from 350 and 752.5 to 560 and 1,204 pS, respectively). The latter ensures that a glutamatergic synaptic event at a given spine produces a similar depolarization at its equivalent dendritic location (i.e., as measured in the dendritic tree at the location of its spine neck). As a result, the number of clustered inputs that was just subthreshold for upstate gener- ation (15) was preserved. Any qualitative differences (small increases in “quarterdrop dura- tion” (B and C) and somatic amplitude (B, C, F, and I) can be attributed to the effect of reduced dendritic filtering resulting from reduced Ra. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10387118. (EPS) S3 Fig. Temporal profile of interaction between glutamatergic and GABAergic synaptic activity. (A) On-path phasic GABAergic activity is delivered to a dendrite positioned more proximally on the same path to the soma as the dendrite receiving glutamatergic synaptic input. As before, synaptic potentials are illustrated at (B) dendritic site of glutamatergic activity and (C) cell soma. L-glutamate mediated excitation alone (15 neighboring spines activated at 1 ms intervals along the chosen dendrite; black trace), phasic GABAergic stimulation alone (3 simultaneous synaptic stimulations delivered 4 times at an interval of 1 ms to the proximal on- path dendritic site giving a total of 12 GABA synapses activated; light gray) and the effect of varying glutamatergic stimulation at temporal intervals relative to this fixed GABAergic input described (blue to orange illustrate 5 traces with Δt = tGLUT−tGABA in the range of −10 to 30 ms, respectively, at 10 ms intervals) are illustrated. The absolute amplitude of the synaptic potential in the dendrite was measured in the absence of GABAergic activity (P1) or either rel- ative to the amplitude of the underlying depolarizing phasic GABAergic potential response (P2) or relative to the underlying baseline (P3). (D) and (E) show the supralinear effect of vary- ing glutamatergic spine activation relative to a fixed distal off-path GABA synaptic input on P2 and P3 normalized to P1, respectively. For comparison, data from the off-path distally located phasic GABAergic activity is overlayed (D and E). The widened temporal window for supra- linear summation in this simulation arises from the reduced path length from the site of GABA activation to the site of glutamatergic clustered activity which manifests as an increase in magnitude of GABAergic PSP at this location (e.g., compare GABAergic PSPs; light gray trace in panel B to its equivalent in Fig 6G). The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo.10387118. (EPS) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 22 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state S4 Fig. NMDAR dependence and GABAergic threshold for dendritic spike generation. (A) Offsite phasic GABAergic activation delivered to 4 distal dendritic locations with the same clustered glutamatergic input (15 spines; 1 ms interval) delivered to the dendrite as in Fig 5G but with NMDAR conductance set to zero. This resulted in a synaptic potential (red) at the dendritic site of delivery (B) and soma (C). Similarly, the effect of only activating GABAergic synapses simultaneously at each of the 4 offsite dendritic locations (3 per dendrite at 1 ms intervals; 12 in total) resulted in a moderate postsynaptic potential (blue trace). When deliv- ered sequentially, with GABAergic activation preceding glutamatergic by 10 ms, the previously subthreshold glutamatergic input did not result in the generation of an up-state (black trace). Thus, supralinear summation to phasic GABAergic input is dependent on the synaptic activa- tion of NMDARs. (D) Offsite phasic GABAergic activation delivered to the midpoint of 12 most distal dendritic locations. As before, synaptic potentials are illustrated at (E) dendritic site of glutamatergic activity and (F) cell soma. As before, L-glutamate mediated excitation alone (15 neighboring spines activated at 1 ms intervals along the chosen dendrite; black trace), phasic GABAergic stimulation alone (single synaptic stimulations delivered simulta- neously to the 12 most distal dendrites giving a total of 12 GABA synapses activated; light gray) and the effect of varying glutamatergic stimulation at temporal intervals relative to this fixed GABAergic input described (blue to orange illustrate 5 traces with Δt = tGLUT−tGABA in the range of −10 to 30 ms, respectively, at 10 ms intervals) are illustrated. (G) and (H) show the supralinear effect of varying glutamatergic spine activation relative to a fixed off-path distal GABA synaptic inputs on P2 and P3 normalized to P1, respectively. For comparison, data from the off-path distally located phasic GABAergic activity with 4 sites each receiving 3 inputs at 1 ms intervals delivered simultaneous across the 4 dendritic locations is overlayed (taken from Fig 6I and 6J). In this example, the widened temporal window for supralinear summation in this simulation most likely arises from the temporal differences in activation (12 delivered simultaneously at 12 sites vs. 12 delivered as a burst of 3 at 1 ms intervals at 4 sites simulta- neously; compare GABAergic PSPs: light gray trace in panel E to its equivalent in Fig 6G). This manifests as an increased GABAergic potential at the glutamatergic site of clustered exci- tation. This simulation confirms that the observed effect of phasic offsite GABAergic activity can be extended to simultaneous activation at many dendritic locations and provides a simpli- fied model to examine the threshold activity required (as it allows synapse number to be varied in a unitary fashion). Synaptic potentials are illustrated at (I) dendritic site of glutamatergic activity and (J) cell soma for GABAergic activity delivered simultaneously to the midpoint of either 8 and 9 (offset for clarity) most distal dendritic locations. The threshold number of GABAergic synapses necessary for dendritic up-state generation in this model was 9. The data underlying the graphs shown in the figure can be found in dx.doi.org/10.5281/zenodo. 10387118. (EPS) S1 Text. Model parameters and ion channel distributions. Table A provides the maximum conductance for all cation channels in the model (S/cm2). Table B gives the maximum perme- ability of Ca2+ channels in the model (cm/s). The equations are identical to that provided in the original modeling code (Du and colleagues, Lindroos and colleagues, Lindroos and Kotal- seki) and retain the original variable names for generating the following distributions: sigmoi- dal: (a4+a5/(1+exp((x-a6)/a7))) gmax; exponential: (a4+a5exp((x-a6)/a7)) *gmax; uniform: gmax. The initial dendrite measurement is taken at x = 6.010 μm (i.e., the radius of the soma; x is measured from center of cell soma); final dendrite values is taken at x = 265.268 μm which is the path length of the furthest dendritic point from the mid-point of the soma. (DOCX) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 23 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state Acknowledgments We wish to thank Sasha Ulrich, Yu Chen, Enrico Zampese, and Abdelhak Belmadani for expert technical assistance; George Augustine for the Clomeleon plasmid and Jun Ding for help in getting the NEURON modeling work started. Author Contributions Conceptualization: Michelle Day, Alexandria Melendez, David Wokosin, Tatiana Tkatch, Vernon R. J. Clarke, D. James Surmeier. Data curation: Michelle Day, Marziyeh Belal, William C. Surmeier, Alexandria Melendez, David Wokosin, Tatiana Tkatch, Vernon R. J. Clarke. Formal analysis: Michelle Day, Marziyeh Belal, Alexandria Melendez, David Wokosin, Tati- ana Tkatch, Vernon R. J. Clarke. Funding acquisition: Michelle Day, D. James Surmeier. Investigation: Michelle Day, David Wokosin, Tatiana Tkatch. Methodology: Michelle Day, Tatiana Tkatch, Vernon R. J. Clarke, D. James Surmeier. Project administration: D. James Surmeier. Resources: D. James Surmeier. Software: William C. Surmeier, David Wokosin, Vernon R. J. Clarke. Supervision: Michelle Day, D. James Surmeier. Validation: Michelle Day, Marziyeh Belal, David Wokosin, Vernon R. J. Clarke. Writing – original draft: Michelle Day, David Wokosin, Vernon R. J. Clarke, D. James Surmeier. Writing – review & editing: Michelle Day, Marziyeh Belal, William C. Surmeier, Alexandria Melendez, David Wokosin, Tatiana Tkatch, Vernon R. J. Clarke, D. James Surmeier. References 1. Balleine BW, Delgado MR, Hikosaka O. The role of the dorsal striatum in reward and decision-making. J Neurosci. 2007; 27(31):8161 5. https://doi.org/10.1523/JNEUROSCI.1554-07.2007 PMID: 17670959 2. Klaus A, da Silva JA, Costa RM. What, If, and When to Move: Basal Ganglia Circuits and Self-Paced Action Initiation. Annu Rev Neurosci. 2016; 42(1):1–25. https://doi.org/10.1146/annurev-neuro-072116- 031033 PMID: 31018098 3. Wilson CJ. GABAergic inhibition in the neostriatum. Progress Brain Res. 2007; 160(91):110. https://doi. org/10.1016/S0079-6123(06)60006-X PMID: 17499110 4. Tepper JM, Koo´ s T, Ibanez-Sandoval O, Tecuapetla F, Faust TW, Assous M. Heterogeneity and Diver- sity of Striatal GABAergic Interneurons: Update 2018. Front Neuroanat. 2018; 12:91. https://doi.org/10. 3389/fnana.2018.00091 PMID: 30467465 5. Plenz D. When inhibition goes incognito: feedback interaction between spiny projection neurons in stria- tal function. Trends Neurosci. 2003; 26(8):436–443. https://doi.org/10.1016/S0166-2236(03)00196-6 PMID: 12900175 6. Bracci E, Panzeri S. Excitatory GABAergic effects in striatal projection neurons. J Neurophysiol. 2006; 95:90. https://doi.org/10.1152/jn.00598.2005 PMID: 16251264 7. Paille V, Fino E, Du K, Morera-Herreras T, Perez S, Kotaleski JH, et al. GABAergic circuits control spike-timing-dependent plasticity. J Neurosci. 2013; 33(22):9353–9363. https://doi.org/10.1523/ JNEUROSCI.5796-12.2013 PMID: 23719804 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 24 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state 8. Kaila K, Price TJ, Payne JA, Puskarjov M, Voipio J. Cation-chloride cotransporters in neuronal develop- ment, plasticity and disease. Nat Rev Neurosci. 2014; 15(10):637–654. https://doi.org/10.1038/nrn3819 PMID: 25234263 9. Tepper JM, Sharpe NA, Koo´s TZ, Trent F. Postnatal Development of the Rat Neostriatum: Electrophysi- ological. Light- and Electron-Microscopic Studies. Dev Neurosci. 1998; 20(2–3):125–145. https://doi. org/10.1159/000017308 PMID: 9691188 10. Stern EA, Jaeger D, Wilson CJ. Membrane potential synchrony of simultaneously recorded striatal spiny neurons in vivo. Nature. 1998; 394(6692):475:8. https://doi.org/10.1038/28848 PMID: 9697769 11. Plotkin JL, Day M, Surmeier DJ. Synaptically driven state transitions in distal dendrites of striatal spiny neurons. Nat Neurosci. 2011; 14(7):881–8. https://doi.org/10.1038/nn.2848 PMID: 21666674 12. Prager EM, Dorman DB, Hobel ZB, Malgady JM, Blackwell KT, Plotkin JL. Dopamine Oppositely Modu- lates State Transitions in Striosome and Matrix Direct Pathway Striatal Spiny Neurons. Neuron. 2020. https://doi.org/10.1016/j.neuron.2020.09.028 PMID: 33080228 13. Du K, Wu Y-W, Lindroos R, Liu Y, Ro´zsa B, Katona G, et al. Cell-type–specific inhibition of the dendritic plateau potential in striatal spiny projection neurons. Proc Natl Acad Sci U S A. 2017; 114(36):E7612 E21. https://doi.org/10.1073/pnas.1704893114 PMID: 28827326 14. Macdonald RL, Olsen RW. GABA A Receptor Channels. Annu Rev Neurosci. 1994; 17(1):569–602. https://doi.org/10.1146/annurev.ne.17.030194.003033 PMID: 7516126 15. Farrant M, Kaila K. The cellular, molecular and ionic basis of GABAA receptor signalling. Progress Brain Res. 2007; 160:59–87. https://doi.org/10.1016/S0079-6123(06)60005-8 PMID: 17499109 16. Ben-Ari Y, Gaiarsa J-L, Tyzio R, Khazipov R. GABA: A Pioneer Transmitter That Excites Immature Neu- rons and Generates Primitive Oscillations. Physiol Rev. 2007; 87(4):1215–1284. https://doi.org/10. 1152/physrev.00017.2006 PMID: 17928584 17. Sanz E, Yang L, Su T, Morris DR, McKnight GS, Amieux PS. Cell-type-specific isolation of ribosome- associated mRNA from complex tissues. Proc Natl Acad Sci U S A. 2009; 106(33):13939 44. https://doi. org/10.1073/pnas.0907143106 PMID: 19666516 18. Grimley JS, Li L, Wang W, Wen L, Beese LS, Hellinga HW, et al. Visualization of synaptic inhibition with an optogenetic sensor developed by cell-free protein engineering automation. J Neurosci. 2013; 33 (41):16297–309. https://doi.org/10.1523/JNEUROSCI.4616-11.2013 PMID: 24107961 19. Ruusuvuori E, Huebner AK, Kirilkin I, Yukin AY, Blaesse P, Helmy M, et al. Neuronal carbonic anhy- drase VII provides GABAergic excitatory drive to exacerbate febrile seizures. EMBO J. 2013; 32 (16):2275–2286. https://doi.org/10.1038/emboj.2013.160 PMID: 23881097 20. Faust TW, Assous M, Tepper JM, Koo´ s T. Neostriatal GABAergic Interneurons Mediate Cholinergic Inhibition of Spiny Projection Neurons. J Neurosci. 2016; 36(36):9505–11. https://doi.org/10.1523/ JNEUROSCI.0466-16.2016 PMID: 27605623 21. Kocaturk S, Guven EB, Shah F, Tepper JM, Assous M. Cholinergic control of striatal GABAergic micro- circuits. Cell Rep. 2022; 41(4):111531. https://doi.org/10.1016/j.celrep.2022.111531 PMID: 36288709 22. Tepper JM, Koo´ s T, Wilson CJ. GABAergic microcircuits in the neostriatum. Trends Neurosci. 2004; 27 (11):662–9. https://doi.org/10.1016/j.tins.2004.08.007 PMID: 15474166 23. Gulledge AT, Stuart GJ. Excitatory Actions of GABA in the Cortex. Neuron. 2003; 37(2):299–309. https://doi.org/10.1016/s0896-6273(02)01146-7 PMID: 12546824 24. Carrillo-Reid L, Day M, Xie Z, Melendez AE, Kondapalli J, Plotkin JL, et al. Mutant huntingtin enhances activation of dendritic Kv4 K+ channels in striatal spiny projection neurons. Elife. 2019; 8:e40818. https://doi.org/10.7554/eLife.40818 PMID: 31017573 25. Day M, Wokosin D, Plotkin JL, Tian X, Surmeier DJ. Differential Excitability and Modulation of Striatal Medium Spiny Neuron Dendrites. J Neurosci. 2008; 28(45):11603–14. https://doi.org/10.1523/ JNEUROSCI.1840-08.2008 PMID: 18987196 26. Fieblinger T, Graves SM, Sebel LE, Alcacer C, Plotkin JL, Gertler TS, et al. Cell type-specific plasticity of striatal projection neurons in parkinsonism and L-DOPA-induced dyskinesia. Nat Commun. 2014; 5 (1):5316. https://doi.org/10.1038/ncomms6316 PMID: 25360704 27. Carter AG, Soler-Llavina GJ, Sabatini BL. Timing and location of synaptic inputs determine modes of subthreshold integration in striatal medium spiny neurons. J Neurosci. 2007; 27(33):8967–77. https:// doi.org/10.1523/JNEUROSCI.2798-07.2007 PMID: 17699678 28. Wilson CJ, Chang HT, Kitai ST. Firing patterns and synaptic potentials of identified giant aspiny inter- neurons in the rat neostriatum. J Neurosci. 1990; 10(2):508–519. https://doi.org/10.1523/JNEUROSCI. 10-02-00508.1990 PMID: 2303856 29. Lindroos R, Kotaleski JH. Predicting complex spikes in striatal projection neurons of the direct pathway following neuromodulation by acetylcholine and dopamine. Eur J Neurosci. 2020. https://doi.org/10. 1111/ejn.14891 PMID: 32609903 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 25 / 26 PLOS BIOLOGYGABAergic regulation of striatal spiny projection neurons depends upon their activity state 30. Lindroos R, Dorst MC, Du K, Filipović M, Keller D, Ketzef M, et al. Basal Ganglia Neuromodulation Over Multiple Temporal and Structural Scales-Simulations of Direct Pathway MSNs Investigate the Fast Onset of Dopaminergic Effects and Predict the Role of Kv4.2. Front Neural Circuits. 2018; 12:3. https:// doi.org/10.3389/fncir.2018.00003 PMID: 29467627 31. Hines ML, Carnevale NT. The NEURON Simulation Environment. Neural Comput. 1997; 9(6):1179– 1209. https://doi.org/10.1162/neco.1997.9.6.1179 PMID: 9248061 32. Roth A, Ha¨ usser M. Compartmental models of rat cerebellar Purkinje cells based on simultaneous somatic and dendritic patch-clamp recordings. J Physiol. 2001; 535(2):445–472. https://doi.org/10. 1111/j.1469-7793.2001.00445.x PMID: 11533136 33. Gidon A, Segev I. Principles Governing the Operation of Synaptic Inhibition in Dendrites. Neuron. 2012; 75(2):330–41. https://doi.org/10.1016/j.neuron.2012.05.015 PMID: 22841317 34. Jadi M, Polsky A, Schiller J, Mel BW. Location-Dependent Effects of Inhibition on Local Spiking in Pyra- midal Neuron Dendrites. PLoS Comput Biol. 2012; 8(6):e1002550. https://doi.org/10.1371/journal.pcbi. 1002550 PMID: 22719240 35. Hao J, Wang X-d, Dan Y, Poo M-m, Zhang X-h.An arithmetic rule for spatial summation of excitatory and inhibitory inputs in pyramidal neurons. Proc Natl Acad Sci U S A. 2009; 106(51):21906 11. https:// doi.org/10.1073/pnas.0912022106 PMID: 19955407 36. Ade KK, Janssen MJ, Ortinski PI, Vicini S. Differential tonic GABA conductances in striatal medium spiny neurons. J Neurosci. 2008; 28(5):1185–97. https://doi.org/10.1523/JNEUROSCI.3908-07.2008 PMID: 18234896 37. Ben-Ari Y. Excitatory actions of GABA during development: the nature of the nurture. Nat Rev Neurosci. 2002; 3(9):728–739. https://doi.org/10.1038/nrn920 PMID: 12209121 38. Ruusuvuori E, Kaila K. Carbonic Anhydrase: Mechanism, Regulation, Links to Disease, and Industrial Applications. Subcell Biochem. 2014; 75:271–290. https://doi.org/10.1007/978-94-007-7359-2_14 PMID: 24146384 39. Wilson CJ, Kawaguchi Y. The origins of two-state spontaneous membrane potential fluctuations of neostriatal spiny neurons. J Neurosci. 1996; 16(7):2397–2410. https://doi.org/10.1523/JNEUROSCI. 16-07-02397.1996 PMID: 8601819 40. Hibino H, Inanobe A, Furutani K, Murakami S, Findlay I, Kurachi Y. Inwardly rectifying potassium chan- nels: their structure, function, and physiological roles. Physiol Rev. 2010; 90(1):291–366. https://doi. org/10.1152/physrev.00021.2009 PMID: 20086079 41. Schiller J, Major G, Koester HJ, Schiller Y. NMDA spikes in basal dendrites of cortical pyramidal neu- rons. Nature. 2000; 404(6775):285–289. https://doi.org/10.1038/35005094 PMID: 10749211 42. Golding NL, Spruston N. Dendritic Sodium Spikes Are Variable Triggers of Axonal Action Potentials in Hippocampal CA1 Pyramidal Neurons. Neuron. 1998; 21(5):1189–1200. https://doi.org/10.1016/s0896- 6273(00)80635-2 PMID: 9856473 43. Balleine BW. The Neural Basis of Choice and Decision Making. J Neurosci. 2007; 27(31):8159–8160. https://doi.org/10.1523/jneurosci.1939-07.2007 44. Reig R, Silberberg G. Multisensory integration in the mouse striatum. Neuron. 2014; 83(5):1200–1212. https://doi.org/10.1016/j.neuron.2014.07.033 PMID: 25155959 45. Alegre-Corte´ s J, Sa´ ez M, Montanari R, Reig R. Medium spiny neurons activity reveals the discrete seg- regation of mouse dorsal striatum. Elife. 2021; 10:e60580. https://doi.org/10.7554/eLife.60580 PMID: 33599609 46. Ketzef M, Spigolon G, Johansson Y, Bonito-Oliva A, Fisone G, Silberberg G. Dopamine Depletion Impairs Bilateral Sensory Processing in the Striatum in a Pathway-Dependent Manner. Neuron. 2017; 94(4):855–65.e5. https://doi.org/10.1016/j.neuron.2017.05.004 PMID: 28521136 47. Tepper JM, Tecuapetla F, Koo´ s T, Ibañez-Sandoval O. Heterogeneity and diversity of striatal GABAer- gic interneurons. Front Neuroanat. 2010; 4:150. https://doi.org/10.3389/fnana.2010.00150 PMID: 21228905 48. Park J, Coddington LT, Dudman JT. Basal Ganglia Circuits for Action Specification. Annu Rev Neurosci. 2020; 43(1):1–23. https://doi.org/10.1146/annurev-neuro-070918-050452 PMID: 32303147 49. Heiman M, Kulicke R, Fenster RJ, Greengard P, Heintz N. Cell type–specific mRNA purification by translating ribosome affinity purification (TRAP). Nat Protoc. 2014; 9(6):1282–1291. https://doi.org/10. 1038/nprot.2014.085 PMID: 24810037 50. R Core Team. A language and environment for statistical computing. R Foundation for Statistical Com- puting, Vienna, Austria; 2023. Available from: https://www.R-project.org/. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002483 January 31, 2024 26 / 26 PLOS BIOLOGY
10.1371_journal.pgen.1011003	RESEARCH ARTICLE Kombucha Tea-associated microbes remodel host metabolic pathways to suppress lipid accumulation Rachel N. DuMez-KornegayID M. DeLoachID 2, Robert H. DowenID 1, Lillian S. BakerID 1,2,3,4* 2, Alexis J. Morris2, Whitney L. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: DuMez-Kornegay RN, Baker LS, Morris AJ, DeLoach WLM, Dowen RH (2024) Kombucha Tea-associated microbes remodel host metabolic pathways to suppress lipid accumulation. PLoS Genet 20(3): e1011003. https://doi.org/10.1371/ journal.pgen.1011003 Editor: Sean P. Curran, University of Southern California, UNITED STATES Received: October 4, 2023 Accepted: February 22, 2024 Published: March 28, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pgen.1011003 Copyright: © 2024 DuMez-Kornegay et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. The whole genome sequencing data are available at the Sequencing Read Archive 1 Curriculum in Genetics and Molecular Biology, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States of America, 2 Department of Biology, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States of America, 3 Department of Cell Biology and Physiology, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States of America, 4 Integrative Program for Biological and Genome Sciences, The University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, United States of America * [email protected] Abstract The popularity of the ancient, probiotic-rich beverage Kombucha Tea (KT) has surged in part due to its purported health benefits, which include protection against metabolic dis- eases; however, these claims have not been rigorously tested and the mechanisms underly- ing host response to the probiotics in KT are unknown. Here, we establish a reproducible method to maintain C. elegans on a diet exclusively consisting of Kombucha Tea-associated microbes (KTM), which mirrors the microbial community found in the fermenting culture. KT microbes robustly colonize the gut of KTM-fed animals and confer normal development and fecundity. Intriguingly, animals consuming KTMs display a marked reduction in total lipid stores and lipid droplet size. We find that the reduced fat accumulation phenotype is not due to impaired nutrient absorption, but rather it is sustained by a programed metabolic response in the intestine of the host. KTM consumption triggers widespread transcriptional changes within core lipid metabolism pathways, including upregulation of a suite of lyso- somal lipase genes that are induced during lipophagy. The elevated lysosomal lipase activ- ity, coupled with a decrease in lipid droplet biogenesis, is partially required for the reduction in host lipid content. We propose that KTM consumption stimulates a fasting-like response in the C. elegans intestine by rewiring transcriptional programs to promote lipid utilization. Our results provide mechanistic insight into how the probiotics in Kombucha Tea reshape host metabolism and how this popular beverage may impact human metabolism. Author summary Kombucha is a popular fermented tea that has been purported to have many human health benefits, including protection against metabolic diseases like diabetes and obesity. These health benefits are thought to be conferred by the probiotic microbes found in Kombucha Tea, which includes both bacterial and yeast species, that may be able to PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 1 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism (PRJNA1044129). Raw and processed mRNA-Seq data have been deposited in GEO (GSE236037). Funding: This work was supported by NIGMS grant T32GM007092 to R.N.D., NCCIH grant F31AT012138 to R.N.D., and NIGMS grant R35GM137985 to R.H.D. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. colonize the human intestine and alter host physiology. The mechanisms by which the Kombucha Tea-associated probiotic microorganisms (KTMs) impact host physiology are largely unknown. Using the nematode Caenorhabditis elegans as an animal model system to study the host physiological response to KTMs, we show that KTMs colonize the C. ele- gans intestine and impart widespread changes in the expression of evolutionarily con- served lipid metabolism genes, resulting in reduced fat levels in the host. The host metabolic response to actively fermenting KTMs requires an increase in proteins that break down lipids paired with a reduction in a protein that builds triglycerides, which mirrors the events that occur during fasting. These findings are consistent with the reported human health benefits of Kombucha Tea and provide new insights into the host response to Kombucha-associated microbes, which could inform the use of Kombucha in complementary health care approaches in the future. Introduction Since the discovery of antibiotics, humans have been successfully eliminating microbes to cure infections and sterilize our environments, but this nonspecific approach to eliminate patho- genic microbes has also made it increasingly evident just how much we rely on interactions with commensal microbes to remain healthy. Antibiotic use, western diets, a sedentary life- style, and many disease states can trigger dysbiosis, or a reduction in microbial diversity, which has been linked to metabolic syndromes, chronic inflammation, and mental health dis- orders [1–3]. For example, C. difficile colitis can arise from antibiotic use and a subsequent loss of microbial diversity in the gut, resulting in severe gastrointestinal symptoms and potentially death [4]. Consumption of probiotics, or live microbes associated with health benefits, can promote, or maintain, a healthy gut microbiome while supplying the host with crucial micro- bially-derived metabolites [5–7]. Understanding the molecular mechanisms underlying the host response to microbes, particularly probiotics [8], is critical for their incorporation into complementary health care approaches. Kombucha tea (KT) is a semi-sweet, fermented beverage that is widely consumed as a func- tional food (i.e., providing health benefits beyond nutritional value) and contains probiotic microbes that have been purported to confer health benefits, including lowering blood pressure, protection against metabolic disease, improved hepatoprotective activity (i.e., protection against liver toxins), and anticancer effects [9–13]. These probiotic microbes include members of the Acetobacter, Lactobacillus, and Komagataeibacter genera [14,15]. While some of these health ben- efits have begun to be tested in animal models, including the ability of KT to ameliorate diabetic symptoms or limit weight gain in adult mice [16–19], the mechanistic underpinnings of these phenotypes have not been rigorously investigated. Moreover, the interactions between the microbes in Kombucha Tea, which include both bacterial and yeast species, and the host remain completely unexplored. Because this beverage contains live probiotic microbes and is widely con- sumed under the largely unsubstantiated claim that it confers health benefits, it is imperative to gain mechanistic insight into the host physiological and cellular response to KT consumption. The impact of individual probiotic microbes, or in this case the small community of Kom- bucha-associated microbes, on human physiology is difficult to deconvolute as humans con- sume a complex diet, have trillions of microbes colonizing their gut, and mechanistic investigation of host-microbe interactions is not feasible in human subjects. Therefore, use of animal models is essential to investigate how probiotic consumption influences host physio- logical processes. Caenorhabditis elegans has been widely used to investigate mechanisms of metabolic regulation and how nutrient sensing pathways govern organismal homeostasis PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 2 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism [20,21]. C. elegans is also an emerging model for studying the impact of the gut microbiome on host physiology [22,23]. Axenic preparation of C. elegans cultures renders these bacterivore animals microbe-free at the onset of life, allowing for complete experimental control over which microbes are consumed during their lifetime (i.e., animals are germ-free before encoun- tering their microbial food source). Additionally, microbes that escape mechanical disruption during feeding can robustly colonize the intestinal lumen [22–24]. Thus, the simple digestive tract of C. elegans is effectively colonized by bacteria that are provided as a food source, making it an ideal system to interrogate the host metabolic response to consumption of specific microbes. Indeed, previous studies have successfully used C. elegans to investigate how individ- ual species of microbes, including probiotics, can elicit physiological changes by rewiring con- served genetic pathways [25–31]. Here, we use C. elegans to investigate whether intestinal colonization with Kombucha-asso- ciated microbial species (KT microbes or KTMs) rewires host metabolism. We developed a reproducible method to culture animals on lawns of KT microbes consisting of microbes found in all commercial and homebrewed KTs (i.e., bacteria from the Acetobacter and Koma- gataeibacter genera and a yeast species). We found that animals feeding ad libitum on KT microbes accumulate significantly less fat than animals consuming either an E. coli diet, any of the individual three KT-associated microbial species, or a simple non-fermenting mix of these three species. Furthermore, our data suggest that KT consumption reduces fat storage by mod- ulating host lipid metabolism pathways rather than restricting caloric intake. To gain insight into the mechanisms that underlie this reduction in lipid levels, we performed a transcriptomic analysis of KT microbe-fed animals, which revealed that a class of lysosomal lipases that func- tion in lipophagy was up-regulated and that a crucial enzyme in triglyceride synthesis was down-regulated in response to KT microbes. Our results suggest that Kombucha Tea con- sumption may alter lipid droplet dynamics by promoting their degradation via lipophagy, while simultaneously restricting lipid droplet expansion through down-regulation of triglycer- ide synthesis. This investigation lays crucial groundwork to deconvolute the molecular mecha- nisms that may underlie the purported health benefits of KT using a genetically tractable animal model. Results Rearing C. elegans on a lawn of KT microbes results in reproducible colonization of the gut Small batch brewing of KT is a serial fermentation process in which the microbially-generated biofilm and a small amount of fully fermented liquid culture are transferred to a fresh prepara- tion of sucrose media, which then ferments for at least a week prior to consumption. This tra- ditional method of brewing KT results in a dynamic microbial community and pH shift over the course of fermentation (pH decreases from 7 to ~4). Contamination by environmental microbes is limited since these species are outcompeted by the core KT microbes (KTMs) as the pH drops [13,32–34]. Furthermore, construction of the protective pellicular biofilm, collo- quially referred to as a SCOBY (symbiotic culture of bacteria and yeast), by the KTMs reduces outside contamination [35]. To investigate the physiological and metabolic effects of Kombu- cha Tea consumption using a genetic model system, we first sought to establish a reproducible method to deliver KTMs to C. elegans animals via feeding on our standard agar-based nema- tode growth media (NGM), which do not contain any antibiotics or antifungals. We found that seeding KTMs that are actively growing in a KT homebrew onto NGM plates is sufficient to generate a lawn of microbes that expands in population and produces a biofilm over the course of 4 days (S1A and S1B Fig). PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 3 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism To gain a better understanding of the microbial community dynamics in our KT culture and to assess our ability to recapitulate the KT microbial community on NGM plates, we per- formed 16S rDNA sequencing of the fermenting KT culture and the KTMs washed from NGM plates isolated from three representative brew cycles (S1 Table). After six days of fer- mentation, the microbial communities in the culture and on NGM plates were similar and were dominated by the expected set of Kombucha-associated microbes (i.e., Acetobacter and Komagataeibacter species), which are essential components of all commercial or homebrewed KTs (Figs 1A, 1B and S1C–S1E) [36]. Notably, the KT culture microbial community remained similar through day 12 of fermentation; however, the community on NGM plates was no lon- ger dominated by the expected KTMs at day 12, which may be due to the expansion of envi- ronmental microbes (Fig 1A and 1B). Thus, we exclusively used NGM plates between days 4–8 after KTM seeding for our subsequent experiments. Establishing this method to reproducibly culture the KT microbial community on NGM plates was essential to leveraging C. elegans as a model to study the host response to KT consumption. Using this standardized method of KTM culturing, we next sought to determine if popula- tions of C. elegans could be reared on a diet exclusively consisting of KTMs. Given that KTMs are a mix of microbial and yeast species, we first conducted an avoidance assay to assess whether C. elegans animals would remain on or flee from the lawn, which is a typical response to a pathogen [37–40]. Importantly, animals remained on the KTM microbial lawn throughout development and into adulthood at levels similar to the E. coli controls (Fig 1C and 1D), indi- cating that C. elegans animals can be successfully reared on KT microbes. These comparisons, as well as all our subsequent characterizations of the KTMs, were conducted in conjunction with two standard laboratory E. coli diets (OP50 and HT115 [41]) and a strain of the bacterium Acetobacter tropicalis that we isolated from our KT culture. A. tropicalis is a major constituent of all KTs and produces vitamin B12 among other bioactive molecules found in KTs [14,15,34,35,42–45]. OP50 and HT115 E. coli strains modulate C. elegans physiology differ- ently, which can be partially attributed to differences in vitamin B12 levels [28,46–48]. Interest- ingly, when presented with a choice of diets animals did not select the KTM lawn (S2A and S2B Fig). This behavior was consistent across different C. elegans wild isolates and other Cae- norhabditis species (S2C–S2I Fig), suggesting that these animals are either attracted to the E. coli or are repelled by a component of the KTM culture. Though animals seem to prefer other food sources, animals offered only KTMs do not flee the lawn, demonstrating that C. elegans can be reliably reared on a KTM-exclusive diet using standard ad libitum feeding methods. To test whether KTMs altered feeding behavior, which might result in reduced caloric intake, we measured pumping rates (i.e., the rate at which animals’ pharyngeal muscle con- tracts to intake food) of individual animals consuming KTMs or control food sources. We found no significant difference in pumping rates of animals consuming KTMs compared to any of the other food sources (Fig 1E), suggesting feeding behavior is not altered on KTM lawns. Finally, we assessed whether KTMs colonize the intestinal lumen of C. elegans animals, as would be predicted for these probiotic microbes in the human gastrointestinal tract. After rearing animals on different diets, we removed surface microbes, extracted the intestinal microbes, and quantified the colony forming units (CFUs) present. Animals consuming KTMs contained at least 5 times more CFUs than animals consuming any other diet, indicat- ing that KTMs robustly colonize the C. elegans gut (Fig 1F). To further investigate this intesti- nal colonization, we used scanning electron microscopy to image the intestine of animals consuming KTMs and found intact microbial cells present in the intestinal lumen (Fig 1G). Together, these results demonstrate that C. elegans animals can be successfully reared on a KTM-exclusive diet, which closely mirrors the microbial community found in the KT culture, resulting in robust KTM colonization of the gut. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 4 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 5 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 1. Consumption of Kombucha-associated microbes (KTMs) does not impact feeding behavior and results in robust gut colonization of C. elegans animals. (A) 16S rDNA sequencing of traditionally cultured KTMs, the biofilm of an actively fermenting Kombucha Tea, or KTMs grown on NGM agar plates. The frequency of known KTMs (green) is plotted relative to environmental microbial contaminants (gray) for three biological replicates across twelve days (d1-d12). (B) A Principal Component Analysis of the weighted unifrac beta diversity derived from the 16S rDNA sequencing revealed similarity between d5 plates and the d6 culture, but divergence of the d12 plates. (C) Images (scale bar, 500 μm) and (D) quantification (mean ± SEM) of day 1 adult animals on lawns of the indicated microbes (72 hr timepoint compared, *, P<0.001, ns, not significant, one-way ANOVA). (E) Measurements of pumping rates for day 1 adults consuming each microbial food source (mean ± SD, ns, not significant, one-way ANOVA). (F) Quantification of the microbial CFUs from animals consuming each diet shows KTMs colonize the gut at higher levels compared to the control diets (mean ± SEM, , P<0.0001, one-way ANOVA, three biological replicates, 10 animals per replicate). (G) Representative scanning electron microscopy images of KTMs in the intestinal lumen (black arrows point to the intestinal microvilli and green arrow heads indicate intact KTMs; left scale bar, 2 μm; right scale bars, 1 μm). Expanded data for panel A can be found in S1 Table and the raw data underlying panels D, E, and F can be found in S1 Data. https://doi.org/10.1371/journal.pgen.1011003.g001 Animals consuming Kombucha microbes exhibit reduced fat accumulation Dietary components, including those produced by probiotic microbes, can play a substan- tial role in modulating host metabolism, including lipid storage and lipolysis [49–51]. Con- sistently, C. elegans metabolism is remarkably sensitive to differences in microbial diets, as even highly similar strains of E. coli promote markedly different levels of fat content [28,29,41]. Given the purported metabolic benefits of KT in humans, including decreased risk of obesity [9–13], we reasoned that consumption of KTMs may impact lipid levels in C. elegans. The majority of fat in C. elegans animals is stored in intestinal epithelial cells within lipid droplets in the form of triglycerides (TAGs), with smaller lipid deposits found in the hypodermis and germline [52]. Using the well-established lipophilic dyes Oil Red O and Nile Red, which both stain neutral lipids, we examined the fat content of animals consum- ing KTMs and control microbes [52,53]. Animals consuming KT microbes accumulated significantly less fat than animals consuming other food sources, including A. tropicalis, which is particularly noteworthy given that A. tropicalis is the most abundant microbial spe- cies in KT (Fig 2A–2D). These trends continued during and after the reproductive period (S3A Fig), suggesting that KTMs restrict host lipid accumulation throughout reproduction and during the aging process. Importantly, the KTM-fed animals successfully commit a sig- nificant proportion of their somatic fat stores to the germline and developing embryos at adulthood (Fig 2C), suggesting that reproductive programs are not impaired despite the overall reduction in lipid levels. The decrease in Oil Red O and Nile Red staining suggests that animals consuming KTMs may have reduced TAG levels compared to animals on con- trol diets. Therefore, we used a biochemical assay to quantify the total amount of TAGs in populations of animals fed each diet [54,55]. Consistent with our previous observations, animals consuming KTMs had an ~85% or ~90% decrease in TAG levels compared to ani- mals consuming E. coli OP50 or A. tropicalis, respectively (Fig 2E). Together, these data clearly demonstrate that animals consuming KT microbes accumulate less fat than E. coli- fed animals and that the most abundant microbe in KT, A. tropicalis, is not sufficient to recapitulate this phenotype. This finding is particularly relevant to human health, as KT consumption has been shown to restrict weight gain and alleviate diabetic symptoms to a similar degree as metformin in rodent models [16–19]. Given that the major site of lipid storage in C. elegans is in intestinal lipid droplets (LD), we hypothesized that LD size or abundance may be impacted in the intestine of KTM-fed animals. Taking advantage of a transgenic strain that expresses the LD-residing DHS-3::GFP protein (dhs, dehydrogenase, short chain), we measured LD abundance and size in intestinal cells of animals fed each diet. Both lipid droplet size and abundance were dramatically reduced in ani- mals consuming KTMs relative to E. coli- or A. tropicalis-fed animals (Figs 2F–2H and S3B). Together, these results suggest that regulation of lipid droplet synthesis or stability may account for the reduced lipid accumulation that we observed in KTM-fed animals. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 6 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 2. KTMs restrict lipid accumulation in the host. (A) Representative images (scale bar, 500 μm) and (B) quantification (mean ± SD, , P<0.0001, one-way ANOVA) of day 1 adults stained with Oil Red O. (C) Representative fluorescence images (scale bar, 500 μm) and (D) quantification (mean ± SD, , P<0.0001, one-way ANOVA) of day 1 adults stained with Nile Red. (E) Biochemical quantification of the triglycerides (TAGs per animal) in animals consuming each food source (mean ± SEM, , P<0.001, , P<0.05, ns, not significant, one-way ANOVA). (F) Representative fluorescence images of DHS-3::GFP (dhs, dehydrogenase, short chain) at intestinal lipid droplets in animals consuming the indicated microbial diets (scale bar, 5 μm). (G) Lipid droplet size measurements with each datapoint representing the average intestinal lipid droplet diameter for a single animal (mean ± SD, **, P<0.0001, one-way ANOVA). (H) Lipid droplet density measurements with each datapoint representing the number of lipid droplets per μm2 for a single animal (mean ± SD, *, P<0.0001, , P<0.05, ns, not significant, one-way ANOVA). Raw data underlying panels B, D, E, G, and H can be found in S2 Data. https://doi.org/10.1371/journal.pgen.1011003.g002 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 7 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism KTM consumption accelerates growth rates and does not substantially alter fecundity Different microbial diets can have a profound impact on C. elegans growth rate and fecundity [29,30]. A KTM diet could restrict developmental rate or alter reproductive programs. More- over, reduced nutrient absorption stemming from a KTM diet could result in caloric restric- tion and reduced lipid accumulation. Indeed, genetic or nutritional models of caloric restriction cause animals to develop more slowly, to accumulate less intestinal fat, and to have a delayed reproductive period that ultimately results in less progeny production [28,56–58]. Therefore, we sought to determine whether animals consuming KTMs exhibit slower devel- opmental rates and smaller brood sizes than animals consuming either an E. coli or A. tropica- lis diet. To investigate variations in developmental rate, we employed a transgenic strain expressing a GFP-PEST protein under the control of the mlt-10 promoter (Pmlt-10:: GFP-PEST), which is specifically expressed during each of the four molt stages, resulting in four peaks of GFP fluorescence throughout development (Fig 3A). The PEST amino acid sequence ensures rapid GFP turnover by proteolytic degradation and allows for precise tempo- ral analyses. Animals consuming KTMs molt at a similar, if not an accelerated rate relative to animals on the control food sources (Fig 3A–3C), clearly indicating that KTM consumption does not decrease developmental rate. To gain a more comprehensive view of animal develop- ment during KTM consumption, we performed mRNA sequencing (mRNA-Seq) of adult ani- mals consuming E. coli, A. tropicalis, or KTMs. Upon inspection of 2,229 genes previously associated with C. elegans development [29], we observed very few gene expression differences between KTM-fed animals and those fed control diets (Fig 3D–3F), suggesting that the KTM- fed population reaches adulthood synchronously. Together, these results suggest that animals consuming KT microbes exhibit wild-type development. Caloric restriction has a profound impact on C. elegans physiology, including reduced devel- opmental rate [58]. The eat-2 mutant is a genetic model of caloric restriction, as loss of eat-2 results in impaired pharyngeal pumping and reduced nutrient intake [58]. Reducing nutrient availability (i.e., E. coli OP50 lawns with concentrations � 109 CFU/ml) provides a second effec- tive method of caloric restriction [59]. Therefore, to further evaluate whether animals consum- ing KTMs are calorically restricted (CR), we conducted developmental rate assays with wild- type and eat-2 mutant animals consuming ad libitum E. coli lawns, CR E. coli lawns (108–109 CFU/ml), or our standard ad libitum lawns of KTMs. This analysis revealed that both wild-type and eat-2 animals exhibited accelerated developmental rates when consuming KTMs compared to the E. coli OP50 diet (Fig 3G). Importantly, eat-2 animals showed reduced developmental rates on CR E. coli lawns relative to ad libitum E. coli lawns, indicating that the effects of the eat- 2 mutation are further enhanced by additional caloric restriction; however, KTM-feeding par- tially suppressed the developmental defects of the eat-2 mutation (Fig 3G). These data demon- strate that KTM consumption does not mimic the effects of restricted caloric intake. Reproductive output (i.e., brood size) of C. elegans is modulated by diet, possibly through the tuning of reproductive programs at the transcriptional level [29,60]. Therefore, we mea- sured the brood sizes of animals consuming KTMs and control diets, finding that the average brood size of animals consuming KTMs was only modestly lower than those consuming E. coli OP50 (Figs 3H and S4; 295 versus 240, P<0.05). Additionally, we found that animals consum- ing KTMs lay their eggs at a similar rate relative to E. coli-fed animals (Figs 3I and S4). In con- trast, calorically restricted animals, such as eat-2 mutants, have extended egg laying periods, up to 12 days, and have substantially reduced brood sizes, with eat-2 mutants averaging 100– 175 progeny [28,57]. Thus, the ~20% reduction in fertility for KTM-fed animals is inconsistent with the more severe reduction in brood size of CR animals. It could, however, be consistent PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 8 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 3. Developmental timing is accelerated and reproductive output is only modestly reduced during KTM feeding, suggesting that caloric intake is not restricted by KTM consumption. (A-C) Profiles of Pmlt-10::GFP-PEST expression throughout development after dropping PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 9 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism synchronized L1s on the indicated microbes. The reporter is expressed exclusively during the larval molts (shown in gray in A). A single representative experiment is displayed in panels A-C. (D-F) Scatter plots comparing the expression of 2,229 developmental genes as determined by mRNA-Seq (RPKM, reads per kilobase of transcript per million mapped reads). A linear regression analysis and the corresponding R2 value is reported for each comparison. (G) The frequency (mean ± SEM) of wild-type N2 and eat-2(ad465) individuals at the indicated developmental stages after 48 hours of growth on ad libitum KTM, ad libitum E. coli, or caloric restriction E. coli (108 or 109 CFUs/mL) plates (**, P<0.0001, chi- squared test). (H) Brood sizes of wild-type animals reared on the different diets (mean ± SD, , P<0.001, , P<0.05, one-way ANOVA). (I) A plot of progeny production for each day during the reproductive period demonstrating that KTM-fed animals exhibit a similar egg laying rate compared to E. coli OP50-fed animals. (J) Normalized vit-2 gene expression values (RPKM, reads per kilobase of transcript per million mapped reads; mean ± SEM, , P<0.05, T-test) and (K) quantification of VIT-2::GFP fluorescence in early embryos (mean ± SD, , P<0.001, T-test) from animals consuming an E. coli OP50 or KTM diet. (L-N) Scatter plots and a linear regression analysis (R2 value reported) comparing the expression of 2,367 reproduction genes as determined by mRNA-Seq. Raw data underlying panels A-N can be found in S3 Data. https://doi.org/10.1371/journal.pgen.1011003.g003 with impaired maternal provisioning of lipid-rich yolk to oocytes from intestinal fat stores, a process termed vitellogenesis. Thus, we next examined the mRNA levels of vit-2, which encodes a vitellogenin protein that mediates the intestine-to-oocyte transport of lipids, finding that vit-2 levels are increased in animals fed a KTM diet compared to E. coli-fed animals (Fig 3J). Consistently, vitellogenin protein levels, which we measured in early embryos (prior to the 44-cell stage) using an endogenously tagged VIT-2::GFP protein, were also elevated in KTM-fed animals (Fig 3K), further substantiating that KTM consumption does not impair maternal lipid provisioning. Finally, we inspected the expression of 2,367 genes implicated in reproduction [29], finding that KTM consumption does not broadly alter reproductive gene expression programs rel- ative to control diets (Fig 3L–3N). Together, our results indicate that reproductive programs are not dramatically altered in animals consuming KTMs. This finding, along with the observation that animals consuming KTMs exhibit wild-type developmental rates, is consistent with the con- tention that caloric intake is not impaired during KTM consumption and substantiates C. elegans as model to investigate the impact of Kombucha-associated microbes on host metabolic pathways. Long-term KTM co-culturing is required to remodel host metabolism Sequencing of commercially available and non-commercial-small-batch KTs has revealed that a reproducible set of core microbes are found in KT [14,15,34,36,61]. These include bacteria in the Acetobacter, Komagataeibacter, Gluconacetobacter, Gluconobacter, and Lactobacillus gen- era, as well as yeast in the Brettanomyces, Zygosaccharomyces, Candida, Dekkera, Lachancea, and Schizosaccharomyces genera. Furthermore, Huang and colleagues recently established a minimal KT microbiome that recapitulates key aspects of traditionally brewed KT based on the criteria that this microbial mix could (1) coexist as in KT, (2) produce a KT-like biochemi- cal composition, and (3) build a pellicle. Intriguingly, regardless of the ratio of bacteria-to- yeast at the onset of fermentation, by day 6 this ratio stabilizes with relatively equal representa- tion of each species regardless of the concentration of the microbial species combined [36]. Given that KTMs robustly colonize the C. elegans gut and that feeding animals the known dominant KT microbe, A. tropicalis, fails to recapitulate the host response to KTMs, we sought to identify additional microbes from our KT culture that can colonize the intestine of animals after KTM consumption. Isolation of these species would facilitate the creation of a KTM cul- ture consisting of a minimal microbiome core, which may be sufficient to confer metabolic phenotypes in C. elegans animals. Our initial extraction of intestinal microbes from KTM-fed animals (Fig 1F) isolated a bacterial species, Acetobacter tropicalis, and a yeast species, of either the Zygosaccharomyces or Brettanomyces genera, which we identified by 16S and 18S rDNA sequencing, respectively (S5A and S5B Fig). While these microbes represent two of the species commonly found in KT, they do not constitute a minimal KT culture because they cannot form a pellicle [36]. Therefore, we sought to isolate the cellulose-producing species from our culture that is responsible for building the pellicle. We removed a small piece of the biofilm PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 10 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism from our KT culture and used a combination of enzymatic digestion (driselase) and mechani- cal disruption (sonication) to dislodge the bacteria from the cellulose matrix. The cellulose- producing bacterium was isolated on mannitol agar plates containing Calcofluor White, which stains cellulose and chitin and fluoresces under ultraviolet light [62–64]. This strategy resulted in the isolation of an additional KT microbe that was identified as a member of the Komaga- taeibacter genus by 16S rDNA sequencing (S5C Fig). To gain additional genetic information about our individually isolated KT microbes, we performed short read whole genome sequencing of the genomic DNA. Subsequently, the Kra- ken algorithm [65], a bioinformatic pipeline for metagenomic classification, was used to deter- mine the approximate taxonomy of our individual KT microbes. Based on these taxonomical classifications, as well as a compiled list of previously published KT-associated microbes, we aligned our KT microbe sequences to several available reference genomes to gain species level information (Fig 4A and S2 Table). This strategy identified our KT microbes as Komagataei- bacter rhaeticus (98.76.% alignment rate to strain ENS_9a1a), Acetobacter tropicalis (87.55% alignment rate to strain NBRC101654), and Zygosaccharomyces bailli (86.88% alignment rate to strain CLIB213) [66]. The isolation and identification of the dominant KT microbes from our culture allowed us to further investigate how consumption of the individual KT microbes, or mixtures of microbes, alter C. elegans lipid metabolism (Fig 4B). We initially fed the individual KT microbes to ani- mals, finding that diets of A. tropicalis or K. rhaeticus promoted lipid accumulation at levels similar to E. coli-fed animals, while a diet of the yeast species Z. bailli failed to support animal development (Figs 4C and S6A–S6D). Surprisingly, increasing the concentration of KTMs pres- ent in the lawn fivefold (5x KTM) further reduced lipid levels compared to our standard KTM lawn (S6A–S6C Fig), indicating that an increase in the microbial concentration, which likely results in additional available nutrients, does not increase host lipid accumulation. We then hypothesized that a mixture of K. rhaeticus, Z. bailii, and A. tropicalis would repre- sent the minimal core of KT microbes, which when co-cultured would ferment sucrose, build a pellicle, and produce a biochemical composition similar to Kombucha tea. Therefore, we combined the three KT microbe isolates in filter sterilized KT media (i.e., steeped black and green tea containing ~5% sucrose) and allowed them to ferment for several weeks until a pelli- cle was formed. We refer to this de novo KT as KTM-Fermented Mix or “KTM-FM” (Fig 4B). To assess the ability of our KTM-FM culture to alter host lipid metabolism, we performed Oil Red O staining on animals consuming KTM-FM or a simple non-fermenting mix of the three KT microbes (referred to as KTM-Mix, abbreviated “KTM-M”, Fig 4B). Intriguingly, we found that the KTM-M diet did not reduce lipid accumulation, lipid droplet size, or lipid drop- let abundance in the host (Figs 4C–4E and S6E); however, consumption of KTM-FM reduced lipid levels to a similar degree as the original KTM diet (Fig 4F). Importantly, neither the KTM-M nor the KTM-FM diet impaired developmental or behavioral programs (S6F–S6I Fig). These results suggest that long-term fermentation is necessary for the host metabolic response to KT consumption. Furthermore, the observation that a non-fermented mix of KT microbes fails to restrict host lipid accumulation further supports our conclusion that KTM- fed animals are not calorically restricted. To better understand the importance of fermentation time, we fed animals KTM-FM cultures after different lengths of fermentation and measured lipid levels using Oil Red O staining. Ani- mals that were fed KTM-FMs with fermentation times less than one week had elevated lipid lev- els; however, KTM-FMs fermented for 2 weeks or more promoted the depletion of host lipids (Fig 4G). Additionally, removal of the fully fermented KTM supernatant followed by repeated washes of the KTMs with a 5% sucrose solution prior to seeding the NGM plates did not alter host lipid accumulation in response to KTMs (Fig 4H and 4I), suggesting that the small molecules PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 11 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 4. The host lipid depletion response is specific to actively fermenting KTM cultures and is not conferred by individual microbes or a non- fermenting mixture. (A) Purification and whole genome sequencing of the microbes from our Kombucha culture resulted in species-level PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 12 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism identification of the core KTMs. (B) A schematic of the preparation and delivery methods for the three KT-derived diets (orange, A. tropicalis; tan, K. rhaeticus; gray, Z. bailii; d, days). Sweet tea media consists of a black and green tea mix with 5% cane sugar that has been filter sterilized. KTM cultures are maintained via serial fermentation, while the KTM-M and KTM-FM are de novo cultures. (C) Quantification of Oil Red O staining of day 1 adults fed the indicated diets (mean ± SD, , P<0.0001, one-way ANOVA). A Z. bailii diet does not support animal development. (D) Representative fluorescence images (scale bars, 5 μm) and (E) quantification of lipid droplet diameter (mean ± SD, n = 10 individuals, , P<0.0001, unpaired T-test) in the intestines of DHS-3::GFP transgenic animals. The KTM lipid droplet image and size measurements shown in (D-E) are also displayed in Fig 2F and 2G, as all these samples were processed in parallel. (F-G) Quantification of Oil Red O staining of day 1 adults fed the indicated KT diets (mean ± SD, , P<0.0001, ns, not significant, one-way ANOVA). (F) Animals consuming KTM-FM have similar lipid levels as KTM-fed animals, while (G) KTM-M must be co-cultured for at least 14 days to restrict host lipid accumulation. (H) The experimental design to test whether the KTM culture supernatant is required for host lipid depletion. (I) Quantification of Oil Red O staining of day 1 adults following a diet of KTMs, KTMs washed extensively with 5% sucrose, or KTM-M (mean ± SD, , P<0.0001, ns, not significant, one-way ANOVA). (J) Quantification of Oil Red O staining of day 1 adult animals consuming E. coli, KTMs, and Z. bailii with or without dead E. coli supplementation as an inert nutrient source (mean ± SD, , P<0.0001, ns, not significant, one-way ANOVA). (K) Representative DIC and fluorescence images of the intestine after feeding the indicated diets supplemented with C1-BODIPY-C12 (stars indicate the intestinal lumen and arrowheads indicate the intestinal epithelial cells; scale bars, 10 μm). Expanded data for panel A can be found in S2 Table and the raw data underlying panels C, E, F, G, I, and J can be found in S4 Data. https://doi.org/10.1371/journal.pgen.1011003.g004 in the green and black tea may be dispensable for conferring host lipid phenotypes. This result, however, does not rule out the possibility that the tea-derived metabolites are crucial for establish- ing the symbiotic Kombucha culture. Together, these data argue that KT microbes must form an established community to reconfigure host lipid metabolism pathways. Although we observed colonization of C. elegans gut with A. tropicalis (Fig 1F), it is unclear whether the other KT isolates, K. rhaeticus or Z. bailii, are ingested by animals. To visualize these microbes in the gut of live animals, we stained animals fed E. coli, K. rhaeticus, or KTMs with Calcofluor White, which selectively stains the polysaccharides in chitin and cellulose. We observed the cellulose-producing microbe K. rhaeticus, which supports animal development, in the intestinal lumen (S6J Fig), suggesting that K. rhaeticus bacteria can colonize the gut while synthesizing cellulose. Surprisingly, we also observed chitin-producing yeast cells in the intestinal lumen, indicating that Z. bailii can be consumed by animals at the adult stage (S6K and S6L Fig). Importantly, these results are consistent with the presumption that all three of KT microbes (Z. bailii, K. rhaeticus, and A. tropicalis) isolated from our KT culture can escape mechanical disruption in the pharynx and can be found in the intestinal lumen of C. elegans. To further assess the ability of the KT microbes to colonize the gut, we quantified the intestinal lumen size of animals reared on the E. coli OP50, KTM, and KTM-M diets. Using animals expressing ERM-1::GFP, which localizes to the apical surface of intestinal cells and facilitates luminal measurements, we found that individuals consuming a KTM diet had an increased intestinal lumen diameter compared to animals consuming E. coli OP50 but not the KTM-M diet, suggesting that any diet containing KT microbes stimulates intestinal bloating (S6M Fig). The presence of Z. bailii in the gut, which may contribute to intestinal bloating, raised the possibly that the yeast (or the other KT microbes) may restrict nutrient absorption, resulting in caloric restriction. Therefore, we supplemented KTM and Z. bailii diets with heat killed E. coli OP50 and assessed lipid levels using Oil Red O staining. While E. coli supplementation had little impact on the ability of the KTM diet to restrict host lipid accumulation, supplemen- tation to a Z. bailii diet supported animal development and promoted lipid accumulation despite the presence of the yeast (Fig 4J). Next, we assessed nutrient absorption in KTM-fed animals by supplementing the KTM lawn with the vital dye C1-BODIPY-C12, which is con- sumed with the food and can readily cross the intestinal apical membrane. Following a three- hour pulse of BODIPY, animals consuming E. coli OP50, KTM, and KTM-M all have detect- able levels of BODIPY in their intestinal epithelial cells (Figs 4K and S7). Since animals con- suming KTMs have very few, small lipid droplets the BODIPY staining was diffusely distributed throughout the intestinal cells; however, in the E. coli OP50 and KTM-M-fed ani- mals the dye localized to intestinal lipid droplets and lysosome-related organelles [53,67]. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 13 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Together, these findings are consistent with our previous observations that animals consuming KT microbes, either individually or in combination, are not impaired in their ability to absorb nutrients; but rather, the KTM diet likely restricts lipid accumulation by modulating host met- abolic pathways. An intestinally driven metabolic response to KTM consumption KTM-fed animals undergo normal development and show no detectable impairment in nutri- ent absorption, yet store markedly less lipids than control animals, including those fed the KTM-M diet. While our transcriptomics suggested that the expression of genes involved in development or reproduction were consistent across diets, we hypothesized that the expression of metabolic genes may be specifically altered by KTM consumption. Therefore, we performed additional analyses of our mRNA-Seq data derived from day one adult animals consuming either KTM, KTM-M, A. tropicalis, or the two E. coli diets to investigate if specific metabolic programs are altered by these diets. A PCA analysis revealed that the transcriptomes of animals fed the same diet cluster, with the transcriptomes of animals fed KTM, KTM-M, and A. tropica- lis distinctly clustering apart from the transcriptomes of the E. coli-fed animals (Fig 5A), indicat- ing that there is at least some commonality between the transcriptional responses of animals consuming any of the KT-associated diets that is different from E. coli diets. To eliminate the possibility of transgenerational epigenetic effects of the KTM diet, we compared the transcrip- tomes of animals fed KTMs for one generation to animals subjected to five generations of the KTM diet, finding no significant difference between these transcriptomes (Figs 5A and S8A). Deeper investigation of our mRNA-Seq data revealed that each KT-associated diet did indeed result in some level of differential gene expression compared to the E. coli OP50 diet (A. tropicalis, 3,952 genes; KTM, 1,237 genes; KTM-M, 1,007 genes; 1% FDR; Figs 5B and S8B– S8F). Intriguingly, 295 genes were unique to the KTM diet (Fig 5B). Altered expression of these KTM-unique genes could be a major driver of the reduced lipid levels that we observed specifi- cally in the KTM-fed animals. A gene ontology (GO) enrichment analysis [68] of the KTM- unique genes revealed an enrichment for genes annotated to have functional roles in lipid metabolism (Fig 5C). Since misexpression of core metabolic genes alters longevity and stress resistance pathways [69,70], we queried whether these same genes were also misexpressed in animals with reduced levels of DAF-2 (i.e., the insulin receptor), which results in increased stress resistance, improved healthspan, and extended lifespan [71,72]. Indeed, depletion of DAF-2 in different tissues [72], including the intestine, results in transcriptional changes that are consistent with those seen in KTM-fed animals (S8G–S8I Fig). Together, these data suggest that consumption of fermenting KT microbes may remodel host lipid metabolism and stress resilience pathways to restrict fat accumulation and improve healthspan. In C. elegans, the intestine functions as the primary hub for nutrient absorption, lipid storage, and metabolic regulation [52]. Our transcriptome data indicated that genes involved in lipid metabolism are modulated by KTM consumption, prompting us to investigate whether the host transcriptional response to KTMs occurs in the intestine. Using previously established gene expression data for the major tissues, we queried whether each set of diet-induced differentially expressed genes were enriched for a specific tissue [73,74]. We found that in response to KTM consumption there was a striking enrichment for differential expression of intestinal genes, as well as a depletion of neuronal and germline genes (Fig 5D). These data indicate that while genes expressed in the intestine are commonly differentially expressed in animals consuming KTMs, genes expressed in other tissue types tend not to be differentially expressed in KTM-fed animals. To identify candidate genes that may be responsible for the metabolic effects of KTM con- sumption, we analyzed the expression levels of 5,676 genes that are annotated to function in PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 14 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 5. Host lipid metabolism gene expression is modulated by KTM consumption. (A) A Principal Component Analysis of the normalized mRNA-Seq data for the indicated diets (1G, KTM feeding for one generation; 5G, KTM feeding for five consecutive generations prior to collection). PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 15 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism (B) The overlap of the differentially expressed genes, determined relative to E. coli OP50, between each food source. (C) A Gene Ontology enrichment analysis performed on the 295 genes that are uniquely differentially expressed in animals consuming KTMs. (D) Enrichment for differential expression of genes that are expressed in the indicated tissues (observed/expected, hypergeometric P values reported). Values <1 indicate that genes expressed in the indicated tissue type tend not to be differentially expressed (under-enriched), while values >1 indicate tissues where differential expression is more common than expected by random chance (over-enriched). (E) A scatter plot and linear regression (R2 = 0.9556) of the RPMK values for 5,676 metabolism-related genes (the genes of interest are indicated with arrows). (F) A schematic and gene expression heatmap (Log2 fold change values relative to E. coli OP50) for the indicated lipid metabolism genes for each diet (boxes from left to right: KTM, A. tropicalis, KTM-M, E. coli HT115). Raw data underlying panels A-F can be found in S5 Data and S4 Table. https://doi.org/10.1371/journal.pgen.1011003.g005 metabolism [29]. This revealed that several genes known to function in lipid biology have altered expression in KTM-fed animals (Fig 5E and 5F). These included down-regulated genes that act in the β-oxidation of lipids (acdh-1, acdh-2), fatty acid desaturation (fat-5, fat-6, fat-7), or triglyceride synthesis (dgat-2), as well as up-regulated genes that act in lipolysis (lipl-1, lipl- 2, lipl-3). These data suggest that expression of specific lipid metabolism genes in the intestine is modulated by KTM consumption. Consistently, intestinal expression of a GFP-based tran- scriptional reporter for the acdh-1 gene, which encodes a conserved acyl-CoA dehydrogenase that catabolizes short chain fatty acids and branch chained amino acids, was reduced when animals were fed a KTM diet (S8J and S8K Fig). Together, our results suggest that transcrip- tional regulation of metabolic genes may, at least in part, underlie the reduction in intestinal lipids that we observed in KTM-fed animals. KTM consumption restricts lipid accumulation by regulating lipid droplet dynamics Coordination of intestinal lipid stores is governed by both transcriptional and post-translational mechanisms that dynamically alter lipid droplets in response to external signals. Expansion of LDs is carried out via de novo lipogenesis and the action of acyl CoA:diacylglycerol acyltransfer- ase (DGAT) enzymes, which catalyze the final step in TAG synthesis [75,76]. In contrast, lipases and lipophagy, a selective LD autophagy pathway, restrict LD size and number, respectively, and promote lipid catabolism [77–81]. Given that KTM-fed animals display a reduction in lipid levels and lipid droplet size, we reasoned that the expression of triglyceride lipases may be induced in response to KTMs; however, we found that expression of the adipocyte triglyceride lipase gene (atgl-1/ATGL), which encodes a LD-associated and starvation-responsive TAG lipase [52,82], and the hormone-sensitive lipase gene (hosl-1/HSL), which encodes a hormone- responsive TAG lipase, are not altered by KTM feeding (Fig 6A and 6B). We then inspected the expression of the remaining lipase genes within our mRNA-Seq data (Fig 5F), finding that three ATGL-like lipase genes (i.e., lipl-1, lipl-2, and lipl-3) were markedly up-regulated in KTM-fed animals relative to those consuming the E. coli or KTM-M diets (Fig 6C–6E). Interestingly, lipl- 1,2,3 gene expression is known to increase upon fasting and the encoded proteins all localize to the lysosomes in the intestine where they break down LD-associated TAGs via lipophagy [83]. Consistent with these observations, expression of a single-copy Plipl-1::mCherry transcriptional reporter was specifically induced in the intestine in response to KTMs compared to the other food sources (Figs 6F and S9A). Up-regulation of the lipl-1,2,3 lysosomal lipase genes, as well as the concomitant reduction in TAGs, suggests that KTM-fed animals may experience a fasting- like state even in the presence of sufficient nutrient availability. To assess whether lysosomal lipases are required for the host response to KT, we used Oil Red O staining to determine the levels of intestinal lipids in previously generated lipl mutants [83,84]. We found that lipid levels were elevated in lipl-1(tm1954); lipl-2(ttTi14801) double mutants and lipl-1(tm1954); lipl-2(ttTi14801); lipl-3(tm4498) triple mutants relative to wild- type animals upon KTM consumption (Fig 6G). We also generated putative loss-of-function PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 16 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Fig 6. KTM consumption stimulates host lipid catabolism and impairs TAG synthesis. (A-E) Normalized gene expression values (RPKM, reads per kilobase of transcript per million mapped reads; mean ± SEM, *, P<0.0001, , P<0.05, ns, not significant, one-way ANOVA) for the PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 17 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism indicated lipase genes. (F) Representative images of animals expressing a Plipl-1::mCherry transcriptional reporter upon consumption of the indicated diets (white arrow heads point to the intestine; scale bars, 500μm). (G) Quantification of Oil Red O stained intestinal lipids in day 1 adult wild-type N2 and lipl mutant animals after consumption of the KTM-M (left group) or KTM (right group) diets (mean ± SD, **, P<0.0001, , P<0.001, , P<0. 01, ns, not significant, one-way ANOVA). Data are shown for the following mutants: lipl-1(tm1954) lipl-2(ttTi14801) in circles, lipl-1(rhd279) lipl-2(rhd282) in triangles, lipl-1(tm1954) lipl-2(ttTi14801) lipl-3(tm4498) in diamonds, and lipl-1(rhd279) lipl-2(rhd282) lipl-3 (tm4498) in hexagons. (H) Representative images (scale bars, 5 μm) and (I) lipid droplet density measurements (mean ± SD, *, P<0.0001, ns, not significant, one-way ANOVA) of DHS-3::GFP-containing lipid droplets in wild-type N2 and lipl-1(tm1954) lipl-2(ttTi14801) lipl-3(tm4498) mutant animals. (J) Normalized gene expression values for the TAG synthesis gene dgat-2 (mean ± SEM, , P<0.0001, one-way ANOVA). (K) Quantification of Oil Red O staining of intestinal lipids in wild-type N2 and DGAT-2::GFP transgenic animals, which constitutively overexpress DGAT-2 in the intestine (dgat-2 OE; mean ± SD, *, P<0.0001, T-test). (L) A model of KTM modulation of host lipid metabolism pathways showing 1) the induction of the lysosomal lipases that are essential to lipophagy and 2) the down-regulation of the TAG synthesis gene dgat-2 thereby restricting lipid droplet initiation/expansion. Raw data underlying panels A-E, G, and I-K can be found in S6 Data. https://doi.org/10.1371/journal.pgen.1011003.g006 nonsense mutations in the lipl-1 and lipl-2 genes using CRISPR/Cas-9, crossed these alleles to the existing lipl-3(tm4498) mutant [83], and performed Oil Red O staining of the resulting tri- ple mutant. Consistent with our initial observations, simultaneous loss of lipl-1,2 or lipl-1,2,3 increased lipid levels in KTM-fed animals (Fig 6G). Since the LIPL-1,2,3 proteins localize to lysosomes and catabolize LD-associated TAGs, we reasoned that LD size or abundance may be altered in lipl-1,2,3 mutants consuming KTMs. Therefore, we crossed the DHS-3::GFP reporter into the lipl-1(tm1954); lipl-2(ttTi14801); lipl-3(tm4498) triple mutant and measured intestinal LDs. Triple mutant animals fed the KTM diet, but not the KTM-M diet, had more LDs compared to wild-type animals; however, the LD size was similar between wild-type and mutant animals (Figs 6H–6I and S9B), suggesting that the LIPL-1,2,3 proteins promote LD degradation, but not LD shrinking, in KTM-fed animals. Together, these results indicate that up-regulation of the lipl-1,2,3 lysosomal lipases in response to Kombucha Tea consumption partially governs the host metabolic response to KTMs and facilitates lipid catabolism. In addition to induction of lipid catabolism pathways, Kombucha-associated microbes may impair TAG accumulation or LD expansion. To investigate this further using our mRNA-Seq data, we compared the expression levels of genes that are known to function in LD synthesis or expansion for animals fed E. coli OP50, KTMs, or the KTM-M [75,85]. Although levels of seipin (seip-1), lipin (lpin-1), and acs-22/FATP4 (a long-chain fatty acid transporter and acyl-CoA syn- thetase enzyme) were not altered in response to KTM feeding, the dgat-2/DGAT2 gene was dra- matically and specifically down-regulated upon KTM consumption (Figs 6J and S9C–S9E), suggesting that TAG synthesis may be impaired in these animals. To test whether down-regula- tion of dgat-2 restricts lipid accumulation in KTM-fed animals, we employed a strain that expresses dgat-2 under the control of a constitutive intestinal promoter (Pvha-6::GFP::dgat-2), which is not predicted to respond to KTM consumption. Indeed, constitutive expression of dgat- 2 partially suppressed the KTM-dependent depletion of intestinal lipid stores (Fig 6K). Together, these results support a model where the concomitant down-regulation of dgat-2 and up-regula- tion of the lysosomal lipase genes limits TAG synthesis while promoting LD breakdown, which together restricts intestinal lipid accumulation in response to Kombucha consumption (Fig 6L). Discussion The first records of Kombucha Tea consumption can be traced to ancient China where it was incorporated into common medical practices [86]. While its popularity has expanded through- out history, a recent surge in worldwide consumption makes it one of the most popular probi- otic-containing fermented beverages, with its numerous purported human health benefits being a major contributor to its popularity [86]. Despite this long history and widespread anec- dotal evidence that it improves metabolic health [9–13], little is known about whether Kombu- cha Tea consumption alters host metabolism and, if so, by which mechanisms this may occur. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 18 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism To investigate Kombucha Tea’s action in an animal model system, we established a reproduc- ible method to deliver a diet of KT-associated microbes (KTM) to C. elegans though standard ad libitum feeding practices. Delivery of KTMs by feeding supports normal C. elegans develop- ment and fecundity, and importantly, results in robust KTM colonization of the intestinal lumen. Our study is the first to leverage a well-established animal model system to elucidate the molecular mechanisms of Kombucha Tea action in the host. Here, we demonstrate that animals consuming KTMs are markedly devoid of lipids relative to animals fed other microbial diets, as determined by Oil Red O and Nile Red staining, biochemical triglyceride measurements, and size calculations of intestinal lipid droplets. Together, our results suggest that KTM consumption stimulates a fasting-like state in C. elegans that is distinct from tra- ditional models of caloric restriction. Indeed, there are several lines of evidence that argue that KTM-fed animals are not experiencing caloric restriction, including 1) KTM feeding supports an increased rate of development for both wild-type and calorically restricted animals (i.e., eat-2 mutants), 2) KTM-fed animals are fertile (i.e., they exhibit nearly normal brood sizes, reproductive lifespans, and expression of reproduction genes), 3) the individual KT microbes (A. tropicalis, K. rhaeticus, and Z. bailli supplemented with dead E. coli), as well as a simple mixture of the three microbes (KTM-M), fail to deplete host lipid stores, and 4) supplementation of KTMs with addi- tional nutrients, either dead E. coli or higher concentrations of KTMs, did not increase lipid accu- mulation. Importantly, calorically restricted animals have severe growth and fertility defects [28,56–59],phenotypes that are inconsistent with those produced by KTM consumption. Finally, we found that host lipid utilization was maintained after washing the concentrated KT microbes with naïve, sucrose-only media prior to plating, supporting the hypothesis that the bioactive mole- cule(s) responsible for altering host lipid metabolism are intrinsic to the KTM microbes rather than found in the cell-free, fermented tea supernatant. Identification of these KTM-derived metabolites will be crucial to gain insight into the molecular mechanisms of KT action. To gain a comprehensive view of the host metabolic response to Kombucha, we performed mRNA sequencing of animals consuming KTMs. While expression of developmental or repro- duction genes were globally unchanged, expression of numerous lipid metabolism genes were specifically altered in response to KTMs, with a strong enrichment for genes known to func- tion in the intestine. These include gene products that function in various aspects of lipid biol- ogy, including β-oxidation of lipids (acdh-1 and acdh-2), fatty acid desaturation (fat-5 and fat- 7), triglyceride synthesis (dgat-2), and lipolysis (lipl-1, lipl-2, and lipl-3). The stearoyl-CoA desaturase genes, in particular fat-5 and fat-7, were down-regulated in KTM-fed animals. This finding is notable since the C. elegans desaturases have lipid substrate preferences, and thus, differential expression of individual fat genes can result in alterations in the abundance of spe- cific monounsaturated or polyunsaturated fatty acids [87]. FAT-5, which desaturates palmitic acid (16:0) to generate palmitoleic acid (16:1n-7), is transcriptionally down-regulated in KTM- fed animals, possibly resulting in a decrease in palmitoleic acid and increase in palmitic acid or other unsaturated fatty acids that are derived from palmitic acid. Specific changes in the abun- dance of monounsaturated or polyunsaturated fatty acids may contribute to the fasting-like state displayed by KTM-fed animals; however, lipidomic studies, paired with fatty acid supple- mentation experiments and genetic analyses, will be needed to resolve the role of the C. elegans desaturases in mediating the host response to KTM consumption. In this study, we focused on three intestinal ATGL-like lipase genes lipl-1, lipl-2, and lipl-3 that were specifically upregulated in KTM-fed animals, while the other 5 lipl genes, as well as the lipid droplet lipase genes atgl-1 and hosl-1, remained unchanged. These findings argue that Kombucha consumption triggers a specific catabolic response to restrict lipid accumulation. The lipl-1,2,3 genes encode three, likely redundantly acting lysosomal lipases that function in lipophagy-mediated break down of LD-associated TAGs [83]. Here, we demonstrate that the PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 19 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism lipl-1,2,3 genes are partially required for KTM-mediated lipid catabolism, suggesting that lipo- phagy is induced by KTM consumption. Lipophagy, which is a selective form of autophagy that targets lipid droplet TAGs to liberate free fatty acids for further catabolism, is essential for lipid homeostasis and survival in times of low nutrient availability or during states of fasting. In addition to these conditions, homeostatic pathways can dynamically govern lipophagy induction under different nutrient- and stress-related conditions (i.e., fed, fasted, and oxidative stress states) [80]. For example, lipl-3 transcription can be governed by the interplay between the DAF-16/FOXO, PHA-4/FoxA, and HLH-30/TFEB transcription factors in specific con- texts [80]. We propose that KTM consumption stimulates a fasting-like state in C. elegans to promote lipid utilization via lipophagy; however, future studies will be needed to dissect the precise molecular mechanisms that lead to lipophagy induction in response to KTMs. It’s nota- ble that a recent study by Xu and colleagues [19] in rodents lends substantial physiological evi- dence supporting the health claims made regarding human KT consumption, including protection against obesity and Type 2 Diabetes, which are disease states that are commonly associated with impaired lipid utilization or dyslipidemia [81,88–90]. Our discovery that C. ele- gans animals consuming a KTM diet may have elevated levels of lipophagy, and potentially a broader autophagy-driven metabolic reprograming, is consistent with these claims and sug- gests that future studies deconvoluting the host response to Kombucha consumption at the molecular level will provide insight into how Kombucha Tea may alter human metabolism. Our mRNA-Seq data also revealed that dgat-2, which encodes an acyl-CoA:diacylglycerol acyltransferase (DGAT) enzyme, is dramatically down-regulated in response to KTM con- sumption. The DGAT enzymes catalyze the synthesis of triglycerides from diacylglycerol and a fatty acyl-CoA, resulting in TAG production and the expansion of lipid droplets. Constitutive over-expression of dgat-2 increased lipid accumulation in animals consuming KTMs, suggest- ing that down-regulation of dgat-2 expression, and consequently reduced TAG synthesis, may be part of the programmed host response to KT. Notably, induction of dgat-2 in C. elegans sup- ports the expansion of LDs in response to the pathogen Stenotrophomonas maltophilia [91], suggesting that dgat-2 expression may be dynamically regulated by nutrient sensing or innate immunity pathways to govern lipid storage levels within LDs. It is possible that dgat-2 expres- sion is controlled by the same signaling networks that control the expression of the LIPL-1,2,3 lysosomal lipases, which together restrict the accumulation of lipids during KTM feeding. This could also explain why loss of lipl-1,2,3 increases LD abundance but not size, as dgat-2 likely remains down-regulated in these animals following KTM consumption. Recently, it has become increasingly evident that C. elegans offers a powerful system to investigate potential human probiotic microbes to gain insight into their mechanisms of action and to identify potential human health benefits [23,25,26,31,92–94]. Our study establishes a rigorous, reproducible, and widely applicable system that leverages the genetic tractability of C. elegans to interrogate the physiological and mechanistic host response to probiotic microbes. While this is an exciting proposition, it is imperative to note that this work, as with other studies conducted using C. elegans as a model to investigate human-probiotic interac- tions, is not directly translatable to human health outcomes and offers no clinical advice or context for human Kombucha Tea consumption. We also acknowledge that the origin of this now popular fermented beverage has deep roots in ancient Chinese medical practices and was created by a culture different from our own. Therefore, we want to make it explicitly clear that we are not making judgements, conclusions or claims regarding Kombucha Tea’s use in any human medical practices or its recreational consumption. Our findings do, however, offer exciting insights into possible mechanisms of KT microbe-mediated host metabolic repro- gramming and lays the foundation for future studies in mammalian model systems that could deconvolute the biological underpinnings of Kombucha Tea’s potential health benefits. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 20 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Materials and methods C. elegans strains and maintenance All Caenorhabditis strains were maintained at 20˚C on Nematode Growth Media (NGM) agar plates containing E. coli OP50 as previously described [95]. A full list of the strains used in this study are shown in S3 Table. All C. elegans strains were well-fed for at least three generations before use in experiments. Unless otherwise stated, eggs were harvested from gravid adults reared on E. coli OP50 by bleaching and animals were synchronized to the L1 stage by incubat- ing the eggs overnight at room temperature. Preparation of L1 animals by bleaching was required to prevent E. coli contamination of Kombucha NGM plates in the following generation. The Plipl-1::mCherry transgenic strain was constructed using Mos1-mediated Single Copy Insertion (mosSCI). The lipl-1 promoter (1,228 bp; chromosome V: 12,918,779–12,920,006; WS288) was amplified by PCR and fused to mCherry::unc-54 3’UTR in pCFJ151 by Gibson Assembly. The resulting plasmid, pRD172[Plipl-1::mCherry::unc-54 3’UTR + cb-unc-119(+)], was microinjected into EG6699 to isolate the single-copy integrant rhdSi53[Plipl-1::mCherry:: unc-54 3’UTR + cb-unc-119(+)] as previously described [96]. The lipl-1(rhd279[A391]) and lipl-2(rhd282[A423]) nonsense alleles were generated by CRISPR/Cas9 gene editing. Briefly, single-stranded oligonucleotide HR donor molecules and the Cas9::crRNA:tracrRNA com- plexes (crRNA sequence: 5’-UAGAGAACUUCUACUCAAAA-3’) were microinjected into the germline of wild-type animals as previously described [97]. The HR donor sequence included a new XbaI cut site which allowed for genotyping via PCR followed by restriction digest prior to Sanger sequencing. The transcriptional reporter strains rhdSi53[Plipl-1:: mCherry::unc-54 3’UTR + cb-unc-119(+)] and wwIs24[Pacdh-1::GFP + cb-unc-119(+)] were imaged with a Nikon SMZ-18 stereo microscope equipped with a DS-Qi2 monochrome cam- era at 10X zoom. Kombucha brewing Kombucha was brewed using a serial fermentation method adapted from a homebrewing Kombucha kit (The Kombucha Shop). Ultrapure water (1L) was boiled for 3 minutes, removed from the heat, and dried tea leaves (2.5 g of Assam Black Tea and 2.5 g Green Tea) were steeped for 5 minutes using an infuser. After removal of the tea infuser, 128 g of granulated cane sugar (Domino) was dissolved in the tea and the solution was poured into a clean 5L glass brewing jar before 3L of chilled ultrapure water was added. Once the solution cooled to below 30˚C, the SCOBY and ~500 mL of the previous fermented Kombucha broth was added to the brew jar and a tight weave muslin cloth was affixed to the jar opening to limit contamination during fermentation. The jar was then placed in indirect sunlight at room temperature (between 24–28˚C) and allowed to ferment for a minimum of 8 days before a new culture was started, which allowed for complete fermentation and a pH of ~4. NGM Kombucha plates For the single microbial diets, NGM plates were seeded with either E. coli strains (OP50 or HT115) after ~16 hours of growth at 25˚C or with A. tropicalis, K. rhaeticus, or Z. bailii grown for at least 3 days at 25˚C. The E. coli strains were grown in 25 mL of LB with shaking (250 rpm) while A. tropicalis, K. rhaeticus, and Z. bailii were grown in 25 mL of mannitol growth media (5 g Yeast Extract, 3 g Peptone, and 25 g Mannitol in 1 L) supplemented with 1% D-glu- cose and 1% glycerol with shaking (250 rpm). The strains were concentrated via centrifugation at 4,000 rcf for 5 minutes followed by resuspension in 5 mL of the appropriate media before PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 21 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism seeding on NGM plates. To calculate the microbial concentration of food sources, OD600 readings were taken followed by serial dilution and CFU quantification. To prepare KTM NGM agar plates, 50 mL of the Kombucha Tea culture on day 2 or 3 of fermentation was removed and the microbes were concentrated via centrifugation for 5 min- utes at 4,000 rcf. The supernatant was removed leaving 5 mL to resuspend the pelleted KTMs. Following resuspension via vortexing, 300 μL or 2 mL of concentrated KT was added to the middle of a 6cm or 10cm NMG plate, respectively. For 5x KTM plates, 250 mL of culture was concentrated to 5 mL. Plates were allowed to mature for 4 days at room temperature before being used in experiments. The KTM-M NGM plates were prepared by first individually grow- ing up 20 mL cultures of A. tropicalis, K. rhaeticus, and Z. bailii. The microbes were then con- centrated by centrifugation, resuspended with filter sterilized tea media (2.5 g of Assam Black Tea, 2.5 g Green Tea, 128 g of granulated cane sugar, 1L of water), combined into a single cul- ture, washed with sterilized tea media, reconcentrated by centrifugation, resuspended in 5 mL of the supernatant, seeded onto NGM plates, and incubated for 4 days at room temperature. To confirm that the filter sterilized tea media was free from microbes and/or spores, the steril- ized tea media was plated on NGM plates and monitored for growth over 14 days, which resulted in plates free of microbial growth. Similarly, microbes were grown independently, harvested, and combined in sterilized tea media to generate the small-scale KTM-FM cultures, which were maintained in 50 mL conical tubes with loosely tapped lids at room temperature. At different timepoints, 30 mL was removed from the culture and replaced with 30 mL of fresh sterilized tea media. The following day 25 mL was removed from the culture, concentrated by centrifugation, seeded onto NGM plates, and incubated at room temperature for 4 days prior to use. A long-term, established KTM-FM cul- ture was started in a similar fashion, but the culture was fermented in a 500 mL graduated cylin- der covered in a cheese cloth, which was serially fermented over time by removing 50 mL of fermented culture (used for plates) prior to the addition of 50 mL of fresh sterilized tea media. 16S rDNA sequencing of Kombucha culture and plates The Kombucha Tea culture was initiated and KTMs were seeded onto plates as described above. For the day 1 culture timepoint, 1 mL of 10x concentrated Kombucha was subjected to further centrifugation at 16,000 krcf for 10 minutes, the supernatant was removed, and the pel- let was flash frozen in liquid nitrogen. The KTM plates were prepared for 16S sequencing at the same time using 10x concentrated Kombucha. For the subsequent culture sampling, 10 mL of KT was collected and the KTMs were harvested by centrifugation. For KTM plate samples, the microbes were removed from NGM plates at different timepoints using a cell scrapper and were collected into 1 mL of UltraPure DNase/RNase free water, concentrated by centrifuga- tion, and frozen. All 16S rDNA sequencing was performed by the UNC Microbiome Core on an Illumina MiSeq instrument (PE 250). The data analysis was performed on 32,000–95,000 raw reads per sample using Qiime2 [98]. Lawn avoidance assay Approximately 50 synchronized L1 animals were dropped outside of each microbial lawn and the number of animals on each lawn was counted at 48, 72, and 96 hours later. The proportion of animals off the lawn was calculated as Noff lawn/Ntotal for each timepoint. Each biological rep- licate was averaged from three technical replicates and the data were plotted as the mean ± SEM using Prism 9. An ordinary one-way ANOVA followed by Sidak’s multiple com- parisons test was used to calculate statistical significance between groups. PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 22 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism Food choice assay NGM plates were seeded in four quadrants, each with 30 μL of one of the four food sources (E. coli OP50, HT115, A. tropicalis, and KTMs). Approximately 50 synchronized L1 animals were dropped in the middle of the plate and the fraction of animals on the different microbial lawns was counted 48 hours later. Pumping rate measurements Animals were grown on each diet from synchronized L1s and the pumping rate of 15 day one adult animals was manually counted using a Nikon SMZ800N Stereo microscope. The number of pharyngeal contractions over a one-minute span was counted and data were plotted as the mean ± SD using Prism 9. A one-way ANOVA followed by Tukey’s multiple comparisons test was used to calculate statistical significance between groups. Gut colonization assay Measurement of the bacterial loads in C. elegans animals after consumption of each diet was per- formed as previously described [99]. Briefly, ~150 animals were grown from synchronized L1s on each diet to adulthood and ~30 animals were picked to an empty plate for 30 minutes to minimize bacterial transfer from lawn. Ten animals were hand-picked to M9 media containing 100 μg/mL levamisole, allowed to settle, and were washed three times with M9 media containing levamisole and gentamicin (100 μg/mL). Animals were lysed in 250 μL 1% Triton X-100 using thirty 1.5 mm sterile zirconium oxide beads (Next Advance) with an electric benchtop homogenizer (BioSpec Mini-beadbeater). The 1.5 mL tubes were shaken twice for 90 sec before serial dilution of the lysates and plating onto standard NGM plates. The CFUs/animal values were calculated as described [99]. Data were plotted using Prism 9 and the statistical significance between food sources was determined by one-way ANOVA followed by Sidak’s multiple comparisons test. SEM imaging of the C. elegans Intestine Day 1 adult animals were fixed with 2% paraformaldehyde in 150 mM sodium phosphate buffer (PB, pH 7.4) at room temperature and stored at 4˚C. Samples were washed 3 times with PB, followed by 3 water rinses, dehydrated using an ethanol gradient (30%, 50%, 75%, 100%, 100%, 100%), washed with two hexamethyldisilazane (HMDS) exchanges, and allowed to dry in HMDS. Dried animals were brushed onto double-sided carbon adhesive mounted to a 13 mm aluminum stub and a scalpel was used to slice the C. elegans animals open by drawing the blade upward though the body of the animal while they were adhered to the adhesive. Mounted samples were then sputter coated with 5 nm of gold-palladium (60 Au:40 Pd, Cres- sington Sputter Coater 208HR, model 8000–220, Ted Pella Inc). Images were taken using a Zeiss Supra 25 FESEM operating at 5 kV, using the InLens detector, ~7 mm working distance, and 30 μm aperture (Carl Zeiss SMT Inc) at 5,000X and 15,000X zoom. Calcofluor White Staining Calcofluor White (or Fluorescent Brightener #28), which stains chitin and cellulose, was added to levamisole paralyzing solution at approximately 1 mg/mL. The animals fed different diets were picked to agar pads containing levamisole and Calcofluor White, covered with a cover- slip, and stained for 10 minutes. For K. rhaeticus and E. coli OP50 imaging (S6J Fig), animals were imaged with a Leica DMI8 with an xLIGHT V3 confocal microscope with a spinning disk head (89 North) equipped with a Hamamatsu ORCAFusion GENIII sCMOS camera using a 63X oil objective (Plan-Apochromat, 1.4 NA). For imaging of the KTMs (S6L Fig), PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 23 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism animals were imaged with a Ti2 widefield microscope equipped with a Hamamatsu ORCA-- Fusion BT camara using a 100X oil objective (Plan Apo λ). Importantly, the stain was prone to rapid photobleaching, and thus, areas of interest were found using the DIC channel and ani- mals were only exposed to the florescent light during image acquisition. Nikon Elements was used to denoise and deconvolute the KTM images and both sets of images were processed in Fiji v2.9.0 [100] to introduce pseudo coloring. Oil Red O and Nile Red staining Approximately 150 animals were grown from synchronized L1s on NGM plates containing different food sources for 72 hours at 20˚C. Day 1 adult animals were washed off the plates in M9 media, allowed to settle on ice, washed three times with S-basal media, and fixed in 60% isopropanol. For Oil Red O staining, fixed animals were treated with filtered 0.5% Oil Red O for 7 hours before washing the animals with 0.01% Triton X-100 in S-basal as previously described [101]. For Nile Red staining, isopropanol-fixed animals were stained for 2 hours with fresh Nile Red/isopropanol solution (150 μL Nile Red stock at 0.5 mg/mL per 1 mL of 40% isopropanol) [29]. For whole body analyses, animals were mounted on agar pads and imaged for Oil Red O staining at 3X zoom with a Nikon SMZ-18 Stereo microscope equipped with a DS-Qi2 monochrome camera or for Nile Red staining at 4X zoom using a Ti2 widefield microscope equipped with a Hamamatsu ORCA-Fusion BT camera. Color images of Oil Red O-stained animals were obtained at 10X magnification using the Ti2 widefield microscope equipped with a Nikon DS-FI3 color camara. For analysis of intestinal Oil Red O staining (Fig 6G and 6K), animals were imaged at 10x with the Ti2 widefield microscope equipped with a Hamamatsu ORCA-Fusion BT camera. For quantification of Oil Red O staining, whole animals were outlined using Fiji and the average gray value (0 to 65,536) for each individual was measured. The resulting values were subtracted from 65,536, which inverts the scale so that strongly stained animals now have higher values. True background values are the unstained regions within each animal; however, these regions are impossible to identify objectively. Thus, no background subtraction was per- formed, which compresses the data to a small range of values (55,000 to 65,000, which we report 5.5 to 6.5). We found this approach to be highly reproducible. For the intestinal Oil Red O staining analyses (Fig 6G and 6K), the mean gray values were calculated in a box (25x25 pix- els) drawn within the first two intestinal cells and the analysis was performed as described above. For Nile Red staining, average fluorescence intensities were also measured using Fiji and no background subtraction was performed. All data were plotted using Prism 9 as the mean ± SD and a one-way ANOVA followed by Tukey’s multiple comparisons statistical test was performed for each experiment. For each set of staining experiments, at least three biologi- cal replicates were performed and yielded similar results. Quantification of Lipid Droplets Live day 1 adult ldrIs1[Pdhs-3::dhs-3::GFP] animals were mounted on agar pads with levami- sole and imaged with a Ti2 widefield microscope equipped with a Hamamatsu ORCA-Fusion BT camara using a 63X oil objective. Bright field DIC and GFP images capturing the last two intestinal cell pairs were imaged in 0.2 μm slices using the same settings across samples. Fol- lowing acquisition, Nikon Elements was used to select a representative slice in the middle of the stack for downstream analysis using Fiji. The DIC image was used to outline the intestinal cell pair and the diameter of the lipid droplets were measured by hand in the GFP channel using the line tool and ROI manager. Measurements were collected from the last intestinal cell pair for 10 representative animals for each food source and the average lipid droplet PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 24 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism measurements for each animal was plotted using Prism 9. A one-way ANOVA followed by Tukey’s multiple comparisons test was used to compare groups. Two independent experi- ments were conducted with similar results. Biochemical triglyceride measurements Day 1 adult animals consuming each food source were harvested and processed as previously described [55]. A Triglyceride Assay Kit was used to measure triglycerides per the manufac- tures’ instructions (Abcam, ab65336). Three biological replicates were performed, the data were plotted in Prism 9, and a one-way ANOVA followed by Tukey’s multiple comparisons test was carried out to compare groups. Molting assays Between 1–8 synchronized mgIs49[Pmlt-10::GFP::PEST]L1s were dropped into each well of a 24-well plate containing NGM media and seeded with 20 μL of each food source (one plate per food source). Animals were reared at 20˚C and visualized by fluorescence microscopy every hour for 70 hours on a Nikon SMZ-18 Stereo microscope. At each time point, animals were scored as green (molting) or nongreen (not molting). Wells without animals were censored. The fraction of animals molting for each timepoint was calculated and plotted with Prism 9. The experiments were performed at least twice (except for KTM-M and KTM-FM) with similar results. Developmental timing measurements Wild-type N2 and eat-2(ad465) mutants were grown to adulthood and egg prepped as described above. The caloric restriction plates containing 108 or 109 CFUs/mL of E. coli OP50 were prepared as previously described [59]. Approximately 20 synchronized L1s were dropped on each food source in technical triplicates and grown at 20˚C. After exactly 48 hours the ani- mals were scored based on vulva morphology as young adults, L4 larval stage, or less than L4 larval stage. The percent of animals at each larval stage was calculated for three biological repli- cates and the data were plotted as the mean ± SEM using Prism 9. Brood size measurements Animals were grown on their respective food sources for 48 hours at which time 15 L4s were singled to the corresponding food source and allowed to mature. The animals were moved to fresh plates every 24 hours for 6 days. Two days after the adult hermaphrodite was moved to a new plate, the L3/L4 progeny were counted and removed. The unhatched eggs were not counted. Total progeny for each individual hermaphrodite was plotted as mean ± SD using Prism 9 and a one-way ANOVA test with a Tukey’s multiple comparison correction was per- formed. The average reproductive output per day was also calculated and an unpaired T-test was performed to identify differences between these means. VIT-2::GFP quantification Animals expressing VIT-2::GFP at endogenous levels (strain BCN9071) were grown to adult- hood, egg prepped, and hatched over night at room temperature as described above. The starved L1s were dropped on NGM plates seeded with their respective food sources and grown for 72 hours at 20˚C. Gravid day 1 adult animals were washed off the NGM plates and eggs were liberated by bleaching. Following three washes with M9 media, embryos were mounted on agar pads and imaged with a Ti2 widefield microscope equipped with a Hamamatsu ORCA-Fusion BT camara using a 20X objective. Bright field DIC and GFP images were PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 25 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism captured and the fluorescent intensity of 30 early-stage embryos (prior to the 44-cell stage) for each condition was measured using Fiji. The mean fluorescent intensity ± SD was plotted using Prism 10 and statistical differences between the groups was calculated using an unpaired T-test. Three independent biological replicates were performed and yielded similar results. Intestinal lumen measurements Day 1 adult ERM-1::GFP animals (strain BOX213) consuming each diet were mounted on agar pads with levamisole and imaged using a Nikon SMZ-18 Stereo microscope equipped with a DS-Qi2 monochrome camera. Bight field and GFP images were acquired for at least 10 individ- uals. Three measurements of the intestinal lumen diameter (positioned at the anterior intestine, the vulva, and the posterior intestine) were performed using Fiji. For each individual, a ratio of the lumen width relative to body width was calculated at each of the three positions along the animal and the three values were averaged. Ten individuals were measured for each biological replicate and the data are reported as the mean ± SEM of three biological replicates. A one-way ANOVA test with a Tukey’s multiple comparison correction was performed in Prism 10. BODIPY staining C1-BODIPY-C12 (Thermo Fisher, D3823) was resuspended in DMSO to generate a 10 mM stock solution. This solution was diluted in S-basal media and overlaid onto the microbial lawn to produce a final concentration of 10 μM within the NGM plate. The plates were allowed to dry for at least 1 hour in the dark before day 1 adult animals were picked to the BODIPY plates. After 3 hours, animals were mounted on agar pads with levamisole and imaged on a Ti2 widefield microscope equipped with a Hamamatsu ORCA-Fusion BT camara using a 40X oil objective. Bright field DIC and GFP images were captured for at least 20 animals at both the anterior and posterior sections of the intestine. Representative images showing detectible levels of BODIPY staining were selected from two independent experiments and are displayed in Figs 4K and S7. Whole genome sequencing and analysis of Kombucha microbes Mannitol growth media supplemented with 1% D-glucose and 1% glycerol was inoculated with the KT associated microbes and the cultures were grown for 48 hrs at 25˚C with shaking. The gDNA was isolated from cell pellets using the Wizard Genomic DNA Purification kit (Promega, A1120). Preparation and Illumina short read sequencing (PE 150) of DNA-Seq libraries was performed by Novogene (Sacramento, CA). Initially, an unbiased metagenomic analysis was performed using Kraken 2 [65] to identify candidate microbial species for each Kombucha-associated microbe. Next, we downloaded the whole genome reference sequences for various strains for each candidate species from the NCBI Genome database and mapped our reads against those reference genomes using Bowtie 2 with the default settings [102]. The overall alignment rate generated by the Bowtie 2 algorithm was reported. The whole genome sequencing data are available at the Sequencing Read Archive (PRJNA1044129). mRNA sequencing Wild-type N2 animals were grown on 10 cm NGM agarose plates (1000 animals/plate) in the presence of their respective food sources. Day 1 adults were harvested, washed 3 times in M9 buffer, and flash frozen. The total RNA was isolated using Trizol (Thermo Fisher), followed by two rounds of chloroform extraction, RNA precipitation with isopropanol, and an 80% etha- nol wash of the RNA pellet. In some cases, an RNA Clean & Concentrator-25 kit (Zymo, PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 26 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism R1017) was used to increase the purity of the sample. The mRNA-Seq libraries were prepared and sequenced by Novogene (Illumina, PE 150). The data were processed exactly as previously described [103]. RPKM values and identification of differentially expressed genes (1% FDR) were calculated using the DESeq2 algorithm [104], which can be found in S4 Table. Lists of developmental, reproduction, and metabolic genes have been previously described [29], and scatter plots showing expression levels of these genes were generated using the DESeq2 RPKM values. Heatmaps and PCA plots were generated with the pheatmap [105] and tidyverse [106] R packages, respectively. All other plots showing mRNA-Seq data were made in Prism 9. Raw and processed mRNA-Seq data have been deposited in GEO (GSE236037). Supporting information S1 Fig. The phylogenetic profile of the KTMs on NGM plates is similar to that of the KT cul- ture. (A) Images of NGM worm plates seeded with a KTM lawn. The preparation starts at day 0 when a new KT brew cycle is initiated, the microbes are seeded on day 1, and incubated at room temperature to day 5 before the KTM plates are used. (B) Representative photos of KT brews at day 1 and day 7 of fermentation. The KTMs are extracted from the culture at day 1 and plated. (C) A comprehensive view of 16S rDNA sequencing results of the KT microbes from fermenting Kom- bucha culture, seeded NGM plates, or the pellicular biofilm from the Kombucha culture. The plot shows the frequency of each species (8 most abundant microbes displayed; a complete list can be found in S1 Table). (D) A plot of Faith’s phylogenetic diversity index showing the difference in α- diversity between the indicated samples (, p<0.01, one-way ANOVA). (E) The Pielou Evenness Diversity Index, measuring the microbial diversity and species richness in the indicated samples (ns, not significant, one-way ANOVA). Raw data underlying panels C-E can be found in S7 Data. (TIF) S2 Fig. Worms choose other diets over a KTM diet. (A) A schematic depicting the food choice assay. (B) The portion of wild-type N2 animals at the L4 stage on each food source 48 hours after dropping L1s (n>200/trial, 3 biological replicates). (C-E) Food choice assays for the N2, MY10, and JU1212 C. elegans strains scored at the L4 stage (48h post L1 drop, n>150/ trial, 3 biological replicates). (F-I) The portion of L4 stage worms on each food source at 48h post L1 drop for the N2 C. elegans, PB2801 C. brenneri, AF16 C. briggsae, and PB4641 C. rema- nei strains (n>75/trial, 3 biological replicates). All food choice data are plotted as the mean ± SEM. All food choice assays include n>150 animals per replicate and the data are plot- ted as the mean ± SEM (*, P<0.0001, , P<0.001, , P<0.01, , P<0.05, ns, not signifi- cant; one-way ANOVA). Raw data underlying panels B-I can be found in S8 Data. (TIF) S3 Fig. Host lipid distributions during reproduction and across individuals. (A) Quantifi- cation of day 3 and day 5 adults stained with Oil Red O (mean ± SD, **, P<0.0001, one-way ANOVA). (B) Measurements of individual lipid droplet sizes measured across ten individuals consuming E. coli OP50, KTMs, or KTM-Mix (mean ± SD, n = 10 animals/trial, 2 biological replicates). The distribution of lipid droplet sizes is similar across individuals fed the same diet. Raw data underlying panels A and B can be found in S9 Data. (TIF) S4 Fig. Average progeny per day. A table displaying the average progeny laid per day of the reproductive period demonstrates that KTM-fed animals exhibit a similar egg laying rate rela- tive to E. coli OP50-fed animals (mean, , P<0.0001, , P<0.001, , P<0. 01, , P<0.05, ns, not significant, T-test). Raw data underlying the figure can be found in S10 Data. (TIF) PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 27 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism S5 Fig. rDNA sequencing identifies candidate KTMs. Results from 16S rDNA sequencing of the isolated bacterial KTMs indicate that (A) A. tropicalis and (C) a member of the Komaga- taeibacter genus are components of our Kombucha culture. (B) Sequencing of the ITS region of the KTM yeast isolate revealed that the strain belongs to the Brettanomyces or Zygosaccharo- myces genus. Raw data underlying panels A-C can be found in S1 Table. (TIF) S6 Fig. Deconvolution of Kombucha Tea facilitates the creation of fermenting and non-fer- menting mixes of KTM. (A-B) Measurements of the microbial concentrations in each of the indicated microbial mixes or single microbial cultures (mean ± SEM, **, P<0.0001, one-way ANOVA). (C) Oil Red O staining of day 1 adult animals fed an E. coli OP50, KTM-Mix, or KTMs diet, as well as a 5X concentrated version of the KTM diet (mean ± SD, , P<0.0001, one-way ANOVA). Increasing the concentration of KTMs decreases lipid storage. (D) Represen- tative images of animals off and on a lawn of Z. bailii yeast 72 hours post L1 drop, which shows that animals fail to develop when consuming a Z. bailii diet (worms are indicated with white arrow heads; scale bar, 500 μm). (E) Lipid droplet density measurements with each datapoint rep- resenting the number of lipid droplets per μm2 for the last two intestinal cells of animals consum- ing a KTM or KTM-M diet (the KTM data are also shown in Fig 2H; mean ± SD, , P<0.01, T- test). (F) A scatter plot comparing the expression of 2,229 developmental genes in animals fed E. coli versus KTM-M as determined by mRNA-Seq (RPKM, reads per kilobase of transcript per million mapped reads). A linear regression analysis and the corresponding R2 value is reported. (G) A choice assay showing the portion of wild-type N2 animals at the L4 stage on the indicated food sources 48 hours after dropping L1s (n>200/trial, 3 biological replicates; mean ± SEM, ***, P<0.0001, , P<0.05, ns, not significant, one-way ANOVA). (H-I) The developmental rate of ani- mals expressing a Pmlt-10::GFP-PEST reporter when fed a KTM, KTM-Mix, or a KTM-FM diet. Synchronized L1 worms were reared at 20˚C for ~72 hours and scored hourly. (J) Representative images of animals consuming K. rhaeticus or E. coli OP50 after staining with Calcofluor White, which selectively labels intestinal microbes producing chitin or cellulose (white arrow heads indi- cate the intestinal lumen; scale bars, 10 μm). (K) Representative brightfield DIC images showing yeast cells in the intestine of animals consuming KTMs and yeast cells on the slide (gray arrow heads indicate yeast cells, magnified inset image shown for clarity, scale bar 5 μm). (L) Represen- tative images of animals consuming KTMs after staining with Calcofluor White (gray arrow heads in the inset indicate yeast cells; scale bars, 5 μm). (M) Intestinal lumen width measurements of animals consuming the E. coli OP50, KTM, and KTM-M diets. Data are reported as the percent of the body width taken up by the intestinal lumen (mean± SEM, , P<0.01, ns, not significant, one-way ANOVA). Raw data underlying panels A-C, E-I, and M can be found in S11 Data. (TIF) S7 Fig. BODIPY lipids are absorbed into intestinal cells of KTM-fed animals. Representa- tive DIC and fluorescence images showing C1-BODIPY-C12 absorption into the intestinal epi- thelial cells of animals feeding on an E. coli OP50, KTM, or KTM-M diet. The pink stars indicate BODIPY remaining in the intestinal lumen, the pink arrowheads point to partial BODIPY absorption into the intestinal cells, white stars indicate a lack of BODIPY remaining in the intestinal lumen, and white arrowheads point to fully stained cells that have absorbed BODIPY (scale bars, 10 μm). (TIF) S8 Fig. KTM consumption results in widespread changes in gene expression. (A) A scatter plot and linear regression analysis comparing the expression of all genes in animals fed KTMs for one PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 28 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism generation (1G) or for five generations (5G), suggesting that pervasive transgenerational epige- netic regulation of gene expression by KTMs is unlikely. (B-F) Volcano plots showing the differ- entially expressed genes for the indicated samples relative to the E. coli OP50 sample. (G) Enrichment (observed/expected, hypergeometric P values reported) for differentially expressed genes common between KTM-fed animals and animals depleted of DAF-2::AID in the indicated tissues using the auxin degron system [72]. Values >1 indicate over-enrichment, or that the same genes tend to be differently expressed in both animals consuming KTMs and animals depleted of DAF-2 compared to random chance. The overlap between differentially expressed genes that are either (H) up-regulated or (I) down-regulated in animals consuming KTMs and animals depleted DAF-2::AID in the intestine (hypergeometric P values are shown). (J) Representative fluorescent images (scale bar, 500 μm) and (K) quantification of the acyl-CoA dehydrogenase Pacdh-1::GFP reporter on the indicated microbial diets (n = 40, mean ± SD, , P<0.0001, ns, not significant, one-way ANOVA). Raw data underlying panels A-I and K can be found in S12 Data. (TIF) S9 Fig. Expression of the lipl-1 gene is modulated in the intestine upon KTM consumption, but the lysosomal lipases genes lipl-1,2,3 are not required to restrict lipid droplet size. (A) Quantification of the expression levels of the lysosomal lipase Plipl-1::mCherry reporter in ani- mals grown on E. coli OP50, KTM, and KTM-M (n>200, mean ± SD, , P<0.0001, one- way ANOVA). (B) Lipid droplet size measurements in wild-type N2 and lipl-1(tm1954) lipl-2 (ttTi14801) lipl-3(tm4498) mutant animals with each datapoint representing the average intes- tinal lipid droplet diameter for a single animal (mean ± SD, , P<0.001, ns, not significant, one-way ANOVA). (C-E) Normalized gene expression values for the indicated TAG synthesis genes (mean ± SEM, *, P<0.001, , P<0.05, ns, not significant, one-way ANOVA). Raw data underlying panels A-E can be found in S13 Data. (TIF) S1 Table. An Excel spreadsheet containing the taxonomy report from 16S rDNA sequenc- ing. Shown are individual sequencing results for biological replicates of the Kombucha Tea cultures, the Kombucha Tea biofilm (one replicate), and Kombucha Tea-associated microbes isolated from C. elegans NGM plates. (XLSX) S2 Table. Sequencing read alignment rates from whole genome sequencing of Kombucha Tea-associated microbes. (PDF) S3 Table. The C. elegans strains used in this study. The strain names, genotypes, and associ- ated references are shown. (PDF) S4 Table. The DESeq2 outputs from the mRNA-Seq analysis. An Excel spreadsheet contain- ing, in separate tabs, gene counts (RPKM, reads per kilobase of transcript per million mapped reads) for all genes, as well as the differential gene expression calls for the following compari- sons: E. coli OP50 vs. E. coli HT115, E. coli OP50 vs. Acetobacter tropicalis, E. coli OP50 vs. KTM, and E. coli OP50 vs. KTM-M. (XLSX) S1 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 1D, 1E and 1F. (XLSX) PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 29 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism S2 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 2B, 2D, 2E, 2G and 2H. (XLSX) S3 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 3A, 3B, 3C, 3D, 3E, 3F, 3G, 3H, 3I, 3J, 3K, 3L, 3M and 3N. (XLSX) S4 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 4C, 4E, 4F, 4G, 4I and 4J. (XLSX) S5 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 5A, 5B, 5C, 5D, 5E and 5F. (XLSX) S6 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying Fig 6A, 6B, 6C, 6D, 6E, 6G, 6I, 6J and 6K. (XLSX) S7 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S1C, S1D and S1E Fig. (XLSX) S8 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S2B, S2C, S2D, S2E, S2F, S2G, S2H and S2I Fig. (XLSX) S9 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S3A and S3B Fig. (XLSX) S10 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S4 Fig. (XLSX) S11 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S6A, S6B, S6C, S6E, S6F, S6G, S6H, S6I and S6M Fig. (XLSX) S12 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S8A, S8B, S8C, S8D, S8E, S8F, S8G, S8H, S8I and S8K Fig. (XLSX) S13 Data. Excel spreadsheet containing, in separate tabs, the numerical data underlying S9A, S9B, S9C, S9D and S9E Fig. (XLSX) Acknowledgments Some of the strains used in this study were provided by the Caenorhabditis Genetics Center, which is supported by the NIH Office of Research Infrastructure Programs (P40 OD010440). The lipl-3(tm4498) lipl-2(ttTi14801) lipl-1(tm1954) strain was generously provided by Dr. Eyleen O’Rourke (UVA). We want to thank Dr. Anne Matthysse (UNC) for her helpful sug- gestions regarding the isolation of KT-associated microbes and for some of the reagents used PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 30 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism in this study, including the Calcofluor White stain and driselase. The authors would also like to thank Kristen K. White and the Microscopy Services Laboratory for their assistance with SEM preparation and imaging. The Microscopy Services Laboratory in the Department of Pathology and Laboratory Medicine at UNC is supported in part by the P30 CA016086 Cancer Center Core Support Grant to the UNC Lineberger Comprehensive Cancer Center. The 16S rDNA sequencing was performed by the UNC Microbiome Core, which is overseen by the director Dr. Andrea Azcarate-Peril and is supported by the Center for Gastrointestinal Biology and Disease (CGIBD P30 DK034987) and the UNC Nutrition Obesity Research Center (NORC P30 DK056350). We would like to thank Monica Macharios for assisting with the Oil Red O analyses, Sarah Torzone for assisting with Oil Red O image acquisition, and Peter Breen and Sarah Torzone for assisting with data collection for the developmental timing experi- ments. Finally, we would like to thank Dr. Mark Peifer (UNC) and Dr. Bob Duronio (UNC) for critical reading of the manuscript. Author Contributions Conceptualization: Rachel N. DuMez-Kornegay, Robert H. Dowen. Data curation: Rachel N. DuMez-Kornegay, Lillian S. Baker, Alexis J. Morris, Whitney L. M. DeLoach. Formal analysis: Rachel N. DuMez-Kornegay, Lillian S. Baker, Whitney L. M. DeLoach, Rob- ert H. Dowen. Funding acquisition: Rachel N. DuMez-Kornegay, Robert H. Dowen. Investigation: Rachel N. DuMez-Kornegay, Robert H. Dowen. Project administration: Robert H. Dowen. Supervision: Rachel N. DuMez-Kornegay, Robert H. Dowen. Writing – original draft: Rachel N. DuMez-Kornegay. Writing – review & editing: Rachel N. DuMez-Kornegay, Robert H. Dowen. References 1. DeGruttola AK, Low D, Mizoguchi A, Mizoguchi E. Current Understanding of Dysbiosis in Disease in Human and Animal Models: Inflamm Bowel Dis. 2016; 22: 1137–1150. https://doi.org/10.1097/MIB. 0000000000000750 PMID: 27070911 2. Levy M, Kolodziejczyk AA, Thaiss CA, Elinav E. Dysbiosis and the immune system. Nat Rev Immunol. 2017; 17: 219–232. https://doi.org/10.1038/nri.2017.7 PMID: 28260787 3. Wang P-X, Deng X-R, Zhang C-H, Yuan H-J. Gut microbiota and metabolic syndrome. Chin Med J (Engl). 2020; 133: 808–816. https://doi.org/10.1097/CM9.0000000000000696 PMID: 32106124 4. Smits WK, Lyras D, Lacy DB, Wilcox MH, Kuijper EJ. Clostridium difficile infection. Nat Rev Dis Primer. 2016; 2: 16020. https://doi.org/10.1038/nrdp.2016.20 PMID: 27158839 5. Tsai Y-L, Lin T-L, Chang C-J, Wu T-R, Lai W-F, Lu C-C, et al. Probiotics, prebiotics and amelioration of diseases. J Biomed Sci. 2019; 26: 3. https://doi.org/10.1186/s12929-018-0493-6 PMID: 30609922 6. Kumar R, Sood U, Gupta V, Singh M, Scaria J, Lal R. Recent Advancements in the Development of Modern Probiotics for Restoring Human Gut Microbiome Dysbiosis. Indian J Microbiol. 2020; 60: 12– 25. https://doi.org/10.1007/s12088-019-00808-y PMID: 32089570 7. Aponte M, Murru N, Shoukat M. Therapeutic, Prophylactic, and Functional Use of Probiotics: A Cur- rent Perspective. Front Microbiol. 2020; 11: 562048. https://doi.org/10.3389/fmicb.2020.562048 PMID: 33042069 8. Hill C, Guarner F, Reid G, Gibson GR, Merenstein DJ, Pot B, et al. The International Scientific Associa- tion for Probiotics and Prebiotics consensus statement on the scope and appropriate use of the term PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 31 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism probiotic. Nat Rev Gastroenterol Hepatol. 2014; 11: 506–514. https://doi.org/10.1038/nrgastro.2014. 66 PMID: 24912386 9. Martı´nez Leal J, Valenzuela Sua´rez L, Jayabalan R, Huerta Oros J, Escalante-Aburto A. A review on health benefits of kombucha nutritional compounds and metabolites. CyTA—J Food. 2018; 16: 390– 399. https://doi.org/10.1080/19476337.2017.1410499 10. Jayabalan R, Malbasˇa RV, Lončar ES, Vitas JS, Sathishkumar M. A Review on Kombucha Tea-Micro- biology, Composition, Fermentation, Beneficial Effects, Toxicity, and Tea Fungus: A review on kombucha. . .. Compr Rev Food Sci Food Saf. 2014; 13: 538–550. https://doi.org/10.1111/1541-4337. 12073 PMID: 33412713 11. Ernst E. Kombucha: A Systematic Review of the Clinical Evidence. Complement Med Res. 2003; 10: 85–87. https://doi.org/10.1159/000071667 PMID: 12808367 12. Kapp JM, Sumner W. Kombucha: a systematic review of the empirical evidence of human health ben- efit. Ann Epidemiol. 2019; 30: 66–70. https://doi.org/10.1016/j.annepidem.2018.11.001 PMID: 30527803 13. Kaczmarczyk D, Lochyński S. PRODUCTS OF BIOTRANSFORMATION OF TEA INFUSION–PROP- ERTIES AND APPLICATION. 14. Marsh AJ, O’Sullivan O, Hill C, Ross RP, Cotter PD. Sequence-based analysis of the bacterial and fun- gal compositions of multiple kombucha (tea fungus) samples. Food Microbiol. 2014; 38: 171–178. https://doi.org/10.1016/j.fm.2013.09.003 PMID: 24290641 15. Coton M, Pawtowski A, Taminiau B, Burgaud G, Deniel F, Coulloumme-Labarthe L, et al. Unraveling microbial ecology of industrial-scale Kombucha fermentations by metabarcoding and culture-based methods. FEMS Microbiol Ecol. 2017;93. https://doi.org/10.1093/femsec/fix048 PMID: 28430940 16. Shenoy C. Hypoglycemic activity of bio-tea in mice. 17. Hartmann AM, Burleson LE, Holmes AK, Geist CR. Effects of chronic kombucha ingestion on open- field behaviors, longevity, appetitive behaviors, and organs in c57-bl/6 mice: a pilot study. Nutrition. 2000; 16: 755–761. https://doi.org/10.1016/s0899-9007(00)00380-4 PMID: 10978857 18. Sai Ram M, B A, T P, Prasad D, Kain AK, Mongia SS, et al. Effect of Kombucha tea on chromate(VI)- induced oxidative stress in albino rats. J Ethnopharmacol. 2000; 71: 235–240. https://doi.org/10.1016/ s0378-8741(00)00161-6 PMID: 10904168 19. Xu S, Wang Y, Wang J, Geng W. Kombucha Reduces Hyperglycemia in Type 2 Diabetes of Mice by Regulating Gut Microbiota and Its Metabolites. Foods. 2022; 11: 754. https://doi.org/10.3390/ foods11050754 PMID: 35267387 20. An L, Fu X, Chen J, Ma J. Application of Caenorhabditis elegans in Lipid Metabolism Research. Int J Mol Sci. 2023; 24: 1173. https://doi.org/10.3390/ijms24021173 PMID: 36674689 21. Hashmi S, Wang Y, Parhar RS, Collison KS, Conca W, Al-Mohanna F, et al. A C. elegans model to study human metabolic regulation. Nutr Metab. 2013; 10: 31. https://doi.org/10.1186/1743-7075-10- 31 PMID: 23557393 22. Cabreiro F, Gems D. Worms need microbes too: microbiota, health and aging in Caenorhabditis ele- gans. EMBO Mol Med. 2013; 5: 1300–1310. https://doi.org/10.1002/emmm.201100972 PMID: 23913848 23. 24. Zhang F, Berg M, Dierking K, Fe´lix M-A, Shapira M, Samuel BS, et al. Caenorhabditis elegans as a Model for Microbiome Research. Front Microbiol. 2017; 8. https://doi.org/10.3389/fmicb.2017.00485 PMID: 28386252 Labrousse A, Chauvet S, Couillault C, Le´ opold Kurz C, Ewbank JJ. Caenorhabditis elegans is a model host for Salmonella typhimurium. Curr Biol. 2000; 10: 1543–1545. https://doi.org/10.1016/s0960-9822 (00)00833-2 PMID: 11114526 25. Higurashi S, Tsukada S, Aleogho BM, Park JH, Al Y, Tanaka M, et al. Bacterial diet affects the age– dependent decline of associative learning in. 26. Nakagawa H, Shiozaki T, Kobatake E, Hosoya T, Moriya T, Sakai F, et al. Effects and mechanisms of prolongevity induced by Lactobacillus gasseri SBT2055 in Caenorhabditis elegans. Aging Cell. 2016; 15: 227–236. https://doi.org/10.1111/acel.12431 PMID: 26710940 27. Watson E, MacNeil LT, Arda HE, Zhu LJ, Walhout AJM. Integration of Metabolic and Gene Regulatory Networks Modulates the C. elegans Dietary Response. Cell. 2013; 153: 253–266. https://doi.org/10. 1016/j.cell.2013.02.050 PMID: 23540702 28. Brooks KK, Liang B, Watts JL. The Influence of Bacterial Diet on Fat Storage in C. elegans. Melov S, editor. PLoS ONE. 2009; 4: e7545. https://doi.org/10.1371/journal.pone.0007545 PMID: 19844570 29. Stuhr NL, Curran SP. Bacterial diets differentially alter lifespan and healthspan trajectories in C. ele- gans. Commun Biol. 2020; 3: 653. https://doi.org/10.1038/s42003-020-01379-1 PMID: 33159120 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 32 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism 30. Samuel BS, Rowedder H, Braendle C, Fe´lix M-A, Ruvkun G. Caenorhabditis elegans responses to bacteria from its natural habitats. Proc Natl Acad Sci. 2016;113. https://doi.org/10.1073/pnas. 1607183113 PMID: 27317746 31. Goya ME, Xue F, Sampedro-Torres-Quevedo C, Arnaouteli S, Riquelme-Dominguez L, Romanowski 32. A, et al. Probiotic Bacillus subtilis Protects against α-Synuclein Aggregation in C. elegans. Cell Rep. 2020; 30: 367–380.e7. https://doi.org/10.1016/j.celrep.2019.12.078 PMID: 31940482 Jose´ Santos Ju´ nior R, Andrade Batista R, Alves Rodrigues S, Xavier Filho L, Silva Lima A´ . Antimicro- bial Activity of Broth Fermented with Kombucha Colonies. J Microb Biochem Technol. 2009; 01: 072– 078. https://doi.org/10.4172/1948-5948.1000014 33. May A, Narayanan S, Alcock J, Varsani A, Maley C, Aktipis A. Kombucha: a novel model system for cooperation and conflict in a complex multi-species microbial ecosystem. PeerJ. 2019; 7: e7565. https://doi.org/10.7717/peerj.7565 PMID: 31534844 34. Villarreal-Soto SA, Beaufort S, Bouajila J, Souchard J-P, Taillandier P. Understanding Kombucha Tea Fermentation: A Review: Understanding Kombucha tea fermentation. . .. J Food Sci. 2018; 83: 580– 588. https://doi.org/10.1111/1750-3841.14068 PMID: 29508944 35. Jayabalan R, Malini K, Sathishkumar M, Swaminathan K, Yun S-E. Biochemical characteristics of tea fungus produced during kombucha fermentation. Food Sci Biotechnol. 2010; 19: 843–847. https://doi. org/10.1007/s10068-010-0119-6 36. Huang X, Xin Y, Lu T. A systematic, complexity-reduction approach to dissect the kombucha tea microbiome. eLife. 2022; 11: e76401. https://doi.org/10.7554/eLife.76401 PMID: 35950909 37. Melo JA, Ruvkun G. Inactivation of Conserved C. elegans Genes Engages Pathogen- and Xenobiotic- Associated Defenses. Cell. 2012; 149: 452–466. https://doi.org/10.1016/j.cell.2012.02.050 PMID: 22500807 38. Pujol N, Cypowyj S, Ziegler K, Millet A, Astrain A, Goncharov A, et al. Distinct Innate Immune Responses to Infection and Wounding in the C. elegans Epidermis. Curr Biol. 2008; 18: 481–489. https://doi.org/10.1016/j.cub.2008.02.079 PMID: 18394898 39. Zhang Y, Lu H, Bargmann CI. Pathogenic bacteria induce aversive olfactory learning in Caenorhabdi- tis elegans. Nature. 2005; 438: 179–184. https://doi.org/10.1038/nature04216 PMID: 16281027 40. Schulenburg H, Ewbank JJ. The genetics of pathogen avoidance in Caenorhabditis elegans. Mol Microbiol. 2007; 66: 563–570. https://doi.org/10.1111/j.1365-2958.2007.05946.x PMID: 17877707 41. Revtovich AV, Lee R, Kirienko NV. Interplay between mitochondria and diet mediates pathogen and stress resistance in Caenorhabditis elegans. Garsin DA, editor. PLOS Genet. 2019; 15: e1008011. https://doi.org/10.1371/journal.pgen.1008011 PMID: 30865620 42. Malbasˇa RV, Lončar ES, Vitas JS, Čanadanović-Brunet JM. Influence of starter cultures on the antioxi- dant activity of kombucha beverage. Food Chem. 2011; 127: 1727–1731. https://doi.org/10.1016/j. foodchem.2011.02.048 43. Chakravorty S, Bhattacharya S, Chatzinotas A, Chakraborty W, Bhattacharya D, Gachhui R. Kombu- cha tea fermentation: Microbial and biochemical dynamics. Int J Food Microbiol. 2016; 220: 63–72. https://doi.org/10.1016/j.ijfoodmicro.2015.12.015 PMID: 26796581 44. Bauer-Petrovska B, Petrushevska-Tozi L. Mineral and water soluble vitamin content in the Kombucha drink. Int J Food Sci Technol. 2000; 35: 201–205. https://doi.org/10.1046/j.1365-2621.2000.00342.x 45. Mousavi SM, Hashemi SA, Zarei M, Gholami A, Lai CW, Chiang WH, et al. Recent Progress in Chemi- cal Composition, Production, and Pharmaceutical Effects of Kombucha Beverage: A Complementary and Alternative Medicine. Chen J, editor. Evid Based Complement Alternat Med. 2020; 2020: 1–14. https://doi.org/10.1155/2020/4397543 PMID: 33281911 46. Winter AD, Tjahjono E, Beltra´n LJ, Johnstone IL, Bulleid NJ, Page AP. Dietary-derived vitamin B12 protects Caenorhabditis elegans from thiol-reducing agents. BMC Biol. 2022; 20: 228. https://doi.org/ 10.1186/s12915-022-01415-y PMID: 36209095 47. Watson E, Olin-Sandoval V, Hoy MJ, Li C-H, Louisse T, Yao V, et al. Metabolic network rewiring of pro- pionate flux compensates vitamin B12 deficiency in C. elegans. eLife. 2016; 5: e17670. https://doi.org/ 10.7554/eLife.17670 PMID: 27383050 48. Kadner RJ. Vitamin B 12 transport in Escherichia coli: energy coupling between membranes. Mol Microbiol. 1990; 4: 2027–2033. https://doi.org/10.1111/j.1365-2958.1990.tb00562.x PMID: 2089218 49. Hemarajata P, Versalovic J. Effects of probiotics on gut microbiota: mechanisms of intestinal immuno- modulation and neuromodulation. Ther Adv Gastroenterol. 2013; 6: 39–51. https://doi.org/10.1177/ 1756283X12459294 PMID: 23320049 50. Visconti A, Le Roy CI, Rosa F, Rossi N, Martin TC, Mohney RP, et al. Interplay between the human gut microbiome and host metabolism. Nat Commun. 2019; 10: 4505. https://doi.org/10.1038/s41467- 019-12476-z PMID: 31582752 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 33 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism 51. Cox TO, Lundgren P, Nath K, Thaiss CA. Metabolic control by the microbiome. Genome Med. 2022; 14: 80. https://doi.org/10.1186/s13073-022-01092-0 PMID: 35906678 52. Srinivasan S. Regulation of Body Fat in C. elegans. 2016. 53. O’Rourke EJ, Soukas AA, Carr CE, Ruvkun G. C. elegans Major Fats Are Stored in Vesicles Distinct from Lysosome-Related Organelles. Cell Metab. 2009; 10: 430–435. https://doi.org/10.1016/j.cmet. 2009.10.002 PMID: 19883620 54. Klapper M, Ehmke M, Palgunow D, Bo¨hme M, Mattha¨us C, Bergner G, et al. Fluorescence-based fixa- tive and vital staining of lipid droplets in Caenorhabditis elegans reveal fat stores using microscopy and flow cytometry approaches. J Lipid Res. 2011; 52: 1281–1293. https://doi.org/10.1194/jlr. D011940 PMID: 21421847 55. Sandhu A, Singh V. Total Triglyceride Quantification in Caenorhabditis elegans. BIO-Protoc. 2020; 10. https://doi.org/10.21769/BioProtoc.3819 PMID: 33659471 56. Walker G, Houthoofd K, Vanfleteren JR, Gems D. Dietary restriction in C. elegans: From rate-of-living effects to nutrient sensing pathways. Mech Ageing Dev. 2005; 126: 929–937. https://doi.org/10.1016/j. mad.2005.03.014 PMID: 15896824 57. 58. Farris M, Fang L, Aslamy A, Pineda V. Steroid signaling mediates longevity responses to the eat-2 genetic model of dietary restriction in Caenorhabditis elegans. Transl Med Aging. 2019; 3: 90–97. https://doi.org/10.1016/j.tma.2019.09.003 Lakowski B, Hekimi S. The genetics of caloric restriction in Caenorhabditis elegans. Proc Natl Acad Sci. 1998; 95: 13091–13096. https://doi.org/10.1073/pnas.95.22.13091 PMID: 9789046 59. Ching T-T, Hsu A-L. Solid Plate-based Dietary Restriction in Caenorhabditis elegans. J Vis Exp. 2011; 2701. https://doi.org/10.3791/2701 PMID: 21654629 60. 61. Franco-Jua´ rez B, Go´mez-Manzo S, Herna´ndez-Ochoa B, Ca´ rdenas-Rodrı´guez N, Arreguin-Espinosa R, Pe´ rez De La Cruz V, et al. Effects of High Dietary Carbohydrate and Lipid Intake on the Lifespan of C. elegans. Cells. 2021; 10: 2359. https://doi.org/10.3390/cells10092359 PMID: 34572007 Teoh AL, Heard G, Cox J. Yeast ecology of Kombucha fermentation. Int J Food Microbiol. 2004; 95: 119–126. https://doi.org/10.1016/j.ijfoodmicro.2003.12.020 PMID: 15282124 62. Harrington BJ, Hageage GJ. Calcofluor White: A Review of its Uses and Applications in Clinical Mycol- ogy and Parasitology. Lab Med. 2003; 34: 361–367. https://doi.org/10.1309/EPH2TDT8335GH0R3 63. Lavasani PS, Motevaseli E, Shirzad M, Modarressi MH. Isolation and identification of Komagataeibac- ter xylinus from Iranian traditional vinegars and molecular analyses. 64. Monheit EJ, Conwan DF, Moor D. Rapid detection of fungi in tissues using calcofluor white and fluo- rescence microscopy. Arch Pathol Lab Med. 1984; 108: 616–8. PMID: 6204621 65. Wood DE, Lu J, Langmead B. Improved metagenomic analysis with Kraken 2. Genome Biol. 2019; 20: 257. https://doi.org/10.1186/s13059-019-1891-0 PMID: 31779668 66. Suh S-O, Gujjari P, Beres C, Beck B, Zhou J. Proposal of Zygosaccharomyces parabailii sp. nov. and Zygosaccharomyces pseudobailii sp. nov., novel species closely related to Zygosaccharomyces bailii. Int J Syst Evol Microbiol. 2013; 63: 1922–1929. https://doi.org/10.1099/ijs.0.048058-0 PMID: 23524351 67. Yen K, Le TT, Bansal A, Narasimhan SD, Cheng J-X, Tissenbaum HA. A Comparative Study of Fat Storage Quantitation in Nematode Caenorhabditis elegans Using Label and Label-Free Methods. Melov S, editor. PLoS ONE. 2010; 5: e12810. https://doi.org/10.1371/journal.pone.0012810 PMID: 20862331 68. Holdorf AD, Higgins DP, Hart AC, Boag PR, Pazour GJ, Walhout AJM, et al. WormCat: An Online Tool for Annotation and Visualization of Caenorhabditis elegans Genome-Scale Data. GSA Journals; 2019. p. 97266344 Bytes. https://doi.org/10.1534/genetics.119.302919 PMID: 31810987 69. Hansen M, Flatt T, Aguilaniu H. Reproduction, Fat Metabolism, and Life Span: What Is the Connec- tion? Cell Metab. 2013; 17: 10–19. https://doi.org/10.1016/j.cmet.2012.12.003 PMID: 23312280 70. Templeman NM, Murphy CT. Regulation of reproduction and longevity by nutrient-sensing pathways. J Cell Biol. 2018; 217: 93–106. https://doi.org/10.1083/jcb.201707168 PMID: 29074705 71. Podshivalova K, Kerr RA, Kenyon C. How a Mutation that Slows Aging Can Also Disproportionately Extend End-of-Life Decrepitude. Cell Rep. 2017; 19: 441–450. https://doi.org/10.1016/j.celrep.2017. 03.062 PMID: 28423308 72. Zhang Y-P, Zhang W-H, Zhang P, Li Q, Sun Y, Wang J-W, et al. Intestine-specific removal of DAF-2 nearly doubles lifespan in Caenorhabditis elegans with little fitness cost. Nat Commun. 2022; 13: 6339. https://doi.org/10.1038/s41467-022-33850-4 PMID: 36284093 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 34 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism 73. Kaletsky R, Lakhina V, Arey R, Williams A, Landis J, Ashraf J, et al. The C. elegans adult neuronal IIS/ FOXO transcriptome reveals adult phenotype regulators. Nature. 2016; 529: 92–96. https://doi.org/10. 1038/nature16483 PMID: 26675724 74. Pauli F, Liu Y, Kim YA, Chen P-J, Kim SK. Chromosomal clustering and GATA transcriptional regula- tion of intestine-expressed genes in C. elegans. Development. 2006; 133: 287–295. https://doi.org/10. 1242/dev.02185 PMID: 16354718 75. Xu N, Zhang SO, Cole RA, McKinney SA, Guo F, Haas JT, et al. The FATP1–DGAT2 complex facili- tates lipid droplet expansion at the ER–lipid droplet interface. J Cell Biol. 2012; 198: 895–911. https:// doi.org/10.1083/jcb.201201139 PMID: 22927462 76. Yen C-LE, Stone SJ, Koliwad S, Harris C, Farese RV. Thematic Review Series: Glycerolipids. DGAT enzymes and triacylglycerol biosynthesis. J Lipid Res. 2008; 49: 2283–2301. https://doi.org/10.1194/ jlr.R800018-JLR200 PMID: 18757836 77. Singh R, Kaushik S, Wang Y, Xiang Y, Novak I, Komatsu M, et al. Autophagy regulates lipid metabo- lism. Nature. 2009; 458: 1131–1135. https://doi.org/10.1038/nature07976 PMID: 19339967 78. Czaja MJ, Cuervo AM. Lipases in lysosomes, what for? Autophagy. 2009; 5: 866–867. https://doi.org/ 10.4161/auto.9040 PMID: 19502773 79. Kounakis K, Chaniotakis M, Markaki M, Tavernarakis N. Emerging Roles of Lipophagy in Health and Disease. Front Cell Dev Biol. 2019; 7: 185. https://doi.org/10.3389/fcell.2019.00185 PMID: 31552248 80. Mony VK, Drangowska-Way A, Albert R, Harrison E, Ghaddar A, Horak MK, et al. Context-specific regulation of lysosomal lipolysis through network-level diverting of transcription factor interactions. Proc Natl Acad Sci. 2021; 118: e2104832118. https://doi.org/10.1073/pnas.2104832118 PMID: 34607947 81. Shin DW. Lipophagy: Molecular Mechanisms and Implications in Metabolic Disorders. 82. Lee JH, Kong J, Jang JY, Han JS, Ji Y, Lee J, et al. Lipid Droplet Protein LID-1 Mediates ATGL-1- Dependent Lipolysis during Fasting in Caenorhabditis elegans. Mol Cell Biol. 2014; 34: 4165–4176. https://doi.org/10.1128/MCB.00722-14 PMID: 25202121 83. O’Rourke EJ, Ruvkun G. MXL-3 and HLH-30 transcriptionally link lipolysis and autophagy to nutrient availability. Nat Cell Biol. 2013; 15: 668–676. https://doi.org/10.1038/ncb2741 PMID: 23604316 84. Bazopoulou D, Tavernarakis N. The NemaGENETAG initiative: large scale transposon insertion gene-tagging in Caenorhabditis elegans. Genetica. 2009; 137: 39–46. https://doi.org/10.1007/s10709- 009-9361-3 PMID: 19343510 85. Papsdorf K, Miklas JW, Hosseini A, Cabruja M, Morrow CS, Savini M, et al. Lipid droplets and peroxi- somes are co-regulated to drive lifespan extension in response to mono-unsaturated fatty acids. Nat Cell Biol. 2023; 25: 672–684. https://doi.org/10.1038/s41556-023-01136-6 PMID: 37127715 86. Antolak H, Piechota D, Kucharska A. Kombucha Tea—A Double Power of Bioactive Compounds from Tea and Symbiotic Culture of Bacteria and Yeasts (SCOBY). Antioxidants. 2021; 10: 1541. https://doi. org/10.3390/antiox10101541 PMID: 34679676 87. Watts JL, Ristow M. Lipid and Carbohydrate Metabolism in Caenorhabditis elegans. Genetics. 2017; 207: 413–446. https://doi.org/10.1534/genetics.117.300106 PMID: 28978773 88. Wondmkun YT. Obesity, Insulin Resistance, and Type 2 Diabetes: Associations and Therapeutic Implications. Diabetes Metab Syndr Obes Targets Ther. 2020;Volume 13: 3611–3616. https://doi.org/ 10.2147/DMSO.S275898 PMID: 33116712 89. Mooradian AD. Dyslipidemia in type 2 diabetes mellitus. Nat Rev Endocrinol. 2009; 5: 150–159. https://doi.org/10.1038/ncpendmet1066 PMID: 19229235 90. Kaze AD, Santhanam P, Musani SK, Ahima R, Echouffo-Tcheugui JB. Metabolic Dyslipidemia and Cardiovascular Outcomes in Type 2 Diabetes Mellitus: Findings From the Look AHEAD Study. J Am Heart Assoc. 2021; 10: e016947. https://doi.org/10.1161/JAHA.120.016947 PMID: 33728932 91. Xie K, Liu Y, Li X, Zhang H, Zhang S, Mak HY, et al. Dietary S. maltophilia induces supersized lipid droplets by enhancing lipogenesis and ER-LD contacts in C. elegans. Gut Microbes. 2022; 14: 2013762. https://doi.org/10.1080/19490976.2021.2013762 PMID: 35112996 92. Sang Y, Ren J, Aballay A. The transcription factor HLH-26 controls probiotic-mediated protection against intestinal infection through up-regulation of the Wnt/BAR-1 pathway. Cadwell K, editor. PLOS Biol. 2022; 20: e3001581. https://doi.org/10.1371/journal.pbio.3001581 PMID: 35263319 93. Kumar A, Baruah A, Tomioka M, Iino Y, Kalita MC, Khan M. Caenorhabditis elegans: a model to under- stand host–microbe interactions. Cell Mol Life Sci. 2020; 77: 1229–1249. https://doi.org/10.1007/ s00018-019-03319-7 PMID: 31584128 94. Kim Y, Mylonakis E. Caenorhabditis elegans Immune Conditioning with the Probiotic Bacterium Lacto- bacillus acidophilus Strain NCFM Enhances Gram-Positive Immune Responses. Urban JF, editor. Infect Immun. 2012; 80: 2500–2508. https://doi.org/10.1128/IAI.06350-11 PMID: 22585961 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 35 / 36 PLOS GENETICSKombucha microbes stimulate host lipid catabolism 95. Brenner S. The genetics of Caenorhabditis elegans. Genetics. 1974; 77: 71–94. https://doi.org/10. 1093/genetics/77.1.71 PMID: 4366476 96. Frøkjær-Jensen C, Wayne Davis M, Hopkins CE, Newman BJ, Thummel JM, Olesen S-P, et al. Sin- gle-copy insertion of transgenes in Caenorhabditis elegans. Nat Genet. 2008; 40: 1375–1383. https:// doi.org/10.1038/ng.248 PMID: 18953339 97. Ghanta KS, Mello CC. Melting dsDNA Donor Molecules Greatly Improves Precision Genome Editing in Caenorhabditis elegans. Genetics. 2020; 216: 643–650. https://doi.org/10.1534/genetics.120. 303564 PMID: 32963112 98. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA, et al. Reproducible, interac- tive, scalable and extensible microbiome data science using QIIME 2. Nat Biotechnol. 2019; 37: 852– 857. https://doi.org/10.1038/s41587-019-0209-9 PMID: 31341288 99. Walker A, Bhargava R, Vaziriyan-Sani A, Brust A, Czyz D. Quantification of Bacterial Loads in Caenor- habditis elegans. BIO-Protoc. 2022; 12. https://doi.org/10.21769/BioProtoc.4291 PMID: 35127981 100. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, et al. Fiji: an open-source platform for biological-image analysis. Nat Methods. 2012; 9: 676–682. https://doi.org/10.1038/nmeth. 2019 PMID: 22743772 101. Wa¨ hlby C, Lee Conery A, Bray M-A, Kamentsky L, Larkins-Ford J, Sokolnicki KL, et al. High- and low- throughput scoring of fat mass and body fat distribution in C. elegans. Methods. 2014; 68: 492–499. https://doi.org/10.1016/j.ymeth.2014.04.017 PMID: 24784529 102. Langmead B, Salzberg SL. Fast gapped-read alignment with Bowtie 2. Nat Methods. 2012; 9: 357– 359. https://doi.org/10.1038/nmeth.1923 PMID: 22388286 103. Dowen RH. CEH-60/PBX and UNC-62/MEIS Coordinate a Metabolic Switch that Supports Reproduc- tion in C. elegans. Dev Cell. 2019; 49: 235–250.e7. https://doi.org/10.1016/j.devcel.2019.03.002 PMID: 30956009 104. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014; 15: 550. https://doi.org/10.1186/s13059-014-0550-8 PMID: 25516281 105. Kolde R. Pheatmap: pretty heatmaps. R Package Version. 2012; 1: 726. 106. Wickham H, Averick M, Bryan J, Chang W, McGowan L, Franc¸ois R, et al. Welcome to the Tidyverse. J Open Source Softw. 2019; 4: 1686. https://doi.org/10.21105/joss.01686 PLOS Genetics \| https://doi.org/10.1371/journal.pgen.1011003 March 28, 2024 36 / 36 PLOS GENETICS
10.1128_jvi.00781-23	\| Virology \| Full-Length Text Human cytomegalovirus mediates APOBEC3B relocalization early during infection through a ribonucleotide reductase- independent mechanism Elisa Fanunza,1,2 Adam Z. Cheng,3 Ashley A. Auerbach,1 Bojana Stefanovska,1,4 Sofia N. Moraes,3 James R. Lokensgard,5 Matteo Biolatti,6 Valentina Dell'Oste,6 Craig J. Bierle,7 Wade A. Bresnahan,8 Reuben S. Harris1,4 AUTHOR AFFILIATIONS See affiliation list on p. 15. ABSTRACT The APOBEC3 family of DNA cytosine deaminases comprises an important arm of the innate antiviral defense system. The gamma-herpesviruses Epstein-Barr virus and Kaposi’s sarcoma-associated herpesvirus and the alpha-herpesviruses herpes simplex virus (HSV)-1 and HSV-2 have evolved an efficient mechanism to avoid APOBEC3 restriction by directly binding to APOBEC3B and facilitating its exclusion from the nuclear compartment. The only viral protein required for APOBEC3B relocalization is the large subunit of the ribonucleotide reductase (RNR). Here, we ask whether this APOBEC3B relocalization mechanism is conserved with the beta-herpesvirus human cytomegalovi rus (HCMV). Although HCMV infection causes APOBEC3B relocalization from the nucleus to the cytoplasm in multiple cell types, the viral RNR (UL45) is not required. APOBEC3B relocalization occurs rapidly following infection suggesting the involvement of an immediate early or early (IE/E) viral protein. In support of this possibility, genetic (IE1 mutant) and pharmacologic (cycloheximide) strategies that prevent the expression of IE/E viral proteins also block APOBEC3B relocalization. In comparison, the treatment of infected cells with phosphonoacetic acid, which interferes with viral late protein expression, still permits A3B relocalization. These results combine to indicate that the beta-herpesvirus HCMV uses an RNR-independent, yet phenotypically similar, molecular mechanism to antagonize APOBEC3B. IMPORTANCE Human cytomegalovirus (HCMV) infections can range from asympto matic to severe, particularly in neonates and immunocompromised patients. HCMV has evolved strategies to overcome host-encoded antiviral defenses to achieve lytic viral DNA replication and dissemination and, under some conditions, latency and long-term persistence. Here, we show that HCMV infection causes the antiviral factor, APOBEC3B, to relocalize from the nuclear compartment to the cytoplasm. This overall strategy resembles that used by related herpesviruses. However, the HCMV relocaliza tion mechanism utilizes a different viral factor(s) and available evidence suggests the involvement of at least one protein expressed at the early stages of infection. This knowledge is important because a greater understanding of this mechanism could lead to novel antiviral strategies that enable APOBEC3B to naturally restrict HCMV infection. KEYWORDS APOBEC3B (A3B), herpesviruses, human cytomegalovirus, immediate- early genes, innate immunity, ribonucleotide reductase T he APOBEC3 (A3) system is an essential part of the cellular innate immune response to viral infections [reviewed by Green and Weitzman (1), Harris and Dudley (2), and Hakata and Miyazawa (3)]. A3-mediated restriction has been reported for a broad number of DNA-based viruses, including exogenous viruses (retroviruses, Editor Felicia Goodrum, The University of Arizona, Tucson, Arizona, USA Address correspondence to Reuben S. Harris, [email protected]. The authors declare no conflict of interest. See the funding table on p. 15. Received 25 May 2023 Accepted 21 June 2023 Published 11 August 2023 Copyright © 2023 Fanunza et al. This is an open- access article distributed under the terms of the Creative Commons Attribution 4.0 International license. August Volume 97 Issue 8 10.1128/jvi.00781-23 1 Full-Length Text Journal of Virology polyomaviruses, papillomaviruses, parvoviruses, hepadnaviruses, and herpesviruses) and endogenous viruses and transposable elements. The mechanism by which virus restriction occurs is well documented and dependent partly on the ability of A3 enzymes to introduce mutations in the viral genome by catalyzing cytosine deamination in exposed single-stranded (ss)DNA intermediates. In addition, deaminase-independent antiviral activity has been reported against endogenous retroelements, reverse-tran scribing viruses, adeno-associated viruses, and RNA viruses, and this may be attributed to strong nucleic acid binding activity. The continuous arms race between host and viruses leads to the selection of viral factors that are able to counteract innate immune factors, including the A3 antiviral enzymes. For example, HIV-1, HIV-2, and related lentiviruses encode a viral accessory protein Vif that mediates the degradation of restrictive A3s (4, 5). Recently, a novel mechanism of A3 counteraction was discovered for the gamma-herpesviruses Epstein- Barr virus (EBV), which uses the viral ribonucleotide reductase (RNR) large subunit, BORF2, to directly bind, inhibit, and relocalize APOBEC3B (A3B) from the nucleus to the cytoplasm, thus preserving viral genome integrity (6). This mechanism of A3 neutral ization is likely to be conserved because at least two other herpesviruses, Kaposi’s sarcoma-associated herpesvirus (KSHV) and herpes simplex virus 1 (HSV-1), whose RNRs [open reading frame (ORF) 61 and ICP6, respectively] physically interact with A3B, as well with APOBEC3A (A3A), and trigger their redistribution from the nucleus to the cytoplasmic compartment (7–10). In further support of evolutionary conservation, a systematic analysis of a large panel of present-day gamma-herpesvirus RNRs and primate A3B proteins indicates that the evolution of this viral RNR-mediated A3B neutralization mechanism was likely selected by the birth of the A3B gene by unequal crossing-over in an ancestral Old World primate approximately 29–43 million years ago (8, 11). Human cytomegalovirus (HCMV) is a member of the beta-herpesvirus subfamily. HCMV is a ubiquitous virus, found in approximately 90% of the worldwide popula tion. HCMV infection is usually asymptomatic in healthy individuals, but it can cause severe disease in immunocompromised hosts [reviewed by Tyl et al. (12) and Grif fiths and Reeves (13)]. Congenital HCMV infections are also a leading cause of birth defects [reviewed by Manicklal et al. (14) and Britt (15)]. HCMV has a large double-stran ded (ds)DNA genome of 235 kb—the largest among known human herpesviruses— containing 165 canonical ORFs and several alternative transcripts [reviewed by Turner and Mathias (16)]. Lytic HCMV infection involves a temporal cascade of gene expression. A small subset of genes, termed immediate-early (IE) genes, are the first to be expressed. Transcription of IE genes does not require de novo protein synthesis. IE proteins together with host factors mediate the expression of the kinetically distinct early genes (E), whose products in large part promote viral genome replication and the expression of late genes (L) [reviewed by Turner and Mathias (16)]. Several HCMV gene products have acquired the ability to subvert different signaling pathways and modulate various components of the immune response to make the host cell more permissive to viral replication and survival [reviewed by Biolatti et al. (17) and Patro (18)]. Given the ability of gamma- and alpha-herpesviruses (EBV/KSHV and HSV-1/2, respectively) to inhibit A3B, we sought to investigate whether HCMV possesses a similar RNR-mediated A3 neutralization mechanism. Our results demonstrate that HCMV infection is also capable of inducing the selective nuclear to cytoplasmic relocalization of A3B. However, surprisingly, results with multiple independent viral strains and cell lines indicate that the relocalization mechanism of A3B by HCMV is not conserved with other human herpesviruses and, instead, occurs independently of the HCMV UL45 RNR. In addition to this strong mechanistic distinction, multiple lines of evidence including rapid A3B relocalization kinetics suggest involvement of at least one viral IE-E protein in A3B relocalization. August Volume 97 Issue 8 10.1128/jvi.00781-23 2 Journal of Virology Full-Length Text RESULTS HCMV mediates A3B relocalization independently of viral strain and cell type We previously reported the ability of gamma- and alpha-herpesviruses to bind to A3B and mediate its relocalization from the nuclear compartment into cytoplasmic aggre gates (6–8). To investigate whether the beta-herpesvirus HCMV has similar functionality, immunofluorescence (IF) microscopy experiments were done using infected human retinal pigment epithelial cells, ARPE19. First, ARPE19 cells were stably transduced with a lentivirus expressing C-terminally HA-tagged A3B. As reported for other human cell types (11, 19–21), A3B localizes primarily to the nuclear compartment of mock/non- infected ARPE19 cells (representative image in Fig. 1A). Next, ARPE19 transduced cells were infected with HCMV strain TB40/E that expresses the mCherry protein (TB40- mCherry) and analyzed for A3B localization by IF microscopy 72 hours post-infection (hpi). Infected, mCherry-positive cells are visibly enlarged, as expected for productive cytomegalovirus infection, and A3B becomes predominantly cytoplasmic (representative image in Fig. 1A and quantification in Fig. 1F). TB40-mCherry infection also causes A3B-HA relocalization in other cell types including primary human foreskin fibroblast cells (HFF-1) and the human glioma cell line U373 (Fig. 1B and C and quantification in Fig. 1F). Moreover, when ARPE19 cells were transfected to express each of the seven different human A3 family members, A3B is the only protein to show a major change in subcellular distribution following HCMV infection (Fig. S1). To ask whether the A3B relocalization mechanism extends to other HCMV strains, HFF-1 and U373 stably transduced with HA-tagged A3B were infected with the labora tory-adapted GFP-expressing AD169 strain (AD169-GFP), and IF microscopy was done 72 hpi. As mentioned above, AD169-GFP infection induces strong relocalization of A3B from the nuclear compartment to the cytoplasm (Fig. 1D and E and quantification in Fig. 1G). Similar A3B-HA relocalization is observed during infection of HFF-1 cells with the Merlin strain (Fig. S2A). As an additional control for specificity, A3B-EGFP but not EGFP alone relocalizes to the cytoplasmic compartment following infection of ARPE19 cells with TB40-mCherry (Fig. S2B). Thus, the A3B relocalization phenotype is evident following infection with multiple HCMV strains and in a range of different cell types (both primary and immortalized) permissive for HCMV infection. Catalytic mutant and endogenous A3B are relocalized upon HCMV infection Overexpression of wildtype (wt) A3B causes chromosomal DNA deamination, strong DNA damage responses, cell cycle perturbations, and eventually cell death (22–24). These phenotypes require the catalytic activity of A3B. To address the possibility that A3B relocalization may be triggered indirectly by one of these events, HFF-1, U373, and ARPE19 cells were transduced with a lentiviral construct expressing the catalytically inactive A3B mutant (A3B-E255A). HCMV infection and IF microscopy experiments were done as above. In all instances, A3B-E255A relocalizes from the nucleus to the cytoplasm following infection with TB40-mCherry or AD169-GFP (representative images in Fig. 2A through E and quantification in Fig. 2F). Importantly, the magnitude of relocalization was indistinguishable for wt A3B and A3B-E255A (Fig. S3A). These results demonstrate that the relocalization of A3B occurs independent of its DNA deamination activity and is unlikely to be part of a general DNA damage response. To further confirm that the relocalization phenotype is not a general effect of A3B overexpression, we next evaluated the subcellular localization of the endogenous protein. ARPE19 cells were infected with TB40-mCherry, allowing 72 h for infection to progress, and then performing IF microscopy with the rabbit anti-human A3B monoclo nal antibody 5210-87-13 (25). As observed above with overexpressed A3B-HA (with or without catalytic activity), the endogenous A3B protein also shows strong relocalization from the nucleus to the cytoplasm (Fig. 2G and H). These results indicate that the A3B relocalization mechanism of HCMV is not likely to be an artifact of protein overexpression August Volume 97 Issue 8 10.1128/jvi.00781-23 3 Full-Length Text Journal of Virology FIG 1 A3B relocalization occurs with multiple HCMV strains in different cell types. (A–E) Representative IF microscopy images of the indicated cell types stably expressing A3B-HA incubated with medium alone (mock) or infected with the indicated HCMV strains for 72 h (10 µm scale). High-magnification images in Fig. (Continued on next page) August Volume 97 Issue 8 10.1128/jvi.00781-23 4 Full-Length Text FIG 1 (Continued) Journal of Virology 1A were taken with a Nikon Eclipse Ti2 system, and all other images were taken with an EVOS cell imaging system. (F and G) Quantification of A3B-HA subcellular localization phenotypes shown in panels A–E. Each histogram bar reports the percentage of cells with cytoplasmic A3B-HA (n > 100 cells per condition; mean ± SD with indicated P values from unpaired Student’s t-tests). because endogenous A3B also exhibits a clear relocalization phenotype following virus infection. HCMV UL45 is incapable of binding, inhibiting, or relocalizing human A3B The only gamma- and alpha-herpesvirus protein required for A3B relocalization is the large subunit of the viral RNR (6–8). The large RNR subunit of EBV, BORF2, directly binds A3B, inhibits its catalytic activity, and relocalizes the protein from the nucleus to the cytoplasm. To address whether the HCMV large RNR subunit, UL45, is capable of similarly binding to A3B, we performed a series of coimmunoprecipitation (co-IP) experiments. 293T cells were transfected with empty vector or FLAG-tagged HCMV UL45 or EBV BORF2 together with a HA-tagged human A3B or other A3 constructs as negative controls. As expected, EBV BORF2 robustly co-IPs A3B but not the negative control A3G (Fig. 3A). In parallel experiments, HCMV UL45 appears incapable of co-IP of either A3B or A3A (Fig. 3A). However, conclusions from these experiments are limited by relatively low UL45 expression levels in cell extracts, and multiple expressed products including likely monomeric and dimeric forms (full-length UL45 is predicted to be ~108 kDa). We, therefore, turned to other approaches to ask whether HCMV UL45 is capable of interfering with A3B catalytic activity. First, 293T cells were transfected with human A3B, together with empty vector, HCMV UL45, or EBV BORF2. 48 h post-transfection, whole cell lysates were incubated with a fluorescently labeled ssDNA substrate containing a single 5′-TC deamination motif. If A3B catalyzes the deamination of this cytosine to uracil, cellular uracil N-DNA glycosylase 2 excises the uracil and the resulting abasic site is cleaved by sodium hydroxide, leading to the formation of a short product oligonucleo tide. Consistent with previous results (6), A3B exhibits robust ssDNA C-to-U activity in cell extracts, and its activity is strongly inhibited by BORF2 (Fig. 3B). In comparison, HCMV UL45 co-expression has a negligible effect on the ssDNA C-to-U activity of A3B in cell extracts (Fig. 3B). Next, IF microscopy experiments were done by cotransfecting HeLa cells with A3B-HA and viral RNR-FLAG constructs, allowing 48 h for expression, and imaging with specific antibodies. In contrast to EBV BORF2, which relocalizes A3B from the nuclear to the cytoplasmic compartment, the expression of HCMV UL45 has no effect on A3B subcellular localization (Fig. 3C). Taken together, negative results from co-IP, deaminase inhibition, and colocalization experiments indicate that the large RNR subunit of HCMV, UL45, is incapable of affecting the subcellular localization or ssDNA deamina tion activity of A3B. To directly ask whether HCMV UL45 is required for A3B relocalization, we compared the subcellular localization phenotypes of A3B in U373 cells following infection by AD169-GFP or a derivative virus engineered to lack UL45 [AD169-GFP ΔUL45 (26)]. U373 cells were stably transduced with HA-tagged A3B 48 h prior to mock infection or infection with AD169-GFP or AD169-GFP ΔUL45. After 72 h of infection, cells were fixed, permeabilized, and imaged by IF microscopy. As described above, infection by AD169- GFP causes the relocalization of A3B from the nuclear to the cytoplasmic compartment (Fig. 3D). As expected, cells infected with AD169-GFP ΔUL45 show an indistinguishable A3B cytoplasmic relocalization phenotype (Fig. 3D and quantification in Fig. S3B). This key result was confirmed by IF microscopy experiments using two other HCMV strains (TB40/E and FIX) and otherwise isogenic UL45-null derivatives (TB40/E ΔUL45 and FIX ΔUL45) (Fig. 3E). These results demonstrate that UL45 is dispensable for HCMV-mediated relocalization of A3B and, together with the results above, that this beta-herpesvirus does not share the RNR-dependent mechanism of gamma- and alpha-herpesviruses. August Volume 97 Issue 8 10.1128/jvi.00781-23 5 Full-Length Text Journal of Virology FIG 2 Catalytic mutant and endogenous A3B are relocalized by HCMV. (A–E) Representative IF microscopy images of the indicated cell types stably expressing A3B-E255A-HA incubated with medium alone (mock) or infected with the indicated HCMV strains for 72 h (10 µm scale). (F) Quantification of A3B-E255A-HA subcellular localization phenotypes shown in panels A–E. Each histogram bar reports the percentage of cells with cytoplasmic A3B-HA (n > 100 cells per condition; mean ± SD with indicated P values from unpaired Student’s t-tests). (G) Representative IF microscopy images of ARPE19 cells incubated with medium alone (mock) or infected with TB40-mCherry for 72 h and then stained for endogenous A3B (10 µm scale). (H) Quantification of endogenous A3B subcellular localization phenotype shown in panel G. The dot-plot chart shows the ratio between nuclear and cytoplasmic fluorescence intensity (n > 50 cells per condition; P values obtained with unpaired Student’s t-tests). August Volume 97 Issue 8 10.1128/jvi.00781-23 6 Full-Length Text Journal of Virology FIG 3 A3B relocalization is UL45 independent. (A) Co-IP of transfected HCMV UL45-FLAG with the indicated A3-HA constructs in 293T cells. Cells co-transfected with EBV BORF2 and A3B or A3G are used as positive and negative controls, respectively. (B) TBE-urea PAGE analysis of A3B deaminase activity in the presence of empty vector, HCMV UL45, or EBV BORF2. (C) Representative IF microscopy images of HeLa cells transiently expressing A3B-HA together with empty vector, HCMV UL45-FLAG, or EBV BORF2-FLAG (10 µm scale). (D and E) Representative IF microscopy images of the indicated cell types stably expressing A3B-HA incubated with medium alone (mock) or infected with the indicated HCMV strains and UL45-null derivatives for 72 h (10 µm scale). The N-terminal domain of A3B is sufficient for HCMV-mediated relocalization A3B is comprised of two conserved deaminase domains: an inactive N-terminal domain (A3B-NTD) and a catalytically active C-terminal domain (A3B-CTD) (9, 27). A3B-NTD is thought to be regulatory in nature and is alone sufficient for nuclear localization (11, 19). EBV BORF2 mediates A3B relocalization by binding to the CTD and not to the NTD (6). To ask whether domain requirements might further distinguish the A3B relocalization August Volume 97 Issue 8 10.1128/jvi.00781-23 7 Full-Length Text Journal of Virology mechanism of HCMV, IF microscopy experiments were done with ARPE19 cells transfec ted with EGFP-tagged full-length A3B (A3B-FL), A3B-NTD, or A3B-CTD constructs. After 72 h infection with TB40-mCherry, A3B-FL shows clear relocalization to the cytoplasmic compartment in comparison to the unchanged cell-wide EGFP control (Fig. 4A and B). Surprisingly, A3B-NTD, which shows nuclear localization in mock-infected cells, becomes predominantly cytoplasmic after infection (Fig. 4A and B). A3B-CTD has a cell-wide localization pattern that is not changed by virus infection (Fig. 4A and B). In contrast, EBV BORF2 has no effect on A3B-NTD nuclear localization, and it strongly promotes the relocalization of A3B-FL and A3B-CTD into cytoplasmic aggregates (Fig. 4C). These data combine to show that A3B-NTD is sufficient for A3B subcellular redistribution during HCMV infection and additionally distinguish the molecular mechanism from that mediated by the large RNR subunit of gamma- and alpha-herpesvirus. A3B relocalization occurs early during infection and requires de novo HCMV protein expression During a productive HCMV infection, viral genes are expressed chronologically in three main groups (28). IE genes are expressed at between 2 and 6 hpi, early (E) genes between 4 and 12 hpi, and late (L) genes following the onset of viral DNA replication (~24 hpi). To investigate the kinetics of A3B relocalization during HCMV infection, HFF-1 cells stably expressing A3B-HA were infected with TB40-mCherry or AD169-GFP and IF microscopy was performed at multiple timepoints after infection (6, 24, 48, and 72 hpi; Fig. 5A through D, respectively). This experiment shows that relocalization begins to occur rapidly with most infected cells exhibiting partial or full A3B-HA relocalization at the earliest timepoint 6 hpi. Moreover, the percentage of cells exhibiting cytoplasmic A3B- HA increases over time and is complete by 72 hpi. These kinetics suggest that an HCMV IE or E protein may be responsible for A3B relocalization during infection. To further investigate whether de novo viral protein expression is required for A3B relocalization, HFF-1 cells stably expressing A3B-HA were infected with AD169-GFP, treated for 24 h with the translation inhibitor cycloheximide (CHX) or dimethylsulfoxide (DMSO) as a control, and then subjected to IF microscopy (workflow in Fig. 6A). CHX treatment strongly prevents A3B-HA from relocalizing to the cytoplasm, whereas DMSO treatment does not (Fig. 6B and E). Similarly, cells infected with a recombinant AD169 lacking expression of the IE1 protein (AD169ΔIE1), which is required for expression of IE viral gene products (29), show no signs of A3B relocalization (Fig. 6C). In contrast, treating infected cells with phosphonoacetic acid (PAA), which blocks viral DNA synthesis and therefore also late protein expression (Fig. S4), fails to block A3B relocalization (Fig. 6D and E). Taken together with the rapid relocalization kinetics described above, these additional experiments implicate at least one HCMV IE/E protein in the A3B relocalization mechanism. DISCUSSION The recent discovery that alpha- and gamma-herpesviruses have evolved strategies to escape from A3-mediated restriction suggested that the beta-herpesvirus HCMV might utilize a similar mechanism to counteract this potent innate immune defense system. Our results demonstrate that HCMV, similar to other herpesviruses, is able to alter the subcellular localization of the A3B enzyme, relocating it from the nucleus to the cyto plasm. However, this A3B relocalization mechanism is mechanistically distinct from other herpesvirus families, first, by occurring in an RNR-independent manner and, second, by targeting the regulatory NTD of A3B. In contrast, gamma- and alpha-herpesviruses utilize the large viral RNR subunit to bind to the catalytic CTD of A3B to mediate relocalization. Moreover, the rapid kinetics of A3B relocalization and pharmacologic (CHX) and genetic (IE1) requirements described above suggest the involvement of at least one IE/E viral gene product. These results combine to support a working model in which at least one HCMV IE/E protein binds to the regulatory NTD of A3B, promotes its relocalization to the August Volume 97 Issue 8 10.1128/jvi.00781-23 8 Full-Length Text Journal of Virology FIG 4 The NTD of A3B is sufficient for A3B relocalization mediated by HCMV. (A) Representative IF microscopy images of ARPE19 cells transiently expressing EGFP alone, A3B-FL-EGFP, A3B-NTD-EGFP, and A3B-CTD-EGFP, incubated with medium alone (mock) or infected with TB40-mCherry for 72 h (10 µm scale). (B) Quantification of A3B-FL, A3B-NTD, and A3B-CTD subcellular localization phenotype shown in panel A. The dot-plot chart shows the ratio between nuclear and cytoplasmic fluorescence intensity (n > 25 cells per condition; P values obtained with unpaired Student’s t-tests). (C) Representative IF microscopy images of HeLa cells transiently expressing EBV BORF2-FLAG together with EGFP alone, A3B-FL-EGFP, A3B-NTD-EGFP, and A3B-CTD-EGFP (10 µm scale). cytoplasm, and thereby protects viral lytic DNA replication intermediates in the nucleus of the cell. A3-mediated restriction of herpesviruses, including HCMV, has been reported by multiple groups (30–33). A3A is upregulated in HCMV-infected decidual tissues and associated with hypermutation of the viral genome (30). Another study reported that A3G is upregulated after HCMV infection of fibroblasts, even if the upregulation does not appear to modulate HCMV replication (32). These studies are certainly interesting, and our work here has not formally excluded these A3s in HCMV restriction. However, given that neither of these A3s appears to be counteracted by HCMV (i.e., degraded or August Volume 97 Issue 8 10.1128/jvi.00781-23 9 Full-Length Text Journal of Virology FIG 5 A3B relocalization occurs early during HCMV infection. (A and C) Representative IF microscopy images of HFF-1 cells stably expressing A3B-HA incubated with medium alone (mock) or infected with the indicated HCMV strains for the indicated timepoints (10 µm scale). (B and D) Quantification of A3B-HA subcellular localization phenotypes shown in panels A and C. Each histogram bar reports the percentage of cells with whole cell, cytoplasmic, and nuclear A3B-HA (n > 100 cells per condition; mean ± SD with P values from unpaired Student’s t-tests). relocalized), in contrast to A3B described here, they are not likely to pose a significant threat to viral genomic integrity in vivo. In contrast, A3B is relocalized away from sites of viral replication by HCMV, which suggests that it may be a bona fide threat to the virus during lytic replication. This possibility is supported by the preferred sites of A3B- mediated deamination (5’TC) being depleted from HCMV genomes, consistent with a longer-term evolutionary conflict between this enzyme and HCMV (33, 34). However, this likelihood is difficult to quantify experimentally until the factor involved in A3B neutrali zation is identified, mutated, and shown to be essential for virus replication in the presence (but not absence) of A3B. A strength of the studies here is demonstrating A3B relocalization with multiple HCMV isolates and multiple different human cell types. Relocalization is also shown for August Volume 97 Issue 8 10.1128/jvi.00781-23 10 Full-Length Text Journal of Virology FIG 6 A3B relocalization requires de novo translation of HCMV proteins but not viral DNA synthesis. (A) Schematic representation of experimental workflows of CHX and PAA treatment of infected cells. Image created with BioRender. (B) Representative IF microscopy images of HFF-1 cells stably expressing A3B-HA incubated with medium alone (mock) or infected with AD169-GFP and treated with DMSO or CHX for 24 h (10 µm scale). (C) Representative IF microscopy images of U373 cells stably expressing A3B-HA incubated with medium alone (mock) or infected with AD169-GFP or AD169-GFP ΔIE1 for 72 h (10 µm scale). (D) Representative IF microscopy images of HFF-1 cells stably expressing A3B-HA incubated with medium alone (mock) or infected with AD169-GFP and treated with DMSO or PAA for 48 h (10 µm scale). (E) Quantification of A3B-HA subcellular localization phenotypes shown in panels B and D. Each histogram bar reports the percentage of cells with cytoplasmic A3B-HA (n > 80 cells per condition; mean ± SD with P values from unpaired Student’s t-tests). endogenous A3B as well as epitope tagged constructs. The overall A3B cytoplasmic localization phenotype is similar to that of gamma- and alpha-herpesviruses except that it is an ambiguously RNR independent. However, a limitation to the studies here is an inability to determine whether the molecular mechanism involves active nuclear export of pre-existing A3B and/or a block to nuclear import of newly translated protein. Another August Volume 97 Issue 8 10.1128/jvi.00781-23 11 Full-Length Text Journal of Virology limitation is the exclusive use of fluorescent microscopy for phenotypic determination. However, the speed at which relocalization is first detectable (6 hpi), pharmacologic inhibition through protein synthesis inhibition (CHX treatment), and genetic inhibition by preventing IE gene expression (ΔIE1) combine to support a model in which expression of at least one HCMV IE/E gene product is required for A3B relocalization. An alternative possibility is the involvement of a late gene product due to residual late gene expression following PAA treatment (Fig. S4). A major open question is, therefore, the identity of the viral factor(s) and potentially cellular co-factor(s) involved in this process. Our studies here add HCMV to the list of herpesviruses that modulate A3B subcellular localization, suggesting that this function may be part of a counteraction mechanism essential for viral infection. Our studies are also consistent with the likelihood that this host–pathogen conflict is conserved evolutionarily (8, 11) with ancient origins and ongoing functionality to present day. It is surprising, however, that the mechanisms differ so dramatically on the molecular level such that HCMV (and perhaps other beta-herpes viruses) has evolved a distinct RNR-independent mechanism. If A3B neutralization proves essential for HCMV replication and pathogenesis, it may be possible in the future to drug the neutralization mechanism and enable natural restriction of viral infections. MATERIALS AND METHODS Cell culture Cells were cultured at 37°C in a 5% CO2 atmosphere in a Thermo Forma incubator (Thermo Fisher Scientific, Waltham, MA, USA). HFF-1 (ATCC, Manassas, VA, USA), U373 (ATCC), and 293T cells were cultured in Dulbecco's Modified Eagle Medium (DMEM; Cytiva, Marlborough, MA, USA) supplemented with 10% fetal bovine serum (Gibco, Billings, MT, USA) and 1% penicillin/streptomycin (Gibco). ARPE19 cells (ATCC) were cultured in DMEM:F12 media (Gibco) supplemented with 10% fetal bovine serum (Gibco) and 1% penicillin/streptomycin (Gibco). HeLa cells (ATCC) were cultured in RPMI 1640 (Corning) supplemented with 10% fetal bovine serum (Gibco) and 1% penicillin/strepto mycin (Gibco). All cells were checked periodically for Mycoplasma and they always tested negative. Viruses and infections Viruses used in this study were as follows: TB40-mCherry [construction described in reference (35)]; AD169-GFP [construction described in reference (36)]; AD169- GFP-ΔUL45 [construction described in reference (26)]; and AD169ΔIE1 [construc tion described in reference (29)]. HCMV strain Merlin (GenBank accession NC 006273.2) was purchased from the ATCC. The strain FIX and its mutant FIXΔUL45 IRCCS Policlinico San Matteo, were a gift by Dr. Elena Percivalle (Fondazione in Fig. Italy) (37). The TB40-BAC4 and TB40-BAC4-UL45Stop strains used Pavia, 3E were produced using a markerless two-step RED-GAM recombination protocol (38, 39). To obtain the BAC of the mutant TB40/E UL45stop, the following pri mers were employed: UL45Stop_Fw: 5′- ATCTACCTGATTTCTTTGTTCTTTTCCTCGTAAACT TATGTAGACTCCGGCTGACGCGGACGAAGGATGACGACGATAAGTAGGG -3′; UL45Stop_Rv: 5′- CCGAGGACACCCGCTGTTCCTCGTCCGCGTCAGCCGGAGTCTACATAAGTTTACGAGGAA AAGCAACCAATTAACCAATTCTGATTAG -3′. All generated recombinant BAC DNAs were controlled for integrity and correctness by sequencing the mutated region. HFF-1 cells were used for the reconstitution of recombinant viruses and virus stock production. Viruses were then propagated by standard procedures as described (40). Briefly, HFF-1 cells were infected with a multiplicity of infection (MOI) of 0.01. When robust cytopathic effect was observed (between 7 and 14 d) cells were harvested. Then, centrifugation was performed at 15,000 × g for 30 min. Cell pellets were resuspended in complete media plus 15% sucrose phosphate buffer and sonicated on ice 4× for 10 s with 15 s between pulse. Centrifugation was performed at 1,300 × g for 5 min. Supernatant was collected, August Volume 97 Issue 8 10.1128/jvi.00781-23 12 Full-Length Text Journal of Virology aliquoted, and frozen at −80°C. The viral titers were calculated using the 50% tissue culture infection dose method upon infection of HFF-1 cells with serially diluted viral supernatants. In all experiments, HFF-1, U373, and ARPE19 were infected with HCMV at an MOI of 3 PFU/cell by diluting the virus into the medium, allowing adsorption for 2 h and replacing the viral dilution with fresh medium. IF microscopy For IF imaging of HCMV-infected cells, 5 × 104 cells/well were seeded in a 24-well plate. After 24 h, cells were transduced with lentiviruses encoding for human A3B-HA or A3B-E255A-HA (Fig. 1A through E; Fig. 2A through E; Fig. 3D). 48 h after transduction, cells were infected with TB40-mCherry or AD169-GFP for up to 72 h as indicated in figure legends. In Fig. 6B and D, DMSO, CHX (100 µg/mL), or PAA (100 µg/mL) was added to the virus dilution, and after 2 h, when virus was removed, cells were incubated with fresh media and compounds for 24 h (CHX) and 48 h (PAA). Cells were fixed in 4% formaldehyde for 15 min, permeabilized in 0.2% Triton X-100 in phosphate-buffered saline (PBS) for 10 min, washed three times for 5 min in PBS, and incubated in blocking buffer [2.8 mM KH2PO4, 7.2 mM K2HPO4, 5% goat serum (Gibco), 5% glycerol, 1% cold water fish gelatin (Sigma, St Louis, MO, USA), 0.04% sodium azide (pH 7.2)] for 1 h. Cells were then incubated with primary rabbit anti-HA (1:2,000) (cat #3724, Cell Signaling, Danvers, MA, USA) or purified rabbit anti-A3B 5210-87-13 [1:300 (25); Fig. 2G], or mouse anti-HCMV-IE1 (1:2,000) (cat #MAB810R, EMD Millipore-Sigma, Burlington, MA, USA) (Fig. S2A) overnight at 4°C. Cells were washed three times for 5 min with PBS and then incubated with the secondary antibodies goat anti-rabbit IgG Alexa Fluor 488 (1:500) (cat #A11034, Invitrogen, Waltham, MA, USA), or goat anti-rabbit IgG Alexa Fluor 594 (1:500) (cat #A11037, Invitrogen), or goat anti-mouse IgG Alexa Fluor 488 (1:500) (cat #A11001, Invitrogen) for 2 h at room temperature in the dark. Cells were then counterstained with 1 µg/mL Hoechst 33342 for 20 min and rinsed twice for 5 min in PBS. For IF imaging of transfected cells, 5 × 104/well HeLa cells were transfected with plasmids expressing 200 ng pcDNA4-BORF2-FLAG or 200 ng pcDNA4-UL45-FLAG, and 100 ng pcDNA3.1-A3B-HA (Fig. 3C). 5 × 104/well ARPE19 cells were transfected with plasmids expressing 100 ng pcDNA5TO-A3B-EGFP, pcDNA5TO-A3B-NTD-EGFP, pcDNA5TO-A3B-CTD-EGFP (Fig. 4A), and pcDNA4-BORF2-FLAG (Fig. 4C). Empty vector pcDNA3.1 or pcDNA3.1 encoding A3B-HA or other A3x-HA proteins is used in Fig. S1. After 48 h, IF was performed as described above. Cells were stained with primary antibodies mouse anti-FLAG (1:2,000) (cat #F1804, Sigma) and rabbit anti-HA (1:2,000) (cat #3724, Cell Signaling, Danvers, MA, USA) overnight at 4°C to detect FLAG-tagged RNRs and HA-tagged A3B, respectively. Goat anti-mouse IgG Alexa Fluor 488 (1:500) (cat #A11001, Invitrogen) and goat anti-rabbit IgG Alexa Fluor 594 (1:500) (cat #A11037, Invitrogen) were used as secondary antibodies. Images in Fig. 1A (×60 magnification) and Fig. 3E (×20 magnification) were collected using an Eclipse Ti2 (Nikon). All other images were collected at ×20 magnification using an EVOS FL Cell Imaging System (Thermo Fisher Scientific). Scale bars are indicated in each panel and figure legend. Quantification was performed using Image J software, counting the percentage of cells with relocalized A3B or the ratio of nuclear/cytoplasmic A3B. Quantification was performed by counting cells from n = 3 independent experimen tal replicates. GraphPad Prism 9 was used to prepare graphs and statistical analyses (unpaired Student’s t-test). Co-IP experiments 293T (2.5 × 105/well) cells were grown in 6-well plates and transfected with pcDNA3.1 plasmids encoding human A3A-HA, A3B-HA, and A3G-HA together or not with pcDNA4- BORF2-FLAG or pcDNA4-UL45-FLAG, and 6 µL TransIT-LT1 (Mirus, Madison, WI, USA) in 200 µL serum-free Opti-MEM (Thermo Fisher Scientific). After 48 h, whole cells were harvested in 300 µL of ice-cold lysis buffer [150 mM NaCl, 50 mM Tris-HCl, 10% glycerol, 1% IGEPAL (Sigma), and complete ethylenediaminetetraacetic acid (EDTA)-free protease August Volume 97 Issue 8 10.1128/jvi.00781-23 13 Full-Length Text Journal of Virology inhibitor cocktail (Roche); pH 7.4]. Cells were vortexed, incubated on ice for 30 min, and then sonicated. Whole-cell lysates (30 µL) were aliquoted for input detection. Lysed cells were centrifuged at 13,000 rpm for 15 min to pellet debris, and the supernatant was resuspended with 25 µL anti-FLAG M2 magnetic beads (Sigma) for overnight incubation at 4°C with gentle rotation. Beads were washed three times in 700 µL of lysis buffer. Bound protein was eluted in 30 µL of elution buffer [0.15 mg/mL 3xFLAG peptide (Sigma) in 150 mM NaCl, 50 mM Tris-HCl, 10% glycerol, and 0.05% Tergitol; pH 7.4]. Input and eluted proteins were analyzed by immunoblots. Immunoblots In Fig. 3A, membranes were stained with mouse anti-FLAG (1:5,000) (cat #3724, Sigma), mouse anti-tubulin (1:10,000) (cat # T5168, Sigma), and rabbit anti-HA (1:3,000) (cat #3724, Cell Signaling). After washing, membranes were incubated with an anti-rabbit IgG horseradish peroxidase-conjugated secondary antibody (1:10,000) (cat #211032171, Jackson ImmunoResearch, West Grove, PA, USA) and an anti-mouse IRDye 800CW (1:10,000) (cat #C70919-05, LI-COR, Lincoln, NE, USA). In Fig. S4, membranes were stained with mouse anti-CMV pp65 antibody (cat #53489, Abcam, Cambridge, UK) (1:1,000) and mouse anti-tubulin (1:10,000) (cat # T5168, Sigma). Secondary antibody was a goat anti-mouse IRDye 680LT (1:10,000) (cat #926-68020, LI-COR). DNA deaminase activity assays 293T (5 × 105/well) cells were seeded into 6-well plates and, after 24 h, transfected with 200 ng pcDNA4-BORF2-FLAG or 200 ng pcDNA4-UL45-FLAG, and 100 ng pcDNA3.1-A3B- HA. After 48 h, cells were harvested, resuspended in 100 µL of reaction buffer (25 mM HEPES, 15 mM EDTA, 10% glycerol, 1 tablet of Sigma-Aldrich cOmplete Protease Inhibitor Cocktail), and sonicated at the lowest setting. Whole-cell lysates were then centrifuged at 10,000 × g for 20 min. The clarified supernatant was incubated with 4 pmol of oligonu cleotide (5′- ATTATTATTATTCAAATGGATTTATTTATTTATTTATTTATTT-fluorescein), 0.025 U uracil DNA glycosylase (UDG), 1x UDG buffer (NEB), and 1.75 U RNase A at 37°C for 2 h. Deamination mixtures were treated with 100 mM NaOH at 95°C for 10 min. Samples were then separated on 15% Tris-borate-EDTA-urea gel. Fluorescence was measured using a Typhoon FLA-7000 image reader (Fig. 3B). ACKNOWLEDGMENTS We thank members of the Harris laboratory for support and constructive feedback. These studies were supported by NIAID R37-AI064046 to R.S.H., NCI P01-CA234228 to R.S.H. and a Recruitment of Established Investigators Award from the Cancer Prevention and Research Institute of Texas (CPRIT RR220053 to RSH). Salary support for A.Z.C. was provided in part by NIH training grants NCI F30-CA200432 and NIGMS T32-GM008244. Salary support for A.A.A. was provided in part by NIAID T32-AI83196 from the Univer sity of Minnesota’s Institute for Molecular Virology Training program. Salary support for S.N.M. was provided by NIAID F31-AI161910 and subsequently an HHMI Gilliam Fellowship. R.S.H. is an Investigator of the Howard Hughes Medical Institute and the Ewing Halsell President’s Council Distinguished Chair. The authors have no competing interests to declare. E.F. and R.S.H. conceptualized the study. E.F., A.Z.C., A.A.A., and B.S. performed experiments. E.F. curated the data, generated figures, and was responsible for formal data analyses. E.F. and R.S.H. wrote the initial draft of the paper and all authors contributed to revisions. S.N.M., J.R.L., C.J.B., W.A.B., V.D.O., M.B. and R.S.H. provided resources. R.S.H. was responsible for funding acquisition. August Volume 97 Issue 8 10.1128/jvi.00781-23 14 Journal of Virology Full-Length Text AUTHOR AFFILIATIONS 1Department of Biochemistry and Structural Biology, University of Texas Health San Antonio, San Antonio, Texas, USA 2Department of Life and Environmental Sciences, University of Cagliari, Cittadella Universitaria di Monserrato, Cagilari, Italy 3Department of Biochemistry, Molecular Biology and Biophysics, University of Minnesota, Minneapolis, Minnesota, USA 4Howard Hughes Medical Institute, University of Texas Health San Antonio, San Antonio, Texas, USA 5Department of Medicine, University of Minnesota, Minneapolis, Minnesota, USA 6Department of Public Health and Pediatric Sciences, University of Turin, Turin, Italy 7Department of Pediatrics, Division of Pediatric Infectious Diseases and Immunology, University of Minnesota, Minneapolis, Minnesota, USA 8Department of Microbiology and Immunology, University of Minnesota, Minneapolis, Minnesota, USA AUTHOR ORCIDs Elisa Fanunza Reuben S. Harris http://orcid.org/0000-0002-2044-8473 http://orcid.org/0000-0002-9034-9112 FUNDING Funder HHS \| NIH \| National Institute of Allergy and Infectious Diseases (NIAID) Grant(s) Author(s) R37-AI064046 Reuben S. Harris HHS \| NIH \| National Cancer Institute (NCI) P01-CA234228 Reuben S. Harris AUTHOR CONTRIBUTIONS Elisa Fanunza, Conceptualization, Formal analysis, Investigation, Methodology, Visualiza tion, Writing – original draft, Writing – review and editing \| Adam Z. Cheng, Formal analysis, Investigation, Supervision, Writing – review and editing \| Ashley A. Auerbach, Investigation, Methodology, Writing – review and editing \| Bojana Stefanovska, Formal analysis, Writing – review and editing \| Sofia N. Moraes, Formal analysis, Writing – review and editing \| James R. Lokensgard, Formal analysis, Writing – review and editing \| Matteo Biolatti, Formal analysis, Writing – review and editing \| Valentina Dell'Oste, Formal analysis, Writing – review and editing \| Craig J. Bierle, Formal analysis, Writing – review and editing \| Wade A. Bresnahan, Formal analysis, Writing – review and editing \| Reuben S. Harris, Conceptualization, Formal analysis, Funding acquisition, Project administration, Supervision, Writing – original draft, Writing – review and editing ADDITIONAL FILES The following material is available online. Supplemental Material Supplemental figures (JVI00781-23-S0001.pdf). Figures S1 to S4 REFERENCES 1. 2. 3. Green AM, Weitzman MD. 2019. The spectrum of APOBEC3 activity: from anti-viral agents to anti-cancer opportunities. DNA Repair (Amst) 83:102700. https://doi.org/10.1016/j.dnarep.2019.102700 Harris RS, Dudley JP. 2015. APOBECs and virus restriction. Virology 479– 480:131–145. https://doi.org/10.1016/j.virol.2015.03.012 Hakata Y, Miyazawa M. 2020. Deaminase-independent mode of in human and Mouse APOBEC3 proteins. antiretroviral action 4. 8:1976. https://doi.org/10.3390/microorgan Microorganisms isms8121976 Jäger S, Kim DY, Hultquist JF, Shindo K, LaRue RS, Kwon E, Li M, Anderson BD, Yen L, Stanley D, Mahon C, Kane J, Franks-Skiba K, Cimermancic P, Burlingame A, Sali A, Craik CS, Harris RS, Gross JD, Krogan NJ. 2011. Vif hijacks CBF-β to degrade APOBEC3G and promote HIV-1 infection. Nature 481:371–375. https://doi.org/10.1038/nature10693 August Volume 97 Issue 8 10.1128/jvi.00781-23 15 Full-Length Text Journal of Virology 5. 6. 7. 8. 9. 10. Sheehy AM, Gaddis NC, Choi JD, Malim MH. 2002. Isolation of a human gene that inhibits HIV-1 infection and is suppressed by the viral Vif protein. Nature 418:646–650. https://doi.org/10.1038/nature00939 Cheng AZ, Yockteng-Melgar J, Jarvis MC, Malik-Soni N, Borozan I, Carpenter MA, McCann JL, Ebrahimi D, Shaban NM, Marcon E, Greenblatt J, Brown WL, Frappier L, Harris RS. 2019. Epstein-Barr virus BORF2 inhibits cellular APOBEC3B to preserve viral genome integrity. Nat Microbiol 4:78–88. https://doi.org/10.1038/s41564-018-0284-6 Cheng AZ, Moraes SN, Attarian C, Yockteng-Melgar J, Jarvis MC, Biolatti M, Galitska G, Dell’Oste V, Frappier L, Bierle CJ, Rice SA, Harris RS. 2019. A conserved mechanism of APOBEC3 relocalization by herpesviral ribonucleotide reductase large subunits. J Virol 93:e01539-19. https:// doi.org/10.1128/JVI.01539-19 Moraes SN, Becker JT, Moghadasi SA, Shaban NM, Auerbach AA, Cheng AZ, Harris RS. 2022. Evidence linking APOBEC3B genesis and evolution of innate immune antagonism by gamma-herpesvirus ribonucleotide reductases. Elife 11:e83893. https://doi.org/10.7554/eLife.83893 Cheng AZ, Moraes SN, Shaban NM, Fanunza E, Bierle CJ, Southern PJ, Bresnahan WA, Rice SA, Harris RS. 2021. APOBECs and herpesviruses. Viruses 13:390. https://doi.org/10.3390/v13030390 Stewart JA, Holland TC, Bhagwat AS. 2019. Human herpes simplex virus-1 depletes APOBEC3A from nuclei. Virology 537:104–109. https:// doi.org/10.1016/j.virol.2019.08.012 11. Auerbach AA, Becker JT, Moraes SN, Moghadasi SA, Duda JM, Salamango DJ, Harris RS. 2022. Ancestral APOBEC3B nuclear is maintained in humans and apes and altered in most other old world primate species. mSphere 7:e0045122. https://doi.org/10.1128/msphere. 00451-22 Tyl MD, Betsinger CN, Cristea IM. 2022. Virus–host protein interactions as footprints of human cytomegalovirus replication. Curr Opin Virol 52:135–147. https://doi.org/10.1016/j.coviro.2021.11.016 localization 12. 13. Griffiths P, Reeves M. 2021. Pathogenesis of human cytomegalovirus in the immunocompromised host. Nat Rev Microbiol 19:759–773. https:// doi.org/10.1038/s41579-021-00582-z 17. 16. 19. 15. 18. 14. Manicklal S, Emery VC, Lazzarotto T, Boppana SB, Gupta RK. 2013. The "silent" global burden of congenital cytomegalovirus. Clin Microbiol Rev 26:86–102. https://doi.org/10.1128/CMR.00062-12 Britt WJ. 2017. Congenital human cytomegalovirus infection and the enigma of maternal immunity. J Virol 91:15. https://doi.org/10.1128/JVI. 02392-16 Turner DL, Mathias RA. 2022. The human cytomegalovirus decathlon: ten critical replication events provide opportunities for restriction. Front Cell Dev Biol 10:1053139. https://doi.org/10.3389/fcell.2022.1053139 Biolatti M, Gugliesi F, Dell’Oste V, Landolfo S. 2018. Modulation of the innate immune response by human cytomegalovirus. Infect Genet Evol 64:105–114. https://doi.org/10.1016/j.meegid.2018.06.025 Patro ARK. 2019. Subversion of immune response by human cytomega lovirus. Front Immunol 10:1155. https://doi.org/10.3389/fimmu.2019. 01155 Salamango DJ, McCann JL, Demir Ö, Brown WL, Amaro RE, Harris RS. 2018. APOBEC3B nuclear localization requires two distinct N-terminal domain surfaces. J Mol Biol 430:2695–2708. https://doi.org/10.1016/j. jmb.2018.04.044 Pak V, Heidecker G, Pathak VK, Derse D. 2011. The role of amino-terminal sequences in cellular localization and antiviral activity of APOBEC3B. J Virol 85:8538–8547. https://doi.org/10.1128/JVI.02645-10 Caval V, Bouzidi MS, Suspène R, Laude H, Dumargne MC, Bashamboo A, Krey T, Vartanian JP, Wain-Hobson S. 2015. Molecular basis of the attenuated phenotype of human APOBEC3B DNA mutator enzyme. Nucleic Acids Res 43:9340–9349. https://doi.org/10.1093/nar/gkv935 Burns MB, Lackey L, Carpenter MA, Rathore A, Land AM, Leonard B, Refsland EW, Kotandeniya D, Tretyakova N, Nikas JB, Yee D, Temiz NA, Donohue DE, McDougle RM, Brown WL, Law EK, Harris RS. 2013. APOBEC3B is an enzymatic source of mutation in breast cancer. Nature 494:366–370. https://doi.org/10.1038/nature11881 Lackey L, Law EK, Brown WL, Harris RS. 2013. Subcellular localization of the APOBEC3 proteins during mitosis and implications for genomic DNA deamination. Cell Cycle 12:762–772. https://doi.org/10.4161/cc.23713 24. Nikkilä J, Kumar R, Campbell J, Brandsma I, Pemberton HN, Wallberg F, Nagy K, Scheer I, Vertessy BG, Serebrenik AA, Monni V, Harris RS, Pettitt SJ, Ashworth A, Lord CJ. 2017. Elevated APOBEC3B expression drives a 21. 23. 22. 20. 25. 26. 27. replication stress-related signature and kataegic-like mutation therapeutic vulnerabilities in p53-defective cells. Br J Cancer 117:113– 123. https://doi.org/10.1038/bjc.2017.133 Brown WL, Law EK, Argyris PP, Carpenter MA, Levin-Klein R, Ranum AN, Molan AM, Forster CL, Anderson BD, Lackey L, Harris RS. 2019. A rabbit monoclonal antibody against the antiviral and cancer genomic DNA mutating enzyme APOBEC3B. Antibodies (Basel) 8:47. https://doi.org/10. 3390/antib8030047 Yu D, Silva MC, Shenk T. 2003. Functional map of human cytomegalovi rus AD169 defined by global mutational analysis. Proc Natl Acad Sci U S A 100:12396–12401. https://doi.org/10.1073/pnas.1635160100 Shaban NM, Yan R, Shi K, Moraes SN, Cheng AZ, Carpenter MA, McLellan JS, Yu Z, Harris RS. 2022. Cryo-EM structure of the EBV ribonucleotide reductase BORF2 and mechanism of APOBEC3B inhibition. Sci Adv 8:eabm2827. https://doi.org/10.1126/sciadv.abm2827 28. Weekes MP, Tomasec P, Huttlin EL, Fielding CA, Nusinow D, Stanton RJ, Wang ECY, Aicheler R, Murrell I, Wilkinson GWG, Lehner PJ, Gygi SP. 2014. Quantitative temporal viromics: an approach to investigate host- pathogen interaction. Cell 157:1460–1472. https://doi.org/10.1016/j.cell. 2014.04.028 29. Mocarski ES, Kemble GW, Lyle JM, Greaves RF. 1996. A deletion mutant in the human cytomegalovirus gene encoding IE1491aa is replication defective due to a failure in autoregulation. Proc Natl Acad Sci U S A 93:11321–11326. https://doi.org/10.1073/pnas.93.21.11321 32. 31. 30. Weisblum Y, Oiknine-Djian E, Zakay-Rones Z, Vorontsov O, Haimov- Kochman R, Nevo Y, Stockheim D, Yagel S, Panet A, Wolf DG. 2017. APOBEC3A is upregulated by human cytomegalovirus (HCMV) in the maternal-fetal interface. J Virol 91:e01296-17. https://doi.org/10.1128/ JVI.01296-17 Suspène R, Aynaud M-M, Koch S, Pasdeloup D, Labetoulle M, Gaertner B, Vartanian J-P, Meyerhans A, Wain-Hobson S. 2011. Genetic editing of herpes simplex virus 1 and Epstein-Barr herpesvirus genomes by human APOBEC3 cytidine deaminases in culture and in vivo. J Virol 85:7594– 7602. https://doi.org/10.1128/JVI.00290-11 Pautasso S, Galitska G, Dell’Oste V, Biolatti M, Cagliani R, Forni D, De Andrea M, Gariglio M, Sironi M, Landolfo S. 2018. Strategy of human cytomegalovirus to escape interferon beta-induced APOBEC3G editing activity. J Virol 92:e01224-18. https://doi.org/10.1128/JVI.01224-18 Poulain F, Lejeune N, Willemart K, Gillet NA. 2020. Footprint of the host restriction factors APOBEC3 on the genome of human viruses. PLoS Pathog 16:e1008718. https://doi.org/10.1371/journal.ppat.1008718 Shapiro M, Meier S, MacCarthy T. 2018. The cytidine deaminase under- representation reporter (CDUR) as a tool to study evolution of sequences under deaminase mutational pressure. BMC Bioinformatics 19:256. https://doi.org/10.1186/s12859-018-2259-2 34. 33. 35. O’Connor CM, Shenk T. 2011. Human cytomegalovirus pUS27 G protein- coupled receptor homologue is required for efficient spread by the extracellular route but not for direct cell-to-cell spread. J Virol 85:3700– 3707. https://doi.org/10.1128/JVI.02442-10 Cantrell SR, Bresnahan WA. 2005. Interaction between the human cytomegalovirus UL82 gene product (pp71) and hDaxx regulates immediate-early gene expression and viral replication. J Virol 79:7792– 7802. https://doi.org/10.1128/JVI.79.12.7792-7802.2005 36. 38. Virol characteristics. 37. Hahn G, Khan H, Baldanti F, Koszinowski UH, Revello MG, Gerna G. 2002. The human cytomegalovirus ribonucleotide reductase homolog UL45 is dispensable for growth in endothelial cells, as determined by a BAC- cloned clinical isolate of human cytomegalovirus with preserved wild- type 76:9551–9555. J https://doi.org/10.1128/jvi.76.18.9551-9555.2002 Sinzger C, Hahn G, Digel M, Katona R, Sampaio KL, Messerle M, Hengel H, Koszinowski U, Brune W, Adler B. 2008. Cloning and sequencing of a highly productive, endotheliotropic virus strain derived from human cytomegalovirus TB40/E. J Gen Virol 89:359–368. https://doi.org/10. 1099/vir.0.83286-0 Tischer BK, Smith GA, Osterrieder N. 2010. En passant mutagenesis: a two markerless red recombination system. Methods Mol Biol 634:421– 430. https://doi.org/10.1007/978-1-60761-652-8_30 Britt WJ. 2010. Human cytomegalovirus: propagation, quantification, and storage. Curr Protoc Microbiol Chapter 14:Unit 14E.3. https://doi. org/10.1002/9780471729259.mc14e03s18 39. 40. August Volume 97 Issue 8 10.1128/jvi.00781-23 16
10.1371_journal.pbio.3002093	RESEARCH ARTICLE Fibroblast-induced mammary epithelial branching depends on fibroblast contractility Jakub Sumbal1,2,3, Silvia Fre2, Zuzana Sumbalova KoledovaID 1¤* 1 Masaryk University, Faculty of Medicine, Department of Histology and Embryology, Brno, Czech Republic, 2 Institut Curie, Laboratory of Genetics and Developmental Biology, INSERM U934, CNRS UMR3215, PSL Universite´ Paris, Paris, France, 3 Sorbonne Universite´, Collège Doctoral, Paris, France ¤ Current address: Institute of Molecular Genetics of the Czech Academy of Sciences, Prague, Czech Republic * [email protected], [email protected] Abstract Epithelial branching morphogenesis is an essential process in living organisms, through which organ-specific epithelial shapes are created. Interactions between epithelial cells and their stromal microenvironment instruct branching morphogenesis but remain incompletely understood. Here, we employed fibroblast-organoid or fibroblast-spheroid co-culture sys- tems and time-lapse imaging to reveal that physical contact between fibroblasts and epithe- lial cells and fibroblast contractility are required to induce mammary epithelial branching. Pharmacological inhibition of ROCK or non-muscle myosin II, or fibroblast-specific knock- out of Myh9 abrogate fibroblast-induced epithelial branching. The process of fibroblast- induced branching requires epithelial proliferation and is associated with distinctive epithelial patterning of yes associated protein (YAP) activity along organoid branches, which is dependent on fibroblast contractility. Moreover, we provide evidence for the in vivo exis- tence of contractile fibroblasts specifically surrounding terminal end buds (TEBs) of pubertal murine mammary glands, advocating for an important role of fibroblast contractility in branching in vivo. Together, we identify fibroblast contractility as a novel stromal factor driv- ing mammary epithelial morphogenesis. Our study contributes to comprehensive under- standing of overlapping but divergent employment of mechanically active fibroblasts in developmental versus tumorigenic programs. Introduction Efficient formation of large epithelial surfaces in limited organ volumes is achieved through branching morphogenesis [1]. The underlying processes of epithelial morphogenesis, includ- ing epithelial cell proliferation, migration, intercalation, differentiation, and death, are regu- lated by both internal genetic programs as well as external cues provided by systemic signals (such as hormones) and local organ-specific microenvironment [1–3]. The mammary gland is the ideal tissue paradigm for stochastically branching epithelia. Mammary morphogenesis starts in the embryo, but the majority of branch bifurcations and ductal elongation takes place postnatally during puberty. During this time epithelial morphogenesis is driven by terminal a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Sumbal J, Fre S, Sumbalova Koledova Z (2024) Fibroblast-induced mammary epithelial branching depends on fibroblast contractility. PLoS Biol 22(1): e3002093. https://doi.org/10.1371/ journal.pbio.3002093 Academic Editor: Emma Rawlins, University of Cambridge, UNITED KINGDOM Received: March 13, 2023 Accepted: November 24, 2023 Published: January 10, 2024 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pbio.3002093 Copyright: © 2024 Sumbal et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and its Supporting Information files. Funding: This work was supported by grants from the Grant Agency of Masaryk University (MU) PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 1 / 29 (https://gamu.muni.cz/; grants no. MUNI/G/1446/ 2018, MUNI/G/1775/2020 to Z.S.K., MUNI/A/1398/ 2021 and MUNI/A/1301/2022), from Internal Grant Agency of Faculty of Medicine MU (MUNI/11/SUP/ 06/2022 to Z.S.K. and MUNI/IGA/1314/2021 to J. S.), from Foundation pour la Recherche Me´dicale (FRM) (https://www.frm.org/; grant no. "FRM Equipes" EQU201903007821 to S.F., the Association for Research against Cancer (ARC) (https://www.fondation-arc.org/the-fondation-arc/; grant no. ARCPGA2021120004232_4874) to S.F., and from Czech Science Foundation (GAČR) (https://gacr.cz/; grant no. GA23-04974S to Z.S.K.). J.S. is supported by Barrande Fellowship (Ministry of Education, Youth and Sports; https://www. msmt.cz/), Fondation pour la Recherche Me´dicale (https://www.frm.org/; grant no. FDM202106013570), and by Brno PhD. Talent Scholarship, funded by the Brno City Municipality (https://www.jcmm.cz/projekt/brno_phd_talent/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. Abbreviations: 3SB, 3D staining buffer; CAF, cancer-associated fibroblast; ECM, extracellular matrix; FBS, fetal bovine serum; FGF2, fibroblast growth factor 2; FGFR, FGF receptor; IHC-IF, immunohistochemistry-immunofluorescence; ROI, region of interest; TEB, terminal end bud; YAP, yes associated protein. Fibroblast contractility is involved in epithelial branching end buds (TEBs), bulb-shaped structures containing proliferative stratified epithelium that invades the surrounding mammary stroma [4]. The microenvironment of the mammary epithelium is a dynamic entity that consists of extracellular matrix (ECM) and stromal cells, including fibroblasts. Fibroblasts lay adjacent to the epithelium and have been well recognized as master regulators of mammary epithelial morphogenesis during puberty through production of growth factors [5–9] and ECM mole- cules [5,7,9–14] necessary for mammary epithelial growth and branching [15]. However, the dynamics of the epithelial–fibroblast interactions during mammary branching morphogenesis as well as whether fibroblasts contribute to shaping of mammary epithelium through addi- tional mechanisms have remained unknown. Microenvironment of several developing organs has been shown to govern epithelial pat- terning by dynamic cues of mechanically active cells. Dermal cells in chick skin determine feather buds by mechanical contraction [16], intestinal vilification is dependent on compres- sion by smooth muscle cells [17], and embryonic lung mesenchyme promotes epithelial bifur- cation by mechanical forces [18–20]. However, it has not been elucidated whether the mammary microenvironment contains an instructive component of mechanically active cells as well. To answer this question, we performed live imaging and functional analysis of co-cultures of primary mammary epithelial organoids (isolated epithelial fragments with in vivo like archi- tecture consisting of inner luminal and outer myoepithelial (basal) cells) with primary mam- mary fibroblasts. Analogously to primary mammary organoids treated with fibroblast growth factor 2 (FGF2), a well-established model of mammary branching morphogenesis driven by paracrine signals [21], our in vitro co-culture model provides a unique window into fibro- blast–epithelial interactions during pubertal mammary branching morphogenesis. It enables visualization of stromal fibroblasts during dynamic morphogenetic processes, which are other- wise largely inaccessible in vivo due to light-scattering properties of mammary adipose tissue. In this work, we show that physical contact between fibroblasts and epithelial cells, and acto- myosin-dependent contractility of fibroblasts are required for branching morphogenesis. We demonstrate successful reconstitution of budding morphogenesis by 3D co-culture of contrac- tile fibroblasts in breast cancer spheroids that normally do not form buds. Moreover, by com- bining contractile fibroblasts with strong proliferative signals we reproduce TEB-like branching morphogenesis in organoid cultures and reveal localization of contractile fibroblasts around TEBs in the mammary glands, suggesting a role for fibroblast contractility in vivo. Our results reveal a novel role of fibroblast contractility in driving epithelial branching morphogenesis. Results Fibroblast-induced branching of organoids does not reproduce FGF2-induced budding To uncover the role of fibroblasts in epithelial morphogenesis, we investigated differences between organoid budding induced solely by paracrine factors (using primary mammary orga- noids exposed to exogenous FGF2) and organoid branching induced by fibroblasts (using organoids co-cultured with primary mammary fibroblasts in the absence of any exogenous growth factor). Addition of FGF2 or fibroblasts to mammary organoid cultures both induced branching of epithelial organoids (Fig 1A and 1B and S1 Movie) but examination of the result- ing organoid morphogenesis revealed important differences in dynamics and epithelial archi- tecture in the 2 conditions. First, organoids co-cultured with fibroblasts developed bigger branches, but the branches were less numerous (Fig 1A, 1C and 1D). Second, they branched PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 2 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 1. Fibroblast-induced branching of organoids does not reproduce FGF2-induced budding. (A) Snapshots from time- lapse imaging of primary mammary organoids in basal organoid medium (basal M) without any FGF supplementation (top), basal M with FGF2 (middle), or co-cultured with primary mammary fibroblasts (fibro) in basal M with no FGF supplementation (bottom). Scale bar: 100 μm. Full videos are presented in S1 Movie. (B) Quantification of percentage of branched organoids per all organoids in the conditions from (A). The plot shows mean ± SD, each dot represents biologically independent experiment, n = 3. Statistical analysis: Two-tailored t test. (C) Quantification of branch thickness from experiments in (A). The plot shows mean ± SD, each dot represents a biologically independent experiment, n = 3. Statistical analysis: Two-tailored t test. (D) Quantification of number of branches per branched organoids in conditions from (A). The plot shows mean ± SD, each lined dot shows mean from each experiment, each faint dot shows single organoid measurement, n = 3 biologically independent experiments, N = 20 organoids per experiment. Statistical analysis: Two-tailored t test. (E) Quantification of organoid circularity in conditions from (A). The lines represent mean, the shadows and error bars represent ± SD, n = 3 biologically independent experiments, N = 20 organoids per experiment. The schemes show representative shape of indicated circularity. (F) Detailed images of branch development in co-culture with fibroblasts from (A). Scale bar: 20 μm. (G) MIP of F-actin (red), DAPI (blue), and PDGFRα (white) in organoid with exogenous FGF2 or with fibroblasts (fibro). Zoom-in area from the box is depicted as MIP and single z slices. The asterisks denote lumen. Scale bar: 100 μm. (H) A scheme depicting differences between organoid budding induced by exogenous FGF2 and organoid branching in a co-culture with fibroblasts. The data underlying the graphs shown in the figure can be found in S1 Data. FGF2, fibroblast growth factor 2; MIP, maximum intensity projection. https://doi.org/10.1371/journal.pbio.3002093.g001 half-day to 1 day earlier than organoids treated with FGF2 (Fig 1A and 1E) and the branches were developed rapidly, including the development of negative curvature at the root of the branch (Fig 1F). Third, while FGF2-induced epithelial branching involved epithelial stratifica- tion as previously reported [21], co-culture with fibroblasts did not perturb the epithelial bilayer with its lumen (Fig 1A and 1G). These results suggest that fibroblasts and exogenous FGF2 drive organoid branching by different mechanisms. While the mechanism of FGF2-induced organoid budding was previously described in detail to begin with epithelial proliferation and stratification [22] followed by budding and ERK-dependent and proliferation-independent bud elongation [23], how fibroblasts induce organoid branching remains unanswered (Fig 1H). PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 3 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Endogenous paracrine signals are not sufficient to induce organoid branching in co-cultures The ability of exogenous FGF2 to promote organoid budding [21] (Fig 1A) well demonstrates the importance of paracrine signals for epithelial branching, although FGF2 amount used in in vitro branching assays likely exceeds physiological values in vivo. Therefore, we sought to determine, whether endogenous FGFs or other paracrine signals produced by fibroblasts in co-cultures [5,7] are sufficient to drive organoid branching. First to test the involvement of FGF signaling, we inhibited either FGF receptors (FGFRs) using SU5402, or ERK, a common signaling node of all receptor tyrosine kinases using U0126. As expected, both inhibitors abol- ished branching induced by exogenous FGF2 (Fig 2A and 2B). However, in the co-cultures with fibroblasts, the same concentration of inhibitors did not abolish branching, albeit slightly reduced organoid growth (Fig 2A and 2B), suggesting that paracrine signaling via FGFR-ERK pathway is not the only mechanism driving organoid branching. To probe the involvement of other paracrine signaling pathways in fibroblast-induced orga- noid branching, namely to test if fibroblast paracrine signaling alone is sufficient to induce organoid branching, or if other mechanisms involving fibroblast–epithelial proximity or con- tact are involved, we performed an array of different types of organoid (co-)culture set-ups (Fig 2C, top). When we provided unidirectional fibroblasts-to-epithelium paracrine signals by culture of organoids in fibroblast-conditioned medium, no organoid branching was observed (Fig 2C and 2D). When we allowed bidirectional paracrine signals by co-culture of organoids with fibroblasts in the same well but the organoids and fibroblasts were separated by a trans- well membrane or by a thick layer of Matrigel, no organoid branching was observed either (Fig 2C and 2D). However, when we allowed both paracrine signals and physical contact between organoids and fibroblasts by co-culturing them together either dispersed in Matrigel or as aggregates of fibroblasts on top of organoids embedded in Matrigel, we observed orga- noid branching (Fig 2C–2E). These results demonstrated an essential requirement of fibro- blast–epithelium contact for fibroblast-induced organoid branching, thus revealing that fibroblast-secreted paracrine factors are not sufficient to initiate branching. MCF7-ras spheroids recapitulate fibroblast-induced branching of organoids To further test the requirement of fibroblast–epithelium contact for fibroblast-induced epithe- lial branching, we developed a simpler co-culture system, where mammary fibroblasts were co-cultured with MCF7-ras breast cancer cell line spheroids (Fig 2F) instead of organoids from normal mammary epithelium. The advantage of MCF7-ras spheroids is that the spher- oids grow constantly due to constitutively active RAS GTPase and unlike normal epithelium they do not respond to exogenous FGF2 (Fig 2G) or EGF (S1A Fig) by branching. We found that similarly to normal epithelium, MCF7-ras spheroids remained round in fibroblast co-cul- tures, which did not allow physical contact with fibroblasts, but developed numerous buds when physical contact with fibroblasts was allowed (Fig 2H and S1B Fig). These results demon- strated that fibroblasts are able to promote epithelial budding even in a system that is morpho- genetically unresponsive to paracrine signals. Fibroblasts form physical contact with organoids To gain more insights into the mechanism of fibroblast-induced organoid branching, we examined organoid branching in the dispersed co-cultures in more detail. A day-by-day analy- sis of the contacts between fibroblasts and organoids revealed that the contacts are established PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 4 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 2. Endogenous paracrine signals are not sufficient to induce organoid branching. (A) Representative organoids cultured with exogenous FGF2 or with fibroblasts (fibro), treated with inhibitors of FGFR (SU5402) and MEK (U0126). Scale bar: 100 μm. (B) Quantification of branched organoid per all organoids, relative to mock. The plots show mean ± SD, each dot represents biologically independent experiment, n = 3–5, N = 20 organoids per experiment. Statistical analysis: Two- tailored t test. (C) Schemes and images on day 0 and day 4 of different organoid-fibroblast co-culture set-ups. Scale bar: 100 μm. (D) Quantification of organoid branching in different co-culture set-ups. The plot shows mean ± SD, each dot represents biologically independent experiment, n = 16 independent experiments for “no fibro,” 5 for “cond. medium,” 5 for “transwell,” 12 for ”bottom,” 16 for “dispersed,” and 4 for “aggregate,” N = 20 organoids per experiment. Statistical analysis: Multiple t tests compared to control “no fibro,” or indicated by the line. (E) Dispersed and aggregated co-culture of LifeAct- GFP fibroblast (white) and tdTomato organoid (red) at the beginning of the culture. Scale bar: 100 μm. (F) A scheme of MCF7-ras spheroid co-culture setup. (G) Representative MCF7-ras spheroids cultured in basal organoid medium (basal M) or basal M with exogenous FGF2. Scale bar: 100 μm. (H) Schemes and images on day 0 and day 4 of different spheroid- fibroblast co-culture set-ups. Top gray and red bars indicate proportion of branched spheroids out of all spheroids per condition, n = 3–5 independent experiments, N = 20 spheroids per experiment. Scale bar: 100 μm. The data underlying the graphs shown in the figure can be found in S1 Data. FGF2, fibroblast growth factor 2; FGFR, FGF receptor. https://doi.org/10.1371/journal.pbio.3002093.g002 from the first day and thus precede the branching events that occur on days 3 and 4 (S2A–S2C Fig). To corroborate this finding, we performed a co-culture experiment with organoids labeled by tdTomato and GFP-tagged fibroblasts. The time-lapse movies confirmed that PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 5 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching fibroblasts came in close contact with the epithelium early during the co-culture and remained there during branching (Fig 3A and S2 Movie). Confocal imaging analysis revealed that fibro- blasts (marked by PDGFRα) came in contact with all organoids (100% of 59 organoids analyzed in 3 independent biological replicates, S2D Fig) and contacted a larger proportion of the orga- noid middle sectional perimeter in round organoids than in branched organoids (S2E Fig). On the branched organoids, fibroblasts were predominantly located around the necks of the nascent branches and sat directly in contact with the epithelium (Fig 3B and 3C). Immuno- fluorescence staining of epithelial markers revealed that fibroblasts formed contacts with KRT5 positive myoepithelial cells (Fig 3D and 3E). Transmission electron microscopy of the co-cultures confirmed the close proximity between the fibroblasts and the epithelium, with a thin layer of ECM in between (Fig 3F). Using immunostaining we detected laminin 5, a basal membrane component, between the organoid and the adjacent fibroblast (Fig 3G and 3H). These data suggest that fibroblasts form contacts with epithelium via ECM. Fibroblast-induced epithelial branching depends on fibroblast contractility Based on observations from the time-lapse videos of organoids branched by fibroblasts (S1 and S2 Movies), we hypothesized that fibroblasts could constrict epithelium, folding it into branches. Immunofluorescence investigation of fibroblast-branched organoids revealed that fibroblasts in contact with the organoid formed a cellular loop, encircling the branch neck (Figs 4A and S3A–S3C and S3 Movie), and contained F-actin cables oriented mostly perpen- dicularly to the branch longitudinal axis (Fig 4A). Moreover, the fibroblasts stained positively for phosphorylated myosin light chain 2 (P-MLC2), a marker of active non-muscle myosin II (Fig 4B). Therefore, we examined the involvement of fibroblast contractility in fibroblast- induced organoid branching using small molecule inhibitors of non-muscle myosin II (bleb- bistatin) or ROCK1/2 (Y27632), 2 major nodes of cell contractility. The contractility inhibitors abrogated branching in co-cultures but did not inhibit organoid budding induced by exoge- nous FGF2 (Fig 4C and 4D; ROCK inhibition in FGF2-induced organoids even led to hyper- branched phenotype as previously described [21]). Similarly to organoid co-cultures, in the MCF7-ras spheroid co-culture model, spheroid budding was inhibited by addition of contrac- tility inhibitors (S4A–S4D Fig). Importantly, the contractility inhibitors did not diminish the capability of fibroblasts to migrate towards and contact the organoid (S5A–S5D Fig), in agree- ment with previous reports that showed that fibroblast migration in 3D is not abrogated by non-muscle myosin II inhibition [7]. Noteworthy, addition of the contractility inhibitors on day 3 of the co-culture, when branches were already formed, led to retraction of formed branches (S6A–S6D Fig), suggesting a role of contractility in branch maintenance as well as initiation. Because exogeneous treatment with pharmacological inhibitors in the culture medium affects both epithelial cells and fibroblasts, we genetically targeted exclusively in fibroblasts the contractility machinery gene myosin heavy chain 9 (Myh9), one of the 2 non-muscle myosin II heavy chain genes expressed in mammary fibroblasts (S7A and S7B Fig). Both siRNA-medi- ated Myh9 knock-down in wild-type fibroblasts and adenoviral Cre-mediated knock-out in Myh9fl/fl fibroblasts led to a decrease of organoid branching in co-cultures (Figs 4E, 4F and S8A–S8E and S4 and S5 Movies). To analyze if Myh9 knock-out affected the ability of fibro- blasts to migrate towards and contact the organoid, we took advantage of the mosaic nature of the adenovirus-mediated gene delivery. We quantified the amount of GFP+ fibroblasts (trans- duced by either Ad-Cre-GFP or Ad-GFP) and GFP-fibroblasts (not transduced by either of the adenoviruses) and compared their migration towards and contact with epithelium. We found no differences in their migration and epithelium-contacting abilities (S9A and S9B Fig). These PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 6 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 3. Fibroblasts in co-cultures are in physical contact with the epithelium. (A) Snapshots from time-lapse brightfield and fluorescence imaging of organoid (tdTomato) and fibroblast (GFP) co-culture (dispersed culture). Scale bar: 100 μm. Top line shows detail of fibroblast- organoid close interaction. Scale bar: 20 μm. (B, C) Images (B) and quantification (C) of the contact point between organoid (tdTomato) and PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 7 / 29 Bday 0day 1day 2day 3day 4t = 0h00't = 2h15't = 4h30't = 6h45't = 9h00'tdTomato (organoid)GFP (fibroblasts)tdTomato (organoid)GFP (fibroblasts)1detail 1detail 2detail 2detail 1AFDAPIfibroblastsepitheliumfibroblastLCMeCtdTomatoDAPILAMA5F-actintdTomatoF-actinLAMA5DAPI00.20.40.60.810.01.22.43.64.86.07.28.49.610.812.013.214.415.616.8Relative intensityDistance [μm]tipneck050100**P < 0.0001Structure in contactwith fibroblast [%]tdTomatoF-actinLAMA5fibroblastLCMeCGHC050100P = 0.0085Fibroblast in contactwith epith. cell [%]KRT8KRT5EDDAPIKRT5KRT8VIMPLOS BIOLOGYFibroblast contractility is involved in epithelial branching fibroblasts (GFP) on day 4 of co-culture (dispersed culture). Scale bar: 100 μm, scale bar in detail: 20 μm. (C) The plot shows mean ± SD, each dot represents 1 organoid, n = 5 experiments, N = 21 organoids. Statistical analysis: Two-tailored t test. (D) Images of the contact point between organoid (luminal (KRT8) and myoepithelial (KRT5) cells) and fibroblasts (VIM) on day 5 of co-culture (dispersed culture). Scale bar: 20 μm. (E) Quantification of fibroblasts in contact with KRT5+ or KRT8+ epithelial cells. The plot shows mean ± SD, each dot represents average from 1 biological replicate, n = 3 experiments, N = 14 organoids, 219 fibroblasts. Statistical analysis: Two-tailored t test. (F) Transmission electron micrographs and scheme (inset) of the contact point between luminal (LC, blue) and myoepithelial (MeC, magenta) cells and fibroblasts (green) on day 4 of co-culture (dispersed culture). Scale bar: 20 μm, scale bar in detail: 2 μm. In agreement with a published study (Ewald and colleagues), luminal cells are defined as lumen-facing cells, which present microvilli and numerous vesicles and granules. Myoepithelial cells are basally oriented, more elongated cells with less vesicles, granules, and organelles in the cytoplasm and they show a different electron density in their cytoplasm (it appears darker than the cytoplasm of luminal cells). The white arrowheads denote ECM between fibroblast and organoid. (G) Optical slice of organoid-fibroblast co-culture (dispersed culture), laminin 5 (cyan), DAPI (blue), F-actin (red), fibroblasts were isolated from R26-mT/mG mice (tdTomato, white). Scale bar: 100 μm, scale bar in detail: 10 μm. (H) A representative 1D relative fluorescence intensity plot. The measurement line is depicted in yellow (right). The data underlying the graphs shown in the figure can be found in S1 Data. ECM, extracellular matrix. https://doi.org/10.1371/journal.pbio.3002093.g003 results demonstrate that fibroblast contractility is not necessary for fibroblast migration towards organoids but is required to induce organoid branching in co-cultures. The need of fibroblast contractility for inducing epithelial branching suggests a mechani- cal signal transduction from fibroblasts to epithelium; therefore, we examined the subcellu- lar localization of yes associated protein (YAP), a mechano-sensor that in a resting cell resides in the cytoplasm but translocates to the nucleus upon mechanical stress [24]. We found that YAP specifically accumulated in the nuclei of epithelial cells in the neck region of epithelial branch of the co-culture, the region in contact with contractile fibroblasts (Fig 4G and 4H). Importantly, this pattern was not present in branches induced with FGF2 (Fig 4G and 4I), indicating that YAP activation in epithelial cells at the necks of elongating buds is induced by the contact with contractile fibroblasts and not simply by the overall shape of the epithelial bud. However, knockout of Myh9 in fibroblasts resulted in round organoids with no branching and a homogeneous distribution of nuclear YAP (Fig 4J). Our results show that while fibroblast contractility is necessary for the formation of branch with patterned YAP signal, the nuclear translocation of YAP can happen even in the absence of fibroblast contractility. Fibroblast-induced epithelial branching requires epithelial proliferation Activation of YAP signaling followed by its nuclear translocation is often associated with cell proliferation [24]. To investigate if such association is manifested in the organoids, we per- formed EdU labeling for proliferative cells. In the co-cultures, we detected highest cell prolifera- tion in the stalks of the branches (S10A–S10C Fig), i.e., in the areas of YAP nuclear localization (Fig 4G). In contrast, in the FGF2-treated organoids, no such pattern of EdU+ cells was observed (S10A, S10B and S10D Fig). To test whether epithelial proliferation (and thus expansion) plays a role in organoid branching in co-cultures, we inhibited cell proliferation using aphidicolin (DNA polymerase inhibitor), upon which we observed a severe defect in organoid branching (S10E–S10G Fig). To test for the possibility that the observed effect could be caused by inhibition of fibroblast proliferation, we performed the experiment also with fibroblasts pretreated with mitomycin C, an irreversible DNA synthesis blocker (S10E Fig). The pretreatment of fibroblast with mitomy- cin C had no effect on the result (S10E–S10G Fig), demonstrating that fibroblast proliferation is dispensable while epithelial proliferation is necessary for organoid branching in co-cultures. In concordance with the results from organoid co-cultures, in the MCF7-ras spheroid co-cul- ture model the blockage of spheroid proliferation by mitomycin C pretreatment of spheroids decreased spheroid size expansion and dramatically decreased branching of the MCF7-ras spheroids (S11A and S11B Fig). PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 8 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 4. Fibroblast-induced branching requires fibroblast contractility. (A) Maximum projection (left), detailed maximum projection (middle), and detail optical sections (right) of organoid branch induced in co-culture with fibroblasts (day 4 of dispersed culture). DAPI (blue), fibroblasts prelabeled with Ad-LifeAct-GFP (white). Scale bar: 50 μm, scale bar in detail: 20 μm. (B) Full maximum projection (left) and 10 μm maximum projection (both middle and right) of organoid co-cultured with fibroblasts (dispersed culture). F-actin (red), PDGFRα (white), DAPI (blue), and phosphorylated myosin light chain 2 (P-MLC2, fire LUT). Scale bar: 50 μm, scale bar in detail: 20 μm. (C) Images of organoids on day 5 of culture (with day 0 insets) with FGF2 or PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 9 / 29 Ad-GFPAd-Cre-GFPMyh9 fibroblasts pretreated withfl/flABranchedorganoidsrel.tocontrol(%)d-GFPAd-Cre050100P = 0.0078-GFPmock10 μMY2763210 μM BlebbistatinFGF2ACEfibroBranchedorganoidsrel.tocontrol(%)050100150200250FGF2YBlebBranchedorganoidsrel.tocontrol(%)050100*YBleb fibroDFP < 0.0001P = 0.0110n.s.n.s.PDGFRαF-actinDAPIminP-MLC2maxfibroblastsepitheliumepitheliumepitheliumfibroblastsepitheliumepitheliumBmax projection10 μmmax projection10 μmmax projection10 μmmax projectionAd-LifeAct (fibroblasts)DAPImax projectionmax projectionz = 0 μmz = 1 μmz = 2 μmz = 3 μmz = 4 μmz = 5 μmz = 6 μmz = 7 μmIYAPDAPI+YAP+DAPI+YAP-Ad-GFPAd-Cre-GFPMyh9 fibroblasts pretreated withfl/flmergeDAPIYAPrelative distanceHiiiiiiimergeDAPIYAPiiiiiFGF2fibro01Relative distance0.41.21.8-110FGF2YAP n/c ratioRelative distance0.61.21.8-110*fibroJG1-1Relative distanceBranchtipBranchrootYAPDAPI+YAP+DAPI+YAP-tdTomato (organoid)GFP (fibroblast)P = 9.76E-31P = 0.454mult. R =0.514mult. R = 0.043n.s.PLOS BIOLOGYFibroblast contractility is involved in epithelial branching with fibroblasts (dispersed culture) and treated with mock (DMSO) or contraction inhibitors (ROCK1/2 inhibitor Y27632 or non- muscle myosin II inhibitor blebbistatin). Scale bar: 100 μm. (D) The plots show organoid branching with contraction inhibitors as mean ± SD. Statistical analysis: Multiple t tests between each treatment and the mock-treated control; n = 4–18 (each dot represents a biologically independent experiment), N = 20 organoids per experiment. (E) Images of tdTomato organoids on day 5 of co-culture with control or Myh9 knock-out fibroblasts (dispersed culture). Scale bar: 100 μm. Videos from the 5-day experiment are presented in S5 Movie. (F) The plot shows organoid branching with control of Myh9 knock-out fibroblasts from experiment in (E) as mean ± SD. Statistical analysis: two-tailored paired t test; n = 3 (each dot represents a biologically independent experiment), N = 20 organoids per experiment. (G) Staining of YAP in an organoid co-cultured with fibroblasts (dispersed culture) or with FGF2. Scale bar: 20 μm. Full arrowheads point to cells with nuclearly localized YAP, empty arrowheads point to cells with cytoplasmic YAP. (H, I) Quantification of YAP nuclear/cytoplasmic signal ratio. The scheme explains relative distance: −1 is branch root, +1 is branch tip. Each dot represents a single cell, n = 436 cells from 19 branches of 10 organoids (fibro, H); n = 306 cells from 12 branches of 10 organoids (FGF2, I). Statistical analysis: Linear regression, mult. R indicates correlation coefficient; P is the result of ANOVA F-test. (J) Staining of organoids co-cultured with control or Myh9 knock-out fibroblasts. Scale bar: 200 μm. The data underlying the graphs shown in the figure can be found in S1 Data. FGF2, fibroblast growth factor 2; YAP, yes associated protein. https://doi.org/10.1371/journal.pbio.3002093.g004 Evidence for the role of fibroblast contractility in epithelial morphogenesis in vivo During puberty, the period of major mammary epithelial growth and branching, new primary mammary epithelial branches arise through bifurcation of TEBs. Could contractile fibroblasts play a role in this process? TEBs are large and highly proliferative stratified epithelial structures consisting of multiple (5–10) layers of luminal cells (called body cells) and an outer layer of basal cells (or cap cells). Such structures are not replicated in our co-culture model because we model only a part of the complex in vivo microenvironment—the effect of fibroblasts. In vivo the TEBs are surrounded by a complex stroma, which provides instructions for epithelial mor- phogenesis, including besides fibroblasts several more stromal cell types (adipocytes, immune cells) that secrete paracrine signals for epithelial proliferation [25–30]. While our reductionist in vitro co-culture model (consisting of fibroblasts and epithelial organoids in basal medium with no exogenous morphogenetic growth factors) was essential for untangling the importance of the contact versus paracrine signaling in fibroblast-induced branching, to model TEBs we needed to modify our co-culture model to promote epithelial proliferation. The classic mammary organoid model cultured in Matrigel with FGF2 [21] mimics epithe- lial stratification to some extent (reaching 3–4 layers of luminal cells in TEB-like ends of the branches) but does not support full myoepithelial coverage of branches [31]. We revealed that a stabilized form of FGF2 (FGF2-STAB; [32,33]) induces several TEB-like features in the orga- noids, including highly proliferative phenotype, multiple layers of luminal cells, and full myoe- pithelial cell coverage [34]. When we exposed the dispersed organoid-fibroblast co-cultures to FGF2-STAB, the organoids developed large branches with a set of features typical of TEBs in vivo, including stratified luminal cells, full myoepithelial coverage, and presence of basal-in- luminal cells (similar to cap-in-body cells in vivo [4]) (Fig 5A–5G). These results demonstrate that combination of contractile fibroblasts and strong proliferative signals can reproduce sev- eral aspects typical for TEB branching in organoid cultures (Fig 5H). Finally, we sought to determine whether the contractility-dependent mechanism of fibro- blast-induced branching could take place in vivo. We found fibroblasts expressing a contractil- ity marker alpha smooth muscle actin (αSMA) in developing mammary glands during puberty (Fig 6A). The αSMA+ fibroblasts specifically populate the stroma surrounding the TEBs (Fig 6A and 6B), the actively growing part of epithelium, which produces new branches. Importantly, the fibroblasts in co-cultures do express αSMA as well (Fig 6C). To visualize the organization of fibroblasts in the peri-TEB stroma, we performed whole organ immunostain- ing, clearing and imaging of the mammary gland. We observed that fibroblasts (stained for their cytoskeletal marker vimentin) were organized perpendicularly to the nascent bud in PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 10 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 5. Combination of fibroblasts and FGF2-STAB induces TEB-like phenotype of organoids. (A) Time-lapse snap-shots of organoids grown in basal organoid medium with no exogenous growth factors (basal M), with FGF2-STAB, co-cultured with fibroblasts or co-cultured with fibroblasts with FGF2-STAB. Scale bar: 100 μm. (B) Quantification of organoid branching. The plot shows mean ± SD. n = 2 independent biological replicates, N = 20 organoids per experiment. (C) Quantification of number of branches per branched organoid. The plot shows mean ± SD. n = 2 independent biological replicates, N = 12–19 branching organoids per experiment. (D) Examples of luminized and full branch on bright-field imaging and quantification of the branch phenotypes. n = 2 independent biological replicates, N = 12–19 branching organoids per experiment. (E) Representative confocal images of organoids PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 11 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching on day 5 of culture with FGF2-STAB or fibroblasts. Scale bar: 100 μm. (F) Quantification of maximum number of cell layers in a branch in confocal images. The plot shows mean ± SD. The dots represent averages from individual experiments. Statistical analysis: two-tailored t test; n = 3 independent biological replicates, N = 9–13 organoids per experiment. (G) Quantification of the percentage of organoids with KRT5+ cells present within the layers of KRT5-cells (basal-in-luminal, BIL cells) in confocal images. The plot shows mean ± SD. Statistical analysis: two- tailored t test; n = 3 independent biological replicates, N = 9–13 organoids per experiment. (H) A schematic representation of uncoupling fibroblast contraction and growth factor signaling in organoids. The data underlying the graphs shown in the figure can be found in S1 Data. FGF2, fibroblast growth factor 2; TEB, terminal end bud. https://doi.org/10.1371/journal.pbio.3002093.g005 bifurcating TEB (Fig 6D–i and S6 Movie) and perpendicularly to the epithelial growth direc- tion at the TEB neck (Fig 6D-ii-1 and 6E and S7 Movie), forming loops similar to those observed in in vitro co-cultures (Figs 4A and 6F). On the other hand, fibroblasts surrounding subtending duct formed a less organized, mesh-like structure (Fig 6D-ii-2 and 6E). Together, our findings suggest that contractile fibroblasts could play a role in bifurcation of TEBs during branching morphogenesis in puberty. Discussion Mechanical forces are an integral part and a driving factor of tissue morphogenesis. However, the sources of mechanical forces in different tissues are still unclear and little is understood of how force sensing is translated into cell fate during organ formation. Our work reveals the crit- ical role of fibroblast-derived mechanical forces in regulation of mammary epithelial branch- ing morphogenesis. It demonstrates that mammary fibroblasts generate mechanical forces via their actomyosin apparatus and transmit them to the epithelium, which leads to epithelial deformation and patterning of epithelial intracellular signaling, resulting in epithelial folding into branched structures. Fibroblast-generated mechanical forces as part of complex tissue mechanics The role of intraepithelial forces in morphogenetic processes involving tissue folding, such as gastrulation, tubulogenesis, or buckling has been long recognized and intensively studied [35]. Similarly, the instructive role of mechanical properties and 3D organization of the ECM in determination of cell fate and behavior during organ formation, including mammary epithelial branching morphogenesis, has been well established [10,31,36]. However, the evidence for reg- ulation of epithelial morphogenesis by mechanical stimuli from mesenchymal cells was discov- ered only recently and has been scarce, limited to the morphogenesis of feather buds in chick skin by mechanically active dermal cells [16], gut villification [17], and lung epithelial bifurca- tion and alveologenesis induced by smooth muscle cells or myofibroblasts [18–20,37]. Fibroblasts as central regulators of epithelial morphogenesis and homeostasis: Evidence for mechanically active fibroblasts in vivo Fibroblasts accompany mammary epithelial cells from early development through homeostasis to aging and disease and employ different functions to meet epithelial needs [15]. The multiple fibroblast functions are facilitated by fibroblast heterogeneity, which has only recently begun to be resolved using single-cell RNA sequencing approaches [38,39]. These studies confirmed well-established fibroblast roles in epithelial development and tissue homeostasis via produc- tion of paracrine signals and ECM, and fibroblast roles in regulation of immune landscape of the mammary gland. Though they did not detect mechanically active fibroblasts. However, these studies included only adult and aged mammary glands and omitted puberty, the stage of active mammary epithelial branching morphogenesis. Using immunostaining on mammary PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 12 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Fig 6. Contractile fibroblasts surround mammary TEBs in vivo. (A) PDGFRα and αSMA staining on pubertal mammary gland sections and detail of peri-ductal and peri-TEB fibroblasts. Scale bar: 50 μm and 10 μm in detail. (B) Quantification of αSMA+ cells out of PDGFRα+ fibroblasts. The plot shows mean ± SD, each dot represents 1 field of view, n = 7 TEBs and 8 ducts; statistical analysis: t test. (C) αSMA staining in a dispersed co-culture (day 5) and a detail of an αSMA+ fibroblast in contact with the organoid. Scale bar: 50 μm; scale bar in detail 10 μm. (D) Whole-mount PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 13 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching imaging of pubertal mammary gland stained for vimentin (VIM, a fibroblast marker). Detail (i) shows a bifurcating TEB. Detail (ii) shows an invading TEB, with close-up showing partial MIPs of upper, middle, and lower part of the TEB neck (1) and a subtending duct (2). The positions of the orthogonal YZ views are indicated with dashed yellow lines. Scale bar: 1 mm in the whole gland MIP, 50 μm in other images. (E) A schematic representation of fibroblasts surrounding TEBs in vivo and an organoid in vitro. Schemes were drawn from Figs 6D and 3B. (F) A schematic representation of our hypothesis on the role of contractile fibroblasts in TEB branching. Two mechanisms, paracrine signaling (growth factors) and mechanical cues (fibroblast contractility), which we uncoupled in vitro, work together to support mammary branching morphogenesis (TEB bifurcation) in vivo. The data underlying the graphs shown in the figure can be found in S1 Data. MIP, maximum intensity projection; TEB, terminal end bud. https://doi.org/10.1371/journal.pbio.3002093.g006 glands in puberty, we discovered that mechanically active fibroblasts (contractile fibroblasts expressing αSMA, myofibroblasts) are localized specifically around TEBs, the main structures of epithelial branching during puberty, where they organize into structures similar to the loops observed in our fibroblast-organoid co-culture models. These data suggest that fibroblast con- tractility could play a role in mammary epithelial branching in vivo. Future studies employing myofibroblast-specific mouse models are needed to determine functional requirement of fibroblast contractility for mammary epithelial branching. Importantly, the presence and func- tion of αSMA+ fibroblasts has been well documented in other developing or homeostatic organs, such as lung [40,41], intestine [42–44], or dermal sheath of the hair follicle [45,46]. While in the intestine the αSMA+ fibroblasts serve as a source of paracrine niche signals [42,44,47], the contractility of myofibroblasts is actively employed in alveolar septation [37] or relocation of the stem cell niche during hair cycle regression [45]. The mechanism of fibroblast-induced mammary morphogenesis: Connection to ECM remodeling It was previously proposed that mechanical forces generated by mesenchymal/stromal cells regulate epithelial morphogenesis indirectly via changes of ECM mechanics, including colla- gen I remodeling in embryonic gut [48] and postnatal mammary gland [5,14,15], or elastin deposition in the lung [40]. However, while not excluding contribution of such mechanism to mammary epithelial branching in vivo, our investigations in vitro in organoid-fibroblast co- cultures devoid of collagen I demonstrate that collagen I fibers are not required for induction of epithelial folding by fibroblast contractility. Mammary fibroblasts form direct, highly dynamic contacts with mammary epithelial cells and induce a mechanosensitive response in the epithelium, resulting in patterning of a key morphogenetic regulator YAP. The direct con- tact between mammary fibroblasts and epithelial cells in vivo could be enabled by immature, thin basement membrane of TEBs [49], the highly proliferative epithelial structures, which drive pubertal mammary branching morphogenesis, and active remodeling of ECM by matrix metalloproteinases produced by both epithelial cells and fibroblasts [50], which is essential for mammary branching morphogenesis [50,51]. Our work does not exclude the importance of ECM remodeling by fibroblast mechanical forces in epithelial branching. We speculate that in vivo the highly dynamic mechanically active fibroblasts could initiate formation of epithelial clefts and further reinforce them by subsequent deposition and remodeling of ECM. Recently published simulations suggest that mammary pubertal branching is highly stochastic; however, its overall shape depends on the angle of TEB bifurcation [13,52]. Thus, the local effect of fibroblast contractility could affect the random branching pattern. The intimate relationship between fibroblasts and epithelium Although our co-culture experiments demonstrate the need for direct contact between fibro- blasts and epithelium for epithelial branching, paracrine interactions between the 2 are also PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 14 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching likely to be involved. Particularly at the beginning of the co-culture, as the fibroblasts migrate towards the organoid, they may follow epithelial signals such as PDGF or FGF ligands, which have been shown to be produced by epithelial cells [53,54] and to stimulate the directional migration of fibroblasts in vitro [7,53]. The degree of fibroblast motility in vivo and their directed migration to the epithelium during mammary development remains an open ques- tion, and the use of intravital imaging may shed light on this issue. Direct molecular contact via heterotypic adhesions has been reported to promote cancer cell migration and invasion [55,56], but it has been less studied in normal epithelium. In our co-culture, we did not observe fibroblasts promoting invasive behavior in either normal orga- noids or cancer cell spheroids and we detected an ECM layer separating organoid from fibro- blasts, suggesting that a different mode of cell–cell contact takes place. Moreover, the ability of fibroblasts to promote contact-dependent branching regardless whether in contact with myoe- pithelial cells of organoids or luminal cells of cancer spheroids suggest that fibroblasts could promote organoid branching without a proper molecular connection, just by forming a supra- cellular fibroblast structure that envelops the epithelium and applies contractile forces, as it was suggested for cancer-associated fibroblasts (CAFs) interacting with intestinal tumors [57]. The mechanism of fibroblast-induced mammary morphogenesis: Requirement of paracrine signaling Importantly, our results do not rule out the importance of fibroblast-secreted factors in mam- mary epithelial morphogenesis. However, we demonstrate that paracrine signals are not suffi- cient to drive organoid branching in the 3D in vitro cultures of organoids with fibroblasts without addition of any branching-inducing growth factors and show the importance of fibro- blast-epithelium contact, so short-distance paracrine or juxtacrine signals could be important in the process. Several growth factors, including FGF2, FGF7, EGF, or TGFα can induce orga- noid branching in the absence of fibroblasts when added to the medium in nanomolar concen- trations [58], including bifurcation of the organoid branches [59]. However, the evidence for requirement of those growth factors’ expression in mammary fibroblasts for mammary epithe- lial branching in vivo is missing. The mechanism of fibroblast-induced mammary morphogenesis: Epithelial response Our work reveals that the mechanical strain imposed on mammary epithelial cells by fibro- blasts results in epithelial folding with negative curvature in the epithelial–fibroblast contact points. The part of epithelium with negative curvature, the stalk of the branch shows presence of epithelial cells with nuclear YAP and increased epithelial proliferation. In contrast, the orga- noids that branched in response to exogenous FGF2 did not show patterned cell proliferation or YAP nuclear localization, further accentuating different mechanisms underlying epithelial branching in response to growth factors and contractile fibroblasts. It remains unclear though whether fibroblasts induce YAP activation to promote epithelial proliferation at the neck and thus bud elongation, or if the patterned YAP activation in epithelial buds reflects the prolifer- ative status of cells located in different regions of the organoid. Our data suggest that epithelial proliferation in the co-cultures is mechanistically linked to the contact with the fibroblasts and/or to the mechanical stress imposed by the contractile fibroblasts in the underlying epithe- lium and in its vicinity, possibly through juxtacrine signaling or mechanochemical interplay. Budding morphogenesis of stratified epithelium, such as in the FGF2-induced mammary orga- noids, might employ self-organizing mechanisms, including preferential cell-ECM adhesion versus cell–cell adhesion as demonstrated in salivary gland organoids [60]. We propose that in PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 15 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching vivo, in the complex microenvironment of the stroma-invading, growing and bifurcating TEB, it is the combined action of contractile fibroblasts and strong proliferative signals from the stroma that governs the morphogenetic process. It was demonstrated that although the whole TEB contains proliferative cells, it is the cells localized in the neck of the TEB, which will mainly contribute to the growth of the adjacent duct, not the cells localized in the TEB tip [61]. Because contractile fibroblasts surround specifically the neck region of the TEB, we speculate that they play an essential role in this process. Importantly, the direct interactions between mammary epithelium (including both orga- noids from normal epithelium and spheroids from breast cancer cells) and fibroblasts do not lead to invasive dissemination of epithelial cells, unlike in co-cultures of squamous cell carci- noma with CAFs [55]. Interestingly, a recent study described mechanical compression of intes- tinal tumors by CAFs forming a mechanically active tumor capsule [57], providing further evidence for context-dependent employment of fibroblast-derived mechanical forces in tissue morphogenesis and tumorigenesis. In conclusion, we find that fibroblasts drive branching morphogenesis in mammary organoids by exerting mechanical forces on epithelial cells. These observations support the hypothesis that contractile fibroblasts drive pubertal mammary branching; however, future in vivo studies will be needed to formally demonstrate this. It is conceivable that such conserved mechanism could be used to regulate morphogenesis of other branched organs, providing a comprehensive understanding of overlapping but divergent employment of mechanically active fibroblasts in developmental versus tumorigenic programs. Materials and methods Animals All procedures involving animals were performed under the approval of the Ministry of Edu- cation, Youth and Sports of the Czech Republic (license # MSMT-9232/2020-2), supervised by the Expert Committee for Laboratory Animal Welfare of the Faculty of Medicine, Masaryk University, at the Laboratory Animal Breeding and Experimental Facility of the Faculty of Medicine, Masaryk University (facility license #58013/2017-MZE-17214), or under the approval of the ethics committee of the Institut Curie and the French Ministry of Research (reference #34364–202112151422480) in the Animal Facility of Institut Curie (facility license #C75–05–18). ICR mice were obtained from the Laboratory Animal Breeding and Experimen- tal Facility of the Faculty of Medicine, Masaryk University. R26-mT/mG [62] and Acta2- CreERT2 mice [63] were acquired from the Jackson Laboratories. LifeAct-GFP mice [64] were created by Wedlich-So¨ldner team, Myh9fl/fl mice [65] were kindly provided by Dr. Sara Wick- stro¨m. Transgenic animals were maintained on a C57/BL6 background. Experimental animals were obtained by breeding of the parental strains, the genotypes were determined by genotyp- ing. The mice were housed in individually ventilated or open cages, all with ambient tempera- ture of 22˚C, a 12 h:12 h light:dark cycle, and food and water ad libitum. Female 4 to 8 weeks old virgin mice were used in the experiments. Mice were euthanized by cervical dislocation and mammary gland tissues were collected immediately. Primary mammary organoid and fibroblast isolation and culture Primary mammary fibroblasts and organoids were isolated from 6 to 8 weeks old female virgin mice (ICR, unless otherwise specified) as previously described [66]. The mammary glands were chopped and partially digested in a solution of collagenase and trypsin [2 mg/ml collage- nase A, 2 mg/ml trypsin, 5 μg/ml insulin, 50 μg/ml gentamicin (all Merck), 5% fetal bovine serum (FBS; Hyclone/GE Healthcare) in DMEM/F12 (Thermo Fisher Scientific)] for 30 min PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 16 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching at 37˚C. Resulting tissue suspension was treated with DNase I (20 U/ml; Merck) and submitted to 5 rounds of differential centrifugation (450 × g for 10 s) to separate epithelial (organoid) and stromal fractions. The organoids were resuspended in basal organoid medium [1 × ITS (10 μg/ml insulin, 5.5 μg/ml transferrin, 6.7 ng/ml sodium selenite), 100 U/ml of penicillin, and 100 μg/ml of streptomycin in DMEM/F12] and kept on ice until co-culture setup. The cells from the stromal fraction were pelleted by centrifugation, suspended in fibroblast cultiva- tion medium (10% FBS, 1× ITS, 100 U/ml of penicillin, and 100 μg/ml of streptomycin in DMEM) and incubated on cell culture dishes at 37˚C, 5% CO2 for 30 min. Afterwards, the unattached (non-fibroblast) cells were washed away, the cell culture dishes were washed with PBS and fresh fibroblast medium was provided for the cells. The cells were cultured until about 80% confluence. During the first cell subculture by trypsinization, a second round of selection by differential attachment was performed, when the cells were allowed to attach only for 15 min at 37˚C and 5% CO2. The fibroblasts were expanded and used for the experiments until passage 5. To inhibit fibroblast proliferation for specific assays, the fibroblasts were treated with 10 μg/ ml mitomycin C in fibroblast medium for 3 h at 37˚C, 5% CO2. Afterwards, the fibroblasts were washed 3 times with PBS and 1 time with basal organoid medium, trypsinized and used to set up co-cultures. To prepare fibroblast-conditioned medium, the fibroblasts were seeded in cell culture dishes in fibroblast medium and the next day, the cells were washed 3 times with PBS and incubated with basal organoid medium for 24 h. Afterwards, the medium was collected from the dishes, sterile-filtered through a 0.22 μm filter, and used immediately in the experiment, or aliquoted, stored at −20˚C and used within 5 days of conditioned medium preparation. 3D culture of mammary organoids and fibroblasts 3D culture of mammary organoids and fibroblasts was performed as previously described [67]. Freshly isolated mammary organoids were embedded in Matrigel either alone (300 orga- noids in 45 μl of Matrigel per well) or with 5 × 104 mammary fibroblasts per well and plated in domes in 24-well plates. For transwell experiments, organoids were plated in domes in the transwell (8 μm pore size, Falcon-Corning), fibroblasts were plated in lower chamber. After setting the gel for 45 to 60 min at 37˚C, the cultures were overlaid with basal organoid medium (1× ITS, 100 U/ml of penicillin, and 100 μg/ml of streptomycin in DMEM/F12), not supple- mented or supplemented with growth factors [2.5 nM FGF2 (Enantis) or 2.5 nM FGF2-STAB (Enantis)] or small molecule inhibitors (S1 Table) according to the experiment. The cultures were incubated in humidified atmosphere of 5% CO2 at 37˚C on Olympus IX81 microscope equipped with Hamamatsu camera and CellR system for time-lapse imaging. The organoids/ co-cultures were photographed every 60 min for 5 days with manual refocusing every day (high-detail imaging) or photographed only once per day for 5 days (low-detail imaging). The images were exported and analyzed using Image J. Organoid branching and retraction was evaluated from videos and it was defined as formation (or loss) of a new bud/branch from the organoid. Organoids that fused with another organoid or collapsed after attachment to the bottom of the dish were excluded from the quantification. Quantification of fibroblast-orga- noid contacts was performed manually in ImageJ. Quantification of branch thickness was per- formed on images from day 5 of culture, manually in ImageJ. For fluorescent time-lapse imaging, organoids were isolated from R26-mT/mG mammary glands on day of the experiment. Fibroblasts were isolated from Acta2-CreERT2;mT/mG mice, cultured to passage 2–3 and induced in vitro by 0.5 mM 4-OH-tamoxifen (Sigma) treatment for 4 days prior to trypsinization and experimental use. Before experiment, the GFP PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 17 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching fluorescence of fibroblasts was assessed using a microscope and when it was > 95%, the cells were used for co-culture. Co-cultures were seeded on coverslip-bottom 24-well plate (IBIDI) and imaged on Cell Discoverer 7 equipped with PLAN-APOCHROMAT 20×/0.95 autocorr with 0.5× magnification lens. GFP was imaged with 470/40 nm excitation, 525/50 nm emis- sion, tdTomato was imaged with 545/25 nm excitation, 605/70 nm emission filter (all Zeiss). The samples were incubated in a humidified atmosphere of 5% CO2 at 37˚C during the imaging. 3D culture of spheroids and fibroblasts MCF7-ras cells ([68] kindly provided by Dr. Ula Polanska) were expanded in DMEM/F12 sup- plemented with 10% FBS, 100 U/ml of penicillin, and 100 μg/ml of streptomycin and incu- bated in non-adherent PolyHEMA-coated dish overnight to form spheroids. Next day, the spheroids were embedded either alone (200 spheroids in 45 μl of Matrigel per well) or with 5 × 104 mammary fibroblasts per well and plated in domes in 24-well plates. After setting the gel for 45 to 60 min at 37˚C, the cultures were overlaid with basal organoid medium, supple- mented with growth factors [2.5 nM FGF2 (Enantis) or EGF (Peprotech)] small molecule inhibitors (S1 Table) according to the experiment. The (co-)cultures were incubated in a humidified atmosphere of 5% CO2 at 37˚C on Olympus IX81 microscope equipped with Hamamatsu camera and CellR system for time-lapse imaging and photographed every 60 min for 5 days with manual refocusing every day (high-detail imaging) or photographed only once per day for 5 days (low-detail imaging). The images were exported and analyzed using Image J. Spheroid budding was evaluated from the videos and it was defined as formation of a new bud from the spheroid. Spheroids that fused with other spheroids were excluded from the quantification. Knockdown and knockout of Myh9 in mammary fibroblasts For Myh9 knockdown, the pre-designed Silencer Select siRNAs against Myh9 (IDs s70267 and s70268, Myh9si#1 and Myh9si#2, respectively) and the scrambled negative control siRNA (Silencer Select negative control or Stealth negative control siRNA), all from Thermo Fisher Scientific, were transfected into wild-type (ICR) fibroblasts with Lipofectamine 3000 Reagent (Thermo Fisher Scientific) according to manufacturer’s instructions at 20 nM siRNA. For Myh9 knockout, Myh9fl/fl fibroblasts were transduced with adenoviruses Adeno-Cre-GFP (Ad-Cre-GFP) or Adeno-GFP (Ad-GFP) from Vector Biolabs at 200 MOI for 4 h. Next day, the transfected/transduced fibroblasts were put in co-culture with organoids and submitted to bright-field or fluorescent time-lapse imaging. A part of the fibroblasts was further cultured and knockdown/knockout efficiency was determined 72 h after transfection/transduction by qPCR analysis of Myh9 mRNA levels, normalized to housekeeping genes Actb and Eef1g, and by immunostaining for MYH9. LifeAct adenoviral transduction For imaging experiments with LifeAct, fibroblasts were infected with LifeAct adenoviral parti- cles (IBIDI) according to the manufacturer’s instructions prior to co-culture set-up. Briefly, the adenovirus particles were reconstituted in fibroblast cultivation medium at concentration of 500 MOI and incubated with adherent fibroblasts at 37˚C for 4 h. After that, adenovirus- containing medium was washed out, and the cells were kept overnight in fibroblast cultivation medium. The next day, GFP fluorescence was checked under the microscope and when >50% of cells appeared green, fibroblasts were used for co-culture. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 18 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Immunofluorescence staining of 2D fibroblasts For immunofluorescent analysis, fibroblasts were cultured directly on glass coverslips, fixed with 10% neutral buffered formalin, permeabilized with 0.05% Triton X-100 in PBS and blocked with PBS with 10% FBS. Then, the coverslips were incubated with primary antibodies (S2 Table) for 2 h at RT or overnight at 4˚C. After washing, the coverslips were incubated with secondary antibodies and phalloidin AlexaFluor 488 (S2 Table) for 2 h at RT. Then, the cover- slips were washed, stained with DAPI (1 μg/ml; Merck) for 10 min and mounted in Mowiol (Merck). The cells were photographed using Axio Observer Z1 microscope with laser scanning confocal unit LSM 800 with 405, 488, 561, and 640 nm lasers, GaAsp PMT detector and objec- tive Plan-Apochromat 40×/1.20 and C-Apochromat 63× /1.20 with water immersion (all Zeiss). The brightness of each channel was linearly enhanced in Zen Blue software (Zeiss) and pictures were cropped to final size in Photo Studio 18 (Zoner). Immunofluorescence staining of 3D co-cultures For immunofluorescent analysis of 3D co-cultures, the co-cultures were fixed with 10% neutral buffered formalin, washed, and stored in PBS. Next, organoid co-cultures were stained accord- ing to the droplet-based method as described [69]. Briefly, the fixed co-cultures were placed on stereoscope (Leica FM165C) and pieces containing an organoid with approximately 100 μm of surrounding Matrigel with fibroblasts were manually cut out with 25G needles and moved on parafilm-covered cell culture dish for staining. All the staining steps were done on the parafilm in 20 μl drops, and all solutions were changed under direct visual control using the stereoscope. The co-cultures were permeabilized with 0.5% Triton X-100 in PBS, blocked with 8% FBS and 0.1% Triton X-100 in PBS (3D staining buffer, 3SB) and incubated with primary antibodies (S2 Table) in 3SB over 1 to 3 nights at 4˚C. Then, the co-cultures were washed for 3 h with 0.05% Tween-20 in PBS and incubated with secondary antibodies, phalloidin AlexaFluor 488 (S2 Table) and DAPI (1 μg/ml; Merck) in 3SB over 1 to 2 nights at 4˚C in dark. Then, the co- cultures were washed for 3 h with 0.05% Tween-20 in PBS, cleared with 60% glycerol and 2.5 M fructose solution overnight at RT in dark and mounted between slide and coverslip with dou- ble-sided tape as a spacer. The co-cultures were imaged using inverted microscope Axio Observer 7 with laser scanning confocal unit LSM 880 with 405, 488, 561, and 633 nm lasers, GaAsp PMT spectral detector and objective C-Apochromat 40×/1.20 or C-Apochromat 63×/ 1.20 with water immersion (all Zeiss). The co-cultures were photographed either as one optical slice or as 3D z-stacks of various z-step as required per experiment. The brightness of each chan- nel was linearly enhanced in Zen Blue software (Zeiss) and pictures were cropped to final size in Photo Studio 18 (Zoner). Image analysis was done manually in ImageJ. Contact of fibroblasts and organoids in confocal images was analyzed in organoid middle section by measuring the perimeter of the organoid in contact with PDGFRα signal and without it. Fibroblast loop was defined as crescent shaped tdTomato signal that wrapped at least half of organoid branch. EdU signal was quantified in 3 to 5 optical sections of organoid 20 μm apart to avoid multiple counts from the same cell. Contact between fibroblast and KRT5/KRT8 cells, number of basal-in-lumi- nal cells was counted manually in ImageJ. Number of cell layers in organoids was counted man- ually in the thickest part of a branch. LAMA5 signal along a line was measured in ImageJ. EdU incorporation assay For proliferation analysis, 5-ethynyl-20-deoxyuridine (EdU) incorporation click-it kit (Thermo Fisher Scientific) was used. EdU was administered to the organoid co-cultures 2 h prior to fixation and the EdU signal was developed according to the manufacturer’s instructions prior to immuno- fluorescence staining. The volumes were adjusted for the droplet-based staining as above. PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 19 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Immunohistochemistry-immunofluorescence (IHC-IF) Mammary glands #4 were harvested from 6 weeks old females, fixed in 10% neutral buffered formalin overnight, dehydrated in ethanol with increasing concentration and xylene and embedded in paraffin, and 5 μm sections were cut on rotational microtome (Thermo Scientific, Microm HM340E). After rehydration, sections were boiled in pH9 Tris-EDTA buffer to retrieve antigens, blocked in 10% FBS and incubated with primary and secondary antibodies, mounted (Aqua Poly/Mount, Polysciences) and imaged on laser scanning confocal microscope (LSM780/880, Zeiss). The quantification of PDGFRα and αSMA positive cells was done manu- ally in ImageJ, considering DAPI signal to distinguish single cells and continuous αSMA signal as a border of epithelium. The fields of view were scored “duct” or “TEB” based on morphology of the structures (TEBs: stratified epithelium, bulb-like shape, presence of cap-in-body cells, cuboidal cap cells; duct: one layer of luminal cells, elongated myoepithelial cells) and on the position of the structure at the distal part of the mammary epithelium (invasive front). Immunofluorescence staining of whole-mount cleared mammary gland Staining and clearing of mammary glands was done following clear, unobstructed brain imag- ing cocktails (CUBIC) protocol [70,71]. Briefly, mammary glands #3 of 4 weeks old females were harvested and fixed in 10% neutral buffered formalin overnight, washed and incubated in CUBIC reagent 1 (25% (w/w) urea, 25% (w/w) N,N,N’,N’-tetrakis(2-hydroxypropyl)ethyle- nediamine, 15% (w/w) Triton X-100 in distilled water) for 4 days shaking at RT. After washing, the glands were blocked using blocking buffer (5% FBS, 2% BSA, 1% Triton X-100, 0.02% sodium azide in PBS) overnight at RT, incubated with primary antibodies diluted in blocking buffer for 3 days at RT with rocking, washed 3 times for 2 h (0.05% Tween 20 in PBS) and incubated with secondary antibodies and DAPI (1 μg/ml) in blocking buffer. Then, the glands were transferred to CUBIC reagent 2 (50% (w/w) sucrose, 25% (w/w) urea, 10% (w/w) 2,2’,2”- nitrilotriethanol, 0.1% (w/w) Triton X-100 in distilled water) for 2 days at RT with rocking. The samples were mounted with CUBIC reagent 2 between 2 coverslips with double-sided tape as a spacer to enable imaging from both sides and they were imaged on laser scanning confocal microscope LSM780 (Zeiss). Image analysis of signal distribution The analysis of YAP nuclear to cytoplasmic ratio was done in ImageJ (NIH). Cells in optical section in the middle of an organoid branch were manually annotated and segmented for tar- get protein signal (YAP channel) and nuclei (DAPI channel) and density of pixels in YAP channel in the regions of interest (ROIs) was measured. The nuclear to cytoplasmic ratio of YAP was calculated in Excel (Microsoft). The spatial information of each ROI was manually measured on a line parallel to the branch longitudinal axis and normalized, with the value “1” set for the tip of the branch and the value “−1” set for the root of the branch. The graphs and linear regression line were created in Prism 6 (GraphPad) or Excel. Colocalization analysis of YAP and DAPI channels was done in Zen Black (Zeiss) and presented as color-coded (blue DAPI+YAP- and red DAPI+YAP+). The same cut-off for the colocalization analysis was applied for all images from the same experiment. Real-time quantitative PCR (qPCR) RNA from fibroblasts was isolated using RNeasy Mini Kit (Qiagen) according to the manufac- turer’s instruction. RNA concentration was measured using NanoDrop 2000 (Thermo Fisher Scientific). RNA was transcribed into cDNA by using Transcriptor First Strand cDNA PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 20 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching Synthesis Kit (Roche) or TaqMan Reverse Transcription kit (Life Technologies). Real-time qPCR was performed using 5 ng cDNA, 5 pmol of the forward and reverse gene-specific prim- ers each (primer sequences are shown in S3 Table) in Light Cycler SYBR Green I Master mix (Roche) on LightCycler 480 II (Roche). Relative gene expression was calculated using the ΔΔCt method and normalization to 2 housekeeping genes, β-actin (Actb) and eukaryotic elon- gation factor 1 γ (Eef1g). Transmission electron microscopy The 3D co-cultures were fixed with 3% glutaraldehyde in 100 mM sodium cacodylate buffer, pH 7.4 for 45 min, postfixed in 1% OsO4 for 50 min, and washed with cacodylate buffer. After embedding in 1% agar blocks, the samples were dehydrated in increasing ethanol series (50, 70, 96, and 100%), treated with 100% acetone, and embedded in Durcupan resin (Merck). Ultrathin sections were prepared using LKB 8802A Ultramicrotome, stained with uranyl ace- tate and Reynold’s lead citrate (Merck), and examined with FEI Morgagni 286(D) transmis- sion electron microscope. The cells in the schematics were segmented manually. Statistical analysis Sample size was not determined a priori and investigators were not blinded to experimental conditions. Statistical analysis was performed using GraphPad Prism software. Student’s t test (unpaired, two-tailed) was used for comparison of 2 groups. Bar plots were generated by Graph- Pad Prism and show mean ± standard deviation (SD). P < 0.05, P < 0.01, P < 0.001, ***P < 0.0001. The number of independent biological replicates is indicated as n. Limitations of the study The critical experiment that demonstrates the need of fibroblasts’ physical contact with the epi- thelium for epithelial branching does not allow to distinguish between direct physical contact and potential juxtacrine or very short-distance paracrine signaling between the epithelium and fibroblasts, which may contribute to epithelial morphogenesis. Supporting information S1 Fig. MCF7-ras spheroids do not respond to exogenous growth factors by branching. (A) Time-lapse snapshots of MCF7-ras spheroids cultured in basal medium with no exogenous growth factors (basal M) or with FGF2 or EGF. Scale bar: 100 μm. (B) Time-lapse snapshots of MCF7-ras spheroids co-cultured with no stromal cells (basal M) or with fibroblasts (fibro). Scale bar: 100 μm. (TIFF) S2 Fig. Fibroblast-organoid contacts precede organoid branching. (A) Time-lapse snapshots of an organoid-fibroblast co-culture. Scale bar: 100 μm. (B) Detailed snapshots of 3 examples of fibroblast-organoid contact establishment in the co-cultures shown in (A) on days 1, 2, and 3. Red arrowheads indicate fibroblasts of interest. Scale bar: 50 μm. (C) Quantification of orga- noid circularity (data from Fig 1), number of new branches and number of established fibro- blast-organoid contacts from matched experiments. The plot shows mean ± SD; n = 3 (each dot represents the average from a biologically independent experiment, N = 20 organoids per experiment). (D) Maximum intensity projection (MIP) and optical section images of a dis- persed co-culture on day 2.5, representative images of cystic and budding organoids (tdTo- mato). Fibroblasts were detected by immunostaining for PDGFRα. Scale bar: 100 μm. (E) Quantification of organoid middle section perimeter in contact with PDGFRα signal. The plot PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 21 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching shows mean ± SD. Each dot represents an average from 1 experiment. Statistical analysis: two- tailored t test; n = 3 independent biological samples, N = 15–24 organoids per sample. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S3 Fig. Quantification of fibroblast loops. (A) A representative confocal image of a dispersed co-culture on day 4. Scale bar: 20 μm, scale bar in detail: 10 μm. (B) A representative confocal image of a dispersed organoid-fibroblast co-culture on day 3. The arrowhead indicates the fibroblast loop at the branch neck. Scale bar: 50 μm. (C) Quantification of the presence of fibroblast loops around organoid branches in dispersed co-cultures. The plot shows mean ± SD. Statistical analysis: two-tailored t test; n = 3 independent biological replicates, N = 5–12 organoids per experiment; 56 branches in total. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S4 Fig. MCF7-ras spheroid budding in co-cultures requires cell contractility. (A, C) Photo- graphs of spheroids on day 4 of dispersed co-culture with fibroblasts upon treatment with no inhibitor (mock), with blebbistatin (Bleb, A) or with Y27632 (C). Top gray and red bars indi- cate proportion of branched spheroids out of all spheroids per condition. Scale bar: 100 μm. (B, D) Quantification of number of branches/buds per branched spheroid in conditions from (A). The plot shows mean ± SD, each lined dot shows mean from each experiment, each faint dot shows single spheroid measurement, n = 4 (B) or 5 (D) biologically independent experi- ments, N = 20 spheroids per experiment. Statistical analysis: two-tailored t test. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S5 Fig. Contractility inhibitors do not impede fibroblast motility. (A) Representative end- point images of organoids in dispersed co-cultures with contractility inhibitors. Scale bar: 100 μm. (B) Detailed time-lapse snapshots of fibroblast-organoid contact establishment in co- cultures with or without the inhibitors. Scale bar: 50 μm. White arrowhead indicates the fibro- blast of interest. (C, D) Quantification of fibroblast-organoid contacts established in co-cul- tures with inhibitors (Y = 10 μM Y27632, C; Bleb = 10 μM Blebbistatin, D) within the first 2 days. The plots show mean ± SD. Statistical analysis: two-tailored t test; n = 3 independent bio- logical replicates, N = 10 organoids per experiment. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S6 Fig. Fibroblast contractility is necessary for branch maintenance. (A) Experimental scheme (top) and time-lapse snapshots of dispersed co-cultures treated with contractility inhibitors on day 3 of culture. Scale bar: 100 μm. White arrowheads indicate organoid branches. (B–D) Quantification of organoids with retracted branches (B), number of formed branches per branched organoids (C) and number of retracted branches per organoid (D). The plots show mean ± SD. Statistical analysis: two-tailored t test; n = 4 independent biological replicates, N = 20 organoids per experiment. The data underlying the graphs shown in the fig- ure can be found in S1 Data. (TIFF) S7 Fig. Mammary fibroblasts express MYH9 and MYH10. (A) Real-time qPCR analysis of non-muscle myosin II heavy chain genes Myh9, Myh10, and Myh14 in mammary fibroblasts (fib) and epithelium (organoids, org). Plots show mean ± SD. Statistical analysis: two-tailored t test; n = 3 independent biological samples. (B) Representative images of MYH9 and MYH10 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 22 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching immunostaining in mammary fibroblasts in the first passage. Scale bar: 50 μm. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S8 Fig. Knockdown of Myh9 in mammary fibroblasts abrogates fibroblast-induced branch- ing of mammary organoids. (A, B) Representative images (day 5 of culture) (A) and quantifi- cation (B) of organoid branching in dispersed co-cultures with wild-type fibroblasts pre- treated with nonsense (siNC) or Myh9 targeting (siMyh9) siRNA. Plot indicates mean ± SD. Statistical analysis: two-tailored paired t test; n = 6 independent Myh9 knockdown experi- ments; N = 20 organoids per each treatment of each independent experiment. Videos from the 5-day experiment are presented in S4 Movie. (C–E) Quantification of MYH9 down-regulation in Myh9 KO fibroblasts by immunofluorescence. The plot (C) shows mean ± SD, n = 2 inde- pendent experiments. Representative images (D) show MYH9 (cyan) and F-actin (phalloidin, magenta) staining in cultured primary mammary fibroblasts from Myh9fl/fl mice, treated with adeno-GFP (Ad-GFP) or adeno-Cre-GFP (Ad-Cre-GFP) vector, including details (E) of cyto- skeleton organization. Scale bars: 1 mm (D), 20 μm (E, first and third row), and 5 μm (E, sec- ond and fourth row). The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S9 Fig. Myh9 knock-out does not impede fibroblast motility. (A) Detailed time-lapse snap- shots of fibroblast-organoid contact establishment in dispersed co-cultures with control or Myh9-KO fibroblasts and tdTomato+ organoids. Scale bar: 50 μm. (B) Quantification of fibro- blast-organoid contacts established in the first 3 days of co-culture, comparing GFP+ and GFP- fibroblasts (GFP is a marker of adenoviral transduction). The plot shows mean ± SD. Statistical analysis: two-tailored t test; n = 3 independent biological replicates, N = 20 orga- noids per experiment. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S10 Fig. Proliferation in co-culture system. (A) Representative images of organoids on day 4 of culture in basal medium (basal M), in dispersed co-culture with fibroblasts or with FGF2, EdU administered 2 h pre-fix, EPCAM (red), DAPI (blue), EdU (cyan), fibroblasts were iso- lated from R26-mT/mG mice (tdTomato, white). Scale bar: 100 μm. (B) Optical section of a branch from (A) (top), a scheme of branch regions (bottom). (C, D) Quantification of percent- age of EdU+ cells from epithelial cells in different branch regions in fibroblast-organoid dis- persed co-culture (C) and in FGF2-treated organoid (D). The box and whiskers plot shows minimum, median, and maximum values, and second and third quartiles of data distribution. n = 3 independent experiments, N = 6 organoids, 2,202 analyzed cells in (C); N = 11 organoids, 3,104 analyzed cells in (D). Statistical analysis: Multiple t tests. (E) A scheme of the prolifera- tion-inhibition experiment. (F) Co-cultures at day 5 (dispersed culture), fibroblasts pretreated with +/- mitomycin C (MMC), co-cultures treated with +/- aphidicolin (Aph). Scale bar: 100 μm. (G) Quantification of the percentage of branched organoids from experiment in (F). The plot shows mean ± SD, each dot represents a biologically independent experiment, n = 2, N = 51–77 organoids per sample, statistical analysis: t test. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S11 Fig. Spheroid proliferation is necessary for its branching in co-culture with fibroblasts. (A) Representative images of MCF7-ras spheroids in dispersed co-culture with fibroblasts on day 4 with spheroids formed from mock- or mitomycin C-treated MCF7-ras cells. The insets PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 23 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching (top red bars) show proportion of branched spheroids out of all spheroids per condition. Scale bar: 100 μm. (B) The plot shows number of spheroid branches/buds formed, with mean ± SD. Each lined dot represents mean of each experiment, each faint dot represents 1 spheroid, n = 4 independent experiments (coded by dot colors), N = 15–20 spheroids per experiment. Statisti- cal analysis: two-tailored t test. The data underlying the graphs shown in the figure can be found in S1 Data. (TIFF) S1 Movie. Mammary epithelial branching morphogenesis upon FGF2 treatment or fibro- blast co-culture. The video is composed of time-lapse videos capturing 5 days of epithelial morphogenesis in 3D organoid culture with no growth factor in the basal organoid medium (left), with FGF2 in the basal organoid medium (middle), or in fibroblast-organoid co-culture without addition of any growth factors to the basal organoid medium. Snapshots from the vid- eos are depicted in Fig 1A. (AVI) S2 Movie. Fibroblasts dynamically interact with the epithelium. Time-lapse video (bright- field and fluorescence imaging) shows 4 days of epithelial morphogenesis in fibroblast (cyan)- organoid (red) co-culture (day 0–4). Scale bar: 100 μm. Snapshots from the movie are depicted in Fig 3A. (AVI) S3 Movie. Fibroblasts form close contacts with epithelium in the organoid branching points. 3D structure of organoid-fibroblast interaction. Single images are shown in Fig 3D. Luminal cells (KRT8), red; basal cells (KRT5), blue; all cells (F-actin), white. (AVI) S4 Movie. Myh9 knock-down in fibroblasts decreases their morphogenetic potential. Time- lapse videos show 5 days of epithelial morphogenesis in co-culture with either control (left) or Myh9 knocked-down fibroblasts (siRNA-mediated knockdown; siMyh9; right). Time is in hours. Snapshots from the video are depicted in Fig 4. (AVI) S5 Movie. Myh9 knock-out in fibroblast decreases their morphogenetic potential. Time- lapse video captures 4 days of epithelial morphogenesis in fibroblast (cyan)-organoid(red) co- culture with either control (Ad-GFP; left) or Myh9 knocked-out fibroblasts (adeno-Cre-medi- ated knock-out; Ad-Cre-GFP; right). Snapshots from the video are depicted in Fig 4. Scale bar: 100 μm. (AVI) S6 Movie. Fibroblasts organization around bifurcating TEB. Z-stack scroll-through of mammary gland whole-organ imaging, showing a bifurcating TEB. DAPI in blue, vimentin in white, tdTomato in red. MIP and appropriate scale bar are depicted in Fig 6. (AVI) S7 Movie. Fibroblasts organization around invading TEB. Z-stack scroll-through of mam- mary gland whole-organ imaging, showing an invading TEB. MIP and appropriate scale bar are depicted in Fig 6. (AVI) S1 Data. Excel spreadsheet with individual numerical data underlying plots and statistical analyses. The data are organized into separate sheets corresponding to the following figure panels: 1B, 1C, 1D, 1E, 2B, 2D, 3C, 3G, 4D, 4F, 4H, 4I, 5B, 5C, 5F, 5G, 6B, S2C, S2E, S3B, S4B, PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 24 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching S4D, S5C, S5D, S6B, S6C, S6D, S7A, S8B, S8C, S9B, S10C, S10D, S10G, and S11B. (XLSX) S1 Table. The list of pharmacological and viral compounds. (DOCX) S2 Table. The list of detection agents used in this study. (DOCX) S3 Table. The list of primers used for qPCR in this study. (DOCX) Acknowledgments We are grateful to Danijela Matic Vignjevic for critical review of the manuscript, to Denisa Belisova for mouse husbandry, and to Maria Luisa Martin Faraldo for the LAMA5 antibody. We are thankful to Enantis for providing FGF2 and FGF2-STAB. We acknowledge the core facility CELLIM of CEITEC, supported by the Czech-BioImaging large RI project (LM2023050 funded by MEYS CR), for their support with obtaining scientific data presented in this paper. We gratefully acknowledge the Cell and Tissue Imaging Platform (PICT-IBiSA) at Institut Curie, member of the French National Research Infrastructure France-BioImaging (ANR-10-INBS-04). Author Contributions Conceptualization: Jakub Sumbal, Zuzana Sumbalova Koledova. Funding acquisition: Jakub Sumbal, Silvia Fre, Zuzana Sumbalova Koledova. Investigation: Jakub Sumbal, Zuzana Sumbalova Koledova. Methodology: Jakub Sumbal, Zuzana Sumbalova Koledova. Project administration: Zuzana Sumbalova Koledova. Resources: Silvia Fre, Zuzana Sumbalova Koledova. Supervision: Silvia Fre, Zuzana Sumbalova Koledova. Validation: Jakub Sumbal. Writing – original draft: Jakub Sumbal, Zuzana Sumbalova Koledova. Writing – review & editing: Jakub Sumbal, Silvia Fre, Zuzana Sumbalova Koledova. References 1. Affolter M, Zeller R, Caussinus E. Tissue remodelling through branching morphogenesis. Nat Rev Mol Cell Biol. 2009; 10:831–842. https://doi.org/10.1038/nrm2797 PMID: 19888266 2. Goodwin K, Nelson CM. Branching morphogenesis. Development. 2020; 147:dev184499. https://doi. org/10.1242/dev.184499 PMID: 32444428 3. Wang S, Sekiguchi R, Daley WP, Yamada KM. Patterned cell and matrix dynamics in branching mor- phogenesis. J Cell Biol. 2017; 216:559–570. https://doi.org/10.1083/jcb.201610048 PMID: 28174204 4. Paine IS, Lewis MT. The Terminal End Bud: the Little Engine that Could. J Mammary Gland Biol Neo- plasia. 2017; 22:93–108. https://doi.org/10.1007/s10911-017-9372-0 PMID: 28168376 5. Koledova Z, Zhang X, Streuli C, Clarke RB, Klein OD, Werb Z, et al. SPRY1 regulates mammary epithe- lial morphogenesis by modulating EGFR-dependent stromal paracrine signaling and ECM remodeling. Proc Natl Acad Sci U S A. 2016; 113:E5731–5740. https://doi.org/10.1073/pnas.1611532113 PMID: 27621461 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 25 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching 6. Kouros-Mehr H, Werb Z. Candidate regulators of mammary branching morphogenesis identified by genome-wide transcript analysis. Dev Dyn. 2006; 235:3404–3412. https://doi.org/10.1002/dvdy.20978 PMID: 17039550 7. Sumbal J, Koledova Z. FGF signaling in mammary gland fibroblasts regulates multiple fibroblast func- tions and mammary epithelial morphogenesis. Development. 2019; 146. https://doi.org/10.1242/dev. 185306 PMID: 31699800 8. Wiseman BS, Werb Z. Stromal effects on mammary gland development and breast cancer. Science. 2002; 296:1046–1049. https://doi.org/10.1126/science.1067431 PMID: 12004111 9. Zhao C, Cai S, Shin K, Lim A, Kalisky T, Lu W-J, et al. Stromal Gli2 activity coordinates a niche signaling program for mammary epithelial stem cells. Science. 2017; 356. https://doi.org/10.1126/science. aal3485 PMID: 28280246 10. Brownfield DG, Venugopalan G, Lo A, Mori H, Tanner K, Fletcher DA, et al. Patterned collagen fibers orient branching mammary epithelium through distinct signaling modules. Curr Biol. 2013; 23:703–709. https://doi.org/10.1016/j.cub.2013.03.032 PMID: 23562267 11. Hammer AM, Sizemore GM, Shukla VC, Avendano A, Sizemore ST, Chang JJ, et al. Stromal PDGFR- α Activation Enhances Matrix Stiffness, Impedes Mammary Ductal Development, and Accelerates Tumor Growth. Neoplasia. 2017; 19:496–508. https://doi.org/10.1016/j.neo.2017.04.004 PMID: 28501760 12. Jones CE, Hammer AM, Cho Y, Sizemore GM, Cukierman E, Yee LD, et al. Stromal PTEN Regulates Extracellular Matrix Organization in the Mammary Gland. Neoplasia. 2019; 21:132–145. https://doi.org/ 10.1016/j.neo.2018.10.010 PMID: 30550871 13. Nerger BA, Jaslove JM, Elashal HE, Mao S, Kosˇmrlj A, Link AJ, et al. Local accumulation of extracellu- lar matrix regulates global morphogenetic patterning in the developing mammary gland. Curr Biol. 2021 [cited 2021 Mar 19]. https://doi.org/10.1016/j.cub.2021.02.015 PMID: 33705716 14. Peuhu E, Kaukonen R, Lerche M, Saari M, Guzma´ n C, Rantakari P, et al. SHARPIN regulates collagen architecture and ductal outgrowth in the developing mouse mammary gland. EMBO J. 2017; 36:165– 182. https://doi.org/10.15252/embj.201694387 PMID: 27974362 15. Sumbal J, Belisova D, Koledova Z. Fibroblasts: The grey eminence of mammary gland development. Semin Cell Dev Biol. 2020. https://doi.org/10.1016/j.semcdb.2020.10.012 PMID: 33158729 16. Shyer AE, Rodrigues AR, Schroeder GG, Kassianidou E, Kumar S, Harland RM. Emergent cellular self- organization and mechanosensation initiate follicle pattern in the avian skin. Science. 2017; 357:811– 815. https://doi.org/10.1126/science.aai7868 PMID: 28705989 17. Shyer AE, Tallinen T, Nerurkar NL, Wei Z, Gil ES, Kaplan DL, et al. Villification: how the gut gets its villi. Science. 2013; 342:212–218. https://doi.org/10.1126/science.1238842 PMID: 23989955 18. Goodwin K, Mao S, Guyomar T, Miller E, Radisky DC, Kosˇmrlj A, et al. Smooth muscle differentiation shapes domain branches during mouse lung development. Development. 2019; 146. https://doi.org/10. 1242/dev.181172 PMID: 31645357 19. Kim HY, Pang M-F, Varner VD, Kojima L, Miller E, Radisky DC, et al. Localized Smooth Muscle Differ- entiation Is Essential for Epithelial Bifurcation during Branching Morphogenesis of the Mammalian Lung. Dev Cell. 2015; 34:719–726. https://doi.org/10.1016/j.devcel.2015.08.012 PMID: 26387457 20. Palmer MA, Nerger BA, Goodwin K, Sudhakar A, Lemke SB, Ravindran PT, et al. Stress ball morpho- genesis: How the lizard builds its lung. Sci Adv. 2021 [cited 2022 Jan 15]. https://doi.org/10.1126/ sciadv.abk0161 PMID: 34936466 21. Ewald AJ, Brenot A, Duong M, Chan BS, Werb Z. Collective epithelial migration and cell rearrange- ments drive mammary branching morphogenesis. Dev Cell. 2008; 14:570–581. https://doi.org/10.1016/ j.devcel.2008.03.003 PMID: 18410732 22. Huebner RJ, Lechler T, Ewald AJ. Developmental stratification of the mammary epithelium occurs through symmetry-breaking vertical divisions of apically positioned luminal cells. Development. 2014; 141:1085–1094. https://doi.org/10.1242/dev.103333 PMID: 24550116 23. Huebner RJ, Neumann NM, Ewald AJ. Mammary epithelial tubes elongate through MAPK-dependent coordination of cell migration. Development. 2016; 143:983–993. https://doi.org/10.1242/dev.127944 PMID: 26839364 24. Panciera T, Azzolin L, Cordenonsi M, Piccolo S. Mechanobiology of YAP and TAZ in physiology and disease. Nat Rev Mol Cell Biol. 2017; 18:758–770. https://doi.org/10.1038/nrm.2017.87 PMID: 28951564 25. Gouon-Evans V, Rothenberg ME, Pollard JW. Postnatal mammary gland development requires macro- phages and eosinophils. Development. 2000; 127:2269–2282. https://doi.org/10.1242/dev.127.11. 2269 PMID: 10804170 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 26 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching 26. Gouon-Evans V, Lin EY, Pollard JW. Requirement of macrophages and eosinophils and their cyto- kines/chemokines for mammary gland development. Breast Cancer Res. 2002; 4:155–164. https://doi. org/10.1186/bcr441 PMID: 12100741 27. Gyorki DE, Asselin-Labat M-L, van Rooijen N, Lindeman GJ, Visvader JE. Resident macrophages influ- ence stem cell activity in the mammary gland. Breast Cancer Res. 2009; 11:R62. https://doi.org/10. 1186/bcr2353 PMID: 19706193 28. Lilla JN, Werb Z. Mast cells contribute to the stromal microenvironment in mammary gland branching morphogenesis. Dev Biol. 2010; 337:124–133. https://doi.org/10.1016/j.ydbio.2009.10.021 PMID: 19850030 29. Parsa S, Ramasamy SK, De Langhe S, Gupte VV, Haigh JJ, Medina D, et al. Terminal end bud mainte- nance in mammary gland is dependent upon FGFR2b signaling. Dev Biol. 2008; 317:121–131. https:// doi.org/10.1016/j.ydbio.2008.02.014 PMID: 18381212 30. Sferruzzi-Perri AN, Robertson SA, Dent LA. Interleukin-5 transgene expression and eosinophilia are associated with retarded mammary gland development in mice. Biol Reprod. 2003; 69:224–233. https:// doi.org/10.1095/biolreprod.102.010611 PMID: 12620930 31. Nguyen-Ngoc K-V, Ewald AJ. Mammary ductal elongation and myoepithelial migration are regulated by the composition of the extracellular matrix. J Microsc. 2013; 251:212–223. https://doi.org/10.1111/jmi. 12017 PMID: 23432616 32. Dvorak P, Bednar D, Vanacek P, Balek L, Eiselleova L, Stepankova V, et al. Computer-assisted engi- neering of hyperstable fibroblast growth factor 2. Biotechnol Bioeng. 2018; 115:850–862. https://doi. org/10.1002/bit.26531 PMID: 29278409 33. Koledova Z, Sumbal J, Rabata A, de La Bourdonnaye G, Chaloupkova R, Hrdlickova B, et al. Fibroblast Growth Factor 2 Protein Stability Provides Decreased Dependence on Heparin for Induction of FGFR Signaling and Alters ERK Signaling Dynamics. Front Cell Dev Biol. 2019; 7:331. https://doi.org/10. 3389/fcell.2019.00331 PMID: 31921844 34. Sumbal J, Vranova T, Koledova Z. FGF signaling dynamics regulates epithelial patterning and morpho- genesis. bioRxiv; 2020. p. 2020.11.17.386607. https://doi.org/10.1101/2020.11.17.386607 35. Heisenberg C-P, Bellaïche Y. Forces in tissue morphogenesis and patterning. Cell. 2013; 153:948– 962. https://doi.org/10.1016/j.cell.2013.05.008 PMID: 23706734 36. Bonnans C, Chou J, Werb Z. Remodelling the extracellular matrix in development and disease. Nat Rev Mol Cell Biol. 2014; 15:786. https://doi.org/10.1038/nrm3904 PMID: 25415508 37. 38. Li R, Li X, Hagood J, Zhu M-S, Sun X. Myofibroblast contraction is essential for generating and regener- ating the gas-exchange surface. J Clin Invest. 2020; 130:2859–2871. https://doi.org/10.1172/ JCI132189 PMID: 32338642 Li CM-C, Shapiro H, Tsiobikas C, Selfors LM, Chen H, Rosenbluth J, et al. Aging-Associated Alterations in Mammary Epithelia and Stroma Revealed by Single-Cell RNA Sequencing. Cell Rep. 2020; 33:108566. https://doi.org/10.1016/j.celrep.2020.108566 PMID: 33378681 39. Yoshitake R, Chang G, Saeki K, Ha D, Wu X, Wang J, et al. Single-Cell Transcriptomics Identifies Het- erogeneity of Mouse Mammary Gland Fibroblasts With Distinct Functions, Estrogen Responses, Differ- entiation Processes, and Crosstalks With Epithelium. Front Cell Dev Biol. 2022; 10:850568. https://doi. org/10.3389/fcell.2022.850568 PMID: 35300413 40. Li R, Bernau K, Sandbo N, Gu J, Preissl S, Sun X. Pdgfra marks a cellular lineage with distinct contribu- tions to myofibroblasts in lung maturation and injury response. Morrisey E, Dietz HC, editors. Elife. 2018; 7:e36865. https://doi.org/10.7554/eLife.36865 PMID: 30178747 41. Branchfield K, Li R, Lungova V, Verheyden JM, McCulley D, Sun X. A three-dimensional study of alveo- logenesis in mouse lung. Dev Biol. 2016; 409:429–441. https://doi.org/10.1016/j.ydbio.2015.11.017 PMID: 26632490 42. McCarthy N, Manieri E, Storm EE, Saadatpour A, Luoma AM, Kapoor VN, et al. Distinct Mesenchymal Cell Populations Generate the Essential Intestinal BMP Signaling Gradient. Cell Stem Cell. 2020; 26:391–402.e5. https://doi.org/10.1016/j.stem.2020.01.008 PMID: 32084389 43. Powell DW, Pinchuk IV, Saada JI, Chen X, Mifflin RC. Mesenchymal cells of the intestinal lamina pro- pria. Annu Rev Physiol. 2011; 73:213–237. https://doi.org/10.1146/annurev.physiol.70.113006.100646 PMID: 21054163 44. Xiang J, Guo J, Zhang S, Wu H, Chen Y-G, Wang J, et al. A stromal lineage maintains crypt structure and villus homeostasis in the intestinal stem cell niche. BMC Biol. 2023; 21:169. https://doi.org/10.1186/ s12915-023-01667-2 PMID: 37553612 45. Heitman N, Sennett R, Mok K-W, Saxena N, Srivastava D, Martino P, et al. Dermal sheath contraction powers stem cell niche relocation during hair cycle regression. Science. 2020; 367:161–166. https://doi. org/10.1126/science.aax9131 PMID: 31857493 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 27 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching 46. Ahlers JMD, Falckenhayn C, Holzscheck N, Sole´ -Boldo L, Schu¨ tz S, Wenck H, et al. Single-Cell RNA Profiling of Human Skin Reveals Age-Related Loss of Dermal Sheath Cells and Their Contribution to a Juvenile Phenotype. Front Genet. 2022; 12:797747. https://doi.org/10.3389/fgene.2021.797747 PMID: 35069694 47. Shoshkes-Carmel M, Wang YJ, Wangensteen KJ, To´ th B, Kondo A, Massasa EE, et al. Subepithelial telocytes are an important source of Wnts that supports intestinal crypts. Nature. 2018; 557:242–246. https://doi.org/10.1038/s41586-018-0084-4 PMID: 29720649 48. Hughes AJ, Miyazaki H, Coyle MC, Zhang J, Laurie MT, Chu D, et al. Engineered Tissue Folding by Mechanical Compaction of the Mesenchyme. Dev Cell. 2018; 44:165–178.e6. https://doi.org/10.1016/j. devcel.2017.12.004 PMID: 29290586 49. Silberstein GB, Daniel CW. Glycosaminoglycans in the basal lamina and extracellular matrix of the developing mouse mammary duct. Dev Biol. 1982; 90:215–222. https://doi.org/10.1016/0012-1606(82) 90228-7 PMID: 6800862 50. 51. Feinberg TY, Zheng H, Liu R, Wicha MS, Yu SM, Weiss SJ. Divergent Matrix-Remodeling Strategies Distinguish Developmental from Neoplastic Mammary Epithelial Cell Invasion Programs. Dev Cell. 2018; 47:145–160.e6. https://doi.org/10.1016/j.devcel.2018.08.025 PMID: 30269950 Fata JE, Werb Z, Bissell MJ. Regulation of mammary gland branching morphogenesis by the extracellu- lar matrix and its remodeling enzymes. Breast Cancer Res. 2004; 6:1–11. https://doi.org/10.1186/ bcr634 PMID: 14680479 52. Hannezo E, Clgj S, Moad M, Drogo N, Heer R, Sampogna RV, et al. A Unifying Theory of Branching Morphogenesis. Cell. 2017; 171. https://doi.org/10.1016/j.cell.2017.08.026 PMID: 28938116 53. Joshi PA, Waterhouse PD, Kasaian K, Fang H, Gulyaeva O, Sul HS, et al. PDGFRα+ stromal adipocyte progenitors transition into epithelial cells during lobulo-alveologenesis in the murine mammary gland. Nat Commun. 2019; 10:1760. https://doi.org/10.1038/s41467-019-09748-z PMID: 30988300 54. Macias H, Moran A, Samara Y, Moreno M, Compton JE, Harburg G, et al. SLIT/ROBO1 signaling sup- presses mammary branching morphogenesis by limiting basal cell number. Dev Cell. 2011; 20:827– 840. https://doi.org/10.1016/j.devcel.2011.05.012 PMID: 21664580 55. Labernadie A, Kato T, Brugue´s A, Serra-Picamal X, Derzsi S, Arwert E, et al. A mechanically active het- erotypic E-cadherin/N-cadherin adhesion enables fibroblasts to drive cancer cell invasion. Nat Cell Biol. 2017; 19:224–237. https://doi.org/10.1038/ncb3478 PMID: 28218910 56. Omelchenko T, Fetisova E, Ivanova O, Bonder EM, Feder H, Vasiliev JM, et al. Contact interactions between epitheliocytes and fibroblasts: formation of heterotypic cadherin-containing adhesion sites is accompanied by local cytoskeletal reorganization. Proc Natl Acad Sci U S A. 2001; 98:8632–8637. https://doi.org/10.1073/pnas.151247698 PMID: 11447275 57. Barbazan J, Pe´rez-Gonza´ lez C, Go´ mez-Gonza´ lez M, Dedenon M, Richon S, Latorre E, et al. Cancer- associated fibroblasts actively compress cancer cells and modulate mechanotransduction. Nat Com- mun. 2023; 14:6966. https://doi.org/10.1038/s41467-023-42382-4 PMID: 37907483 58. Sternlicht MD, Sunnarborg SW, Kouros-Mehr H, Yu Y, Lee DC, Werb Z. Mammary ductal morphogene- sis requires paracrine activation of stromal EGFR via ADAM17-dependent shedding of epithelial amphiregulin. Development. 2005; 132:3923–3933. https://doi.org/10.1242/dev.01966 PMID: 16079154 59. Neumann NM, Kim DM, Huebner RJ, Ewald AJ. Collective cell migration is spatiotemporally regulated during mammary epithelial bifurcation. J Cell Sci. 2023; 136:jcs259275. https://doi.org/10.1242/jcs. 259275 PMID: 36602106 60. Wang S, Matsumoto K, Lish SR, Cartagena-Rivera AX, Yamada KM. Budding epithelial morphogenesis driven by cell-matrix versus cell-cell adhesion. Cell. 2021; 184:3702–3716.e30. https://doi.org/10.1016/ j.cell.2021.05.015 PMID: 34133940 61. Scheele CLGJ, Hannezo E, Muraro MJ, Zomer A, Langedijk NSM, van Oudenaarden A, et al. Identity and dynamics of mammary stem cells during branching morphogenesis. Nature. 2017; 542:313–317. https://doi.org/10.1038/nature21046 PMID: 28135720 62. Muzumdar MD, Tasic B, Miyamichi K, Li L, Luo L. A global double-fluorescent Cre reporter mouse. Gen- esis. 2007; 45:593–605. https://doi.org/10.1002/dvg.20335 PMID: 17868096 63. Wendling O, Bornert J-M, Chambon P, Metzger D. Efficient temporally-controlled targeted mutagenesis in smooth muscle cells of the adult mouse. Genesis. 2009; 47:14–18. https://doi.org/10.1002/dvg. 20448 PMID: 18942088 64. Riedl J, Flynn KC, Raducanu A, Ga¨ rtner F, Beck G, Bo¨ sl M, et al. Lifeact mice for studying F-actin dynamics. Nat Methods. 2010; 7:168–169. https://doi.org/10.1038/nmeth0310-168 PMID: 20195247 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 28 / 29 PLOS BIOLOGYFibroblast contractility is involved in epithelial branching 65. Conti MA, Even-Ram S, Liu C, Yamada KM, Adelstein RS. Defects in cell adhesion and the visceral endoderm following ablation of nonmuscle myosin heavy chain II-A in mice. J Biol Chem. 2004; 279:41263–41266. https://doi.org/10.1074/jbc.C400352200 PMID: 15292239 66. Koledova Z. 3D Coculture of Mammary Organoids with Fibrospheres: A Model for Studying Epithelial- Stromal Interactions During Mammary Branching Morphogenesis. Methods Mol Biol. 2017; 1612:107– 124. https://doi.org/10.1007/978-1-4939-7021-6_8 PMID: 28634938 67. Koledova Z, Lu P. A 3D Fibroblast-Epithelium Co-culture Model for Understanding Microenvironmental Role in Branching Morphogenesis of the Mammary Gland. Methods Mol Biol. 2017; 1501:217–231. https://doi.org/10.1007/978-1-4939-6475-8_10 PMID: 27796955 68. Kasid A, Lippman ME, Papageorge AG, Lowy DR, Gelmann EP. Transfection of v-rasH DNA into MCF- 7 human breast cancer cells bypasses dependence on estrogen for tumorigenicity. Science. 1985; 228:725–728. https://doi.org/10.1126/science.4039465 PMID: 4039465 69. Sumbal J, Koledova Z. Single Organoids Droplet-Based Staining Method for High-End 3D Imaging of Mammary Organoids. Methods Mol Biol. 2022; 2471:259–269. https://doi.org/10.1007/978-1-0716- 2193-6_14 PMID: 35175602 70. Susaki EA, Tainaka K, Perrin D, Kishino F, Tawara T, Watanabe TM, et al. Whole-brain imaging with single-cell resolution using chemical cocktails and computational analysis. Cell. 2014; 157:726–739. https://doi.org/10.1016/j.cell.2014.03.042 PMID: 24746791 71. Lloyd-Lewis B, Davis FM, Harris OB, Hitchcock JR, Lourenco FC, Pasche M, et al. Imaging the mam- mary gland and mammary tumours in 3D: optical tissue clearing and immunofluorescence methods. Breast Cancer Res. 2016; 18:127. https://doi.org/10.1186/s13058-016-0754-9 PMID: 27964754 PLOS Biology \| https://doi.org/10.1371/journal.pbio.3002093 January 10, 2024 29 / 29 PLOS BIOLOGY
10.1099_mgen.0.001019	SHORT COMMUNICATION Narunsky et al., Microbial Genomics 2023;9:001019 DOI 10.1099/mgen.0.001019 A conserved uORF in the ilvBNC mRNA of Corynebacterium species regulates ilv operon expression Aya Narunsky1†, Kumari Kavita1†, Shanker S. S. Panchapakesan1†, Megan E. Fris2‡ and Ronald R. Breaker1,2,3,* Abstract Computational methods can be used to identify putative structured noncoding RNAs (ncRNAs) in bacteria, which can then be validated using various biochemical and genetic approaches. In a search for ncRNAs in Corynebacterium pseudotuberculosis, we observed a conserved region called the ilvB- II motif located upstream of the ilvB gene that is also present in other members of this genus. This gene codes for an enzyme involved in the production of branched- chain amino acids (BCAAs). The ilvB gene in some bacteria is regulated by members of a ppGpp- sensing riboswitch class, but previous and current data suggest that the ilvB- II motif regulates expression by a transcription attenuation mechanism involving protein translation from an upstream open reading frame (uORF or leader peptide). All representatives of this RNA motif carry a start codon positioned in- frame with a nearby stop codon, and the peptides resulting from translation of this uORF are enriched for BCAAs, suggesting that expres- sion of the ilvB gene in the host cells is controlled by attenuation. Furthermore, recently discovered RNA motifs also associated with ilvB genes in other bacterial species appear to carry distinct uORFs, suggesting that transcription attenuation by uORF translation is a common mechanism for regulating ilvB genes. INTRODUCTION Acetohydroxyacid synthase I (AHAS) and isomeroreductase (IR) are two key enzymes for the biosynthesis of the three branched- chain amino acids (BCAAs). These enzymes are encoded by the ilvBNC operon (wherein ‘ilv’ represents isoleucine, leucine and valine). Specifically, AHAS is encoded by ilvBN and IR is encoded by ilvC [1, 2]. Although BCAAs are not synthesized by mammals, genes involved in their synthesis are vital for cell growth in many bacteria [3]. The phylogenetic distribution of these genes makes them potential targets for novel antibiotics, and thus the regulation of their expression has been studied extensively. Computational methods offer powerful approaches for finding new structured RNA motifs, including those involved in gene regulation [4, 5]. For example, these methods were proven useful for identifying riboswitch candidates in bacteria [6–9]. Riboswitches are structured noncoding RNA (ncRNA) domains that regulate the expression of adjacent genes by forming a selective ligand- binding pocket or ‘aptamer’ [10–12]. As part of a search for structured ncRNAs in bacterial genomes, we observed an aptamer candidate upstream of the ilvB gene in Corynebacterium pseudotuberculosis. Previously, representatives of a ppGpp riboswitch class were reported to control this gene in Firmicutes [13]. Therefore, we hypothesized that the newly recognized RNA structure called the ilvB- II motif might represent a different riboswitch class selective for ppGpp. Genetic reporter assays conducted in a surrogate organism with a representative RNA derived from Corynebacterium ulcerans, an emerging human pathogen that causes respiratory diphtheria [14], indicate that the motif indeed regulates the expression of its downstream gene. Received 09 November 2022; Accepted 23 March 2023; Published 26 May 2023 Author affiliations: 1Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, CT 06511, USA; 2Howard Hughes Medical Institute, Yale University, New Haven, CT 06511, USA; 3Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT 06511, USA. *Correspondence: Ronald R. Breaker, ronald. breaker@ yale. edu Keywords: attenuation; gene regulation; intergenic region; RNA motif. Abbreviations: AHAS, acetohydroxyacid synthase; BCAA, branched- chain amino acid; GMM, Spizizen glucose minimal medium; IGR, intergenic region; IR, isomeroreductase; LB, lysogeny broth; NCBI, National Center for Biotechnology Information; ncRNA, noncoding RNA; ORF, open reading frame; uORF, upstream open reading frame; UTR, untranslated region. ‡Present address: Abcam, Branford, CT 06405, USA. †These authors contributed equally to this work Data statement: All supporting data, code and protocols have been provided within the article or through supplementary data files.Three supplementary files are available with the online version of this article. 001019 © 2023 The Authors This is an open- access article distributed under the terms of the Creative Commons Attribution License. This article was made open access via a Publish and Read agreement between the Microbiology Society and the corresponding author’s institution. 1 OPENDATAOPENACCESSImpact Statement Gene regulation by upstream open reading frames (uORFs) is common in bacteria and they are often involved in adjusting the levels of proteins involved in fundamental metabolic processes. Here we determined that a conserved RNA sequence and structural motif present upstream of the ilvBNC operon in many Corynebacterium species is composed of numerous representa- tives of a uORF- encoding translation attenuation system reported previously. The main operon codes for proteins involved in branched- chain amino acid biosynthesis, which are sometimes regulated by ppGpp riboswitches. Analysis of more than 100 unique representatives of the ‘ilvB motif’ reveals that competing hairpins predicted to function as intrinsic terminator and anti- terminator stems are likely to be differentially formed based on the speed of uORF translation. The computational and genetics results presented indicate that attenuation mechanisms associated with ilvBNC operons and related genetic elements are widespread in bacteria. However, experimental data also suggest that the RNA motif does not function as a ppGpp riboswitch, which prompted further examination of the mechanism of ilvB gene regulation in Corynebacterium. Previous studies have shown that in Escherichia coli and in Corynebacterium glutamicum, an upstream open reading frame (uORF) associated with the ilvBNC operon controls its expression through an attenuation mechanism involving terminator and antiterminator sequences [15–17]. Different speeds of translation of the leader peptide encoded by the uORF result in changes to the levels of ilvB gene expression, often by affecting the formation of terminator and antiterminator stems. This precedent is consistent with features we observe in the motif identified in the present study. Specifically, the ilvB- II motif includes a conserved start codon always in- frame with a stop codon, and its conservation pattern is consistent with a protein coding region. Further- more, we show that previously reported RNA motifs associated with ilvB genes in various bacterial species might also carry uORFs. These findings support the hypothesis that attenuation mechanisms are widely used for bacterial ilvB gene regulation. METHODS Computational search for novel RNA motifs in Corynebacterium The ilvB- II motif was detected as part of an effort to identify and classify functional ncRNAs in a collection of 50 bacterial genomes (Breaker Laboratory, unpublished data). Specifically, the motif was identified while searching the C. pseudotubercu- losis genome [National Center for Biotechnology Information (NCBI) reference sequence: NC_017945.2]. The computational search pipeline [8, 9] is briefly described as follows. For each genome examined, all the intergenic regions (IGRs) that are long and enriched in G and C nucleotides relative to the average for the species were first identified. IGRs that met the length and GC content criteria were used as queries for Infernal [18]. Infernal enables the identification of additional IGRs that share sequence and structure similarities with query alignments, and was applied for the analysis of Reference Sequence (RefSeq) database release 80 and of metagenomic datasets [6, 19]. After an initial set of representatives was identified, CMfinder [20] was used to create a 2D structural model for the putative structured RNA class, based on covariance analysis. The Infernal search was then repeated with the resulting structural model to identify additional representatives. This collection of sequences is referred to as a ‘motif ’, and the BLISS server (Breaker Laboratory Intergenic Sequence Server) [21] is used to aid in evaluating the annotations of genes in the same genomic neighbourhood. The motif is assigned a putative function based on various factors, such as its sequence, structure, orientation relative to surrounding genes and the annotations of the functions of surrounding genes. Finally, the R2R software program [22] is used to depict a consensus sequence and secondary structure model, highlighting covariation and conservation patterns. In-line probing assays In- line probing assays [23] were used to evaluate the structural model and to assess direct binding of ligands using a described previously protocol [24]. The DNA oligonucleotides used to prepare (via PCR amplification) the double- stranded DNA template for in vitro transcription of the RNA construct are as follows: ilvB- 103F 5′- TAATACGACTCACTATAgg AACA TTAT TCGA CTTG TAGT GCTA TCCG AGCG GCAC CTGC CGTA ACGG CCACCA; ilvB- 103R 5′- GCCC CCGA TCAG CACT TGTG ATGC TGGC GAGG GCGC TTAC GTTA CGAC TTGG TGGC CGTT ACGG CAGG TGCCGC. Genetic constructs and bacterial cells DNA constructs used for in vitro transcription or for cloning of ilvB- II motif RNAs based on the genome of C. ulcerans 809 were supplied by Integrated DNA Technologies (Coralville, IA, USA). DNAs were amplified using PCR, and reporter gene constructs were cloned between a constitutive Bacillus subtilis lysC promoter and the E. coli lacZ gene on plasmid pDG1661 to 2 Narunsky et al., Microbial Genomics 2023;9:001019create a transcriptional fusion. Plasmids carrying the ilvB- II motif reporter fusion constructs were subsequently transformed into B. subtilis 1A1 cells and integrated into the bacterial chromosome. Selection for proper chromosomal integration was achieved by using 5 µg ml−1 of chloramphenicol. Counter selection was accomplished using 100 µg ml−1 spectinomycin. Genetic reporter assays Reporter assays were performed in liquid culture by inoculating B. subtilis in 2 ml of either lysogeny broth (LB) or Spizizen glucose minimal medium (GMM) [25], supplemented with 5 µg ml−1 chloramphenicol. The cultures were incubated overnight at 37 °C and then diluted 1/100 in LB or 1/10 in GMM with appropriate supplements [chloramphenicol at 5 µg ml−1 and X- gal (5- bromo- 4- chloro- 3- indolyl-β- d - galactopyranoside) at 100 µg ml−1]. The resulting mixtures were incubated at 37 °C for 24 h and images were recorded. Experiments were performed in triplicate, and representative data are depicted herein. Computational analysis of the uORF protein sequence The uORF peptide sequence was translated using the Transeq tool [26]. The MAFFT sequence aligner was used to create a multiple sequence alignment of the peptides [27], and the WebLogo v3.7.4 tool was used to create a visual figure representing this alignment [28]. RESULTS AND DISCUSSION Identifying the ilvB-II motif in Corynebacterium We identified 103 unique sequence representatives in the genus Corynebacterium that conform to a specific sequence and secondary structure pattern we have named the ilvB- II motif due to their consistent location in the 5′ untranslated region (UTR) of an ilvB gene (File S1, available in the online version of this article). The structural model corresponding to the alignment includes two predicted stem loops called P1 (pairing element 1) and P2 (Fig. 1a). Both P1 and P2 carry several conserved nucleotides and exhibit a pattern of covariation in some locations. P2 is followed by a stretch of uridine nucleotides, which matches the arrangement observed for most intrinsic transcription terminator stems [29]. We also noted that the motif appears to include a uORF spanning most of the P1 region of the RNA (Fig. 1b). Previous bioinformatic searches also revealed that some ilvB genes are likely controlled by uORF- mediated transcription attenuation [30–32]. Nonetheless, because the motif is located upstream of a gene that in some bacteria is known to be regulated by a ppGpp riboswitch [13], we were motivated to test whether it represents a new riboswitch class for this nucleotide- like signalling molecule. A structural probing assay called in- line probing [23, 24] was applied, which revealed that a representative ilvB- II motif RNA from C. ulcerans (Fig. 2a) indeed conforms to the predicted structure (Fig. 2b). However, the RNA does not exhibit structural modulation when ppGpp is introduced into the in- line probing reaction, suggesting that the RNA is unlikely to function as a riboswitch that directly senses ppGpp. Riboswitch aptamers commonly exhibit RNA folding differences in the absence versus presence of their target ligands [13]. Given the lack of evidence for ppGpp binding, we decided to investigate whether the motif regulates the expression of its downstream gene, perhaps by another mechanism. In vivo regulation of gene expression by an ilvB-II motif representative We evaluated the gene control function of the representative ilvB- II motif RNA from C. ulcerans by transforming B. subtilis with a plasmid containing a transcriptional fusion of wild- type (WT) or mutant (M1) examples of the motif and a lacZ reporter gene (Fig. 3a). This construct has the potential to form two distinct structures corresponding either to the ‘OFF’ state that forms a terminator stem (Fig. 3a, top), or to the ‘ON’ state that forms the antiterminator structure (Fig. 3a, bottom). Specifically, the right shoulder of the lower portion of P2 (denoted P2a) (orange shading) can form an alternative base- pairing interaction with a region of the loop of P1 to form a putative antiterminator stem (blue shading). Reporter strains carrying either the WT or M1 constructs were grown in either rich (LB) or minimal (GMM) media containing the β-galactosidase substrate X- gal, and reporter gene regulation was evaluated by visual inspection (Fig. 3b). The relative intensity of blue colour of the culture media is indicative of lacZ expression. With the WT construct, reporter gene expression was high in GMM in comparison to LB, revealing that the ilvB- II motif representative strongly suppresses reporter gene expression in rich media. This finding is consistent with the motif functioning as a genetic OFF switch, wherein the signalling molecule is abundant in rich media. Furthermore, this conclusion matches that reported previously for the regulation of ilvB gene expression [13, 16]. The M1 construct, which carries a mutation in a strictly conserved nucleotide in the consensus model (Fig. 3a) (U23A) exhibits no difference in the levels of gene expression (Fig. 3b), suggesting that this nucleotide is not essential for the regulatory effect. A conserved nucleotide in a riboswitch candidate is usually indicative of a functionally important nucleotide, and therefore a mutation at this position would be expected to disrupt riboswitch activity. If the motif functioned as a riboswitch that turns off expression when its ligand is present, a mutation that disrupted the aptamer would likely activate expression in rich media compared to the WT sequence. Given these results, we considered other hypotheses regarding the mechanism of gene control that do not involve riboswitch function. 3 Narunsky et al., Microbial Genomics 2023;9:001019Fig. 1. A conserved RNA sequence and structure upstream of ilvB genes in the genus Corynebacterium. (a) Consensus sequence and a secondary structure model of the ilvB- II motif, based on 103 representatives from various species of Corynebacterium. I/L/V identifies nucleotide triplets coding for isoleucine, leucine and valine amino acids. The denoted start and stop codons identify a predicted uORF. Predicted terminator and antiterminator base- paired substructures are also identified. (b) Consensus amino acid sequence of the short peptide encoded by the predicted ilvB uORF, as derived from the representatives identified in Corynebacterium (File S1) and generated by WebLogo [24]. Although the start codon was AUG, GUG and UUG, we set the first amino acid of the peptide to methionine. The ilvB-II motif exhibits the characteristics of a uORF Previous studies in various bacterial species revealed that some ilvB genes are regulated either by a riboswitch or by a uORF [13, 16]. Riboswitches for ppGpp frequently regulate more than a single mRNA within the genomes of species that use them [13], whereas the RNA motif we identified is always found in a consistent location and only once in each genome. Representatives are typically located ~150 nucleotides upstream of the start codon for the ORF of the ilvB gene. Previous research revealed that ilvB in C. glutamicum is regulated by a uORF [16]. Importantly, our alignment of ilvB- II motif sequences (File S1) includes a representative from a strain of this same species (NCBI reference sequence: NC_009342.1). This observation reveals that this previously validated uORF matches the consensus for the ilvB- II motif identified in the current study (Fig. 1a). As noted above, we recognized that a conserved start codon (AUG/GUG) resides at the beginning of the motif (Fig. 1a), always positioned in- frame with a downstream stop codon. The distance between the beginning of the start codon and the beginning of the associated stop codons is usually 45, 48, or 51 nucleotides, which defines a uORF coding for a peptide of 15–17 amino acids (Fig. 1b). The previously reported uORF is 45 nucleotides in length [16], which conforms to the consensus we have developed. However, some 4 Narunsky et al., Microbial Genomics 2023;9:001019Fig. 2. In- line probing of the ilvB- II motif from C. ulcerans confirms the bioinformatically predicted secondary structure. (a) Sequence and secondary structure model of the RNA construct subjected to in- line probing. The RNA sequence (called 4–106 ilvB- II) encompasses nucleotide positions 4 through 106 of the natural C. ulcerans sequence plus two G nucleotides (lowercase ‘g’ letters) added at the 5′ end to facilitate preparation by in vitro transcription. (b) Autoradiogram of the products of in- line probing reactions with the 5′ 32P- labelled 4–106 ilvB- II construct after separation by denaturing (8M urea) 10 % polyacrylamide gel electrophoresis (PAGE). T1 indicates the precursor RNA (Pre) was RNA subjected to partial degradation by treatment with RNase T1, which cleaves after G nucleotides (certain bands labelled). ‒OH indicates partial digestion of Pre RNA under alkaline conditions, which cleaves after all nucleotides. The asterisk identifies a compression site where several product bands migrate to the same location. In- line probing reactions were conducted either in the absence (‒) or presence (+) of 100 µM ppGpp. Bands corresponding to nucleotides involved in the predicted secondary (p1 and p2) structures are highlighted, and generally exhibit reduced spontaneous strand scission relative to regions in bulges, joining regions or loops (such as L1 and L2). 5 Narunsky et al., Microbial Genomics 2023;9:001019Fig. 3. Genetic reporter fusion assays for the evaluation of an ilvB- II motif RNA representative. (a) Sequence and secondary structure models of the C. ulcerans ilvB- II motif representative prepared as a transcriptional fusion with the E. coli lacZ gene. Top: predicted structure of the RNA in the genetic ‘OFF’ state, wherein the lower part of stem P2 (denoted P2a; orange shading) participates in forming an intrinsic terminator stem. Bottom: predicted structure of the RNA in the genetic ‘ON’ state, wherein an antiterminator stem (blue shading) forms at the expense of the P2a stem. Red nucleotides identify positions in the consensus model (Fig. 1a) that are conserved in at least 97 % of the sequences present in the alignment. Encircled numbers represent natural sequences from the C. ulcerans ilvB gene that are not depicted. Construct M1 is a mutant version of the WT sequence that carries a U23A mutation. (b) lacZ reporter gene assays of the WT and M1 ilvB- II reporter constructs depicted in (a) cultured with rich (LB) or minimal (GMM) media. Blue colour indicates high lacZ expression. 6 Narunsky et al., Microbial Genomics 2023;9:001019Fig. 4. Previously identified RNA motifs associated with bacterial ilvB genes. (a) Consensus sequence and structural model for the ilvB- OMG motif updated from a model reported previously [6] (File S2). Annotations are as described for Fig. 1(a). (b) Consensus sequence and structural model for ilvB motif updated from a model reported previously [9]. (c) Consensus amino acid sequence of the short peptide encoded by the uORF predicted to be present in the ilvB- OMG motif [6] (File S3). Details are as described for Fig. 1(b). The height of each letter in the figure represents its frequency, corrected to the number of peptide sequences in the alignment. (d) Consensus amino acid sequence of the short peptide encoded by the uORF predicted to be present in the ilvB motif [9]. There are only four unique peptide sequences, and the height of each letter in the figure is corrected accordingly. representative uORFs are as short as 30 nucleotides or as long as 72 nucleotides (excluding the stop codon) (File S1), but all lengths are divisible by three, suggesting codon function. The consensus sequence also is enriched in codons for the three BCAAs, which are primarily located in four positions (Fig. 1a, I/L/V annotations). The M1 construct examined in our reporter gene assays alters the second of these codons by changing a valine codon to a glutamate codon. However, there are other I/L/V codons in the sequence, which might explain why the mutation does not disrupt gene regulation. Additional RNA motifs associated with ilvB genes suggest broader use of attenuation mechanisms Two additional RNA motifs associated with ilvB genes from other bacterial lineages have been reported by our laboratory previously [6, 9]. Based on the rarity and the lack of clues regarding the possible functions of these motifs, our hypotheses regarding possible functions were uncertain. Specifically, the first RNA structure, called the ilvB- OMG motif (Fig. 4a; File S2), was noted as an RNA of unknown function [6]. The second structure, simply called the ilvB motif (Fig. 4b; File S3), was considered a weak riboswitch candidate [9], meaning that there was low confidence in the riboswitch hypothesis. Thus, we had prioritized other candidate riboswitch classes for experimental investigation over the analysis of these motifs. Considering the findings described herein with the Corynebacterium ilvB- II motif, we reexamined the sequence alignments and features of the ilvB- OMG and ilvB motifs to determine whether the uORF mechanism could be more broadly distributed than originally known. The ilvB- OMG motif [6] consensus sequence and structural model was updated based on 41 representatives from various bacterial phyla. All but two carry a start codon in- frame with a downstream stop codon, suggesting that these sequences might function as uORFs. Three other representatives were much longer than others, and were removed from the protein alignment despite carrying a similar amino acid consensus sequence. Although the peptides that would result from translation of these predicted uORFs are highly variable in length and sequence, they are rich in codons for BCAAs and threonine (Fig. 4c). These features of ilvB- OMG motif RNAs are suggestive of attenuation function in response to the availability of aminoacylated BCAA tRNAs. The ilvB motif [9] only has seven representatives present in species of Leptospira. After updating the consensus sequence and structural model (Fig. 4b), this motif appears to carry a start codon in- frame with a downstream stop codon, located ~70 nucleotides upstream of the main ORF. The RNA sequence exhibits a conservation pattern consistent with a uORF region that is also rich in BCAAs and in threonine (Fig. 4d). For example, all representatives have at least three consecutive valine codons. Again, this architecture is consistent 7 Narunsky et al., Microbial Genomics 2023;9:001019with an attenuation function, although confirmation of the functions of the ilvB- OMG and ilvB motifs requires additional genetic experimentation. CONCLUSION Previous studies [15–17] have demonstrated that the leader sequence of the ilvB gene can control the expression levels of ilvB by exploiting a uORF regulating the formation of an intrinsic terminator stem [29, 33]. Likewise, we observe a conserved uORF associated with a strong terminator stem (pairing element P2) in the leader sequences of the ilvB mRNAs of many Corynebacterium species. Furthermore, when fused to a lacZ reporter gene, the ilvB- II motif from the mRNA leader sequence of C. ulcerans exhibits high expression levels in minimal media and low levels in rich media (Fig. 3b). These observations suggest that the motif representative from this species also regulates the expression of its associated main ORF by encoding a short peptide enriched in BCAAs, which can only be translated quickly when sufficient concentrations of isoleucine, leucine and valine are present in the media. Thus, we speculate that the secondary structure features characteristic of the ilvB motif are differentially formed based on the speed of translation of the uORF, which would cause transcription termination if the P2 stem (intrinsic terminator hairpin) is formed. This proposed uORF- mediated gene regulation mechanism for ilvB resembles the attenuation mechanism originally reported for the regulation of tryptophan biosynthesis genes [33]. In the abundance of tryptophan, a uORF is translated quickly, preventing the formation of an antiterminator hairpin. This, in turn, allows a terminator stem to form, resulting in decreased levels of RNA transcription of the adjacent gene. Most known classes of riboswitches are widespread among diverse lineages of bacteria [12, 34]. However, the riboswitch classes discovered more recently tend to be rarer and more narrowly distributed phylogenetically. Therefore, structured RNA motifs that are only present in a few closely related species might be less likely to function as metabolite- binding riboswitches, but rather serve as gene control devices that are bound by protein- based genetic factors or as uORFs like those described herein. Regardless, we anticipate that many novel types of RNA- based regulatory systems remain to be discovered. Funding information This research was supported by NIH grants (GM022778 and AI136794) to R.R.B. Research in the Breaker laboratory is also supported by the Howard Hughes Medical Institute. Acknowledgements We thank members of the Breaker Laboratory for helpful discussions and comments. We thank Breaker laboratory members Dr Gadareth Higgs for preliminary bioinformatic identification of the ilvB- II motif, Randall Kras for his help constructing the strains and Narasimhan Sudarsan for construc- tive comments on the manuscript. Author contributions A.N., K.K., S.S.S.P. and R.R.B. established the experimental validation research plan, A.N., K.K., S.S.S.P. and M.F. performed the experiments, and all authors analysed the data. R.R.B. supervised the project. A.N., K.K. and R.R.B. wrote the manuscript, and all authors contributed edits. Conflicts of interest The authors declare that there are no conflicts of interest. References 1. Keilhauer C, Eggeling L, Sahm H. Isoleucine synthesis in Corynebac- terium glutamicum: molecular analysis of the ilvB- ilvN- ilvC operon. J Bacteriol 1993;175:5595–5603. 2. Singh BK, Shaner DL. Biosynthesis of branched chain amino acids: from test tube to field. Plant Cell 1995;7:935–944. 3. Neinast M, Murashige D, Arany Z. Branched chain amino acids. Annu Rev Physiol 2019;81:139–164. 4. Backofen R, Amman F, Costa F, Findeiß S, Richter AS, et al. Bioin- formatics of prokaryotic RNAs. RNA Biol 2014;11:470–483. 5. Backofen R, Engelhardt J, Erxleben A, Fallmann J, Grüning B, et al. RNA- bioinformatics: Tools, services and databases for the analysis of RNA- based regulation. J Biotechnol 2017;261:76–84. 6. Weinberg Z, Lünse CE, Corbino KA, Ames TD, Nelson JW, et al. Detection of 224 candidate structured RNAs by comparative anal- ysis of specific subsets of intergenic regions. Nucleic Acids Res 2017;45:10811–10823. 7. Weinberg Z, Nelson JW, Lünse CE, Sherlock ME, Breaker RR. Bioinformatic analysis of riboswitch structures uncovers variant ligand specificity. Proc Natl Acad Sci classes with altered 2017;114:E2077–E2085. 8. Stav S, Atilho RM, Mirihana Arachchilage G, Nguyen G, Higgs G, et al. Genome- wide discovery of structured noncoding RNAs in bacteria. BMC Microbiol 2019;19:66. 9. Brewer KI, Greenlee EB, Higgs G, Yu D, Mirihana Arachchilage G, et al. Comprehensive discovery of novel structured noncoding RNAs in 26 bacterial genomes. RNA Biol 2021;18:2417–2432. 10. Sherwood AV, Henkin TM. Riboswitch- mediated gene regulation: novel RNA architectures dictate gene expression responses. Annu Rev Microbiol 2016;70:361–374. 11. Lotz TS, Suess B. Small- molecule- binding riboswitches. Microbiol Spect 2018;6:4. 12. Kavita K, Breaker RR. Discovering riboswitches: the past and the future. Trends Biochem Sci 2023;48:119–141. 13. Sherlock ME, Sudarsan N, Breaker RR. Riboswitches for the alarmone ppGpp expand the collection of RNA- based signaling systems. Proc Natl Acad Sci 2018;115:6052–6057. 14. Hacker E, Antunes CA, Mattos- Guaraldi AL, Burkovski A, Tauch A. Corynebacterium ulcerans, an emerging human pathogen. Future Microbiol 2016;11:1191–1208. 15. Hauser CA, Hatfield GW. Attenuation of the ilvB operon by amino acids reflecting substrates or products of the ilvB gene product. Proc Natl Acad Sci 1984;81:76–79. 8 Narunsky et al., Microbial Genomics 2023;9:001019 16. Morbach S, Junger C, Sahm H, Eggeling L. Attenuation control of ilvBNC in Corynebacterium glutamicum: evidence of leader peptide formation without the presence of a ribosome binding site. J Biosci Bioeng 2000;90:501–507. 17. Salmon KA, Yang CR, Hatfield GW. Biosynthesis and regulation of the branched- chain amino acids. EcoSal Plus 2006;2. 18. Nawrocki EP, Eddy SR. Infernal 1.1: 100- fold faster RNA homology searches. Bioinformatics 2013;29:2933–2935. 19. Tatusova T, DiCuccio M, Badretdin A, Chetvernin V, Nawrocki EP, et al. NCBI prokaryotic genome annotation pipeline. Nucleic Acids Res 2016;44:6614–6624. 20. Yao Z, Weinberg Z, Ruzzo WL. CMfinder—a covariance model based RNA motif finding algorithm. Bioinformatics 2006;22:445–452. 21. Barrick JE, Corbino KA, Winkler WC, Nahvi A, Mandal M, et al. New RNA motifs suggest an expanded scope for riboswitches in bacte- rial genetic control. Proc Natl Acad Sci 2004;101:6421–6426. 22. Weinberg Z, Breaker RR. R2R- software to speed the depiction of aesthetic consensus RNA secondary structures. BMC Bioinfor- matics 2011;12:3. 23. Soukup GA, Breaker RR. Relationship between internucleotide linkage geometry and the stability of RNA. RNA 1999;5:1308–1325. 24. Regulski EE, Breaker RR. In- line probing analysis of riboswitches. Methods Mol Biol 2008;419:53–67. 26. Rice P, Longden I, Bleasby A. EMBOSS: the European Molecular Biology Open Software Suite. Trends Genet 2000;16:276–277. 27. Katoh K, Standley DM. MAFFT multiple sequence alignment soft- ware version 7: improvements in performance and usability. Mol Biol Evol 2013;30:772–780. 28. Crooks GE, Hon G, Chandonia JM, Brenner SE. WebLogo: a sequence logo generator. Genome Res 2004;14:1188–1190. 29. Roberts JW. Mechanisms of bacterial transcription termination. J Mol Biol 2019;431:4030–4039. 30. Seliverstov AV, Putzer H, Gelfand MS, Lyubetsky VA. Comparative analysis of RNA regulatory elements of amino acid metabolism genes in Actinobacteria. BMC Microbiol 2005;5:54. 31. Vitreschak AG, Lyubetskaya EV, Shirshin MA, Gelfand MS, Lyubetsky VA. Attenuation regulation of amino acid biosynthetic operons in proteobacteria: comparative genomics analysis. FEMS Microbiol Lett 2004;234:357–370. 32. Lopatovskaya KV, Seliverstov AV, Lyubetsky VA. Attenuation regulation of the amino acid and aminoacyl- tRNA biosynthesis operons in bacteria: a comparative genomic analysis. Mol Biol 2010;44:128–139. 33. Bertrand K, Squires C, Yanofsky C. Transcription termination in vivo in the leader region of the tryptophan operon of Escherichia coli. J Mol Biol 1976;103:319–337. 25. Anagnostopoulos C, Spizizen J. Requirements for transformation 34. McCown PJ, Corbino KA, Stav S, Sherlock ME, Breaker RR. Ribos- in Bacillus subtilis. J Bacteriol 1961;81:741–746. witch diversity and distribution. RNA 2017;23:995–1011. Five reasons to publish your next article with a Microbiology Society journal When you submit to our journals, you are supporting Society activities for your community. 1. Experience a fair, transparent process and critical, constructive review. 2. If you are at a Publish and Read institution, you’ll enjoy the benefits of Open Access across 3. our journal portfolio. Author feedback says our Editors are ‘thorough and fair’ and ‘patient and caring’. Increase your reach and impact and share your research more widely. 4. 5. Find out more and submit your article at microbiologyresearch.org. 9 Narunsky et al., Microbial Genomics 2023;9:001019
10.1093_nar_gkad633	Published online 31 July 2023 Nucleic Acids Research, 2023, Vol. 51, No. 17 8957–8969 https://doi.org/10.1093/nar/gkad633 Linking folding dynamics and function of SAM / SAH riboswitches at the single molecule level Ting-Wei Liao 1 , Lin Huang 2 , Timothy J. Wilson 3 , Laura R. Ganser 1 , David M.J. Lilley 3 and Taekjip Ha 1 , 4 , 5 , * 1 Department of Biophysics, Johns Hopkins University, Baltimore, MD 21218, USA, 2 Guangdong Provincial Key Laboratory of Malignant Tumor Epigenetics and Gene Regulation, Guangdong-Hong Kong Joint Laboratory for RNA Medicine, Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University, Guangzhou 510120, China, 3 Nucleic Acid Structure Research Group, MSI / WTB Complex, The University of Dundee, Dundee, Dow Street, Dundee DD1 5EH, UK, 4 Department of Biophysics and Biophysical Chemistry, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA and 5 Ho w ard Hughes Medical Institute, Baltimore, MD, USA Received April 19, 2023; Revised June 27, 2023; Editorial Decision July 12, 2023; Accepted July 18, 2023 ABSTRACT GRAPHICAL ABSTRACT Riboswitches are regulatory elements found in bacterial mRNAs that control downstream gene expression through ligand-induced conformational chang es. Here , we used single-molecule FRET to map the conformational landscape of the translational SAM / SAH riboswitch and probe how ligand-induced conformational co-transcriptional c hanges aff ect its translation regulation function. Ri- boswitch folding is highly heterog eneous, sugg est- ing a rugged conformational landscape that allows for sampling of the ligand-bound conformation even in the absence of ligand. The addition of ligand shifts the landscape, favoring the ligand-bound con- formation. Mutation studies identiﬁed a key struc- tural element, the pseudoknot helix, that is crucial for determining ligand-free conformations and their ligand responsiveness. We also investigated ribo- somal binding site accessibility under tw o scenar - ios: pre-folding and co-transcriptional folding. The regulatory function of the SAM / SAH riboswitch in- volves kinetically favoring ligand binding, but co- transcriptional folding reduces this preference with a less compact initial conformation that exposes the Shine–Dalgarno sequence and takes min to redis- tribute to more compact conformations of the pre- folded riboswitc h. Suc h slow equilibration decreases the effective ligand afﬁnity. Overall, our study pro- vides a deeper understanding of the complex folding process and how the riboswitch adapts its folding pattern in response to ligand, modulates ribosome accessibility and the role of co-transcriptional fold- ing in these processes. INTRODUCTION Riboswitches are regulatory units of RNA that mediate gene expression in response to binding of specific metabo- lites. They are widely found in bacteria ( 1–3 ) but also exist in archaea ( 4 ), plants ( 5 ) and fungi ( 6 , 7 ). To date, > 40 classes of riboswitches have been discovered, and they bind chem- ically di v erse ligands and contribute up to 4% to the bacte- ria genetic control, especially in gram positi v e bacteria. Ri- (cid:2) -untransla ted regions boswitches are mostly loca ted a t the 5 (cid:2) -UTR), upstream of the regulated genes, and include an (5 a ptamer domain ca pable of binding a particular metabolite with exceptionally high specificity. The riboswitch adopts a specific fold on binding the ligand, leading to up- or down-regulation of the gene either by altering transcrip- tion or translation. Since the riboswitch folds and acts as a regulatory unit during transcription, the timing of ligand * To whom correspondence should be addressed. Tel: +1 217 398 0865; Email: [email protected] C (cid:3) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 8958 Nucleic Acids Research, 2023, Vol. 51, No. 17 binding and conformational change is critical, necessitating investigation into its folding kinetics. Substantial r esear ch has been devoted to the origins of specificity, correlation of sequence and structure ( 8 , 9 ), folding kinetics ( 10–12 ), and identification of candidates that can be adapted for drug- deli v ery ( 13 ) and in vivo imaging ( 14 ). (SAM)-binding S-adenosylmethionine riboswitches comprise one of the largest classes of riboswitches ( 1 ). SAM is synthesized from methionine and ATP by SAM synthetase, encoded by the metK gene. SAM is an es- sential co-substrate of methyltransferases, supplying the methyl group for methyl transfer. Once the methyl group is donated, the resulting S -adenosyl- L -homocysteine (SAH) is degraded due to its toxicity ( 15 , 16 ). To main- tain SAM concentration, SAM acts as an inhibitor of MetK synthesis ( 17–20 ) . This regulation is achie v ed by the SAM-riboswitches, which bind SAM and acts as negati v e feedback unit for genes in methionine or cysteine biosynthesis. W hen SAM concentra tion goes up, expres- sion of genes in methionine or cysteine biosynthesis is (cid:2) -UTR adopting transla tion-of f reduced by its upstream 5 conformation. Six sub-classes of SAM riboswitches (SAM-I to SAM- VI) have been identified, classified into three families ac- cording to their structural features ( 17–25 ). In general, SAM riboswitches exhibit strong discrimination between SAM and SAH by electrostatically interacting with the positi v e-charged sulfonium cation of the SAM molecule ( 26–33 ), as previously shown by X-ray crystallography and single-molecule methods ( 22 , 34 ). By contrast, the SAM / SAH riboswitch does not discriminate between SAM and SAH ( 35 , 36 ). The ligand binding interactions of this particular SAM / SAH riboswitch have been pr eviously r evealed by NMR and X-ray crystallo gra phy ( 35 , 36 ). Binding of SAM or SAH is accompanied by the formation of three base pairs that extend the helix of the stem-loop P1, and formation of a pseudoknot helix PK (Supplementary Figure S1A). The ligand binds in the major groove of the extended he- lix, with the methionyl nitrogen and the adenine moiety hy- drogen bonded to a specific cytosine nucleobase. The me- thionine side chain containing the sulfonium of SAM or the thioether of SAH does not make any direct contact with the RN A, w hich explains the inability to distinguish between the two ligands. Single-molecule FRET (fluorescence res- onance energy transfer) ( 37 ) was utilized to compare the binding of SAM and SAH and their kinetic characteristics ( 36 ), and no significant differences were observed between the two ligands. Although the ligand-bound state and basic kinetics have been characterized, important features such as ligand-free conformations, binding, folding kinetics, and its role in modula ting transla tion initia tion activity are still unknown. Here, we used single-molecule FRET to investigate the ligand-free and ligand-bound conformations of the SAM / SAH riboswitch and map the energy landscape of folding dynamics and altered ribosome accessibility. Fold- ing of the riboswitch is highly heterogeneous, suggesting a rugged conformational landscape that allows for sam- pling of the ligand-bound conformation e v en in the ab- sence of ligand. The addition of ligand shifts the land- scape, favoring the ligand-bound conformation. Site spe- cific mutations showed that the PK helix is crucial for de- ter mining ligand-free confor mations and their ligand re- sponsi v eness. In addition, we investigated the accessibility of the ribosomal binding site under two scenarios: ( i ) pre- folding of the riboswitch: folding equilibrium is reached in advance and ( i i) vectorial release of the RNA by mim- icking co-transcriptional folding. Vectorial folding initially favors an open conformation that exposes the ribosome binding site, and it takes min bef ore conf ormational redis- tribution to that of the pre-folded riboswitch. Such slow equilibration decreases the effecti v e ligand affinity. Over- all, our studies offer a deeper understanding of the com- plexity of the folding process, revealing the mechanism by which the riboswitch adapts its folding pattern in response to ligand and modulates ribosome accessibility, and how co- transcriptional folding influences these processes. MATERIALS AND METHODS Riboswitch ligands SAM (A7007), SAH (A9384) were all obtained from Sigma. The RNAs (wide-typed and mutants) are synthesized as described in the following sections. DNA oligonucleotides for mimicking ribosome binding were pur- chased from Integrated DNA Technologies (Coralville). RNA synthesis for single-molecule experiments The wild-type and mutated SAM / SAH riboswitches for single molecule measurements contain a Cy3 flu- orophore attached to the O2’ of U20 generated by Cu 2+ -catalyzed reaction of alkyne-modified RNA with an azide-attached fluorophore (Lumiprobe Corp). (cid:2) DNA exten- The wild-typed RNA had an 18 nt 3 sion for base-pairing to the anchor DNA, and the complete sequence was (DNA starts with d and under- scored): GAUACCUGUCACAACGGCU(U-Cy3)CCU GGCGUGA CGAGGUGA CCUCAGUGGAGCAA d( ACCGCTGCCGT CGCT CCG ), and all the other mu- tated sequences were showed in Supplementary Table S1. (cid:2) -biotin and 3 The anchor DNA had a 5 (cid:2) Cy5 flu- orophore and was complementary to the 18-nt ex- tension of the SAMSAH riboswitch strand. Its se- quence was: biotin-CGGA GCGACGGCA GCGGT-Cy5. RNA oligonucleotides were synthesized using t-BDMS phosphoramidite chemistry ( 38 ) as described in Wil- son et al. ( 39 ), implemented on an Applied Biosys- tems 394 DN A / RN A synthesizer. RN A was synthesized using ribonucleotide phosphoramidites with 2 -Otter- butyldimethyl-silyl (t-BDMS) protection (Link Technolo- gies) ( 40 , 41 ). Oligonucleotides containing 5-bromocytidine (ChemGenes) were deprotected in a 25% ethanol / ammonia ◦C. All oligoribonucleotides were re- solution for 36 h at 20 dissolved in 100 (cid:2)l of anhydrous DMSO and 125 (cid:2)l tri- eth ylamine trih ydrofluoride (Sigma-Aldrich) to remove t- ◦C in the dark for 2.5 h. BDMS groups, and agitated at 65 After cooling on ice for 10 min, the RNA was precipitated with 1 ml of butanol, washed once with 70% ethanol and suspended in double-distilled w ater. RNA w as further puri- fied by gel electrophoresis in polyacrylamide under denatur- ing conditions in the presence of 7 M urea. The full-length RNA product was visualized by UV shadowing. The band was excised and electroeluted using an Elutrap Electroelu- tion System (GE Healthcare) into 45 mM Tris-borate (pH ◦C . The 8.5), 5 mM EDTA buf fer for 12 h. a t 150 V a t 4 RNA was precipitated with isopropanol, washed once with 70% ethanol and suspended in water or ITC buffer (40 mM HEPES-K (pH 7.0), 100 mM KCl, 10 mM MgCl 2 ). Ligand titration of pre-folded riboswitches: wild-typed and mutated riboswitches 40 pM of the pre-annealed SAM / SAH riboswitch immobilized on a neutravidin- molecules were functionalized, polymer-passivated surface and free molecules were washed out with T50 b uffer. Ima ge b uffer containing an oxygen-scavenging system was freshly mixed befor e measur ements , comprising 1% (w / v) dextrose , 2 mM Trolo x, glucose o xidase (1 mg / ml; Sigma-Aldrich), and catalase (500 U / ml; Sigma-Aldrich)] in buffer containing 40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 . All the ligands were diluted with image buffer immediately prior to measurements. The ligands were incubated for 5 min before imaging. Short movies (duration of 1.5 s: 20 frames) were collected for 30 field of view for generating distribution of FRET efficiencies ( E FRET ). The distribution is then fitted by two individual Gaussian function, and the high E FRET ratio is estimated accordingly. Single-molecule imaging and data acquisition Single-molecule FRET data were obtained using a prism- based total internal reflection fluorescence (TIRF) micro- scope. The Cy3 and Cy5 fluorophores were excited by a 532- nm laser (Coherent Compass 315M) and a 638-nm laser (Cobolt 06-MLD) respecti v ely. The fluorescence emission was collected by a water immersion objecti v e (Olympus NA 1.2, 60 ×) and recorded by a back-illuminated electron- m ultipl ying charge-coupled device camera (iXON, Andor Technology) with a dual-vie w setup. The dual-vie w setup used a long-pass emission filter (Semrock BLP02-561R- 25) for eliminating the 532-nm laser, and a notch filter (Chroma ZET633TopNotch) for eliminating the 638-nm laser. The fluorescence emission was separated into donor and acceptor emission by a long-pass dichr oic mirr or (Sem- rock FF640-FDi01-25X36). The passivated PEG quartz slides and coverslips were purchased from Johns Hopkins Slides Core and were assembled into a reaction cham- ber. ( 42 ) Spots detection, background subtraction, donor leakage and acceptor direct-excitation correction followed our previous protocol ( 42 ). Custom codes are available on GitHub ( https://github.com/Ha-SingleMoleculeLab ) and archi v ed in Zenodo with the following doi. Data acqui- sition DOI: 10.5281 / zenodo.4925630; Raw data analysis DOI: 10.5281 / zenodo.4925617. Nucleic Acids Research, 2023, Vol. 51, No. 17 8959 E FRET fluctuated between the middle and high values. To characterize further the dynamic species, the regions of dy- namics were collected and analyzed by ebFRET ( 43 ) and the two-state dwell time was plotted into log-scale scatter plot. Single-molecule data analysis of vectorially folded riboswitch Single-molecule traces showing the immobilized heterodu- plex was unwound were categorized into four types of be- havior. We classified the riboswitch folding behavior into four types (Supplementary Figure S7A): (i) molecules tran- sitioned from the heteroduplex state to the closed confor- mation without any detectable intermediate, then remain- ing there, (ii) molecules transitioned from the heteroduplex state to the open conformation, then remaining there, (iii) molecules transitioned from the heteroduplex state to one undergoing fluctuations between the open and closed con- formations, (iv) molecules transitioned from the heterodu- plex to fluctuating states after which they became locked in the closed conformation. Pseudo-functional readout of ribosome accessibility of pre- folded riboswitch 40 pM of the pre-annealed SAM / SAH riboswitch immobilized on a neutravidin- molecules were functionalized, polymer-passivated surface and free molecules were washed out with T50 buffer. All the ligand (SAM) and DNA oligonucleotides with a designated con- centration was freshly mixed with ima ge b uffer containing an oxygen-scavenging system. Short movies (duration of 1.5 sec: 20 frames) were collected for 30 field of view immedia tely or a t designa ted time (5-min or 1 h) after injection for generating distribution of FRET efficiencies ( E FRET ). Sample pr epar ation f or the single-molecule FRET measur e- ments For preparation of the pre-folded assay, 10 (cid:2)M of the SAM / SAH riboswitch molecule or the riboswitch mutants with internal Cy3 labeled was annealed with 15 (cid:2)M an- chored DNA with Cy5 and biotin label under 1 × T50 [10 mM Tris (pH 8.0), 50 mM NaCl) buffer follo wed by slo w ◦C to room temperature. cooling from 95 For preparing of the vectorial folding assays, 10 (cid:2)M an- chored DNA with Cy5 and biotin label was annealed with 20 (cid:2)M of the Cy3-labeled SAM / SAH riboswitch and 40 (cid:2)M complementary DNA oligos (cDNA) with dT30 over- hang in 10 (cid:2)l of 1 × T50 [10 mM Tris (pH 8.0), 50 mM NaCl] ◦C for 5 min, by incubating the mixture at 95 ◦C for 15 min and finally equilibrating at room tempera- 37 ture for 5 min ( 44 , 45 ). ◦C for 1 min, 75 Single-molecule data analysis of pr e-f olded riboswitch Vectorial folding as a mimic of riboswitch folding and ligand binding Single-molecule traces showing E FRET as a function of time were categorized into three types of behavior. (i) E FRET re- mained middle for the duration of observation, up to 1 min, (ii) E FRET remained high for the duration of observation (iii) Labeled and biotinylated heteroduplex es wer e immobilized on a neutravidin-functionalized surface and free heterodu- plex es wer e washed out. 50 nM of Rep-X ( 46 ) was incu- bated for 2 min with heteroduplexes in the imaging buffer 8960 Nucleic Acids Research, 2023, Vol. 51, No. 17 (40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 ) containing an oxygen-scavenging system, and images (du- ration of 1.5 s: 20 frames) were collected for 30 field of view for confirming the heteroduplex conformation. Unwind- ing was initiated by mixing unwinding buffer with / without ligands at designated concentration. Unless specified oth- erwise, the unwinding buffer contained 40 mM HEPES (pH 7.5), 100 mM KCl, 2 mM MgCl 2 , 2 mM ATP with an oxygen-scavenging system. Buffer with ATP then trig- gered the pre-bound RepX into unwinding the anchored heteroduplex. For the real-time observation (‘flow-in’ experiments) of riboswitch released from heteroduplex, imaging was started 12 s before the addition of the unwinding buffer. For char- acterization of VFA products after helicase unwinding, im- ages were taken after the addition of the unwinding buffer a t designa ted time. The loading and unwinding buf fers used during imaging contained additional 1% (w / v) dextrose, 2 mM Trolo x, glucose o xidase (1 mg / ml; Sigma-Aldrich), and catalase (500 U / ml; Sigma-Aldrich). Vectorial folding with ligand and ribosome mimic addition si- multaneously Labeled and biotinylated heteroduplex es wer e incubated with 50 nM Rep-X as previously described, images were taken before addition to confirm the heteroduplex confor- mations. Ligands, ribosome mimics (oligonucleotides with 9-nt or 15-nt complementary to ribosome binding site), and additional Rep-X (50 nM for 9-nt; 100 nM for 15-nt) wer e mix ed with unwinding buffer. The additional Rep-X is added in order to reduce the competition of free oligonu- cleotides to the Rep-X pre-incubated before unwinding mixture. This optimized condition is tested with a nega- ti v e control, where additional Rep-X with dT9 or dT15 were added sim ultaneousl y into the heteroduplex, no ob- servable loss of unwinding efficiency in this condition. The negati v e control experiments were shown in Supplementary Figure S10. RESULTS Heterogeneous folding energy landscape of ligand-free ri- boswitch First, we determined the conforma tional d ynamics of the SAM / SAH riboswitch in the absence of ligand. Single ri- boswitch molecules were tethered to the quartz slide by hy- (cid:2) extension to an oligonucleotide carry- bridization of a 3 (cid:2) termin us. The ribos witch con- ing a Cy5 acceptor at its 3 struct has a Cy3 donor attached internally within the loop region such that FRET efficiency, E FRET , between the two fluorophores can be used to distinguish between conforma- tions. We anticipated two major conformations: the open state with a stem-loop structure previously determined by in-line probing and the closed state with a H-type pseudo- knot (Figure 1 A, ( 31 )). The closed conformation likely has a global conformation similar to the crystal structure of the liganded riboswitch that showed the 8-bp extended P1 he- lix and 5-bp pseudoknot (PK) helix coaxially stacked with each other ( 36 ). We previously showed that Cy3 labeling in the loop region does not perturb folding ( 36 ). Single-molecule histograms of E FRET showed two major peaks, likely corresponding to the open conformation (mid- E FRET = 0.4, Figure 1 B) and the closed conformation (high- E FRET = 0.84, Figure 1 B), suggesting the ligand-bound con- formation is adopted e v en without ligand. Lowering mag- nesium concentration reduced the high- E FRET population but a significant percentage ( > 30%) of high- E FRET popula- tion remained e v en in the absence of Mg 2+ (Supplementary Figure S1B), suggesting divalent cations promote the closed conformation, but are not required. Single-molecule time traces of E FRET displayed three types of behavior: (i) constant mid-FRET ( E FRET = 0.4) (ii) constant high-FRET ( E FRET = 0.84) and (iii) dynamic be- havior showing transitions between mid- and high-FRET values (Figure 1 C). The majority (55%, Figure 1 D) of traces showed dynamic behavior, further indicating that e v en in the ligand-free state the closed conformation is sampled. The interconversion kinetics of the d ynamic popula tion was quantified by calculating the average dwell times for high and mid- E FRET states for each molecule and visualized as a log-scale scatter plot. The average dwell times covered a wide range, spanning up to 3 orders of magnitude (Fig- ure 1 E and F). In most cases, the open conformation was longer-li v ed than the closed conformation (Figure 1 G). The dynamic transitioning was a long-lasting characteristic with no clear population interconversions to or from constant mid- or high-FRET states within our experimental window, up to 50 min long (a typical trace shown in Supplemen- tary Figure S2A with zoom-in traces in Supplementary Fig- ure S2B-D, with intermittent 30 s e xposure e v ery 5 min). We attribute this ‘static heterogeneity’ to deep energy wells, and our preliminary investigation at higher temperature still showed static heterogeneity. Ligand binding reshapes the folding energy landscape Next, we measured riboswitch folding in the presence of the cognate ligand SAM. The high-FRET state indeed r epr e- sents the closed conformation because SAM increased the high-FRET population (Figure 2 A). SAM concentrations we used in our study are similar to the physiological con- centration in E. coli, ranging from 28 (cid:2)M to 228 (cid:2)M ( 20 ). The fraction of molecules in the high-FRET population vs ligand concentration could be fitted using a simple two- state binding isotherm, yielding a dissociation constant ( K d ) of 10 (cid:2)M, similar to those measured in bulk solution us- ing isothermal calorimetry ( 36 ). The fraction of molecules in the constant high-FRET species increased from 0.16 to 0.43, and this increase appeared to occur at the expense of the dynamic species while the population of the constant mid-FRET species remained unchanged upon ligand bind- ing (Figure 2 B). This suggests that the molecules already in dynamic exchange with the closed conformation were mor e r eadily locked into the closed conformation via lig- and binding, while the constant mid-FRET population may be trapped in a misfolded state that is not easily rescued by ligand binding. Indeed, flow experiments demonstrated that ∼43% of dynamic species (37 of 86) showed clear lock- ing into the closed conformation after addition of 1 mM SAM (Figure 2 C). Notably, a significant fraction (38%) of molecules still exhibited dynamic transitioning e v en when Nucleic Acids Research, 2023, Vol. 51, No. 17 8961 Figure 1. Studies of ligand-free conformations of the SAM / SAH riboswitch by single-molecule FRET. ( A ) A scheme showing the probable folding of SAM / SAH riboswitch RNA. An 18 nt DNA molecule with a 3 (cid:2) Cy5 acceptor (red circle) was attached via its biotinylated 5 (cid:2) terminus to a quartz slide. Cy3 donor (green circle) was attached to the bulged nucleotide in the PK helix of the riboswitch, and an 18 nt 3 (cid:2) DNA extension complementary to the surface-attached DNA allowed the riboswitch to be tethered to the slide. If the pseudoknot helix is not formed the fluorophores should be separated (the open conformation with mid FRET efficiency) whereas in the folded structure the fluorophores should be much closer (the closed conformation with high FRET efficiency). ( B ) Distribution of FRET efficiencies ( E FRET ) for SAM / SAH riboswitch molecules corresponding to the open and closed conformations. ( C ) Char acteristic tr aces of E FRET as a function of time r ecorded. Thr ee r epr esentati v e tr aces are shown, illustr ati v e of constant high FRET (top), constant mid FRET (middle) and dynamic molecules (bottom) undergoing transitions between states of high and middle E FRET . ( D ) Histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey with line). ( E ) Among traces showing dynamic transitioning, two characteristic traces are shown, indicating the kinetics of transitioning is di v erse and heterogenous. (F, G) Such di v erse tr ansitioning kinetics is then fitted into two states tr ansitioning by ebFRET. Indi vidual molecule av erage dwell time is plotted into log- scale scatter plot ( F ), indicating transitioning heterogeneity. And the high FRET and middle FRET probability within the d ynamic popula tion is plotted individually ( G ). excess ligand was added. However, the average high-FRET dwell time became significantly longer upon ligand addition (Figure 2 D). A schematic model is presented in Figure 2 F, which illustra tes tha t in the absence of ligand, the riboswitch is in a dynamic equilibrium between folded and unfolded sta tes, and tha t the addition of a ligand results in a shift to- wards the closed conformation. Structural perturbations provide insights into the folding en- ergy landscape We ne xt e xamined alterations in the conformational land- scape caused by mutations designed to impact the local structural stability. As shown in Figure 1 A and Figure 3 A, the ligand-bound riboswitch adopts H-type pseudoknot structure with three stabilizing features: ( i ) P1x: the exten- sion to helix P1, comprising one W-C base pair and two non-W–C pairs, ( i i) PK: the pseudoknot helix, involving the Shine–Dalgarno sequence and ( i ii) T: a triple base interac- tion (G47:C16–G16) that is part of the PK helix (Figure 3 A). These structural features are abbreviated here as P1x, PK and T, respecti v ely. To perturb the closed conformation, we designed four dif- ferent mutants, named according to the location of muta- tion: P1x C26Z, P1x A14P, PK C18A / G49U / C50U, and T G16P. For P1x mutants, the original base pairing was al- tered by introducing a modified nucleotide: zebularine (Z: cytosine with N4 removed) or purine (P: adenine with N6 r emoved) (Figur e 3 A). For mutation of the PK helix, two original CG base pairings were replaced with weaker pair- ings: AU and GU. For mutation of the base triple T G16P, G16 was replaced by purine, disrupting the interaction with the Hoogsteen edge of G47. In choosing the muta- tion sites, we avoided altering nucleotides that interact di- rectly with the ligand to minimize disruption of the binding site. Additionally, the number of hydrogen bonds removed was kept to a minimum. The positions of these sequence variations can be found in Supplementary Figure S3, and the sequences of the mutants are listed in Supplementary Table S1. All mutants exhibited an increase in the high FRET population with increasing ligand concentrations, show- ing that the mutations did not eliminate the ligand’s abil- ity to stimulate riboswitch folding (Supplementary Figure S4A–D). The fraction of the high FRET state versus lig- and concentration could be fitted using a two-state bind- ing isotherm, yielding K d values (Table 1 ). Mutants that af- fect the PK helix stability (PK and T m utants) greatl y re- duced binding affinity: K d (PK C18A / G49U / C50U) > 1 mM; K d (T G16P) = 607 (cid:2)M. In contrast, muta tions a t the 8962 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 2. Studies of ligand-induced conformations of the SAM / SAH riboswitch by single-molecule FRET. ( A ) Distribution of FRET efficiencies ( E FRET ) as a function of SAM and ( B ) the histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey with line) in the presence of 1 mM SAM. ( C ) Two typical trajectories of riboswitches showed populations converted from dynamics to constant high FRET while SAM was flowed into the reaction chamber at 20 s, corresponding to the change of the relati v e populations in the presence of SAM. ( D, E ) Among molecules remained tr ansitioning, tr ansitioning kinetics was then fitted into two states by ebFRET. Individual molecule average dwell time is plotted into log-scale scatter plot ( D ). And the high FRET and middle FRET probability within the dynamic population is plotted individually ( E ). ( F ) A scheme showing in the presence of ligand, conformations are shifted toward the closed conformations. P1x region had milder effects: K d (P1x C26Z: 41 (cid:2)M; P1x A14P: 61 (cid:2)M). Ne xt, we e xamined the impact of mutations on ligand- free folding dynamics and ligand responsi v eness. In the absence of ligand, all mutants displayed peaks at similar FRET values as the wild type, indica ting tha t open and closed conformations themselves were not significantly al- tered, but their relati v e populations changed (Figure 3 B and C). For example, the destabilization of the closed con- formation in the PK C18A / G49U / C50U mutant led to a near-complete depletion of the closed conformations (Fig- ure 3 D middle panel). The other three mutants showed a more modest decrease in high FRET peak, by < 3% for P1x C26Z, 20% for P1x A14P and 22% for T G16P (Figure 3 B and C). Considering both the binding affinity and the ligand-free conformations, our findings provide evidence for the notion that greater disruptions to the original ligand- free conformations result in greater reduction in binding affinity. Upon ligand introduction, the general behavior observed in the wild-type riboswitch was observed for all mutants, but the relati v e populations and their ligand-induced changes wer e mutation-dependent (Figur e 3 D and Supplementary Figur e S5). Compar ed to mutants targeting the extended P1 stem, mutants that were specifically designed to re- duce the PK helix stability (PK C18A / G49U / C50U and T G16P) exhibited larger changes in conformation, cor- r esponding to r educed binding affinities. As an example, the PK C18A / G49U / C50U mutant had the constant high FRET population almost depleted, replaced by the domi- nant constant mid FRET populations (Figure 3 E). Addi- tionally, in the presence of the ligand, the dynamic popula- tion became more prevalent at the expense of the constant mid FRET population, while the constant high FRET pop- ulation still remained nearly depleted (Figure 3 E). We also tested another mutant P1x A14C / C26U, which stabilizes the extended P1 helix by introducing extra hy- drogen bond (P1x A14C, C26U) by replacing two original non-WC base pairs ( cis sugar-Hoogsteen A13:C26, trans Hoogsteen-sugar A14:G25) with WC base pairs (A13:U26, C14:G25). Despite the introduction of extra hydrogen bonds, the mutants showed a reduction in closed confor- mation (Figure 3 C bottom panel). In addition, the dynamic species increased in population accompanied by a loss of the constant high FRET species (Figure 3 D). The substitution of the WC base pairs may have disturbed the original stack- ing geometry of the three extended P1 base pairs, resulting in less stable PK conformation and affecting the base pair C15:G24 that interacts with the ligand, and thus leading to a reduced binding affinity ( K d = 593 (cid:2)M). These findings highlight the complex nature of riboswitch folding and how small, localized changes can alter the overall folding equi- librium and responsi v eness to ligands, ultimately impacting binding affinity. Ribosome accessibility assay in the absence of ligand We next explored the riboswitch function of blocking trans- la tional initia tion in a ligand dependent manner by mim- (cid:2) region of the riboswitch icking ribosome binding. The 3 Nucleic Acids Research, 2023, Vol. 51, No. 17 8963 Figure 3. Studies of muta tions a t local structures affect conformations and ligand responsi v eness. ( A ) The ligand-bound riboswitch previously re v ealed by X-r ay crystallogr aphy ( 36 ) adopts H-type pseudoknot structure with three stabilizing features: ( i ) P1x: the extension to helix P1, comprising one W-C base pair and two non-W-C pairs, ( i i) PK: the pseudoknot helix, involving the Shine–Dalgarno sequence and ( i ii) a triple base interaction (G47:C16-G16). (B, C) Distribution of FRET efficiencies ( E FRET ) of the ligand-free conformations of, ( B ) wild-type, P1x C26Z and P1x A14P mutants; ( C ) T G16P, PK C18A / G49U / C50U and P1x A14C / C26U m utants. All m utations shared similar folding behaviors of constant high FRET, constant middle FRET, and dynamics. ( D , E ) The histograms showing the relati v e fraction of constant high FRET (black), constant middle FRET (dark gray) and dynamic molecules (light grey) of wild-type and all mutants (D). Among all mutants, PK C18A / G49U / C50U showed the most different populations both in the absence or in the presence of ligands (E). Table 1. Binding affinity of SAM Mutation Wild-typed P1x C26Z P1x A14P T G16P PK C18A, G49U, C50U P1xA14C, C26U Binding affinity ( K d ) in (cid:2)M 11 42 61 607 > 1000 593 contains Shine–Dalgarno (S–D) site to which the ribosome binds to initiate translation. To fully cover the S–D sequence and form stable binding, we used a 9 nt oligonucleotide complementary to the S–D region in assessing the accessi- bility of the transla tion initia tion site (Figure 4 A). Binding of this oligonucleotide was easily monitored using our sin- gle molecule experiment. In the absence of ligands, adding the oligonucleotides de- creased the two major FRET populations ( E FRET = 0.4 & 0.84) and created a new population with an E FRET of a pproximatel y 0.22 (Figure 4 B). To identify a condition that the oligonucleotides are saturated to the pre-exposed S–D region, we conducted experiments with varying con- centrations of oligonucleotides, and no discernible changes in conformation were observed above 500 nM (Supple- mentary Figure S6A). Consequently, all subsequent exper- iments were performed with the sa tura ting oligonucleotide concentration of 500 nM. We attribute the low-FRET state to the lengthening of 9 nt S–D region upon ribos witch-oligon ucleotide complex formation. A similar experiment using a 15 nt oligonu- cleotide showed the low-FRET peak at a slightly lower value ( ∼0.18), consistent with the longer helix that would be formed (Figure 4 C and Supplementary Figure S6B). Duplex formation was very stable and persisted even after w ashing aw ay fr ee oligonucleotides (Supplementary Figur e S6C), in contrast to transient duplex formation observed us- ing a shorter 7 nt long ribosome mimic for 7-aminomethyl- 7-deazaguanine (preQ 1 )-sensing riboswitch ( 47 ). We further examined the oligonucleotide binding reac- tion in real time by flowing in the oligonucleotide dur- ing observation. Most molecules remained unchanged or showed photobleaching because oligo binding generally took longer than our observation window of ∼180 s. Among molecules showing evidence of oligonucleotide binding, d ynamic fluctua tions between 0.4 and 0.84 FRET states were observed before they were locked into the low- FRET state (Figure 4 D). In addition, most low-FRET states (75% and 91% for the 9-nt oligo and 15-nt oligo, respecti v ely) were reached from the mid-FRET, open con- formation (Figure 4 D), indicating that the S–D region be- comes accessible in the open conformation prior to oligonu- cleotide binding. 8964 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 4. Pseudo-functional studies for assessing the accessibility of the translation initiation site. ( A ) A scheme showing 9 or 15-nt complementary oligonu- cleotides bind to the open conformation of the SAM / SAH riboswitch. ( B ) Distribution of E FRET before and after addition of sa tura ted concentra tion ( = 500 nM) with various time points. ( C ) Distribution of E FRET comparing the oligo-free (top), 9-nt (middle) and 15-nt (bottom) oligonucleotide-bound conformations. ( D ) Typical trajectories of riboswitches showed population converted from the mid-FRET to low FRET while oligonucleotides were flowed into the reaction chamber at 12 s, corresponding to the generated conformations observed in distribution of E FRET . ( E ) Distribution of E FRET after si- multaneous addition of 9-nt oligonucleotides ( = 500 nM) and SAM ( = 1 (cid:2)M) at various time points, both additions are a t sa tura ted concentra tion. ( F ) Distribution of E FRET showing the ligand responsi v eness after the riboswitches were pre-bound by oligonucleotides. Ligand-induced conformational change outcompetes ribo- some mimic binding We ne xt e xamined the effect of the SAM ligand on the S– D region accessibility and riboswitch folding by simulta- neously adding the ligand SAM and oligonucleotide ribo- some mimics. Within min, the high-FRET population in- creased at the expense of the mid-FRET population. Only a small fraction of molecules ( < 20%) went to the low-FRET conformation corresponding to the ribosome-mimic bound state (Figure 4 E). Therefore, under our experimental con- ditions (1 mM SAM and 500 nM oligonucleotide), ligand binding outcompetes 9-nt oligonucleotide binding at early timepoints. The riboswitch remained in the high-FRET state for the whole observation window for the 9-nt oligonucleotide, up to 60 min. Howe v er, for the 15-nt oligonucleotides, ∼40% of high-FRET conformation was lost by 10 min and by 1 h, the low-FRET conformation became dominant (Supple- mentary Figure S6D), suggesting that the ligand bound ri- boswitch can still undergo occasional visits to the open con- formation. The additional foothold of the 15-nt oligo en- ables capturing the transiently exposed S–D site. At least for the longer ribosome-mimic, the equilibrium favors the oligo-bound ‘gene-on’ state w hereas earl y in the process, ligand binding can trap the molecule in the closed con- formation, momentarily blocking access to the ribosome. Howe v er, it is also possible that the extra f oothold f or the long oligo may facilitate pseudoknot unwinding by par- tially hybridizing to an unpaired region followed by strand displacement. To test if the oligo-bound riboswitch remains responsi v e to ligand, we added the ligand after pre-incubation with oligonucleotide. Only ∼25% of 9-nt oligo-bound structures converted to the ligand-bound ( E FRET = 0.84 state), and a negligible fraction of the 15-nt oligo-bound riboswitch was responsi v e to the ligand (Figure 4 F and Supplemen- tary Figure S6E). Therefore, to function as a transla- tional riboswitch, the decision must be made before the ar- rival of the ribosome, assuming ribosome binding happens only once. In the more realistic case of multiple ribosome molecules arriving and initiating translation in succession, the riboswitch activity can be gradational. Vectorial folding assay disfavors ligand binding compared to the pr e-f olded riboswitch Because RNA folds co-transcriptionally, riboswitch func- tion should be examined in the context of ongoing tran- scription ( 48 ). Se v eral different methods of mimicking co- transcriptional riboswitch folding and function are avail- able ( 44 , 45 , 49–55 ). Here we used the vectorial folding assay w here a DN A helicase is used to mimic co-transcriptional RNA folding ( 44 , 45 ). The riboswitch was hybridized with a complementary DNA oligonucleotide to form an RNA- (cid:2) overhang at the DNA ter- DNA heteroduplex with a 3 minus. The same Cy3–Cy5 FRET pair as in our pre- vious experiments was used to determine the riboswitch Nucleic Acids Research, 2023, Vol. 51, No. 17 8965 Figur e 5. Vectoriall y f olded assa ys f or mimicking co-transcriptional f olding. ( A ) A scheme showing the engineered superhelicase Rep-X was preincubated and initia ted a t designa ted time for unwinding the heteroduplex. The riboswitch was hybridized with a complementary DNA oligonucleotide to form an RN A-DN A heteroduplex with a 3 (cid:2) overhang at the DNA terminus. The same Cy3–Cy5 FRET pair was used to determine the riboswitch conformations. ( B ) Distribution of E FRET before introducing ATP, corresponding to the heteroduplex conformation. ( C ) Distribution of E FRET after vectorially folding, cor- responding to the conformations released from the heteroduplex. ( D ) A typical trajectory showing a heteroduplex is unwound and folded into the constant high FRET conformation, where the ATP is flowed in at 12 s. ( E ) Histograms of relati v e populations in the presence of various SAM concentrations. conformation (Figure 5 A). A highly processi v e, engineered DNA helicase, Rep-X ( 46 ), was used to unwind the het- eroduplex unidirectionally by translocating on the DNA (cid:2) direction, to release the RNA strand (cid:2) to 5 strand in the 3 (cid:2) direction of transcription and at (cid:2) to 3 progressi v ely in the 5 the speed of transcription, about ∼60 nt per second ( 44 , 45 ). Upon Rep-X addition without ATP, we observed low FRET efficiency ( E FRET = 0.2) because the fluorophores remain separated by the heteroduplex (Figure 5 B). After ATP addition, two new populations centered at 0.4 and 0.84 emerged, corresponding to the open and closed con- formations, respecti v ely (Figure 5 C). A representati v e v ec- torial folding trace shows two features (Figure 5 D). First, Cy3 intensity shows a transient increase due to protein- induced fluorescence enhancement ( 56 , 57 ), signifying Rep- X approaching the Cy3 fluorophore on the RNA strand. Second, the heteroduplex unwinds and riboswitch fold- ing begins. We classified the riboswitch folding behavior into four types (Supplementary Figure S7A) : (i) molecules that transitioned from the heteroduplex state to the stable closed conformation without any detectable intermediate, (ii) molecules that transitioned from the heteroduplex state to the stable open conformation, (iii) molecules that transi- tioned from the heteroduplex state to one undergoing fluc- tuations between the open and closed conformations and (iv) molecules that transitioned from the heteroduplex to the stable closed conformation after first fluctuating be- tween open and closed states. These distinct populations are in agreement with the observed conformations for pre- folded riboswitches except for the type (iv), likely because this behavior is observable only on the path to reach fold- ing equilibrium. Addition of ligand during vectorial folding changed the relati v e populations of the four types of folding behavior. Type I, direct transition to stable high-FRET state, became more popula ted a t higher ligand concentra tions (Figure 5 E) and its fraction vs ligand concentration could be well fit- ted using a two-state binding isotherm (Supplementary Fig- ure S7B), yielding an apparent K d value of 108 (cid:2)M. This apparent K d is an order of magnitude higher than the K d value of 10 (cid:2)M we observed for pre-folded RNA, suggest- ing ligand-responsi v e conforma tion is not immedia tely ob- tained during co-transcriptional folding. We hypothesized that the increase in K d is due to insufficient time for the nascent riboswitch to reach the stead y-sta te conforma tions. Indeed, as shown in Supplementary Figure S7C, the con- formational analyses of time points after vectorial folding e xhibited noticeab le differences. Specifically, the conforma- tions observed at 5 seconds post-folding displayed a lower fraction of high-FRET population and decreased respon- si v eness to ligand. These results imply that the nascent ri- boswitch necessitates more than a few seconds to attain a 8966 Nucleic Acids Research, 2023, Vol. 51, No. 17 Figure 6. Pseudo-functional studies for assessing the accessibility of the translation initiation site during vectorially folding. ( A ) A scheme showing SAM, oligonucleotides, and ATP are flowed in sim ultaneousl y for simulating competition over mutually e xclusi v e conformations during co-transcriptional fold- ing. The oligonucleotide-bound state is termed ‘ON’ sta te, indica ting transla tion can be initia ted, whereas the ligand-bound sta te is termed ‘OFF’ state, indicating the Shine–Dalgarno site is blocked. ( B ) Distribution of E FRET for pre-folded riboswitches under simultaneous addition of 9-nt oligonucleotides and various concentrations of SAM. Similar competitions between oligonucleotides and ligands were carried out while the riboswitches are vectorially folded and distribution of E FRET with various SAM concentrations is shown in ( D ). Similar competition experiments were carried out with longer 15-nt oligonucleotides, f or pre-f olded competition, ( C ) distribution of E FRET with various SAM concentrations; for vectorially folded competition ( E ) distribu- tion of E FRET with various SAM concentrations. stead y sta te conforma tional distribution, ther eby r educing its ligand-binding affinity during transcription. Vectorial folding favors ribosome mimic binding In the ribosome accessibility assay on pre-folded ri- boswitches, we found that ligand binding is kinetically fa- vored over ribosome mimic binding. To test if this result holds for vectorial folding, we included sa tura ting concen- tration of-9 nt or 15-nt oligonucleotides during vectorial folding (Figure 6 A). In the absence of ligand, the closed conformation was rarely observed, likely because the S–D r egion r e v ealed thr ough heter oduplex unwinding is bound by the ribosome mimic before the aptamer can fold into the closed form (top histogram of Figure 6 D and E). This find- ing is also in line with our observa tion tha t it takes min for the nascent riboswitch to attain a stead y-sta te confor- mational distribution (as demonstrated in Supplementary Figure S7C). In the presence of ligand, the closed conformation was obtained in a ligand-concentration dependent manner. The efficacy of the ligand in converting the riboswitch to its closed conforma tion sta te was diminished in the vectorial folding condition compared to the pre-folded condition for both ribosome mimics (Figure 6 B–E and Supplementary Figure S9). The nascent riboswitch likely first adopts the open conformation, which facilitates ribosome mimic bind- ing, and reduces the m utuall y e xclusi v e ligand bound con- formation. DISCUSSION We propose a model describing the folding scheme and its energy landscape based on our findings of multiple populations of static folds, open or closed and dynamic switching, and the highly heterogenous switching rates. The FRET values (0.4 and 0.84) of the switching molecules wer e indiffer entiable from those with sta tic conforma tions, suggesting there is significant structural resemblance. We were surprised that the majority ( ∼55%) of ligand-free ri- boswitches showed dynamic switching between the closed and open conformations. Most of the dynamically switch- ing molecules are responsi v e to ligand, either by population conversion to the static closed conformation or rate alter- ations. The observed heterogeneity is likely to be a prop- erty inherent to the riboswitch because our constructs with their modifications for surface tethering and fluorescence imaging showed comparable binding affinity to what was determined from unmodified RNA in bulk solution. Fur- thermore, all fiv e single-site mutants we tested show simi- larl y hetero geneous behavior with onl y their relati v e popu- lations and kinetics changed. Mutants examined in this study showed that not only the populations of static and dynamic populations were strongly affected, but the rates of switching between con- formations changed (Supplementary Figure S8). We specu- la te tha t an y incomplete base pair f ormation of the extended P1 stem (E-P1) or PK helix may introduce metastable conformations, leading to heterogeneous folding / unfolding rates for this riboswitch, and potentially for other func- tional RNAs that also contain the H-type pseudoknot ( 58–61 ). Such heterogeneity, if present in vivo , may buffer the riboswitch activity against a wide range of ligand concentrations. Our findings are most consistent with the previously pro- posed hybrid model combining conformation selection & induced-fit ( 10 ): whereas all conformations are sampled in the absence of ligand (conformation selection), ligand addi- tion repopulates the population ensemble by imparting fur- ther stability to the ligand-bound state (induced-fit). A pre- vious SAM-II riboswitch study reported that transient con- formational excursions occur in the absence of ligand, sug- gesting conformational sampling ( 10 ). Howe v er, they could not determine if those transient conformations were respon- si v e to ligands or how folding and ligand binding are pro- moted through specific structural motifs. Relevant to our evaluation of the riboswitch’s accessibil- ity for ribosome mimics, a previous study probed the folding of the 7-aminomethyl-7-deazaguanine-sensing riboswitch using a 7 nt long fluorescently labeled oligonucleotides as transla tional initia tion mimic ( 46 ). They observed bursts of probe binding and showed that ligand addition reduces burst duration and extends the intervals between bursts. Howe v er, the use of fluorescently labeled probes limited their analysis to sub- K d concentrations. By employing un- labeled oligonucleotides, we were able to mimic translation initiation under conditions of saturating ribosome mimic so that the exposure of the binding site is rate-limiting and show that the nascent folds adopted have yet to reach an equilibrium, thus leading to a reduced ligand binding affinity. In the vectorial f olding assa y, we observed a decrease in ligand binding affinity (Figure 6 and Supplementary Fig- ur e S9), r esulting in a r eduction in the effecti v eness of lig- and binding when competing with a ribosome mimic. These differ ences between pr e-f olded and vectorially f olded ri- boswitches suggest that the timing of regulatory decision is critical to the effecti v eness of the riboswitch and may ex- plain the r equir ement f or higher ligand concentration f or ef- fecti v e regulatory control in vivo ( 62 ) . It is possible that there ar e differ ent modes of r egulating accessibility, and the tim- ing of transcription and translation coupling. For tighter regulation, riboswitch needs to reach equilibrium first, thus transcription needs to be carried out in advance of trans- lation. Howe v er, when regulation needs not to be tight, the transcription and translation can happen simultaneously. In conclusion, our studies on this small SAM / SAH ri- boswitch provide valuable insights into the complexities of the folding landscape, including individual folding hetero- geneity and the role of RNA folding kinetics. Furthermore, our findings have implications for the translational control governed by the riboswitch, highlighting the critical influ- ence of folding equilibrium on the efficacy of regulatory decisions. DA T A A V AILABILITY Analyses and data acquisition codes are upload on lab GitHub account and archi v ed in Zenodo with the following doi. Additionally, raw data that support our findings have Nucleic Acids Research, 2023, Vol. 51, No. 17 8967 been uploaded and archi v ed in Zenodo, corresponding to each individual figure. GitHub: https://github.com/Ha-SingleMoleculeLab Analyses , data acquisition codes , and raw data are archi v ed in Zenodo: Raw data analysis DOI: 10.5281 / zenodo.4925617 Data acquisition DOI: 10.5281 / zenodo.4925630 Raw data DOI: 10.5281 / zenodo.8088172 SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We thank Prof. Hui-Ting Lee, Dr Olivia Yang, Dr Boyang Hua, and the members of the Ha laboratory and Sua My- ong laboratory for their input and support. All the authors w ould lik e to expr ess their gratitude to the funding sour ce for their generous support. FUNDING US National Institutes of Health [R35 GM 122569 to T.H. and F32 GM 139268 to L.R.G.]; Cancer Research UK [Progr am gr ant A18604]; EPSRC [EP / X01567X / 1 to D.M.J.L.]; National Natural Science Foundation of China [32171191 to L.H.]; Guangdong Science and Technol- ogy Department [2022A1515010328, 2020B1212060018, 2020B1212030004 to L.H.]; T.H. is an investigator of the Howard Hughes Medical Institute. Funding for open ac- cess charge: U.S. Department of Health and Human Ser- vices [R35 GM 122569]. Conflict of interest statement. None declared. REFERENCES 1. Mccown,P.J., Corbino,K.A., Stav,S., Sherlock,M.E. and Breaker,R.R. (2017) Riboswitch di v ersity and distribution. RNA , 23 , 995–1011. 2. Serganov,A. and Nudler,E. (2013) A decade of riboswitches. Cell , 152 , 17–24. 3. Sherwood,A.V. and Henkin,T.M. (2016) Riboswitch-mediated gene regulation: novel RNA architectures dictate gene expression responses. Annu. Rev. Microbiol. , 70 , 361–374. 4. Speed,M.C., Burkhart,B.W., Picking,J.W. and Santangelo,J. (2018) An Archaeal Fluoride-Responsi v e Riboswitch Provides an Inducible Expression System for Hyperthermophiles. Appl. Environ. Microbiol. , 84 , e02306-17. 5. Wachter,A., Tunc-Ozdemir,M., Grove,B.C., Green,P.J., Shintani,D.K. and Breaker,R.R. (2007) Riboswitch control of gene expression in plants by splicing and alternative 3 (cid:2) end processing of mRNAs. Plant Cell , 19 , 3437–3450. 6. Moldo van,M.A., Petro va,S.A. and Gelfand,M.S. (2018) Comparati v e genomic analysis of fungal TPP-riboswitches. Fungal Genet. Biol. , 114 , 34–41. 7. Li,S. and Breaker,R.R. (2013) Eukaryotic TPP riboswitch regulation of alternati v e splicing involving long-distance base pairing. Nucleic Acids Res. , 41 , 3022–3031. 8. Andreasson,J.O.L., Savinov,A., Block,S.M. and Greenleaf,W.J. (2020) Comprehensi v e sequence-to-function mapping of cofactor-dependent RN A catal ysis in the glmS ribozyme. Nat. Commun. , 11 , 1663. 9. Valeri,J.A., Collins,K.M., Ramesh,P., Alcantar,M.A., Lepe,B.A., Lu,T.K. and Camacho,D.M. (2020) Sequence-to-function deep learning frame wor ks for engineer ed ribor egulators. Nat. Commun. , 11 , 5058. 8968 Nucleic Acids Research, 2023, Vol. 51, No. 17 10. Haller,A., Rieder,U., Aigner,M., Blanchard,S.C. and Micura,R. (2011) Conformational capture of the SAM-II riboswitch. Nat. Chem. Biol. , 7 , 393–400. 11. Savinov,A., Perez,C.F. and Block,S.M. (2014) Single-molecule studies of riboswitch folding. Biochim. Biophys. Acta - Gene Regul. Mech. , 1839 , 1030–1045. 12. Holmstrom,E.D., Polaski,J.T., Batey,R.T. and Nesbitt,D.J. (2014) Single-molecule conformational dynamics of a biolo gicall y functional hydro x ocobalamin riboswitch. J. Am. Chem. Soc. , 136 , 16832–16843. 13. Panchal,V. and Brenk,R. (2021) Riboswitches as drug targets for antibiotics. Antibiotics , 10 , 45. 14. Hallberg,Z.F., Su,Y., Kitto,R.Z. and Hammond,M.C. (2017) Engineering and in vivo applications of riboswitches. Annu. Rev. Biochem. , 86 , 515–539. 15. Loenen,W.A.M. (2006) S-Adenosylmethionine: jack of all trades and master of e v erything? Biochem. Soc. T r ans . , 34 , 330–333. 16. Kredich,N.M. and Hershfield,M.S. (1979) S-adenosylhomocysteine toxicity in normal and adenosine kinase-deficient lymphoblasts of human origin. Proc. Natl. Acad. Sci. U.S. A. , 76 , 2450–2454. 17. Schubert,H.L., Blumenthal,R.M. and Cheng,X. (2003) Many paths to methyltr ansfer : a chronicle of convergence. T r ends Biochem. Sci. , 28 , 329–335. 18. Roje,S. (2006) S-Adenosyl-l-methionine: beyond the uni v ersal methyl group donor. Phytoc hemistr y , 67 , 1686–1698. 19. Cantoni,G.L. (1975) Biological methylation: selected aspects. Annu. Rev. Biochem. , 44 , 435–451. 20. Posnick,L.M. and Samson,L.D. (1999) Influence of S-adenosylmethionine pool size on spontaneous mutation, dam methylation, and cell growth of Escherichia coli. J. Bacteriol. , 181 , 6756–6762. 21. Price,I.R., Grigg,J.C. and Ke,A. (2014) Common themes and differences in SAM recognition among SAM riboswitches. Biochim. Biophys. Acta - Gene Regul. Mech. , 1839 , 931–938. 34. Gilbert,S.D., Rambo,R.P., Van Tyne,D. and Batey,R.T. (2008) Structure of the SAM-II riboswitch bound to S-adenosylmethionine. Nat. Struct. Mol. Biol. , 15 , 177–182. 35. Weickhmann,A.K., Keller,H., Wurm,J.P., Strebitzer,E., Juen,M.A., Kremser,J., Weinberg,Z., Kreutz,C., Duchardt-Ferner,E. and W ¨ohnert,J. (2019) The structure of the SAM / SAH-binding riboswitch. Nucleic Acids Res. , 47 , 2654–2665. 36. Huang,L., Liao,T.W., Wang,J., Ha,T. and Lilley,D.M.J. (2020) Crystal structure and ligand-induced folding of the SAM / SAH riboswitch. Nucleic Acids Res. , 48 , 7545–7556. 37. Ha,T ., Enderle,T ., Ogletree,D .F., Chemla,D .S., Selvin,P.R. and Weiss,S. (1996) Probing the interaction between two single molecules: fluor escence r esonance energy transfer between a single donor and a single acceptor. Proc. Natl. Acad. Sci. U.S.A. , 93 , 6264–6268. 38. Beaucage,S.L. and Caruthers,M.H. (1981) Deoxynucleoside phosphoramidites-A new class of key intermediates for deoxypolynucleotide synthesis. Tetr ahedr on Lett. , 22 , 1859–1862. 39. Wilson,T.J., Zhao,Z.Y., Maxwell,K., Kontogiannis,L. and Lilley,D.M.J. (2001) Importance of specific nucleotides in the folding of the natural form of the hairpin ribozyme. Bioc hemistr y , 40 , 2291–2302. 40. Hakimelahi,G.H., Proba,Z.A. and Ogilvie,K.K. (1981) High yield selecti v e 3 (cid:2) -silylation of ribonucleosides. Tetrahedron Lett. , 22 , 5243–5246. 41. Perreault,J., Wutt,T., Cousineau,B., Ogilviett,K.K. and Cedergren,R. (1990) Mixed deoxyribo- and ribo-oligonucleotides with catalytic activity. Nature , 344 , 1988–1990. 42. Roy,R., Hohng,S. and Ha,T. (2008) A practical guide to single-molecule FRET. Nat. Methods , 5 , 507–516. 43. Van De Meent,J.W., Bronson,J.E., Wiggins,C.H. and Gonzalez,R.L. (2014) Empirical bayes methods enable advanced population-level analyses of single-molecule FRET experiments. Biophys. J. , 106 , 1327–1337. 22. Huang,L. and Lilley,D.M.J. (2018) Structure and ligand binding of 44. Hua,B., Panja,S., Wang,Y., Woodson,S.A. and Ha,T. (2018) the SAM-V riboswitch. Nucleic Acids Res. , 46 , 6869–6879. 23. Fuchs,R.T., Grundy,F.J. and Henkin,T.M. (2006) The SMK box is a Mimicking co-transcriptional RNA folding using a superhelicase. J. Am. Chem. Soc. , 140 , 10067–10070. new SAM-binding RNA for translational regulation of SAM synthetase. Nat. Struct. Mol. Biol. , 13 , 226–233. 24. Lu,C., Smith,A.M., Fuchs,R.T., Ding,F., Rajashankar,K., Henkin,T.M. and Ke,A. (2008) Crystal structures of the SAM-III / SMK riboswitch re v eal the SAM-dependent translation inhibition mechanism. Nat. Struct. Mol. Biol. , 15 , 1076–1083. 25. Sun,A., Gasser,C., Li,F., Chen,H., Mair,S., Krasheninina,O., Micura,R. and Ren,A. (2019) SAM-VI riboswitch structure and signature for ligand discrimination. Nat. Commun. , 10 , 5728. 26. Wang,J.X., Lee,E.R., Morales,D.R., Lim,J. and Breaker,R.R. (2008) Riboswitches that sense S-adenosylhomocysteine and activate genes involved in coenzyme recycling. Mol. Cell , 29 , 691–702. 27. Montange,R.K. and Batey,R.T. (2006) Structure of the S-adenosylmethionine riboswitch regulatory mRNA element. Nature , 441 , 1172–1175. 28. Mirihana Arachchilage,G., Sherlock,M.E., Weinberg,Z. and Breaker,R.R. (2018) SAM-VI RNAs selecti v ely bind S-adenosylmethionine and exhibit similarities to SAM-III riboswitches. RNA Biol , 15 , 371–378. 29. Winkler,W.C., Nahvi,A., Sudarsan,N., Barrick,J.E. and Breaker,R.R. (2003) An mRNA structure that controls gene expression by binding S-adenosylmethionine. Nat. Struct. Biol. , 10 , 701–707. 30. Weinberg,Z., Regulski,E.E., Hammond,M.C., Barrick,J.E., Yao,Z., Ruzzo,W.L. and Breaker,R.R. (2008) The aptamer core of SAM-IV riboswitches mimics the ligand-binding site of SAM-I riboswitches. RNA , 14 , 822–828. 31. Weinberg,Z., Wang,J.X., Bogue,J., Yang,J., Corbino,K., Moy,R.H. and Breaker,R.R. (2010) Comparati v e genomics re v eals 104 candidate structured RNAs from bacteria, archaea, and their metagenomes. Genome Biol. , 11 , R31. 32. Corbino,K.A., Barrick,J.E., Lim,J., Welz,R., Tucker,B.J., Puskarz,I., Mandal,M., Rudnick,N.D. and Breaker,R.R. (2005) Evidence for a second class of S-adenosylmethionine riboswitches and other regulatory RNA motifs in alpha-proteobacteria. Genome Biol. , 6 , R70. 45. Hua,B., Jones,C.P., Mitra,J., Murray,P.J., Rosenthal,R., Ferr ´e-D’Amar ´e,A.R. and Ha,T. (2020) Real-time monitoring of single ZTP riboswitches re v eals a complex and kinetically controlled decision landscape. Nat. Commun. , 11 , 4531. 46. Arslan,S., Khafizov,R., Thomas,C.D., Chemla,Y.R. and Ha,T. (2015) Engineering of a superhelicase through conformational control. Science , 348 , 344–347. 47. Rinaldi,A.J., Lund,P.E., Blanco,M.R. and Walter,N.G. (2016) The Shine–Dalgarno sequence of riboswitch-regulated single mRNAs shows ligand-dependent accessibility bursts. Nat. Commun. , 7 , 8976. 48. Gong,S., Wang,Y., Wang,Z. and Zhang,W. (2017) Co-transcriptional folding and regulation mechanisms of riboswitches. Molecules , 22 , 1–14. 49. Duss,O., Stepanyuk,G.A., Puglisi,J.D. and Williamson,J.R. (2019) Transient protein-RNA interactions guide nascent ribosomal RNA folding. Cell , 179 , 1357–1369.e16. 50. Rodgers,M.L. and Woodson,S.A. (2019) Transcription increases the cooperativity of ribonucleoprotein assembly. Cell , 179 , 1370–1381. 51. Juette,M.F., Terry,D.S., Wasserman,M.R., Altman,R.B., Zhou,Z., Zhao,H. and Blanchard,S.C. (2016) Single-molecule imaging of non-equilibrium molecular ensembles on the millisecond timescale. Nat. Methods , 13 , 341–344. 52. Uhm,H., Kang,W., Ha,K.S., Kang,C. and Hohng,S. (2017) Single-molecule FRET studies on the cotranscriptional folding of a thiamine pyrophosphate riboswitch. Proc. Natl. Acad. Sci. U.S.A. , 115 , 331–336. 53. Yu,A.M., Gasper,P.M., Cheng,L., Lai,L.B., Kaur,S., Gopalan,V., Chen,A.A. and Lucks,J.B. (2021) Computationally reconstructing cotranscriptional RNA folding from experimental data reveals rearrangement of non-nati v e folding intermediates. Mol. Cell , 81 , 870–883. 54. Strobel,E.J., Watters,K.E., Nedialkov,Y., Artsimovitch,I. and Lucks,J.B. (2017) Distributed biotin-streptavidin transcription roadblocks for mapping cotranscriptional RNA folding. Nucleic Acids Res. , 45 , e109. 33. Poiata,E., Meyer,M.M., Ames,T.D. and Breaker,R.R. (2009) A 55. Chauvier,A., St-Pierre,P., Nadon,J.F., Hien,E.D.M., variant riboswitch aptamer class for S-adenosylmethionine common in marine bacteria. RNA , 15 , 2046–2056. Perez-Gonzalez,C., Eschbach,S.H., Lamontagne,A.M., Carlos Penedo,J. and Lafontaine,D.A. (2021) Monitoring RNA dynamics in Nucleic Acids Research, 2023, Vol. 51, No. 17 8969 nati v e transcriptional complexes. Proc. Natl. Acad. Sci. U.S.A. , 118 , e2106564118. 59. Staple,D.W. and Butcher,S.E. (2005) Pseudoknots: RNA structures with di v erse functions. PLoS Biol. , 3 , 0956–0959. 56. Pettigrew,N.R., Thomas,A.C., Mcmanus,M.A., Paduan,J.D., 60. Shen,L.X. and Tinoco,I. (1995) The structure of an RNA Chavez,F.P., Nielsen,T.G., Desiderio,R.A., Carr,M., Osborn,T.R., Abe,T. et al. (2009) Cytosolic viral sensor RIG-I is a 5 (cid:2) -triphosphate-dependent translocase on double-stranded RNA. Science , 323 , 1070–1074. 57. Hwang,H., Kim,H. and Myong,S. (2011) Protein induced fluorescence enhancement as a single molecule assay with short distance sensitivity. Proc. Natl. Acad. Sci. U.S.A. , 108 , 7414–7418. 58. Egli,M., Minasov,G., Su,L. and Rich,A. (2002) Metal ions and flexibility in a viral RNA pseudoknot at atomic resolution. Proc. Natl. Acad. Sci. U.S.A. , 99 , 4302–4307. pseudoknot that causes efficient frameshifting in mouse mammary tumor virus. J. Mol. Biol. , 247 , 963–978. 61. Wang,Y., Yesselman,J.D., Zhang,Q., Kang,M. and Feigon,J. (2016) Structural conservation in the template / pseudoknot domain of vertebr ate telomer ase RNA from teleost fish to human. Proc. Natl. Acad. Sci. U.S.A. , 113 , E5125–E5134. 62. Neuner,E., Frener,M., Lusser,A. and Micura,R. (2018) Superior cellular activities of azido- over amino-functionalized ligands for engineer ed pr eQ1 riboswitches in E.coli. RNA Biol. , 15 , 1376–1383. C (cid:3) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
10.1126_sciadv.aat9488	M A T E R I A L S C I E N C E Nanoscale strain engineering of giant pseudo-magnetic fields, valley polarization, and topological channels in graphene C.-C. Hsu1, M. L. Teague1, J.-Q. Wang1, N.-C. Yeh1,2† The existence of nontrivial Berry phases associated with two inequivalent valleys in graphene provides interesting opportunities for investigating the valley-projected topological states. Examples of such studies include observation of anomalous quantum Hall effect in monolayer graphene, demonstration of topological zero modes in “molecular graphene” assembled by scanning tunneling microscopy, and detection of topological valley transport either in graphene superlattices or at bilayer graphene domain walls. However, all aforementioned experiments involved nonscalable approaches of either mechanically exfoliated flakes or atom-by-atom constructions. Here, we report an approach to manipulating the topological states in monolayer graphene via nanoscale strain engineering at room temperature. By placing strain-free monolayer graphene on architected nanostructures to induce global inversion symmetry breaking, we demonstrate the development of giant pseudo-magnetic fields (up to ~800 T), valley polarization, and periodic one-dimensional topological channels for protected propagation of chiral modes in strained graphene, thus paving a pathway toward scalable graphene-based valleytronics. Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works. Distributed under a Creative Commons Attribution NonCommercial License 4.0 (CC BY-NC). INTRODUCTION It has been well recognized that the Berry curvature of electronic wave functions can have a profound effect on the physical properties of materials (1–3). For instance, nontrivial Berry curvatures in the event of either broken time-reversal symmetry or broken inversion symmetry are known to be responsible for various (quantum, anom- alous, spin, and valley) Hall effects (3–6). In the case of graphene with gapless Dirac cones at the two inequivalent valleys K and K′, the spinor-type wave functions of the Dirac fermions result in nontrivial Berry phases of  and − (7–9). For perfect and flat monolayer graphene in the absence of an external magnetic field, the Berry flux from K and K′ exactly cancels each other so that the Hall conduc- tance vanishes under both time and inversion symmetries (7–9). On the other hand, inversion symmetry can be broken by either atomi- cally aligning monolayer graphene on top of hexagonal boron nitride (h-BN) (10) or artificially building in the broken inversion symmetry in strained “molecular graphene” assembled by scanning tunneling microscopy (STM) (11). The former leads to the realization of the valley Hall effect (VHE) (10), and the latter manifests Landau quan- tization and site-dependent topological zero modes in the tunneling conductance spectra (11). In addition, one-dimensional (1D) valley- polarized conducting channels associated with the protected chiral edge states of quantum valley Hall insulators have been demonstrated at the domain walls between AB- and BA-stacked bilayer graphene (12). These interesting results underscore the rich opportunities pro- vided by graphene-based systems for the studies of valley-projected topological states. In contrast, while monolayer transition-metal dichalcogenides (TMDCs) in the 2H-phase are 2D semiconducting crystals with an in-plane structure similar to graphene and exhibit strong spin-valley coupling and interesting optical properties (13–16), no discernible 1Department of Physics, California Institute of Technology, Pasadena, CA 91125, USA. 2Kavli Nanoscience Institute, California Institute of Technology, Pasadena, CA 91125, USA. These authors contributed equally to this work. †Corresponding author. Email: [email protected] out-of-the-plane pseudo-magnetic fields can be induced by strain due to negligible Berry curvatures around the gapped bands at the K and K′ points (17). Thus, the primary strain-induced effect on 2D TMDCs is only associated with the modification to the semiconducting en- ergy gaps, leading to brighter photoluminescence in stronger strained areas due to the resulting smaller energy gaps (17, 18). Earlier experimental investigations of strained graphene gener- ally involve approaches of either stacking mechanically exfoliated, microscale flakes of graphene on h-BN (10) or assembling atom by atom using a scanning tunneling microscope (11). Neither method is scalable for realistic device applications. Recently, efforts have been made to pre-design the substrate to induce controlled strain on mono- layer graphene (19–24). However, most of these strained graphene samples have only been characterized by imaging with scanning electron microscopy (SEM) and/or atomic force microscopy (AFM) and by Raman spectroscopy without explicit investigation of the strain-induced pseudo-magnetic field (19–23). In the only case of direct measurements of the pseudo-magnetic fields by means of STM/ scanning tunneling spectroscopy (STS), studies were solely carried out in nearly flat areas with much smoother topography so that the re- sulting pseudo-magnetic fields were very small (~7 T) and were taken at low temperatures (~4.6 K) (24). Here, we report a scalable exper- imental approach that successfully induces giant pseudo-magnetic fields (up to ~800 T) and achieves manipulation of the topological states and in monolayer graphene by means of nanoscale strain en- gineering at room temperature. Our methodology involves placing a nearly strain-free, large-area (~1 cm2) monolayer graphene sheet on top of properly architected nanostructures to induce global in- version symmetry breaking. We demonstrate the development of strain-induced giant pseudo-magnetic fields and global valley po- larization by direct STM/STS studies. The experimental investiga- tions are further corroborated by simulations using the molecular dynamics (MD) method elaborated in the Supplementary Materials and figs. S1 to S3. We also verify the feasibility of periodic 1D topo- logical channels for protected propagation of chiral modes in graphene based on our empirically demonstrated periodic strain patterns and MD simulations. This methodology is shown to be scalable and 1 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLEcontrollable, which paves a new pathway toward realizing realistic graphene-based valleytronic applications. RESULTS Nanoscale strain engineering of graphene Our experimental approach is based on the notion that the electronic properties of graphene exhibit significant dependence on the nanoscale structural distortion and the resulting strain (7, 25–31). In general, structural distortion-induced strain in the graphene lattice gives rise to two primary effects on the Dirac fermions (13, 15). One is an ef- fective scalar potential , and the other is an effective gauge poten- tial A related to the Berry connection in the reciprocal space (25, 27). These strain-induced effective scalar and gauge potentials in graphene are the consequences of the changes in the distance or angle between the pz orbitals due to structural distortions, which modify the hop- ping energies between Dirac electrons at different lattice sites, there- by giving rise to the addition of an effective gauge potential A and a scalar potential  to the original Dirac Hamiltonian of ideal mono- layer graphene. The excess scalar potential can cause scattering of Dirac fermions and changes in the local charge densities, whereas the excess gauge potential can result in a pseudo-magnetic field that is related to the Berry curvatures in the reciprocal space and couples with the valley (pseudo-spin) degrees of freedom. Thus, proper na- noscale engineering of the strain in graphene can provide unique means to manipulate the valley degrees of freedom (29). Moreover, by combining large-scale strain-free graphene and modern nano- fabrication technology, it becomes feasible to devise scalable structures to achieve desirable control of the valley-associated topological states. To achieve large-area (~1 cm2) nearly strain-free graphene, we used the plasma-enhanced chemical vapor deposition (PECVD) technique, as detailed previously (32). To induce controlled nanoscale strain on initially strain-free graphene, we carried out the procedures schematically illustrated in Fig. 1A. Specifically, two different ap- proaches were taken to induce substantial strain. The first approach involved fabricating Pd nanocrystals (NCs) that were in the form of tetrahedrons with a typical base of ~30 nm in length, using a wet chemical method first developed by Zhang et al. (33) and further elaborated in Materials and Methods. The second approach involved fabricating periodic arrays of nanostructures on silicon substrates using either electron beam (e-beam) lithography or focused ion beam (FIB) lithography, as detailed in Materials and Methods and further illustrated in fig. S4. The NCs were dispersed on a silicon substrate, as exemplified by the SEM images over an area of 3 m × 3 m shown in Fig. 1B and for a zoom-in view of two NCs shown in Fig. 1C. Next, we transferred a monolayer of h-BN and then a monolayer of our PECVD-grown monolayer graphene on top of the NCs by means of a polymer-free technique (34), where the h-BN was mechanically exfoliated either from an h-BN single crystal or by the CVD growth method developed by Lin et al. (35). Both the h-BN and graphene monolayers were shown to conform well to separated NCs, as illus- trated in Fig. 1D and the upper panel in Fig. 1E. However, for closely spaced NCs, a wrinkle in graphene appeared, as exemplified in the lower panel of Fig. 1E. Here, we note that the purpose of inserting an h-BN monolayer between graphene and silicon nanostructures is to minimize the effects of phonon coupling between Si and graphene (28) and of charge impurities in the Si substrate (28) on graphene. The structural distortion-induced strain on graphene can be evaluated using the strain tensors. For a spatially varying displace- ment field u ≡ u x ˆ x + u y ˆ y + h ˆ z , the strain tensor components uxx, uxy, and uyy are given by (25, 27) u xx = ∂ u x ─ ∂ x + 1 ─ 2 ( ∂ h ─ ) ∂ x 2 , u xy = 1 ─ 2 ( ∂ u x ─ ∂ y + ∂ u y ─ ) ∂ x u yy = ∂ u y ─ ∂ y + 1 ─ 2 ( 2 ∂ h ─ ) ∂ y + 1 ─ 2 ( ∂ h ─ ∂ x ∂ h ─ ) ∂ y , (1) If the x axis is chosen along the zigzag direction, the 2D strain- induced gauge potential in the real-space A = A x ˆ x + A y ˆ y can be expressed in terms of the tensor components by the following rela- tions (in the first-order approximation) (7)    ─  ─ A x = ± a 0 u xy a 0 ( u xx − u yy ) , A y = ∓ 2 (2) Fig. 1. Nanoscale strain engineering of graphene. (A) Schematic illustrations showing the steps taken to induce strain on graphene by Pd tetrahedron NCs. (B) SEM image of randomly distributed Pd tetrahedron NCs distributed on a Si substrate over an area of 3 m × 3 m. (C) A zoom-in SEM image of two Pd tetrahedron NCs. (D) Exemplifying AFM image of graphene/h-BN/Pd tetrahedron NCs. (E) Top: AFM image of graphene/BN on a single Pd tetrahedron NC, showing excellent conformation of graphene/BN to the single Pd tetrahedron NC. Bottom: AFM image of graphene/BN on two closely spaced Pd tetrahedron NCs, showing the formation of a graphene wrinkle between the two Pd tetrahedron NCs. 2 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLE where a0 ≈ 0.142 nm is the nearest carbon-carbon distance for equi- librium graphene,  is a constant ranging from 2 to 3 in units of the flux quantum, and the upper and lower signs are associated with the K and K′ valleys, respectively (7). From Eq. 1, we note that in the event of strong z-axis corrugation, the strain components resulting from the height variations becomes dominant over the in-plane strain components, as exemplified in figs. S1 to S3. Using Eqs. 1 and 2, the spatial distribution of the pseudo-magnetic field can be obtained according to BS (r) = ∇ × A (r), with opposite signs associated with the K and K′ valleys so that the global time reversal symmetry is preserved and the total flux integrated over the entire sample is zero. Hence, from given 3D atomic structural distortions of graphene, which may be determined by either STM or high-resolution AFM, the spatial variations of BS (r) can be derived. Alternatively, the magnitude of pseudo-magnetic field, \|BS (r)\|, can be independently verified by spatially resolved STS, where the pseudo-magnetic field–induced quantized Landau levels En (with n being integers) in the tunneling conductance (dI/dV) versus biased voltage (V = E/e, with E being quasi-particle energy) spectrum at a given position r satisfy the following relation E n = sgn(n ) √ ______________ 2 ev F 2 ℏ ∣ nB S ∣ , ⇒ ∣ B S ∣ = [ ( E n+1 ) 2 − ( E n ) 2 ] / (2 ev F 2 ℏ) (3) Using Eq. 3, the magnitude of the pseudo-magnetic field at a given position can be determined rigorously by the energy spacing of dif- ferent Landau levels of varying indices n in the local tunneling spec- trum. The consistency of such spectroscopic studies with the value obtained from the strain tensors can be verified by comparing with the atomically resolved topographic studies. Topographic and spectroscopic evidences for the formation of giant pseudo-magnetic fields In Fig. 2, we illustrate the comparison of the strain-induced pseudo- magnetic fields for the K valley from both topographic and spec- troscopic studies at room temperature. The main panels of Fig. 2 (A and B) are respectively zoom-out AFM and STM topographic images over an area of 100 nm × 100 nm that cover the full view of mono- layer graphene/h-BN over an isolated Pd tetrahedron. In the inset of Fig. 2A, a zoom-in atomically resolved STM topography of graphene over an area of 3 nm × 3 nm near the tip of the tetrahedron reveals strong structural distortion in graphene with significant height dis- placements. Assuming the validity of first-order strain-induced perturbation to the Dirac Hamiltonian and using the MD method as detailed in the Supplementary Materials, we obtain the resulting pseudo-magnetic field distributions in Fig. 2C for the topography shown in Fig. 2B. In addition, maps of the corresponding strain ten- sors are provided in fig. S5 (A to C). Given the significant structural distortions in graphene, we note the resulting large magnitudes of the pseudo-magnetic field, up to ~800 T in maximum values if com- puted from the topographic information. Concurrent spectroscopic studies of the strained graphene over the isolated tetrahedron also revealed spatially varying tunneling spectra, as exemplified in Fig. 2D for a collection of high-resolution tunneling conductance versus bias voltage spectra along the black line indicated in Fig. 2C. Here, the horizontal axis in Fig. 2D corre- sponds to the bias voltage, the vertical axis corresponds to the spatial dimension along the black line (from lower left to upper right) in Fig. 2C, and the colors represent the tunneling conductance difference from the unstrained graphene. The 3D representation of the tunneling spectra taken along the same line cut is shown in Fig. 2E. Specifically, a typical V-shaped tunneling spectrum for ideal graphene is shown Fig. 2. Topographic and spectroscopic studies of strain-induced effects on graphene at room temperature due to one Pd tetrahedron NC. (A) 3D topographic images of the distorted graphene taken by AFM (main panel) and by STM (inset, zoom-in image with atomic resolution). (B) 3D topographic image of the distorted graphene taken by STM. (C) Pseudo-magnetic field map calculated from the topography over the same area as shown in (B). (D) Tunneling conductance spectral differ- ence relative to the Dirac spectrum of strain-free graphene is shown along the line cut indicated by the black arrow in (C), revealing spatially varying strengths of strain-induced pseudo-magnetic fields as manifested by the variations in the Landau-level separation. a.u., arbitrary units. (E) Representative spectra of tunneling conductance versus energy of strained graphene along the black line cut in (C), showing quantized conductance peaks in strained regions and the V-shape Dirac spectrum in strain-free regions as exemplified by the white curve located at r ~ 36 nm. (F) 3D topographic map of graphene/h-BN deformation on an ideal tetrahedron, as computed from MD simulations described in the Supplementary Materials. (G) Pseudo-magnetic field map computed from the topographic distortion in (F). (H) Comparison of the absolute values of pseudo-magnetic fields \|BS (r)\| derived from topographic studies (red line) and from the Landau-level separations in STS (black diamonds), showing overall satisfactory agreement. Here, r denotes the distance measured from the lower-left end to the upper-left end of the black arrow shown in (C). 3 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLE in the strain-free region, as exemplified by the white curve in Fig. 2E, whereas increasing larger energy separations for consecutive peak features are found for the tunneling spectra taken at increasingly strained regions, showing a consistent increase in the Landau-level energy separations with the increasing magnitude of strain found in the topographic studies. To further verify the consistency between the magnitude of pseudo- magnetic field determined from topography and from spectroscopy, we compare in Fig. 2H the absolute values of pseudo-magnetic fields \|BS (r)\| derived from topographic studies (Fig. 2C) and those from the Landau-level separations (Fig. 2D) using Eq. 3 and find overall reasonable agreement. Here, r denotes the distance measured from the lower-left end to the upper-left end of the black arrow in Fig. 2C. In addition, we carried out MD simulations for the topography and pseudo-magnetic field map of monolayer graphene/h-BN strained by a perfect tetrahedron with a base dimension of 30 nm, as shown in Fig. 2 (F and G). These MD simulations are largely consistent with the experimental results shown in Fig. 2 (B and C), although it is difficult to achieve detailed agreement due to unknown micro- scopic interaction parameters between the monolayer graphene/h-BN and the underlying nano-tetrahedron that are required to carry out the MD simulations. To better manifest the characteristics of point spectra taken on areas of strained graphene, we show, in fig. S6A, four point spectra taken on highly strained locations indicated as , , , and  on the pseudo-magnetic field map in fig. S6B (which is the same pseudo- magnetic field map as in Fig. 2C), where the corresponding pseudo- magnetic fields are \|BS (r)\| ~ 600 T and the resulting Landau levels n = 0, ±1, ±2, and ±3 are explicitly indicated. In addition, a theoretical fitting curve for one of the point spectra with \|BS (r)\| = 592 T is shown in fig. S6C, demonstrating that the superposition of Lorentzian Landau levels on top of a background Dirac spectrum achieves good agreement with the experimental data. We further note that for all point spectra taken in strained graphene areas, approximately half of the spectra reveal a zero-bias conductance peak that corre- sponds to a Landau level with n = 0, whereas the other half of the spectra are without a zero-bias Landau level. This phenomenon is the result of two zero modes associated with spontaneous local time- reversal symmetry breaking, which will be investigated further in Discussion. Next, we consider the strain on graphene induced by two closely spaced nano-tetrahedrons, as manifested by the topography in Fig. 3 (A and B) and the corresponding pseudo-magnetic field map for the K valley in Fig. 3C. In addition, maps of the strain tensors associated with Fig. 3B are given in fig. S7 (A to C). We found that the maximum magnitude for the pseudo-magnetic field computed from the struc- tural distortion was ~600 T (Fig. 3C), smaller than that found in the case of single tetrahedron (Fig. 2C). This is because comparable height displacements to those in Fig. 2B were spread over a larger lateral dimension in the case of two closely spaced tetrahedrons so that the magnitude of (∂h/∂i) (∂h/∂j) becomes significantly reduced, where i and j denote either x or y coordinate. Moreover, detailed compari- sons of the spectroscopically determined pseudo-magnetic fields [as exemplified in Fig. 3D for the line cut spectra along the white dashed line and in Fig. 3 (E and H) for the line cut spectra along the black dashed lines in Fig. 3C] with those determined topographically (Fig. 3C) were found to be in good agreement quantitatively. In addition to verifying the consistency between the topographic and spectroscopic derivations of strain-induced pseudo-magnetic fields, the development of a topographic “wrinkle” between two nearby nanostructures is noteworthy. Moreover, the resulting pseudo-magnetic fields along the wrinkle direction appeared to form quasi-1D “channels” of nearly uniform pseudo-magnetic fields, whereas those perpen- dicular to the wrinkle exhibited relatively rapid and continuous spa- tial variations with alternating signs. This formation of a topographic wrinkle in graphene between two nanostructures provides a hint for developing controlled and spatially extended strain to achieve global inversion symmetry breaking, which is the subject of our following exploration. Formation of periodic parallel graphene wrinkles for valley splitting and as topological channels Next, we used nanofabrication technology to develop regular arrays of nano-cones on silicon with processes described in Materials and Fig. 3. Topographic and spectroscopic studies of strain-induced effects on graphene due to two closely separated Pd tetrahedron NCs. (A) 3D topographic image of the distorted graphene taken by AFM. (B) 3D topographic image of the distorted graphene taken by STM. (C) Pseudo-magnetic field map calculated from the topography over the same area as shown in (B). (D) Tunneling conductance spectral difference from the Dirac spectrum along the line cut shown by the white dashed line in (C). (E) Spatially resolved tunneling spectra of strained graphene along the black dashed line in (C), showing strain-induced quantized conductance peaks. (F) 3D topographic map of graphene/h-BN on two ideal tetrahedrons computed from MD simulations. (G) Pseudo-magnetic field map computed from topographic distortion shown in (F). (H) Tunneling conductance spectral difference relative to the Dirac spectrum along the line cut shown by the black dashed line in (C). 4 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLEMethods and schematically illustrated in fig. S4. Two types of peri- odic arrays were explored. One was a triangular lattice structure and the other was a rectangular lattice structure, as shown by the SEM images in the top panels of Fig. 4, A and B, respectively. We found that the wrinkles induced on monolayer graphene by a triangular lattice had the tendency of forming along any of the three equivalent directions, as shown by the SEM image in the bottom panel of Fig. 4A. In contrast, wrinkles induced by the rectangular lattice were generally well aligned and parallel to each other, as exemplified by the SEM image in the bottom panel of Fig. 4B and the AFM images in the top panels of Fig. 4 (C and D). The corresponding pseudo-magnetic fields associated with the graphene distortions in the top panels of Fig. 4 (C and D) are computed from the topography and shown in bottom panels of Fig. 4 (C and D). It is worth noting that each extended graphene wrinkle results in four parallel, relatively uniform pseudo-magnetic fields along one direction and varying with alternating signs perpendicular to the chan- nels, as illustrated in the bottom panel of Fig. 4D. Given that the pseudo-magnetic fields as observed by K and K′ Dirac fermions are opposite in sign, the formation of parallel channels of pseudo-magnetic fields can effectively result in valley splitting and valley polarization. As illustrated by the theoretical simulations in the upper panels of Fig. 5 (A and B) and further detailed in the Supplementary Materials, for valley-degenerate Dirac fermions incident perpendicular to the parallel channels of pseudo-magnetic fields, K- and K′-valley fermions can become spatially separated and the lateral separation will increase with the increasing number of wrinkles they pass over, provided that the average separation (d) of consecutive wrinkles is less than the ballistic length (lB) of Dirac fermions. Specifically, the ballistic length lB is related to the conductance (G), mobility (), and carrier density (n2D) of Dirac fermions in mono- layer graphene by the following relation (7) G = 2 e 2 ─ 2ℏ ( k F l B ) = n 2D e, ⇒ l B = ( 2ℏ ─ ) 2e n 2D  = ( ℏ ─ e )  √  ─ k F _  n 2D (4) where kF = (n2D)1/2 is the Fermi momentum, e is the electron charge, and 2ℏ denotes the Planck constant. For typical values of n2D = 1010 to 1012 cm−2 and  ~ 105 cm2/V-s for our PECVD-grown graphene, we find that lB = 120 nm to 1.2 m. Thus, by proper nanofabrication to design the d value and by gating the PECVD-grown graphene for suitable n2D and lB, the condition d < lB can be satisfied within realistic experimental parameters to achieve valley splitting and therefore valley polarized currents. In addition to yielding valley splitting as discussed above, the par- allel distributions of alternating signs of pseudo-magnetic fields can serve as topological channels for chiral fermions. As shown in Fig. 5C, theoretical simulations for realistic arrays of nanostructures reveal that chiral Dirac fermions (i.e., either K or K′ fermions) can be pre- served when propagate along the parallel channels of strain-induced pseudo-magnetic fields, as illustrated by the simulations shown in the top panel of Fig. 5C. In addition, valley-polarized Dirac fermions can even be collimated along the topological channels if the incident angle deviates slightly from the channel direction, as exemplified in the bottom panel of Fig. 5C. Thus, parallel graphene wrinkles can serve as an effective conduit for protected propagation of valley- polarized Dirac fermions. DISCUSSION Spontaneous local time-reversal symmetry breaking and the resulting two zero modes Although the strain-induced pseudo-magnetic fields do not break the global time-reversal symmetry, the gauge potentials A and A* Fig. 4. Extended strain effects induced by periodic arrays of nano-cones on graphene. (A) Top: SEM image of triangular arrays of cone-shaped nanostructures fabricated on a SiO2/Si substrate. Bottom: SEM image of monolayer-graphene/h-BN films on the triangular arrays shown in the top panel, showing graphene wrinkles that appeared randomly along three equivalent directions. (B) Top: SEM image of rectangular arrays of cone-shaped nanostructures fabricated on a SiO2/Si substrate. Bottom: SEM image of monolayer graphene/h-BN films on the rectangular arrays shown in the top panel, showing graphene wrinkles parallel to the axis of closer spaced nano- structures. (C) AFM image (top) of three parallel graphene wrinkles and the corresponding map of pseudo-magnetic fields derived from the strain tensors (bottom). (D) AFM image (top) of the graphene wrinkle enclosed by the blue dashed box in (C) and the corresponding map of pseudo-magnetic fields derived from the strain tensors (bottom). 5 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLE Fig. 5. Parallel graphene wrinkles as topological channels for valley splitting and valley-polarized propagation. (A) Simulations for parallel graphene wrinkles as a valley splitter, showing the trajectories of initially valley-degenerate (K + K′) fermions from strain-free regions becoming split when injected vertically into regions with strain-induced periodic channels of pseudo-magnetic fields. Top: Trajectories of K and K′ fermions for an incident angle perpendicular to the parallel channels ( = 0°). Bottom: Trajectories of K- and K′-valley fermions for an incident angle at  = 15° relative to the normal vector of the parallel channels. (B) Top: Simulated trajectories of K and K′ fermions for an incident angle perpendicular to the realistic strain-induced parallel pseudo-magnetic fields ( = 0°) shown in Fig. 4C. Bottom: Simulated trajec- tories of K and K′ fermions for an incident angle at  = 15° relative to the normal vector of the realistic strain-induced parallel pseudo-magnetic fields shown in Fig. 4C. (C) Simulations for parallel graphene wrinkles as a valley propagator, showing the collimation of valley-polarized fermions. Top: Trajectories of K-valley fermions inci- dent at an angle parallel to the channels ( = 90°). Bottom: Trajectories of K-valley fermions incident at an angle  = 75° relative to the normal vector of the parallel channels. associated with the two valleys (also known as two pseudo-spins) K and K′ in reciprocal space are opposite in sign and give rise to a peculiar zero mode (36). This zero mode corresponds to a condensate where the Dirac fermions are delocalized over the entire sample, and yet they remain alternately localized and anti-localized for the pseudo- spin projection in the real space, yielding local spontaneous time- reversal symmetry breaking (36). Empirically, this spontaneous symmetry breaking may be manifested by the alternating presence and absence of the tunneling conductance peak at n = 0 for two in- equivalent sublattices in graphene, which has been previously demon- strated by STS studies of molecular graphene (11). In this study, we also found that the point spectra of all strained regions exhibit statistically equal probabilities of the two zero modes. That is, the tunneling spectra at zero bias (V = 0) exhibit either a conductance peak or a conductance gap, as exemplified in fig. S8A for the zero- bias conductance map of strained graphene over a Pd tetrahedron and the corresponding histograms in fig. S8B. This finding there- fore provides supporting evidence for spontaneous local time- reversal symmetry breaking due to strain-induced gauge potentials in real graphene. Nanoscale strain engineering of graphene-based valleytronic/spintronic devices The formation of periodic parallel graphene wrinkles by means of modern nanofabrication technology provides a pathway toward re- alizing controlled strain-induced effects for scalable development of graphene-based valleytronic devices. For instance, by patterning a valley Hall device configuration with the long-axis parallel to graphene wrinkles as schematically illustrated in fig. S9A, strong nonlocal re- sistance and VHEs may be detected under proper back-gated voltages, leading to a valley Hall transistor similar to previous observation of the VHE in exfoliated monolayer graphene–on–h-BN flakes (10). It is also conceivable to obtain highly valley-polarized currents through the combination of valley splitters (Fig. 5, A and B) and valley prop- agators (Fig. 5C), as conceptually illustrated in fig. S9A. Furthermore, by injecting valley-polarized currents into strong spin-orbit cou- pled materials, the outgoing currents can become spin-polarized for spintronic applications (fig. S9B). Last, we note that many such de- vices can be developed by means of scalable and reproducible nano- fabrication technology on large-area PECVD-grown graphene sheets (32), thus making the applications of graphene-based nanoscale valleytronic/spintronic devices closer to reality. Strain-induced superconductivity in monolayer graphene The discovery of superconductivity in bilayer graphene twisted at a “magic angle” (37) has kindled great interest in exploring “flat-band” materials (38) (i.e., materials with dispersionless energy versus momentum relation) for induction of superconductivity. A recent theoretical proposal (39) suggests that superconductivity may be more easily realized in topological flat bands induced by strain in graphene through periodic ripples and by including the effect of electronic correlation. It is argued that the chiral d-wave supercon- ductivity may be stabilized under strain even for slightly doped graphene and that superconductivity thus derived could exhibit the long-sought-after superconducting states with nonvanishing center- of-mass momentum for Cooper pairs (39). In the limit of (J/t) ~1 where J represents the antiferromagnetic coupling and t is the nearest-neighbor hopping energy, the theoretical conditions necessary for the occurrence of superconductivity are found to be (h/L) ≥ 0.05 and h2/(La0) ≥ 1, where h and L denote the height and periodic separation of the ripples, respectively, and a0 ≈ 0.142 nm is the nearest carbon-carbon distance for equilibrium graphene (39). For typical values of h = 20 nm and L = 300 nm in this work, we find (h/L) ≈ 0.067 and h2/(La0) ≈ 9.39 so that both 6 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLEtheoretical conditions are satisfied, implying possible occurrence of superconductivity if the premise of strong electronic correlation is justifiable (39). Given this intriguing prospect, it would be worthwhile to empirically explore possible strain-induced superconductivity in monolayer graphene with architected parallel wrinkles and to verify the validity of strong electronic correlation under giant pseudo- magnetic fields. However, empirical verifications of superconductivity require measurements at cryogenic temperatures. Given that the thermal expansion coefficient for graphene is negative and those for typical substrate materials (such as silicon) are positive, the strain in- duced by architected substrates in graphene is expected to decrease with decreasing temperature. Therefore, proper consideration of such complications will be necessary in the investigation of possible strain- induced superconductivity in graphene. In summary, we have demonstrated a controlled approach to ma- nipulating the topological states in monolayer graphene via nanoscale strain engineering. By placing strain-free monolayer graphene on architected nanostructures to induce global inversion symmetry break- ing, we are able to induce giant pseudo-magnetic fields (up to ~800 T) with desirable spatial distributions, realize global valley polarization, and achieve periodic 1D topological channels for protected propaga- tion of chiral fermion modes in strained graphene. The methodology presented in this work not only provides a platform for designing and controlling the gauge potential and Berry curvatures in graphene but also is promising for realizing scalable graphene-based valleytronic devices and strain-induced superconductivity. MATERIALS AND METHODS Graphene/BN/Pd tetrahedron sample preparation In this work, the Pd tetrahedron NCs were synthesized by a wet- chemical method (33). The preparation procedure is briefly summa- rized below. We mixed 7.6 mg of palladium (II) acetylacetonate [Pd(acac)2], 16.5 mg of iron (II) acetylacetonate [Fe(acac)2], 50.0 mg of polyvinylpyrrolidone, and 10.0 ml of N,N′-dimethylformamide into a 30-ml vial. After ultrasonication for 5 min, the mixture was heated at 120°C for 10 hours in an oil bath on a hotplate. The resulting precipitant were collected by centrifugation and rinsed with ethanol several times. A Si substrate was first ultrasonicated in acetone and subsequently in isopropyl alcohol for 10 min each, blown dry with dry nitrogen, and then loaded into a 100-W O2 plasma for 5 min to remove any traces of hydrocarbon residue. The Pd tetrahedron sus- pension was dropped onto the Si substrate and spun at 1500 RPM for 1 min. After the spin-coating process, the sample was loaded into a 100-W O2 plasma for 5 min again to remove any residue on the Pd tetrahedron NCs. The resulting typical size of the Pd tetrahedron NCs ranges from 50 to 70 nm and the height ranges from 40 to 60 nm, as exemplified in Fig. 1 (B and C). Next, a monolayer BN was trans- ferred over the substrate covered by the Pd tetrahedron NCs, fol- lowed by the transfer of PECVD-graphene onto the BN/Pd-NCs. AFM and SEM measurements were performed on every step of the process. We found that graphene/BN conformed very well to the Pd tetrahedron NCs if they were well separated from each other, as ex- emplified by the AFM image in Fig. 1 (D and E, top). However, we found that graphene tended to form wrinkles along the Pd tetrahe- drons if they were sufficiently closed to each other, as exemplified by the AFM images in the bottom panel of Fig. 1E. This situation is sim- ilar to our previous observation of graphene/h-BN on Au nanoparticles and graphene/h-BN on Si nanostructures (31). Procedures for fabricating periodic arrays of graphene/h-BN/SiO2 nano-cones SiO2 nano-cone substrate fabrication is schematically shown in fig. S4 and further described here. First, a typical e-beam lithography method was used to pattern an array of discs with ~50 nm diameters on a Si substrate with a 300-nm oxide layer. After development, 15-nm-thick Ni is deposited and used as a mask in a C4F8/O2 reac- tive ion etching environment to create Si nano-pillars. After etching, the substrate was immersed in the buffered oxide etch for ~20 s until the Ni discs fell from the top of the nano-cones. A typical size of nano-cone is ~40 nm in diameter and ~20 nm in height. A mono- layer h-BN was transferred over the SiO2 nano-cones, followed by the transfer of PECVD-grown graphene. Scanning tunneling microscopic and spectroscopic studies of strain-engineered graphene Two types of monolayer strained graphene samples were investigat- ed using STM/STS at room temperature. The samples were loaded onto our homemade STM system and pumped down to a vacuum level of 1.6 × 10−6 torr. Atomically resolved topographic and spec- troscopic measurements were carried out on samples at room temperature using a Pt/Ir STM tip with a typical tunnel junction resistance at 2 gigaohms. SUPPLEMENTARY MATERIALS Supplementary materials for this article is available at http://advances.sciencemag.org/cgi/ content/full/6/19/eaat9488/DC1 REFERENCES AND NOTES 1. M. V. Berry, Quantal phase factors accompanying adiabatic changes. Proc. R. Soc. London A392, 45–57 (1984). 2. G. P. Mikitik, Y. V. Sharlai, Manifestation of Berry’s phase in metal physics. Phys. Rev. Lett. 82, 2147–2250 (1999). 3. D. Xiao, M.-C. Chang, Q. Niu, Berry phase effects on electronic properties. Rev. Mod. Phys. 82, 1959–2007 (2010). 4. D. J. Thouless, M. Kohmoto, M. P. Nightingale, M. den Nijs, Quantized Hall conductance in a two-dimensional periodic potential. Phys. Rev. Lett. 49, 405 (1982). 5. X.-L. Qi, S.-C. Zhang, The quantum spin Hall effect and topological insulators. Phys. Today 63, 33–38 (2010). 6. M. Z. Hasan, C. L. Kane, Colloquium: Topological insulators. Rev. Mod. Phys. 82, 3045–3067 (2010). 7. A. H. Castro Neto, F. Guinea, N. M. R. Peres, K. S. Novoselov, A. K. Geim, The electronic properties of graphene. Rev. Mod. Phys. 81, 109–162 (2009). 8. F. Zhang, A. H. MacDonald, E. J. Mele, Valley Chern numbers and boundary modes in gapped bilayer graphene. Proc. Natl. Acad. Sci. U.S.A. 110, 10546–10551 (2013). 9. Y. Zhang, Y.-W. Tan, H. L. Stormer, P. Kim, Experimental observation of the quantum Hall effect and Berry’s phase in graphene. Nature 438, 201–204 (2005). 10. R. V. Gorbachev, J. C. W. Song, G. L. Yu, A. V. Kretinin, F. Withers, Y. Cao, A. Mishchenko, I. V. Grigorieva, K. S. Novoselov, L. S. Levitov, A. K. Geim, Detecting topological currents in graphene superlattices. Science 346, 448–451 (2014). 11. K. K. Gomes, W. Mar, W. Ko, F. Guinea, H. C. Manoharan, Designer Dirac fermions and topological phases in molecular graphene. Nature 483, 306–310 (2012). 12. L. Ju, Z. Shi, N. Nair, Y. Lv, C. Jin, J. Velasco Jr., C. Ojeda-Aristizabal, H. A. Bechtel, M. C. Martin, A. Zettl, J. Analytis, F. Wang, Topological valley transport at bilayer graphene domain walls. Nature 520, 650–655 (2015). 13. D. Xiao, G.-B. Liu, W. Feng, X. Xu, W. Yao, Coupled spin and valley physics in monolayers of MoS2 and other group-VI dichalcogenides. Phys. Rev. Lett. 108, 196802 (2012). 14. K. F. Mak, K. He, J. Shan, T. F. Heinz, Control of valley polarization in monolayer MoS2 by optical helicity. Nat. Nanotechnol. 7, 494–498 (2012). 15. K. S. Novoselov, A. Mishchenko, A. Carvalho, A. H. C. Neto, 2D materials and van der Waals heterostructures. Science 353, 6298 (2016). 16. S. Manzeli, D. Ovchinnikov, D. Pasquier, O. V. Yazyev, A. Kis, 2D transition metal dichalcogenides. Nat. Rev. Mater. 2, 17033 (2017). 17. A. J. Pearce, E. Mariani, G. Burkard, Tight-binding approach to strain and curvature in monolayer transition-metal dichalcogenides. Phys. Rev. B 94, 155416 (2016). 7 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLE 18. H. Li, A. W. Contryman, X. Qian, S. M. Ardakani, Y. Gong, X. Wang, J. M. Weisse, C. H. Lee, J. Zhao, P. M. Ajayan, J. Li, H. C. Manoharan, X. Zheng, Optoelectronic crystal of artificial atoms in strain-textured molybdenum disulphide. Nat. Commun. 6, 7381 (2015). 19. A. Reserbat-Plantey, D. Kalita, Z. Han, L. Ferlazzo, S. Autier-Laurent, K. Komatsu, C. Li, R. Weil, A. Ralko, L. Marty, S. Guéron, N. Bendiab, H. Bouchiat, V. Bouchiat, Strain superlattices and macroscale suspension of graphene induced by corrugated substrates. Nano Lett. 14, 5044–5051 (2014). 20. H. Tomori, A. Kanda, H. Goto, Y. Ootuka, K. Tsukagoshi, S. Moriyama, E. Watanabe, 36. D. Tsuya, Introducing nonuniform strain to graphene using dielectric nanopillars. Appl. Phys. Express 4, 075102 (2011). 21. J. Choi, H. J. Kim, M. C. Wang, J. Leem, W. P. King, S. W. Nam, Three-dimensional integration of graphene via swelling, shrinking, and adaptation. Nano Lett. 15, 4525–4531 (2015). 22. B. Pacakova, T. Verhagen, M. Bousa, U. Hübner, J. Vejpravova, M. Kalbac, O. Frank, Mastering the wrinkling of self-supported graphene. Sci. Rep. 7, 10003 (2017). 23. Y. Zhang, M. Heiranian, B. Janicek, Z. Budrikis, S. Zapperi, P. Y. Huang, H. T. Johnson, N. R. Aluru, J. W. Lyding, N. Mason, Strain modulation of graphene by nanoscale substrate curvatures: A molecular view. Nano Lett. 18, 2098–2104 (2018). 24. Y. Jiang, J. Mao, J. Duan, X. Lai, K. Watanabe, T. Taniguchi, E. Y. Andrei, Visualizing strain-induced pseudomagnetic fields in graphene through an hBN magnifying glass. Nano Lett. 17, 2839–2843 (2017). 25. J. L. Mañes, Symmetry-based approach to electron-phonon interactions in graphene. Phys. Rev. B 76, 045430 (2007). 26. F. Guinea, M. I. Katsnelson, M. A. H. Vozmediano, Midgap states and charge inhomogeneities in corrugated graphene. Phys. Rev. B 77, 075422 (2008). 27. F. Guinea, M. I. Katsnelson, A. K. Geim, Energy gaps and a zero-field quantum Hall effect in graphene by strain engineering. Nat. Phys. 6, 30–33 (2010). 28. M. L. Teague, A. P. Lai, J. Velasco, C. R. Hughes, A. D. Beyer, M. W. Bockrath, C. N. Lau, N.-C. Yeh, Evidence for strain-induced local conductance modulations in single-layer graphene on SiO2. Nano Lett. 9, 2542–2546 (2009). 29. N. Levy, S. A. Burke, K. L. Meaker, M. Panlasigui, A. Zettl, F. Guinea, A. H. C. Neto, M. F. Crommie, Strain-induced Pseudo-Magnetic fields greater than 300 Tesla in graphene nanobubbles. Science 329, 544–547 (2010). 30. N.-C. Yeh, M.-L. Teague, S. Yeom, B. L. Standley, R. T.-P. Wu, D. A. Boyd, M. W. Bockrath, Strain-induced pseudo-magnetic fields and charging effects on CVD-grown graphene. Surf. Sci. 605, 1649–1656 (2011). 31. N.-C. Yeh, C.-C. Hsu, M. L. Teague, J.-Q. Wang, D. A. Boyd, C.-C. Chen, Nanoscale strain engineering of graphene and graphene-based devices. Acta Mech. Sin. 32, 497–509 (2016). 32. D. A. Boyd, W.-H. Lin, C.-C. Hsu, M. L. Teague, C.-C. Chen, Y.-Y. Lo, W.-Y. Chan, W.-B. Su, T.-C. Cheng, C.-S. Chang, C.-I. Wu, N.-C. Yeh, Single-step deposition of high-mobility graphene at reduced temperatures. Nat. Commun. 6, 6620 (2015). 34. W.-H. Lin, T.-H. Chen, J.-K. Chang, J.-I. Taur, Y.-Y. Lo, W.-L. Lee, C.-S. Chang, W.-B. Su, C.-I. Wu, A direct and polymer-free method for transferring graphene grown by chemical vapor deposition to any substrate. ACS Nano 8, 1784–1791 (2014). 35. W.-H. Lin, V. W. Brar, D. Jariwala, M. C. Sherrott, W.-S. Tseng, C.-I. Wu, N.-C. Yeh, H. A. Atwater, Atomic-scale structural and chemical characterization of hexagonal boron nitride layers synthesized at the wafer-scale with monolayer thickness control. Chem. Mater. 29, 4700–4707 (2017). I. F. Herbut, Pseudomagnetic catalysis of the time-reversal symmetry breaking in graphene. Phys. Rev. B 78, 205433 (2008). 37. Y. Cao, V. Fatemi, A. Demir, S. Fang, S. L. Tomarken, J. Y. Luo, J. D. Sanchez-Yamagishi, K. Watanabe, T. Taniguchi, E. Kaxiras, R. C. Ashoori, P. Jarillo-Herrero, Correlated insulator behaviour at half-filling in magic-angle graphene superlattices. Nature 556, 80–84 (2018). 38. R. Bistritzer, A. H. MacDonald, Moiré bands in twisted double-layer graphene. Proc. Natl. Acad. Sci. U.S.A. 108, 12233–12237 (2011). 39. F. Xu, P.-H. Chou, C.-H. Chung, T.-K. Lee, C.-Y. Mou, Strain-induced superconducting pair density wave states in graphene. Phys. Rev. B 98, 205103 (2018). 40. M. Neek-Amal, F. M. Peeters, Graphene on boron-nitride: Moiré pattern in the van der Waals energy. Appl. Phys. Lett. 104, 041909 (2014). Acknowledgments Funding: The authors gratefully acknowledge joint support for this work by the Army Research Office under the MURI program (award #W911NF-16-1-0472), the National Science Foundation under the Physics Frontier Centers program for the Institute for Quantum Information and Matter (IQIM) at the California Institute of Technology (award #1733907), and the Kavli Foundation. Author contributions: N.-C.Y. conceived the ideas and coordinated the research project. C.-C.H. synthesized and characterized the strain-free monolayer graphene, developed architected nanostructures, transferred monolayer graphene and monolayer h-BN to the architected nanostructures for strain engineering, and carried out the SEM and AFM studies. M.L.T. performed the STM/STS studies on strained graphene and analyzed the topographic and spectroscopic data. J.-Q.W. carried out the MD simulations to map out the strain-induced pseudo-magnetic fields and developed a semi-classical model to determine the trajectories of valley-polarized Dirac fermions. N.-C.Y. wrote the paper with contributions from all coauthors. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. Submitted 29 April 2019 Accepted 14 February 2020 Published 8 May 2020 10.1126/sciadv.aat9488 33. Y. Zhang, M. Wang, E. Zhu, Y. Zheng, Y. Huang, X. Huang, Seedless growth of palladium nanocrystals with tunable structures: From tetrahedra to nanosheets. Nano Lett. 15, 7519–7525 (2015). Citation: C.-C. Hsu, M. L. Teague, J.-Q. Wang, N.-C. Yeh, Nanoscale strain engineering of giant pseudo-magnetic fields, valley polarization, and topological channels in graphene. Sci. Adv. 6, eaat9488 (2020). 8 of 8 Hsu et al., Sci. Adv. 2020; 6 : eaat9488 8 May 2020SCIENCE ADVANCES \| RESEARCH ARTICLE
10.1126_science.add5327	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 July 03. Published in final edited form as: Science. 2023 May 12; 380(6645): eadd5327. doi:10.1126/science.add5327. Epigenetic plasticity cooperates with cell-cell interactions to direct pancreatic tumorigenesis Cassandra Burdziak1,2,†, Direna Alonso-Curbelo3,4,†, Thomas Walle1,5,6,7, José Reyes1,3, Francisco M. Barriga3, Doron Haviv1,2, Yubin Xie1,2, Zhen Zhao3,8, Chujun Julia Zhao1,9, Hsuan-An Chen3, Ojasvi Chaudhary1,10, Ignas Masilionis1,10, Zi-Ning Choo1, Vianne Gao1,2, Wei Luan3, Alexandra Wuest3, Yu-Jui Ho3, Yuhong Wei11, Daniela F Quail11, Richard Koche12, Linas Mazutis1,9,13, Ronan Chaligné1,10, Tal Nawy1, Scott W. Lowe3,14,, Dana Pe’er1,14, 1Computational and Systems Biology Program, Sloan Kettering Institute, Memorial Sloan Kettering Cancer Center; New York, NY 10065, USA 2Tri-Institutional Training Program in Computational Biology and Medicine, Weill Cornell Medicine; New York, NY 10065, USA 3Cancer Biology and Genetics Program, Sloan Kettering Institute, Memorial Sloan Kettering Cancer Center; New York, NY 10065, USA 4Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology; Barcelona 08028, Spain 5Clinical Cooperation Unit Virotherapy, German Cancer Research Center (DKFZ); Heidelberg 69120, Germany This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Corresponding authors. [email protected], [email protected]. †These authors contributed equally to this work. Author contributions: Cassandra Burdziak: Conceptualization, Methodology, Software, Formal analysis, Data curation, Investigation, Writing-original draft presentation, Visualization, Funding Acquisition; Direna Alonso-Curbelo: Conceptualization, Methodology, Data curation, Investigation, Writing-original draft presentation, Visualization, Funding Acquisition; Thomas Walle: Formal analysis, Data curation, Investigation, Writing-review and editing, Visualization; José Reyes: Data curation, Investigation, Writing-review and editing; Francisco M. Barriga: Investigation; Doron Haviv: Formal analysis, Data curation, Investigation; Yubin Xie: Formal analysis, Data curation, Investigation; Zhen Zhao: Investigation; Chujun Julia Zhao: Formal analysis; Hsuan-An Chen: Investigation; Ojasvi Chaudhary: Investigation; Ignas Masilionis: Investigation; Zi-Ning Choo: Resources; Vianne Gao: Data curation; Wei Luan: Investigation; Alexandra Wuest: Investigation; Yu-Jui Ho: Data curation; Yuhong Wei: Resources; Daniela Quail: Resources; Richard Koche: Formal analysis; Linas Mazutis: Investigation; Ronan Chaligné: Investigation; Tal Nawy: Writing-original draft presentation; Scott W. Lowe: Conceptualization, Methodology, Writing-original draft presentation, Funding Acquisition, Study supervision; Dana Pe’er: Conceptualization, Methodology, Writing-original draft presentation, Funding Acquisition, Study supervision. Competing interests: Scott W. Lowe is a consultant and holds equity in Blueprint Medicines, ORIC Pharmaceuticals, Mirimus Inc., PMV Pharmaceuticals, Faeth Therapeutics, and Constellation Pharmaceuticals. A patent application (PTC/US2019/041670, internationally filing date 12 July 2019) has been submitted covering methods for preventing or treating KRAS mutant pancreas cancer with inhibitors of Type 2 cytokine signaling. Direna Alonso-Curbelo and Scott W. Lowe are listed as the inventors. Dana Pe’er is on the scientific advisory board of Insitro. Thomas Walle reports stock ownership for Roche, Bayer, Innate Pharma, Illumina and 10x Genomics as well as research funding (not related to this study) from CanVirex AG, Basel Switzerland and Institut für Klinische Krebsforschung GmbH, Frankfurt, Germany. Cassandra Burdziak, Direna Alonso-Curbelo, Scott W. Lowe, and Dana Pe’er are listed as inventors on a provisional patent application (63/390,075) related to aspects of this work, where Memorial Sloan Kettering Cancer Center is the applicant. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 2 6Department of Medical Oncology, National Center for Tumor Diseases; Heidelberg University Hospital, Heidelberg 69120, Germany 7German Cancer Consortium (DKTK); Heidelberg 69120, Germany 8Department of Pathology, Molecular and Cell-Based Medicine, Icahn School of Medicine at Mount Sinai; New York, NY 10029, USA 9Department of Biomedical Engineering, Columbia University; New York, NY 10027, USA 10Alan and Sandra Gerry Metastasis and Tumor Ecosystems Center; Memorial Sloan Kettering Cancer Center, New York 10065, NY, USA 11Rosalind and Morris Goodman Cancer Institute, McGill University; Montreal, QC H3A 1A3, Canada 12Center for Epigenetics Research, Memorial Sloan Kettering Cancer Center; New York, NY 10065, USA 13Institute of Biotechnology, Life Sciences Centre; Vilnius University, Vilnius LT 02158, Lithuania 14Howard Hughes Medical Institute; Chevy Chase, MD 20815, USA Abstract Introduction—Virtually all cancers begin with genetic alteration in healthy cells, yet mounting evidence suggests that non-genetic events such as environmental signaling play a crucial role in unleashing tumorigenesis. In the pancreas, epithelial cells harboring an activating mutation in the Kras proto-oncogene can remain phenotypically normal until an inflammatory event, which drives cellular plasticity and tissue remodeling. The inflammation-driven molecular, cellular, and tissue changes that precede and direct tumor formation remain poorly understood. Rationale—Understanding tumorigenesis requires a high-resolution view of events spanning cancer progression. We leveraged genetically engineered mouse models (GEMMs), single-cell genomic (RNA-seq and ATAC-seq) and imaging technologies to measure pancreatic epithelial cell-states across physiological, premalignant, and malignant stages. To analyze this rich and complex dataset, we developed computational approaches to characterize epigenetic plasticity and to infer cell-cell communication impacts on tissue remodeling. Results—Our data revealed that early in tumorigenesis, Kras-mutant cells are capable of acquiring multiple highly reproducible cell-states that are undetectable in normal or regenerating pancreata. Several such states align with experimentally validated cells-of-origin of neoplastic lesions, some of which display a high degree of plasticity upon inflammatory insult. These diverse Kras-mutant cell populations are defined by distinct chromatin accessibility patterns and undergo inflammation-driven cell fate transitions that precede pre-neoplastic and premalignant lesion formation. Furthermore, a subset of early Kras-mutant cell-states exhibit marked similarity to either the benign or malignant fates that emerge weeks to months later; for instance, Kras-mutant Nestin-positive progenitor-like cells display accessible chromatin near genes active in malignant tumors. We defined and quantified epigenetic plasticity as the diversity in transcriptional phenotypes that is enabled or restricted by a given epigenetic accessibility landscape. Intriguingly, these Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 3 plastic cell-states are enriched for open chromatin near cell-cell communication genes encoding ligands and cell-surface receptors, suggesting an increased propensity to communicate with the microenvironment. Given the rapid remodeling of both the epithelial and immune compartments during inflammation, we hypothesize that this epigenetically enabled communication is a major driver of tumorigenesis. We found that the premalignant epithelium displays extraordinary modularity with respect to communication gene co-expression patterns; distinct cell subpopulations each express a unique set of receptors and ligands that define the nature of incoming and outgoing signals that they can receive and send. Through the development of Calligraphy, an algorithm that utilizes this receptor-ligand modularity to robustly infer the cell-cell communication underlying tissue remodeling, we showed that the enhanced signaling repertoire of early neoplastic tissue specifically endows plastic epithelial populations with greater capability for crosstalk, including numerous communication routes with immune populations. As one example, we identified a feedback loop between inflammation- driven epithelial and immune cell-states involving IL-33, previously implicated in pancreatic tumorigenesis. Using a new GEMM that enables spatiotemporally controlled suppression of epithelial Il33 expression during Kras-initiated neoplasia, we functionally demonstrated that the loop initiated by epithelial IL-33 directs exit from a highly plastic inflammation-induced epithelial state, enabling progression towards typical neoplasia. Conclusion—Multimodal single-cell profiling of tumorigenesis in mouse models identified the cellular and tissue determinants of pancreatic cancer initiation, and a rigorous quantification of plasticity enabled the discovery of plasticity-associated gene programs. We found that Kras- mutant subpopulations markedly increase epigenetic plasticity upon inflammation, reshaping their communication potential with immune cells, and establishing aberrant cell-cell communication loops that drive their progression towards neoplastic lesions. Abstract The response to tumor-initiating inflammatory and genetic insults can vary amongst morphologically indistinguishable cells, suggesting yet uncharacterized roles for epigenetic plasticity during early neoplasia. To investigate the origins and impact of such plasticity, we performed single-cell analyses on normal, inflamed, pre-malignant, and malignant tissues in autochthonous models of pancreatic cancer. We reproducibly identified heterogeneous cell- states that are primed for diverse late-emerging neoplastic fates and linked these to chromatin remodeling at cell-cell communication loci. Using an inference approach, we revealed signaling gene modules and tissue-level crosstalk, including a neoplasia-driving feedback loop between discrete epithelial and immune cell populations that was validated in mice. Our results uncover a neoplasia-specific tissue remodeling program that may be exploited for pancreas cancer interception. One-Sentence Summary: Single-cell analysis reveals that enhanced epigenetic plasticity drives pro-neoplastic crosstalk in early pancreatic tumors. Graphical Abstract Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 4 The initial events by which tissues diverge from normalcy to form benign neoplasms and malignant tumors remain poorly understood. It is well established that this process is driven by genetic mutations (1); however, the discovery of prevalent cancer driver mutations in phenotypically normal epithelia (2) challenges the classic notion of cancer pathogenesis and underscores the essential role of cellular and environmental context (3–5). Indeed, non-mutagenic environmental insults promote tumor initiation in mice (6, 7) and chronic inflammatory conditions substantially increase cancer risk in humans (8, 9). These events can have heterogeneous effects even amongst morphologically indistinguishable and genetically identical cells from the same tissue (10). Genetic tracing studies similarly reveal that all such cells are not equally prone to undergo neoplastic and malignant reprogramming (11). This heterogeneity suggests that for tumorigenesis to proceed, select mutant cells either possess or gain an enhanced ability to change cell-states, a phenomenon known as cellular plasticity (12, 13). Developmental, regenerative, and pathologic plasticity is largely determined at the chromatin level as increases or decreases in the repertoire of transcriptional programs that can be accessed by a given cell (13, 14). Cells showing a high degree of plasticity, such as stem cells, often have a more ‘open’ or accessible chromatin landscape that becomes restricted during differentiation (15, 16). Previous work has used de-differentiation with respect to normal cell-states to characterize cancer cell plasticity with single-cell genomics from lung cancer models (17, 18). However, we still do not know how plasticity emerges in the earliest stages of tumorigenesis, particularly in concert with the environmental insults that accelerate these initiating events. Learning how plasticity is triggered to arise in pre-malignant tissues and how it contributes to early tumor evolution is paramount to understanding and intercepting cancer at its earliest stages. Pancreatic ductal adenocarcinoma (PDAC) is typically diagnosed too late for curative treatment and arises from cooperativity between genetic and epigenetic reprogramming events (19). Unlike more genetically heterogeneous cancers, PDAC is invariably initiated by an activating mutation in the proto-oncogene KRAS. However, KRAS-mutant epithelia Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 5 can remain phenotypically normal and depend on inflammatory stimuli (pancreatitis) to transform into pre-neoplastic and neoplastic lesions (20–22). We (23) and others (24, 25) have reported that oncogenic KRAS, in the absence of further mutation, cooperates with inflammation to trigger large-scale chromatin remodeling events that promote tumor initiation. However, important questions remain: How does KRAS-mediated plasticity give rise to neoplastic lineages and enable their subsequent evolution to invasive disease? What are critical cell-intrinsic and cell-extrinsic factors that determine a cell’s propensity to acquire a plastic and ultimately a tumorigenic cell-state? Understanding the answers to these questions may point to intervention strategies to prevent PDAC progression. To shed light on neoplastic plasticity in PDAC, we compared physiological, pre-malignant, and malignant cell-state heterogeneity using single-cell genomics, applying computational methods and functional perturbation in autochthonous genetically engineered mouse models (GEMMs) that accurately recapitulate many aspects of the human disease. Beyond providing a comprehensive charting of epithelial dynamics from normal metaplasia through malignant tissue states, our approach allowed us to expose, quantify, and perturb early plasticity traits endowed by oncogene-environment interaction, and define molecular, cellular, and tissue-level principles of pre-malignant tumor evolution. Results Targeted high-resolution profiling of epithelial dynamics during damage-induced neoplasia The study of epithelial dynamics in pancreatic cancer has been limited by the inability to capture early and transitional cell-states, which tend to be rare, short-lived, and difficult to identify. To characterize the full spectrum of epithelial cell-states in both normal and pathological tissue remodeling, we generated a single-cell transcriptomic (scRNA-seq) atlas of healthy, regenerating, benign neoplastic, and malignant epithelia using GEMMs that faithfully model cancer from initiation to metastasis. Our GEMMs incorporate a Ptf1a- Cre-dependent mKate2 fluorescent reporter to enrich pancreatic epithelial cells (23, 26, 27), allowing us to comprehensively profile pancreatic epithelial dynamics in well-defined tissue states. Specifically, we profiled pancreatic epithelial cells from healthy pancreas (N1) undergoing reversible metaplasia associated with normal regeneration after injury (N1→N2), and the metaplasia-neoplasia-adenocarcinoma sequence that initiates PDAC in the presence of mutant Kras (K1→K6) (Figs. 1A, S1A and Table S1). In this setting, as in human cancer (10), Kras-mutant metaplasia is accelerated by an inflammatory insult (pancreatitis) (pre-neoplasia; K1→K2), proceeds to benign pancreatic intraepithelial neoplasia, (PanIN; K3, K4), and ultimately, malignant PDAC (K5) and distal metastases (K6; Figs. 1A and S1A). Using a lineage tagging reporter to enrich for epithelial (mKate2+) cells, we captured both abundant and rare constituents of normal, regenerating, and Kras-mutant epithelia, such as progenitor-like tuft (Pou2f3+, Dclk1+), EMT-like (Zeb1+), neuroendocrine (Syp+ Chga+), and other previously reported subpopulations (28–30) (Figs. 1B, S1B–E and Table S2). We also characterized highly granular routes of acinar-to-ductal metaplasia (ADM) associated with regeneration and tumor initiation (23, 31) (Fig. S2A,B). Compared to healthy and regenerating pancreata, we uncovered a staggering expansion of new cell-states that emerge Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 6 during the earliest stages of KRAS-driven tumorigenesis (K1-K2, Figs. 1B, S1B–E and S2A–C). Despite such heterogeneity, the distinct cell-states captured within pre-malignant tissues were reproducible across biological replicates (individual mice) (Fig. S2D). In stark contrast, and consistent with studies analyzing human PDAC (32), malignant tumors isolated from different mice showed extensive inter-tumor variability, only sharing one small cell cluster (Fig. S2D,E). We next used diffusion maps (33) to characterize the major axes of transcriptional variation in our data, ordering cells along components associated with coherent gene expression patterns. The top component of variation closely matches progression from normal to regenerating, early tumorigenic, and finally late-stage disease, and is consistent with gene signatures that distinguish advanced human PDAC from normal pancreas (34, 35). Specifically, genes upregulated in human and mouse PDAC rise along the first diffusion component (DC), while normal pancreas programs are downregulated (Fig. 1C). Consistent with prior reports analyzing bulk RNA-seq data (23), the combined effects of Kras mutation and injury-driven inflammation are sufficient to induce signatures of human PDAC in pre-malignancy, as early as 24 to 48 hours post-injury (hpi) (Fig. 1C,D). However, the added granularity of our single cell analyses revealed that cancer-specific signatures are not induced uniformly across pre-malignant epithelial cell-states; for example, some rare early Kras-mutant cells express high levels of EMT gene programs (Zeb1, Vim; (30, 36)) (Figs. 1D and S1E). Kras-mutant cell-states are also observed with varying degrees of de-differentiation (downregulation of acinar genes) and reactivation of developmental (Clu) or oncogenic (Kras, Myc) programs (Fig. 1D). Thus, in the presence of inflammation, Kras- mutant pancreatic epithelial cells rapidly undergo specific and highly reproducible changes that endow select subpopulations with the capacity to activate disease-relevant programs long before malignant progression. Aberrant, highly plastic cell-states emerge early in PDAC progression To map the cellular origins and processes underlying this diversity in transcriptional cell- states, we first visualized heterogeneity in all Kras-mutant epithelia using a force-directed layout (FDL), which emphasizes cell-state transitions along axes toward malignancy. As expected, we found that the Kras-mutant pancreatic epithelium undergoes progressive gene expression changes that activate metaplastic (Clu+, Krt19+; (22)), neoplastic (Agr2+, Muc5ac+, Tff1+; (37)) and ultimately, invasive cancer (Foxa1+; (38)) programs (Fig. 2A). The relatively low heterogeneity in apoptotic or proliferative signatures (Fig. S3A) implies that much of the change in observed cellular states is likely due to cell-state transitions rather than population dynamics. To better characterize sources of cell-state variation, we applied CellRank (39), a data- driven approach that infers transcriptional dynamics from cell-cell similarity coupled with RNA velocity information (27, 40, 41). RNA velocities derived from the proportion of spliced to unspliced transcripts in each cell can indicate likely future states in neighboring phenotypic space. CellRank integrates directional information from per-cell velocity estimates with standard pseudotime inference based on cell-cell similarity to infer global transcriptional dynamics that can robustly pinpoint the origins of cell-state trajectories (39). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 7 Applying CellRank to early Kras-mutant cells acutely responding to an inflammatory insult (K1→K2) identified multiple states that potentially act as distinct origins for the observed heterogeneity (Fig. S3B). The four inferred origin or ‘apex’ states include one differentiated acinar (Ptf1a+) and three de-differentiated (Nes+, Aldh1b1+ and Tff2+) populations. Most of these states align with independent genetic lineage tracing studies that demonstrate their ‘cell-of-origin’ potential individually (26, 42–45) (Fig. S3B,C). These relationships are further supported by single molecule fluorescence in situ hybridization (smFISH) data derived from inflamed Kras- mutant tissue (K2), which revealed clear transitional states (Anxa10+ Nes+ Msn+) in lesions containing both apex cells (Nes+ Msn+) and the gastric-like cells (Anxa10+) predominant in neoplastic tissue weeks later (K3-K4, Fig. S3D,E). Moreover, several of the inferred apex states are highly responsive to inflammation, with apparent cell-state shifts along the cell-cell similarity graph emerging in the context of tissue injury (K2). For instance, during pancreatitis, well-differentiated acinar cells generate a metaplastic population with transcriptional features that are intermediate for acinar (Zg16, Cpa1) and tumorigenesis-associated (S100a6) programs within 24 hpi (ADM- PDAC “Bridge”) (Fig. S3F), and Nes+ progenitor-like cells shift into a state showing reduced activation of tumor suppressive programs (Cdkn2a). Our findings thus suggest that oncogenic Kras enables the emergence of diverse high-potential states (not observed in healthy nor regenerating pancreata (see Figs. 1B and S2)), each exhibiting distinct responses to inflammatory triggers, but all upregulating cancer-associated programs (see Fig. 1D). An epigenetic basis for high plasticity states Given the important role for chromatin dynamics in driving neoplasia (23), we hypothesized that the expansive phenotypic diversity in Kras-mutant apex states and their injury-driven progeny arises through a diversification of permissive chromatin states. To determine how chromatin dynamics correspond to changes observed in our longitudinal scRNA-seq atlas, we first analyzed bulk ATAC (assay for transposase-accessible chromatin) sequencing data matching the above tissue stages (23, 27). The dominant principal components of variation revealed accessibility patterns specific to each stage of progression (Fig. 2B). Compared to samples from normal pancreas epithelium (N1-N2), the chromatin landscapes of pre-neoplastic Kras-mutant epithelia (K1-K2) reproducibly shift toward states acquired in early neoplasia and sustained in advanced disease (K3-K6). Nevertheless, we observed that the chromatin landscapes of benign neoplastic lesions (K3-K4) and malignant tumors (K5-K6) are highly divergent. Consistent with this, large sets of regulatory elements (“chromatin modules”) exhibit mutually exclusive accessibility patterns across benign and malignant stages, with one set of ATAC-seq peaks showing increased accessibility in benign lesions but not in malignant counterparts (Benign Neoplasia chromatin module) and another set behaving opposite (Malignant chromatin module, Fig. 2C and Table S3). The modular structure of these data suggests that chromatin accessibility at benign (K3-K4) and malignant (K5-K6) stages corresponds to discrete, stable cell-states, which may underlie the clearly distinct morphologies (see Fig. S1A) and expression patterns (see Fig. S2D) characteristic to cells of these advanced stages. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 8 Mapping diverging Benign Neoplasia and Malignant chromatin accessibility modules to single-cell gene expression reveals a concordant pattern; genes proximal to Benign Neoplasia module loci are up-regulated in pre-malignant disease (K3-K4) and genes proximal to Malignant module loci increase in malignant disease (K5) (Fig. S4A). Chromatin modules associated with advanced stages are induced remarkably early in tumor development, such that Benign neoplasia and Malignant chromatin module associated genes are expressed and restricted to transcriptionally distinct populations within 24–48 hpi (K1- K2, Fig. 2C and Table S3). Furthermore, bulk ATAC-seq data show an initial increase in accessible chromatin in both modules from samples collected at 48 hpi, or without injury (K1-K2)—well before the emergence of benign lesions or malignant disease (K3- K6, Fig. 2C). These observations imply that the transcriptional diversity of pre-neoplastic Kras-mutant cells is established at the chromatin level and involves Benign Neoplasia or Malignant module activation prior to the development of PanINs or PDAC. This activation requires both oncogenic Kras and inflammation, as these programs are not similarly accessible or expressed in normal regeneration (N1-N2) (Figs. 2C and S4A). The early establishment of a permissive chromatin landscape (K1-K2) that is later specified into a restricted, distinct set of accessible regulatory elements (K3-K6) is reminiscent of cell-fate determination occurring in developmental systems (16). We thus refer to the cell-states that pre-neoplastic Kras-mutant cells may eventually acquire as ‘cell-fates’— those associated with benign neoplasia (K3-K4) or malignancy (K5-K6). We postulated that pre-neoplastic Kras-mutant cells (K1, K2) expressing programs associated with the chromatin landscape of a single distinct fate (Benign or Malignant) may be epigenetically primed toward that fate, conferring greater propensity to acquire its phenotype over time or in response to certain exogenous triggers. We further reasoned that similarities between the transcriptomes of Kras-mutant cells from pre-neoplastic (K1, K2) and later neoplastic stages (K3–K6) would indicate such fate potential. We therefore developed a classification- based approach that first identifies gene expression patterns that accurately discriminate between cell populations in benign lesions or cancer, and then uses these patterns to assign cell-fate probabilities to pre-neoplastic cells based on the activation of fate-associated genes. Specifically, we trained a logistic-regression classifier to distinguish between benign neoplasia (K3, K4) and malignancy (K5, K6), and used it to classify pre-neoplastic (K1, K2) cells (27). This classifier is highly accurate (99%) in assigning fate to PanIN and PDAC cells of known fate and identified a set of discriminative genes which have been linked to either fate (Fig. S4B). Applied to pre-neoplastic cells, this approach indeed pinpointed Kras-mutant cells that are strongly skewed toward one or the other fate (Figs. 2D and S4C). Most pre-neoplastic cells are classified as only having the potential to acquire a single fate, with cells responding to inflammation (K2) assigned a higher probability of acquiring a malignant fate. We also identified an intriguing set of Kras-mutant cells that are not well classified (Figs. 2D and S4D), the majority of which express a composite program of otherwise divergent fate-associated genes (Fig. S4E). These dual-primed subpopulations exist largely in the absence of tissue damage and overlap with initiating apex states (Ptf1a+ acinar and Nes+ progenitor) captured independently by CellRank (see Fig. S3B,C). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 9 Collectively, our results imply that tumorigenesis can proceed from multiple well- differentiated or progenitor-like states, and that their neoplastic progression is not dictated solely by cell intrinsic determinants (Kras gene mutation) but impacted by inflammatory signals that epigenetically prime them towards diverse fates that can be predicted early in disease progression. Epigenetic plasticity is enhanced by inflammation To map the epigenomic landscape at higher resolution, we generated single-cell chromatin accessibility (scATAC-seq) profiles of pre-neoplastic (K1), pre-neoplastic inflamed (K2), benign neoplastic (K3), and adenocarcinoma (K5) epithelia. Consistent with an epigenetic basis for the observed pre-malignant diversity (see Fig. 2C), we found considerable heterogeneity in chromatin accessibility within Kras-mutant epithelial cells at each stage (Figs. 3A and S5A). A major axis of variation in accessibility reproduced the divergence between benign and malignant fates seen in bulk analyses (Figs. 3B and S5B). Substantial variation in accessibility near fate-associated genes occurred across both stages and clusters (Fig. S5C), with Kras-mutant apex cells exhibiting a composite state defined by open chromatin at benign-associated and malignant-associated loci. This pattern extends to variation in open chromatin near other genes that define the benign and malignant chromatin modules (Fig. S5D). These data support bona fide epigenetic priming of divergent fate- associated programs in early, pre-neoplastic Kras-mutant cells. To better connect primed chromatin landscapes to their transcriptional outputs, we next sought to integrate scATAC-seq and scRNA-seq profiles from comparable stages. Clustering and cell-state annotation demonstrated that cell-states derived from scRNA-seq data largely match those derived from scATAC-seq data at the broad cluster level, including those corresponding to Nr5a2+ acinar, Neurod1+ neuroendocrine, Pou2f3+ tuft, and Nes+ progenitor cells (Fig. S5A). However, we also found substantial epigenomic heterogeneity within each scATAC-seq cluster. To explore this heterogeneity in more detail, we applied an algorithm that aggregates highly similar cells into granular cell-states, or metacells (27, 46, 47). Metacells provide much higher resolution than clusters, but aggregate cells sufficiently to reduce sparsity and improve statistical power for comparison. After separately identifying metacells for each scRNA-seq and scATAC-seq modality, we developed a framework to map between them based on similarity between a gene’s expression and its proximal chromatin accessibility (Fig. S6A–C) (27). This integrative analysis showed the expected correspondence between the accessibility and expression programs of comparable cell-states (Fig. 3C). However, we also observed extensive off-diagonal correspondence, indicating that features of the chromatin landscape are shared across diverse gene expression states. Specifically, we found diverse pre-malignant transcriptional states (tuft, neuroendocrine, progenitor, and gastric) to broadly correlate with the ADM epigenomic state, reflecting the known acinar history of these Ptf1a lineage-sorted cells (23, 26). In other cases, these correspondences may indicate widespread transcriptional “poising” of regulatory elements near unexpressed genes. Such effects were particularly evident in apex Nes+ progenitor cells, which exist in pre-neoplastic tissues at 48 hpi but establish chromatin landscapes that are highly correlated with those of late-stage malignant Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 10 populations (Figs. 3C and S5B). While these similarities can be partially explained by lineage relationships within each data modality, the greater off-diagonal (inter-cell-state) correlation existing across modalities (Fig. S6D) suggests that subpopulations of pre- neoplastic Kras-mutant cells are epigenetically primed to engage neoplasia transcriptional programs later in progression (see Fig. 2C). We sought to quantify the degree of epigenetic plasticity, which we define as the amount of diversity in transcriptional phenotypes that is enabled (or restricted) by a given chromatin accessibility landscape. To first determine these potential transcriptional phenotypes, we used a simple classifier to identify gene expression patterns that discriminate cell-states. Assuming that proximal open chromatin conveys the potential for a gene’s activation, we then applied the classifier to predict cell-states based on accessibility proximal to genes, rather than gene expression (Fig. 3D). We reasoned that for a given epigenomic state, uncertainty in such predictions serves as a measure of epigenetic plasticity. Following this logic, high plasticity is characterized by many accessible loci that define multiple discrete transcriptional states and thus produce high classifier prediction uncertainty. In contrast, low plasticity is defined by restricted potential diversity and prediction certainty. Applying this approach to epigenomic metacells identified populations of varying plasticity (Figs. 3E,F and S6E), with the most plastic states exhibiting striking overlap with the apex cells identified by CellRank (such as Nes+ progenitors, Tff2+ gastric cells) and experimentally validated cells-of-origin from lineage tracing studies (such as Pou2f3+ tuft cells; (28, 42– 44)) (Fig. 3F). Some of these states arise largely from pre-neoplastic conditions (K1-K2), aligning with observations on priming toward future neoplastic states. Notably, all plastic states identified in this analysis have no clear analog in normal or regenerating pancreata (see Figs. 1B, S2A,B), although a deeper exploration of physiological plasticity would be required to fully contrast normal and disease mechanisms. To expose potential unifying features of distinct plastic cell-states, we used gene set enrichment analysis (GSEA) (48) to identify gene signatures within populations displaying high plasticity scores (Table S4). This analysis revealed robust and consistent upregulation of sets related to cell-cell communication (Fig. S6F), with Cytokine-Cytokine Receptor Interaction yielding the top association (normalized enrichment score = 2.155, adjusted p value = 0.000) (Fig. 3G). A substantial fraction of these plasticity-associated genes encoded inflammatory mediators, receptors, or ligands involved in cell-cell communication, including those previously associated with malignant progression (Csf2, Cxcl1, and Cxcr2 (49)) (Table S5). Accordingly, plasticity increases significantly (p value = 0.006; one-tailed t-test, t = 2.5511) upon injury in the context of Kras mutation (K2 vs. K1) (Fig. 3H), suggesting an interplay between highly plastic cells and immune infiltrates flooding the pre-neoplastic tissue environment in this context (Fig. 3I). Together, our results indicate that epithelial plasticity in pre-malignant cells is directly associated with an increased, epigenetically-encoded propensity for ligand-receptor mediated communication with the immune microenvironment. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 11 Calligraphy charts cell-state-specific communication repertoires and their interactions The dominance of the association between plasticity and cell-cell communication drove us to investigate how heterotypic interactions may result from plasticity or enhance it in the Kras-mutant pancreatic epithelium. We hypothesized that chromatin remodeling of receptor and ligand gene loci (hereafter, ‘communication genes’) contributes to plasticity in pre-neoplasia by enabling cells to respond to inductive signals from the environment. The delineation of communication events requires an assessment of the communication propensities of each cell-state (defined by its expressed receptors and ligands), which may then be used to link interacting cell-states based on prior knowledge of receptor-ligand binding partners. As a first step, we characterized communication gene accessibility and expression across cell-states of the pre-malignant epithelium using the scATAC-seq and scRNAseq datasets generated above. Each plastic cell-state reveals substantial variability in chromatin accessibility near communication genes, consistent with distinct molecular repertoires for potential communication (Fig. S7A). To identify trends of coordinated gene expression, we searched for co-expression between any two communication genes (testing all combinations of two receptors, two ligands, and one of each) in individual cells across the pre-malignant epithelium and found a high degree of block structure in pairwise co-expression. This pattern implies that communication capabilities are driven by ‘communication modules’; sets of communication genes that are mutually expressed in the same cell populations (Fig. 4A). We next sought to infer actual cell-cell signaling interactions that may occur between cells expressing different communication modules. Although several methods have been developed to predict cellular interactions from single-cell data (50), their inference relies on weak signals arising from the noisy expression of a single cognate receptor- ligand (R-L) pair across fixed cell-states. We therefore developed our own approach, Calligraphy, that leverages the observed modularity in communication gene expression to infer potential cell-cell signaling events (27). Calligraphy first identifies communication modules—thereby establishing the incoming and outgoing communication each cell-state can participate in. Next, Calligraphy identifies communication events between cell-states based on prior knowledge of cognate R-L binding partners. Unlike previous methods which test interactions on individual R-L pairs, Calligraphy draws inferences across entire sets of genes, making the output insensitive to noise in any single gene (50). Using Calligraphy (27), we obtained seven communication modules of genes that are co-expressed across the pre-malignant pancreatic epithelium (Fig. 4A and Table S6) and are reproducible in situ in smFISH data (Fig. S7B). Mapping average expression of communication genes back onto the pre-malignant epithelium revealed that most cells express a single dominant module, making it possible to annotate cells by their corresponding module (Fig. 4B). Strikingly, cell-states defined solely by communication gene expression coincide with those identified by clustering the entire transcriptome (Figs. S7C–E). We further observe similar patterns in scATAC-seq data, where each subpopulation maintains open chromatin around genes of distinct communication modules, supporting an epigenetic basis for the emergence of these programs (Fig. S7F). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 12 Of note, normal and regenerating epithelial cells (N1, N2) showed much less diversity in communication module expression compared to their Kras-mutant counterparts, with most cells maintaining very low module expression (Fig. 4C). Among cells with wild-type Kras, a small injury-induced population does express communication modules associated with the Gastric (E6) cell-state, likely reflecting the expected trans/dedifferentiation of acinar cells under inflammatory conditions (51). These cells also expressed high levels of transcripts derived from a mutant Kras signature (23), implying that high level RAS signaling may play a role in normal regeneration and that KRAS mutation might stabilize such transient, injury-induced communication modules. Moreover, communication modules established in the pre-malignant pancreas are maintained in advanced cancers (K5, K6), with most cells expressing at least one of the Gastric (E6), Progenitor (E7), or Bridge (E3) modules (Fig. 4D). These modules (as well as their corresponding cell-states) are observed in an analogous mouse model of pre-neoplasia with activation of mutant Kras in adult acinar cells (52) (Fig. S8). Furthermore, these communication modules are conserved in human PDAC derived from multiple patients (32) (Fig. 4E). The distinct behavior of communication modules in early neoplasia across multiple model systems and their persistence in advanced murine and human PDAC implies a functional role in pancreatic tumorigenesis. Extensive epithelial-immune interactions drive oncogenic tissue remodeling The striking distinction of communication modules that arise in Kras-mutant epithelial cells during inflammation compared to normal regeneration implicate one or more signaling nodes in early PDAC development (20–22). Tissue damage produces inflammation and changes in immune cell-state and composition that contribute to neoplasia (53); thus we investigated how mutant Kras-driven epithelial communication modules interact with infiltrating and tissue-resident immune cells. scRNA-seq analysis of immune cells (CD45+ sorted) from Kras-mutant tissues, before and after induction of pancreatitis (K1–K3), identified all expected immune subtypes, including both abundant (macrophage) and rare (Treg, ILC) types (Fig. S9) (27). As expected, injury-induced inflammation causes dramatic remodeling of the immune cell landscape, including the enrichment or depletion of specific lymphoid and myeloid cell-states (Fig. S10A). Applying Calligraphy to these data identified consistent and structured communication modules defining distinct immune populations. To achieve even greater resolution, we ran Calligraphy separately on T cells/ILCs/NK cells, myeloid cells, and B cells, and found numerous modules containing known regulators as well as candidates for pancreatic tumorigenesis (Fig. S10B,C). For example, T cell/ILC/NK cell module 8 is highly expressed in ILC2, ILC3/LTi and Treg cells; these cells express the receptor for IL-33 (Il1rl1/Il1rap), a ligand that accelerates the formation of mucinous PanIN lesions (23). Reasoning that such rapid immune and epithelial remodeling could arise through heterotypic crosstalk in pancreata undergoing neoplastic transformation, we utilized a feature in Calligraphy to nominate potential cell-cell interactions that drive this process (Fig. S11A) (27). To limit our search to tumorigenesis-specific crosstalk following tissue inflammation, we filtered Calligraphy modules to retain those cognate R-L pairs in which at least one Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 13 partner is selectively upregulated in Kras-mutant (K2) relative to Kras-wild type (N2) epithelial cells (23). This filtering reduced the space of possible interacting molecules from 340 total communication genes down to 55 receptors and 46 ligands potentially involved in tumorigenesis-associated communication. We then assumed that two cell-states potentially interact if their associated modules are enriched in the number of shared cognate R-L pairs spanning them (27). Whereas CellphoneDB (50) predicts a highly dense network of 720 out of 729 (98.8%) of possible interactions, involving nearly all pairwise combinations of cell- states, Calligraphy identified a sparser, more interpretable network of potential neoplasia- specific interactions between the Kras-mutant epithelium and the immune environment (5.6% of possible interactions, Fig. S11B,C and Table S7). Within Calligraphy’s context-specific network were apparent ‘master communication hubs’ that participate in numerous interactions. We calculated a receiving score (ability to sense the environment via expressed receptors) and a transmission score (ability to remodel the environment via expressed ligands) based on the number of Calligraphy’s statistically significant incoming and outgoing edges for each module (p value < 0.1) (27). The two most prominent hubs for transmitting and receiving interactions are the epithelial Gastric (E6) and Progenitor (E7) modules, respectively (Fig. 4F), which correspond to ‘high-plasticity’ populations identified above. These same communication hubs are enriched in advanced mouse and human PDAC (Fig. 4D,E). To validate predicted communication networks, we mapped interacting modules to their spatial context, leveraging smFISH data including probes for transcripts marking distinct cell-states and their corresponding communication genes. We identified cells with concordant communication gene expression patterns across space for module E6-expressing gastric cells (Anxa10+ Il18+ Spp1+) and module E7-expressing progenitor cells (Nes+ Il18rl+ Cd44+) (Figs. 4G and S7B) in pre-neoplastic cells subject to inflammation (K2). Calculating distances between receiving (progenitor) cells and their closest sending (gastric) cells, against a random control set of gastric cells which do not express these ligands, showed a significant enrichment (t-test, p value < 0.01) of receiving cells in the vicinity of their interacting sending cells (Fig. 4H). Neoplastic tissue remodeling involves feedback communication loops One of our most striking observations is the dramatic remodeling of epithelial and immune compartments within 24 to 48 hpi (see Figs. S2D and S10A). This remodeling is highly reproducible and expansive, involving numerous new cell-states that are quickly adopted by most cells. The dynamics of such a rapid and robust response suggest a feedback loop, by which immune cell intermediates may amplify tumor-promoting epithelial cues (54). We probed the Calligraphy module-module interaction network to systematically enumerate cycles involving any epithelial or immune subsets and identified only one feedback loop in the system (Fig. 5A) (27). The loop includes the Gastric (E6) hub module, which is maintained in late disease, and further involves cytokines and receptors with reported roles in KRAS-driven pancreatic tumorigenesis, including IL1A, IL-33, and IL4RA (55, 56). Specifically, it engages Treg and ILC2 cells via IL-33 signaling before feeding back to epithelial cells (Fig. 5B). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 14 Il33 is expressed during pancreatitis by a small subset (4%) of TFF1/ANXA10+ Kras- mutant Gastric module-expressing epithelial cells and is predicted to initiate signaling to Tregs and ILCs (Module T8) by binding with its cognate receptor Il1rl1 and co- receptor Il1rap (Figs. 5B and S12A–D). Supporting the relevance of these interactions, immunofluorescence (IF) data reveal that IL-33-expressing epithelial cells (IL-33+ mKate2+) and rare Tregs (Foxp3+) are in close spatial proximity in Kras-mutant pancreata under injury conditions (distance vs. randomly permuted positions, t-test p value < 0.01 for all IF images collected across five independent mice) (Figs. 5C,D and S12E). Subsequently, many receiving cells that express Il1rl1 (Module T8) also express the Th2 cytokine gene Il4 (Fisher’s exact test; odds ratio = 21.47, p value = 9.88 × 10−35), consistent with the known role of IL-33 in triggering Th2-type immune responses (57). Module T8 cells then apparently signal through IL-4 (Il4) back to the Gastric Module (E6) via the IL-4 receptor (Il4ra), thereby closing the loop and potentially propagating signals to other modules in both immune and epithelial compartments (Fig. 5B). The broad expression of the IL-4 receptor across Kras-mutant epithelial cell-states, including gastric, tuft cell and Nes+ progenitor populations (Figs. 5B and S12A,C), implies that this signaling loop has a system-wide impact on pre-malignant tissue (45% of pre-malignant epithelial cells appear impacted). In contrast, few wild-type normal pancreas cells express both sending (Il33) and receiving (Il4ra) factors, and do so at low levels, even during injury-induced regeneration (Figs. S12F,G). A particular strength of Calligraphy is its ability to dissect the complexity inherent to tissue crosstalk by constructing communication circuits linking a cascade of signaling events between multiple communication modules in a serial fashion, thus mapping cell populations that are potentially both directly and indirectly affected by epithelial-derived IL-33. Calligraphy predicts that IL-33-driven communication has a large impact on pre- malignant tissue, and the proportion of affected tissue increases via signaling cascades between communication modules that each utilize multiple cognate R-L pairs, ultimately reaching the vast majority of the pre-malignant pancreas (72% of cells, Fig. 5E,F). While it is unlikely that the IL-33 loop is solely responsible for KRAS-driven tumor progression in the context of the high complexity of observed intercellular communication in the pre- malignant tissue, the number of populations that appear directly and indirectly impacted by epithelial IL-33 expression suggests that this communication circuit plays an important role in driving tumorigenesis. KRAS-dependent IL-33 feedback loop directs rapid tissue remodeling in early tumorigenesis Whereas previous studies have implicated stroma-derived IL-33 in disease phenotypes (58, 59), Calligraphy identified a feedback loop driven by IL-33 expressed from epithelial cell-states. To determine how IL-33 derived specifically from the epithelium contributes to early neoplasia, we developed a GEMM that enables specific Il33 suppression in lineage- traced Kras-mutant epithelial cells. Animals were produced from multi-allelic embryonic stem cells engineered to harbor a conditional KrasG12D allele together with a doxycycline (dox)-inducible GFP-coupled short hairpin RNA (shRNA) capable of suppressing Il33 Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 15 (KC-shIl33), allowing potent Il33 suppression in the epithelial compartment following dox administration (Fig. 6A,B). Additionally, a separate cohort of animals was produced harboring a control shRNA (shRen) to control for potential perturbations of the RNA interference machinery and dox. scRNA-seq and spatial imaging (Imaging Mass Cytometry, IMC) were performed on each model assessing epithelial and immune compartments at an early time-point (48 hpi), when inflammation unleashes neoplastic remodeling (K2), and later (3 weeks post-injury, or wpi), when PanIN lesions normally emerge (K3). As expected, IL-33 expression remains intact in non-epithelial pancreatic cells (Il33+ Vim+ or Il33+ aSMA+) (Fig. S13A–D) in shIl33 animals on dox, and is specifically abrogated in Kras mutant cells expressing the gastric markers TFF1 and ANXA10 (Figs. 6B and S13A,B). Analysis of co-embedded scRNA-seq data derived from control or KC-shIl33 on-dox mice show that IL-33 perturbation profoundly shifted the observed cell-states within both epithelial and immune compartments. We applied the Milo algorithm (60), which characterizes such local shifts (as opposed to loss or gain of entire clusters) by grouping similar cells into ‘neighborhoods’ and identifying those neighborhoods which are differentially abundant between perturbed and control conditions. Consistent with an epithelial-to-immune crosstalk, we found that Il33 suppression in the Kras-mutant epithelium results in rapid remodeling of the immune landscape, with multiple immune subpopulations shifting in abundance by 48 hpi (Fig. 6C). Epithelial remodeling is delayed by comparison; a lack of substantial changes at the early time point is followed by dramatic remodeling of many epithelial cell-states at 3 wpi (K3). By this time, the perturbation of IL-33-mediated crosstalk generates evidence of neoplastic epithelial remodeling, with a shift to more cells in progenitor-like and fewer in gastric-like states (Figs. 6D and S13E,F), also seen by immunofluorescence data (Fig. 6E,F). To gain insights into the impact of IL-33 on the dynamics of cell-state transitions, we used Palantir (61) to infer a pseudotime ordering of epithelial neighborhoods at 3 wpi, beginning from the Nes+ progenitor state (Fig. S13E). Ordering Milo log fold-change values along this pseudotime axis confirms that more IL-33-perturbed epithelial cells accumulate in earlier states expressing progenitor markers (Nes) and other genes associated with a plastic state (plasticity score correlation p value < 0.01) (Figs. 6D,G). Although Il33 is expressed in only a small fraction of Kras-mutant epithelial cells, Il33 perturbation results in marked changes in the cell-state composition of the pre-malignant pancreas, apparently by preventing the transition from a plastic progenitor-like state into distinct PanIN populations, such as the gastric-like cells that are normally abundant in unperturbed epithelia by 3 wpi. The widespread changes in cell-state due to IL-33 perturbation support Calligraphy’s prediction of the relevance of this feedback loop. To more directly link the specific predicted interacting partners with observed perturbation-induced changes, we mapped each Milo neighborhood to Calligraphy communication modules (Fig. S13G) and evaluated the extent to which cell-states predicted to be downstream of IL-33-mediated crosstalk overlap with cell-states impacted by the perturbation. Qualitatively, we found the largest impact of Il33 perturbation on Progenitor and Bridge modules, both of which are predicted to be downstream of IL-33 and express IL-4 receptor (Il4ra) (see Fig. 5B,E); whereas the only two modules not downstream of IL-33 are those with the smallest effect sizes (E2 and E5). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 16 Quantitatively, cell-states predicted by Calligraphy to participate in the IL-33 network were more significantly affected by the Il33 perturbation (one-sided t-test; t = −5.25, p value = 1.24 X 10−7) (Fig. 6H). These results functionally validate Calligraphy as an approach to infer both communication circuits and the specific subpopulations impacted (directly and indirectly) upon perturbation of such networks. Discussion While much is known about the molecular processes affecting tumor progression to advanced PDAC, pancreatic cancer is diagnosed late, and the paucity in molecular studies of early neoplasia has left us with little knowledge of how it emerges from a relatively homogeneous epithelium. By combining single-cell sequencing of mouse models with computational analysis, we found that permissive chromatin states in Kras-mutant cells diversify the communication programs available to pre-neoplastic tissue, expanding downstream crosstalk throughout the tumor microenvironment. Moreover, in the Kras- mutant context, epigenetic reprogramming and the emergence of cancer-driving populations is remarkably dynamic, occurring within two days of insult by inflammation. Mutation is known to drive plasticity in lung cancer via the loss of AT1 or AT2 lineage identity and acquisition of a phenotype intermediate between these states (17, 18). In the pancreas, a similar loss of acinar identity and gain of an intermediate acinar-ductal state have long been observed in both tumorigenesis and regeneration; thus, traditional notions of plasticity are insufficient to describe its contribution to disease. We defined plasticity as the potential of a cell to manifest diverse future fates, motivating a generalizable plasticity score that tracks with the degree of epigenetic priming. This score nominated several highly plastic cell-states, in which open chromatin unlocks access to multiple distinct gene programs observed in benign lesions or malignant disease and revealed that inflammation enhances plasticity across these states. To better elucidate the emergence of plastic states, we sought to reconcile prior work proposing different cells-of-origin for neoplasia. Our GEMMs harbor mutant Kras in all acinar cells, allowing us to comprehensively explore which states can initiate tumorigenesis. Using CellRank (39), we traced the origins of epithelial transcriptional diversity to multiple ‘apex’ progenitor populations that correspond with experimentally determined cells-of- origin. These populations also exhibit high plasticity scores and unify prior work by suggesting that neoplasia can arise from multiple Kras-mutant cell-states through distinct responses to inflammation. Moreover, our Ptf1a-Cre model traces this diversity back to a predominantly acinar-like state, supported by the fact that nearly all pre-malignant epithelial cell-states have an acinar-like chromatin state (epigenetic ‘memory’), which itself maps to a CellRank-predicted apex state. While non-acinar lineage cells can also undergo neoplastic transformation in mice (21, 62), our results agree with the observed loss of normal lineage identity upon Kras mutation and inflammation (63) and reveal apex states which may emerge following this transition. Among apex states, multiple unbiased analyses in particular support a Nes+ progenitor-like state (44, 45), which displays PDAC-associated chromatin alterations, expresses progenitor-associated genes, and scores highest for our plasticity metric—all hallmarks of highly plastic cells. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 17 Our plasticity score was most correlated with cytokine and receptor genes, implying that plastic populations are primed to both signal and respond to the environment. Addressing this, we asked how cell-cell communication may drive rapid tissue remodeling. Communication inference approaches failed to find specific signals among the large number of cytokines and receptors expressed across cell populations. We therefore developed Calligraphy to leverage modularity in gene expression for greater power and robustness over testing individual receptor-ligand gene pairs, allowing us to focus on neoplasia-specific communication networks. Calligraphy identified modules of co-expressed communication genes that, surprisingly, mapped one-to-one to transcriptional cell-states, implying that communication is critical for establishing cell-state diversity within the pre-malignant pancreas. These networks were largely absent from normal pancreas, with only one being induced in a rare subpopulation of cells that emerges upon tissue damage. The same module has the highest propensity for tissue remodeling and persists in advanced murine and human cancers, demonstrating that cancer commandeers gene programs used during normal regeneration. Our analyses revealed a feedback loop initiated by IL-33 signaling from epithelial cells expressing the Gastric module to Th2 cytokine-expressing Tregs and ILCs, which signal back to the epithelium (among multiple other routes). These findings link previous results on the relevance of Th2 signaling in PDAC tumorigenesis (56) to those on the role IL-33 in this process (23). Spatial analysis revealed co-localization of signaling populations in the loop, and epithelial Il33 knockdown in a GEMM impaired inflammation-driven remodeling of plastic populations, blocking the emergence of gastric-like state cells that are otherwise abundant in PanIN lesions. This mechanism can be driven solely by epithelium- derived IL-33, despite the high stromal IL-33 expression previously implicated in disease phenotypes (58, 59). Further, the results of in vivo Il33 perturbation support Calligraphy inference, by matching predictions of which populations are perturbed, to what degree, and in what temporal sequence. Other modules defined herein are likely also to have functional importance. Future work can extend this approach to other niche components such as fibroblasts or endothelial cells and should expose additional communication with potential for therapeutic or diagnostic exploitation. PDAC is frequently detected too late for curative intervention, a detailed understanding of early neoplastic events may enable the development of rational strategies to prevent, detect, and intercept tumors before they progress to an intractable stage. Our results show that GEMMs can be used to study and perturb early events, revealing epigenetically plastic cell-states in neoplasia that are not observed in the normal or regenerative pancreas. Further efforts to understand neoplasia-specific communication networks driving PDAC initiation hold promise for the development of therapeutics that block early cancer progression, and may also be effective against advanced disease. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 18 Materials and methods summary Experimental design: Samples from GEMMs were collected to span the entire range of PDAC progression (K1-K6) as well as regenerating pancreata (N1-N2). Additional samples were collected from GEMMs enabling selective genetic perturbation of pre-malignant Kras-mutant cells. Table S8 summarizes experimental conditions (27). All animal experiments were performed in accordance with the Institutional Animal Care and Use Committee (IACUC)-approved protocol (11–06–018). Generation of bulk and single-cell omics data: Tissue dissociation and cell preparations for bulk and single-cell ATAC-seq were performed as previously described (23). For scRNA-seq analysis, FACS-sorted epithelial or immune cells were encapsulated and processed following 10x Genomics user manual (Reagent Kit 3’ v2) as described (27). Spatial and immunophenotyping data: Tissues were processed and stained for imaging (IF/IHC, IMC, H&E, and smFISH) or FACS analyses (27). Tables S9 and S10 summarize panels used for multiplexed IMC and smFISH. IMC data was collected using Hyperion Imaging System and CyTOF Software v7.0.8493.0 (Fluidigm). smFISH imaging was performed on a Nikon Ti2 inverted microscope. FACS data was acquired using a 5-laser BD LSRFortessa and analyzed using FlowJo v10.0. Computational analysis: scRNA-seq data were processed with SEQC (64), filtered with a custom pipeline (27), and log library size normalized. scATAC-seq data were processed with ArchR (65). Processed transcriptomic and epigenomic datasets were analyzed with custom Python scripts for visualization, cell-state annotation, metacell inference, multimodal integration, plasticity scoring, and Calligraphy communication inference, among other analyses fully described in (27). smFISH image analysis was performed on maximum projection images with segmentation on the DAPI channel using Mesmer (66) and Python code for phenotyping and spatial analyses. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgments: We thank J. Simon and the MSKCC animal facility for technical support with animal colonies; MSKCC Flow Cytometry facility for support with FACS sorting experiments; the Single Cell and Imaging Mass Cytometry Platform at Goodman Cancer Research Centre; the members of the Sloan Kettering Institute’s Single Cell Analytics and Innovation Lab (SAIL) computational unit, and members of the Lowe and Pe’er laboratories for advice and discussions, in particular Manu Setty and Andrea Chaikovsky for foundational scRNA-seq analyses and critically revising the final manuscript, respectively. We thank Jeff Moffitt, Brianna Watson, Jenna Hurley and Sam Aviles for generously sharing their knowledge, protocols and guidance to help us set up the multiplex smFISH platform in the lab. We thank Siting Gan and Dig Vijay Kumar Yarlagadda for their pivotal contributions to establish our infrastructure for multiplex smFISH imaging. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Funding: Page 19 C.B. is supported by the Ruth L. Kirschtein Predoctoral Fellowship (NCI grant F31CA24690). D.A.C. is supported by the La Caixa Junior Leader Fellowship (LCF/BQ/PI20/11760006), FERO-ASEICA (BFERO2021), and the Spanish Ministry of Science and Innovation Grant (PID2021–128102OA-I00). F.M.B. is supported by the Edward P. Evans Young Investigator Award. T.W. is supported by a fellowship of the DKFZ Clinician Scientist Program, supported by the Dieter Morszeck Foundation. SCIMAP is supported by the Fraser Memorial Trust and a McGill MI4 Platform grant. J.R. is a Howard Hughes Medical Institute Fellow of the Damon Runyon Cancer Research Foundation (DRG-2382–19). C.J.Z. is supported by NIH grant R25 CA233208. The work was supported by Alan and Sandra Gerry Metastasis and Tumor Ecosystems Center (GMTEC) funding and NCI Cancer Center Support Grant P30 (CA008748). S.W. L. is an investigator in the Howard Hughes Medical Institute and the Geoffrey Beene Chair for Cancer Biology. D.P. is an investigator in the Howard Hughes Medical Institute, Alan and Sandra Gerry Endowed Chair, and is supported by NCI U54 (CA209975), NICHD DP1 (HD084071), and the Starr Cancer Consortium. Data and materials availability: All sequencing data have been deposited at the Gene Expression Omnibus (GEO) under accession GSE207943. An interactive data browser to plot gene expression trends on tSNE or FDL visualizations of scRNA-seq data is accessible at http://pdac-progression- browser.us-east-1.elasticbeanstalk.com. Code for data analysis is available at https:// github.com/dpeerlab/pdac-progression (DOI: 10.5281/zenodo.7738450). KC-shIL33 ESCs for the production of EPO-GEMMs are available from the corresponding author (S.W.L.) upon request. References and Notes 1. Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, Diaz LA Jr, Kinzler KW, Cancer genome landscapes. Science. 339, 1546–1558 (2013). [PubMed: 23539594] 2. Martincorena I, Roshan A, Gerstung M, Ellis P, Van Loo P, McLaren S, Wedge DC, Fullam A, Alexandrov LB, Tubio JM, Stebbings L, Menzies A, Widaa S, Stratton MR, Jones PH, Campbell PJ, Tumor evolution. High burden and pervasive positive selection of somatic mutations in normal human skin. Science. 348, 880–886 (2015). [PubMed: 25999502] 3. Wijewardhane N, Dressler L, Ciccarelli FD, Normal Somatic Mutations in Cancer Transformation. Cancer Cell. 39, 125–129 (2021). [PubMed: 33220180] 4. Hanahan D, Hallmarks of Cancer: New Dimensions. Cancer Discov. 12, 31–46 (2022). [PubMed: 35022204] 5. Nam AS, Chaligne R, Landau DA, Integrating genetic and non-genetic determinants of cancer evolution by single-cell multi-omics. Nat. Rev. Genet. 22, 3–18 (2021). [PubMed: 32807900] 6. Guerra C, Schuhmacher AJ, Cañamero M, Grippo PJ, Verdaguer L, Pérez-Gallego L, Dubus P, Sandgren EP, Barbacid M, Chronic pancreatitis is essential for induction of pancreatic ductal adenocarcinoma by K-Ras oncogenes in adult mice. Cancer Cell. 11, 291–302 (2007). [PubMed: 17349585] 7. Carrière C, Young AL, Gunn JR, Longnecker DS, Korc M, Acute pancreatitis markedly accelerates pancreatic cancer progression in mice expressing oncogenic Kras. Biochem. Biophys. Res. Commun. 382, 561–565 (2009). [PubMed: 19292977] 8. Lowenfels AB, Maisonneuve P, DiMagno EP, Elitsur Y, Gates LK, Perrault J, Whitcomb DC, Hereditary Pancreatitis and the Risk of Pancreatic Cancer. J. Natl. Cancer Inst. 89, 442–446 (1997). [PubMed: 9091646] 9. Coussens LM, Werb Z, Inflammation and cancer. Nature. 420, 860–867 (2002). [PubMed: 12490959] 10. Giroux V, Rustgi AK, Metaplasia: tissue injury adaptation and a precursor to the dysplasia-cancer sequence. Nat. Rev. Cancer. 17, 594–604 (2017). [PubMed: 28860646] 11. Maddipati R, Stanger BZ, Pancreatic Cancer Metastases Harbor Evidence of Polyclonality. Cancer Discovery. 5 (2015), pp. 1086–1097. [PubMed: 26209539] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 20 12. Torborg SR, Li Z, Chan JE, Tammela T, Cellular and molecular mechanisms of plasticity in cancer. Trends Cancer Res. (2022), doi:10.1016/j.trecan.2022.04.007. 13. Flavahan WA, Gaskell E, Bernstein BE, Epigenetic plasticity and the hallmarks of cancer. Science. 357 (2017), doi:10.1126/science.aal2380. 14. Dawson MA, The cancer epigenome: Concepts, challenges, and therapeutic opportunities. Science. 355, 1147–1152 (2017). [PubMed: 28302822] 15. Xie W, Schultz MD, Lister R, Hou Z, Rajagopal N, Ray P, Whitaker JW, Tian S, Hawkins RD, Leung D, Yang H, Wang T, Lee AY, Swanson SA, Zhang J, Zhu Y, Kim A, Nery JR, Urich MA, Kuan S, Yen C-A, Klugman S, Yu P, Suknuntha K, Propson NE, Chen H, Edsall LE, Wagner U, Li Y, Ye Z, Kulkarni A, Xuan Z, Chung W-Y, Chi NC, Antosiewicz-Bourget JE, Slukvin I, Stewart R, Zhang MQ, Wang W, Thomson JA, Ecker JR, Ren B, Epigenomic analysis of multilineage differentiation of human embryonic stem cells. Cell. 153, 1134–1148 (2013). [PubMed: 23664764] 16. Gifford CA, Ziller MJ, Gu H, Trapnell C, Donaghey J, Tsankov A, Shalek AK, Kelley DR, Shishkin AA, Issner R, Zhang X, Coyne M, Fostel JL, Holmes L, Meldrim J, Guttman M, Epstein C, Park H, Kohlbacher O, Rinn J, Gnirke A, Lander ES, Bernstein BE, Meissner A, Transcriptional and epigenetic dynamics during specification of human embryonic stem cells. Cell. 153, 1149–1163 (2013). [PubMed: 23664763] 17. LaFave LM, Kartha VK, Ma S, Meli K, Del Priore I, Lareau C, Naranjo S, Westcott PMK, Duarte FM, Sankar V, Chiang Z, Brack A, Law T, Hauck H, Okimoto A, Regev A, Buenrostro JD, Jacks T, Epigenomic State Transitions Characterize Tumor Progression in Mouse Lung Adenocarcinoma. Cancer Cell. 38, 212–228.e13 (2020). [PubMed: 32707078] 18. Marjanovic ND, Hofree M, Chan JE, Canner D, Wu K, Trakala M, Hartmann GG, Smith OC, Kim JY, Evans KV, Hudson A, Ashenberg O, Porter CBM, Bejnood A, Subramanian A, Pitter K, Yan Y, Delorey T, Phillips DR, Shah N, Chaudhary O, Tsankov A, Hollmann T, Rekhtman N, Massion PP, Poirier JT, Mazutis L, Li R, Lee J-H, Amon A, Rudin CM, Jacks T, Regev A, Tammela T, Emergence of a High-Plasticity Cell State during Lung Cancer Evolution. Cancer Cell. 38, 229–246.e13 (2020). [PubMed: 32707077] 19. Storz P, Acinar cell plasticity and development of pancreatic ductal adenocarcinoma. Nat. Rev. Gastroenterol. Hepatol. 14, 296–304 (2017). [PubMed: 28270694] 20. Guerra C, Collado M, Navas C, Schuhmacher AJ, Hernández-Porras I, Cañamero M, Rodriguez- Justo M, Serrano M, Barbacid M, Pancreatitis-induced inflammation contributes to pancreatic cancer by inhibiting oncogene-induced senescence. Cancer Cell. 19, 728–739 (2011). [PubMed: 21665147] 21. Gidekel Friedlander SY, Chu GC, Snyder EL, Girnius N, Dibelius G, Crowley D, Vasile E, DePinho RA, Jacks T, Context-dependent transformation of adult pancreatic cells by oncogenic K-Ras. Cancer Cell. 16, 379–389 (2009). [PubMed: 19878870] 22. Iv JPM, Cano DA, Sekine S, Wang SC, Hebrok M, β-catenin blocks Kras-dependent reprogramming of acini into pancreatic cancer precursor lesions in mice. J. Clin. Invest. 120, 508–520 (2 2010). [PubMed: 20071774] 23. Alonso-Curbelo D, Ho Y-J, Burdziak C, Maag JLV, Morris JP 4th, Chandwani R, Chen H-A, Tsanov KM, Barriga FM, Luan W, Tasdemir N, Livshits G, Azizi E, Chun J, Wilkinson JE, Mazutis L, Leach SD, Koche R, Pe’er D, Lowe SW, A gene-environment-induced epigenetic program initiates tumorigenesis. Nature. 590, 642–648 (2021). [PubMed: 33536616] 24. Del Poggetto E, Ho I-L, Balestrieri C, Yen E-Y, Zhang S, Citron F, Shah R, Corti D, Diaferia GR, Li C-Y, Loponte S, Carbone F, Hayakawa Y, Valenti G, Jiang S, Sapio L, Jiang H, Dey P, Gao S, Deem AK, Rose-John S, Yao W, Ying H, Rhim AD, Genovese G, Heffernan TP, Maitra A, Wang TC, Wang L, Draetta GF, Carugo A, Natoli G, Viale A, Epithelial memory of inflammation limits tissue damage while promoting pancreatic tumorigenesis. Science. 373, eabj0486 (2021). [PubMed: 34529467] 25. Li Y, He Y, Peng J, Su Z, Li Z, Zhang B, Ma J, Zhuo M, Zou D, Liu X, Liu X, Wang W, Huang D, Xu M, Wang J, Deng H, Xue J, Xie W, Lan X, Chen M, Zhao Y, Wu W, David CJ, Mutant Kras co-opts a proto-oncogenic enhancer network in inflammation-induced metaplastic progenitor cells to initiate pancreatic cancer. Nat Cancer. 2, 49–65 (2021). [PubMed: 35121887] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 21 26. Kawaguchi Y, Cooper B, Gannon M, Ray M, MacDonald RJ, Wright CVE, The role of the transcriptional regulator Ptf1a in converting intestinal to pancreatic progenitors. Nat. Genet. 32, 128–134 (2002). [PubMed: 12185368] 27. See Materials and Methods. 28. Westphalen CB, Takemoto Y, Tanaka T, Macchini M, Jiang Z, Renz BW, Chen X, Ormanns S, Nagar K, Tailor Y, May R, Cho Y, Asfaha S, Worthley DL, Hayakawa Y, Urbanska AM, Quante M, Reichert M, Broyde J, Subramaniam PS, Remotti H, Su GH, Rustgi AK, Friedman RA, Honig B, Califano A, Houchen CW, Olive KP, Wang TC, Dclk1 Defines Quiescent Pancreatic Progenitors that Promote Injury-Induced Regeneration and Tumorigenesis. Cell Stem Cell. 18, 441–455 (2016). [PubMed: 27058937] 29. Sinha S, Fu Y-Y, Grimont A, Ketcham M, Lafaro K, Saglimbeni JA, Askan G, Bailey JM, Melchor JP, Zhong Y, Joo MG, Grbovic-Huezo O, Yang I-H, Basturk O, Baker L, Park Y, Kurtz RC, Tuveson D, Leach SD, Pasricha PJ, PanIN Neuroendocrine Cells Promote Tumorigenesis via Neuronal Cross-talk. Cancer Res. 77, 1868–1879 (2017). [PubMed: 28386018] 30. Rhim AD, Mirek ET, Aiello NM, Maitra A, Bailey JM, McAllister F, Reichert M, Beatty GL, Rustgi AK, Vonderheide RH, Leach SD, Stanger BZ, EMT and dissemination precede pancreatic tumor formation. Cell. 148, 349–361 (2012). [PubMed: 22265420] 31. Kopp JL, von Figura G, Mayes E, Liu F-F, Dubois CL, Morris JP, Pan FC, Akiyama H, Wright CVE, Jensen K, Hebrok M, Sander M, Identification of Sox9-Dependent Acinar-to-Ductal Reprogramming as the Principal Mechanism for Initiation of Pancreatic Ductal Adenocarcinoma. Cancer Cell. 22 (2012), pp. 737–750. [PubMed: 23201164] 32. Peng J, Sun B-F, Chen C-Y, Zhou J-Y, Chen Y-S, Chen H, Liu L, Huang D, Jiang J, Cui G-S, Yang Y, Wang W, Guo D, Dai M, Guo J, Zhang T, Liao Q, Liu Y, Zhao Y-L, Han D-L, Zhao Y, Yang Y-G, Wu W, Author Correction: Single-cell RNA-seq highlights intra-tumoral heterogeneity and malignant progression in pancreatic ductal adenocarcinoma. Cell Res. 29, 777 (2019). [PubMed: 31409908] 33. Coifman RR, Lafon S, Diffusion maps. Appl. Comput. Harmon. Anal. 21, 5–30 (2006). 34. Yang S, He P, Wang J, Schetter A, Tang W, Funamizu N, Yanaga K, Uwagawa T, Satoskar AR, Gaedcke J, Bernhardt M, Ghadimi BM, Gaida MM, Bergmann F, Werner J, Ried T, Hanna N, Alexander HR, Hussain SP, A Novel MIF Signaling Pathway Drives the Malignant Character of Pancreatic Cancer by Targeting NR3C2. Cancer Res. 76, 3838–3850 (2016). [PubMed: 27197190] 35. Moffitt RA, Marayati R, Flate EL, Volmar KE, Loeza SGH, Hoadley KA, Rashid NU, Williams LA, Eaton SC, Chung AH, Smyla JK, Anderson JM, Kim HJ, Bentrem DJ, Talamonti MS, Iacobuzio-Donahue CA, Hollingsworth MA, Yeh JJ, Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma. Nat. Genet. 47, 1168– 1178 (2015). [PubMed: 26343385] 36. Gibbons DL, Creighton CJ, Pan-cancer survey of epithelial-mesenchymal transition markers across the Cancer Genome Atlas. Dev. Dyn. 247, 555–564 (2018). [PubMed: 28073171] 37. Jahan R, Ganguly K, Smith LM, Atri P, Carmicheal J, Sheinin Y, Rachagani S, Natarajan G, Brand RE, Macha MA, Grandgenett PM, Kaur S, Batra SK, Trefoil factor(s) and CA19.9: A promising panel for early detection of pancreatic cancer. EBioMedicine. 42, 375–385 (2019). [PubMed: 30956167] 38. Roe J-S, Hwang C-I, Somerville TDD, Milazzo JP, Lee EJ, Da Silva B, Maiorino L, Tiriac H, Young CM, Miyabayashi K, Filippini D, Creighton B, Burkhart RA, Buscaglia JM, Kim EJ, Grem JL, Lazenby AJ, Grunkemeyer JA, Hollingsworth MA, Grandgenett PM, Egeblad M, Park Y, Tuveson DA, Vakoc CR, Enhancer Reprogramming Promotes Pancreatic Cancer Metastasis. Cell. 170, 875–888.e20 (2017). [PubMed: 28757253] 39. Lange M, Bergen V, Klein M, Setty M, Reuter B, Bakhti M, Lickert H, Ansari M, Schniering J, Schiller HB, Pe’er D, Theis FJ, CellRank for directed single-cell fate mapping. Nat. Methods (2022), doi:10.1038/s41592-021-01346-6. 40. Bergen V, Lange M, Peidli S, Wolf FA, Theis FJ, Generalizing RNA velocity to transient cell states through dynamical modeling. Nat. Biotechnol. 38, 1408–1414 (2020). [PubMed: 32747759] 41. La Manno G, Soldatov R, Zeisel A, Braun E, Hochgerner H, Petukhov V, Lidschreiber K, Kastriti ME, Lönnerberg P, Furlan A, Fan J, Borm LE, Liu Z, van Bruggen D, Guo J, He X, Barker R, Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 22 Sundström E, Castelo-Branco G, Cramer P, Adameyko I, Linnarsson S, Kharchenko PV, RNA velocity of single cells. Nature. 560, 494–498 (2018). [PubMed: 30089906] 42. Ioannou M, Serafimidis I, Arnes L, Sussel L, Singh S, Vasiliou V, Gavalas A, ALDH1B1 is a potential stem/progenitor marker for multiple pancreas progenitor pools. Dev. Biol. 374, 153–163 (2013). [PubMed: 23142317] 43. Mameishvili E, Serafimidis I, Iwaszkiewicz S, Lesche M, Reinhardt S, Bölicke N, Büttner M, Stellas D, Papadimitropoulou A, Szabolcs M, Anastassiadis K, Dahl A, Theis F, Efstratiadis A, Gavalas A, Aldh1b1 expression defines progenitor cells in the adult pancreas and is required for Kras-induced pancreatic cancer. Proc. Natl. Acad. Sci. U. S. A. 116, 20679–20688 (2019). [PubMed: 31548432] 44. Carrière C, Seeley ES, Goetze T, Longnecker DS, Korc M, The Nestin progenitor lineage is the compartment of origin for pancreatic intraepithelial neoplasia. Proc. Natl. Acad. Sci. U. S. A. 104, 4437–4442 (2007). [PubMed: 17360542] 45. Carrière C, Young AL, Gunn JR, Longnecker DS, Korc M, Acute pancreatitis accelerates initiation and progression to pancreatic cancer in mice expressing oncogenic Kras in the nestin cell lineage. PLoS One. 6, e27725 (2011). [PubMed: 22140463] 46. Baran Y, Bercovich A, Sebe-Pedros A, Lubling Y, Giladi A, Chomsky E, Meir Z, Hoichman M, Lifshitz A, Tanay A, MetaCell: analysis of single-cell RNA-seq data using K-nn graph partitions. Genome Biol. 20, 206 (2019). [PubMed: 31604482] 47. Persad S, Choo Z-N, Dien C, Masilionis I, Chaligné R, Nawy T, Brown CC, Pe’er I, Setty M, Pe’er D, SEACells: Inference of transcriptional and epigenomic cellular states from single-cell genomics data. bioRxiv (2022), p. 2022.04.02.486748, doi:10.1101/2022.04.02.486748. 48. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP, Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U. S. A. 102, 15545–15550 (2005). [PubMed: 16199517] 49. Pylayeva-Gupta Y, Lee KE, Hajdu CH, Miller G, Bar-Sagi D, Oncogenic Kras-induced GM-CSF production promotes the development of pancreatic neoplasia. Cancer Cell. 21, 836–847 (2012). [PubMed: 22698407] 50. Efremova M, Vento-Tormo M, Teichmann SA, Vento-Tormo R, CellPhoneDB: inferring cell- cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat. Protoc. 15, 1484–1506 (2020). [PubMed: 32103204] 51. Strobel O, Dor Y, Alsina J, Stirman A, Lauwers G, Trainor A, Castillo CF-D, Warshaw AL, Thayer SP, In vivo lineage tracing defines the role of acinar-to-ductal transdifferentiation in inflammatory ductal metaplasia. Gastroenterology. 133, 1999–2009 (2007). [PubMed: 18054571] 52. Schlesinger Y, Yosefov-Levi O, Kolodkin-Gal D, Granit RZ, Peters L, Kalifa R, Xia L, Nasereddin A, Shiff I, Amran O, Nevo Y, Elgavish S, Atlan K, Zamir G, Parnas O, Single-cell transcriptomes of pancreatic preinvasive lesions and cancer reveal acinar metaplastic cells’ heterogeneity. Nat. Commun. 11, 4516 (2020). [PubMed: 32908137] 53. Gonzalez H, Hagerling C, Werb Z, Roles of the immune system in cancer: from tumor initiation to metastatic progression. Genes Dev. 32, 1267–1284 (2018). [PubMed: 30275043] 54. Alon U, Network motifs: theory and experimental approaches. Nat. Rev. Genet. 8, 450–461 (2007). [PubMed: 17510665] 55. Das S, Shapiro B, Vucic EA, Vogt S, Bar-Sagi D, Tumor Cell-Derived IL1β Promotes Desmoplasia and Immune Suppression in Pancreatic Cancer. Cancer Res. 80, 1088–1101 (2020). [PubMed: 31915130] 56. Alam A, Levanduski E, Denz P, Villavicencio HS, Bhatta M, Alhorebi L, Zhang Y, Gomez EC, Morreale B, Senchanthisai S, Li J, Turowski SG, Sexton S, Sait SJ, Singh PK, Wang J, Maitra A, Kalinski P, DePinho RA, Wang H, Liao W, Abrams SI, Segal BH, Dey P, Fungal mycobiome drives IL-33 secretion and type 2 immunity in pancreatic cancer. Cancer Cell. 40, 153–167.e11 (2022). [PubMed: 35120601] 57. Drake LY, Kita H, IL-33: biological properties, functions, and roles in airway disease. Immunol. Rev. 278, 173–184 (2017). [PubMed: 28658560] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 23 58. Andersson P, Yang Y, Hosaka K, Zhang Y, Fischer C, Braun H, Liu S, Yu G, Liu S, Beyaert R, Chang M, Li Q, Cao Y, Molecular mechanisms of IL-33-mediated stromal interactions in cancer metastasis. JCI Insight. 3 (2018), doi:10.1172/jci.insight.122375. 59. Velez-Delgado A, Donahue KL, Brown KL, Du W, Irizarry-Negron V, Menjivar RE, Lasse Opsahl EL, Steele NG, The S, Lazarus J, Sirihorachai VR, Yan W, Kemp SB, Kerk SA, Bollampally M, Yang S, Scales MK, Avritt FR, Lima F, Lyssiotis CA, Rao A, Crawford HC, Bednar F, Frankel TL, Allen BL, Zhang Y, Pasca di Magliano M, Extrinsic KRAS signaling shapes the pancreatic microenvironment through fibroblast reprogramming. Cell. Mol. Gastroenterol. Hepatol. 13, 1673–1699 (2022). [PubMed: 35245687] 60. Dann E, Henderson NC, Teichmann SA, Morgan MD, Marioni JC, Differential abundance testing on single-cell data using k-nearest neighbor graphs. Nat. Biotechnol. (2021), doi:10.1038/ s41587-021-01033-z. 61. Setty M, Kiseliovas V, Levine J, Gayoso A, Mazutis L, Pe’er D, Characterization of cell fate probabilities in single-cell data with Palantir. Nat. Biotechnol. 37, 451–460 (2019). [PubMed: 30899105] 62. Bailey JM, Hendley AM, Lafaro KJ, Pruski MA, Jones NC, Alsina J, Younes M, Maitra A, McAllister F, Iacobuzio-Donahue CA, Leach SD, p53 mutations cooperate with oncogenic Kras to promote adenocarcinoma from pancreatic ductal cells. Oncogene. 35, 4282–4288 (2016). [PubMed: 26592447] 63. Malinova A, Veghini L, Real FX, Corbo V, Cell lineage infidelity in PDAC progression and therapy resistance. Front. Cell Dev. Biol. 9, 795251 (2021). [PubMed: 34926472] 64. Azizi E, Carr AJ, Plitas G, Cornish AE, Konopacki C, Prabhakaran S, Nainys J, Wu K, Kiseliovas V, Setty M, Choi K, Fromme RM, Dao P, McKenney PT, Wasti RC, Kadaveru K, Mazutis L, Rudensky AY, Pe’er D, Single-Cell Map of Diverse Immune Phenotypes in the Breast Tumor Microenvironment. Cell. 174, 1293–1308.e36 (2018). [PubMed: 29961579] 65. Granja JM, Corces MR, Pierce SE, Bagdatli ST, Choudhry H, Chang HY, Greenleaf WJ, ArchR is a scalable software package for integrative single-cell chromatin accessibility analysis. Nat. Genet. 53, 403–411 (2021). [PubMed: 33633365] 66. Greenwald NF, Miller G, Moen E, Kong A, Kagel A, Dougherty T, Fullaway CC, McIntosh BJ, Leow KX, Schwartz MS, Pavelchek C, Cui S, Camplisson I, Bar-Tal O, Singh J, Fong M, Chaudhry G, Abraham Z, Moseley J, Warshawsky S, Soon E, Greenbaum S, Risom T, Hollmann T, Bendall SC, Keren L, Graf W, Angelo M, Van Valen D, Whole-cell segmentation of tissue images with human-level performance using large-scale data annotation and deep learning. Nat. Biotechnol. 40, 555–565 (2022). [PubMed: 34795433] 67. Krah NM, De La O J-P, Swift GH, Hoang CQ, Willet SG, Chen Pan F, Cash GM, Bronner MP, Wright CV, MacDonald RJ, Murtaugh LC, The acinar differentiation determinant PTF1A inhibits initiation of pancreatic ductal adenocarcinoma. Elife. 4 (2015), doi:10.7554/eLife.07125. 68. Schaefer CF, Anthony K, Krupa S, Buchoff J, Day M, Hannay T, Buetow KH, PID: the Pathway Interaction Database. Nucleic Acids Res. 37, D674–9 (2009). [PubMed: 18832364] 69. van Dijk D, Sharma R, Nainys J, Yim K, Kathail P, Carr AJ, Burdziak C, Moon KR, Chaffer CL, Pattabiraman D, Bierie B, Mazutis L, Wolf G, Krishnaswamy S, Pe’er D, Recovering Gene Interactions from Single-Cell Data Using Data Diffusion. Cell. 174, 716–729.e27 (2018). [PubMed: 29961576] 70. Saborowski M, Saborowski A, Morris JP 4th, Bosbach B, Dow LE, Pelletier J, Klimstra DS, Lowe SW, A modular and flexible ESC-based mouse model of pancreatic cancer. Genes Dev. 28, 85–97 (2014). [PubMed: 24395249] 71. Fellmann C, Hoffmann T, Sridhar V, Hopfgartner B, Muhar M, Roth M, Lai DY, Barbosa IAM, Kwon JS, Guan Y, Sinha N, Zuber J, An optimized microRNA backbone for effective single-copy RNAi. Cell Rep. 5, 1704–1713 (2013). [PubMed: 24332856] 72. Dow LE, Premsrirut PK, Zuber J, Fellmann C, McJunkin K, Miething C, Park Y, Dickins RA, Hannon GJ, Lowe SW, A pipeline for the generation of shRNA transgenic mice. Nat. Protoc. 7, 374–393 (2012). [PubMed: 22301776] 73. Gertsenstein M, Nutter LMJ, Reid T, Pereira M, Stanford WL, Rossant J, Nagy A, Efficient generation of germ line transmitting chimeras from C57BL/6N ES cells by aggregation with outbred host embryos. PLoS One. 5, e11260 (2010). [PubMed: 20582321] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 24 74. Jackson EL, Willis N, Mercer K, Bronson RT, Crowley D, Montoya R, Jacks T, Tuveson DA, Analysis of lung tumor initiation and progression using conditional expression of oncogenic K-ras. Genes Dev. 15, 3243–3248 (2001). [PubMed: 11751630] 75. Hingorani SR, Wang L, Multani AS, Combs C, Deramaudt TB, Hruban RH, Rustgi AK, Chang S, Tuveson DA, Trp53R172H and KrasG12D cooperate to promote chromosomal instability and widely metastatic pancreatic ductal adenocarcinoma in mice. Cancer Cell. 7, 469–483 (2005). [PubMed: 15894267] 76. Beard C, Hochedlinger K, Plath K, Wutz A, Jaenisch R, Efficient method to generate single-copy transgenic mice by site-specific integration in embryonic stem cells. genesis. 44 (2006), pp. 23–28. [PubMed: 16400644] 77. Dow LE, Nasr Z, Saborowski M, Ebbesen SH, Manchado E, Tasdemir N, Lee T, Pelletier J, Lowe SW, Conditional reverse tet-transactivator mouse strains for the efficient induction of TRE- regulated transgenes in mice. PLoS One. 9, e95236 (2014). [PubMed: 24743474] 78. Moffitt JR, Bambah-Mukku D, Eichhorn SW, Vaughn E, Shekhar K, Perez JD, Rubinstein ND, Hao J, Regev A, Dulac C, Zhuang X, Molecular, spatial, and functional single-cell profiling of the hypothalamic preoptic region. Science. 362, eaau5324 (2018). [PubMed: 30385464] 79. Moffitt JR, Hao J, Wang G, Chen KH, Babcock HP, Zhuang X, High-throughput single-cell gene-expression profiling with multiplexed error-robust fluorescence in situ hybridization. Proc. Natl. Acad. Sci. U. S. A. 113, 11046–11051 (2016). [PubMed: 27625426] 80. Wang G, Moffitt JR, Zhuang X, Multiplexed imaging of high-density libraries of RNAs with MERFISH and expansion microscopy. Sci. Rep. 8, 4847 (2018). [PubMed: 29555914] 81. Farack L, Itzkovitz S, Protocol for single-molecule fluorescence in situ hybridization for intact pancreatic tissue. STAR Protoc. 1, 100007 (2020). [PubMed: 33111069] 82. Moffitt JR, Hao J, Bambah-Mukku D, Lu T, Dulac C, Zhuang X, High-performance multiplexed fluorescence in situ hybridization in culture and tissue with matrix imprinting and clearing. Proc. Natl. Acad. Sci. U. S. A. 113, 14456–14461 (2016). [PubMed: 27911841] 83. Lin J-R, Fallahi-Sichani M, Sorger PK, Highly multiplexed imaging of single cells using a high-throughput cyclic immunofluorescence method. Nat. Commun. 6, 8390 (2015). [PubMed: 26399630] 84. Chen KH, Boettiger AN, Moffitt JR, Wang S, Zhuang X, RNA imaging. Spatially resolved, highly multiplexed RNA profiling in single cells. Science. 348, aaa6090 (2015). [PubMed: 25858977] 85. Levine JH, Simonds EF, Bendall SC, Davis KL, Amir E-AD, Tadmor MD, Litvin O, Fienberg HG, Jager A, Zunder ER, Finck R, Gedman AL, Radtke I, Downing JR, Pe’er D, Nolan GP, Data-Driven Phenotypic Dissection of AML Reveals Progenitor-like Cells that Correlate with Prognosis. Cell. 162, 184–197 (2015). [PubMed: 26095251] 86. Satpathy AT, Granja JM, Yost KE, Qi Y, Meschi F, McDermott GP, Olsen BN, Mumbach MR, Pierce SE, Corces MR, Shah P, Bell JC, Jhutty D, Nemec CM, Wang J, Wang L, Yin Y, Giresi PG, Chang ALS, Zheng GXY, Greenleaf WJ, Chang HY, Massively parallel single-cell chromatin landscapes of human immune cell development and intratumoral T cell exhaustion. Nat. Biotechnol. 37, 925–936 (2019). [PubMed: 31375813] 87. Gayoso A, Shor J, JonathanShor/DoubletDetection: doubletdetection v3.0 (2020; https:// zenodo.org/record/4359992). 88. Smillie CS, Biton M, Ordovas-Montanes J, Sullivan KM, Burgin G, Graham DB, Herbst RH, Rogel N, Slyper M, Waldman J, Sud M, Andrews E, Velonias G, Haber AL, Jagadeesh K, Vickovic S, Yao J, Stevens C, Dionne D, Nguyen LT, Villani A-C, Hofree M, Creasey EA, Huang H, Rozenblatt-Rosen O, Garber JJ, Khalili H, Nicole Desch A, Daly MJ, Ananthakrishnan AN, Shalek AK, Xavier RJ, Regev A, Intra- and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis. Cell. 178 (2019), pp. 714–730.e22. [PubMed: 31348891] 89. van der Maaten L, Accelerating t-SNE using Tree-Based Algorithms. J. Mach. Learn. Res. 15, 3221–3245 (2014). 90. Nowotschin S, Setty M, Kuo Y-Y, Liu V, Garg V, Sharma R, Simon CS, Saiz N, Gardner R, Boutet SC, Church DM, Hoodless PA, Hadjantonakis A-K, Pe’er D, The emergent landscape of the mouse gut endoderm at single-cell resolution. Nature. 569, 361–367 (2019). [PubMed: 30959515] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 25 91. Lun ATL, Bach K, Marioni JC, Pooling across cells to normalize single-cell RNA sequencing data with many zero counts. Genome Biol. 17, 75 (2016). [PubMed: 27122128] 92. Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, Liu XS, Model-based analysis of ChIP-Seq (MACS). Genome Biol. 9, R137 (2008). [PubMed: 18798982] 93. Love MI, Huber W, Anders S, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). [PubMed: 25516281] 94. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Others, Scikit-learn: Machine learning in Python. the Journal of machine Learning research. 12, 2825–2830 (2011). 95. Li H, Tabix: fast retrieval of sequence features from generic TAB-delimited files. Bioinformatics. 27, 718–719 (2011). [PubMed: 21208982] 96. Papailiopoulos D, Kyrillidis A, Boutsidis C, “Provable deterministic leverage score sampling” in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining (Association for Computing Machinery, New York, NY, USA, 2014), KDD ‘14, pp. 997–1006. 97. Kanehisa M, Goto S, KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 28, 27–30 (2000). [PubMed: 10592173] 98. Hagberg A, Swart P, S Chult D, “Exploring network structure, dynamics, and function using networkx” (LA-UR-08–05495; LA-UR-08–5495, Los Alamos National Lab. (LANL), Los Alamos, NM (United States), 2008), (available at https://www.osti.gov/biblio/960616). 99. Lancichinetti A, Radicchi F, Ramasco JJ, Fortunato S, Finding statistically significant communities in networks. PLoS One. 6, e18961 (2011). [PubMed: 21559480] 100. Steele NG, Carpenter ES, Kemp SB, Sirihorachai VR, The S, Delrosario L, Lazarus J, Amir E-AD, Gunchick V, Espinoza C, Bell S, Harris L, Lima F, Irizarry-Negron V, Paglia D, Macchia J, Chu AKY, Schofield H, Wamsteker E-J, Kwon R, Schulman A, Prabhu A, Law R, Sondhi A, Yu J, Patel A, Donahue K, Nathan H, Cho C, Anderson MA, Sahai V, Lyssiotis CA, Zou W, Allen BL, Rao A, Crawford HC, Bednar F, Frankel TL, Pasca di Magliano M, Multimodal Mapping of the Tumor and Peripheral Blood Immune Landscape in Human Pancreatic Cancer. Nat Cancer. 1, 1097–1112 (2020). [PubMed: 34296197] Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 26 Figure 1. A single-cell transcriptomic atlas of pancreatic regeneration and tumorigenesis. (A) Experimental design for tissue collection. GEMMs expressing Ptf1a-Cre enable FACS- based enrichment of mKate2-labeled exocrine pancreas epithelial cells (23). mKate2+ cells were isolated from wild-type Kras mice before injury with caerulein (N1) or 48 hours post-injury (N2); and from KrasG12D mice (KC genotype) before injury (K1), and 24–48 hours (K2) or 3 weeks after caerulein (K3, PanIN stage), as well as uninjured older KC mice (K4). PDAC primary tumors (K5) and liver and lung metastases (K6) were harvested from KC mice with a p53 floxed (p53fl/+) or mutant (p53R172H/+) allele (KPC genotype). Mouse illustration was created with BioRender (https://biorender.com/). (B) tSNE visualization of pancreatic epithelial scRNA-seq profiles from all collected stages (n = 17 mice), colored Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 27 as in (A) and labeled by cell-state (27). ADM denotes cells undergoing acinar-to-ductal metaplasia (31), and ‘Bridge’ denotes cells between acinar-like and malignant programs, which express genes from both. (C) Expression of PDAC-associated gene sets (rows) across all pancreatic epithelial cells (columns) (34, 35). Cells are ordered by the first diffusion component (DC1), representing the major axis of progression from normal (N1) to metastatic (K6) states. Plot at top displays frequency (from 0 to 1) of cells per stage, in bins of n = 2000 cells. Gene set score for each cell is computed as the average of log-normalized expression, z-scored for each gene to obtain a comparable scale. Heatmap is standardized to compare cells within each gene set. (D) tSNE plots as in (B), with pre-malignant (K1–K4) Kras-mutant cells colored by the expression of genes (from left to right) upregulated in bulk RNA-seq of Kras-mutant (Kras) pancreas relative to normal (67), associated with Myc activity (68), EMT (36), or down-regulated upon Ptf1a knockout (67). Colors are scaled from 5th to 95th percentile of expression. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 28 Figure 2. Differential epigenetic priming of Kras-mutant cells. (A) Force-directed layout (FDL) of all Kras-mutant scRNA-seq profiles (K1–K6, n = 11 mice). Cells colored by stage as in Fig. 1A. Stars highlight ‘apex’ states inferred by CellRank (39) (see Fig. S3B). (B) Principal component analysis (PCA) of bulk ATAC-seq profiles from pancreatic epithelial cells. Each point shows the position of a single biological replicate (individual mouse), colored by stage as in (A). Arrows indicate a transition upon injury and Kras mutation (N1-N2, K1-K2; green arrow) and a divergence between benign neoplastic (K3-K4; pink arrow) and malignant (K5-K6; purple arrow) stages. (C) Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 29 Left: Chromatin accessibility along progression. Subsets of differentially accessible ATAC- seq peaks (rows) are organized into three modules by clustering (27); bulk ATAC-seq replicates (columns) are ordered and colored by stage as in (A). Peaks organize into distinct accessibility patterns, denoted as chromatin modules (27). Right: Expression of genes corresponding to chromatin accessibility modules in pre-neoplastic cells (K1, K2). FDL map as in (A), colored by module expression score computed by z-scoring each cell to emphasize dominant gene programs per cell, and averaging genes nearest to module peaks. Color (expression scores) are scaled between the 40th and 90th percentiles. (D) Probability of classifying pre-neoplastic cells (K1, K2) as more similar to benign neoplastic (K3-K4) or malignant (K5-K6) scRNA-seq profiles, based on expression similarity. Sampled cells (rows) are ordered from highest benign fate probability (top) to highest malignant fate probability (bottom); bars represent probability of classification from 0 to 1 to K3, K4, K5, or K6 labels, colored as in (A). A fraction of cells exhibit composite states with probability for both fates. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 30 Figure 3. Kras-mutant cells display elevated epigenetic plasticity, which is associated with cell- cell communication propensity. (A) FDL of scATAC-seq profiles from Kras-mutant epithelial cells from pre-malignant (K1– K3) and malignant (K5) stages (n = 9 mice), colored by stage. (B) Frequency of cells from each stage along second high-variance component from latent semantic indexing (LSI) of scATAC-seq profiles (65). (C) Pairwise Pearson correlation coefficients of metacells from scATAC-seq ArchR gene accessibility scores (rows) and scRNA-seq expression values (columns). Annotated cell-states, determined by refined PhenoGraph clustering of scRNA- Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 31 seq (Fig. S1C) and scATAC-seq (Fig. S5A), are colored according to their annotation as in labels from (A). Blocks of positive correlation along diagonal represent similar cell-states across the two modalities, whereas off-diagonal correlations indicate similarity across distinct cell-states. (D) Cartoon of classifier-based approach to quantify plasticity (27). (E) Classifier confusion matrix based on procedure in (D). Cell-states, determined by scRNA-seq (Fig. S1C) and scATAC-seq (Fig. S5A) metacell clusters, are colored as in labels in (A). Values represent number of metacells from an epigenomic cluster that classify to a transcriptomic cluster, normalized within each row. Dashed box highlights high plasticity epigenomic states. (F) Plasticity scores for epigenomic clusters in (E). Boxes represent interquartile range (IQR) of plasticity scores for all epigenomic metacells assigned to that cluster, computed as per-cell Shannon entropy in the classifier’s predicted probability distribution across transcriptomic states. Lines represent medians and whiskers represent 1.5x IQR. (G) GSEA plot based on Spearman rank correlation between plasticity score and each gene’s accessibility score. (H) Plasticity scores for epigenomic metacells from K1 and K2, showing significant increase in plasticity in K2 (one-tailed t-test; t = 2.5511, p value = 0.006). (I) Immunohistochemistry of CD45 (brown) marking immune cells in K1 and K2 tissue, showing increase in immune infiltrate in response to injury. Scale bar, 200 μm. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 32 Figure 4. Inferred epithelial-immune crosstalk in plastic neoplastic states. (A) Calligraphy-inferred ‘communication modules’ in pre-malignant Kras-mutant epithelial cells (K1-K3, n = 6 mice). Each row or column represents one receptor or ligand; value at intersection indicates correlation in expression (Pearson r) of that gene pair across pre- malignant cells. Blocks of highly correlated genes denote partially overlapping modules (annotated at right) that tend to co-express in the same cell-states. Schematic (far right) describes the second step of Calligraphy (see Fig. S11A). (B) FDL of K1–K3 epithelial cells with color values based on relative communication module gene expression (27). (C) FDL Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 33 of KrasWT pancreas cells before and after injury (N1-N2, n = 4 mice), colored by K1–K3 communication module expression as in (B) (top) or Kras mutant signature gene expression (bottom, (23)), scaled between 1st and 99th percentile. (D) FDL of malignant cells (K5- K6, n = 3 mice), colored as in (B). (E) Communication module expression in human pancreatic tumor scRNA-seq data (32), colored as in (B). (F) Pairwise crosstalk between communication modules inferred by Calligraphy from epithelial or immune scRNA-seq data (one module per row or column), colored gray for immune or as in (A) for epithelial modules. Heat values represent number of inferred cognate R-L pair interactions across each communicating module pair; some contributing receptors or ligands are shown at right. Bars quantify total inferred edge counts, representing remodeling (row) or sensing (column) interactions for that module. (G) Two smFISH fields of view reveal the spatial proximity of sending (magenta box) to receiving (green box) mKate2+ epithelial cells. The expression of two Gastric (E6) module ligands (cyan and red), as well as two Progenitor (E7) receptors (magenta and green) overlap spatially in these three examples. Scale bars, 20 μm. ( H) Distance between each receiving progenitor cell (Il18r1hi Cd44hi) and double- positive sending gastric cell (Il18hi Spp1hi), versus randomly selected non-sending gastric cells (Il18lo Spp1lo). Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 34 Figure 5. Kras-mutant epithelial states participate in a feedback loop with immune populations. (A) A feedback loop identified by Calligraphy in the pre-malignant pancreas. Arrows depict cognate R-L interactions. (B) tSNEs of immune and epithelial scRNA-seq data from pre-malignant stages (K1–K3, n = 6 mice), displaying imputed expression (69) of key genes from the loop in (A). Arrows between plots indicate sequential steps of the loop. (C) Co-immunofluorescence (co-IF) images showing co-expression of IL-33 and E-cadherin (epithelial marker), and apposition of FOXP3-expressing Tregs (arrows) and IL-33-expressing epithelial cells. Scale bar, 10 μm. ( D) Distance in pixels (0.325 μm per pixel) of IL-33+ epithelial cells to Tregs against a null model of spatial distribution in Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 35 co-IF data pooled across all biological replicates from K2 tissue. Distances are calculated between each IL-33+ epithelial (E-cadherin+) cell and its closest Treg (CD3+ FOXP3+). Asterisks, significant difference (one-tailed, un-paired t-test, p value < 0.0001) compared to random distances calculated by permuting epithelial cells. (E) IL-33-centric crosstalk paths originating from epithelial Gastric module E6 (central circle, magenta), with each outward concentric circle illustrating possible communication paths from inner to outer modules based on links inferred by Calligraphy. Arc length is proportional to the number of inner- module ligands that can bind to cognate receptors in the outer module. (F) tSNE as in (B), colored according to the step in which communication events from the IL-33-centric path in (E) reach the module expressed by that cell. Cells are assigned to their highest-expressed module, and each module is scored by the earliest step in which it appears along any paths through the Calligraphy network emanating from E6-derived IL-33. Cells expressing modules which are not downstream of IL-33 are colored in gray. Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 36 Figure 6. Spatiotemporal in vivo perturbation of Il33 impairs neoplastic progression. (A) Mouse models for inducible repression of Il33 (KC-shIl33, 2 independent strains) or Renilla control (KC-shRen), restricted to Kras-mutant epithelial cells by Ptf1a-Cre expression. (B) Representative IF of pancreata from control (top) or KC-shIl33 (bottom) mice placed on dox at 5 weeks of age and analyzed 9 days later at the 48 hpi timepoint (K2). Kras-mutant epithelial cells, not surrounding stroma, express Il33 shRNA marked by GFP in KC-shIl33; TFF1 marks cell-state in which IL-33 is activated at 48 hpi in control but not shIl33 animals. Dashed lines demark epithelium-stroma boundary, asterisks Science. Author manuscript; available in PMC 2023 July 03. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Burdziak et al. Page 37 highlight suppression of Il33 in TFF1+ metaplastic cells of KC-shIl33 mice. DAPI marks nuclei (blue). Scale bar, 20 μm. ( C) Milo (60) log fold change (logFC) magnitudes across cell neighborhoods (n = 5 mice), with higher values indicating greater impact of IL-33 perturbation. Rightward distribution shifts (dotted lines) indicate a larger impact on particular cell-states; vertical dashed lines indicate neighborhoods with significant (adjusted p < 0.1) shifts according to Milo, appearing only in K3 epithelia. (D) FDL of Milo neighborhoods colored by logFC of abundance in shIl33 samples relative to controls, at the late (3 wpi) timepoint. (E,F) Representative IF in pancreata from KC-shIl33 mice placed on-dox (bottom) or off-dox (top) at 3 wpi (K3), showing (E) aberrant accumulation of progenitor-like state (MSN+) in epithelial cells (E-cadherin+) of IL-33-perturbed animals at 48 hpi and (F) depletion of gastric-like (AGR2+) states upon epithelial IL-33 suppression. DAPI marks nuclei (blue). Scale bar, 100 μm. ( G) Impacts of Il33 perturbation across Kras- mutant epithelial neighborhoods at K3 (3 wpi). Top, pseudotime-ordered neighborhoods (columns) colored by cell-state. Middle, neighborhoods plotted and colored by Milo logFC; higher logFC denotes greater abundance in shIl33 relative to control. Bottom, Nes and plasticity-associated gene expression (Fig. 3F) (27); heatmap colors scaled to ±2 s.d. from mean. (H) Milo logFC of neighborhoods mapped to modules that are (left) or are not (right) downstream of Calligraphy’s IL-33-centric network; asterisks, indicate significance (unpaired, one-tailed t-test, p value = 1.24 X 10−7). Science. Author manuscript; available in PMC 2023 July 03.
10.1126_sciimmunol.ade2860	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Sci Immunol. Author manuscript; available in PMC 2023 July 21. Published in final edited form as: Sci Immunol. 2023 April 21; 8(82): eade2860. doi:10.1126/sciimmunol.ade2860. Encephalitis and poor neuronal death-mediated control of herpes simplex virus in human inherited RIPK3 deficiency Zhiyong Liu1, Eduardo J. Garcia Reino1,#, Oliver Harschnitz2,3,#, Hongyan Guo4,5,6,#, Yi-Hao Chan1, Noopur Khobrekar2, Mary L. Hasek1, Kerry Dobbs7, Darawan Rinchai1, Marie Materna8,9, Daniela Matuozzo8,9, Danyel Lee1,8,9, Paul Bastard1,8,9,10, Jie Chen1, Yoon Seung Lee1, Seong K. Kim5, Shuxiang Zhao1, Param Amin2, Lazaro Lorenzo8,9, Yoann Seeleuthner8,9, Remi Chevalier8,9, Laure Mazzola11, Claire Gay11, Jean-Louis Stephan11, Baptiste Milisavljevic1, Soraya Boucherit8,9, Flore Rozenberg12, Rebeca Perez de Diego13,14,15, Richard D. Dix16,17, Nico Marr18,19, Vivien Béziat1,8,9, Aurelie Cobat1,8,9, Mélodie Aubart8,20, Laurent Abel1,8,9, Stephane Chabrier11, Gregory A. Smith21,&, Luigi D. Notarangelo7,&, Edward S. Mocarski4,&, Lorenz Studer2,&, Jean-Laurent Casanova1,8,9,22,23,,@, Shen-Ying Zhang1,8,9,,@ 1.St. Giles Laboratory of Human Genetics of Infectious Diseases, Rockefeller Branch, The Rockefeller University, New York, NY, USA. 2.The Center for Stem Cell Biology, Sloan Kettering Institute for Cancer Research, New York, NY, USA. 3.Human Technopole, Viale Rita Levi-Montalcini, Milan, Italy, EU. 4.Department of Microbiology and Immunology, Emory Vaccine Center, Emory University, GA, USA. 5.School of Medicine, Atlanta, GA, USA. 6.Louisiana State University Health Sciences Center at Shreveport (LSUHSC-S), Shreveport, Louisiana, USA. 7.Laboratory of Clinical Immunology and Microbiology, National Institute of Allergy and Infectious Diseases, NIH, Bethesda, MD, USA. 8.Laboratory of Human Genetics of Infectious Diseases, Necker Branch, INSERM U1163, Necker Hospital for Sick Children, Paris, France, EU. 9.Paris City University, Imagine Institute, Paris, France, EU. This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. @Corresponding authors. Author contributions: Z.L., E.J.G.R., O.H., H.G., Y.-H.C., N.K., M.L.H., K.D., M.M., D.L., P.B., J.C., Y.S.L., S.K.K., S.Z., L.L., R.C., F.R., R.P.D., R.D.D., N.M., V.B., G.A.S., L.D.N., E.S.M., L.S., S.-Y.Z., and J.-L.C. performed or supervised experiments, generated and analyzed data, and contributed to the manuscript by providing figures and tables. D.R., D.M., Y.S., B.M., A.C. and L.A. performed computational analysis of data. L.M., C.G., J.-L.S., S.B. and S.C. evaluated and recruited patients. Z.L, J.-L.C and S.-Y.Z., wrote the manuscript. J.-L.C. and S.-Y.Z. conceptualized and supervised the project. All the authors edited the manuscript. #, &, equal contributions. Competing interests: Gregory A. Smith discloses a significant financial interest in Thyreos, Inc. All other authors declare no competing interests. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 2 10.Pediatric Hematology-Immunology and Rheumatology Unit, Necker Hospital for Sick Children, AP-HP, Paris, France, EU 11.Department of Pediatrics, CHU Saint-Etienne, Saint-Etienne, Paris, France, EU. 12.Laboratory of Virology, Assistance Publique-Hôpitaux de Paris (AP-HP), Cochin Hospital, Paris, France, EU. 13.Laboratory of Immunogenetics of Human Diseases, IdiPAZ Institute for Health Research, La Paz Hospital, Madrid, Spain, EU. 14.Innate Immunity Group, IdiPAZ Institute for Health Research, La Paz Hospital, Madrid, Spain, EU. 15.Interdepartmental Group of Immunodeficiencies, Madrid, Spain, EU. 16.Viral Immunology Center, Department of Biology, Georgia State University, Atlanta, GA, USA. 17.Department of Ophthalmology, Emory University School of Medicine, Atlanta, GA, USA. 18.Research Branch, Sidra Medicine, Doha, Qatar. 19.Institute of Translational Immunology, Brandenburg Medical School, Brandenburg an der Havel, Germany College of Health and Life Sciences, Hamad Bin Khalifa University, Doha, Qatar 20.Pediatric Neurology Department, Necker Hospital for Sick Children, APHP, Paris-City University, France, EU. 21.Department of Microbiology-Immunology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA. 22.Department of Pediatrics, Necker Hospital for Sick Children, Paris, France, EU. 23.Howard Hughes Medical Institute, New York, USA. Abstract Inborn errors of TLR3-dependent type I IFN immunity in cortical neurons underlie forebrain herpes simplex virus-1 (HSV-1) encephalitis (HSE) due to uncontrolled viral growth and subsequent cell death. We report an otherwise healthy patient with HSE who was compound heterozygous for nonsense (R422) and frameshift (P493fs9) RIPK3 variants. Receptor interacting protein kinase 3 (RIPK3) is a ubiquitous cytoplasmic kinase regulating cell death outcomes, including apoptosis and necroptosis. In vitro, the R422 and P493fs9* RIPK3 proteins impaired cellular apoptosis and necroptosis upon TLR3, TLR4 or TNFR1 stimulation, and ZBP1/DAI-mediated necroptotic cell death following HSV-1 infection. The patient’s fibroblasts displayed no detectable RIPK3 expression. Following TNFR1 or TLR3 stimulation, the patient’s cells did not undergo apoptosis or necroptosis. Following HSV-1 infection, the cells supported excessive viral growth despite normal induction of antiviral IFN-β and interferon-stimulated genes (ISGs). This phenotype was, nevertheless, rescued by application of exogenous type I IFN. The patient’s human pluripotent stem cell (hPSC)-derived cortical neurons displayed impaired cell death and enhanced viral growth following HSV-1 infection, as did isogenic RIPK3-knockout hPSC-derived cortical neurons. Inherited RIPK3 deficiency therefore confers a predisposition to Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 3 HSE, by impairing the cell death-dependent control of HSV-1 in cortical neurons independently of type I IFN immunity. One-sentence summary: Autosomal recessive RIPK3 deficiency impairs cell death-dependent control of HSV-1 in cortical neurons, underlying HSV-1 encephalitis. Introduction Herpes simplex virus-1 (HSV-1) is an enveloped double-stranded DNA virus that infects most of the human population (1, 2). Infection is typically benign and limited to recurrent labial lesions, but this virus can cause life-threatening encephalitis in rare patients (3). HSV-1 encephalitis (HSE) occurs at a rate of ~2–4 cases per million inhabitants per year (4–8), corresponding to a prevalence of ~1–2 per 10,000 infected people worldwide. HSE is the most common sporadic (i.e. non-epidemic) encephalitis in the Western world. It can strike at any age, but there are two major peaks in incidence, the first at an age of six months to three years, corresponding essentially to disease following primary infection, and the second at an age of >50 years, probably following a viral reactivation event (7, 8). HSV-1 invades the central nervous system (CNS) via the olfactory bulb to cause forebrain herpes simplex encephalitis (HSE) (~95% of the patients) or, more rarely, via the trigeminal nerve to cause brainstem HSE (~5% of patients) (9, 10). The clinical manifestations depend on the location and extent of lesions, with presentations including fever, seizures, and altered consciousness (3). Remarkably, HSE is not accompanied by the dissemination of the virus to the blood and other organs. Patients do not even display the benign mucocutaneous labial lesions typically caused by HSV-1 in the general population. If untreated, mortality from this disease exceeds 70%. The advent of acyclovir in the 1980s decreased mortality rates to ~20% (11). Despite treatment, 40%-60% of survivors suffer from neurological sequelae, which are severe in about 10%-20% of cases (10, 12, 13). HSV was first identified as the causal agent of HSE in 1941 (14), and HSV-1 was distinguished from HSV-2 in 1977 (15). The pathogenesis of HSE, apart from its viral etiology, long remained unexplained, as the virus remains in the peripheral sensory neural tissues in most infected individuals. Childhood HSE is no more common in children with inherited or acquired deficits of leukocytes, including a complete lack of T cells or B cells, or both, than in children with no such deficits (16). However, since 2006, autosomal recessive (AR) and autosomal dominant (AD) monogenic inborn errors of cell-intrinsic immunity have been identified as contributing to HSE susceptibility in children. Germline mutations of genes governing the TLR3 pathway (TLR3, UNC93B1, TRIF, TRAF3, TBK1, IRF3, NEMO) or the IFNAR1 pathway (IFNAR1, TYK2, STAT1, STAT2) have been implicated in this disease (17–29). The mechanism of HSE has been clarified through studies of human stem cell (hPSC)-derived peripheral and CNS cells. Mutations affecting the two connected pathways impair cortical neuron and oligodendrocyte cell-intrinsic type I IFN immunity to HSV-1 (30–32). TLR3 pathway gene mutations impair tonic and dsRNA- inducible levels of type I IFNs, whereas IFNAR1 pathway gene mutations impair cellular responses to type I IFNs. Mutations of SNORA31 and DBR1 have recently been shown Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 4 to underlie forebrain HSE and brainstem viral encephalitis due to infection with HSV-1 or other viruses, respectively (33, 34). Overall, HSE appears to result from inborn errors of CNS-resident cell-intrinsic antiviral immunity in at least 5–10% of children with this rare disease. The clinical penetrance of these genetic defects is incomplete (8, 35), consistent with the sporadic nature of HSE (36). We hypothesized that additional unknown inborn errors of immunity might underlie HSE in other children. Results Biallelic predicted loss-of-function RIPK3 mutations in a patient with HSE We studied a girl (patient 1, P1) born to healthy non-consanguineous parents originating from and living in France (fig. 1A). She was diagnosed with independent episodes of HSE at 6 and 17 months of age, and with autoimmune encephalitis (AE) one month after the second episode of HSE (fig. 1B, S1A,B and Supplementary clinical report). This girl was an only child. She had no history of any other severe infectious disease before or after these HSE episodes, despite having been infected with many other viruses in the past, as demonstrated by viral serology studies, and she remained otherwise healthy. Deep immunophenotyping by mass cytometry (CyTOF) revealed no obvious abnormalities in blood leukocyte subsets (fig. S1C). We performed whole-exome sequencing (WES) on the patient and both her parents. In P1, we searched for rare (minor allele frequency (MAF) < 0.01 in the Genome Aggregation Database (gnomAD) and in our in-house WES database containing ~15,000 exomes) non- synonymous or essential-splicing variants with a combined annotation-dependent depletion (CADD) score (37) higher than the mutation significance cutoff (MSC) (38), for genes with a gene damage index (GDI) below 13.83 (39). We considered de novo and biallelic variants, in accordance with the guidelines for single-patient genetic studies (40). We detected 20 de novo variants, none of which was deemed a plausible candidate based on biochemical nature or the expression of the corresponding genes or the function of their products (Table S1). This search also identified six genes for which P1 was homozygous or compound- heterozygous for variants (fig. S1D, Table S2), including two heterozygous mutations of RIPK3 predicted to be loss-of-function (pLOF): p. Arg422* (c.1264 C>T, MAF 0.001568, CADD 35) and p. Pro493fs9* (c.1475 C>CC, MAF 0.002611, CADD 24.2). RIPK3 encodes receptor-interacting protein kinase 3 (RIPK3). Sanger sequencing of DNA from the patient and her parents confirmed both mutations in the patient and showed that Arg422* (R422) was inherited from the mother and Pro493fs9 (P493fs9) from the father (fig. S1E). Population genetics of RIPK3 The human RIPK3 gene is under modest negative selection, with a CoNeS of 0.8, consistent with an AR predisposition to life-threatening disease (fig. 1C, S1F). Only one homozygous carrier each was found for the R422 and P493fs9* variants among the 141,456 sequenced individuals from the gnomAD database (41). No other homozygous or compound heterozygous carrier of rare non-synonymous variants was found in our in-house exome database of data for ~15,000 patients with various infectious diseases. Only seven non-synonymous variants (MAF<0.01) were found in the homozygous state (11 homozygous carriers in total) in public databases (gnomAD and BRAVO, fig. 1C, Table S3). The rarity of homozygous carriers of these variants in the general population Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 5 suggests that RIPK3 has an important function and is consistent with the low prevalence and incomplete penetrance of HSE. RIPK3 is a serine/threonine protein kinase involved in the extrinsic apoptosis and necroptosis death pathways (42–49) shown to be involved in host defense against herpesviruses, including HSV-1, both in cell culture and in mice (47, 50, 51). RIPK3 carries an N-terminal (N-ter) kinase domain and a C-terminal (C-ter) RIP homotypic interaction motif (RHIM) (fig. 1D), both of which are required for RIPK3- mediated necroptosis in response to various stimuli, including TNF and dsRNAs (52, 53). The RHIM is also required for RIPK3-mediated apoptosis (54). RIPK3 can homodimerize, oligomerize, or heterodimerize with RIPK1 via the RHIM domains of the two proteins, and the resulting RIPK3 homodimer and RIPK1/RIPK3 complex play different roles in the activation of necroptotic and apoptotic cell death pathways (54). The two pLOF variants of the patient are predicted to have different impacts on protein function, with R422* predicted to encode a protein with no RHIM domain, and P493fs9* predicted to encode a protein with an intact RHIM domain but lacking 26 C-terminal amino acids (a.a.), the impact of this deletion on RIPK3 function being unknown (fig. 1D). We hypothesized that compound heterozygosity for R422* and P493fs9* compromised RIPK3 function in this French child with recurrent HSE. Production of the R422* and P493fs9* RIPK3 proteins following transient transfection of HeLa and HEK293T cells RIPK3 is a nucleocytoplasmic protein located predominantly in the cytosol in the steady state (55, 56). We found that both the R422* and P493fs9* RIPK3 proteins were located in the cytoplasm, like the wild-type (WT) RIPK3, following plasmid-mediated transient expression in HeLa cells (fig. 2A). Similar levels of RIPK3 mRNAs were detected following transient expression of C-ter (C-terminus) Myc-tagged R422, P493fs9 and WT RIPK3 cDNAs in HEK293T cells, which have very low levels of endogenous RIPK3 (fig. 2B). Western blotting with an antibody (Ab) against N-ter (N-terminus) RIPK3, C-ter RIPK3 or Myc, detected the full-length WT RIPK3 as two bands at a molecular weight (MW) of ~60 kDa (fig. 2C). The lower band corresponds to the non-phosphorylated protein, as an antibody against the phosphorylated form (Ser227) of RIPK3 detected only the upper band (fig. 2C). In addition, bands were also detected at ~30 and ~25 kDa with the N-ter and C-ter Ab, respectively, suggesting that RIPK3 was cleaved upon overexpression (fig. 2C). Both the R422* and P493fs9* RIPK3 proteins were C-terminally truncated, as both were detected with the N-ter Ab but not with the C-ter Ab (fig. 2C). Moreover, R422* RIPK3 displayed impaired autophosphorylation and cleavage (fig. 2C). Function of the R422* and P493fs9* RIPK3 proteins following transient transfection of HEK293T cells We investigated the activity of the mutant forms of RIPK3 by measuring NF-κB-dependent luciferase induction following the transient transfection of HEK293T cells. R422* was non- functional in this assay, whereas P493fs9* retained WT RIPK3-like activity (fig.2D, S2A and S2B). We also characterized all the other nine homozygous non-synonymous RIPK3 variants from the gnomAD database in the same assay, seven of which had a MAF<0.01, the other two having a MAF between 0.01 and 0.1 (Table S3). One variant, Leu131Met (MAF = 0.0000398), was mildly hypomorphic, whereas the other eight variants were isomorphic Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 6 (fig. S2C–E). The cumulative frequency of homozygotes for hypomorphic and LOF RIPK3 variants in the general population is, therefore, ~1.5E-5 (confidence interval: 3.8E-7 – 8.4E-5), a frequency consistent with the low prevalence and incomplete penetrance of HSE. RIPK3 has been shown to homodimerize or to heterodimerize with RIPK1 and TRIF via the RHIM domains of the two proteins (57–60). We therefore assessed the capacity of the R422* and P493fs9* variants to interact. P493fs9* was able to homodimerize, and to bind WT RIPK3, RIPK1 and TRIF, whereas R422* was not able to do so (Fig. 2E,F, S2F-H). This finding is consistent with the conservation in P493fs9* of an intact RHIM, which is clearly truncated in R422, and the requirement of this motif for interaction with RHIM-containing proteins and the induction of NF-κB-dependent transcription (49, 61, 62). These data suggest that both the R422 and P493fs9* variants of RIPK3 are expressed as C-terminally truncated proteins, and that R422* is LOF for autophosphorylation, cleavage, NF-κB induction activity, and homo- or heterodimerization, whereas P493fs9* conserves these functions in overexpression conditions. Production of the mutant RIPK3 proteins following stable transduction of HT29 cells We analyzed the production of the R422* and P493fs9* RIPK3 proteins following the lentivirus-mediated transduction of RIPK3 knockout (KO) HT29 cells. The HT29 cell line is a human colorectal adenocarcinoma cell line with an epithelial morphology that is commonly used for cell death assays (48). Similar levels of RIPK3 mRNA were detected in cells stably transduced with R422, P493fs9, and WT RIPK3 (fig. 3A). Western blotting with the N-ter and C-ter RIPK3 antibodies revealed the presence of WT and mutant RIPK3 proteins, respectively, at molecular weights similar to those in HEK293T cells (fig. 3B, S3A). The autophosphorylation of endogenous RIPK3 and exogenous WT and mutant RIPK3 proteins was also detected in this system (fig. S3A). R422* and P493fs9* are C-terminally truncated proteins (fig. 3B, S3A), but only the P493fs9* protein was produced in smaller amounts than the WT RIPK3 (fig. 3B, S3A), suggesting that it may be poorly translated or subject to more intense posttranslational degradation. We compared the half-lives of the mutant and WT RIPK3 proteins in transiently transfected HEK293T cells following treatment with the protein synthesis inhibitor cycloheximide (CHX). The half-life of the P493fs9* protein was much shorter than those of R422* and WT RIPK3, suggesting that the P493fs9* protein is prone to degradation (fig. 3C). RIPK3 degradation has been reported to be mediated by both proteasome-dependent and lysosome-dependent pathways (63–65). We investigated both these degradation pathways in HT29 cells. P493fs9* mutant protein levels were rescued by treatment with proteasome inhibitors (MG132 and BTZ), but not by treatment with lysosomal protease inhibitors (CQ and E64d/Pep A) (fig. 3D). This result suggests that the P493fs9* variant is functionally isomorphic upon transient overexpression, but that the mutant protein is unstable and prone to excessive degradation. Thus, the mutant allele encoding this variant may be hypomorphic or LOF when expressed constitutively, given the low overall levels of the protein in the patient’s cells. Function of the mutant RIPK3 proteins in TLR3/4- or TNF-induced cell death in HT29 cells We analyzed the function of the two mutant proteins following the stable transduction of RIPK3 KO HT29 cells (51). We first assessed the ability of these proteins to induce the phosphorylation of mixed-lineage kinase domain-like pseudokinase (MLKL), the main Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 7 substrate of RIPK3 and the main known effector molecule of the necroptotic pathway (66, 67). Cells were treated with the pan-caspase inhibitor z-VAD, and the SMAC mimetic BV6, in combination with either the TLR3 agonist poly(I:C) (PBZ) or TNF (TBZ), to trigger necroptosis (43, 46, 48). Cells transduced with R422* or P493fs9* had lower levels of MLKL phosphorylation (p-MLKL) than RIPK3 KO HT29 cells transduced with the WT RIPK3 cDNA, after four hours of stimulation with PBZ (TLR3) or two hours of stimulation with TBZ (TNFR1) (fig. 3E, S3B and S3C). We also analyzed caspase 3 cleavage, a marker of apoptosis induction, upon TLR3 or TNFR1 stimulation with BV6 in the same cells, in combination with either poly(I:C) (PB) or TNF (TB). Caspase 3 cleavage levels were similar in the parental and RIPK3 KO HT29 cell lines (fig. S3D and S3E), suggesting that apoptosis was induced. We then assessed apoptosis- and necroptosis-mediated cell death upon TLR3, TLR4 or TNFR1 stimulation. Stimulation with poly(I:C), LPS or TNF, in the presence of BV6 alone (activating apoptotic signaling) or in the presence of both BV6 and z-VAD (necroptotic signaling), led to sustained high levels of cell death in HT29 RIPK3 KO cells transduced with WT RIPK3 (fig. 3F,G). By contrast, in R422-transduced cells, the TLR3-, TLR4- or TNFR1-mediated induction of apoptotic or necroptotic cell death was abolished (fig 3F,G). P493fs9-transduced cells displayed an impairment of the TLR3- or TNFR1-dependent induction of apoptotic or necroptotic cell death at early, but not late time points, as well as impaired but not abolished TLR4-mediated apoptotic or necroptotic cell death (fig. 3F), suggesting that this unstable variant is hypomorphic. As a control for RIPK1/3-independent apoptotic cell death, we used a combination of TNF and cycloheximide (TC) to stimulate the HT29 cells (68). As expected, similar levels of cell death were observed in RIPK3 KO HT29 cells transduced with empty vector, WT or mutant RIPK3 (Fig. S3F), confirming that RIPK3 had no effect on TC-induced apoptosis in HT29 cells, contrasting with its role in TLR3/4- and TNF-induced apoptotic or necroptotic cell death. Function of the mutant RIPK3 proteins in ZBP1/DAI-mediated cell death upon HSV-1 infection in HT29 cells It has been shown in mice that ZBP1/DAI is a RHIM adaptor able to recruit RIPK3 to drive virus-induced necroptosis during infection with RHIM suppressor mutant herpesviruses (51, 69). We assessed the impact of our patient’s RIPK3 mutants on ZBP1/DAI-mediated necroptotic cell death in RIPK3 KO HT29 cells upon HSV-1 infection, comparing the results with those for WT RIPK3 and a previously reported kinase-dead RIPK3 mutant, D142N (70). Cell death rates were high in HT29 RIPK3 KO cells transduced with WT RIPK3, but not in those transduced with D142N RIPK3, as expected (fig. 3H, S3G–I). DAI-mediated necroptotic cell death was abolished in R422-transduced cells and cell death rates in P493fs9-transduced cells were lower than those in WT RIPK3-transduced cells (fig. 3H). Our data therefore suggest that, following the stable transduction of RIPK3 KO HT29 cells, the R422* and P493fs9* RIPK3 proteins abolish and impair, respectively, TLR3-, TLR4-, TNFR1-, and ZBP1/DAI-mediated, RIPK3-dependent apoptotic and necroptotic cell death. Compound heterozygosity for the two mutant alleles may, therefore, underlie a profound form of AR RIPK3 deficiency in the patient with HSE studied. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 8 Impaired RIPK3 protein production in the patient’s cells The RIPK3 mRNA is ubiquitously produced, in all tissues of the human body (https:// proteinatlas.org/). Consistently, we detected RIPK3 mRNA in the various cell types studied, including Epstein-Barr virus-transformed B cells (EBV-B cells), SV40-T antigen- immortalized fibroblasts (SV40-fibroblasts), primary fibroblasts, and human pluripotent stem cell (hPSC)-derived cortical neurons (fig, S4A). However, RIPK3 mRNA levels were generally much lower in these cells than in the HT29 cell line, probably because RIPK3 production would be compromised by the culture of these human cells (48). We measured RIPK3 mRNA and protein levels in the patient’s cells. The total amounts of RIPK3 mRNA in EBV-B cells, SV40-fibroblasts, and primary fibroblasts from P1 were towards the lower end of the range in healthy controls (fig. 4A). Moreover, the TOPO-cloning of cDNAs generated from mRNA from the patient’s SV40-fibroblasts revealed that ~80% of the RIPK3 transcripts were P493fs9* transcripts, whereas only ~20% were R422* (fig. 4B), suggesting that R422* transcripts underwent nonsense-mediated mRNA decay. The RIPK3 protein was undetectable in EBV-B cells, SV40-fibroblasts, and primary fibroblasts from the patient when assessed with Abs recognizing the N-ter or C-ter of the protein (fig. 4C). In human fibroblasts heterozygous for the P493fs9* variant only, RIPK3 mRNA and protein levels were within the range of healthy controls (Fig. S4B, S4C), and only one band corresponding to WT RIPK3 protein was detected by western blotting with both N-ter and C-ter RIPK3 antibodies (Fig. S4C). Finally, following treatment of the patient’s SV40-fibroblasts with proteasome inhibitors (MG132 and BTZ) or lysosomal protease inhibitors (CQ and E64d/ PepA), a band corresponding to the MW of the P493fs9* protein was detected by western blotting with an N-ter, but not a C-ter RIPK3 antibody (fig. 4D). Together with studies of individual alleles in HEK293T and HT29 cells, these findings suggest that RIPK3 proteins were undetectable in the cells of the patient due to the nonsense mRNA-mediated decay of R422* and the instability of the P493fs9* protein. These data suggest that compound heterozygosity for R422* and P493fs9* underlies a very profound, and perhaps complete, form of AR RIPK3 deficiency in this patient. Impaired RIPK3-dependent necroptotic and apoptotic cell death in the patient’s fibroblasts RIPK3 is required for MLKL phosphorylation and the initiation of necroptosis downstream from TLR3/4, TNFR1, ZBP1/DAI, and IFNRs (71) in various cell types. We compared MLKL phosphorylation in SV40-fibroblasts from the patient with that in cells from healthy controls following stimulation with PBZ (TLR3) or TBZ (TNFR1). We detected p-MLKL clearly in control, but not in patient fibroblasts (fig. 4E and S4D). Transient plasmid-mediated exogenous WT RIPK3 expression rescued p-MLKL levels in the patient’s fibroblasts (fig. 4E), suggesting that compound heterozygosity for the two RIPK3 variants resulted in AR RIPK3 deficiency in these cells, accounting for resistance to necroptosis. Interestingly, by contrast to HT29 cells, in which RIPK3 KO did not affect caspase 3 cleavage in response to stimulation with PB or TB, caspase 3 cleavage in response to PB or TB was impaired in P1 fibroblasts (fig. 4F), suggesting that RIPK3 is directly or indirectly involved in TLR3- or TNFR1-mediated apoptosis in human fibroblasts. Primary fibroblasts from the patient were also more resistant than control cells to TLR3- and TNFR1-mediated necroptotic and apoptotic cell death (fig. 4G,H). Moreover, transient lentivirus-mediated exogenous WT RIPK3 expression rescued the sensitivity of the patient fibroblasts to TLR3- Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 9 and TNFR1-mediated necroptotic and apoptotic cell death (fig. 4G,H). However, the kinase- dead mutant, RIPK3 D142N, was able to rescue TLR3- and TNFR1-mediated apoptotic cell death, but not necroptotic cell death in patient fibroblasts (fig. 4G,H and S4G,H), consistent with previous demonstrations of the dependence of RIPK3-mediated necroptosis, but not apoptosis, on the kinase activity of RIPK3 (48, 72, 73). Fibroblasts from other HSE patients with AR complete deficiencies of the TLR3 pathway (TLR3, TRIF, or UNC93B1) (19, 21, 28), serving as controls, displayed impaired TLR3-, but not TNFR1-mediated, necroptotic and apoptotic signaling (fig. 4I–L). By contrast, in fibroblasts from a patient with an AR deficiency of TBK1 (74) — a molecule at the crossroads of various antiviral pathways including TLR3 and functioning in parallel to IKBKE and RIPK1 (74) — necroptotic and apoptotic signaling upon TLR3 or TNFR1 stimulation was intact (fig. 4I–L). Overall, these findings suggest that AR RIPK3 deficiency in this patient has an overarching impact on known RIPK3 functions, as it profoundly impairs TLR3- and TNFR1-mediated necroptotic and apoptotic signaling, at least in fibroblasts. Normal TLR3- and TNFR1-dependent NF-κB and IRF3 pathways in the patient’s fibroblasts RIPK3 has been implicated in cell death-independent signal transduction in addition to the necroptotic and apoptotic signaling pathways (75). It acts particularly through the recruitment of RIPK1, and TRIF (60), via their RHIMs. We compared the activation of NF- κB, IRF3 and MAPKs following the stimulation of TLR3 and TNFR1, between fibroblasts from P1 and from healthy controls and other patients with recessive, complete TLR3 or NEMO deficiency. Following endosomal stimulation with poly(I:C), the activation of P65 and IRF3 was normal in SV40-fibroblasts from P1, with ERK1/2 and JNK1/2 activation only mildly decreased (not abolished), whereas the phosphorylation of P65, IRF3, and MAPKs (including ERK1/2, JNK1/2) was completely abolished in fibroblasts from TLR3- or NEMO-deficient patients (fig. 5A). Following stimulation with TNF, the activation of NF- κB, IRF3 and MAPKs was intact in fibroblasts from P1 and TLR3-deficient patients (fig. 5B–C). In NEMO-deficient cells, which served as a negative control in these experiments (76, 77), the activation of P65, ERK1/2 and IRF3 upon stimulation with poly(I:C) or TNF was abolished. Thus, RIPK3 deficiency had a modest impact on the TLR3-mediated activation of NF-κB, IRF3 and MAPKs, whereas the activation of TNFR1-mediated signaling was intact. These findings suggest that RIPK3 is redundant for the activation of these pathways, at least in fibroblasts. The patient’s cells would therefore probably display normal type I IFN production following the stimulation of TLR3 by dsRNA (including normal baseline IFN-β levels), and during the course of viral infection. Normal TLR3-dependent induction of anti-viral IFNs in the patient’s fibroblasts We measured the production of IFNs and other cytokines in RIPK3-deficient fibroblasts from P1 following TLR3 activation by poly(I:C). Interestingly, unlike TLR3- and NEMO- deficient fibroblasts, in which the TLR3-mediated production of antiviral IFN-β, IFN-λ and other cytokines and chemokines was impaired (fig. 5D and S5A), only the TLR3- mediated production of IL-6, IL-8 and CCL3 was impaired in fibroblasts from P1, which had normal levels of production for IFN-β, IFN-λ, and the other cytokines tested (fig. 5D and S5A). Similar results were obtained in assessments of mRNA levels for IFNB, IFNL, and ISGs (MX1, IFIT1, ISG15) by RT-qPCR in the same cells following TLR3 Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 10 activation (fig. S5B). Consistent with the intact TNFR1-mediated activation of NF-κB, IRF3 and MAPKs in P1’s fibroblasts, the production of IFNs and of the other cytokines tested was similar in fibroblasts from P1 and in those from controls, following TNF stimulation (fig. S5C). Moreover, the patient’s fibroblasts displayed a normal response to IFN-α2b, as shown by normal mRNA induction for ISGs including IFIT1 and MX1 (fig. S5D), indicating that the signaling of the IFNAR1-mediated type I IFN response pathway was intact. Together, these data suggest that RIPK3 is important for the TLR3-mediated production of a few specific proinflammatory cytokines and chemokines, such as IL6, IL-8 and CCL3, probably related to the mild decrease in ERK1/2 and JNK1/2 activation by poly(I:C) stimulation (fig. 5A). However, RIPK3 is largely redundant, at least in fibroblasts, for the production of antiviral type I and III IFNs and other cytokines. A role of impaired TLR3-dependent chemokine production in the pathogenesis of HSE in P1 cannot be excluded, as a previous study suggested that RIPK3 may coordinate immune responses by mediating chemokine production in West Nile virus-infected mouse neurons (78). However, the molecular mechanisms by which human RIPK3 deficiency underlies HSE do not involve an impairment of type I IFN-mediated antiviral immunity, contrasting with the situation in patients with HSE due to mutations of the TLR3-IFNAR1 circuit. The pathogenesis of HSE due to RIPK3 deficiency is probably related to an impairment of TLR3-, and/or ZBP1/ DAI-mediated, and perhaps TNFR1-mediated, necroptotic and apoptotic signaling. Enhanced HSV-1 replication in the patient’s fibroblasts We have shown that HSE-causing TLR3 pathway mutations impair TLR3-dependent type I IFN-mediated cell-intrinsic immunity to HSV-1 in fibroblasts and hPSC-derived cortical neurons (20). We tested SV40-fibroblasts, which have been shown to be a surrogate cell type for studies of type I IFN-dependent and -independent cell-intrinsic immunity to viruses, including HSV-1 (19, 20, 23–27). Consistent with the normal production of IFN-β and IFN-λ by P1’s SV40-fibroblasts upon TLR3 stimulation (fig. 5D and S5A–B), HSV-1 infection (KOS strain) of the patient’s SV40-fibroblasts also induced normal levels of IFNB and IFNL production, similar to those in healthy controls (fig. S6A). Furthermore, the cellular responses to HSV-1, as assessed by bulk RNA sequencing (RNA-seq) on primary fibroblasts from the patient, were normal at genome-wide transcriptome level (Fig. 6A,B), including for the induction of all fibroblastic ISGs (32) assessed (Fig. S6B), as shown by comparisons with cells from healthy controls. Nevertheless, P1’s fibroblasts displayed enhanced HSV-1 replication at various time points after HSV-1 infection at a low multiplicity of infection (MOI=0.001), in multiple cycle conditions (fig. 6C), at levels similar to those in TLR3-deficient or IFNAR1-deficient fibroblasts from other HSE patients, but higher than those in healthy controls. Exogenous IFN-β pretreatment rendered P1’s fibroblasts and TLR3-deficient fibroblasts resistant to HSV-1 infection, but had no such effect on IFNAR1-deficient fibroblasts (fig. 6C). Finally, unlike fibroblasts from AR TLR3- or IFNAR1-deficient patients, P1’s fibroblasts displayed normal control of infection for the other neurotropic viruses tested, including vesicular stomatitis virus (VSV), measles virus (MeV), encephalomyocarditis virus (EMCV) and influenza A virus (IAV), similar to that in cells from healthy controls (Fig. 6D–G). Therefore, despite their intact type I IFN antiviral immunity, P1’s fibroblasts are specifically highly susceptible to HSV-1, probably Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 11 due to impaired control of viral growth via necroptotic/apoptotic cell death, which plays a particularly important role in the control of HSV-1 infection. Low levels of cell death and enhanced HSV-1 replication in the patient’s cortical neurons We have shown that HSE-causing mutations impair cell-intrinsic immunity to HSV-1 in hPSC-derived CNS cortical neurons and oligodendrocytes due to the disruption of TLR3- dependent type I IFN-mediated antiviral immunity (mutations of the TLR3-IFNAR1 circuit) (30) or new antiviral mechanisms (mutations of SNORA31) (34). Programmed cell death pathways are key mechanisms in host antiviral defenses, particularly against viruses that invade the central nervous system (79). We tested CNS neurons from patient-specific hPSCs reprogrammed from P1’s primary fibroblasts (fig. S6C–E). RIPK3 mRNA levels were low in P1’s neurons (fig. 6H), consistent with the results obtained for P1’s EBV-B cells and fibroblasts. TOPO-cloning of the cDNA generated from the mRNA of the patient’s hPSC-derived neurons revealed that ~80% of the RIPK3 transcripts were P493fs9, whereas only ~20% were R422 (fig. 6I), as in fibroblasts, suggesting that R422* mRNA also underwent nonsense-mediated decay in cortical neurons. HSV-1 replication levels were much higher in hPSC-derived cortical neurons from P1 than in those from healthy controls, as were those in TLR3- and IFNAR1-deficient hPSC-derived neurons, at various times points after infection with HSV-1 at a MOI of 0.001 (fig. 6J). As in P1’s fibroblasts, enhanced HSV-1 replication was rescued by pretreatment with exogenous IFN-β in P1’s and TLR3-deficient hPSC-derived neurons, but not in IFNAR1-deficient hPSC-derived neurons (fig. 6J). However, unlike TLR3-deficient neurons, which had higher levels of virus-induced cell death following HSV-1 infection, neurons from P1 displayed enhanced resistance to HSV-1 infection-induced cell death, even relative to healthy control hPSC-derived neurons (fig. 6K). These data suggest that RIPK3-deficient cortical neurons are highly susceptible to HSV-1 due to defective HSV-1-induced necroptotic and apoptotic cell death-dependent antiviral defenses. Enhanced HSV-1 replication and low levels of cell death in RIPK3 KO cortical neurons Finally, as a means of establishing a causal relationship between genotype (RIPK3 deficiency) and phenotype (low levels of cell death-dependent antiviral immunity resulting in enhanced HSV-1 growth), we generated CRISPR/Cas9-mediated RIPK3 knockout (KO) hPSC lines, which were differentiated into cortical neurons (fig. S6F–I). The relative levels of RIPK3 mRNA in RIPK3 KO neurons were low, suggesting that transcripts were destroyed by nonsense-mediated mRNA decay (fig. 6L). RIPK3 KO neurons, like hPSC- derived neurons from P1, had much higher levels of HSV-1 than the parental WT cells, a phenotype that was rescued by IFN-β pretreatment (fig. 6M and S6J). Like hPSC-derived neurons from P1, RIPK3 KO neurons also displayed enhanced resistance to HSV-1-induced cell death (fig. 6N). Together, these data confirm that impaired RIPK3-dependent apoptotic and/or necroptotic cell death signaling-mediated antiviral immunity renders RIPK3-deficient hPSC-derived cortical neurons prone to HSV-1 infection. The activation of apoptotic and necroptotic cell death pathways during HSV-1 infection appears to depend on TLR3 and/or ZBP1/DAI, which are controlled by RIPK3, although a role for other antiviral sensors (e.g. TNFR1) cannot be excluded. Our data provide a plausible molecular and cellular mechanism of HSE in this patient with AR RIPK3 deficiency. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Discussion Page 12 AR RIPK3 deficiency is a new genetic etiology of childhood HSE. Approximately 5% of the children studied in our international HSE cohort have experimentally proven AR or AD deficiencies of the TLR3-IFN-α/β circuit, which governs cell-intrinsic immunity in specific organs of the human body, including the brain and the lung (80–82), and another ~2% have AD or AR deficiencies of snoRNA31 or DBR1, which govern new antiviral mechanisms that appear to be specific to the forebrain and brainstem, respectively (33, 34). TLR3-induced apoptotic and necroptotic signaling is impaired in RIPK3-deficient fibroblasts, resulting in impaired TLR3-dependent apoptotic and necroptotic cell death. TLR3-dependent apoptotic and necroptotic signaling is also impaired in fibroblasts from other HSE patients with deleterious mutations of TLR3 or of some of the genes encoding components of its signaling pathway (UNC93B1, TRIF). Impaired TLR3-RIPK3-dependent apoptotic and necroptotic cell death-mediated antiviral immunity may, therefore, have played a role in HSE pathogenesis in the patient with AR RIPK3 deficiency studied here, and in other patients with deficiencies of the TLR3 pathway. However, in HSE patients with inborn errors of the TLR3 pathway, the TLR3-dependent production of IFN-α/β, -λ and of many other cytokines was also impaired, whereas RIPK3 deficiency impaired the production of only a narrow range of cytokines, not including antiviral IFNs. Moreover, RIPK3 deficiency impaired apoptotic and necroptotic cell death signaling not only via TLR3, but also via ZBP1/DAI, the proposed sensor inducing HSV-1-induced necroptotic cell death (51). It also affects cell death signaling via other sensors, such as TLR4 and TNFR1. RIPK3-deficient hPSC-derived cortical neurons were resistant to HSV-1-induced cell death, leading to excessive virus replication, whereas TLR3-deficient neurons underwent HSV-1- induced cell death earlier, due to enhanced viral replication. Inborn errors of RIPK3 and the TLR3-IFN circuit may, thus, lead to both common and specific molecular mechanisms of disease, impairing cell-intrinsic antiviral immunity in an overlapping manner. In vitro studies have shown that the RIPK1/RIPK3 complex may be involved in regulating virus-induced cell necroptosis, which can be triggered by various viral agents, including influenza A virus (IAV) (44), RHIM suppressor mutants of murine cytomegalovirus (HCMV and MCMV) (45), or herpes simplex viruses (HSV-1 and HSV-2) (46, 47), and E3-deficient vaccinia virus (43). A role for RIPK3-mediated cell death in host antiviral defense was first proposed in 2009, following studies of vaccinia virus infection in vivo (43). Ripk3-deficient mice also display impaired HSV-1-induced necrosis, and uncontrolled HSV-1 replication and pathogenesis in various organs, including the brain (50). Meanwhile, Ripk3-deficient mice displayed levels of murine gammaherpesvirus 68 (MHV68) replication similar to that in the WT mice, at least in the lung and the spleen (83), suggesting that the role of RIPK3 in host antiviral immunity might be virus- or organ-specific. In addition to the contribution of RIPK3-mediated cell death to restricting viral pathogenesis, cell death-independent functions of RIPK3 may also specifically limit viral invasion of the CNS. Ripk3-deficient mice are susceptible to CNS infection by West Nile virus (WNV) and Zika virus (ZIKV) in vivo (78, 84). Interestingly, it has been suggested that the enhanced susceptibility of Ripk3−/− mice to WNV is due to an abolition of neuronal chemokine expression, leading to lower levels of T-lymphocyte and myeloid-cell recruitment to the CNS to restrict viral Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 13 infection (78). In another study, an enhanced viral load in Ripk3−/− mouse brain following infection with ZIKV was attributed to impaired activation of the nucleotide sensor ZBP1 and downstream RIPK1 and RIPK3 signaling to restrict viral replication, due to changes in cellular metabolism due to the upregulation of an enzyme, IRG1, and production of the metabolite itaconate (84). RIPK3 therefore seems to control mouse cortical neuron-intrinsic immunity to viruses through canonical cell apoptosis and necroptosis pathways and non- canonical functions (78, 84). In this study, we observed impaired TLR3-mediated CCL3 production in RIPK3-deficient fibroblasts from our patient, but the pathophysiological role of this impairment of CCL3 production remains unclear. Remarkably, the RIPK3-deficient patient, who is now 24 years old, has not suffered from any severe infectious disease other than HSE, viral or otherwise, despite infection with many common viruses (fig. S1A,B). Human RIPK3 may be crucial to protect the CNS against HSV-1 through control over cortical neuron cell-intrinsic immunity to HSV-1, but otherwise largely redundant (85). This hypothesis is supported by our in vitro data showing that RIPK3-deficient fibroblasts from our patient are susceptible to HSV-1, but able to control the other four viruses tested normally. This is also reminiscent of the normal clearance of MHV68 in Ripk3-deficient mice (83). More patients are required to delineate more accurately the range of infections and other phenotypes associated with this deficiency (85). However, the clinical phenotype of this patient already contrasts sharply with that of inherited RIPK1 deficiency (86, 87). Fourteen patients with AR RIPK1 deficiency, which became symptomatic between the ages of one day and six months and was diagnosed between the ages of one day and 11 years, have been reported (86–88). The main clinical phenotypes of these RIPK1-deficient patients include early-onset inflammatory bowel disease, polyarthritis, and recurrent viral, bacterial, and fungal infections. Viruses as diverse as human cytomegalovirus (HCMV), varicella zoster virus (VZV), respiratory syncytial virus (RSV), and HSV-1 have been reported to underlie severe infections of the skin and digestive and respiratory mucosae. These patients also displayed multiple leukocytic abnormalities, including profound and broad NK, T, and B lymphopenia. The cellular and molecular basis of viral disease in RIPK1 deficiency has yet to be studied, and it is unclear whether RIPK1 deficiency affects non-leukocytic, cell-intrinsic immunity to viruses. HSV-1 infection status is unknown for most patients. However, given their very young age at diagnosis, these patients are unlikely to have been infected with HSV-1 before IgG substitution or hematopoietic stem cell transplantation. In such circumstances, we would predict that RIPK1-deficient patients would probably be prone to HSE. By contrast, the absence of obvious leukocytic abnormalities in the RIPK3-deficient patient is consistent with the absence of serious infections other than HSE in this patient until the age of 24 years. Future studies searching for inborn errors of RIPK1, RIPK3, and related molecules in patients with HSE and other diseases will improve our understanding of RIPK1 and RIPK3 biology in natural conditions (89–92). Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 14 Materials and methods Study design We first performed whole-exome sequencing (WES) on one patient who had recurrent Herpes simplex virus 1 (HSV-1) encephalitis (HSE) during her childhood, and both her parents (Trio design), in order to searched for candidate monogenic inborn error of immunity (IEI) related to HSE, with a focus on rare de novo or biallelic variants. This resulted in the identification of compound-heterozygous predicted to be loss-of-function (pLOF) mutations of RIPK3 in the patient. We then carried out molecular characterization of the two mutant RIPK3 proteins, by studying their expression and function following the transient transfection of HeLa and HEK293T cells or the stable transduction of HT29 cells. Finally, in order to evaluate the causality and mechanism of RIPK3 deficiency in HSE pathogenesis, we studied the known antiviral type I IFN inducing pathways, the RIPK3-dependent necroptotic and apoptotic cell death, as well as HSV-1 replication levels in the patient’s fibroblasts and human pluripotent stem cell (hPSC)-derived cortical neurons, in comparison with those from healthy controls and previously published patients with recessive deficiencies of TLR3 (21) or IFNAR1 (28) also predisposing to HSE. Human subjects Informed consent was obtained in France, in accordance with local regulations and a human subjects research protocol approved by the institutional review board (IRB) of the Institut National de la Santé et de la Recherche Médicale (INSERM). Experiments were conducted in the United States and France, in accordance with local regulations and with the approval of the IRB of The Rockefeller University and INSERM, respectively. Approval was obtained from the appropriate French Ethics Committee (Comité de Protection des Personnes), the French National Agency for Medicine and Health Product Safety, INSERM in Paris, France (protocol no. C10–13), and the Rockefeller University Institutional Review Board in New York, USA (protocol no. SHZ-0676). Cell culture and transfection Primary human fibroblasts were obtained from skin biopsy specimens from controls and P1, and were cultured in DMEM (GIBCO BRL, Invitrogen) supplemented with 10% fetal calf serum (FCS) (GIBCO BRL, Invitrogen). Immortalized SV40-transformed fibroblast cell lines (SV40-F) and Epstein-Barr virus (EBV)-transformed B-cell lines (EBV- B) were generated as previously described (33). More technical details are provided in Supplementary Materials. HEK293T, HeLa and HT29 cells (ATCC) were maintained in DMEM supplemented with 10% FCS. SV40-F, HEK293T and HeLa cells were transiently transfected with the aid of X-tremegen 9 DNA Transfection Reagent (XTG9-RO, Roche). RIPK3 KO cell lines were transduced with a lentiviral system, with a mock vector (Luci), or with WT and mutant RIPK3 constructs, and were then cultured under puromycin selection. Whole-exome and Sanger sequencing Genomic DNA was isolated from peripheral blood cells or primary fibroblasts from the patient. Whoe exome sequencing (WES) was performed as previously described (34). More Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 15 technical details are provided in Supplementary Materials. For the Sanger sequencing of the RIPK3 variants, 300–600-bp genomic regions encompassing the mutation were amplified by PCR with site-specific oligonucleotides. PCR products were purified by ultracentrifugation through Sephadex G-50 Superfine resin (Amersham-Pharmacia-Biotech), and sequenced with the Big Dye Terminator Cycle Sequencing Kit on an ABI Prism 3700 apparatus (Applied Biosystems). SnapGene was used for sequence analysis. Western blots and protein immunoprecipitation (IP) For Western blots, HEK293T, HT29 or SV40-F cell pellets were harvested, washed with PBS and lysed in RIPA buffer supplemented with cOmplete protease inhibitor cocktail (Roche). Total cell lysates were harvested. Equal amounts of protein from each sample were subjected to SDS-PAGE, and the proteins were blotted onto polyvinylidene difluoride membranes (Bio-Rad). The membranes were then probed with the desired primary antibody followed by the appropriate secondary antibody. More technical details are provided in Supplementary Materials. For co-immunoprecipitation assays, HEK293T cells were cotransfected with RIPK3 and/or RIPK1 plasmids with FLAG or Myc tags. The cells were harvested 48 h later, stored at °20°C overnight, then processed to immunoprecipitation (IP) followed by Western blotting. More technical details are provided in Supplementary Materials. Reverse transcription-quantitative PCR (RT-qPCR) Total RNA was extracted with the RNeasy mini kit (Qiagen) from HEK293T cells, primary fibroblasts, SV40-F, EBV-B cells and hPSC-derived cortical neurons. The RNA was reverse-transcribed with the SuperScript III First-Strand Synthesis System (Thermo Fisher Scientific, #180800051). Reverse transcription-quantitative PCR (RT-qPCR) was performed with Applied Biosystems 2 × universal Taqman reaction mixture and Assays-on-Demand probe/primer combinations, in an ABI PRISM 7700 Sequence Detection System. The human β-glucuronidase (GUSB) gene was used for normalization with the VIC™/TAMRA™ probe (Thermo Fisher Scientific, 4310888E). The Hs01011171_g1 (RIPK3 exons 2–3) and Hs01011177_g1 (RIPK3 exons 9–10) probes were used. The other probes used in this study have been reported elsewhere (34). Results are expressed according to the Δ described by the manufacturer. ΔCt method, as Induction of necroptosis and apoptosis For the detection of MLKL phosphorylation, necroptosis was induced by treatment with PBZ complex (25 μg/ml poly(I:C) (Tocris, #4287); 1 μM BV6 (APExBIO, B4653); 20 μM z-VAD-fmk(APExBIO, A1902)) or TBZ complex (1000 units/ml, recombinant human TNF-alfa (TNF) (R&D Systems, 210-TA-020); 1 μM BV6; 20 μM z-VAD-fmk) for 4 h. Whole-cell lysates were collected and used for western blotting. For the detection of caspase 3 cleavage, apoptosis was induced by combined PB (25 μg/ml poly(I:C) and 1 μM BV6) or TB (1000 units/ml TNF and 1 μM BV6) treatment for various times. Whole-cell lysates were collected and western blotting was performed. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Cell viability assay Page 16 Cells (5,000 cells/well) were used to seed Corning 96-well tissue culture plates; 16–24 h post-seeding, cells were treated with the indicated reagents, with levels of the solvent, DMSO, kept constant for all experiments. We used 25 μg/ml poly(I:C), 30 ng/ml TNF, 2 μg/ml LPS, 100 μg/ml cycloheximide (CHX), 1 μM BV6, and 20 μM Z-VAD-fmk. HSV-1 infection-induced DAI-dependent necroptotic cell death was triggered, as previously described (51). Briefly, DAI was stably overexpressed in HT29 cells with a retrovirus system, with selection on 500 μg/ml hygromycin. The cells were then infected with the F strain of HSV-1 FmutRHIM (MOI=5) for 24 h. Cell viability was assessed by measuring intracellular ATP levels with the CellTiter-Glo luminescent cell viability assay kit (Promega, G7571) according to the manufacturer’s instructions. Bulk RNA sequencing (RNAseq) RNA was extracted from primary fibroblasts with the Quick-RNA MicroPre Kit (#R1051, Zymo Research). RNA-Seq libraries were prepared with the Illumina RiboZero TruSeq Stranded Total RNA Library Prep Kit (Illumina) and RNA-sequencing was performed on the Illumina NovaSeq platform, with a read length of 100 bp and a read depth of 40 million reads. All samples were sequenced in technical duplicates. All FASTQ files passed quality control (QC) and were aligned with the GRCh38 reference genome with STAR (2.6.1d). Gene-level features were quantified with featureCounts v1.6.0 based on GRCh38 gene annotation. Count data were normalized through ‘cpm’ (counts per million) in the EdgeR package (93), dimension-reduced through principal component analysis (PCA), and subjected to heatmap analysis with “ComplexHeatmap” (94). Differential expression (DE) analysis was performed with DESeq2 (95). Cell stimulation and cytokine production in a multiplex assay We used a synthetic analog of dsRNA, poly(I:C), as a nonspecific agonist of TLR3 and MDA5/RIG-I. SV40-fibroblast cells were activated in 24-well plates, at a density of 100,000 cells/well, by incubation for 24 h with poly(I:C) at concentrations of 1, 5 and 25 μg/mL. Cells were stimulated with 25 μg/ml poly(I:C) in the presence of Lipofectamine 2000 to activate MDA5/RIG-I signaling. After 24 h, cell supernatants were used for the LEGENDplex multiplex bead assay (BioLegend, #740003 and #740984), then analyzed by flow cytometry on an Attune NxT Flow Cytometer, according to the manufacturer’s instructions. Data were analyzed with LEGENDplex Cloud-based Data Analysis Software (BioLegend), and presented in raw values and heatmaps. The heatmaps are generated using relative values for each sample as normalized to the maximum range of production levels of each cytokine among all samples (X-Min)/(Max-Min). Min: minimum production level; Max: maximum production level; X: the production level of a given cytokine in a given sample (96). Luciferase reporter assays HEK293T cells (2.5 × 105 cells ml−1) were used to seed a 96-well plate. They were transfected, in triplicate, the following day, with WT and mutant RIPK3 plasmids along with 100 ng NF-κB promoter-firefly luciferase reporter plasmid and 100 ng of Renilla luciferase Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 17 reporter plasmid per well, in the presence of X-tremegen 9 DNA Transfection Reagent (Roche). Luciferase activities were assessed 24 h later, in a dual luciferase assay (Promega, E2940). Firefly luciferase activities were normalized against Renilla luciferase activities. Mass cytometry on fresh whole blood Whole-blood mass cytometry was performed on 250 μl of fresh blood with a custom- designed panel (Table S4), according to Fluidigm recommendations. Labeled cells were frozen at °80°C after overnight dead-cell staining, and acquisition was performed on a Helios machine (Fluidigm). Data analysis was performed with OMIQ software. Gene editing with CRISPR-Cas9 Gene-editing experiments were performed as previously described (97). Briefly, guide RNA sequences were generated with the CRISPR design tool (http:// crispr.mit.edu/). Two gRNAs were selected: 5’-GTCGTCGGCAAAGGCGGGTT-3’ and 5’- GCAGTGTTCCGGGCGCAACAT-3’. Forward and reverse oligonucleotides for the gRNAs were then inserted into the MLM3636 vector (a gift from K. Joung, Addgene, no. 43860). The activity of the gRNAs was assessed in HEK293T cells. We then transfected two million H9 human embryonic stem cells (hESCs) with 20 μg Cas9-GFP plasmid and 5 μg gRNA plasmid mixed in electroporation buffer (BTX, no.45–0805). Cells were sorted by FACS, on the basis of GFP signals, 48 hours after electroporation. We replated 50,000 cells with a moderate GFP fluorescence intensity and cultured them for 72 h. The cells were then detached with Accutase, counted and plated at clonal density in 96-well plates. Ten days later, the colonies were passaged and amplified. Genomic DNA was then extracted from each single-cell clone and genotyping was performed by Sanger sequencing with the following primers: Forward: 5’-AGAGGCGCCTATAAGGGAAGT-3’ and Reverse: 5’- TACACTCCAGGAGAGAGCTGG-3’. TOPO cloning and sequencing of cDNAs from the patient’s cells Total RNA was extracted with the RNeasy mini kit (Qiagen) from SV40-F and hPSC- derived cortical neurons. The RNA was reverse-transcribed with the SuperScript III First-Strand Synthesis System (Thermo Fisher Scientific, #18080051) according to the manufacturer’s instructions. PCR was performed with 2 × Taq PCR master mix (APExBIO, K1034) and the following primers: Forward: 5’-CCCAGACTCCAGAGACCTCA-3’ and Reverse: 5’-AGGGGTGGCACTCTTCCTTA-3’). PCR products were inserted into the pCR2.1-TOPO vector (Life Technologies) and used to transform Steller competent cells. We picked at least 100 colonies per subject, for P1 and a healthy control. Finally, we performed PCR on these colonies and sequenced them with the following primers: Forward: 5´-GTAAAACGACGGCCAG-3´ and Reverse: 5´-CAGGAAACAGCTATGAC-3´. Human pluripotent stem cell (hPSC) culture and cortical neuron differentiation Patient-specific induced pluripotent stem cells (iPSCs) were obtained by reprogramming the patients’ primary fibroblasts by infection with the non-integrating CytoTune Sendai viral vector kit (Life Technologies, USA). HESC or iPSC (together referred to as hPSC) cultures were maintained in Essential 8 medium (Life Technologies, A1517001). We used Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 18 one healthy control hESC line (H9) and one healthy control iPSC line (BJ1) in this study. All hPSCs were karyotyped to ensure that the genome was intact. Patient-specific RIPK3 mutations were confirmed by the Sanger sequencing of genomic DNA extracted from the patient’s iPSC lines. Cortical neuron differentiation was performed with hPSCs cultured in E8 essential medium in 10 cm VTN-N (Thermo Fisher Scientific)-coated plates. Cells were maintained at 37℃, under an atmosphere containing 5% CO2. hPSCs were differentiated with a previously described protocol (32). Viral infections and the quantification of viral replication For WT HSV-1 (KOS strain, ATCC, VR-1493) infection, 5 × 104 SV40-F or 1.75 × 105 cortical neurons per well were used to seed 48-well plates and were infected at a MOI of 0.001 in DMEM supplemented with 2% FCS (for fibroblasts), or in neuron culture medium (for neurons). After 2 h, the cells were washed and transferred to 250 μl of fresh medium. Both cells and supernatants were collected at various timepoints and frozen. HSV-1 titers were determined by calculating the TCID50 ml–1, as previously described (34). For VSV, MeV, IAV and EMCV infections, 5 × 104 SV40-F per well were added to 48-well plates in DMEM supplemented with 10% FCS. Cells were infected with VSV, MeV, IAV or EMCV at different MOI. Cells and supernatants were obtained at various timepoints and frozen, and virus levels determined by calculating the TCID50 ml−1 or by viral RNA quantification, as previously described (32, 81, 98). More technical details are provided in Supplementary Materials. Statistical analysis When applicable, results are presented as means ± SD or means ± SEM. Mean values were compared between control cells and mutant cells by Paired t test or one-way ANOVA followed by Tukey tests for multiple comparisons. When appropriate, linear mixed models were used for log-transformed relative values, to account for repeated measurements. Statistical analysis was performed in GraphPad Prism9 (Version 9.1.1). Statistical significance is indicated as follows: ns, P > 0.05; , P < 0.05; and , P < 0.01, , P < 0.001, *, P < 0.0001 in the corresponding figures. Data and materials availability: The materials and reagents used are commercially available and nonproprietary, with the exception of the gene-KO or patient-specific cell lines generated by this study. The cell lines generated by this study are available from S.-Y.Z. and J.-L.C upon request under MTAs from the Rockefeller University and the Imagine Institute. The RNA sequencing data generated by this study are available in the NCBI database under the NCBI-SRA project PRJNA937264. All other data needed to support the conclusions of the paper are in the paper or the supplementary materials. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 19 Acknowledgments: We warmly thank our patients and their families. We thank the members of both branches of the Laboratory of Human Genetics of Infectious Diseases for helpful discussions; Tatiana Kochetkov for technical assistance; Yelena Nemirovskaya for administrative assistance. We thank the Bio-Imaging, Flow Cytometry Resource Center and Genomics Resource Center of The Rockefeller University for technical assistance. We thank Dusan Bogunovic for providing the human fibroblast cells with complete TBK1 deficiency. Funding: This work was conducted in the two branches of the Laboratory of Human Genetics of Infectious Diseases, and was funded in part by the National Center for Advancing Translational Sciences (NCATS), National Institutes of Health (NIH), Clinical and Translational Science Award (CTSA) program, grant UL1TR001866, NIH grants R01AI088364, R01NS072381, R01AI020211 and R21AI151663, grants from the Integrative Biology of Emerging Infectious Diseases Laboratory of Excellence (ANR-10-LABX-62-IBEID) and the French National Research Agency (ANR) under the “Investments for the future” program (ANR-10-IAHU-01), the ANR grants IEIHSEER (ANR-14-CE14-0008-01), SEAeHostFactors (ANR-18-CE15-0020-02), and CNSVIRGEN (ANR-19- CE15-0009-01), the French Foundation for Medical Research (FRM) (EQU201903007798), the Square Foundation, Grandir - Fonds de solidarité pour l’enfance, the SCOR Corporate Foundation for Science, the Rockefeller University, Institut National de la Santé et de la Recherche Médicale (INSERM), Paris Cité University, and the St. Giles Foundation. LDN is supported by the Division of Intramural Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health. Y.H.C. was supported by the ASTAR International Fellowship (AIF). P.B. was supported by the FRM (EA20170638020) and the MD-PhD program of the Imagine Institute (with the support of the Fondation Bettencourt Schueller). E.S.M. is supported by NIH grant R01AI020211. References 1. Roizman B, Knipe DM, Whitley RJ, Herpes simplex viruses. Fields Virology Vol. 2, 1823–1879 (2013).Vol. 2. James C et al. , Herpes simplex virus: global infection prevalence and incidence estimates, 2016. B World Health Organ 98, 315–329 (2020). 3. Gnann JW Jr., Whitley RJ, Herpes Simplex Encephalitis: an Update. Current infectious disease reports 19, 13 (2017). [PubMed: 28251511] 4. Kennedy PG, Steiner I, Recent issues in herpes simplex encephalitis. Journal of neurovirology 19, 346–350 (2013). [PubMed: 23775137] 5. Hjalmarsson A, Blomqvist P, Skoldenberg B, Herpes simplex encephalitis in Sweden, 1990–2001: incidence, morbidity, and mortality. Clin Infect Dis 45, 875–880 (2007). [PubMed: 17806053] 6. Nahmias AJ, Lee FK, S. B-N. in Herpes Simplex Viruses, Studahl CPM, Bergstrom T.. Ed. (Taylor&Francis, New York, 2006), pp. 55–98. 7. Tyler KL, Update on herpes simplex encephalitis. Rev Neurol Dis 1, 169–178 (2004). [PubMed: 16400278] 8. Abel L et al. , Age-Dependent Mendelian Predisposition to Herpes Simplex Virus Type 1 Encephalitis in Childhood. J Pediatr. 9. Jubelt B, Mihai C, Li TM, Veerapaneni P, Rhombencephalitis / brainstem encephalitis. Current neurology and neuroscience reports 11, 543–552 (2011). [PubMed: 21956758] 10. Stahl JP, Mailles A, Herpes simplex virus encephalitis update. Curr Opin Infect Dis 32, 239–243 (2019). [PubMed: 30921087] 11. Whitley RJ et al. , Vidarabine versus acyclovir therapy in herpes simplex encephalitis. N Engl J Med 314, 144–149 (1986). [PubMed: 3001520] 12. Arciniegas DB, Anderson CA, Viral encephalitis: neuropsychiatric and neurobehavioral aspects. Curr Psychiatry Rep 6, 372–379 (2004). [PubMed: 15355760] 13. Mailles A et al. , Long-term Outcome of Patients Presenting With Acute Infectious Encephalitis of Various Causes in France. Clinical Infectious Diseases 54, 1455–1464 (2012). [PubMed: 22460967] 14. Smith MG, Lennette EH, Reames HR, Isolation of the virus of herpes simplex and the demonstration of intranuclear inclusions in a case of acute encephalitis. Am J Pathol 17, 55–68 (1941). [PubMed: 19970544] Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 20 15. Wolman B, Longson M, Herpes encephalitis. Acta Paediatr Scand 66, 243–246 (1977). [PubMed: 190853] 16. Zhang S, et al. , Human inborn errors of immunity to infection affecting cells other than leukocytes: from the immune system to the whole organism. Current opinion in immunology 59, (2019). 17. Puel A et al. , The NEMO Mutation Creating the Most-Upstream Premature Stop Codon Is Hypomorphic Because of a Reinitiation of Translation. Am J Hum Genet 78, 691–701 (2006). [PubMed: 16532398] 18. Dupuis S et al. , Impaired response to interferon-alpha/beta and lethal viral disease in human STAT1 deficiency. Nat Genet 33, 388–391. (2003). [PubMed: 12590259] 19. Casrouge A et al. , Herpes simplex virus encephalitis in human UNC-93B deficiency. Science 314, 308–312 (2006). [PubMed: 16973841] 20. Zhang SY et al. , TLR3 deficiency in patients with herpes simplex encephalitis. Science 317, 1522–1527 (2007). [PubMed: 17872438] 21. Guo Y et al. Herpes simplex virus encephalitis in a patient with complete TLR3 deficiency: TLR3 is otherwise redundant in protective immunity. J Exp Med 208, 2083–2098, doi:jem.20101568 [pii]10.1084/jem.20101568 (2010). 22. Lim HK et al. , TLR3 deficiency in herpes simplex encephalitis: high allelic heterogeneity and recurrence risk. Neurology 83, 1888–1897 (2014). [PubMed: 25339207] 23. Sancho-Shimizu V et al. Herpes simplex encephalitis in children with autosomal recessive and dominant TRIF deficiency. J Clin Invest 121, 4889–4902, doi:59259 [pii]10.1172/JCI59259 (2011). [PubMed: 22105173] 24. Perez de Diego R et al. Human TRAF3 Adaptor Molecule Deficiency Leads to Impaired Toll-like Receptor 3 Response and Susceptibility to Herpes Simplex Encephalitis. Immunity 33, 400–411, doi:S1074–7613(10)00319–5 [pii]10.1016/j.immuni.2010.08.014 (2010). [PubMed: 20832341] 25. Herman M et al. Heterozygous TBK1 mutations impair TLR3 immunity and underlie herpes simplex encephalitis of childhood. J Exp Med, doi:jem.20111316 [pii]10.1084/jem.20111316 (2011). 26. Andersen LL et al. , Functional IRF3 deficiency in a patient with herpes simplex encephalitis. J Exp Med 212, 1371–1379 (2015). [PubMed: 26216125] 27. Bastard P et al. , Herpes simplex encephalitis in a patient with a distinctive form of inherited IFNAR1 deficiency. J Clin Invest 131, (2021). 28. Sancho-Shimizu V et al. , Herpes simplex encephalitis in children with autosomal recessive and dominant TRIF deficiency. J Clin Invest 121, 4889–4902 (2011). [PubMed: 22105173] 29. Casanova JL, Abel L, Mechanisms of viral inflammation and disease in humans. Science 374, 1080–1086 (2021). [PubMed: 34822298] 30. Lafaille FG et al. , Impaired intrinsic immunity to HSV-1 in human iPSC-derived TLR3-deficient CNS cells. Nature. 31. Zimmer BL, S. Y; Smith G; Studer L., Human iPSC-derived trigeminal neurons’ intrinsic immunity to HSV-1 is TLR3-independent. revised manuscript in preparation for PNAS, (2017). 32. Gao D et al. , TLR3 controls constitutive IFN-beta antiviral immunity in human fibroblasts and cortical neurons. J Clin Invest 131, (2021). 33. Zhang SY et al. , Inborn Errors of RNA Lariat Metabolism in Humans with Brainstem Viral Infection. Cell 172, 952–965 e918 (2018). [PubMed: 29474921] 34. Lafaille FG et al. , Human SNORA31 variations impair cortical neuron-intrinsic immunity to HSV-1 and underlie herpes simplex encephalitis. Nat Med 25, 1873–1884 (2019). [PubMed: 31806906] 35. Sancho-Shimizu V et al. , Genetic susceptibility to herpes simplex virus 1 encephalitis in mice and humans. Curr Opin Allergy Clin Immunol 7, 495–505 (2007). [PubMed: 17989525] 36. Abel L et al. , Age-dependent Mendelian predisposition to herpes simplex virus type 1 encephalitis in childhood. J Pediatr 157, 623–629, 629 e621 (2010). [PubMed: 20553844] 37. Kircher M et al. , A general framework for estimating the relative pathogenicity of human genetic variants. Nat Genet 46, 310–315 (2014). [PubMed: 24487276] Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 21 38. Itan Y et al. , The mutation significance cutoff: gene-level thresholds for variant predictions. Nature Methods 13, 109–110 (2016). [PubMed: 26820543] 39. Itan Y et al. , The human gene damage index as a gene-level approach to prioritizing exome variants. Proc Natl Acad Sci U S A 112, 13615–13620 (2015). [PubMed: 26483451] 40. Casanova JL, Conley ME, Seligman SJ, Abel L, Notarangelo LD, Guidelines for genetic studies in single patients: lessons from primary immunodeficiencies. J Exp Med 211, 2137–2149 (2014). [PubMed: 25311508] 41. Karczewski KJ et al. , The mutational constraint spectrum quantified from variation in 141,456 humans. Nature 581, 434–443 (2020). [PubMed: 32461654] 42. Newton K, RIPK1 and RIPK3: critical regulators of inflammation and cell death. Trends Cell Biol 25, 347–353 (2015). [PubMed: 25662614] 43. Cho Y et al. , Phosphorylation-driven assembly of the RIP1-RIP3 complex regulates programmed necrosis and virus-induced inflammation. Cell 137, 1112–1123 (2009). [PubMed: 19524513] 44. Nogusa S et al. , RIPK3 Activates Parallel Pathways of MLKL-Driven Necroptosis and FADD- Mediated Apoptosis to Protect against Influenza A Virus. Cell Host Microbe 20, 13–24 (2016). [PubMed: 27321907] 45. Upton JW, William J Kaiser, and Edward S. Mocarski, Virus inhibition of RIP3-dependent necrosis. Cell Host Microbe 7, 302–313 (2010). [PubMed: 20413098] 46. Guo H, et al. , Herpes simplex virus suppresses necroptosis in human cells. Cell Host Microbe 17, 243–251 (2015). [PubMed: 25674983] 47. Huang Z, et al. , RIP1/RIP3 binding to HSV-1 ICP6 initiates necroptosis to restrict virus propagation in mice. Cell Host Microbe 17, 229–242 (2015). [PubMed: 25674982] 48. He S et al. , Receptor interacting protein kinase-3 determines cellular necrotic response to TNF- alpha. Cell 137, 1100–1111 (2009). [PubMed: 19524512] 49. Yu PW et al. , Identification of RIP3, a RIP-like kinase that activates apoptosis and NFkappaB. Curr Biol 9, 539–542 (1999). [PubMed: 10339433] 50. Wang X et al. , Direct activation of RIP3/MLKL-dependent necrosis by herpes simplex virus 1 (HSV-1) protein ICP6 triggers host antiviral defense. Proc Natl Acad Sci U S A 111, 15438–15443 (2014). [PubMed: 25316792] 51. Guo H et al. , Species-independent contribution of ZBP1/DAI/DLM-1-triggered necroptosis in host defense against HSV1. Cell Death Dis 9, 816 (2018). [PubMed: 30050136] 52. Silke J, Rickard JA, Gerlic M, The diverse role of RIP kinases in necroptosis and inflammation. Nat Immunol 16, 689–697 (2015). [PubMed: 26086143] 53. Orozco S, Oberst A, RIPK3 in cell death and inflammation: the good, the bad, and the ugly. Immunological Reviews 277, 102–112 (2017). [PubMed: 28462521] 54. Mandal P et al. , RIP3 induces apoptosis independent of pronecrotic kinase activity. Mol Cell 56, 481–495 (2014). [PubMed: 25459880] 55. Yang YH, Ma J, Chen YJ, Wu MA, Nucleocytoplasmic shuttling of receptor-interacting protein 3 (RIP3) - Identification of novel nuclear export and import signals in RIP3. J Biol Chem 279, 38820–38829 (2004). [PubMed: 15208320] 56. Weber K, Roelandt R, Bruggeman I, Estornes Y, Vandenabeele P, Nuclear RIPK3 and MLKL contribute to cytosolic necrosome formation and necroptosis. Communications Biology 1, (2018). 57. Sun XQ, Yin JP, Starovasnik MA, Fairbrother WJ, Dixit VM, Identification of a novel homotypic interaction motif required for the phosphorylation of receptor-interacting protein (RIP) by RIP3. J Biol Chem 277, 9505–9511 (2002). [PubMed: 11734559] 58. Wu XN et al. , Distinct roles of RIP1-RIP3 hetero- and RIP3-RIP3 homo-interaction in mediating necroptosis. Cell Death Differ 21, 1709–1720 (2014). [PubMed: 24902902] 59. He SD, Liang YQ, Shao F, Wang XD, Toll-like receptors activate programmed necrosis in macrophages through a receptor-interacting kinase-3-mediated pathway. P Natl Acad Sci USA 108, 20054–20059 (2011). 60. Kaiser WJ et al. , Toll-like receptor 3-mediated necrosis via TRIF, RIP3, and MLKL. J Biol Chem 288, 31268–31279 (2013). [PubMed: 24019532] Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 22 61. Upton JW, Shubina M, Balachandran S, RIPK3-driven cell death during virus infections. Immunol Rev 277, 90–101 (2017). [PubMed: 28462524] 62. Pazdernik NJ, Donner DB, Goebl MG, Harrington MA, Mouse receptor interacting protein 3 does not contain a caspase-recruiting or a death domain but induces apoptosis and activates NF-kappaB. Mol Cell Biol 19, 6500–6508 (1999). [PubMed: 10490590] 63. Seo J et al. , CHIP controls necroptosis through ubiquitylation- and lysosome-dependent degradation of RIPK3. Nature Cell Biology 18, 291–+ (2016). [PubMed: 26900751] 64. Choi SW et al. , PELI1 Selectively Targets Kinase-Active RIP3 for Ubiquitylation-Dependent Proteasomal Degradation. Mol Cell 70, 920–935 e927 (2018). [PubMed: 29883609] 65. Xie YD et al. , Gut epithelial TSC1/mTOR controls RIPK3-dependent necroptosis in intestinal inflammation and cancer. Journal of Clinical Investigation 130, 2111–2128 (2020). [PubMed: 31961824] 66. Sun L, et al. , Mixed Lineage Kinase Domain-like Protein Mediates Necrosis Signaling Downstream of RIP3 Kinase. Cell 148, 213–227 (2012). [PubMed: 22265413] 67. Zhang T et al. , CaMKII is a RIP3 substrate mediating ischemia- and oxidative stress-induced myocardial necroptosis. Nat Med 22, 175–182 (2016). [PubMed: 26726877] 68. Wang L, Du FH, Wang XD, TNF-alpha induces two distinct caspase-8 activation pathways. Cell 133, 693–703 (2008). [PubMed: 18485876] 69. Upton JW, Kaiser WJ, Mocarski ES, DAI/ZBP1/DLM-1 Complexes with RIP3 to Mediate Virus- Induced Programmed Necrosis that Is Targeted by Murine Cytomegalovirus vIRA. Cell Host Microbe 11, 290–297 (2012). [PubMed: 22423968] 70. Meng YX et al. , Human RIPK3 maintains MLKL in an inactive conformation prior to cell death by necroptosis. Nat Commun 12, (2021). 71. Galluzzi L, et al. , Molecular mechanisms of cell death: recommendations of the Nomenclature Committee on Cell Death 2018. Cell Death & Differentiation 25, (2018). 72. Zhang DW et al. , RIP3, an Energy Metabolism Regulator That Switches TNF-Induced Cell Death from Apoptosis to Necrosis. Science 325, 332–336 (2009). [PubMed: 19498109] 73. Cho YS et al. , Phosphorylation-driven assembly of the RIP1-RIP3 complex regulates programmed necrosis and virus-induced inflammation. Cell 137, 1112–1123 (2009). [PubMed: 19524513] 74. Taft J et al. , Human TBK1 deficiency leads to autoinflammation driven by TNF-induced cell death. Cell 184, 4447–4463 e4420 (2021). [PubMed: 34363755] 75. Moriwaki K, Chan FK, Necroptosis-independent signaling by the RIP kinases in inflammation. Cell Mol Life Sci 73, 2325–2334 (2016). [PubMed: 27048814] 76. Fang R et al. , NEMO-IKKbeta Are Essential for IRF3 and NF-kappaB Activation in the cGAS- STING Pathway. J Immunol 199, 3222–3233 (2017). [PubMed: 28939760] 77. Zhao T et al. , The NEMO adaptor bridges the nuclear factor-kappaB and interferon regulatory factor signaling pathways. Nat Immunol 8, 592–600 (2007). [PubMed: 17468758] 78. Daniels BP et al. , RIPK3 Restricts Viral Pathogenesis via Cell Death-Independent Neuroinflammation. Cell 169, 301–313 e3111 (2017). [PubMed: 28366204] 79. Zhang SY, Harschnitz O, Studer L, Casanova JL, Neuron-intrinsic immunity to viruses in mice and humans. Curr Opin Immunol 72, 309–317 (2021). [PubMed: 34425410] 80. Zhang SY, Herpes simplex virus encephalitis of childhood: inborn errors of central nervous system cell-intrinsic immunity. Human Genetics 139, 911–918 (2020). [PubMed: 32040615] 81. Lim HK et al. , Severe influenza pneumonitis in children with inherited TLR3 deficiency. Journal of Experimental Medicine 216, 2038–2056 (2019). [PubMed: 31217193] 82. Zhang Q et al. , Inborn errors of type I IFN immunity in patients with life-threatening COVID-19. Science 370, (2020). 83. Webster JD et al. , RIP1 kinase activity is critical for skin inflammation but not for viral propagation. J Leukocyte Biol 107, 941–952 (2020). [PubMed: 31985117] 84. Daniels BP et al. , The Nucleotide Sensor ZBP1 and Kinase RIPK3 Induce the Enzyme IRG1 to Promote an Antiviral Metabolic State in Neurons. Immunity 50, 64–76 e64 (2019). [PubMed: 30635240] Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 23 85. Casanova JL, Abel L, Human genetics of infectious diseases: Unique insights into immunological redundancy. Seminars in Immunology 36, 1–12 (2018). [PubMed: 29254755] 86. Cuchet-Lourenco D et al. , Biallelic RIPK1 mutations in humans cause severe immunodeficiency, arthritis, and intestinal inflammation. Science 361, 810–813 (2018). [PubMed: 30026316] 87. Li Y et al. , Human RIPK1 deficiency causes combined immunodeficiency and inflammatory bowel diseases. Proc Natl Acad Sci U S A 116, 970–975 (2019). [PubMed: 30591564] 88. Sultan M et al. , Pathogenic Ripk1 Mutations Cause Infantile-Onset Ibd with Inflammatory and Fistulizing Features. Gastroenterology 162, S121–S122 (2022). 89. Casanova JL, Abel L, The human model: A genetic dissection of immunity to infection in natural conditions. Tissue Antigens 64, 346–346 (2004). 90. Casanova JL, Abel L, The human model: a genetic dissection of immunity to infection in natural conditions. Nat Rev Immunol 4, 55–66 (2004). [PubMed: 14704768] 91. Medetgul-Ernar K, Davis MM, Standing on the shoulders of mice. Immunity 55, 1343–1353 (2022). [PubMed: 35947979] 92. Gros P, Casanova JL, Reconciling Mouse and Human Immunology at the Altar of Genetics. Annu Rev Immunol, (2022). 93. Robinson MD, McCarthy DJ, Smyth GK, edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139–140 (2010). [PubMed: 19910308] 94. Gu ZG, Eils R, Schlesner M, Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847–2849 (2016). [PubMed: 27207943] 95. Love MI, Huber W, Anders S, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, (2014). 96. Chan YH et al. , Mutating chikungunya virus non-structural protein produces potent live-attenuated vaccine candidate. Embo Mol Med 11, (2019). 97. Paquet D et al. , Efficient introduction of specific homozygous and heterozygous mutations using CRISPR/Cas9. Nature 533, 125–129 (2016). [PubMed: 27120160] 98. Wang Z, Liu YB, Lin WC, Cui SJ, A real-time PCR to detect and analyze virulent EMCV loads in sows and piglets. Mol Biol Rep 39, 10013–10017 (2012). [PubMed: 22752806] Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 24 Figure 1. Compound heterozygous RIPK3 mutations in a patient with HSE A. Family pedigree with allele segregation of the two RIPK3 mutations. The proband (patient 1, P1), in black, is compound heterozygous for the p.Arg422* (R422) and p. Pro493fs9 (P493fs9) mutations. Each parent is heterozygous for one mutant allele. B. Images of the brain of P1, showing lesions affecting the left insula. C. Graph showing the CADD scores of all homozygous RIPK3 nonsynonymous or essential-splicing variants reported by the gnomAD database, and their minor allele frequency (MAF). MSC 95%: mutation significance cutoff for the 95% confidence interval. D. Schematic representation of the structure of the RIPK3 protein and the impact of the two mutations. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 25 Figure 2. In vitro production and function of the RIPK3 variants upon transient transfection A. Confocal microscopy imaging of HeLa cells 24 h after transfection with wild-type (WT) and mutant RIPK3. Cells were stained with an antibody directed against the N-terminus (N-ter) of RIPK3 and the corresponding Alexa Fluor 488-conjugated secondary antibody (green). The nuclei were stained with DAPI (blue). Scale bar, 10 μm. The results shown are representative of three independent experiments. B. RIPK3 mRNA levels were determined by reverse transcription-quantitative polymerase chain reaction (RT-qPCR) in HEK293T cells 24 h after transfection with a luciferase-expressing vector (Luci), or WT and mutant RIPK3 constructs. We used two probes, targeting exons 2–3 (upper panel) and exons 9–10 (lower panel) of RIPK3. The data shown are the means from two biological replicates from one experiment representative of three independent experiments. C. Immunoblotting analysis of total RIPK3 and autophosphorylated RIPK3 (p-RIPK3, Ser227) levels in HEK293T cells 24 h after transfection with C-terminally Myc-tagged RIPK3 WT and mutant constructs. RIPK3 proteins were detected with antibodies against the N terminus (N-ter) or C terminus (C-ter) of RIPK3, p-RIPK3 (Ser227) and Myc-tag. The results shown are representative of three independent experiments. D. RIPK3 overexpression-mediated NF-κB promoter-driven reporter assay in HEK293T cells, 24 h after transfection with the Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 26 NF-κB reporter plasmid, along with various doses of empty vector (EV), WT and mutant RIPK3 constructs; analysis of luciferase activity. The data shown are the means of at least three biological replicates from three independent experiments. E. Myc-tagged WT and mutant RIPK3 constructs were co-expressed with FLAG-tagged WT and mutant RIPK3 constructs in HEK293T cells, which were then subjected to immunoprecipitation (IP) with anti-Myc antibody-conjugated agarose beads, and immunoblotting with anti-FLAG or anti- Myc antibodies. The results shown are representative of three independent experiments. F. HEK293T cells cotransfected with Myc-tagged RIPK1 and FLAG-tagged WT and mutant RIPK3 WT plasmids, subjected to IP and immunoblotting as in (E). The results shown are representative of three independent experiments. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 27 Figure 3. In vitro production and function of the RIPK3 variants after stable transduction A. RIPK3 mRNA levels were determined by RT-qPCR in parental HT29 cells, or RIPK3 knockout (KO) HT29 cells left non-transfected (NT) or stably transfected with a mock vector (Luci) or with WT or mutant RIPK3 constructs in a lentiviral system. The data shown are the means from three biological replicates from three independent experiments. B. Immunoblotting analysis of RIPK3 protein levels in the parental HT29 cells, and in RIPK3 KO HT29 cells, as in (A). The results shown are representative of three independent experiments. C. Pulse-chase analysis to measure WT and mutant RIPK3 protein degradation. HEK293T cells were transfected with FLAG-tagged WT and mutant RIPK3 plasmids for 24 h. They were then treated with 100 μg/ml cycloheximide (CHX) for the indicated times, and subjected to western blotting (lower panel). The relative amounts of Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 28 RIPK3 were calculated after normalization against GAPDH and are shown in the graph (upper panel). The results shown are representative of three independent experiments. D. Immunoblotting analysis of RIPK3 levels in RIPK3 KO HT29 cells stably expressing WT and mutant RIPK3, treated with protein degradation inhibitors (5 nM bortezomib (BTZ) for 12 h, 10 mM MG132, 50 mM chloroquine diphosphate (CQ), or 10 mg/mL E64d plus 10 mg/mL pepstatin) for 6 h. The red asterisk indicates the bands corresponding to RIPK3. The results shown are representative of three independent experiments. E. Immunoblot analysis of phosphorylated MLKL (p-MLKL, Ser358) in RIPK3 KO HT29 cells stably expressing Luci, WT or mutant RIPK3, treated with DMSO solvent as a control or with PBZ complex (consisting of poly(I:C), BV6 and Z-VAD), or TBZ complex (containing TNF, BV6 and Z-VAD) for the indicated times. The results shown are representative of three independent experiments. F. Viability of RIPK3−/− HT29 cells stably expressing WT and mutant RIPK3 constructs, treated with DMSO solvent (control), or with poly(I:C), TNF, PB (poly(I:C) + BV6), TB (TNF + BV6), PBZ or TBZ complex for the indicated times. The results shown are the means ± SD from three biological replicates from one experiment, representative of three independent experiments. P values were obtained by one-way ANOVA and Tukey’s multiple comparison tests, and P values are indicated for the comparison of R422 or P493fs9* RIPK3 transduced cells with WT RIPK3 transduced cells. ns-not significant, P<0.05, P<0.01, *P<0.0001. G. Viability of RIPK3−/− HT29 cells stably expressing WT and mutant RIPK3 constructs, treated with DMSO solvent (control), or with L (LPS), LB (LPS + BV6) or LBZ (LPS + BV6 + Z-VAD) complex for 24 h. The results shown are the means ± SD from three biological replicates from one experiment, representative of two independent experiments. P values were obtained by one-way ANOVA and Tukey’s multiple comparison tests, and the corresponding P values are indicated. *P<0.0001. H. Viability of RIPK3 KO HT29 cells stably co-expressing FLAG-tagged DAI with WT and mutant RIPK3, left non-infected (NI), or after infection with HSV-1 FmutRHIM (MOI=5) for 24 h. The data shown are means ± SEM from two independent experiments, with three biological replicates per experiment. P values were obtained by one-way ANOVA and Tukey’s multiple comparison tests, and the corresponding P values are indicated. P<0.05, **P<0.0001. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 29 Figure 4. Impaired RIPK3 production and function in P1’s cells A. RIPK3 mRNA levels were measured by RT-qPCR in EBV-B cells (EBV-B), SV40- fibroblasts (SV40-F) and primary fibroblasts (Primary-F) from healthy controls (Ctrls) and P1, with two probes targeting exons 2–3 (upper panel) and exons 9–10 (lower panel) of RIPK3. The results data shown are the means of four biological replicates from two independent experiments. B. Relative abundance (in percentages) of the RIPK3 cDNA generated from mRNA extracted from SV40-F from P1, assessed by TOPO-TA cloning. C. Immunoblot analysis of endogenous RIPK3 levels in EBV-B, SV40-F and Primary-F from healthy controls (C1, C2, C3) and P1, with antibodies against the N-terminus and C-terminus of RIPK3. The results shown are representative of more than three independent experiments. D. Immunoblot analysis of endogenous RIPK3 levels in SV40-F from healthy Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 30 controls (C1, C2) and P1 treated with protein degradation inhibitors (5 nM BTZ for 12 h, 10 mM MG132, 50 mM CQ, or 10 mg/mL E64d plus 10 mg/mL pepstatin) for 6 h. The red asterisks indicates the bands corresponding to RIPK3. The results shown are representative of three independent experiments. E. Immunoblot analysis of p-MLKL levels in SV40-F from a healthy control (C1) and P1, either left non-transfected (NT) or transiently transfected with Luci or WT RIPK3 for 24 h, and then stimulated with PBZ or TBZ for 4 h. The red asterisks indicate the bands corresponding to RIPK3. The results shown are representative of three independent experiments. F. Immunoblot analysis of full-length and cleaved caspase 3 in SV40-F from healthy controls and P1, treated for 8 h with PB complex containing poly(I:C) and BV6, or TB complex containing TNF and BV6. The results shown are representative of three independent experiments. G. Viability of primary fibroblasts from healthy controls (Ctrls, n=3) and P1, either non- transduced or transduced with Luci, WT, variants with P1’s mutations or D142N RIPK3 lentiviruses for 48 h, then treated with DMSO solvent (D), or with poly(I:C), PB or PBZ complex for the times indicated. The results shown are the means ± SEM from three independent experiments (transduction with WT or P1 mutant RIPK3) and two independent experiments (for D142N RIPK3), with three biological replicates per experiment. H. Viability of primary fibroblasts from healthy controls (Ctrls, n=3) and P1, either non-transduced or transduced with Luci, WT, variants with P1’s mutations or D142N RIPK3 lentiviruses for 48 h, and treated with DMSO solvent (D), TNF, TB or TBZ complex for the times indicated. The results shown are the means ± SEM from three independent experiments (for WT or P1 mutant RIPK3) and two independent experiments (for D142N RIPK3), with three biological replicates per experiment. In G and H, P values were obtained by one-way ANOVA with Tukey’s multiple comparison tests, and the corresponding P values are indicated. ns-not significant, P<0.001, **P<0.0001. I-J. Immunoblot analysis of p-MLKL in SV40-F from a healthy control (C1), P1, other HSE patients (TLR3−/−, TRIF−/−, UNC93B1−/−), and a TBK1−/− patient, after stimulation with PBZ (I) or TBZ (J) for 4 h. The results shown are representative of three independent experiments. K-L. Immunoblot analysis of full-length and cleaved caspase 3 in SV40-F from a healthy control, P1, TLR3−/−, TRIF−/−, UNC93B1−/−, or TBK1−/− patients, after stimulation with PB (K) or TB (L) for 8 h. The results shown are representative of three independent experiments. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 31 Figure 5. Intact signaling via the TLR3- and TNFR1-dependent NF-κB, IRF3 and MAPK pathways in P1 fibroblasts A. Immunoblot analysis of total and phosphorylated P65, IRF3, ERK1/2 and JNK1/2 in SV40-F from healthy controls (C1, C2), P1, TLR3−/− and NEMO−/− patients, after stimulation with 25 μg/ml poly(I:C) for the times indicated. The results shown are representative of three independent experiments. B. Immunoblot analysis of total and phosphorylated P65, ERK1/2 and JNK1/2 in SV40-F from healthy controls, P1 and a NEMO−/− patient, after stimulation with 20 ng/ml TNF for the times indicated. The results shown are representative of three independent experiments. C. Immunoblot analysis of total and phosphorylated IRF3 in SV40-F from healthy controls, P1, TLR3−/− and NEMO−/− patients, after stimulation with 20 ng/ml TNF for 24 h. The results shown are representative of three independent experiments. D. SV40-F from healthy controls (Ctrls, n=3), P1 and TLR3−/− and NEMO−/− HSE patients were left unstimulated (NS) or were stimulated with various doses of poly(I:C) alone, Lipofectamine alone (Lipo), or both (poly(I:C)+Lipo), for 24 h. The amounts of IFN-β, IFN-λ1, IL-6 and CCL3 in culture supernatants were determined with Legendplex cytometric bead arrays. The results shown are the means ± Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 32 SEM from two independent experiments, with three biological replicates per experiment. Each dot represents one biological replicate. P values were obtained by one-way ANOVA with Tukey’s multiple comparison tests, and the corresponding P values are indicated. ns-not significant, P<0.05, P<0.01, *P<0.0001. Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 33 Figure 6. Enhanced susceptibility of RIPK3-deficient fibroblasts and hPSC-derived cortical neurons to HSV-1 A. Principal component analysis of the transcriptome of human primary fibroblasts without (NS) or with HSV-1 infection for 24 hours, in cells from healthy controls (Ctrls, n=3), P1 and other patients with recessive TLR3, IFNAR1 or NEMO deficiency. B. Genes differentially expressed between HSV-1 and NS in human primary fibroblasts, as in (A). Heatmap including 3060 genes with relative absolute fold-changes in expression > 2 (in all three healthy controls) in response to HSV-1 relative to NS samples. C. SV40-F from healthy controls (Ctrls, n=3), P1 and other patients with recessive TLR3, IFNAR1 or NEMO deficiencies were left untreated or were treated with IFN-β for 24 h, and then infected with HSV-1 (MOI = 0.001). Virus replication was then evaluated at the indicated Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 34 timepoints post infection. HSV-1 replication was quantified by the TCID50 virus titration method. The data shown are the means ± SEM of two independent experiments with two biological replicates per experiment. P values were obtained by one-way ANOVA with Tukey’s multiple comparison tests, and the P values for the 24 h and 48 h time points are indicated for the comparison of P1’s cells with control cells. P<0.05. D-G, Virus replication levels for VSV (D), measles virus (MeV) (E), influenza A virus (IAV) (F), and EMCV (G) in SV40-F, as in (C), at the indicated times post infection with VSV (MOI = 0.1), MeV (MOI = 0.5), IAV (MOI = 10), or EMCV (MOI = 0.01), as assessed by the TCID50 virus titration method (VSV and MeV), plaque assay (IAV), or expression levels of the three-dimensional region of the EMCV genome, as measured by RT-qPCR. The data shown are the means of two to four biological replicates from two (IAV and EMCV) or four (VSV, MeV) independent experiments. H. RIPK3 mRNA levels, as measured by RT-qPCR, in cortical neurons differentiated from the hPSCs of healthy controls and P1. We used two probes, targeting exons 2–3 (left) and exons 9–10 (right) of RIPK3. The data shown are the means of two biological replicates from one experiment. I. Relative abundance (in percentages) of the RIPK3 cDNA generated from mRNA extracted from hPSC-derived cortical neurons from a healthy control and P1, assessed by TOPO-TA cloning. J. hPSC- derived cortical neurons from a healthy control, P1 and other HSE patients with AR TLR3 or IFNAR1 deficiencies, with or without IFN-β pretreatment for 24 h, were infected with HSV-1 (MOI=0.001) and virus replication levels were measured at the indicated timepoints post infection. HSV-1 replication was quantified by the TCID50 virus titration method. The data shown are means ± SEM for four independent experiments (a healthy control, P1 and TLR3-deficienct cells) and two independent experiments (IFNAR1-deficiency cells). P values were obtained by one-way ANOVA with Tukey’s multiple comparison tests, and the P values for the 72 h time points are indicated for the comparison of P1’s cells with control cells. ***P<0.0001. K. Viability of hPSC-derived cortical neurons from a healthy control, P1 and a TLR3−/− HSE patient, left non-infected (NI), or infected with HSV-1 (MOI=0.001) for the times indicated. The data shown are means ± SEM for two independent experiments, with three biological replicates per experiment. P values were obtained by one-way ANOVA with Tukey’s multiple comparison of P1’s cells with control cells, and the corresponding P values are indicated. P<0.05, *P<0.001. L. RIPK3 mRNA levels, as measured by RT-qPCR, in cortical neurons from parental and RIPK3 KO hPSCs. We used two probes, targeting exons 2–3 (left) and exons 9–10 (right) of RIPK3. The data shown are the means of three biological replicates from one experiment. M. hPSC-derived cortical neurons derived from parental healthy control cells, RIPK3 KO cells and HSE patients with AR TLR3 were infected with HSV-1 (MOI=0.001) for the times indicated. HSV-1 replication was quantified by the TCID50 virus titration method. The data shown are means ± SEM for three independent experiments (the parental cells and RIPK3 KO cells) and two independent experiments (TLR3-deficient cells). P values were obtained by one-way ANOVA with Tukey’s multiple comparison tests, and the P value for the 48 h time point is indicated for the comparison of RIPK3 KO cells with control parental cells. P<0.01. N. Viability of hPSC-derived cortical neurons from healthy parental control cells, RIPK3 KO cells and a TLR3−/− HSE patient, left non-infected, or infected with HSV-1 (MOI=0.001) for the times indicated. The data shown are means ± SEM for two independent experiments, with three biological replicates per experiment. P values were obtained by Sci Immunol. Author manuscript; available in PMC 2023 July 21. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 35 one-way ANOVA with Tukey’s multiple comparison tests, and the P values for the indicated timepoints are indicated for the comparison of RIPK3 KO cells with control parental cells. The corresponding P values are indicated. ns - not significant, *P<0.05. Sci Immunol. Author manuscript; available in PMC 2023 July 21.
10.1093_molbev_msad085	Evolution of the SARS-CoV-2 Mutational Spectrum Jesse D. Bloom ,,1,2,3 Annabel C. Beichman,2 Richard A. Neher ,4,5 and Kelley Harris2 1Basic Sciences Division and Computational Biology Program, Fred Hutchinson Cancer Center, Seattle, WA 2Department of Genome Sciences, University of Washington, Seattle, WA 3Howard Hughes Medical Institute, Seattle, WA 4Biozentrum, University of Basel, Basel, Switzerland 5Swiss Institute of Bioinformatics, Lausanne, Switzerland Corresponding author: E-mail: [email protected]. Associate editor: Crystal Hepp Abstract SARS-CoV-2 evolves rapidly in part because of its high mutation rate. Here, we examine whether this mutational pro- cess itself has changed during viral evolution. To do this, we quantify the relative rates of different types of single-nu- cleotide mutations at 4-fold degenerate sites in the viral genome across millions of human SARS-CoV-2 sequences. We find clear shifts in the relative rates of several types of mutations during SARS-CoV-2 evolution. The most striking trend is a roughly 2-fold decrease in the relative rate of G→T mutations in Omicron versus early clades, as was recently noted by Ruis et al. (2022. Mutational spectra distinguish SARS-CoV-2 replication niches. bioRxiv, doi:10.1101/ 2022.09.27.509649). There is also a decrease in the relative rate of C→T mutations in Delta, and other subtle changes in the mutation spectrum along the phylogeny. We speculate that these changes in the mutation spectrum could arise from viral mutations that affect genome replication, packaging, and antagonization of host innate-immune factors, although environmental factors could also play a role. Interestingly, the mutation spectrum of Omicron is more similar than that of earlier SARS-CoV-2 clades to the spectrum that shaped the long-term evolution of sarbecoviruses. Overall, our work shows that the mutation process is itself a dynamic variable during SARS-CoV-2 evolution and suggests that human SARS-CoV-2 may be trending toward a mutation spectrum more similar to that of other animal sarbecoviruses. Key words: SARS-CoV-2, mutation rate, equilibrium frequencies, mutational spectrum. Introduction The evolution of SARS-CoV-2 is enabled in part by the high underlying rate at which mutations arise in the viral gen- ome during replication. Coronaviruses (and other mem- bers of the nidovirus order) are the only RNA viruses known to have a proofreading mechanism in their RNA-dependent RNA polymerase (Denison et al. 2011; Ogando et al. 2019), but despite that proofreading, coro- naviruses still have mutation rates that dwarf those of cel- lular organisms by several orders of magnitude (Drake 1993; Peck and Lauring 2018). A r t i c l e Europeans and South Asians (Harris 2015; Speidel et al. 2021). The mutation spectrum also diverged more grad- ually during human evolution in Africa and East Asia, re- sulting in profiles that are sufficiently distinctive to identify an individual’s continent of origin. Populations of great apes, mice, and yeast have similarly distinctive mu- tational processes (Lindsay et al. 2019; Jiang et al. 2021; Goldberg and Harris 2022). It remains unclear how much these changes are due to the evolution of the underlying genome-replication machinery versus changes in life his- tory or exposure to environmental mutagens (Mathieson and Reich 2017; Macià et al. 2021; Ruis, Peacock, et al. 2022; Ruis, Weimann, et al. 2022); however, in a few cases, changes in the mutation spectrum have been linked to heritable genetic change affecting the function or expres- sion of proteins involved in genome replication or repair (Couce et al. 2013; Jiang et al. 2021; Robinson et al. 2021; Kaplanis et al. 2022; Sasani et al. 2022). For viruses, the mu- tational process can also be affected by a virus’s ability to evade host innate-immune proteins that mutagenize viral nucleic acids (Sadler et al. 2010; De Maio et al. 2021; Ratcliff and Simmonds 2021; Ringlander et al. 2022). For coronaviruses like SARS-CoV-2, genes encoding pro- teins involved in genome replication and innate-immune Open Access Studies of cellular organisms ranging from bacteria to humans have shown that the mutational process itself can change during evolution (Hwang and Green 2004; Sung, Tucker, et al. 2012; Couce et al. 2013; Long et al. 2015). Many studies of changes in the mutational process during natural evolution focus on the mutation spectrum which represents the distribution of relative rather than absolute rates of different types of mutations. For instance, humans experienced a transient increase in the relative rate of C→T mutations in certain sequence contexts, which affected a 10,000-year-old population of Anatolian hunter-gatherers and spread via gene flow to all living © The Author(s) 2023. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/ licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Mol. Biol. Evol. 40(4):msad085 https://doi.org/10.1093/molbev/msad085 Advance Access publication April 11, 2023 1 Bloom et al. · https://doi.org/10.1093/molbev/msad085 antagonism constitute a substantial fraction of the gen- ome (Ziebuhr 2005; V’kovski et al. 2021), providing an am- ple target for mutations that could potentially alter the mutation process itself. In artificial lab settings, researchers have isolated coronavirus variants with mutations in genome-replication proteins that have dramatically al- tered mutation rates (Eckerle et al. 2007, 2010). However, it is unclear how such mutator variants generally fare during natural evolution (Peck and Lauring 2018). A recent preprint by Ruis, Weimann, et al. (2022) on patho- genic bacterial mutagenesis identified several mutation types whose relative rates correlate with replication niche within the human body. The authors found that bacterial replication within the lower respiratory tract correlated with an in- creased load of G→T mutations, which prompted them to hypothesize that the Omicron lineage of SARS-CoV-2 would have a reduced G→T rate relative to earlier SARS-CoV-2 lineages that may replicate more in the lungs (Ruis, Peacock, et al. 2022). Consistent with this hypothesis, they found a reduced relative number of G→T mutations across all sites for Omicron clades of SARS-CoV-2. Since their study pooled all mutations (nonsynonymous and synonymous), it is not clear the extent to which the signal could be affected by protein-level selection as well as the underlying rate of mu- tation. It is also unclear whether the difference in G→T frac- tion between Omicron and other SARS-CoV-2 viruses is the dominant feature of the SARS-CoV-2 mutational landscape or just one component of the sort of continuous variation that has been observed in cellular organisms. Here, we systematically analyze changes in the relative rates of all single-nucleotide mutation types among differ- ent clades of human SARS-CoV-2. To disentangle under- lying mutation rates from the subsequent action of natural selection, we restrict our analysis to only 4-fold de- generate sites where all mutations are expected to be neu- tral with respect to protein function. We also use rigorous quality control to ensure that our estimates are not biased by technical artifacts related to sequencing or base-calling errors. Using this approach, we confirm that Omicron has a roughly 2-fold decrease in the relative rate of G→T muta- tions relative to early clades. We also find additional shifts in the mutation spectrum, including a decrease in C→T mutations in Delta and a broader correlation between mu- tation spectrum divergence and genetic divergence across the SARS-CoV-2 phylogeny. While our analysis does not de- termine in SARS-CoV-2’s mutational spectrum, the pervasive and phylogenetically correlated nature of the shifts suggests that viral mutations affecting genome replication, pack- aging, and innate-immune antagonism could play a role. the evolutionary the cause of shifts Results Different Clades of Human SARS-CoV-2 Have Different Mutation Spectra We focused our analysis on the roughly 6-million publicly available SARS-CoV-2 sequences in the pre-built UShER 2 MBE phylogenetic tree (McBroome et al. 2021; Turakhia et al. 2021). Each of these sequences represents the consensus sequence of a virus that infected a human individual. We counted the occurrence of each mutation on the branches of the phylogenetic tree (McBroome et al. 2021; Turakhia et al. 2021): these counts represent the number of occur- rences of mutations, not how often the mutations are found in the final sequence alignment (in other words, a mutation that occurred once but then is shared in several sequences by common descent is only counted once). We for each Nextstrain clade tallied counts separately (Aksamentov et al. 2021) and used a variety of quality- control steps to remove sequences and sites likely to be affected by spurious mutations from sequencing or base- calling artifacts (see Materials and Methods). Note that what we count as “mutations” in this approach represent changes that are fixed at the intra-host consensus level, al- though the vast majority never fix them globally. Counting mutations that have fixed intra-host will reflect the under- lying mutation process at sites where mutations are neu- tral, but not at sites where they are not neutral (Kimura 1968). These data are analogous to the polymorphism data that have been used to infer variation of mutational processes within other species (Harris 2015; Harris and Pritchard 2017; Jiang et al. 2021; Goldberg and Harris 2022). Prior analyses of SARS-CoV-2 mutation rates have gener- ally focused on all nucleotide mutations (Neher 2022; Ruis, Peacock, et al. 2022). However, many sites in the viral gen- ome are under strong functional selection, and so the mu- tational patterns at those sites will represent the combined action of mutation and selection. We, therefore, focused our analysis only on 4-fold degenerate sites (sites at the third position in codons where all three possible nucleotide mutations are synonymous), under the assumption that mutations at such sites will tend to be nearly neutral. The SARS-CoV-2 genome has approximately 4,240 such sites (with the exact number differing slightly among viral clades), and we restricted our analysis to only clades with at least 5,000 mutations at such sites (table 1). Table 1. Number of 4-fold Degenerate Synonymous Sites and Total Mutations at Those Sites for the Clades Analyzed Here. Clade Four-fold Degenerate Sites Total Mutations at These Sites 20A 20B 20C 20E 20G 20I 21C 21I 21J 21K 21L 22A 22B 22C 4,247 4,247 4,246 4,246 4,243 4,243 4,245 4,241 4,239 4,241 4,235 4,236 4,234 4,233 17,202 14,121 9,344 10,454 14,019 60,858 6,308 24,117 282,051 113,721 83,475 11,413 64,765 18,958 NOTE.—We only analyzed clades with at least 5,000 mutations at 4-fold degenerate sites. SARS-CoV-2 Mutational Spectrum · https://doi.org/10.1093/molbev/msad085 MBE There were clear differences in the mutation spectrum at 4-fold degenerate sites across viral clades (fig. 1A and interactive plot at https://jbloomlab.github.io/SARS2- mut-spectrum/pca.html). The largest difference was be- tween the Omicron clades and all other clades, but the Delta clades also showed a unique pattern. Importantly, these clade-to-clade differences were robust to analyzing sequences only from specific geographical locations, ex- cluding the most heavily mutated sites, or analyzing each half of the viral genome separately (supplementary fig. S1, Supplementary Material online and https://jbloomlab. github.io/SARS2-mut-spectrum/). The biggest difference between Omicron and other clades was a roughly 2-fold decrease in the rate of G→T mutations (fig. 1B and C and interactive plot at https:// jbloomlab.github.io/SARS2-mut-spectrum/rates-by-clade. html), consistent with a recent study (Ruis, Peacock, et al. 2022) that analyzed all sites (synonymous and nonsynon- ymous). There was also a clear decrease in the rate of C→T mutations in Delta (fig. 1B and C). Some other types of mu- tations with lower rates also showed substantial differ- ences among clades (this is seen most easily by clicking on specific mutation types in the interactive plot at https://jbloomlab.github.io/SARS2-mut-spectrum/rates- by-clade.html). Note also that we confirm previous find- ings that the two types of mutations with the highest rates are C→T transitions and G→T transversions (De Maio et al. 2021). The Mutation Spectrum Has Phylogenetic Signal Beyond G→T Mutations The clade-to-clade differences in relative mutation rates have a visually obvious phylogenetic pattern (fig. 1D). To statistically confirm the visual impression of phylogenetic patterns in the mutation rates, we used Mantel tests (Mantel 1967; Harmon and Glor 2010; Hardy and Pavoine 2012; Legendre and Legendre 2012) to compare the phylo- genetic distances between clades with the differences in their relative mutation rates (fig. 2). These tests showed that the relative mutation rates were indeed significantly correlated with the phylogenetic distances between clades. The correlations remained significant even if we excluded G→T or C→T mutations individually (although not to- gether), or analyzed only Omicron or non-Omicron clades (fig. 2). These results show that the evolution of the muta- tion spectrum goes beyond simply a change in the relative rate of G→T mutations in Omicron. The G→T mutation fraction change observed in Omicron (from an ancestral fraction of about 15% to a de- rived fraction of about 8%, see fig. 1B) could be the result of a 2-fold decrease in the absolute rate of G→T mutations in this lineage if the rates of all other mutations stayed ap- proximately constant. Such a rate change would imply that the overall Omicron mutation rate is about 7% lower than the mutation rate of non-Omicron lineages. More complicated scenarios are also possible, such as an increase in the rates of all non-G→T mutation types in Omicron or compensatory increases and decreases of different muta- tion types that leave the overall rate unchanged. The exist- ence of phylogenetic signal in the mutation spectrum of non-G→T mutations suggests that the rates of multiple mutation types likely changed over time, but none of these shifts necessarily imply a detectable change in the overall SARS-CoV-2 mutation rate. SARS-CoV-2’s Mutation Spectrum Is Becoming More Similar to the Mutation Spectra of Other Sarbecoviruses In the absence of natural selection, the nucleotide com- position of a gene sequence should eventually reach a stable “equilibrium” nucleotide frequency distribution that is determined by its mutation spectrum (Felsenstein 2003). If we assume that the nucleotides at 4-fold degener- ate sites are not under selection, then the actual observed frequencies of nucleotides at these sites should be similar to the equilibrium frequencies predicted by the mutation spectrum if the virus has been evolving with the same mu- tation spectrum for a sufficiently long period of time. We calculated the predicted equilibrium frequencies of nucleotides from the mutation spectra of the various hu- man SARS-CoV-2 clades (fig. 3A). Because the mutation spectra differ somewhat among clades, the predicted equi- librium nucleotide frequencies also differ among clades: for instance, Omicron’s mutation spectrum implies a some- what lower equilibrium frequency of T nucleotides than earlier clades, in part because Omicron has a lower rate of G→T mutations. We compared these predicted equilibrium frequencies from the SARS-CoV-2 clades’ mutation spectra to the ac- tual frequencies of nucleotides observed at 4-fold degener- ate sites in various sarbecoviruses (the subgenus of coronaviruses to which SARS-CoV-2 belongs). As can be seen from figure 3B, the nucleotide frequencies at 4-fold degenerate sites are similar among SARS-CoV-2, two close relatives (e.g., RaTG13 and BANAL-52), and more diverged sarbecoviruses such as SARS-CoV-1 and BtKY72, suggest- ing that the long-term evolution of all these viruses has been shaped by a similar mutation spectrum. However, the equilibrium nucleotide frequencies pre- dicted by the mutation spectrum of human SARS-CoV-2 are quite different from the actual frequencies observed in SARS-CoV-2 and other sarbecoviruses (fig. 3A and B). Some of this difference could be due to natural selection even at 4-fold degenerate sites, or flanking context de- pendence in mutation rates not incorporated into our analysis. To determine if SARS-CoV-2 is unusual in having a large disparity between the empirically observed nucleo- tide frequencies at 4-fold degenerate sites and the equilib- rium frequencies predicted from the mutation spectrum, we performed a similar analysis for a number of other hu- man viruses, including influenza, RSV, dengue, and entero- viruses (fig. 3C; supplementary fig. S2, Supplementary Material online). For all of these other viruses, the empir- ical frequencies and prediction equilibrium frequencies 3 Bloom et al. · https://doi.org/10.1093/molbev/msad085 MBE FIG. 1. Mutation spectrum of SARS-CoV-2 clades at 4-fold degenerate sites. (A) Principal component analysis (PCA) of mutation spectra of Nextstrain clades with sufficient sequences (each point is a clade). (B) Fraction of mutations at 4-fold degenerate sites that are of each type for each clade. (C ) Relative rates of each mutation type, calculated as the fraction of mutations of that type divided by the fraction of sites with the parental nucleotide. (D) Phylogenetic tree of clade founder sequences, with plots showing mutation rates for that clade (ordered as in C) minus rates for clade 20A. Interactive versions of A–C at https://jbloomlab.github.io/SARS2-mut-spectrum/ enable easier identification of individual clades. Supplementary figure S1, Supplementary Material online, shows that the PCA is robust to subsetting on sequences from different geographic locations, excluding top mutations, and partitioning the genome. The numerical values in (B) and (C ) are at https://github. com/jbloomlab/SARS2-mut-spectrum/blob/main/results/synonymous_mut_rates/rates_by_clade.csv. were more similar than for SARS-CoV-2, especially for the from early pre-Omicron clades. mutation spectrum Therefore, early SARS-CoV-2 is unusual among human viruses in having a mutation spectrum that is relatively dif- ferent from its actual nucleotide frequencies at putatively neutral sites. The actual observed nucleotide frequencies of both SARS-CoV-2 and other sarbecoviruses are more similar to the equilibrium nucleotide frequencies implied by the mutation spectra of the Omicron clades are more similar than the frequencies implied by the spectra of earlier hu- man SARS-CoV-2 clades (note how the Omicron clades 4 in fig. 3A look more similar to fig. 3B). The long-term evo- lution of all these sarbecoviruses occurred in bats, and it is possible that some aspect of replication in humans altered the mutation spectrum of SARS-CoV-2 and is now shifting in Omicron back to a spectrum more similar to that of bat coronaviruses. Putative Associations of Protein-Coding Mutations With Changes in the Mutation Spectrum Mutation spectrum changes could potentially be caused by clade-specific amino-acid mutations in viral proteins SARS-CoV-2 Mutational Spectrum · https://doi.org/10.1093/molbev/msad085 MBE FIG. 2. The changes in relative mutation rates correlate with the phylogenetic relationships among clades. The plots show the correlation be- tween the square root of the phylogenetic distance separating each pair of clades and the Euclidean distance between the relative mutation rates for those clades. Assuming that mutation rates evolve neutrally according to a Brownian motion model, mutation rate distances should scale linearly with the square root of phylogenetic distance. The P-values are computed using the Mantel test with 100,000 permutations. The plots show that the mutation rates are significantly correlated (P < 0.05) with phylogenetic distance even if we exclude G→T or C→T mutations individually (although not together) or do the analysis only among Omicron or non-Omicron clades. involved in genome replication, packaging, or antagoniza- tion of host-cell innate-immune proteins that mutagenize foreign nucleic acids (De Maio et al. 2021; Ratcliff and Simmonds 2021; Ringlander et al. 2022). To explore the plausibility of this hypothesis, we tabulated the non-spike amino-acid mutations in each clade relative to the early 20A (B.1) clade (table 2) and identified several viral amino-acid mutations that could speculatively affect the mutation spectrum. The Omicron clades all share mutation I42V in nsp14 (also known as ExoN), which provides proofreading activ- ity during genome replication (Denison et al. 2011; Ogando et al. 2020). Rare polymerase proofreading defects have recently been shown to perturb the G→T mutation rate in human cells (Robinson et al. 2021). The Omicron clades also share mutation P13L in the nucleoprotein, which is part of the genome replication complex and en- capsidates viral RNA (Bessa et al. 2022), and P132H in nsp5, which proteolytically processes the polyprotein en- coding the viral replicase (Roe et al. 2021) and helps antag- onize innate immune responses (Liu et al. 2021). The Delta clades share mutation G671S in the viral polymerase nsp12 (Kirchdoerfer and Ward 2019) as well as several mutations in the nucleoprotein and a mutation in the ORF3a protein that may play a peripheral role in viral replication (Zhang et al. 2022). The Delta clades also share mutations in the ORF7a (V82A and T120I) and nsp13 (P77L) proteins in- volved in innate-immune antagonization (Cao et al. 2021; Fung et al. 2022), which could be relevant as Delta has a decreased relative rate of the C→T mutations, which is the type of change induced by host-cell APOBEC innate-immune proteins (De Maio et al. 2021; Ratcliff and Simmonds 2021). Note also that noncoding mutations or indels (which are not listed in table 1) could also affect the mutation spectrum if they alter the expression of viral genes. However, we also emphasize that there is no direct evi- dence that any of the above mutations actually cause changes in the mutation spectrum, and they could just be associated with clades with different spectra by chance of shared ancestry. In particular, due to the phylogenetic structure of the SARS-CoV-2, sequences in clades share mutations like the ones described above by common des- cent, and so it is not possible to perform meaningful stat- istical association tests because there are not sufficient independent occurrences of clades with each mutation (Felsenstein 1985). Discussion We have demonstrated that there are clear shifts in the mutation spectrum during the evolution of SARS-CoV-2. We corroborate the findings of Ruis, Peacock, et al. (2022) that Omicron has a lower relative rate of G→T mu- tations, but we also show that the changes in the mutation spectrum are not restricted to this one type of mutation. 5 Bloom et al. · https://doi.org/10.1093/molbev/msad085 MBE FIG. 3. Predicted equilibrium frequencies from mutation rates versus actual nucleotide frequencies at 4-fold degenerate sites in sarbecoviruses. (A) Predicted equilibrium frequencies of nucleotides at 4-fold degenerate sites as calculated from the mutation rates for all of the SARS-CoV-2 clades analyzed here. (B) Actual frequencies of nucleotides at 4-fold degenerate sites in various sarbecoviruses. (C ) Manhattan between pre- dicted equilibrium frequencies (from mutation rates) and actual empirically observed nucleotide frequencies at 4-fold degenerate sites for a variety of viruses. SARS-CoV-2 clades are shown in purple squares. See supplementary figure S2, Supplementary Material online for per- nucleotide frequencies for the viruses shown in C. Instead, there are also significant phylogenetically corre- lated shifts in the spectrum among other mutation types, and among both Omicron and non-Omicron clades. In this sense, changes in the SARS-CoV-2 mutation spectrum ap- pear to involve the type of pervasive evolutionary shifts that have been observed among many cellular organisms (Harris 2015; Lindsay et al. 2019; Jiang et al. 2021; Goldberg and Harris 2022). Our analysis cannot determine why the mutation spec- trum differs among clades, although our approach of cal- culating the rates at only 4-fold degenerate sites does rule out confounding effects of protein-level selection. Ruis, Peacock, et al. (2022) proposed that the lower rate of G→T mutations in Omicron is due to a shift to viral rep- lication in the upper rather than lower airways. This is cer- tainly possible, but we suggest that the differences may instead be driven by mutations elsewhere in the viral gen- ome. For instance, Omicron and Delta have clade-specific mutations in proteins involved in genome replication, packaging, and innate-immune antagonism. The latter fac- tor could be important since some mutations (such as the C→T mutations that are relatively rarer in Delta) can be induced by host-cell innate-immune proteins (De Maio et al. 2021; Ratcliff and Simmonds 2021). Ruis et al. accur- ately point out that Omicron does not have any amino-acid mutations in the active site of core genome 6 SARS-CoV-2 Mutational Spectrum · https://doi.org/10.1093/molbev/msad085 MBE Table 2. Non-spike Amino-acid Mutations in the Founder Sequence for Each Clade Relative to the Early 20A (B.1) Clade. Clade 20B (B.1.1) 20C (B.1.367) 20E (B.1.177) 20G (B.1.2) 20I (Alpha) 21C (Epsilon) Non-spike Amino-acid Mutations Relative to Clade 20A N: R203K, G204R ORF3a: Q57H; nsp2: T85I N: A220V; ORF10: V30L N: P67S, P199L; ORF3a: Q57H, G172V; ORF8: S24L; nsp14: N129D; nsp16: R216C; nsp2: T85I; nsp5: L89F N: D3L, R203K, G204R, S235F; ORF8: Q27*, R52I, Y73C; nsp3: T183I, A890D, I1412T N: T205I; ORF3a: Q57H; nsp13: D260Y; nsp2: T85I 21I (Delta) M: I82T; N: D63G, R203M, D377Y; ORF3a: S26L; ORF7a: V82A, T120I; nsp12: G671S; nsp13: P77L; nsp3: P822L; nsp4: A446V; nsp6: V149A, T181I 21J (Delta) M: I82T; N: D63G, R203M, G215C, D377Y; 21K (Omicron BA.1) 21L (Omicron BA.2) 22A (Omicron BA.4) 22B (Omicron BA.5) 22C (Omicron BA.2.12.1) ORF3a: S26L; ORF7a: V82A, T120I; ORF7b: T40I; nsp12: G671S; nsp13: P77L; nsp14: A394V; nsp3: A488S, P1228L, P1469S; nsp4: V167L, T492I; nsp6: T77A E: T9I; M: D3G, Q19E, A63T; N: P13L, R203K, G204R; nsp14: I42V; nsp3: K38R, A1892T; nsp4: T492I; nsp5: P132H; nsp6: I189V E: T9I; M: Q19E, A63T; N: P13L, R203K, G204R, S413R; ORF3a: T223I; ORF6: D61L; nsp1: S135R; nsp13: R392C; nsp14: I42V; nsp15: T112I; nsp3: T24I, G489S; nsp4: L264F, T327I, L438F, T492I; nsp5: P132H E: T9I; M: Q19E, A63T; N: P13L, P151S, R203K, G204R, S413R; ORF3a: T223I; ORF6: D61L; ORF7b: L11F; nsp1: S135R; nsp13: R392C; nsp14: I42V; nsp15: T112I; nsp3: T24I, G489S; nsp4: L264F, T327I, T492I; nsp5: P132H E: T9I; M: D3N, Q19E, A63T; N: P13L, R203K, G204R, S413R; ORF3a: T223I; nsp1: S135R; nsp13: R392C; nsp14: I42V; nsp15: T112I; nsp3: T24I, G489S; nsp4: L264F, T327I, T492I; nsp5: P132H E: T9I; M: Q19E, A63T; N: P13L, R203K, G204R, S413R; ORF3a: T223I; ORF6: D61L; nsp1: S135R; nsp13: R392C; nsp14: I42V; nsp15: T112I; nsp3: T24I, G489S; nsp4: L264F, T327I, L438F, T492I; nsp5: P132H replication proteins, but prior work has shown that the mutation rates of other viruses can be modulated by mu- tations distant from polymerase protein active sites (Vignuzzi et al. 2008; Pauly et al. 2017). Similar subtle mod- ifications could be induced by mutations to the nucleo- protein (which is part of the replication complex and protects viral RNA) as well as proteins that modulate the expression of host-cell innate-immune proteins. However, ultimate determination of the cause of the changes in the mutation spectrum will require experimen- tal work beyond the scope of our study and could also po- tentially be due to a wide range of factors including modifications in the location or speed of replication or transmission. There are several caveats to our study in addition to the inability to determine why the mutation spectrum differs among clades. First, our approach assumes that mutations at 4-fold synonymous sites are neutral. This assumption is probably not completely true, as various studies have shown that synonymous nucleotide composition is often under some selection in viruses for reasons including the physical structure of the genomic RNA, innate-immune evasion, and translation (van der Kuyl and Berkhout 2012; Kuo and Masters 2013; Huston et al. 2021; Kustin and Stern 2021). Such selection could also contribute to the disparity between the mutation spectrum and empir- ical equilibrium frequencies, since previous work has sug- gested that small-effect mutations that affect nucleotide composition may be under incomplete purifying selection on short branches such as those we use to estimate the mutation spectrum (Kustin and Stern 2021). Second, our analysis will be sensitive to sequencing errors among the millions of publicly available SARS-CoV-2 sequences that could be affected by factors such as changes in primer sets that occurred around the same time as the emergence of Omicron (Davis et al. 2021). The fact that our results are robust to excluding top mutated sites and partitioning the genome suggests that such technical factors probably do not seriously affect our results, but such caveats should be kept in mind. Finally, the emergence of Omicron oc- curred coincident with changes in the global level of im- munity to SARS-CoV-2, although this seems unlikely to have affected the mutation spectrum as the immune fac- tors that have been identified to act on viral nucleotide se- quences involve innate rather than adaptive immunity. Our analysis examines the relative rather than absolute rates of different types of nucleotide mutations across SARS-CoV-2 clades. We take this approach because rela- tive mutation rates can be internally calibrated, whereas precise estimation of absolute mutation rates from natural sequence data is harder. However, other recent work sug- gests that the overall absolute mutation rate is fairly simi- lar among human SARS-CoV-2 clades (Neher 2022). But if the 2-fold drop in the relative rate of G→T mutations in Omicron reflects a 2-fold drop in the absolute rate of that mutation type, that would only decrease the absolute rate across all mutations by approximately 7%, which would not be detectable at the resolution of current stud- ies (Neher 2022). Note that much more dramatically ele- vated mutation rates have been observed in rare clusters of human (Hisner 2022) or white-tail deer SARS-CoV-2 (Pickering et al. 2022), but these clusters have not spread widely. Overall, these observations are consistent with the idea that mutation rates might drift moderately during the natural evolution of successful SARS-CoV-2 variants (Sung, Ackerman, et al. 2012). However, so far, there is no evidence for widespread transmission of SARS-CoV-2 variants with extreme changes in mutation rates like those sometimes observed in the lab (Eckerle et al. 2007, 2010; Pauly et al. 2017), although there is evidence of limited hu- man transmission of viruses mutagenized by the drug monlupiravir (Sanderson et al. 2023). Interestingly, the actual nucleotide frequencies at 4-fold in both SARS-CoV-2 and related degenerate sites 7 Bloom et al. · https://doi.org/10.1093/molbev/msad085 sarbecoviruses differ from what would be predicted based on the mutation spectrum of any human SARS-CoV-2 clade. This difference is especially large for the mutation spectrum of early SARS-CoV-2 clades, with the mutation spectrum of Omicron clades being closer to that which shaped the long-term evolution of sarbecoviruses. We ac- knowledge that comparison of the mutation rates esti- mated in our study to nucleotide frequencies in natural sarbecoviruses could be somewhat confounded if there is weak selection on nucleotide identity even at 4-fold syn- onymous sites. But we were able to confirm that 4-fold de- generate nucleotide frequencies are close to their expected equilibrium for a wide range of other human viruses, sug- gesting SARS-CoV-2 may be unusual in having a mutation spectrum that is highly discordant with the actual frequen- cies of nucleotides at putatively neutral sites. One possible explanation is that the mutation spectrum of sarbecov- iruses could be relatively stable in the natural reservoir of bats, but has been altered in SARS-CoV-2 by some as- pect of replication in humans, and is now undergoing rela- tively rapid evolutionary change. The broader implications of shifts in the mutation spec- trum of SARS-CoV-2 for its evolution are unclear. Changes in the mutation spectrum alter the rates at which different potentially adaptive amino-acid mutations arise. But SARS-CoV-2 evolution in humans exhibits high levels of convergence (Martin et al. 2021; Cao et al. 2022), with puta- tively beneficial amino-acid mutations often emerging many independent times in different viral variants. This con- vergence suggests that the virus’s evolution is not generally limited by the underlying rate at which new mutations ap- pear. Therefore, the changes in the mutation spectrum we report are likely to at most modestly impact the overall pro- cess of adaptive evolution. However, our work does suggest that clade-specific estimates of the mutation rate are likely to improve the accuracy of efforts to estimate the fitness ef- fects of viral mutations from their number of observed oc- currences in natural sequences (Neher 2022) and could perhaps be useful for certain types of phylogenetic analyses. In addition, our work shows that the mutation process is clearly dynamic during SARS-CoV-2 evolution, so it will be interesting to see if larger changes in the mutation spectrum accrue as the virus continues to evolve. Materials and Methods Counting Mutations at 4-fold Degenerate Sites We determined the mutation spectrum by counting the number of unique occurrences of each nucleotide muta- tion on the branches of a global phylogenetic tree of all publicly available SARS-CoV-2 sequences. Note we are counting how many times each mutation is inferred to have independently occurred among available consensus SARS-CoV-2 sequences from individual human infections, not its final count in the alignment of such sequences (this distinction is important because a single occurrence of a mutation may be observed in multiple sequences due to shared ancestry). 8 MBE To the pre-built count mutations, we used tree clade-annotated UShER mutation-annotated (McBroome et al. 2021; Turakhia et al. 2021) from November 7, 2022 (http://hgdownload.soe.ucsc.edu/ goldenPath/wuhCor1/UShER_SARS-CoV-2/2022/11/07/ public-2022-11-07.all.masked.nextclade.pangolin.pb.gz). We used matUtils (McBroome et al. 2021; Turakhia et al. 2021) to subset the mutation-annotated tree on samples from each Nextstrain clade, and then extract the mutations on each branch of the subsetted mutation-annotated trees. We next tallied the counts of each mutation on all branches for that clade, excluding mutations on any branches with >4 total mutations, > 1 mutation that was a reversion to either the Wuhan-Hu-1 reference genome (Genbank NC_045512.2), or >1 mutation that was a reversion to the founder for that Nextstrain clade as defined by Neher (2022) (see https://github.com/neherlab/SC2_variant_ rates/blob/62c525dc4238385ec0755b40658f3007fdbfab1a/ data/clade_gts.json). The rationale for these exclusions is that branches with abnormally large numbers of mutations are often indicative of low-quality sequences with lots of er- rors, and branches with abnormally large numbers of rever- sions to the reference or clade founder can be indicative of sequences generated by problematic bioinformation pipe- lines that call low-coverage regions to the reference. For each clade, we then identified sites that are 4-fold degenerate in the clade founder (see https://github.com/ jbloomlab/SARS2-mut-spectrum/blob/main/results/clade _founder_nts/clade_founder_nts.csv). We also manually excluded sites that previous analyses (Turakhia et al. 2020) or our own analysis suggested might be prone to er- rors due to abnormally large numbers of mutations (the excluded sites are listed under sites_to_exclude in https:// github.com/jbloomlab/SARS2-mut-spectrum/blob/main/ config.yaml). Finally, we excluded any sites that differed between the clade founder and the Wuhan-Hu-1 reference (i.e., had fixed mutations in the clade founder relative to Wuhan-Hu-1). This exclusion was designed to avoid any spurious mutations caused by bioinformatics pipelines that call low-coverage sites to reference. The counts for all mutations in each clade are in the file at https:// github.com/jbloomlab/SARS2-mut-spectrum/blob/main/ results/mutation_counts/aggregated.csv, which contains columns indicating which sites are 4-fold degenerate or specified for exclusion. Table 1 presents the number of 4-fold degenerate sites for each clade and the total number of mutations at these sites. Note, we only retained clades with at least 5,000 mutation counts at non-excluded 4-fold degenerate sites. Finally, we tabulated the counts for each type of nucleo- tide mutation for each clade at the non-excluded 4-fold de- generate sites and determined the fraction of all mutations that were of that type (https://github.com/jbloomlab/ SARS2-mut-spectrum/blob/main/results/synonymous_ mut_rates/rates_by_clade.csv). the analyses supplementary figure S1, Supplementary Material online, we repeated the above process but subsetted only sequences from the USA or For in SARS-CoV-2 Mutational Spectrum · https://doi.org/10.1093/molbev/msad085 MBE England (as determined by whether the strain name con- tained that word), after excluding any site that was among the top 5 most mutated sites for any clade, or after parti- tioning the genome into halves. Principal Component Analysis The principal component analyses (PCAs) were performed on the length 12 probability vectors giving the fraction of all mutations at the 4-fold degenerate sites that were of each mutation type. The PCA was done using scikit-learn after first standardizing the vectors to have zero mean and unit variance. As described above, we repeated this analysis on subsets of the data to determine whether the results remained consistent when we restricted our ana- lyses to only sequences from the USA and England, ex- cluded any site that was among the top 5 most mutated sites for any clade, or partitioned the genome into halves. Calculation of Relative Mutation Rates The relative mutation rates plotted in figure 1C were calcu- lated simply by normalizing the fraction of all 4-fold degen- erate mutations that were of a given type by the fraction of all nucleotides at those sites in the clade founder that were of the parental nucleotide identity. For instance, the relative rate of A→T mutations was computed as the fraction of all mutations at non-excluded 4-fold degenerate sites that changed an A to a T, divided by the fraction of all 4-fold de- generate sites that had an A as their identity in the clade founder. Note that the frequencies of the different nucleo- tides at 4-fold degenerate sites are virtually identical among the clade founder sequences (supplementary fig. S3, Supplementary Material online). Phylogenetic Tree The phylogenetic tree in figure 1D was inferred on the clade founder sequences using iqtree (Minh et al. 2020) and then rendered using ete3 (Huerta-Cepas et al. 2016). The tips show the relative rates (as in fig. 1C) for each clade minus those rates for clade 20A, with the mutation types in the same order as in figure 1C. Mantel Test The Mantel test (Mantel 1967; Harmon and Glor 2010; Hardy and Pavoine 2012; Legendre and Legendre 2012) was used to estimate the significance of the correlation be- tween the Euclidean distance between clades’ mutation spectra and the square root of the phylogenetic distance between clade founder sequences (as estimated using iq- tree), also known as phylogenetic signal (fig. 2). The square root of the phylogenetic distance is used because it is ex- pected to scale linearly with Euclidean distance under a Brownian motion model (Hardy and Pavoine 2012). The Mantel test was implemented using the R package vegan (version 2.5–7), with method=“pearson” and 100,000 per- mutations (Oksanen et al. 2022). To determine whether the phylogenetic signal we observe is solely due to Omicron’s G > T fraction, the Mantel test was also carried out after excluding G > T mutations from the mutation spectrum and re-normalizing it. To additionally determine whether the phylogenetic signal is due only to differences between Omicron and non-Omicron clades, we also car- ried out tests for phylogenetic signal on Omicron clades and non-Omicron clades separately. Equilibrium Frequencies of SARS-CoV-2 Nucleotides The predicted equilibrium frequencies of nucleotides shown in figure 3A were calculated as the real component of the principal eigenvector of a rate matrix constructed from the relative rates of each mutation type for that clade. Predicted and Actual Nucleotide Frequencies at 4-fold Degenerate Sites for Other Viruses The predicted and observed nucleotide frequencies for other human viruses in figure 3C were calculated from phylogenetic analyses available on next strain: • the different influenza virus lineages at nextstrain.org/ groups/neherlab • RSV-A and RSV-B at nextstrain.org/rsv • Enterovirus A71 at nextstrain.org/groups/neherlab/ ev/a71 • Enterovirus D68 at nextstrain.org/enterovirus/d68 (Hodcroft et al. 2022) • Dengue virus 1–4 at nextstrain.org/dengue • West Nile Virus (WNV) nextstrain.org/WNV/NA (Hadfield et al. 2019) The empirical nucleotide frequencies were calculated by counting nucleotide states at 4-fold synonymous in the alignment references used in each of these builds (ex- plicitly linked in the script, see below). The mutation spec- trum was calculated by traversing the phylogenetic tree and counting mutations at positions that are 4-fold synonymous in the reference sequence. From the spec- trum and the empirical equilibrium frequencies, the pre- dicted equilibrium frequencies were calculated as for SARS-CoV-2. For influenza, the six largest segments such as PB2, PB1, PA, HA, NP, and NA were used for these ana- lyses, and for the other non-segmented viruses, the entire genome was used. These calculations are explicitly docu- the https://github.com/jbloomlab/SARS2- mented mut-spectrum/blob/main/scripts/compare_other_virus_ spectra.py script. A table listing the originating and sub- mitting labs of influenza sequences used in this analysis is provided at https://github.com/jbloomlab/SARS2- mut-spectrum/blob/main/GISAID_acknowledgments/flu_ acknowledgement.tsv. in The distances plotted in figure 3C represent the Manhattan distance of the empirical nucleotide frequen- cies at 4-fold degenerate sites to the equilibrium frequen- cies predicted from the mutation spectrum. 9 Bloom et al. · https://doi.org/10.1093/molbev/msad085 Supplementary Material Supplementary data are available at Molecular Biology and Evolution online. Acknowledgments We thank Ryan Hisner and Adam Lauring for their helpful comments. This research is based on sequence data from hundreds of laboratories around the world that have gen- erously shared their data. We gratefully acknowledge their contributions. This work was supported in part by the NIH/NIAID grant R01AI141707 J.D.B., NIH/NIA T32AG066574 to A.C.B., NIH/NIGMS grant R35GM 133428 to K.H., a Burroughs Wellcome Career Award at the Scientific Interface to K.H., a Searle scholarship to K.H., a Pew Scholarship to K.H., and a Sloan Fellowship to K.H., J.D.B. is an Investigator of the Howard Hughes Medical Institute. to Data availability The computer code used for the analysis is available at https://github.com/jbloomlab/SARS2-mut-spectrum as a fully reproducible Snakemake pipeline (Mölder et al. 2021). Interactive versions of many of the plots rendered with Altair (VanderPlas et al. 2018) are at https:// jbloomlab.github.io/SARS2-mut-spectrum/. Conflict of interest statement. J.D.B. is on the scientific ad- visory boards of Apriori Bio, Aerium Therapeutics, and Oncorus. J.D.B. receives royalty payments as an inventor on Fred Hutch’s licensed patents related to viral deep mu- tational scanning. References Aksamentov I, Roemer C, Hodcroft E, Neher R. 2021. Nextclade: clade assignment, mutation calling and quality control for viral gen- omes. J Open Source Softw. 6:3773. Bessa LM, Guseva S, Camacho-Zarco AR, Salvi N, Maurin D, Perez LM, Botova M, Malki A, Nanao M, Jensen MR, et al. 2022. The intrin- sically disordered SARS-CoV-2 nucleoprotein in dynamic com- plex with its viral partner nsp3a. Sci Adv. 8:eabm4034. Cao Y, Jian F, Wang J, Yu Y, Song W, Yisimayi A, Wang J, An R, Chen X, Zhang N, et al. 2022. Imprinted SARS-CoV-2 humoral immunity induces convergent Omicron RBD evolution. 2022.09.15.507787. Available from: https://www.biorxiv.org/content/10.1101/2022. 09.15.507787v4 Cao Z, Xia H, Rajsbaum R, Xia X, Wang H, Shi P-Y. 2021. Ubiquitination of SARS-CoV-2 ORF7a promotes antagonism of interferon response. Cell Mol Immunol. 18:746–748. Couce A, Guelfo JR, Blázquez J. 2013. Mutational spectrum drives the rise of mutator bacteria. PLoS Genet. 9:e1003167. Davis JJ, Long SW, Christensen PA, Olsen RJ, Olson R, Shukla M, Subedi S, Stevens R, Musser JM. 2021. Analysis of the ARTIC ver- sion 3 and version 4 SARS-CoV-2 primers and their impact on the detection of the G142D amino acid substitution in the spike protein. Microbiol Spectr. 9:e0180321. De Maio N, Walker CR, Turakhia Y, Lanfear R, Corbett-Detig R, Goldman N. 2021. Mutation rates and selection on synonymous mutations in SARS-CoV-2. Genome Biol Evol. 13:evab087. 10 MBE Denison MR, Graham RL, Donaldson EF, Eckerle LD, Baric RS. 2011. Coronaviruses: an RNA proofreading machine regulates replica- tion fidelity and diversity. RNA Biol. 8:270–279. Drake JW. 1993. Rates of spontaneous mutation among RNA viruses. Proc Natl Acad Sci U S A. 90:4171–4175. Eckerle LD, Becker MM, Halpin RA, Li K, Venter E, Lu X, Scherbakova S, Graham RL, Baric RS, Stockwell TB, et al. 2010. Infidelity of SARS-CoV Nsp14-exonuclease mutant virus replication is re- vealed by complete genome sequencing. PLoS Pathog. 6: e1000896. Eckerle LD, Lu X, Sperry SM, Choi L, Denison MR. 2007. High fidelity of murine hepatitis virus replication is decreased in nsp14 exor- ibonuclease mutants. J Virol. 81:12135–12144. Felsenstein J. 1985. Phylogenies and the comparative method. Am Nat. 125:1–15. Felsenstein J. 2003. Inferring phylogenies. Available from: https:// www.amazon.com/Inferring-Phylogenies-Joseph-Felsenstein/dp/ 0878931775 Fung S-Y, Siu K-L, Lin H, Chan C-P, Yeung ML, Jin D-Y. 2022. SARS-CoV-2 NSP13 helicase suppresses interferon signaling by perturbing JAK1 phosphorylation of STAT1. Cell Biosci. 12:36. Goldberg ME, Harris K. 2022. Mutational signatures of replication timing and epigenetic modification persist through the global di- vergence of mutation spectra across the great ape phylogeny. Genome Biol Evol. 14:evab104. Hadfield J, Brito AF, Swetnam DM, Vogels CBF, Tokarz RE, Andersen KG, Smith RC, Bedford T, Grubaugh ND. 2019. Twenty years of West Nile virus spread and evolution in the Americas visualized by Nextstrain. PLoS Pathog. 15:e1008042. Hardy OJ, Pavoine S. 2012. Assessing phylogenetic signal with meas- urement error: a comparison of Mantel tests, Blomberg et al.’s K, and phylogenetic distograms. Evolution. 66:2614–2621. Harmon LJ, Glor RE. 2010. Poor statistical performance of the Mantel in phylogenetic comparative analyses. Evolution. 64: test 2173–2178. Harris K. 2015. Evidence for recent, population-specific evolution of the human mutation rate. Proc Natl Acad Sci U S A. 112: 3439–3444. Harris K, Pritchard JK. 2017. Rapid evolution of the human mutation spectrum. Elife. 6:e24284. Hisner R. 2022. Sublineage of BM.2 with 8 additional spike mutations (9 seq, Australia) Issue #1286 · cov-lineages/pango-designation. GitHub [Internet]. Available from: https://github.com/cov- lineages/pango-designation/issues/1286 Hodcroft EB, Dyrdak R, Andrés C, Egli A, Reist J, García Martínez de Artola D, Alcoba-Flórez J, Niesters HGM, Antón A, Poelman R, et al. 2022. Evolution, geographic spreading, and demographic distribution of Enterovirus D68. PLoS Pathog. 18:e1010515. Huerta-Cepas J, Serra F, Bork P. 2016. ETE 3: reconstruction, analysis, and visualization of phylogenomic data. Mol Biol Evol. 33: 1635–1638. Huston NC, Wan H, Strine MS, de Cesaris Araujo Tavares R, Wilen CB, Pyle AM. 2021. Comprehensive in vivo secondary structure of the SARS-CoV-2 genome reveals novel regulatory motifs and mechanisms. Mol Cell. 81:584.e5–598.e5. Hwang DG, Green P. 2004. Bayesian Markov chain Monte Carlo se- quence analysis reveals varying neutral substitution patterns in mammalian evolution. Proc Natl Acad Sci U S A. 101: 13994–14001. Jiang P, Ollodart AR, Sudhesh V, Herr AJ, Dunham MJ, Harris K. 2021. A modified fluctuation assay reveals a natural mutator pheno- spectrum variation within that drives mutation type Saccharomyces cerevisiae. Elife 10:e68285. Kaplanis J, Ide B, Sanghvi R, Neville M, Danecek P, Coorens T, Prigmore E, Short P, Gallone G, McRae J, et al. 2022. Genetic and chemotherapeutic influences on germline hypermutation. Nature 605:503–508. Kimura M. 1968. Evolutionary rate at the molecular level. Nature 217:624–626. SARS-CoV-2 Mutational Spectrum · https://doi.org/10.1093/molbev/msad085 MBE Kirchdoerfer RN, Ward AB. 2019. Structure of the SARS-CoV nsp12 polymerase bound to nsp7 and nsp8 co-factors. Nat Commun. 10:2342. Kuo L, Masters PS. 2013. Functional analysis of the murine corona- virus genomic RNA packaging signal. J Virol. 87:5182–5192. Kustin T, Stern A. 2021. Biased mutation and selection in RNA viruses. Mol Biol Evol. 38:575–588. Legendre P, Legendre L. 2012. Numerical ecology, Volume 24—3rd from: https://www.elsevier.com/books/ Edition. Available numerical-ecology/legendre/978-0-444-53868-0 Lindsay SJ, Rahbari R, Kaplanis J, Keane T, Hurles ME. 2019. Similarities and differences in patterns of germline mutation be- tween mice and humans. Nat Commun. 10. Liu Y, Qin C, Rao Y, Ngo C, Feng JJ, Zhao J, Zhang S, Wang T-Y, Carriere J, Savas AC, et al. 2021. SARS-CoV-2 Nsp5 demonstrates two distinct mechanisms targeting RIG-I and MAVS to evade the innate immune response. mBio. 12:e0233521. Long H, Kucukyildirim S, Sung W, Williams E, Lee H, Ackerman M, Doak TG, Tang H, Lynch M. 2015. Background mutational fea- tures of the radiation-resistant bacterium Deinococcus radiodur- ans. Mol Biol Evol. 32:2383–2392. intervals Macià MC, Skov L, Peter BM, Schierup MH. 2021. Different historical inferred from generation Neanderthal fragment lengths and mutation signatures. Nat Commun. 12:5317. https://www.ncbi.nlm.nih.gov/pmc/articles/ PMC8423828/ in human populations Mantel N. 1967. Ranking procedures for arbitrarily restricted obser- vation. Biometrics. 23:65–78. Martin DP, Weaver S, Tegally H, San JE, Shank SD, Wilkinson E, Lucaci AG, Giandhari J, Naidoo S, Pillay Y, et al. 2021. The emergence and ongoing convergent evolution of the SARS-CoV-2 N501Y lineages. Cell. 184:5189.e7–5200.e7. Mathieson I, Reich D. 2017. Differences in the rare variant spectrum among human populations. PLoS Genet. 13:e1006581. McBroome J, Thornlow B, Hinrichs AS, Kramer A, De Maio N, Goldman N, Haussler D, Corbett-Detig R, Turakhia Y. 2021. A daily-updated database comprehensive SARS-CoV-2 mutation-annotated trees. Mol Biol Evol. 38: 5819–5824. tools and for Minh BQ, Schmidt HA, Chernomor O, Schrempf D, Woodhams MD, von Haeseler A, Lanfear R. 2020. IQ-TREE 2: new models and ef- ficient methods for phylogenetic inference in the genomic era. Mol Biol Evol. 37:1530–1534. Mölder F, Jablonski KP, Letcher B, Hall MB, Tomkins-Tinch CH, Sochat V, Forster J, Lee S, Twardziok SO, Kanitz A, et al. 2021. Sustainable data analysis with Snakemake. F1000Res. 10:33. Neher RA. 2022. Contributions of adaptation and purifying selection to SARS-CoV-2 evolution. 2022.08.22.504731. Available from: https://www.biorxiv.org/content/10.1101/2022.08.22.504731v1 Ogando NS, Ferron F, Decroly E, Canard B, Posthuma CC, Snijder EJ. 2019. The curious case of the nidovirus exoribonuclease: its role in RNA synthesis and replication fidelity. Front Microbiol.10. Ogando NS, Zevenhoven-Dobbe JC, van der Meer Y, Bredenbeek PJ, Posthuma CC, Snijder EJ. 2020. The enzymatic activity of the nsp14 exoribonuclease is critical for replication of MERS-CoV and SARS-CoV-2. J Virol. 94:e01246-20. Oksanen J, Simpson GL, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’Hara RB, Solymos P, Stevens MHH, Szoecs E, et al. 2022. ve- gan: community ecology package. Available from: https://CRAN. R-project.org/package=vegan Pauly MD, Lyons DM, Fitzsimmons WJ, Lauring AS. 2017. Epistatic in- teractions within the influenza a virus polymerase complex me- diate mutagen resistance and replication fidelity. mSphere. 2: e00323-17. Peck KM, Lauring AS. 2018. Complexities of viral mutation rates. J Virol. 92:e01031-17. Pickering B, Lung O, Maguire F, Kruczkiewicz P, Kotwa JD, Buchanan T, Gagnier M, Guthrie JL, Jardine CM, Marchand-Austin A, et al. 2022. Highly divergent white-tailed deer SARS-CoV-2 with po- tential 2022.02.22.481551. Available from: https://www.biorxiv.org/content/10.1101/2022. 02.22.481551v1 deer-to-human transmission. Ratcliff J, Simmonds P. 2021. Potential APOBEC-mediated RNA edit- ing of the genomes of SARS-CoV-2 and other coronaviruses and its impact on their longer term evolution. Virology. 556:62–72. Ringlander J, Fingal J, Kann H, Prakash K, Rydell G, Andersson M, Martner A, Lindh M, Horal P, Hellstrand K, et al. 2022. Impact of ADAR-induced editing of minor viral RNA populations on replication and transmission of SARS-CoV-2. Proc Natl Acad Sci U S A. 119:e2112663119. Robinson PS, Coorens THH, Palles C, Mitchell E, Abascal F, Olafsson S, Lee BCH, Lawson ARJ, Lee-Six H, Moore L, et al. 2021. Increased somatic mutation burdens in normal human cells due to defect- ive DNA polymerases. Nat Genet. 53:1434–1442. Roe MK, Junod NA, Young AR, Beachboard DC, Stobart CC. 2021. Targeting novel structural and functional features of coronavirus protease nsp5 (3CLpro, Mpro) in the age of COVID-19. J Gen Virol. 102. Ruis C, Peacock TP, Polo LM, Masone D, Alvarez MS, Hinrichs AS, Turakhia Y, Cheng Y, McBroome J, Corbett-Detig R, et al. 2022. Mutational spectra distinguish SARS-CoV-2 replication niches. 2022.09.27.509649. Available from: https://www.biorxiv.org/ content/10.1101/2022.09.27.509649v1 Ruis C, Weimann A, Tonkin-Hill G, Pandurangan AP, Matuszewska M, Murray GGR, Lévesque RC, Blundell TL, Floto RA, Parkhill J. 2022. Mutational spectra analysis reveals bacterial niche and transmission routes. 2022.07.13.499881. Available from: https:// www.biorxiv.org/content/10.1101/2022.07.13.499881v1 Sadler HA, Stenglein MD, Harris RS, Mansky LM. 2010. APOBEC3G contributes to HIV-1 variation through sublethal mutagenesis. J Virol. 84:7396–7404. Sanderson T, Hisner R, Donovan-Banfield I, Peacock T, Ruis C. 2023. Identification of a molnupiravir-associated mutational signature in SARS-CoV-2 sequencing databases. 2023.01.26.23284998. Available from: https://www.medrxiv.org/content/10.1101/ 2023.01.26.23284998v2 Sasani TA, Ashbrook DG, Beichman AC, Lu L, Palmer AA, Williams RW, Pritchard JK, Harris K. 2022. A natural mutator allele shapes mutation spectrum variation in mice. Nature 605:497–502. Speidel L, Cassidy L, Davies RW, Hellenthal G, Skoglund P, Myers SR. 2021. Inferring population histories for ancient genomes using genome-wide genealogies. Mol Biol Evol. 38:3497–3511. Sung W, Ackerman MS, Miller SF, Doak TG, Lynch M. 2012. Drift-barrier hypothesis and mutation-rate evolution. Proc Natl Acad Sci U S A. 109:18488–18492. Sung W, Tucker AE, Doak TG, Choi E, Thomas WK, Lynch M. 2012. Extraordinary genome stability in the ciliate Paramecium tetraur- elia. Proc Natl Acad Sci U S A. 109:19339–19344. Turakhia Y, Maio ND, Thornlow B, Gozashti L, Lanfear R, Walker CR, Hinrichs AS, Fernandes JD, Borges R, Slodkowicz G, et al. 2020. Stability of SARS-CoV-2 phylogenies. PLoS Genet. 16: e1009175. Turakhia Y, Thornlow B, Hinrichs AS, De Maio N, Gozashti L, Lanfear R, Haussler D, Corbett-Detig R. 2021. Ultrafast sample placement on existing tRees (UShER) enables real-time phylogenetics for the SARS-CoV-2 pandemic. Nat Genet. 53:809–816. V’kovski P, Kratzel A, Steiner S, Stalder H, Thiel V. 2021. Coronavirus biology and replication: implications for SARS-CoV-2. Nat Rev Microbiol. 19:155–170. van der Kuyl AC, Berkhout B. 2012. The biased nucleotide compos- ition of the HIV genome: a constant factor in a highly variable virus. Retrovirology 9:92. 11 Bloom et al. · https://doi.org/10.1093/molbev/msad085 MBE VanderPlas J, Granger BE, Heer J, Moritz D, Wongsuphasawat K, Satyanarayan A, Lees E, Timofeev I, Welsh B, Sievert S. 2018. Altair: interactive statistical visualizations for python. J Open Source Softw. 3:1057. Vignuzzi M, Wendt E, Andino R. 2008. Engineering attenuated virus vaccines by controlling replication fidelity. Nat Med. 14:154–161. Zhang J, Ejikemeuwa A, Gerzanich V, Nasr M, Tang Q, Simard JM, Zhao RY. 2022. Understanding the role of SARS-CoV-2 ORF3a in viral pathogenesis and COVID-19. Front Microbiol. 13. Available from: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8959714/ Ziebuhr J. 2005. The coronavirus replicase. Curr Top Microbiol Immunol. 287:57–94. 12
10.1093_nar_gkad329	Published online 5 May 2023 Nucleic Acids Research, 2023, Vol. 51, No. 12 6355–6369 https://doi.org/10.1093/nar/gkad329 Translational ﬁdelity screens in mammalian cells reveal eIF3 and eIF4G2 as regulators of start codon selectivity Richard She 1 ,* , Jingchuan Luo 1 and Jonathan S. Weissman 1 , 2 , 3 , 4 ,* 1 Whitehead Institute for Biomedical Research, Cambridge, MA, USA, 2 Department of Biology, Massachusetts Institute of Technology, Cambridge, MA, USA, 3 David H. Koch Institute for Integ r ative Cancer Research, Massachusetts Institute of Technology; Cambridge, MA 02142, USA and 4 Ho w ard Hughes Medical Institute, Massachusetts Institute of Technology, Cambridge, MA, USA Received July 01, 2022; Revised April 13, 2023; Editorial Decision April 13, 2023; Accepted May 03, 2023 ABSTRACT GRAPHICAL ABSTRACT The translation initiation machinery and the ribo- some orchestrate a highly dynamic scanning pro- cess to distinguish proper start codons from sur- r ounding nuc leotide sequences. Here, we perf ormed genome-wide CRISPRi screens in human K562 cells to systematically identify modulators of the fre- quency of translation initiation at near-cognate start codons. We observed that depletion of any eIF3 core subunit promoted near-cognate start codon usag e , though sensitivity thresholds of each sub- unit to sgRNA-mediated depletion varied consid- erab ly. Doub le sgRNA depletion experiments sug- gested that enhanced near-cognate usage in eIF3D depleted cells required canonical eIF4E cap-binding and was not driven by eIF2A or eIF2D-dependent leucine tRNA initiation. We further characterized the effects of eIF3D depletion and found that the N- terminus of eIF3D was strictly required for accu- rate start codon selection, whereas disruption of the cap-binding properties of eIF3D had no effect. Lastly, depletion of eIF3D activated TNF (cid:2) signal- ing via NF- (cid:3)B and the interferon gamma response. Similar transcriptional proﬁles were observed upon knockdown of eIF1A and eIF4G2, which also pro- moted near-cognate start codon usag e , sugg esting that enhanced near-cognate usage could potentially contrib ute to NF- (cid:3)B activ ation. Our study thus pr o- vides new avenues to study the mechanisms and consequences of alternative start codon usage. INTRODUCTION The fidelity of transcription and translation ensures the faithful transmission of genetically encoded DNA sequence into functional protein. The accuracy of each step r equir es biochemical discrimination between individual nucleotides or triplet codons. Whereas DNA replication introduces er- rors at rates as low as 1 in 10 8 bp after mismatch repair ( 1 ), translation is the most error prone and energetically costly step in pr otein pr oduction ( 2 , 3 ). Howe v er, because mistranslation e v ents are only transiently encoded in pro- tein products that can be readily degraded, quantifying the exact rates and modalities of aberrant translation in eukary- otic cells remains highly challenging. The production of protein from mRN A is typicall y ini- tia ted a t an AUG start codon, which ensures tha t transla- tion begins in the proper reading frame. Howe v er, a small handful of endogenous proteins are e xclusi v ely initia ted a t * To whom correspondence should be addressed. Tel: +1 6173241483; Email: [email protected] Correspondence may also be addressed to Richard She. Email: [email protected] C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 6356 Nucleic Acids Research, 2023, Vol. 51, No. 12 non-AUG start codons, including eIF4G2, which is initi- a ted a t an evolutionarily conserved GUG start codon ( 4 , 5 ). Experiments to systematically document the full breadth of translation products within a cell have been enabled by ri- bosome profiling ( 6 , 7 ). In particular, high-resolution ribo- some footprinting identified numerous noncanonical trans- lation products, including upstream open reading frames (uORFs), which were initia ted a t both AUG and non- AUG (near-cognate) start codons. While translation initi- a tion a t near-cogna te start codons is generally biochemi- cally disfavored, CUG and GUG start codons are the most commonly found and efficiently utilized near-cognate start codons ( 8 ). Initia tion a t uORFs can modula te the expres- sion of the downstream main ORF and empirically con- strains the length of 5´UTRs ( 9 ). In the human genome, 5´UTRs are only 210 bp on average, whereas 3´UTRs are 1028 bp on average ( 10 ). Furthermor e, the r egulatory ef- fect of uORF initiation has been well documented in the ATF4 and BiP transcripts ( 11 , 12 ), which act as key me- diators of the integrated str ess r esponse. Recent evidence has emerged that the peptide products of uORFs can pos- sess intrinsic functions ( 13 , 14 ). Many such micropeptides exerted effects on cellular growth and functional charac- terization re v ealed se v eral with distinct cellular localiza- tions and protein binding partners. Lastly, hundreds of uORF pr otein pr oducts wer e found to be pr esented by the MHC-I, suggesting that these peptides could com- prise a meaningful fraction of the cell’s antigen r epertoir e ( 13 , 15 , 16 ). The biochemical machinery that orchestrates start codon recognition has been characterized in great detail. In eukaryotes, variants of the consensus Kozak sequence (GCCACC AUG G) are commonly found directly upstream of both canonical AUG and near-cognate start codons ( 17 ). In addition, numerous highly conserved initiation fac- tor complexes play essential roles in directing the start of translation. In the classic model of translation, an initia- tor methionine tRN A (Met-tRN A i ) is bound by eIF2 and loaded onto the 40S small ribosomal subunit. This pro- cess is guided by physical interactions with eIF1, eIF1A, eIF3 and eIF5, and these factors collecti v ely comprise the 43S pre-initiation complex (PIC) ( 18 , 19 ). The 43S PIC is then recruited to the 5’-proximal region of the mRNA via interactions with se v eral cap-associated initiation factors and begins scanning in the 5’ to 3’ direction until it reaches a start codon with a suitable local nucleotide context ( 20–22 ). Base-pairing between Met-tRNA i and start codon facili- ta tes a conforma tional rearrangement of the scanning com- plex and joining of the 60S large ribosomal subunit to form the 80S ribosome, which then begins the process of transla- tion. These final steps in start codon selection are regulated by displacement of eIF1 from the ribosomal P-site, eIF5- induced hydrolysis of eIF2 •GTP, and eIF5B-mediated join- ing of the 60S ( 23–28 ). Previous genetic re v ersion screens in S. cerevisiae identified eIF1, eIF1A, eIF2 and eIF5 as genes involved in rescuing expression of an auxotrophic marker dri v en by a near-cognate start codon ( 26 , 29–32 ). Similarly, biochemical assays in rabbit reticulocyte lysate showed that increased stoichiometry of eIF1, eIF1A, eIF5 and eIF5B alter ed the fr equency of near-cognate start codon usage ( 8 , 33 ). Lastl y, targeted m utagenesis of regions of eIF3C and eIF4G known to interact with eIF1 and eIF5 led to mu- tants capable of promoting near-cognate start codon usage ( 34–37 ). We set out to explore whether near-cogna te initia tion in mammalian cells could be regulated or perturbed by in- trinsic cell biological processes. To do so, we performed genome-wide CRISPR interference (CRISPRi) screens in human K562 cells expressing a fluorescent reporter encoded with a CUG near-cognate start codon. The strongest genes identified by our unbiased screening approach recapitulated genetic re v ersion screens in yeast and included roles for eIF3 and eIF4G2 in regulating near-cognate start codon usage. Follo w-up experiments sho wed tha t alterna ti v e cap- binding by eIF3D / eIF4G2 was not r equir ed but that the N- terminus of eIF3D was essential for rescuing normal start codon stringency. Lastly, transcriptional profiling of cells depleted for eIF1A, eIF3D and eIF4G2 re v ealed acti vation of NF- (cid:2)B targets and the interferon gamma response. MATERIALS AND METHODS Manufacturer Product # Illumina 20020599 E2621X 10767–122 107689 356884 VB100 22400105 97068–085, lot 043K20 10378016 11965118 SLHP033RS P -600–100 352235 NC1105387 50–105-0634 NC0406407 M0202S 636766 Q32854 ZR2052 5067–4626 AM9822 AM9260G 11756050 F416L R0278-50M Sigma ThermoFisher A32965 Life Technologies NP0007 Life Technologies NW04122BOX ThermoFisher BioRad BioRad LI-COR LI-COR PI23227 1704270 1704150 927–90003 926–32213 Reagents Component Illumina TruSeq Stranded Total RNA kit NEBuilder HiFi DNA Assembly Kit Mirus Transfection Reagent Polybrene 1L spinner flask 1x ViralBoost RPMI media FBS NEB VWR Millipore Sigma DWK Life Sciences Alstem Thermo VWR Gibco Thermo Millipore Sigma Goldbio Corning Thermo Thermo Thermo NEB Takara Bio Thermo Zymo Research Penicillin-streptomycin-glutamine DMEM media 0.45 (cid:3)m filter Puromycin Mesh cap tube Macherey-Nagel NucleoSpin Blood XL kit NEBNext Ultra II Q5 Master Mix SPRIselect magnetic beads T4 DNA ligase Stellar Competent Cells Qubit High Sensitivity dsDNA assay Direct-zol RNA miniprep High Sensitivity DNA Bioanalyzer Kit Agilent Ambion 10% SDS Ambion 0.5 M EDTA ThermoFisher SuperScript IV VILO ThermoFisher DyNAmo ColorFlash SYBR Green kit RIPA buffer Pierce protease inhibitor tablets NuPAGE Sample Buffer (4x) Bolt Bis-Tris Plus 4–12% 12 Well Gel Bradford BCA kit Nitrocellulose membrane Bio-Rad Trans-Blot Turbo Intercept (PBS) Blocking Buffer IRDye 800CW Donkey anti-Rabbit IgG Biological r esour ces Cell lines: • K562 CRISPRi cells from ( 38 , 39 ). • HeLa CRISPRi cells from ( 40 ). • Jurkat CRISPRi cells (clone NH7) were obtained from the Berkeley Cell Culture Facility ( 41 ). • HEK293T cells from ( 38–40 ). Plasmids : • pMH0001 UCOE-SFFV-dCas9-BFP-KRAB (Addgene #85969); • pCRISPRia-v2 (Addgene #84832) • pLG GI3 hU6 sgRNA vector (Addgene #111594) Data availability / sequence data resources Sequencing data is available at GEO under accession num- ber GSE207330. Processed RNA-seq data are provided in the supplementary materials. Data av ailability / no v el pr ogr ams, softw ar e, algorithms For CRISPRi screen processing, sequencing data were aligned to the top 5 hCRISPRi-v2 library and quan- tified using the ScreenProcessing pipeline described in ( 42 ) with code available at ( https://github.com/mhorlbeck/ ScreenProcessing ). Websites / database r efer encing Molecular signatures database ( https://www.gsea-msigdb. org/gsea/msigdb/ ). Gene Ontology (GO) Resource ( http://geneontology. org/ ). Media formulations K562 and Jurkat cells were cultured in RPMI (Thermo Fisher Scientific, cat. 22400105) + 10% FBS (VWR, cat. 97068-085, lot 043K20) + 1 × penicillin–streptomycin– glutamine (Gibco, cat. 10378016). HEK293T and HeLa cells were cultured in DMEM (Thermo Fisher Scientific, cat. 11965118) + 10% FBS + 1 × penicillin–streptomycin– glutamine. Reporter cell line construction A series of fluorescence reporters for monitoring the pro- duction of GFP dri v en by a near-cognate start codons was constructed based on a lentiviral expression plasmid back- bone described in ( 38 ) (Addgene #85969). The vector con- tains an SFFV promoter, which dri v es strong expression and also contains no AUG sequences in the 5`UTR. Fur- thermore, a uni v ersal chromatin opening element (UCOE) upstream of the promoter pre v ents epigenetic silencing and a WPRE at the 3` end of the reporter sequence pro- motes transcript stability without causing pr ematur e termi- nation during lentivirus transcription. The original dCas9- BFP-KRAB cassette was replaced with reporter elements consisting of superfolder GFP coded with a near-cognate start codon and mCherry dri v en by the EMCV IRES. Reporter elements were inserted via restriction digest of the original vector with MluI and NotI and Gibson as- semb ly (Ne w England Biolabs, cat. E2621X). Reporter Nucleic Acids Research, 2023, Vol. 51, No. 12 6357 variants with AUG, CUG, GUG or other near-cognate start codons were produced by PCR amplification of the (cid:3) primers with GFP / IRES / mCherry cassette with unique 5 constant overhangs to allow for Gibson assembly (Supple- mentary Table S1). Lentivirus for each fluor escence r eporter was transduced into K562 cells for stable polyclonal expression at MOI < 1. Monoclonal cell lines were isolated for the CUG reporter v ariant b y sorting on a Sony MA900 but provided only a small advantage in terms of the covariance of GFP and mCherry expression – thus polyclonal sorted populations were constructed to pre v ent e xperimental artefacts that might arise from a single cell bottleneck. Lentivirus production for CRISPRi screening Lentivirus containing the hCRISPRi-v2 genome-wide CRISPRi sgRNA library was produced in thirteen 15 cm petri dishes of HEK293T cells. Prior to transfection, HEK293T cells were maintained at < 70% confluence dur- ing expansion. One day prior to transfection, cells were seeded at a density of 30 000 cells / cm 2 such that they reached a confluence of ∼60–70% on the day of trans- fection. For transfection, each 15 cm dish of HEK293Ts was transduced with 20 (cid:3)g sgRNA library, 6.75 (cid:3)g of standard packaging plasmids v3 (for expression of VSV- G, Gag / Pol, Rev and Tat), and 81 (cid:3)l Mirus transfection reagent (VWR, cat. 10767-122) in Opti-MEM. 24 h post- transfection, media was changed and supplemented with 1x ViralBoost (Alstem, ca t. VB100). Superna tant contain- ing lentivirus was harvested at 48 hours post-transfection. Cells were removed by centrifugation at 500g for 2 min and supernatant was filtered through a 0.45 (cid:3)m filter (Millipore ◦C. Lentivirus Sigma, cat. SLHP033RS) and frozen at -80 was then titered via a dilution series in K562s based on BFP expression at day 3 post-infection. CRISPRi screening K562 cells were expanded into six T-175 flasks with 70 ml media per flask. Cells were split each day during expansion to 400 000 cells / ml, such that after 24 h of growth they reached a density of roughly 800 000 cells / ml. Cell counts and viability were evaluated by flow cytometry on an BD Accuri C6 Plus flow cytometer and viability was maintained at > 90% prior to screening. To infect the K562 CRISPRi cells with the hCRISPRi-v2 sgRNA library (top 5 sgRNAs for each gene), cells were spinfected for 2 h to enhance in- fection efficiency. Briefly, 400M initial cells were pelleted by centrifuga tion a t 200g for 5 min. Cells wer e r esuspended in a total of 96 ml of fresh RPMI media + lentivirus + 8 (cid:3)g / ml polybr ene (Millipor e Sigma, cat. 107689). Lentivirus volume was chosen based on prior titration curves with a target infection rate of 30%. The cell suspension was then aliquoted into eight 6-well plates with 2 ml per well and cen- ◦C in a Sorvall Legend XTR trifuged at 1000g for 2 h at 37 centrifuge. After spinfection, cells were manually recovered by pipetting the contents each 6-well plate into three 50mL conical tubes and using 1mL of fresh media to wash each well and enhance the fraction of cells r ecover ed. In prac- tice, ∼80% of cells or 320M in total were recovered. Cells 6358 Nucleic Acids Research, 2023, Vol. 51, No. 12 were again pelleted at 200g for 5 min. Lastly, cells were re- suspended in fresh media at a density of 400 000 cells / ml and transferred to a 1 L spinner flask (DWK Life Sciences, cat. 356884) with a magnetic stir bar. Twenty four hours after infection, cells were split into two biological replica tes a t a density of 400 000 cells / ml. At 48 h post-infection, cells were evaluated for percent infected by measuring the fraction of BFP + cells by flow cytom- etry. In practice, a ∼20% infection rate was achie v ed cor- responding to a library coverage of ∼640 ×. At 48 h post- infection, cells were split to a density of 600 000 cells / ml in a total volume of 800 ml and 1 ug / ml of puromycin (Gold- bio, cat. P -600-100) was added to select for sgRNA express- ing cells. At day 3 post-infection, cells were again split to a density of 600 000 cells / ml and another 1 ug / ml of fresh puromycin was added. At day 3, roughly 30% of cells were BFP+ / sgRNA + by flow cytometry. At day 4 and day 5 post-infection, cells were split into fresh media at 400 000 cells / ml with a total volume of 600 ml to permit recovery fr om pur omycin selection. By day 5, ∼90% of cells were BFP+ / sgRNA+. Cells were sorted on day 6 post-infection on a BD FACS Aria II. For each round of sorting, 40M cells were gently pelleted at 200g for 4 min to help remove cell debris from puromycin selection. To enable a high sort rate, cells were resuspended in 1 ml of fresh media ( ∼40M cells / ml) and filtered through a mesh cap tube (Corning, cat. 352235) to disaggregate cell clumps. Cells were then placed on ice while awaiting flow sorting for a maximum of 2 h. Cells w ere flow ed at a flow rate of 8, which achie v ed up to 25 000 e v ents / second. Cells were sorted based on a cell via- bility gate (FSC versus SSC), a cell doublet gate (FSC-A versus FSC-H), an sgRNA expression gate (BFP+), and a GFP / mCherry ratiometric gate (top 15% and bottom 15% GFP / mCherry). In practice, ∼50–60% of cells passed the cell viability and singlet gates and ∼90% of cells were BFP+. Cells were sorted using custom sort setting with yield mask = 0 (ensuring deflection of only 1 drop and not adja- cent drops) and purity mask = 8 (rejecting drops if a non- targeted particle falls within 4 / 32 of the leading or trail- ing drop). With these settings, ∼60% sort efficiency was achie v ed at flow rates of 20000–25000 e v ents / second. In practice, ∼1000 cells / second were sorted into both the low and high GFP / mCherry populations. 12–20M cells were sorted over the course of 4 hours for each replicate. Sorted cells were then harvested by centrifugation and pellets were snap-frozen and stored at −80 ◦C. Genomic DNA was isolated from cell pellets with the Macherey-Nagel NucleoSpin Blood XL kit (Thermo Fisher Scientific, cat. NC1105387). Genomic DNA was isolated in a PCR free clean room and a small aliquot was quan- tified by NanoDrop, with ∼2.5 mg of total yield per pellet. 100 (cid:3)l PCR reactions with 10 (cid:3)g genomic DNA template each were set up in 96-well plates, using NEBNext Ultra II Q5 Master Mix (Thermo Fisher Scientific, cat. 50-105- 0634). Unique Illumina TruSeq indices were incorporated for each sample. All PCR reactions from each sample were then pooled and 100 (cid:3)l of the pool was size selected by double-sided SPRIselect magnetic bead clean-up (Thermo Fisher Scientific, cat. NC0406407). Libraries were quanti- fied and sequenced by Illumina HiSeq 4000 SE50. Data analysis for primary CRISPRi screen Sequencing data were aligned to the top 5 hCRISPRi- v2 library and quantified using the ScreenProcessing pipeline ( https://github.com/mhorlbeck/ScreenProcessing ) ( 42 ). sgRNA counts for the top 15% sample were divided by sgRNA counts for the bottom 15% sample and log 2 trans- formed into a log 2 enrichment score. An enrichment score for each gene was calculated by taking the mean of the top three sgRNAs targeting the gene. Significance at the gene le v el was calculated as Mann–Whitney P -value of the fiv e sgRNAs targeting the gene compared to the set of 1895 non- targeting sgRNAs. Enrichment scores from the two repli- cates were averaged, while P -values were combined using Fisher’s combined probability test. Enriched and depleted gene sets were defined based on an empirically deri v ed threshold based on the product of the enrichment score × −log 10 P -value. The threshold was cho- sen such that no negati v e control sgRNAs met the thresh- old. GO analysis was performed on enriched and depleted gene sets ( http://geneontology.org/ ) and genes belonging to major GO categories were visualized via volcano scatter plots (Figure 1 D–F). Secondary screening sgRNAs For individual evaluation and re-testing of sgRNA phe- notypes, 96 sgRNA expression plasmids were cloned in arr ay ed f ormat. sgRNA protospacers f or each target gene were inserted by annealing complementary synthetic oligonucleotide pairs (Integrated DN A Technolo gies) with BstXI and BlpI restriction site overhangs and ligation into BstXI / BlpI digested pCRISPRia-v2 (marked with a puromycin resistance cassette and BFP, Addgene #84832) ( 42 ). To promote annealing, the two oligos were added to 1x duplex buffer (Integrated DN A Technolo gies) at a final ◦C on a PCR block concentration of 2 (cid:3)M, heated to 95 for 5 min, and slowly cooled to room temperature. Oligos were then diluted 1:40 in 1 × duplex buffer and added to a ligation reaction with 1 (cid:3)l cut plasmid (25 ng / (cid:3)l), 1 (cid:3)l di- luted oligos, 0.5 (cid:3)l fresh 10 × T4 ligase buffer (with limited freeze thaw cycles), 0.5 T4 ligase (New England Biolabs, cat. M0202S), and 2 (cid:3)l wa ter. Liga tion was performed at RT for 1 h and 1 (cid:3)l of ligation product was transformed into 10 (cid:3)l of Stellar Competent Cells (Takara Bio, cat. 636766). Pro- tospacer sequences used for individual sgRNAs are listed in Supplementary Table S1. Indi vidual sgRNA e xpr ession plasmids wer e transfected into HEK293T cells for lentivirus production in arr ay ed for- mat in 6-well plates. Lentiviruses were then transduced into reporter cell lines by spinfection for K562 cells and Jurkat cells or re v erse transduction for HeLa cells, typically in 24- well format. Reporter cells were then ev aluated b y flow cy- tometry at day 5 post-transduction to allow for depletion of sgRNA target genes. MOI was typically < 1, resulting in ∼15–30% of cells infected. Flow cytometry analysis was then performed using uninfected cells as an internal control for each well and reporter phenotypes were quantified as the difference in GFP ( (cid:2) GFP) and mCherry ( (cid:2) mCherry) be- tween sgRNA-infected and uninfected cells within the same well. Nucleic Acids Research, 2023, Vol. 51, No. 12 6359 Figure 1. Genome-wide CRISPRi screens using a CUG transla tion reporter. ( A ) Schema tic for lentiviral dual-fluorescence reporter to measure near- cognate start codon translation from a CUG start codon. ( B ) Comparati v e e xpression le v els of reporter variants dri v en by CUG v ersus AUG start codons. ( C ) Workflow for FACS-based genome-wide CRISPRi screening in K562 cells. ( D ) Volcano plot of sgRNA enrichment scores for known initiation factors. ( E ) Volcano plot of sgRNA enrichment scores for proteosome or ribosome components. ( F ) Volcano plot of sgRNA enrichment scores for RN A pol y- merase, mediator, or spliceosome components. ( G ) Simplified cartoon of major steps in translation initiation, adapted from ( 45 ). Numbers are abbreviations for initiation factors, e.g. eIF3 is abbreviated as 3. Double knockdown sgRNAs Western blots Dual sgRNA expression vectors were cloned in accordance with the method previously described in ( 43 ). Each of 24 sgRNA protospacers (Supplementary Table S1) were cloned into a variant of the single sgRNA plasmid with a modified human U6 promoter replacing the original mouse U6 promoter (Addgene #111594). After verification by Sanger sequencing, a fragment containing the human U6 and sgRNA components was PCR amplified and Gibson cloned into the XhoI restriction site of an original single sgRNA expression plasmid with sgRNAs targeting eIF3D, eIF4G2, eIF5 or a non-targeting sgRNA. As with single sgRNA e xpression v ectors, dual e xpression v ectors were transfected into HEK293T cells for lentivirus production in arr ay ed format and transduced into K562 reporter cells. Reporter phenotypes were analyzed by internally controlled comparisons to uninfected cells within the same well. K562 cells were infected with lentiviral constructs con- taining sgRNAs targeting eIF1A, eIF2 (cid:4), eIF3A, eIF3D, eIF3G, eIF3H, eIF3M, eIF4G2 and eIF5. Forty-eight hours post-infection, 2 (cid:3)g / ml puromycin was added to RPMI media to select for cells expressing sgRNA. Cells were spun down at 72 h post-infection at 200g for 2 min to help remove debris and dying cells. The pellet was then resuspended in fresh media with 2 (cid:3)g / ml puromycin for an additional day. Cells wer e r ecover ed in normal growth media from 4 days post-infection to 5 days post-infection. sgRNA containing cells were sorted to purity on a Sony MA900 cell sorter at 5 days post-infection and pellets were immediatel y l ysed in 30 (cid:3)l ice-cold RIPA buffer + protease inhibitor (ThermoFisher, cat. A32965) per 1 million cells. ◦C, cells were After 30 min of incubation in lysis buffer at 4 ◦C. Supernatant was centrifuged at 16 000g for 5 min at 4 6360 Nucleic Acids Research, 2023, Vol. 51, No. 12 collected and snap frozen in liquid nitrogen and stored at −80 ◦C. Protein concentrations in each lysate were quantified using a Bradford BCA kit (ThermoFisher, cat. PI23227) Lysate was normalized to 1 (cid:3)g / (cid:3)l in RIPA buffer. 30 (cid:3)l of lysate was added to 10 (cid:3)l of NuPage Sample Buffer ◦C on a PCR thermocycler, and loaded (4 ×), heated to 70 onto a Bolt 4–12% polyacrylamide gel (ThermoFisher, NW04122BOX). Four replicate gels were run for 45 min at 165V in MOPS buffer to allow for blotting of multiple eIF3 subunits of similar molecular weight. Protein was then transferred onto a nitrocellulose membrane (BioRad, cat. 1704270) with a Bio-Rad Trans-Blot Turbo (BioRad, cat. 1704150). The membrane was blocked with Intercept (PBS) Blocking Buffer (LI-COR, cat. 927–90003) for 1 h at RT. ◦C with primary Membrane was incubated overnight at 4 antibod y, with ca talog numbers and dilutions for each an- tibody listed in Supplementary Table S2. Membrane was washed 3 × with TBST and incubated with secondary anti- body (Licor IRDye 800CW Donkey anti-Rabbit IgG, cat. 926-32213) at 1:15000 dilution. Membrane was washed 3 × with TBST and imaged on a LI-COR Odyssey CLX. Quantitative RT-PCR K562 cells with sgRNAs targeting eIF1A, eIF2 (cid:4) and eIF5 were grown in parallel with cells grown for western blotting (see above). Cells were puromycin selected and 200 000 cells were sorted at 5 days post-infection. Cell pellets were im- ◦C . mediatel y l ysed in RN Ase-fr ee Trizol and stor ed a t −80 RNA was extracted with a Dir ect-zol RNA minipr ep kit (Zymo Research, cat. R2051). RNA was re v erse transcribed with SuperScript IV VILO (ThermoFisher, cat. 11756050), and cDNA was amplified with the DyNAmo ColorFlash SYBR Green kit (ThermoFisher, cat. F416L). Primers for GAPDH were used as loading controls and no-RT controls were performed to control for genomic DNA contamina- tion. Amplifications were performed in duplicate and quan- tified on a QuantStudio Flex 7 Real-Time PCR system in 96-well plates. eIF3D structure function variants A donor plasmid containing a cDNA-based eIF3D ORF was ordered from the Harvard CCSB Human ORFeome collection (no longer operational) (BC080515, Internal ID 55224). The eIF3D ORF was cloned into the single sgRNA vector containing an sgRNA targeted against the endoge- nous copy of eIF3D. As the sgRNA targets the endogenous promoter, it does not target the SFFV promoter that dri v es the expression of exogenous eIF3D. The original eIF3D sgRNA plasmid was digested with NheI and EcoRI. The eIF3D ORF was then inserted downstream of Puro-T2A- BFP via three-piece Gibson Assembly, with one PCR frag- ment reconstituting the Puro-T2A-BFP cassette and one PCR fragment consisting of the eIF3D ORF with an up- stream P2A to maintain expression on the same transcript. eIF3D variants were constructed via primers that resulted in N-terminal truncation, C-terminal truncation, modifica- tion of phosphorylation sites proximal to the C-terminus, or triple mutants in helix (cid:4)5 or (cid:4)11. Lentivirus for each variant was transduced into K562 reporter cells. Cells ex- pressing each construct were gated by BFP expression and compared to cells without BFP expression. Reporter phe- notypes were quantified by measuring changes in GFP and mCherry expression. Over expr ession plasmids Donor plasmids containing cDNA-based ORFs for cJun, ABCD1, eIF2A, eIF2D, MCTS1 and DENR were ob- tained from the ORFeome Collaboration Clones (Hori- zon Discovery). Each ORF was cloned into the eIF3D sgRNA vector and an otherwise identical vector contain- ing a non-targeting sgRNA. As with the eIF3D ORF, the ORFs were inserted downstream of Puro-T2A-BFP-P2A. Lentivirus for each over expr ession construct was trans- duced into K562 reporter cells. Cells expressing each con- struct were gated by BFP expression and compared to cells without BFP expression. Reporter phenotypes were quanti- fied by measuring changes in GFP and mCherry expression. Flow cytometry Data were collected on an Attune NxT flow cytometer (Thermo Fisher Scientific) and analyzed with custom Mat- la b scripts. Via b le cells were gated based on forwar d and side scatter with a manually drawn gates. Doublets were fil- tered based on FSC-A and FSC-H with manually drawn gates. Both viable cell and doublet filters were applied to all cells within a well. Next, sgRNA containing cells were dis- tinguished from uninfected cells by BFP expression with a linear gate. Mean GFP and mCherry expression was then calculated for sgRNA expressing cells and uninfected cells. The difference in GFP expression ( (cid:2) GFP) between the two populations was then calculated, r epr esenting the change in near-cognate start codon transla tion. The dif ference in mCherry expression ( (cid:2) mCherry) was also calculated, rep- resenting a change in IRES-dri v en translation. To quantify a normalized reporter score, differences in mCherry were subtracted from differences in GFP and log 2 transformed. Thus, log 2 ( (cid:2) GFP − (cid:2) mCherry), was used as the primary metric for the effect of an sgRNA, dual sgRNA, or ov ere x- pression construct on the reporter. Bulk RNA-seq K562, Jurkat, and HeLa cells were infected with individual sgRNAs targeting eIF3D. K562 cells were also infected with individual sgRNAs targeting eIF4G2, eIF1A or ZNF324. Cells were then expanded for 5 days. At day 5, ∼1M sgRNA expressing cells were sorted on a Sony MA900 cell sorter based on BFP + expression. Cells were then pelleted, snap- ◦C. High quality RNA was ex- frozen and stored at −80 tracted by adding RNAse-free Trizol (Thermo Fisher Sci- entific, cat. 15596026) to each pellet and processing with the Zymo Research Direct-zol RNA miniprep kit (Zymo Research, cat. R2050). RNA-seq was performed using the Illumina TruSeq Stranded Total RNA kit (Illumina, cat. 20020599) according to the manufacturer’s instructions, with the exception of the final PCR step for which only 10 cycles were used to pre v ent ov eramplification. The final pooled library was sequenced with 50 bp single end reads on a HiSeq 2500. aligned RNA-seq sequencing reads were to hg19 / GRCh37 with STAR aligner and quantified with fea- tur eCounts. Fold-changes wer e calculated by comparison of counts between wild-type cells and cells expressing sgR- NAs for gi v en target genes. Transcriptional responses were then compared to annotated gene sets from the Molecular Signa tures Da tabase (MSigDB). Perturb-seq analysis eIF3D did not cluster with any other genes in the original Perturb-seq analysis ( 44 ), as clustering was performed in a robust fashion that only identified the strongest clusters. In this original analysis, not all genes were members of tran- scriptional clusters. Howe v er, the sgRNA targeting eIF3D produced a strong transcriptional phenotype as measured by the number of differentially regulated genes. We thus per- formed a more permissi v e clustering on a high-dimensional (20 dimensions) embedding of the data. The embedding in this case served as a light imputation step that potentially caused perturbations to be drawn closer to their presump- ti v e relati v es. This analysis produced a clustering visual- ization akin to the one presented in ( 44 ). With this anal- ysis, eIF3D knockdown cells clustered with cells depleted for eIF3E / F / H / L / M, eIF4A1, eIF4G2, eIF1A, DDX3X, CSDE1, STRAP and ZNF324. The clustering method did not inherently identify genes that were most responsible for distinguishing the eIF3D cluster from all other clusters. Thus, we rationally picked comparison gene sets likely to influence translation or activate NF- (cid:2)B. These comparison sets consisted of all other known initiation factors, ribosomal pr oteins fr om the small or large subunit, and genes known to activate NF- (cid:2)B upon knockdown. To visualize the differential tran- scriptional response of each gene set, we filtered the re- sponse by genes with maximum normalized z -score > 1 (ab- solute value) and median normalized z -score > 0.1 (abso- lute value) across all genes. sgRNA target genes and tran- scriptional r esponses wer e then cluster ed by k-means clus- tering with fiv e clusters randomly seeded for the sgRNA target genes and three clusters seeded for the transcrip- tional response. Genes in the transcriptional response clus- ters were ev aluated b y GO enrichment and the top GO terms were empirically summarized into cluster labels. The transcriptional response clusters tended to contain se v eral broad categories of genes, with the exception of ribosomal proteins. RESULTS Genome-wide CRISPRi screens identify candidate regulators of alternative start codon usage To enable systematic genetic screening approaches, we de- signed a series of translational fidelity reporters. Our most basic reporter design utilized a bicistronic fluorescence pro- tein construct: superfolder GFP encoded with a CUG near- cognate start codon and mCherry dri v en by the EMCV in- Nucleic Acids Research, 2023, Vol. 51, No. 12 6361 ternal ribosome entry site (IRES) (Figure 1 A). We chose CUG because it is the most frequently utilized near-cognate start codon, while the IRES / mCherry element acted as an internal expression control. To minimize translation initia- tion at other sites within the 5`UTR, we removed all AUG sequences upstream of the CUG start codon. We observed that compared to a standard AUG start codon, a CUG start codon with a consensus Kozak sequence produced ∼920x lower le v els of GFP e xpr ession (Figur e 1 B, normal- ized to mCherry). Howe v er, due to the presence of a strong SFFV promoter, the CUG start codon reporter produced ∼13 × higher le v els of GFP fluor escence compar ed to the background autofluorescence of wild-type K562s. To quan- tify changes in the rate of GFP translation, we normalized total GFP expression by mCherry expression. This ratio- metric approach allowed us to account for changes in tran- script le v els, which would e xert equi valent effects on both GFP and mCherry. In addition, we observed a high degree of correlation (Pearson’s correlation r = 0.96) between GFP and mCherry expression across a polyclonal population of cells, which helped to control for variation in absolute GFP e xpression le v els due to factors such as cell size and lentivi- r al integr a tion site. The coef ficient of varia tion (CV) for the GFP / mCherry ratio was 22%, whereas GFP alone had a CV of 76%. Ne xt, we le v eraged the low degree of variation and noise in our fluorescence reporter for large-scale FACS-based screening. To identify genetic perturbations that would al- ter the frequency of CUG start codon initiation, we in- fected our reporter cell line with a genome-wide lentiviral hCRISPRi-v2 sgRNA library ( 42 ). After 6 days of selec- tion and outgrowth to ensure adequate recovery and knock- down, we sorted cells with high or low GFP / mCherry ra- tios. Gates for the top 15% and bottom 15% of cells dif- fered by only ∼46%, in theory allowing our screen to dis- tinguish sgRNAs with ∼2-fold effects on CUG initiation rates (Figure 1 C). Lastly, we used next-generation sequenc- ing to quantify the sgRNAs present in the high vs. low GFP / mCherry subpopulations. Our screen re v ealed strong enrichment of sgRNAs tar- geting known initiation factor complexes. sgRNAs tar- geting eIF5 / 5B, eIF4G1 or eIF4E were enriched in the low GFP / mCherry population while sgRNAs targeting eIF1 / 1A, eIF2, eIF3 or eIF4G2 were enriched in the high GFP / mCherry population (Figure 1 D). In addition, com- ponents of the spliceosome and ribosome were enriched in the high GFP / mCherry population whereas members of the proteasome, RN A pol ymerase, and mediator complexes wer e depleted (Figur e 1 E, F). In total, we observed 476 can- didate genes whose knockdown increased the ratio of GFP to mCherry and 154 candidate genes that decreased the ra- tio (Supplementary Table S3). The majority of factors iden- tified in our screen were highly essential, demonstrating the utility of our CRISPRi screening approach in evaluating factors involved in essential cell biological processes (Sup- plementary Figure S1A). Howe v er, the majority of essen- tial genes (1923 / 2324) had no effect on our fluorescence re- porter. In addition, the ability of essential genes to either in- cr ease or decr ease GFP / mCherry suggested that simple im- pairment of cellular growth was not responsible for changes in reporter expression. 6362 Nucleic Acids Research, 2023, Vol. 51, No. 12 Secondary screening decouples alternative start codon trans- lation from IRES-dependent translation Ne xt, we indi vidually retested 96 candidate sgRNAs target- ing 96 genes to distinguish increases in GFP translation at the CUG start codon from reductions in IRES-mediated mCherry translation, as both effects would similarly alter the GFP to mCherry ratio (Supplementary Table S4). To do so, we performed sgRNA-mediated knockdowns in inter- nally controlled co-cultures, with uninfected wild-type cells (BFP −) and sgRNA-expressing cells (BFP+) mixed within the same well. As each well contained only a single sgRNA, we decoupled GFP and mCherry fluorescence into average GFP and average mCherry expression levels across a pop- ulation of cells (Figure 2 A). For each sgRNA, we quanti- fied whether the change in GFP / mCherry ratio could be attributed to changes in GFP le v els or in mCherry le v els. sgRNAs tar geting kno wn initiation factors elicited lar ger changes in GFP le v els (Figur e 2 B) compar ed to mCherry le v els (Figure 2 C). By contrast, sgRNAs targeting riboso- mal small subunit proteins and spliceosome factors primar- ily altered the mCherry le v els (Supplementary Figure S1B). Overall, 43 / 96 sgRNAs tested exhibited stronger effects on IRES-media ted transla tion compared to CUG trans- lation – thus a substantial number of hits from the pri- mary screen were likely due to trans-factors required by the EMCV IRES. In theory, some of these false positi v e can- didates could have been mitigated in primary screening via usage the type 4 CRPV IRES, which directly recruits the 40S and does not depend on initiation factors. Howe v er, a false negati v e result would require a factor to change GFP and mCherry le v els by the same amount, and the sensitivity of our assay allowed us to recover candidate genes with quanti- tati v ely different effects on CUG and IRES-mediated trans- lation. Furthermor e, r esults from our secondary screening allowed us to exclude factors that primarily affect IRES- media ted transla tion from further follow-up experiments. We then explored whether each of our candidate genes specifically modula ted initia tion a t CUG start codons or at near-cognate start codons more broadly. To do so, we re- placed the CUG start codon in our bicistronic fluorescence reporter with a GUG start codon, the second most fre- quently used near-cognate start codon. Overall GFP expres- sion from the GUG reporter was comparable to that from the CUG reporter (Supplementary Figure S1C). Across all 96 candidate sgRNAs, changes in alternati v e start codon us- age were virtually identical for the two reporters (Pearson’s correlation r = 0.98) (Figure 2 D). As an additional control, we tested the effect of the knockdowns in reporter cells expressing GFP from a stan- dard AUG start codon. GFP expr ession was r estor ed to high le v els in the A UG reporter , while mCherry le v els re- mained unchanged. We observed that both CUG and AUG r eporters exhibited decr eased GFP le v els upon knockdown of eIF4E or eIF4G1, as both are key factors in the ini- tial recruitment of the 40S small subunit to mRNA, but not in start codon discrimination (Figure 2 E). In contrast, sgRNAs targeting eIF3 subunits, eIF4G2 or eIF1 / 1A in- cr eased expr ession of the CUG reporter but not of the AUG reporter. Conversel y, sgRN As targeting eIF5 / 5B reduced translation from the CUG near-cognate reporter but not the AUG r eporter. These r esults wer e consistent with pr e- vious screens in yeast that identified eIF1 / 1A and eIF5 / 5B as modifiers of start codon stringency ( 26 , 29 , 32 ). In addi- tion, subsequent studies have shown that eIF1A, eIF5 and eIF5B are key factors in catalyzing the formation of the 80S ribosome, the final step in start codon recognition ( 8 , 45 ). Howe v er, the roles of eIF3 and eIF4G2 in alternati v e start codon usage have not been previously described. To further exclude the possibility that the effects of our knockdowns were mediated by interactions with the EMCV IRES, we constructed reporter cell lines with no IRES ele- ments. To maintain the normalization properties of a dual- fluor escence r eporter, we integrated a CUG-encoded GFP and an AUG-encoded mCherry via separ ate lentivir al vec- tors and isolated a monoclonal reporter cell line. We con- firmed that this no-IRES reporter exhibited highly concor- dant phenotypes across most candidate knockdowns, with the exception of IRES-dependent candidates that were also identified during the initial round of validation (Pearson’s correlation r = 0.76) (Supplementary Figure S2A). In con- cordance with previous control experiments, we observed that depletion of eIF3 components, eIF4G2, and eIF1A re- sulted in increased translation from alternati v e start codons whereas eIF5 knockdown repressed alternati v e start codon usage. Lastly, to control for specific RNA binding motifs in the 5´UTR, we replaced the SFFV promoter with an EF-1 (cid:4) promoter and verified that eIF3D knockdown promoted in- creased alternati v e start codon usage (Supplementary Fig- ure S2B). eIF3 plays a major role in start codon discrimination We next investigated the effects of eIF3 depletion on alter- nati v e start codon usage, as four out of the six strongest sgRNAs from our validation screening targeted subunits of eIF3 (eIF3D, eIF3G, eIF3H and eIF3M). We first asked whether depletion of these four subunits uniquely promoted CUG start codon usage compared to other eIF3 subunits. We hypothesized that not all sgRNAs targeting eIF3 would be fully acti v e, as some fraction of sgRNAs would not achie v e sufficient depletion of their target genes. Indeed, se v eral lines of evidence revealed that sgRNA-mediated depletion was highly variable and incomplete, which was likely due to the strong essentiality of eIF3 components. First, we observed depletion of any eIF3 subunit other than eIF3J resulted in enhanced translation of the CUG near- cognate start codon reporter (Figure 3 A). Secondly, we ob- served that the growth defects induced by sgRNAs target- ing eIF3 subunits were strongly correlated to the magnitude of reporter phenotypes. The sgRNAs with the strongest growth defects were comparable to sgRNAs targeting ribo- somal subunits (Supplementary Figure S3A). Lastly, we di- r ectly measur ed the extent of sgRNA-mediated depletion by Western blot for individual sgRNAs targeting eIF3A, eIF3D, eIF3G, eIF3H and eIF3M (Supplementary Figure S3B–E, Supplementary Table S2). Knockdown efficiency was poor for eIF3A, with only 25% depletion. Depletion was intermediate for eIF3D, eIF3H, and eIF3M, ranging from 48% to 80%. Only eIF3G exhibited greater than 90% knockdown efficiency. These results suggest that individual subunits of the eIF3 complex may have distinct depletion Nucleic Acids Research, 2023, Vol. 51, No. 12 6363 Figure 2. Validation of individual sgRNA phenotypes across reporter variants . ( A ) GFP vs . mCherry expression for co-cultures between wild-type cells and sgRNA containing cells. ( B ) Change in CUG-dri v en GFP e xpression upon sgRNA knockdown for 96 individual sgRNAs. ( C ) Change in IRES-dri v en mCherry expression upon sgRNA knockdown for 96 individual sgRNAs. ( D ) Comparison of sgRNA depletion phenotypes in cells expressing a CUG translation reporter versus GUG translation reporter. ( E ) Comparison of sgRNA phenotypes in cells expressing a CUG translation reporter versus AUG translation reporter. thresholds, at which point the overall activity of the com- plex is impaired and cellular translation is disrupted. We extended our comparison to other key initiation fac- tors. Knockdowns of eIF2 produced equally se v ere growth defects as eIF3 depletion but comparati v ely milder in- creases in CUG start codon usage (Figure 3 B). Measure- ments of sgRNA-mediated transcript depletion by qRT- PCR showed ∼73% knockdown efficiency (Supplementary Figure 3F). Meanwhile, eIF1 or eIF1A knockdown also elicited an increase in CUG translation, with depletion of eIF1 causing no impairment to growth (Supplementary Figure S4A). qRT-PCR on an eIF1A sgRNA re v ealed only 37% knockdown efficiency. Double knockdown of eIF1 and eIF1B, which encode the same polypeptide from sep- ara te genomic loca tions, r esulted in a pur ely additi v e effect that remained weaker than knockdown of eIF3 subunits, 6364 Nucleic Acids Research, 2023, Vol. 51, No. 12 Figure 3. eIF3D exerts a dominant effect on start codon selection. ( A ) Scatter plot of CUG reporter score (log 2 ( (cid:2) GFP – (cid:2) mCherry)) versus growth defect (lo g 2 sgRN A depletion per doubling) for sgRN As targeting subunits of eIF3. ( B ) Sca tter plot of CUG transla tion phenotype for sgRNAs targeting subunits of eIF3 and sgRNAs targeting subunits of eIF2. ( C ) Double sgRNA knockdowns with 24 candidate sgRNAs and a non-targeting sgRNA, ( D ) an sgRNA targeting eIF3D, ( E ) an sgRNA targeting eIF5, ( F ) or an sgRNA targeting eIF4G2. indicating a lack of genetic buffering between these fac- tors (Supplementary Figure S4B, C). These results demon- stra ted tha t our CRISPRi screening approach was capable of uncovering cellular phenotypes in highly essential genes whose complete knockouts were unviable. In addition, the variab le acti vity of the fiv e sgRNAs targeting each eIF3 and eIF2 subunit acted as a dose titration e xperiment, re v ealing a stronger relationship between the degree of growth im- pairment and the frequency of near-cognate start codon us- age for eIF3. To explore potential interactions effects between key ini- tiation factors, we performed a set of targeted double- knockdown genetic interaction experiments. We selected a set of 24 potential interaction partners that included ma- jor initiation factor candidates, ribosomal subunits, and ad- ditional genes from our validation screens (Supplementary Tables S5 and S6). We cloned each sgRNA into a dual- sgRNA v ector and v erified that each interacting sgRNA maintained its original activity when paired with a non- targeting sgRNA (Figure 3 C). Next, we introduced sgR- NAs targeting eIF3D, eIF4G2, or eIF5 in combination with each potential interaction partner. We chose eIF3D as a r epr esentati v e of eIF3 due to its strong effect size and be- cause of previous literature indicating that eIF3D depletion is unique among eIF3 subunits in preserving the structural integrity of the remaining eIF3 complex ( 46 , 47 ). In addi- tion, eIF3D has recently been found to play a separate role in alternati v e cap-binding via a physical interaction with eIF4G2 ( 48–50 ). Under an additi v e model with no inter- actions, we would expect that the combined effect of each double knockdown to equal the sum of the two individ- ual knockdowns. Instead, we observed that eIF3D knock- down exerted a maximal effect, with no other knockdowns significantly increasing the extent of CUG translation be- yond eIF3D knockdown alone (Figure 3 D). Howe v er, we observed that knockdowns of eIF2, eIF4E, and eIF5 re- duced the effect of eIF3D knockdown. These data suggest tha t increased transla tion of the CUG reporter was largely mediated by standard eIF4E cap-dependent processes and eIF2-linked initiator methionine tRNA. By contrast, depletion of alternati v e eIF2 initiation factors eIF2A and eIF2D with two independent sgRNAs had no effect on the eIF3D depleted cells (Supplemental Figure S4D). These re- sults indica te tha t initia tion a t the CUG start codon was not primarily dri v en by its cognate leucine tRNA ( 51 , 52 ). In addition, ov ere xpression of alternati v e initiation factors and ribosome recycling factors ABCE1, eIF2A, eIF2D, MCTS1, or DENR had no effect on either wild-type cells or eIF3D depleted cells (Supplementary Figure S4E). In contrast to eIF3D, double knockdowns with eIF5 con- formed exactly to the additive expectation of independent sgRN As. For nearl y all potential interaction partners, dou- ble knockdowns with eIF5 exhibited decreased CUG trans- lation compared to the single sgRNA knockdowns alone (Figure 3 E). This lack of genetic interactions was consistent with the role of eIF5 in mediating the final steps of initia- tion, downstream of other potential factors. Howe v er, we observed that eIF5’s known interaction partner eIF1A de- viated substantially from the additi v e e xpectation. As eIF5 and eIF1A physically interact and compete for binding to eIF5B ( 53–56 ), loss of eIF5 fully negated the eIF1A phe- notype, despite the single sgRNA eIF1A phenotype being stronger than the single sgRNA eIF5 phenotype. Lastly, we performed double knockdown experiments with eIF4G2. eIF4G2 is a homolog of eIF4G1 that can- not bind to eIF4E and instead relies on the cap-binding properties of eIF3D to recruit the ribosome to select mR- NAs ( 48–50 ). eIF4G2 exhibited substantial genetic inter- actions with eIF3 subunits and eIF1A (Figure 3 F). The combined knockdown phenotypes of eIF3H / eIF4G2 and eIF3M / eIF4G2 wer e buffer ed and r emained substantially weaker than the effect of eIF3D knockdown alone. These da ta suggest tha t ef fects of eIF4G2 on near-cogna te start codon usage could depend on its interactions with eIF3. eIF3D N-terminal domain is r equir ed f or stringent start codon selection To systematically explore the structure to function relation- ships within eIF3D, we combined sgRNA-mediated knock- down of endogenous eIF3D with exogenous rescue using various mutants (Figure 4 A, B). Full length eIF3D over- expression fully rescued eIF3D knockdown. C-terminal truncation, which r emoved r esidues 527–548 and a long poly-glutamic acid tract, remained capable of rescuing start codon selecti vity. Ne xt, we tested serine to aspartic acid (S528D / S529D) and serine to asparagine (S528N / S529N) muta tions a t two casein kinase II (CK2) phosphorylation sites near the eIF3D cap-binding domain. Phosphorylation of these residues was recently reported to activate eIF3D in response to metabolic stress ( 49 ). Howe v er, neither the phospho-mimetic aspartic acid substitution nor the non- phosphorylatable asparagine had an effect on eIF3D res- cue. We also tested removal of the ‘RNA gate’, an unstruc- tured loop of 15 amino acids between strand (cid:5)5 and helix (cid:4)6. Previous studies showed that this loop structurally reg- ulates the binding of eIF3D to mRNA and potentially pre- vents promiscuous mRNA binding ( 48 ). This mutant exhib- ited nearly full rescue as well, with only a minor increase in CUG translation. Structure-guided triple mutants of helix (cid:4)5 or (cid:4)11 that fully abolish eIF3D cap-binding activity also Nucleic Acids Research, 2023, Vol. 51, No. 12 6365 rescued depletion of endogenous eIF3D, indicating that the cap-binding activity of eIF3D was dispensable for near- cognate start codon usage. Finally, we tested an N-terminal truncation of residues 1–160. This N-terminal truncation mutant r epr esents the minimal stable human cap-binding domain and was shown to bind to the c-Jun mRNA 5´ cap in vitro ( 48 ). Unlike other mutants we tested, the N- terminal truncation mutant was unable to rescue loss of endogenous eIF3. These data further suggest that rescuing the cap-binding function of eIF3D alone was not sufficient for restoring normal start codon initiation. Recent cryo-EM structures showed that the N-terminal tail of eIF3D inter- acts with eIF3E and eIF3C, suggesting that these interac- tions may be essential in connecting eIF3D to the core eIF3 complex, which may then collectively regulate near-cognate translation ( 57 , 58 ). T r anscriptional signatur es of eIF3D depletion in multiple cell types We performed bulk RNA-seq in K562, HeLa and Jurkat cells to determine the transcriptional signature of eIF3 de- pletion across cell types. RNA-seq in K562 cells depleted for eIF3D re v ealed strong upregulation of immune-related genes, including IL-8, IL-32, IFI6, CD83 and CD44 (Figure 5 A, Supplementary Table S7). Comparison to the Molec- ular Signatures Database (MSigDB) re v ealed TNF (cid:4) sig- naling via NF- (cid:2)B as the dominant transcriptomic signa- −17 ), with 44 / 145 annotated genes upregulated ture ( P < 10 by > 2-fold (Figure 5 B). This transcriptional signature was shared in eIF3D-depleted Jurkat and HeLa cells. Jurkat cells upregulated a highly similar subset of genes compared to K562s, while HeLa cells upregulated a separate subset of genes within the TNF (cid:4) signaling via NF- (cid:2)B annotation set (Figure 5 B). Next, we asked whether the transcriptional profile of eIF3D depletion was similar to any other genetic pertur- ba tion using da ta from genome-wide Perturb-seq in K562 cells ( 44 ). Using a permissi v e clustering scheme on a 20- dimensional embedding of the Perturb-seq data, we ob- served that eIF3D knockdown cells clustered with cells depleted for eIF3E / F / H / L / M, eIF4A1, eIF4G2, eIF1A, DDX3X, CSDE1, STRAP and ZNF324 (Figure 5 C). The grouping of core eIF3 subunits with additional initiation factors demonstrates the power of Perturb-seq to identify functional modules based on shared changes in gene expres- sion and mirrors known biochemical interactions. Among the remaining genes in the cluster, we determined that the major effects of the sgRNA targeting ZNF324 were medi- ated by off-target r epr ession of eIF3H (Supplementary Fig- ure S5A). Secondary screening suggested that CSDE1 and STRAP mainly affected IRES-mediated translation rather than near-cognate start codon usage (Supplementary Fig- ure S5B-D). We further analyzed the genome-wide Perturb-seq data to test whether any part of the eIF3D transcriptional sig- nature was unique compared to all other genes. As a com- parison set, we picked major classes of highly essential genes such as ribosomal proteins, other major initiation fac- tors, and genes known to activate NF- (cid:2)B such as members of the ESCRT complex. Because Perturb-seq captures a 6366 Nucleic Acids Research, 2023, Vol. 51, No. 12 Figure 4. Evaluation of the structure-function relationship of eIF3D mutants. ( A ) Schematic of lentiviral construct for simultaneous knockdown of en- dogenous eIF3D and rescue with exogenous eIF3D mutants. ( B ) CUG reporter phenotypes for simultaneous eIF3D knockdown and rescue with eIF3D mutants. limited number of transcripts per cell, only the most highly expressed 5530 genes were analyzed. After filtering for dif- fer entially expr essed genes, we found that the eIF3D clus- ter did not upregulate an entirely unique set of genes com- pared to other perturbations (Figure 5 C). Instead, the clus- tering reflected subtle but coherent changes across se v eral sets of genes. Compared to depletion of ESCRT subunits, genes in the eIF3D cluster promoted more modest le v els of NF- (cid:2)B acti vation. Howe v er, due to the relati v ely lower read-depth in Perturb-seq, none of the most highly up- r egulated immune-r elated transcripts (IL-8, IL-32, IFI6, CD83 and CD44) from bulk-RNA-seq of eIF3D knock- down cells were detected. To test whether the transcriptional clustering defined by Perturb-seq would extend to lowly expressed genes, we per- formed bulk RNA-seq on cells expressing sgRNAs target- ing eIF3D, eIF4G2, and eIF1A. Across all three genetic perturba tions, we observed upregula tion of 60 / 145 genes in the TNF (cid:4) signaling via NF- (cid:2)B annotation set (Figure 5 D). Many of these genes were lowly expressed in wild-type K562 cells, with 37 / 60 expressed below 1 transcript per million (TPM). In addition, we observed coher ent upr egulation of 54 / 200 genes annotated as part of the interferon gamma re- sponse (Figure 5 E), with only 10 / 54 genes overlapping with the TNF (cid:4) signaling set. We confirmed that activation of NF-kB was not due to transcriptional r epr ession of key NF- kB inhibitors (Supplementary Figure S5E). These data thus show that depletion of genes in the eIF3D cluster promotes a moderate innate immune response dri v en by NF- (cid:2)B and interferon gamma. Lastly, we investigated direction of causality between in- creased near-cognate start codon usage and NF- (cid:2)B acti- vation, as both occurred upon eIF3D depletion. Double knockdown of eIF3D and numerous key NF- (cid:2)B signaling factors or TNF (cid:4) factors yielded no changes to the CUG start codon reporter (Supplementary Figure S6A). Simi- larly, double knockdown with NFKBIA, the (cid:4)-subunit of the inhibitory IKK complex ( 59 ), had no effect on the r eporter phenotype. Over expr ession of c-Jun, part of the AP-1 early response transcription factor complex ( 60 ), had no effect (Supplementary Figure S6B). We thus concluded tha t activa tion of NF- (cid:2)B signaling upon eIF3D depletion was not responsible for increased near-cognate start codon usage. DISCUSSION The fidelity of start codon selection plays a fundamental role in shaping the composition of the proteome. Tran- scripts contain up to hundreds of nucleotides upstream of the canonical start site that can potentially initiate trans- la tion from alterna ti v e reading frames. Initiation at up- stream ORFs (uORFs) can play critical regulatory roles in the translation of their downstream partners, as in the in- tegrated stress response ( 11 , 12 ). Furthermore, the peptide pr oducts fr om uORFs may sometimes exert functions inde- pendent from translational regulation ( 14 ). We performed unbiased genome-wide CRISPRi screen- ing to systematically identify the regulators of near-cognate start codon usage in human K562 cells. Our primary screen broadly recapitulated prior knowledge and identi- fied known initiation factors eIF1, eIF1A, eIF5 and eIF5B. The majority of strong hits occurred in highly essential genes whose full knockouts are unviable, demonstrating the unique value of our CRISPRi approach in eliciting highly tar geted knockdo wn phenotypes. Furthermore, we sho wed that depletion of all core eIF3 subunits and eIF3D in partic- ular led to unusually robust near-cognate start codon usage. While the exact mechanistic role of eIF3 and eIF3D in near-cognate start codon usage remains unclear, we ruled out alternati v e cap-binding by eIF3D / eIF4G2 and leucine tRNA initiation as potential mechanisms. Results from pre- vious structural and biochemical studies provide avenues f or future in vestiga tion. Biochemical characteriza tion of Nucleic Acids Research, 2023, Vol. 51, No. 12 6367 Figure 5. Transcriptional signatures of eIF3D depletion in multiple cell types. ( A ) Volcano plot of RNA-seq expression in K562s with eIF3D knockdown compar ed to wild-type. (B ) RNA expr ession of genes annotated as TNF (cid:4) signaling via NF- (cid:2)B by the Molecular Signa tures Da tabase (MSigDB) in K562, HeLa, and Jurkat cells. ( C ) Genome-wide perturb-seq transcriptional profiles for eIF3D and closely clustered genes compared to control genes. ( D ) Bulk RNA-seq gene expression for genes annotated as TNF (cid:4) signaling via NF- (cid:2)B by the Molecular Signatures Database (MSigDB) in K562 cells depleted for eIF3D, eIF4G2, or eIF1A. ( E ) Bulk RNA-seq gene expression for genes annotated as interferon gamma response by the Molecular Signatures Database (MSigDB) in K562 cells depleted for eIF3D, eIF4G2 or eIF1A. eIF3D knockdown re v ealed that the loss of eIF3D does not compromise the integrity of the rest of the eIF3 com- plex ( 46 , 61 ). This unique property of eIF3D suggests that loss of eIF3D may influence scanning or decoding via con- formational changes, whereas loss of other subunits leads to broader disruption of the eIF3 complex. In addition, our structure function data point to the N-terminal tail of eIF3D as being essential as opposed to its alternati v e cap-binding properties. These data also indicate that the effects of eIF4G2 on near-cognate usage are unlikely to be mediated by alternati v e cap-binding translation initia- tion ( 50 ). Recent cryo-electron microscopy structures have shown that the N-terminal tail of eIF3D physically inter- acts with eIF3C and eIF3E, connecting eIF3D to the rest of the eIF3 complex ( 57 , 58 ) suggesting that eIF3D deple- tion may allosterically affect the conformation of eIF3C, 6368 Nucleic Acids Research, 2023, Vol. 51, No. 12 e v en if eIF3D depletion does not alter the overall assem- bly of eIF3. The altered conformation of eIF3C could then in turn tune the stringency of the decoding site during scan- ning via interactions with eIF1 / 1A and eIF5 ( 36 , 37 ). Estab- lishing the exact structural effects of eIF3D depletion will r equir e structur al char acterizations of the scanning com- plex in the absence of eIF3D or biochemical reconstitution of the initiation machinery with depleted le v els of eIF3D. Lastly, the downstream consequences of eIF3D knock- down extended beyond leaky translation initiation and in- cluded activation of NF-kB and cessation of growth. De- pletion of other eIF3 subunits, eIF1A, or eIF4G2, induced a similar transcriptional response, suggesting that the ac- tiva tion of inna te imm unity could potentiall y be induced by altered start codon usage as opposed to depletion of eIF3 complexes alone. As many viral pathogens disrupt cel- lular translation, we speculate that the production of non- canonical ORFs could act as an intracellular signal that ac- tivates an antiviral response. It remains to be seen whether the inducer for such a response would involve cis -regulatory control of a master regulator by uORFs or whether in- creased production of specific uORF peptide products acti- va te inna te immune pa thways. DA T A A V AILABILITY Raw sequencing data has been submitted to the NCBI SRA under accession numbers SRR19744356-SRR19744369. The corresponding BioSample accession numbers are SAMN29198687-SAMN29198700. Processed RNA-seq data are provided in the supplementary materials. SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS We thank Thomas Norman for help in the initial clustering and analysis of genome-wide perturb-seq data. We thank Dian Yang for initial training in mammalian cell culture techniques. FUNDING Howard Hughes Medical Institute. Funding for open access char ge: Ho ward Hughes Medical Institute. Conflict of interest statement. J.S.W. declares outside interest in 5 AM Venture, Amgen, Chroma Medicine, DEM Bio- sciences , KSQ Therapeutics , Maze Therapeutics , Tenaya Ther apeutics, Tesser a Ther apeutics and Velia Therapeutics. REFERENCES 1. 1000 Genomes Project Consortium, Abecasis,G.R., Altshuler,D., A uton,A., Brooks,L.D ., Durbin,R.M., Gibbs,R.A., Hurles,M.E. and McVean,G.A. (2010) A map of human genome variation from population-scale sequencing. Nature , 467 , 1061–1073. 2. Drummond,D.A. and Wilke,C.O. (2009) The evolutionary consequences of erroneous protein synthesis. Nat. Rev. Genet. , 10 , 715–724. 3. Lane,N. and Martin,W. (2010) The energetics of genome complexity. Nature , 467 , 929–934. 4. Takahashi,K., Maruyama,M., Tokuzawa,Y., Murakami,M., Oda,Y., Yoshikane,N., Makabe,K.W., Ichisaka,T. and Yamanaka,S. (2005) Evolutionarily conserved non-AUG translation initiation in NAT1 / p97 / DAP5 (EIF4G2). Genomics , 85 , 360–371. 5. Imataka,H., Olsen,H.S. and Sonenberg,N. (1997) A new translational regulator with homology to eukaryotic translation initiation factor 4G. EMBO J. , 16 , 817–825. 6. Ingolia,N.T., Ghaemmaghami,S., Newman,J.R.S. and Weissman,J.S. (2009) Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science , 324 , 218–223. 7. Ingolia,N.T., Lareau,L.F. and Weissman,J.S. (2011) Ribosome profiling of mouse embryonic stem cells re v eals the complexity and dynamics of mammalian proteomes. Cell , 147 , 789–802. 8. Kearse,M.G. and Wilusz,J.E. (2017) Non-AUG translation: a new start for protein synthesis in eukaryotes. Genes Dev. , 31 , 1717–1731. 9. Calvo,S.E., Pagliarini,D.J. and Mootha,V.K. (2009) Upstream open reading frames cause widespread reduction of protein expression and are polymorphic among humans. Proc. Natl. Acad. Sci. U. S. A. , 106 , 7507–7512. 10. Pesole,G., Mignone,F., Gissi,C., Grillo,G., Licciulli,F. and Liuni,S. (2001) Structural and functional features of eukaryotic mRNA untranslated regions. Gene , 276 , 73–81. 11. Starck,S.R., Tsai,J.C., Chen,K., Shodiya,M., Wang,L., Yahiro,K., Martins-Green,M., Shastri,N. and Walter,P. (2016) Translation from the 5’ untranslated region shapes the integrated stress response. Science , 351 , aad3867. 12. Vattem,K.M. and Wek,R.C. (2004) Reinitiation involving upstream ORFs regulates ATF4 mRNA translation in mammalian cells. Proc. Natl. Acad. Sci. U.S.A. , 101 , 11269–11274. 13. Chen,J., Brunner,A.-D., Cogan,J.Z., Nu ˜ nez,J.K., Fields,A.P., Adamson,B., Itzhak,D.N., Li,J.Y., Mann,M., Leonetti,M.D. et al. (2020) Pervasi v e functional translation of noncanonical human open reading frames. Science , 367 , 1140–1146. 14. Wright,B.W., Yi,Z., Weissman,J.S. and Chen,J. (2021) The dark proteome: translation from noncanonical open reading frames. T r ends Cell Biol. , 32 , 243–258. 15. Erhard,F., Halenius,A., Zimmermann,C., L’Hernault,A., Kowalewski,D.J., Weekes,M.P., Stevanovic,S., Zimmer,R. and D ¨olken,L. (2018) Improved Ribo-seq enables identification of cryptic translation e v ents. Nat. Methods , 15 , 363–366. 16. Laumont,C.M., Daouda,T., Lav er dure,J.-P., Bonneil, ´E., Caron-Lizotte,O., Hardy,M.-P., Granados,D.P., Durette,C., Lemieux,S., Thibault,P. et al. (2016) Global proteogenomic analysis of human MHC class I-associated peptides deri v ed from non-canonical reading frames. Nat. Commun. , 7 , 10238. 17. Kozak,M. (1986) Point mutations define a sequence flanking the AUG initiator codon that modulates translation by eukaryotic ribosomes. Cell , 44 , 283–292. 18. Kapp,L.D. and Lorsch,J.R. (2004) GTP-dependent recognition of the methionine moiety on initiator tRNA by translation factor eIF2. J. Mol. Biol. , 335 , 923–936. 19. Sokabe,M., Fraser,C.S. and Hershey,J.W.B. (2012) The human transla tion initia tion multi-factor complex promotes methionyl-tRNAi binding to the 40S ribosomal subunit. Nucleic Acids Res. , 40 , 905–913. 20. Lomakin,I.B., Shirokikh,N.E., Yusupov,M.M., Hellen,C.U.T. and Pestova,T.V. (2006) The fidelity of translation initiation: reciprocal activities of eIF1, IF3 and YciH. EMBO J. , 25 , 196–210. 21. Kolitz,S.E., Takacs,J.E. and Lorsch,J.R. (2009) Kinetic and thermodynamic analysis of the role of start codon / anticodon base pairing during eukaryotic translation initiation. RNA N. Y. N , 15 , 138–152. 22. Hinnebusch,A.G. (2011) Molecular Mechanism of Scanning and Start Codon Selection in Eukaryotes. Microbiol. Mol. Biol. Rev. MMBR , 75 , 434–467. 23. Hinnebusch,A.G. (2014) The scanning mechanism of eukaryotic transla tion initia tion. Annu. Rev. Biochem. , 83 , 779–812. 24. Terenin,I.M., Akulich,K.A., Andreev,D.E., Poly anskay a,S.A., Shatsky,I.N. and Dmitriev,S.E. (2016) Sliding of a 43S ribosomal complex from the recognized AUG codon triggered by a delay in eIF2-bound GTP hydrolysis. Nucleic Acids Res. , 44 , 1882–1893. 25. Lomakin,I.B. and Steitz,T.A. (2013) The initiation of mammalian protein synthesis and mRNA scanning mechanism. Nature , 500 , 307–311. 26. Huang,H.K., Yoon,H., Hannig,E.M. and Donahue,T.F. (1997) GTP hydr olysis contr ols stringent selection of the AUG start codon during transla tion initia tion in Saccharomy ces cere visiae. Genes Dev. , 11 , 2396–2413. 27. Unbehaun,A., Borukhov,S.I., Hellen,C.U.T. and Pestova,T.V. (2004) Release of initiation factors from 48S complexes during ribosomal subunit joining and the link between establishment of codon-anticodon base-pairing and hydrolysis of eIF2-bound GTP. Genes Dev. , 18 , 3078–3093. 28. Pestova,T.V., Lomakin,I.B., Lee,J.H., Choi,S.K., De v er,T.E. and Hellen,C.U. (2000) The joining of ribosomal subunits in eukaryotes r equir es eIF5B. Nature , 403 , 332–335. 29. Castilho-Valavicius,B., Yoon,H. and Donahue,T.F. (1990) Genetic characterization of the Saccharomyces cerevisiae translational initiation suppressors sui1, sui2 and SUI3 and their effects on HIS4 expression. Genetics , 124 , 483–495. 30. Yoon,H.J. and Donahue,T.F. (1992) The suil suppressor locus in Saccharomy ces cere visiae encodes a transla tion factor tha t functions during tRNA(iMet) recognition of the start codon. Mol. Cell. Biol. , 12 , 248–260. 31. Fekete,C.A., Applefield,D.J., Blakely,S.A., Shirokikh,N., Pestova,T., Lorsch,J.R. and Hinnebusch,A.G. (2005) The eIF1A C-terminal domain promotes initiation complex assembly, scanning and AUG selection in vivo. EMBO J. , 24 , 3588–3601. 32. Donahue,T.F. and Cigan,A.M. (1988) Genetic selection for mutations that reduce or abolish ribosomal recognition of the HIS4 transla tional initia tor region. Mol. Cell. Biol. , 8 , 2955–2963. 33. Barth-Baus,D., Bhasker,C.R., Zoll,W. and Merrick,W.C. (2013) Influence of translation factor activities on start site selection in six different mRNAs. Transl. Austin Tex , 1 , e24419. 34. He,H., von der Haar,T., Singh,C.R., Ii,M., Li,B., Hinnebusch,A.G., McCarthy,J.E.G. and Asano,K. (2003) The yeast eukaryotic initiation factor 4G (eIF4G) HEAT domain interacts with eIF1 and eIF5 and is involved in stringent AUG selection. Mol. Cell. Biol. , 23 , 5431–5445. 35. Val ´asek,L., Nielsen,K.H., Zhang,F., Fekete,C.A. and Hinnebusch,A.G. (2004) Interactions of eukaryotic translation initiation factor 3 (eIF3) subunit NIP1 / c with eIF1 and eIF5 promote preinitiation complex assembly and regulate start codon selection. Mol. Cell. Biol. , 24 , 9437–9455. 36. Obayashi,E., Luna,R.E., Nagata,T., Martin-Marcos,P., Hiraishi,H., Singh,C.R., Erzberger,J.P., Zhang,F., Arthanari,H., Morris,J. et al. (2017) Molecular landscape of the ribosome pre-initiation complex during mRNA scanning: structural role for eIF3c and its control by eIF5. Cell Rep. , 18 , 2651–2663. 37. Kar ´asko v ´a,M., Guni ˇso v ´a,S., Herrmanno v ´a,A., Wagner,S., Munzarov ´a,V. and Val ´a ˇsek,L.S. (2012) Functional characterization of the role of the N-terminal domain of the c / Nip1 subunit of eukaryotic initiation factor 3 (eIF3) in AUG recognition. J. Biol. Chem. , 287 , 28420–28434. 38. Adamson,B., Norman,T.M., Jost,M., Cho,M.Y., Nu ˜ nez,J.K., Chen,Y., Villalta,J.E., Gilbert,L.A., Horlbeck,M.A., Hein,M.Y. et al. (2016) A multiplexed single-cell CRISPR screening platform enables systematic dissection of the unfolded protein response. Cell , 167 , 1867–1882. 39. Gilbert,L.A., Horlbeck,M.A., Adamson,B., Villalta,J.E., Chen,Y., W hitehead,E.H., Guimaraes,C ., Panning,B., Ploegh,H.L., Bassik,M.C. et al. (2014) Genome-scale CRISPR-mediated control of gene r epr ession and activation. Cell , 159 , 647–661. 40. Gilbert,L.A., Larson,M.H., Morsut,L., Liu,Z., Brar,G.A., Torres,S.E., Stern-Ginossar,N., Brandman,O., Whitehead,E.H., Doudna,J.A. et al. (2013) CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell , 154 , 442–451. 41. Jost,M., Santos,D.A., Saunders,R.A., Horlbeck,M.A., Hawkins,J.S., Scaria,S.M., Norman,T.M., Hussmann,J.A., Liem,C.R., Gross,C.A. et al. (2020) Titrating gene expression using libraries of systematically a ttenua ted CRISPR guide RNAs. Nat. Biotechnol. , 38 , 355–364. 42. Horlbeck,M.A., Gilbert,L.A., Villalta,J.E., Adamson,B., Pak,R.A., Chen,Y., Fields,A.P., Park,C.Y., Corn,J.E., Kampmann,M. et al. (2016) Compact and highly acti v e ne xt-gener ation libr aries for CRISPR-mediated gene r epr ession and activation. Elife , 5 , e19760. 43. Horlbeck,M.A., Xu,A., Wang,M., Bennett,N.K., Park,C.Y., Bogdanoff,D., Adamson,B., Chow,E.D., Kampmann,M., Nucleic Acids Research, 2023, Vol. 51, No. 12 6369 Peterson,T.R. et al. (2018) Mapping the genetic landscape of human cells. Cell , 174 , 953–967. 44. Replogle,J.M., Saunders,R.A., Pogson,A.N., Hussmann,J.A., Lenail,A., Guna,A., Mascibroda,L., Wagner,E.J., Adelman,K., Lithwick-Yanai,G. et al. (2022) Mapping information-rich genotype-phenotype landscapes with genome-scale Perturb-seq. Cell , 185 , 2559–2575. 45. Hinnebusch,A.G. and Lorsch,J.R. (2012) The mechanism of eukaryotic translation initiation: new insights and challenges. Cold Spring Harb. Perspect. Biol. , 4 , a011544. 46. Wagner,S., Herrmannov ´a,A., ˇSikrov ´a,D. and Val ´a ˇsek,L.S. (2016) Human eIF3b and eIF3a serve as the nucleation core for the assembly of eIF3 into two interconnected modules: the yeast-like core and the octamer. Nucleic Acids Res. , 44 , 10772–10788. 47. Wagner,S., Herrmannov ´a,A., Mal ´ık,R., Peclinovsk ´a,L. and Val ´a ˇsek,L.S. (2014) Functional and biochemical characterization of human eukaryotic translation initiation factor 3 in living cells. Mol. Cell. Biol. , 34 , 3041–3052. 48. Lee,A.S., Kranzusch,P.J., Doudna,J.A. and Cate,J.H.D. (2016) eIF3d is an mRNA cap-binding protein that is r equir ed for specialized transla tion initia tion. Natur e , 536 , 96–99. 49. Lamper,A.M., Fleming,R.H., Ladd,K.M. and Lee,A.S.Y. (2020) A phosphoryla tion-regula ted eIF3d translation switch mediates cellular adaptation to metabolic stress. Science , 370 , 853–856. 50. de la Parra,C., Ernlund,A., Alard,A., Ruggles,K., Ueberheide,B. and Schneider,R.J. (2018) A widespread alternate form of cap-dependent mRNA translation initiation. Nat. Commun. , 9 , 3068. 51. Starck,S.R., Jiang,V., Pavon-Eternod,M., Prasad,S., McCarthy,B., Pan,T. and Shastri,N. (2012) Leucine-tRNA initiates at CUG start codons for protein synthesis and presentation by MHC class I. Science , 336 , 1719–1723. 52. Sendoel,A., Dunn,J.G., Rodriguez,E.H., Naik,S., Gomez,N.C., Hurwitz,B., Levorse,J., Dill,B.D., Schramek,D., Molina,H. et al. (2017) Translation from unconventional 5’ start sites drives tumour initia tion. Natur e , 541 , 494–499. 53. Luna,R.E., Arthanari,H., Hiraishi,H., Nanda,J., Martin-Marcos,P., Markus,M.A., Akaba y ov,B., Milbradt,A.G., Luna,L.E., Seo,H.-C. et al. (2012) C-terminal domain of eukaryotic initiation factor 5 promotes start codon recognition by its dynamic interplay with eIF1 and eIF2 (cid:5). Cell Rep. , 1 , 689–702. 54. Ll ´acer,J.L., Hussain,T., Saini,A.K., Nanda,J.S., Kaur,S., Gordiyenko,Y., Kumar,R., Hinnebusch,A.G., Lorsch,J.R. and Ramakrishnan,V. (2018) Translational initiation factor eIF5 replaces eIF1 on the 40S ribosomal subunit to promote start-codon recognition. Elife , 7 , e39273. 55. Lin,K.Y., Nag,N., Pestova,T.V. and Marintchev,A. (2018) Human eIF5 and eIF1A compete for binding to eIF5B. Bioc hemistr y , 57 , 5910–5920. 56. Nanda,J.S., Saini,A.K., Mu ˜ noz,A.M., Hinnebusch,A.G. and Lorsch,J.R. (2013) Coordinated movements of eukaryotic translation initiation factors eIF1, eIF1A, and eIF5 trigger phosphate release from eIF2 in response to start codon recognition by the ribosomal preinitiation complex. J. Biol. Chem. , 288 , 5316–5329. 57. Brito Querido,J., Sokabe,M., Kraatz,S., Gordiyenko,Y., Skehel,J.M., Fraser,C.S. and Ramakrishnan,V. (2020) Structure of a human 48S transla tional initia tion complex. Science , 369 , 1220–1227. 58. Bochler,A., Querido,J.B., Prilepskaja,T., Soufari,H., Simonetti,A., Cistia,M.L.D., Kuhn,L., Ribeiro,A.R., Val ´a ˇsek,L.S. and Hashem,Y. (2020) Structural differences in translation initiation between pa thogenic trypanosoma tids and their mammalian hosts. Cell Rep. , 33 , 108534. 59. Karin,M. and Ben-Neriah,Y. (2000) Phosphorylation meets ubiquitination: the control of NF-(kappa)B activity. Annu. Rev. Immunol. , 18 , 621–663. 60. Shaulian,E. and Karin,M. (2001) AP-1 in cell proliferation and survival. Oncogene , 20 , 2390–2400. 61. Herrmannov ´a,A., Prilepskaja,T., Wagner,S., ˇSikrov ´a,D., Zeman,J., Poncov ´a,K. and Val ´a ˇsek,L.S. (2020) Adapted formaldehyde gradient cr oss-linking pr otocol implicates human eIF3d and eIF3c, k and l subunits in the 43S and 48S pre-initiation complex assembly, respecti v ely. Nucleic Acids Res. , 48 , 1969–1984.
10.1093_nar_gkad460	Published online 1 June 2023 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 https://doi.org/10.1093/nar/gkad460 A method to generate capture baits for targeted sequencing Balaji Sundararaman 1 , 3 ,* , Alisa O. Vershinina 2 , Samantha Hershauer 1 , 3 , Joshua D. Kapp 2 , Shelby Dunn 2 , Beth Shapiro 1 , 2 , 3 , 4 and Richard E. Green 1 , 4 ,* 1 Department of Biomolecular Engineering, University of California Santa Cruz, Santa Cruz, CA 95064, USA, 2 Department of Ecology and Evolutionary Biology, University of California Santa Cruz, Santa Cruz, CA 95064, USA, 3 Ho w ard Hughes Medical Institute, University of California Santa Cruz, Santa Cruz, CA 95064, USA. and 4 UCSC Genomics Institute, University of California Santa Cruz, Santa Cruz, CA 95064, USA Received February 15, 2023; Revised April 19, 2023; Editorial Decision May 09, 2023; Accepted May 15, 2023 ABSTRACT GRAPHICAL ABSTRACT Hybridization capture approaches allow targeted high-throughput sequencing analysis at reduced costs compared to shotgun sequencing. Hybridiza- tion capture is particularly useful in analyses of ge- nomic data from ancient, environmental, and foren- sic samples, where target content is low, DNA is fragmented and multiplex PCR or other targeted ap- proaches often fail. Here, we describe a DNA bait synthesis approach for hybridization capture that we call C ircular N ucleic acid E nrichment R eagent, or CNER (pronounced ‘ snare’ ). The CNER method uses r olling-cir c le ampliﬁcation f ollowed b y restric- tion digestion to discretize microgram quantities of h ybridization pr obes. We demonstrate the utility of the CNER method by generating probes for a panel of 23 771 known sites of single nucleotide polymor- phism in the horse genome. Using these probes, we capture and sequence from a panel of ten an- cient horse DNA libraries, comparing CNER cap- ture efﬁciency to a commerciall y a v ailable appr oach. With about one million read pairs per sample, CN- ERs captured more targets (90.5% versus 66.5%) at greater mean depth than an alternative commercial approach. INTRODUCTION Compared with whole-genome sequencing, targeted se- quencing is a cost-effecti v e method for analyzing specific genomic regions ( 1 ). Targeted sequencing has wide applica- tion in dia gnostics, meta genomic, phylogenetic, ancient and environmental DNA studies, and forensics ( 2 , 3 ). In targeted sequencing, r egions of inter est ar e enriched by hybridiza- tion capture using target-specific probes or by PCR am- plification using target-specific primers, followed by high- throughput next-generation sequencing (NGS). Hybridiza- tion capture methods overcome drawbacks of PCR-based target enrichment, including scalability to a large number of targets, PCR failure and PCR artifacts ( 1 , 2 ). Pioneering hybridization capture experiments used DNA arrays to enrich for targeted sequencing of human sam- ples ( 4–7 ) and Neanderthal ancient DN A (aDN A) ( 8 ). In these array-based hybridization capture methods, NGS li- brary molecules were hybridized to a microarray imprinted * To whom correspondence should be addressed. Tel: +1 831 502 7394; Email: [email protected] Correspondence may also be addressed to Balaji Sundar ar aman. Email: [email protected] C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. e69 Nucleic Acids Research, 2023, Vol. 51, No. 13 PAGE 2 OF 11 with probes targeting human exons. After washing non- hybridized library molecules off the surface of the ar- ray, captured molecules were eluted and sequenced ( 4–8 ). Array-based hybridization capture expanded the capabil- ity to millions of target regions, beyond what is achiev- able with PCR-based enrichment methods ( 1–3 ). However, array-based capture is labor and time-intensi v e and r equir es large amounts of input DNA as well as specialized instru- mentation for capture. In-solution hybridization capture is currently the most commonly used method of targeted sequencing due to the commercial availability of capture probes and the simplic- ity of the approach ( 2 , 3 ). In-solution hybridization capture uses biotinylated DNA or RNA molecules (baits) to capture target regions ( 1–3 , 9 ). A molar excess of biotinylated baits is hybridized with NGS libraries in solution. The resulting library-bait heteroduplex es ar e captur ed on str eptavidin- coated magnetic beads. Unbound non-target molecules are w ashed aw ay, and target molecules ar e r ecover ed for se- quencing ( 9 , 10 ). −2 –10 Current bait synthesis methods require large-scale oligonucleotide chemical synthesis and / or in vitro tran- scription. Both RN A and DN A bait generation r equir es synthesizing template oligonucleotides using phospho- r amidite chemistry. Microarr ay-based synthesis gener ates oligonucleotides in femtomole scales with chemical cou- −3 ( 11 , 12 ). Templates synthe- pling error rates of 10 sized at small-scale r equir e enzymatic amplification before use in hybridization capture. For RNA baits, PCR ampli- fied oligo templates are transcribed in vitro into biotiny- lated RNA baits as initially described by Gnrike et al. ( 9 ). Howe v er , in vitro transcription using T7 RN A pol ymerase can lead to amplification biases based on the templates’ se- quence, length, and GC content ( 13 , 14 ). For DNA baits, ei- ther a small-scale template pool is enzymatically amplified (Twist Biosciences product sheet) or each bait is individu- ally manufactured at scale (IDT product sheet). We present a cost-effecti v e, large-scale DNA bait synthe- sis method that we call C ircular N ucleic acid E nrichment R eagent, or CNER (pronounced as snare ). The CNER method involves circularization of target template oligos that contain a linker region to promote circularization via splint-ligation and a rar e-cutter r estriction enzyme site for subsequent discretization of the capture probes. Circular- ized templates are isothermally amplified by rolling circle amplification (RCA) with the inclusion of biotinylated nu- cleotides. The long RCA products are discretized into single biotinylated baits by restriction digestion (Figure 1 ). The resulting biotinylated CNER probes can be generated in microgram quantities and used for capture enrichments on streptavidin-coated beads. Here, we demonstrate the use of the CNER method for targeted genotyping by producing a set of CNER probes to capture 23771 SNPs in the horse genome. We use these CN- ERs to capture target SNPs from ten ancient horse DNA libraries of varying endogenous DNA content and DNA degradation le v els. We show tha t the CNERs ef fecti v ely perform target enrichment e v en in highly degraded ancient samples comparably to or better than commercially made baits and at a fraction of the cost. MATERIALS AND METHODS DNA isolation We selected ten ancient horse samples of varying DNA preservation (details in Supplementary Table S1 and in ( 15 )) to test the performance of the CNER method. The sam- ples date to the Late Pleistocene between 10 000 and 50 000 years ago, based on stratigraphic information and directly radiocarbon dated collagen (Supplementary Table S1 and in ( 15 )). We extracted ancient DNA following ( 16 ) in a ded- ica ted ancient DNA labora tory a t the UC Santa Cruz Pale- ogenomics Laboratory (PGL) and following standard pro- tocols for handling ancient DNA ( 17 ). We isolated DNA from four modern domestic horses for capture optimization using blood samples drawn in May / June 2017 during routine veterinary checks. We used the DNeasy Blood & Tissue kit (Qiagen) following the man- ufacturer’s protocol. Sequencing library preparation We pr epar ed NGS libraries from each horse extract us- ing the Santa Cruz Reaction (SCR) ( 18 ). For the modern horse, we fragmented genomic DNA using 0.02U DNase ◦C for 15 min with MgCl 2 before I (Thermo Fisher) at 15 proceeding with the SCR. We pr epar ed ancient horse DNA libraries in the dedicated clean at the PGL. For both an- cient and modern samples, we divided adapter-ligated DNA into three aliquots before PCR amplification. We PCR- amplified ancient DNA libraries with Illumina unique dual index primers ( 19 ) using 2x AmpliTaq Gold 360 master mix ◦C for 10 min, followed by 10–15 cy- (Thermo Fisher) at 95 ◦C for 1 min, with a cles of 95 ◦C. final extension at 72 We PCR amplified the modern horse libraries with Illumina unique dual index primers using 2 × KAPA HiFi master mix ◦C for (Roche) at 98 ◦C for 20 s, with a final extension at 30 s, 65 ◦C. We purified the amplified 72 libraries with SPRI ( 20 ) beads at 0.8 × ratio for the modern horse and at 1.2x for the ancient horses, quantified the DNA using Qubit 1 × HS assay (Thermo Fisher), and determined library size by Fragment Analyzer (Agilent). ◦C for 20 s, 72 ◦C for 3 min then hold at 12 ◦C for 3 min, followed by 13 cycles of 98 ◦C for 7 min followed by a hold at 12 ◦C for 30 s, 72 ◦C for 30 s, 60 Horse SNP panel design We designed the horse SNP panel for target enrichment of known nuclear SNPs based on the SNP ascertainment scheme described in ( 15 ). Briefly, we genotyped Batagai ( 21 ), CGG10022 ( 22 ), YG188.42 / YT03-40 and YG303.325 (both from (15) ) ancient horse genomes mapped to Equ- Cab2 (GenBank: GCA 000002305.1; ( 23 )) as described in ( 15 ), using samtools v.1.7 utilities mpileup and bcftools ( 24 ), AntCaller v1.1 ( 25 ), and GATK HaplotypeCaller 3.7 ( 26 ). We intersected variant calls from all three programs using VCFtools v0.1.16 vcf-isec ( 27 ). In downstream anal- yses, we used only variants called by all three programs. We also removed variants with < 20 base call quality, < 5X read coverage, location within 5 bp of indels, singletons and homozygous alternati v e alleles in all four ancient horse PAGE 3 OF 11 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 Figure 1. C ircular N ucleic acid E nrichment R eagent method. An oligonucleotide template pool containing restriction enzyme recognition sites (RES) and oligo-dT linkers is circularized by an oligo-dA splint adapter mediated ligation. Circularized templates are isothermally amplified using oligo-dA and oligo-dT oligos by rolling circle amplification (RCA). RCA products are then digested with restriction enzymes to generate CNERs. CNERs generate both strands (dark and light shades of colors) of the templates. Biotinylated nucleotides (purple diamonds) are incorporated during amplification. genomes. We selected SNPs located outside of gene bound- aries and repetiti v e regions using the filtering strategy de- scribed in ( 15 ). We selected the candidate set of 26944 variant loci for bait designing by Arbor Biosciences. Arbor provided us a list of 74 385 candidate baits. We filtered these to limit to 60K baits based on the chosen synthesis tier. We chose baits with 20– 80% GC content, filtered out baits containing repeats using RepeatMasker and baits with strong secondary structures ( (cid:2) G > –9 kcal / mol). After filtering, we chose a final list of baits to target 22 619 variant loci to proceed with Arbor my- Baits generation. The final Arbor panel targeted 2583 SNPs using one bait, 3391 SNPs using two baits, and 16 645 SNPs using three baits, and 228 Y-chromosome targets represent- ing sequence-tagged sites (STS), AMLEY and SRY genes. All 59528 Arbor myBaits were 80 nt long RNA probes. For CNERs generation, we targeted the same randomly selected 22 619 autosomal SNPs, each with one 80-bp long CNERs centered at the SNP site, plus the same 228 Y chro- mosome targets. To test the effect of CNERs length on cov- erage, we selected two additional sets of 576 SNPs and de- signed 50 bp and 100 bp CNERs with SNPs at the center. In total, the horse SNP panel targets 23 771 SNPs using a total of 23 999 probes. Horse SNP panel CNERs generation We generated CNERs for the horse SNP panel as schemat- ically described in Figure 1 . We appended six deoxy-T (dT) (cid:4) end, and AscI restriction site and (dT) 6 bases at the 5 (cid:4) end to all horse target regions to make CN- at the 3 ERs templates. We synthesized the templates as an DNA oligo pool using silicon chip based phosphoramidite chem- istry (Twist Biosciences). We circularized 100 or 300 fem- tomoles of the oligo pool in a 20 (cid:2)l splint ligation re- action containing 2000 U T4 DNA ligase (NEB), 10 U T4 PNK (NEB) and 1000 fmol (dA) 12 splint oligo in 1 × ◦C for 1 h followed by 25 ◦C T4 DNA ligase buffer at 37 ◦C for 3 min. We amplified for 3 h and dena tured a t 95 the circularized oligo pool in a 50 (cid:2)l RCA reaction con- taining 30U of Phi29 polymerase (NEB), 25 pmol each (cid:4) ) and re- of forward (5 (cid:4) ) RCA primers, 2 verse (5 nmol each of biotin-11-dATP (Perkin Elmer) and biotin-11- dUTP (Thermo Fisher), 25 nmol each dNTPs in 1X Phi29 (cid:4) -AAAAAAAAAGGCGCGCC-3 (cid:4) -GGCGCGCCTTTTTTTTT-3 ◦C , we buffer with BSA. After 40–48 h of RCA reaction a t 30 purified RCA products using SPRI beads (1.2 × ratio) and ◦C to produce digested with 100 U AscI (NEB) for 5 h at 37 monomeric CNERs. We estimated size and concentration of RCA products before and after AscI digestion using cap- illary electrophoresis in a Fragment Analyzer (Agilent) with the genomic DNA kit. We purified post-digestion products using SPRI beads (2 × ratio) and quantified the DNA using a Qubit (Thermo Fisher). CNERs hybridization capture optimization We optimized CNERs capture for adapter blocker con- centration, CNER amount per reaction, and hybridization buffer compositions. To optimize adapter blocker concen- tr ation, we titr ated oligonucleotide blockers at 5 ×–200 × molar excess to 100–300 ng (1.0–2.3 pmol) of the modern horse libraries, 25 ng horse SNP panel CNERs, 2.5 (cid:2)g of Human c0t DNA, and 25 (cid:2)g of salmon sperm DNA in 25 ◦C for 10 min. We (cid:2)l reaction, and then denatured at 95 added this DNA mixture to 25 (cid:2)l prewarmed Hyb buffer (final concentrations: 6 × SSPE, 6 × Denhardt’s solution, 10 mM EDTA, pH 8.0, 0.2% SDS) and hybridized the mix- ◦C . To ture overnight in 50 (cid:2)l total reaction volume a t 65 optimize CNERs amount titrations, we hybridized 300 ng of libraries with 30–90 ng of horse SNP panel CNERs and ◦C 200x molar excess oligo blockers in the Hyb buffer at 65 overnight. We tested four hybridization buffers (HB1: 100 mM MES pH 6.5 and 1 M NaCl; HB2: 6 × SSC, pH 7.0; HB3: 6 × SSPE, pH 7.4; and HB4: 100 mM Tris pH 8.0 and 1 M NaCl) to capture 250 ng of libraries using 50 ng CN- ◦C . All four buffers also contained 0.1% ERs overnight a t 65 SDS, 10 mM EDTA and 10% DMSO at final concentration. We captured CNER hybridized libraries onto 30 (cid:2)l MyOne ◦C for 30 min. C1 streptavidin beads (Thermo Fisher) at 65 We washed beads three times in high stringency wash buffer (0.2 × SSC, 0.1% SDS and 10% DMSO) for 5 min each at ◦C and then three times in low stringency buffer (2 × SSC 65 and 0.1% SDS) at room temperature. We washed beads in 10 mM Tris pH 8.0 before resuspending in the PCR reac- tion. We amplified post-captured libraries using 2 × KAPA HiFi master mix (Roche) and Illumina uni v ersal amplifi- ◦C for 3 min, followed by 15 cycles of cation primers at 98 ◦C for 30 s, with a final ex- ◦C for 30 s, 60 98 ◦C. We purified tension at 72 ◦C for 5 min then hold at 12 ◦C for 30 s, 72 e69 Nucleic Acids Research, 2023, Vol. 51, No. 13 PAGE 4 OF 11 post-capture libraries with 0.9 × SPRI beads, quantified us- ing a Qubit (Thermo Fisher), pooled, and sequenced on an Illumina NextSeq using PE 2 × 150 kit. Ancient horse DNA capture and sequencing For the ancient horse samples, we captured 5 (cid:2)l (constant li- brary volume with varying library mass; see Supplementary Table S2 for details) of individual ancient horse libraries us- ing Arbor myBaits and CNERs. For both Abor myBaits and CNERs captures, we performed two experiments. In experiments A1 (CNERs) and A2 (Arbor myBaits), we fol- lowed the Arbor myBaits protocol and used 50% of capture beads for post-capture amplification and purified libraries with 1.7 × SPRI as per the protocol. In experiments B1 (CN- ERs) and B2 (Arbor myBaits), we followed the optimized CNERs protocol, and used 100% of capture beads for PCR and 0.9 × SPRI for cleanup. Finally, we performed a sepa- rate CNERs Experiment C, in which we captured libraries in 3-plex pools. In experiment C, we also used 100% of cap- tured beads for PCR amplification and purified the post- capture libraries with 0.9 × SPRI. For all experiments using CNERs, we used 2 (cid:2)l ( ∼40 ng) of the horse SNP panel CNERs. For a single sample, UAM:ES:27502, for which little material remained at the start of the experiment, we used only 2 (cid:2)l of library CN- ERs in both experiment A and B. For all other samples, we used 5 (cid:2)l libraries for captures. We added 200 × adapter blocking oligos, 2.5 (cid:2)g of Human c0t DNA and 25 (cid:2)g of salmon sperm DNA to these library-CNERs to a total of ◦C for 10 min. We 30 (cid:2)l volume, and then denatured at 95 ◦C for 5min, mixed with de- preincubated 30 (cid:2)l of HB4 at 62 natured library / CNERs / blockers mixture and hybridized ◦C for 19.5 h. We enriched post-hybridization libraries at 62 onto streptavidin beads as in the optimization experiments except both low and high stringency wash steps were done a t 65 ◦C . For CNERs experiment C (pooled capture), we hy- bridized 67–100 ng of libraries for each of three samples with similar endogenous content with 40–60 ng CNERs (Supplementary Table S2). We repeated the individual cap- ture for UAM:ES:26433, rather than including it in a pool, as it had the lowest pr e-captur e endogenous content. We did not perform pooled captures for Arbor myBaits as it was not recommended by the manufacturer. For all captures using Arbor myBaits, we used 5 (cid:2)l of the same ancient horse libraries that we used in CNERs cap- tures. We used unopened vial of the Arbor myBaits Horse ◦C SNP panel. Although the baits had been stored at -80 continuously since production, they were 15 months older than the labeled use-by date. We followed Arbor Biosciences capture protocol v3 with recommended modifications of hy- bridiza tion a t 55 ◦C f or 41 h f or ancient DNA. We used different approaches to post-capture library am- plification in experiments A compared to experiments B. In A, we resuspended capture beads in 30 (cid:2)l 10 mM Tris pH 8.0 buffer. We then used 15 (cid:2)l of the resuspended beads in 20 cycles of PCR amplification with 2 × KAP A HiFi. W e then purified the product with 1.7 × SPRI, as recommended by Arbor. For B, we resuspended capture beads in 20 (cid:2)l 10 mM Tris pH 8.0 buffer and used all of it in a 50 (cid:2)l PCR reaction and performed 20 cycles of amplification, followed by pu- rification with 0.9 × SPRI. All post-capture libraries were Qubit (Thermo Fisher) quantified, pooled, and sequenced on an Illumina NextSeq with a PE 2 × 75 kit. Bioinformatic processing We trimmed adapter sequences from the reads and merged overlapping paired end reads using SEQPREP2 ( https:// github.com/jeizenga/SeqPrep2 ). We mapped merged and unmerged reads to the EquCab2 reference ( 23 ) genome us- ing BWA aln (version 0.7.17-r1188, 28 ). We marked and removed duplicated reads using Picard MarkDuplicates - v2.21.7 and calculated capture metrics using Picard Col- lectHsMetrics (version 2.21.7, http://broadinstitute.github. io/picard ). We determined read coverage at target SNPs using bedtools multicov (version 2.29.1). We plotted SNP covera ge a gainst CNERs length, GC content, and percent targets using custom python scripts ( https://github.com/ bsun210/CNERs ancient horses ). We used bedtools inter- sect (version 2.29.1) to find sequence reads mapping to the target SNPs to calculate the position of SNPs relati v e to the sequence read insert size. We determined genotype likelihoods for the ancient horses using ANGSD (version 0.935–52-g39eada3) with -GL 2 -minMapQ 20 -nThreads 24 -doGlf 2 -doMajorMinor 1 -SNP pval 1e-6 -doMaf 1 op- tions ( 29 ). We analyzed population clustering and ancestry using PCANGSD (version 1.10) with default settings ( 30 ). We used prcomp and f actoextr a R packages (Kassambara. A and Mundt. F. (2020) Factoextr a: Extr act and Visualize the Results of Multivaria te Da ta Analyses. https://cran.r- project.org/w e b/packages/factoe xtra/inde x.html ) for prin- cipal component analysis (PCA). We calculated endoge- nous content (proportion of unique reads aligned to the horse genome), library complexity (proportion of uniquely- mapped non-duplicated molecules) and insert size distribu- tion using the pipeline described in ( 15 ). We assessed whether the SNP coverage for CNERs with different lengths, changes in endogenous content, library complexity, and insert size between pre and post-capture libraries are normally distributed using the Shapiro-Wilk test. All these groups are not normally distributed; hence we performed a nonparametric Mann-Whitney Wilco x on (MWW) rank test for comparison between groups. For comparison of normalized covera ge distrib ution across GC bins for various experimental groups, we used two sample Kolmogoro v-Smirno v (KS) tests for goodness of fit. RESULTS The CNER method is designed to generate large amounts of biotinylated baits for hybridization captur e (Figur e 1 ). CNER templates are synthesized as oligonucleotides with (cid:4) ends to facilitate circular- (cid:4) and 3 oligo-dT linkers at both 5 ization using a complementary, oligo-dA splint. Because the linkers are oligo-dT, this design limits the impact of incom- plete oligonucleotide chemical synthesis errors at the tem- (cid:4) end upstream of the oligo-dT, a rare- plate ends. In the 3 cutter restriction enzyme recognition site (RES) is also in- corporated (Figure 1 ). Oligo-dT and rare cutter RES are PAGE 5 OF 11 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 appended to all target sequences such that all CNER tem- plates have uniform ends to facilitate bulk circularization by splint ligation using an oligo-dA splint adapter (Figure 1 ). After circulariza tion, CNER templa tes are bulk ampli- fied by rolling circle amplification (RCA) using high proces- sivity phi29 DN A pol ymerase. The RCA reaction includes biotin-dATP and biotin-dUTP (an ine xpensi v e and widely availab le alternati v e f or biotin ylated dTTP) in the reac- tion to generate biotinylated products. An oligo-dA forward primer and oligo-dT re v erse primer initiate forwar d and re v erse RCA reactions. Thus, the RCA products for each CNER template is doub le-stranded, regar dless of which strand the original CNER template was designed against (Figure 1 ). This conveniently generates probes against both strands of each CNER targeted region. Further, inclusion of both forward and reverse primers facilitate branched am- plification during RCA to increase yield. The RCA makes many of copies of the CNERs as concatemers, a single re- striction enzyme digestion of which produces monomeric, biotinylated capture probes (Figure 1 ). The monomeric CNERs can ther efor e be used as baits to capture and enrich target molecules on streptavidin-coated beads for sequencing. We designed a horse SNP panel with 23 771 randomly selected SNPs from a list of high confidence variant sites ascertained in four ancient horse genomes ( 15 ). Chemical synthesis of oligo templates for this panel yielded a 215 ng (6.3 pmol) pool. RCA amplification of 100 femtomoles ( ∼3.3 ng) bulk circularized template pool generated 611 ng of double-stranded high-molecular weight DNA ( ∼77 kB average size, Supplementary Figure S1A), restriction di- gestion of which generated 499 ng of monomeric CNERs with 114 bp average size (Supplementary Figure S1B). The presence of double-stranded DNA indicates that the CN- ERs method generates probes against both strands of the target region. In a separate experiment, we increased the input template to 300 femtomoles. The protocol yielded 1.57 (cid:2)g CNERs in that experiment. Thus, we estimate 100 fmol ( ∼3.3 ng) of circularized CNER templates produces ∼500 ng of CNERs using the protocol as described. CNER hybridization optimization We optimized in-solution hybridization conditions for the horse SNP panel CNERs using the modern horse DNA libraries (see Supplementary Data). We tested hybridiza- tion capture reactions with increasing amounts of adapter blocking oligos to prevent cross-hybridization of library molecules ( 31 ) with a constant amount of CNERs. In a separate set of experiments, we tested increasing amount of CNERs with a constant amount of blocking oligos. Both in- creasing amount of blocking oligos and CNERs modestly improved the enrichment efficiency (Supplementary Table S3, Supplementary Figure S2A and B). We note that con- ventional hybridization buffer like those used by Arbor my- Baits for RNA baits ( 32 ) might be suboptimal for DNA baits. Ther efor e, we tested four hybridiza tion buf fers (HB) to improve the enrichment efficiency for CNERs. Captures in HB4 produced > 50% (by Picard metrics) bases on or near targets for the modern horse libraries (Figure 2 A). Addi- ti v es used in conventional hybridization buffers like Den- hardt’s solution and trimethyl ammonium chloride did not improve and or lowered the percentage of on or near tar- get bases (Supplementary Figure S2C). Hybridization at ◦C also resulted in similar enrichment efficiency 62 (Supplementary Figure S2D). ◦C and 65 Existing capture bait synthesis methods use different probe lengths and tiling to optimize for the GC content of target regions (33, 34). We designed CNERs with three dif- ferent lengths to test the effect of CNER length on SNP coverage. The 80 bp CNERs produce higher SNP coverage than either 50 bp or 100 bp CNERs (Figure 2 B) consis- tently across various hybridization conditions (Supplemen- tary Figure S3). Further, target regions within 43–65% GC bins, which are 47% of the total target SNP regions (aver- age GC = 43.8%), consistently resulted in ≥1 normalized coverage (Figure 2 C, Supplementary Figure S4). CNERs efficiently capture ancient DNA target SNPs We extracted DNA from ten horse bones collected from Late Pleistocene age permafrost deposits in Alaska, USA and Chukotka, Russia (Supplementary Table S1 and ( 15 )). Sequence reads generated from each of these samples, mapped to the EquCab2 r efer ence genome, provided esti- mates of endogenous DNA content. Before SNP enrich- ment, the ancient horse DNA libraries had 18.4% me- dian reads mapped to the horse genome, across a wide range (6.0–91.2%, ‘preCap’ in Figure 3 A, Supplementary Table S2). SNP enrichments using both DNA based CNERs and RNA based Arbor myBaits increased the proportion of reads in the sequencing library that mapped to the r efer ence genome, indicating successful target enrichment. Enrich- ment using CNERs improved median precent of mapped reads to 37.9% in experiment A (individual captures follow- ing the Arbor myBaits protocol), and 30.5% in experiment B (individual captures following the CNERs protocol), and 40.1% in experiment C (pooled-captures with CNERs pro- tocol). Arbor myBaits resulted in 28.8% in experiment A (individual capture following the Arbor myBaits protocol), and 21.1% in experiment B (individual capture following the CNERs protocol) (Figure 3 A, Supplementary Table S4. Comparison of CNERs e xperiments B v ersus C show a con- sistent proportion of mapped reads when a sample was cap- tured indi vidually v ersus as part of a pool (Figure 3 A). The differences between capture probes and protocols are not significant by Mann–Whitney Wilco x on test. Different SPRI bead ratio used in the post-capture pu- rification steps did not affect the proportion of mapped r eads (Figur e 3 A). Howe v er, the dif ferent SPRI ra tio re- sulted in different proportions of merged and unmerged reads identified during data analyses. Short insert size of aDNA molecules result in overlapping read pairs which are merged during data processing, hence called as merged reads. Read pairs that did not overlap are processed as un- merged read pairs. Following the Arbor myBaits protocol which uses 1.2x SPRI beads ratio (experiments A) resulted in a higher proportion of merged compared to unmerged reads for both Arbor myBaits and CNERs (Supplemen- tary Figure S5A, Supplementary Table S4). All experiments that followed the CNERs cleanup protocol resulted in equal e69 Nucleic Acids Research, 2023, Vol. 51, No. 13 PAGE 6 OF 11 Figure 2. Optimization of CNERs hybridization capture of SNPs in four modern horse samples. ( A ) Enrichment efficiency f or f our hybridization buffers with pH varying from 6.5 to 8.0 (HB1 - 4). Light grey bars show the Percent Selected Bases determined using Picard tools and dark grey bars show the SNP enrichment efficiency. Values pr esented ar e the average of three experiments for HB1 and HB4 buffers and exact values for a single experiment for HB2 and HB3. ( B ) Histogram density plots of SNP coverage depth for three CNER lengths. SNPs captured with 80bp CNERs (blue bars) result in significantly higher coverage compared to SNPs captured with 50 bp (grey bars) or 100 bp (orange bars) CNERs; p-value is from a Mann–Whitney Wilco x on test. Dotted lines indicate the mean coverage for each CNERs length. ( C ) Mean of normalized coverage (primary Y-axis) plotted across GC content of CNER tar get regions sho w that regions with 43–65% GC have sample-normalized coverage of 1 or higher. A histogram of GC bins across the target regions is shown in the secondary Y-axis. proportions of merged and unmerged reads regardless of probes, due to the lower SPRI beads ratio (0.9 ×) used dur- ing the post-amplification cleanup. Across all experiments, a greater proportion of merged reads mapped to the refer- ence genome compared to unmerged reads, as expected for aDNA (Supplementary Figure S5B). Previous studies used Picard’s program CollectHsMetric to measure the success of target enrichment ( 35 ) . This tool reports coverage of the targeted base and 100bp flanking re- gions when determining ‘Percent Selected Bases’. We used this metric during the optimization experiments to compare the performance of CNERs to current standar ds. Howe v er, this metric overestimates the SNP enrichment success by in- cluding the regions around the target SNP site. Ther efor e, we elected to measure the success of SNP enrichment in an- cient horses by defining ‘SNP enrichment efficiency’ as the percentage of all or mapped reads that overlap the target SNPs. This is a straightforward and more practically impor- tant measure of SNP enrichment success. For the modern ◦C horse captures with CNERs, hybridization in HB4 at 65 for 18–20 h produced ∼30% SNP enrichment efficiency for mapped reads (Figure 2 A). We followed these hybridization conditions to capture ancient horse samples. SNP enrichment efficiency, or the proportion of reads mapping to the target SNPs, was significantly higher when using CNERs compared to when using Arbor myBaits. In experiments A (Arbor myBaits protocol), the median SNP enrichment efficiency was 15.7% for CNERs ver- sus 4.8% for Arbor (MWW P < 0.05). In experiments B (CNERs protocol), the median SNP enrichment effi- ciency was 14.5% for CNERs versus 4.3% for Arbor my- Baits (MWW P < 1e-2; Figure 3 B, Supplementary Table S4). This pattern holds when considering only reads that map to the r efer ence genome. Experiments A (Arbor my- Baits protocol) resulted in median enrichment efficiencies of mapped reads of 32.4% for CNERs versus 17.7% for Ar- bor myBaits (MWW P < 1e-2), and experiments B (CN- ERs protocol) resulted in median efficiencies of mapped reads of 31.5% for CNERs versus 15.2% for Arbor my- Baits (MWW P < 1e-3; Figure 3 C). The pattern is also consistent when considering merged and unmerged reads separately, both for all reads and mapped reads (Supple- mentary Figure S5C and D), although unmerged reads al- ways had significantly lower enrichment efficiency com- pared to merged reads (Supplementary Figure S5D, Supple- mentary Table S4). Finally, the enrichment efficiency when using CNERs was consistent between individually cap- tured libraries and captures performed in pools (Figures 3 B and C). To test the potential impact of differences in sequencing depth, we subsampled data to one million read pairs per sample in experiments A and B. For this analysis, we con- sidered only the 22 619 target SNPs that were common be- tween CNERs and Arbor myBaits. For experiments A (Ar- bor myBaits protocol), this read depth resulted in a me- dian of 90.5% (20 479) of target SNPs covered by at least one unique read using CNERs versus 66.5% (15 038) for Arbor myBaits (MWW P < 1e-2; Figure 3 D, Supplemen- tary Table S4). We observed a similar trend when following the CNERs protocol (experiments B; Figure 3 D). At this coverage, CNERs captures have fewer SNP dropouts com- pared to Arbor myBaits captures, as estimated using cu- mulati v e distribution plots of SNP coverage as percentage of SNPs less than the x-fold mean coverage (Supplemen- tary Figure S6). When averaged across the 10 horse data sets at this standard coverage, CNERs captures resulted in 2.5-fold higher average SNP coverage than Arbor myBaits (an average of 5.4 reads per SNP compared to an average of 2.2 reads per SNP when using the Arbor myBaits pro- tocol (experiments A; Figure 3 E), and an average of 4.9 reads per SNP compared to an average of 1.9 reads per SNP when following the CNERs protocol (experiments B; Fig- ure 3 E). The average coverage was not significantly different by MWW test due to one outlier sample (UAM:ES:27502), which was the sample for which we had to reduce library volume going into CNERs captures and has low SNP coverage. We evaluated target coverage uniformity using fold-80 base penalty, which estimates additional sequencing re- quired to bring 80% of the zero-coverage targets to mean coverage depth. The smaller the fold-80 base penalty, the more uniform the coverage is across all target regions ( 36 ). PAGE 7 OF 11 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 Figur e 3. SNP ca pture with CNERs and Arbor myBaits for ancient horse samples. ( A ) Endo genous content measur ed as proportion of r eads mapping to horse r efer ence genome for ten ancient horse samples befor e captur e enrichment (gr ey bars), proportion of mapped r eads after captur e with Arbor myBaits (cyan), and proportion of mapped reads after capture with CNERs (yellow). SNP enrichment efficiency measured as proportion of total reads ( B ) and mapped reads ( C ) covering the target SNPs for CNERs and Arbor myBaits. ( D ) Number of target SNPs covered by at least one read. ( E ) Mean coverage of target SNPs at one million raw read pairs. Mann–Whitney Wilco x on test P values are indicated as ns (5.00e-02 < P ≤ 1.00e + 00), * (1.00e-02 < P ≤ 5.00e-02), (1.00e-03 < P ≤ 1.00e-02) and * (1.00e-04 < P ≤ 1.00e-03). The average fold-80 base penalty is 3.7 for CNERs and 5.3 for Arbor myBaits, suggesting that CNERs produces more uniform coverage across all target SNPs. We explored whether probe length or GC content ex- plained coverage unevenness among the ancient horses. As observed in the modern horse enrichments, enrichment of ancient horses resulted significantly higher SNP coverage for CNERs targeting 80bp regions compared to 50bp or 100bp (Supplementary Figure S7). The statistical degree of significance of these comparisons as estimated from MWW test p-values (Supplementary Figure S7) differed among the ancient horses due to differences in percent mapped reads. Enrichments using CNERs resulted in higher normalized coverage for SNPs in target regions that had 42–66% (mode ∼55%) GC content compared to SNP targets in other GC contents and to Arbor myBaits capture data in this GC bin (Supplementary Figure S8). Arbor myBaits resulted in higher SNP normalized coverage for target regions with 30–45% GC content (mode ∼37% GC) compared to other GC contents and to CNERs capture data in this GC bin. W hile this indica tes a shift towards lower GC pr efer ence for Arbor myBaits and higher GC pr efer ence for CNERs, the e69 Nucleic Acids Research, 2023, Vol. 51, No. 13 PAGE 8 OF 11 difference in coverage across GC bins is not statistically dif- ferent by KS test (Supplementary Figure S8). We next compared CNERs captures and Arbor myBaits captures in the mean normalized coverage at 100 bp up- str eam and downstr eam r egions of target SNPs to assess whether coverage around the SNP target region influenced cov erage une v enness. We designed only one CNER per tar- get SNP, centered in the target region, resulting in maxi- mum coverage depth for SNPs and reduced coverage for the surrounding region (Supplementary Figure S9). Arbor my- Baits designed up to three baits per target SNP, tiled 20 bp (cid:4) end, which resulted in an expected maximum cover- from 5 age for ∼20 bp region to the right of the target SNP (Sup- plementary Figure S9). These differences in coverage profile between CNERs and Arbor myBaits are significant by KS test. Post-capture purification steps did not affect the coverage around SNPs; both experiments A (Arbor myBaits proto- col) and experiments B (CNERs protocol) resulted in sim- ilar coverage profiles when comparing enrichments using same probes (Supplementary Figure S9). CNERs and Arbor myBaits produce similar genotypes We calculated genotype likelihoods for target SNPs using the capture data. We did not include sample UAM:ES:27502 because it had few genotyped sites. Av- erage concordance of genotypes of nine ancient horses between experiment A (Arbor myBaits protocol) and B (CNERs protocol) is 97.9% for Arbor myBaits data and 98.1% for CNERs data (Supplementary Figure S10, Sup- plementary Table S5). To increase the read depth for in- dividual SNPs, we merged bam files from the two exper- iments and called genotypes on the merged data. With merged data, both CNERs and Arbor myBaits geno- typed between 4394 and 13 330 sites with 96.7–99.5% concordance for individual horses (Figure 4 A). On aver- age, genotypes called on Arbor myBaits and CNERs data concur 98.6%. CNERs and Arbor myBaits captured reads with differ- ent base substitution patterns in the target SNPs (Figure 4 B). Of the total 18 994 genotyped sites among the nine an- cient horses, 13 893 sites wer e captur ed using both probes, 1334 sites were only captured by Arbor myBaits and 3767 sites were onl y ca ptured by CNERs data. CNERs capture more GC transversions compared to Arbor myBaits (Fig- ure 4 B) because they more efficiently capture higher GC re- gions (Supplementary Figure S8). While CNERs and Arbor myBaits capture reads with comparable patterns of cyto- sine deamination at the ends of reads (Supplementary Fig- ure S11), Arbor myBaits captured more SNPs with tran- sition substitutions (11.5% versus 4.5% for CNERs versus 0.4% shared in both probes, Figure 4 B). This pattern may arise because the right shifted tiling design pr efer entially en- riches for SNPs at the ends of aDNA molecules (Supple- mentary Figure S12) where transition substitutions occur due to cytosine deamina tion. Alterna ti v ely, CNERs enrich for aDNA fragments with SNPs at the center of the read (Supplementary Figure S12), which may lead to higher cov- erage at SNP sites compared to Arbor myBaits (Supplemen- tary Figure S9). We used the enriched genotypes to explore the evolu- tionary relationships between the nine ancient horses for which we generated data. Admixture analysis identified two main ancestry components, both for da ta genera ted using CNERs (Figure 4 C) and Arbor myBaits captures (Supple- mentary Figure S13). Principal component (PC) analysis of genotype likelihood covariance also segregated ancient horses into two major clusters (Figure 4 D), with similar pat- terns observed when using CNERs or Arbor myBaits data. The first principal component (PC1) roughly corresponds to ancestry as in Figure 4 C, and PC2 reflects geo gra phic origin either in Chukotka, Russia (Western Beringia) or Alaska, USA (Eastern Beringia). This pattern is consistent among probe types and with horse population structure pr eviously inferr ed from whole-genome and mitochondrial data ( 15 ). DISCUSSION Targeted sequencing can provide a cost-effecti v e method for data generation f or man y comparati v e genomics ap- plications, in particular when the samples of interest con- tain only trace amounts of degraded DNA. Howe v er, the high cost of producing hybridization baits hinders the widespread adoption of this approach. Our approach, which we call Circular Nucleic acid Enrichment Reagent method, reduces both the cost and time r equir ed for gener- ation of microgram quantities of probes. Incorporation of poly-dT overhangs at both ends in the CNER template de- sign overcomes end synthesis errors in long oligonucleotide baits. The length of the poly-dT limits the circularization of templates by splint ligation using the pol y-dA oligo. Pol y- dA mediated splint ligation ensures that only templates with a certain length of poly-dT are amplified by RCA, thus eliminating incompletely synthesized baits. These template design features and isothermal amplification using RCA overcome many of the artifacts induced by PCR amplifi- ca tion of templa te oligo pools like non-specific amplifica- tion and generation of heterogenous products (Twist Bio- science’s technical note). Further, standard PCR amplifica- tion r equir es inclusion of specific primer binding sequences at the ends that increase oligo length ( 9 ) and may interfere with hybridization captur e. Futur e comparison of the CN- ERs methods with other PCR-based oligonucleotide am- plification methods would be useful to explore the role of amplification biases in hybridization efficiency. We optimized the hybridization conditions for the CN- ERs which differed from conventional hybridization condi- tions used for RNA baits. Enrichments using CNERs re- duces the hybridization time to overnight incubation (18 - 20 hours) instead of the 48–72 h r equir ed in conventional capture methods for degraded DNA ( 32 , 34 ). This increase in efficiency may be useful in clinical diagnostics. Further, conventional baits are designed with multifold tiling baits per target ( 33 , 34 ) to achie v e uniform coverage across dif- fer ent GC r egions, but still underperf orm f or target re- gions with > 50% GC content ( 33 , 35 ). We designed only one CNER tiling per SNP target region to save both CN- ERs production cost and sequencing cost. CNERs capture results in higher coverage for target regions with 45 - 75% GC content than regions with other GC contents, similar PAGE 9 OF 11 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 Figure 4. Genotyping and estimated evolutionary relationships between the ancient horse samples. ( A ) Genotype concordance between SNP capture data generated using CNERs and Arbor myBaits. Numbers above the bars indicate the sites genotyped by both methods in a gi v en horse sample. ( B ) Percentage of substitution types shared between (green) and unique to CNERs (yellow) and Arbor myBaits (cyan). ( C ) Admixture analysis with K = 2 separated the ancient horses into two lineages regardless of their geo gra phic location. ( D ) Principal component analysis of genotype likelihood covariance matrix of 23771 nuclear SNP sites in nine ancient horses. Transitions are filtered out for population analyses due to cytosine deamination in aDNA. PC1 segregated horses into two major clades and PC2 separated horses into the Western (Chukotka) and Eastern (Alaska) Beringian populations. to other DNA baits ( 35 ), whereas Arbor myBaits produced higher coverage for regions with 30–45% GC, similar to other RNA baits ( 34 , 35 ). Difference in the AT / GC bond- ing strength might differently influence the melting tem- perature of DN A-RN A heteroduplex and double stranded DN A molecules, w hich could lead to the observed cover- age differences between the DN A and RN A baits for target r egions with differ ent GC content. It would be inter esting to test whether multi-tiling CNERs for target regions with lower GC content brings their coverage closer to the sample mean coverage. Multi-tiling and probe length also increase the coverage for regions around the targeted region ( 32 , 33 ). This might be desired for some applications like exome cap- ture, but it will reduce the cost-effecti v eness of genotyping- by-sequencing (GBS). CNERs achie v e highest cov erage at the target SNP sites compared to adjacent regions which is desired for GBS applications. To demonstrate the utility of the CNERs approach for GBS, we genotyped ∼23k nuclear SNPs in ten ancient horses using both DNA based CNERs and a commercially available RNA baits from Arbor myBaits. We found that SNP enrichment efficiency using CNERs was consistent across most of our ancient samples, despite their variabil- ity in pre-enrichment precent mapped reads (endogenous content). Further, CNERs pro vided tw o-fold higher SNP enrichment efficiency compared to Arbor myBaits. CNERs r equir ed only one probe per target SNP and enriched a greater number of targeted sites with maximal read depth at the target SNP site. Two-fold higher enrichment efficiency could be due to enrichment of both strands of target regions e69 Nucleic Acids Research, 2023, Vol. 51, No. 13 PAGE 10 OF 11 by the CNERs probes compared to one targeted strand by RNA baits from Arbor. This could be tested using double stranded RNA baits ( 35 ). Both admixture and PC analy- sis of genotype likelihoods grouped the ancient horses into two major clusters (Figure 4 ), like the results based on whole genomes ( 15 ). Future work using the horse SNP panel with a more geo gra phicall y and temporally e xtensi v e sampling of ancient horses will provide new insights into the history of movement and gene flow among Late Pleistocene horses. Although we focused on generating data from individual horse bones, CNERs can also be used for targeted DNA capture and sequencing from other sample types that are difficult to genotype by conventional methods ( 37 ). Cell- free and circulating tumor DN A (cf / ctDN A) isolated from liquid biopsies, for example, can be used to identify muta- tion burden in cancer patients, disease carrier status, and for noninvasi v e prenatal testing ( 38 ). DNA isolated from envi- ronmental samples like water and air and from ancient sed- iments can be used to reconstruct present and past environ- ments noninvasi v el y ( 39 ). DN A isolated fr om single r ootless hair can be used to solve forensic cases ( 40 ). All these sam- ple types are preserved as highly fragmented DNA, how- e v er, and often in complex mixtur es, wher e targeted capture using CNERs provides a straightforward approach to gen- erating useful comparati v e data ( 41 ). The CNER method can be extended to generate whole genome enrichment (WGE) probes. Genome fragments of a r efer ence or r elated species can be circularized by bridge adapters to included restriction enzyme sites, amplified, and digested as in oligo templates to make WGE-CNERs. These would be a DNA alternati v e for the whole-genome in- solution capture (WISC) method’s RNA baits ( 32 ). WGE is valuab le when e xploring an unkno wn or ganism or enrich- ing a taxon in mixtures, as well as when analyzing aDNA samples with low endogenous content. WGE can also be used to generate low-coverage genomes of a few individuals for SNP ascertainment, from which a target SNP panel for population studies can be designed. We expect the CNER method may be adopted by future studies for various GBS and WGE applications. DA T A A V AILABILITY All raw sequencing data generated for this project are sub- mitted to the SRA database under BioProject accession number PRJNA785663. SUPPLEMENT ARY DA T A Supplementary Data are available at NAR Online. ACKNOWLEDGEMENTS The authors thank the members of the UCSC Pale- o genomics lab, especiall y Dr. Rachel Meyer and Remy Nguyen for their critical comments and proofreading the manuscript. We are grateful to Pamela Groves, Daniel Mann and the Uni v ersity of Alaska Museum-Earth Science collections for providing us access to the Alaskan ancient horse specimens analyzed in this study. We are also grateful to Love Dalen, Sergey Vartanyan, Eleftheria Palkopoulou and Alexei Tikhonov for providing the Eurasian ancient horse samples analyzed in this study. We thank Neda De- Ma y o, Celeste Carlisle and ‘Return to Freedom’ foundation for donating modern horse blood specimens. FUNDING Office of Research, Industry Alliances & Technology Com- mercialization of the UC Santa Cruz [to R.E.G., B.S.]; National Institute of Justice [Graduate Research Fellow- ship grant number 2020-R2-CX-0029 to B. Sundar ar aman., 2020-DQ-BX-0014 to R.E.G.]; UCSC Chancellor’s Disser- tation Year Fellowship [to A.O.V.]; National Science Foun- dation [ARC-1417036 to B.Shapiro]; Institute of Museum and Library Services [MG-30-17-0045-17 to B. Shapiro]. Funding for open access charge: National Institute of Jus- tice [2020-DQ-BX-0014]. Conflict of interest statement. B. Sundar ar aman and R.E.G are listed as co-inventors in a PCT application filed by the UCSC describing some of the methods presented here. B. Sundar ar aman is founder and shareholder of GenZ Ge- nomics Private Limited in Chennai, India. REFERENCES 1. Mamanova,L., Coffey,A.J., Scott,C.E., Kozarewa,I., Turner,E.H., Kumar,A., Howard,E., Shendure,J. and Turner,D.J. (2010) Target-enrichment strategies for next-generation sequencing. Nat. Methods , 7 , 111–118. 2. Gasc,C., Peyretaillade,E. and Peyret,P. (2016) Sequence capture by hybridization to explore modern and ancient genomic diversity in model and nonmodel organisms. Nucleic Acids Res. , 44 , 4504–4518. 3. Gaudin,M. and Desnues,C. (2018) Hybrid capture-based next generation sequencing and its application to Human infectious diseases. Front. Microbiol. , 9 , 2924. 4. Hodges,E., Xuan,Z., Balija,V., Kramer,M., Molla,M.N., Smith,S.W., Middle,C.M., Rodesch,M.J., Albert,T.J., Hannon,G.J. et al. (2007) Genome-wide in situ exon capture for selective resequencing. Nat. Genet. , 39 , 1522–1527. 5. Albert,T.J., Molla,M.N., Muzny,D.M., Nazareth,L., Wheeler,D., Song,X., Richmond,T.A., Middle,C.M., Rodesch,M.J., Packard,C.J. et al. (2007) Direct selection of human genomic loci by microarray hybridization. Nat. Methods , 4 , 903–905. 6. Okou,D.T., Steinberg,K.M., Middle,C., Cutler,D.J., Albert,T.J. and Zwick,M.E. (2007) Microarray-based genomic selection for high-throughput resequencing. Nat. Methods , 4 , 907–909. 7. Hodges,E., Rooks,M., Xuan,Z., Bhattacharjee,A., Benjamin Gordon,D., Brizuela,L., Richard McCombie,W. and Hannon,G.J. (2009) Hybrid selection of discrete genomic intervals on custom-designed microarrays for massi v ely parallel sequencing. Nat. Protoc. , 4 , 960–974. 8. Burbano,H.A., Hodges,E., Green,R.E., Briggs,A.W., Krause,J., Meyer,M., Good,J.M., Maricic,T., Johnson,P.L.F., Xuan,Z. et al. (2010) Targeted investigation of the neandertal genome by array-based sequence capture. Science , 328 , 723–725. 9. Gnirk e,A., Melniko v,A., Maguire,J., Rogo v,P., LeProust,E.M., Brockman,W., Fennell,T., Giannoukos,G., Fisher,S., Russ,C. et al. (2009) Solution hybrid selection with ultra-long oligonucleotides for massi v ely parallel targeted sequencing. Nat. Biotechnol. , 27 , 182–189. 10. Maricic,T., Whitten,M. and P ¨a ¨abo,S. (2010) Multiplexed DNA sequence capture of mitochondrial genomes using PCR products. PLoS One , 5 , e14004. 11. Kosuri,S. and Church,G.M. (2014) Large-scale de novo DNA synthesis: technologies and applications. Nat. Methods , 11 , 499–507. 12. Song,L.-F., Deng,Z.-H., Gong,Z.-Y., Li,L.-L. and Li,B.-Z. (2021) Large-scale de novo oligonucleotide synthesis for whole-genome synthesis and data storage: challenges and opportunities. Front. Bioeng. Biotechnol. , 9 , 689797. PAGE 11 OF 11 Nucleic Acids Research, 2023, Vol. 51, No. 13 e69 13. Duftner,N., Lar kins-For d,J., Legendre,M. and Hofmann,H.A. (2008) Efficacy of RNA amplification is dependent on sequence characteristics: implications for gene expression profiling using a cDNA microarray. Genomics , 91 , 108–117. 14. Conrad,T., Plumbom,I., Alcobendas,M., Vidal,R. and Sauer,S. (2020) Maximizing transcription of nucleic acids with efficient T7 promoters. Commun. Biol. , 3 , 439. 15. Vershinina,A.O ., Heintzman,P.D ., Froese,D .G ., Zazula,G ., Cassatt-Johnstone,M., Dal ´en,L., Der Sarkissian,C., Dunn,S.G., Ermini,L., Gamba,C. et al. (2021) Ancient horse genomes re v eal the timing and extent of dispersals across the Bering Land Bridge. Mol. Ecol. , 30 , 6144–6161. 16. Dabney,J., Knapp,M., Glocke,I., Gansauge,M.-T., Weihmann,A., Nickel,B., Valdiosera,C., Garc ´ıa,N., P ¨a ¨abo,S., Arsuaga,J.-L. et al. (2013) Complete mitochondrial genome sequence of a middle pleistocene cave bear reconstructed from ultrashort DNA fragments. Proc. Natl. Acad. Sci. USA. , 110 , 15758–15763. 17. Fulton,T.L. and Shapiro,B. (2019) Setting up an ancient DNA laboratory. Methods Mol. Biol. , 1963 , 1–13. et al. (2011) The variant call format and vcftools. Bioinformatics , 27 , 2156–2158. 28. Li,H. and Durbin,R. (2009) Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics , 25 , 1754–1760. 29. Korneliussen,T.S., Albrechtsen,A. and Nielsen,R. (2014) ANGSD: analysis of next generation sequencing data. BMC Bioinf. , 15 , 356. 30. Meisner,J. and Albrechtsen,A. (2018) Inferring population structure and admixture proportions in low-depth NGS data. Genetics , 210 , 719–731. 31. Lee,H., O’Connor,B.D., Merriman,B., Funari,V.A., Homer,N., Chen,Z., Cohn,D.H. and Nelson,S.F. (2009) Improving the efficiency of genomic loci capture using oligonucleotide arrays for high throughput resequencing. BMC Genomics (Electronic Resource) , 10 , 646. 32. Carpenter,M.L., Buenr ostr o,J.D., Valdiosera,C., Schr oeder,H., Allentoft,M.E., Sikora,M., Rasmussen,M., Gravel,S., Guill ´en,S., Nekhrizov,G. et al. (2013) Pulling out the 1%: whole-genome capture for the targeted enrichment of ancient DNA sequencing libraries. Am. J. Hum. Genet. , 93 , 852–864. 18. Kapp,J.D., Green,R.E. and Shapiro,B. (2021) A fast and efficient 33. Samorodnitsky,E., Datta,J., Jewell,B.M., Hagopian,R., Miya,J., single-stranded genomic library preparation method optimized for ancient DNA. J. Hered. , 112 , 241–249. 19. Kircher,M., Sawyer,S. and Meyer,M. (2012) Double indexing overcomes inaccuracies in multiplex sequencing on the Illumina pla tform. Nuc leic Acids Res . , 40 , e3. 20. Rohland,N. and Reich,D. (2012) Cost-effecti v e, high-throughput DNA sequencing libraries for multiplexed target capture. Genome Res. , 22 , 939–946. 21. Librado,P., Der Sarkissian,C., Ermini,L., Schubert,M., J ´onsson,H., Albrechtsen,A., Fumagalli,M., Yang,M.A., Gamba,C., Seguin-Orlando,A. et al. (2015) Tracking the origins of Yakutian horses and the genetic basis for their fast adaptation to subarctic environments. Proc. Natl. Acad. Sci. U.S.A. , 112 , E6889–E6897. 22. Schubert,M., J ´onsson,H., Chang,D., Der Sarkissian,C., Ermini,L., Ginolhac,A., Albrechtsen,A., Dupanloup,I., Foucal,A., Petersen,B. et al. (2014) Prehistoric genomes re v eal the genetic foundation and cost of horse domestication. Proc. Natl. Acad. Sci. U.S.A. , 111 , E5661–E5669. 23. Wade,C.M., Giulotto,E., Sigurdsson,S., Zoli,M., Gnerre,S., Imsland,F., Lear,T.L., Adelson,D.L., Bailey,E., Bellone,R.R. et al. (2009) Genome sequence, comparati v e analysis, and population genetics of the domestic horse. Science , 326 , 865–867. 24. Li,H. (2011) A statistical frame wor k for SNP calling, mutation discovery, association mapping and population genetical parameter estimation from sequencing data. Bioinformatics , 27 , 2987–2993. 25. Zhou,B., Wen,S., Wang,L., Jin,L., Li,H. and Zhang,H. (2017) AntCaller: an accurate variant caller incorporating ancient DNA damage. Mol. Genet. Genomics , 292 , 1419–1430. 26. Van der Auwera,G.A. and O’Connor,B.D. (2020) In: Genomics in the Cloud: Using Doc k er, GATK, and WDL in Terra (1st Edition) . O’Reilly Media. 27. Danecek,P., Auton,A., Abecasis,G., Albers,C.A., Banks,E., DePristo,M.A., Handsaker,R.E., Lunter,G., Marth,G.T., Sherry,S.T. Wing,M.R., Damodaran,S., Lippus,J.M., Reeser,J.W., Bhatt,D. et al. (2015) Comparison of custom capture for targeted next-generation DNA sequencing. J. Mol. Diagn. , 17 , 64–75. 34. Cruz-D ´avalos,D.I., Llamas,B., Gaunitz,C., Fages,A., Gamba,C., Soubrier,J., Librado,P., Seguin-Orlando,A., Pruvost,M., Alfarhan,A.H. et al. (2017) Experimental conditions improving in-solution target enrichment for ancient DNA. Mol. Ecol. Resour. , 17 , 508–522. 35. Zhou,J., Zhang,M., Li,X., Wang,Z., Pan,D. and Shi,Y. (2021) Performance comparison of four types of target enrichment baits for exome DNA sequencing. Hereditas , 158 , 10. 36. So,A.P., V ilbor g,A., Bouhlal,Y., Koehler,R.T., Grimes,S.M., Pouliot,Y., Mendoza,D., Ziegle,J., Stein,J., Goodsaid,F. et al. (2018) A robust targeted sequencing approach for low input and variable quality DNA from clinical samples. NPJ Genom. Med. , 3 , 2. 37. Diaz,L.A. and Bardelli,A. (2014) Liquid biopsies: genotyping circulating tumor DNA. J. Clin. Oncol. , 32 , 579–586. 38. Szil ´agyi,M., P ¨os,O., M ´arton, ´E., Bugly ´o,G., Solt ´esz,B., Keser ˝u,J., Penyige,A., Szemes,T. and Nagy,B. (2020) Circulating cell-free nucleic acids: main characteristics and clinical application. Int. J. Mol. Sci. , 21 , 6827. 39. Murchie,T.J., Kuch,M., Duggan,A.T., Ledger,M.L., Roche,K., Klunk,J., Karpinski,E., Hackenberger,D., Sadoway,T., MacPhee,R. et al. (2021) Optimizing extraction and targeted capture of ancient environmental DNA for reconstructing past environments using the P alaeoChip Ar ctic-1.0 bait-set. Quat. Res. , 99 , 305–328. 40. Brandhagen,M.D ., Loreille,O . and Irwin,J.A. (2018) Fragmented nuclear DNA is the predominant genetic material in human hair shafts. Genes (Basel) , 9 , 640. 41. Marchini,J. and Howie,B. (2010) Genotype imputation for genome-wide association studies. Nat. Rev. Genet. , 11 , 499–511. C (cid:2) The Author(s) 2023. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creati v e Commons Attribution License (http: // creati v ecommons.org / licenses / by / 4.0 / ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
10.1371_journal.pcbi.1011899	RESEARCH ARTICLE Tipping points emerge from weak mutualism in metacommunities Jonas DenkID 1,2, Oskar HallatschekID 1,2,3* 1 Department of Physics, University of California, Berkeley, California, United States of America, 2 Department of Integrative Biology, University of California, Berkeley, California, United States of America, 3 Peter Debye Institute for Soft Matter Physics, Leipzig University, Leipzig, Germany * [email protected] Abstract The coexistence of obligate mutualists is often precariously close to tipping points where small environmental changes can drive catastrophic shifts in species composition. For example, microbial ecosystems can collapse by the decline of a strain that provides an essential resource on which other strains cross-feed. Here, we show that tipping points, eco- system collapse, bistability and hysteresis arise even with very weak (non-obligate) mutual- ism provided the population is spatially structured. Based on numeric solutions of a metacommunity model and mean-field analyses, we demonstrate that weak mutualism low- ers the minimal dispersal rate necessary to avoid stochastic extinction, while species need to overcome a mean threshold density to survive in this low dispersal rate regime. Our results allow us to make numerous predictions for mutualistic metacommunities regarding tipping points, hysteresis effects, and recovery from external perturbations, and let us draw general conclusions for ecosystems even with random, not necessarily mutualistic, interac- tions and systems with density-dependent dispersal rather than direct mutualistic interactions. Author summary In ecosystems with obligate mutualism, species rely on each other’s cooperation to thrive. Obligate mutualism has been of special interest in theoretical ecology because it generates tipping points between drastically different ecological states (along with bistability). Weak, non-obligate mutualistic interactions have attracted much less interest due to their minimal impact on population growth behavior in well-mixed scenarios. However, in spa- tially structured metacommunities with migration coupling and demographic stochasti- city, we find that weak mutualism can fundamentally alter population growth behavior, leading to bistability and abrupt shifts in population size. We identify a broad range of dis- persal rates in which populations show bistability and go either extinct or reach an equi- librium, depending on their initial population size. Crossing the limiting dispersal rates of this bistable regime, the system undergoes abrupt catastrophic shifts in population size, which would be overlooked under well-mixed assumptions. Our findings have broad a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Denk J, Hallatschek O (2024) Tipping points emerge from weak mutualism in metacommunities. PLoS Comput Biol 20(3): e1011899. https://doi.org/10.1371/journal. pcbi.1011899 Editor: Jacopo Grilli, Abdus Salam International Centre for Theoretical Physics, ITALY Received: October 10, 2023 Accepted: February 6, 2024 Published: March 5, 2024 Copyright: © 2024 Denk, Hallatschek. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the manuscript and its Supporting information files. The code is made available on GitHub (https://github.com/Hallatscheklab/Self- Consistent-Metapopulations). Funding: O.H. acknowledges support by a Humboldt Professorship of the Alexander von Humboldt Foundation and by a National Science Foundation CAREER Award (1555330). J.D. acknowledges support from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through grant 445916943. The PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 1 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Competing interests: The authors have declared that no competing interests exist. biogeographic implications for predicting tipping points, hysteresis effects, and recovery from perturbations in mutualistic metacommunities. Introduction Mutualistic interactions between species are ubiquitous in nature and can be critical for the stability of natural ecosystems as exemplified by cross-feeding microbes in the gut [1–3] and cooperative breakdown of sugar [4]. When one or more species of a community entirely rely on each other for survival, their populations must surpass a certain critical size to prevent extinction. This threshold property is often referred to as strong Allee effect at the community level (a weak Allee effect refers to scenarios where at low population size, the per capita growth rate decreases with decreasing population size, but never becomes negative). Models that incorporate a strong Allee effect are of great interest in ecology and are invoked frequently to explain tipping points accompanied by bistability and catastrophic shifts between survival and extinction in ecosystems [5–7]. As an instructive example, when microbes depend on each other in order to access vital resources, their population dynamics can exhibit a tipping point at which the community undergoes catastrophic shifts upon the variation of experimental parameters such as nutrient levels [4, 8–10]. Although any community contains several populations, its dynamics can often be under- stood in terms of the one-dimensional dynamics of a single effective population (e.g. see [10]), illustrated in Fig 1A. Strong Allee effects induced by obligate mutualism between species there- fore generate a very similar dynamics and threshold phenomena as intra-specific strong Allee effects, which can be observed in natural populations of all length scales, including zooplank- ton [11], plants [12], and polar bears [13]. However, strong Allee effects and the associated tipping points and bistability only occur in well-mixed communities if the mutual interactions are vital (i.e. obligate) for the survival of the community. In fact, well-mixed ecosystems with weakly mutualistic interactions (faculta- tive mutualism) behave very similar to entirely non-mutualistic ecosystems, except that the carrying capacity of their logistic growth is enhanced due to the mutualistic interactions between different species (see Fig 1B). Fig 1. The (well-mixed) dynamics of a population with strong Allee effect vs logistic growth. A. If a population exhibits a strong Allee effect, its growth rate @tN is negative at low population sizes. The dynamics, therefore, heads toward extinction (N = 0) unless the initial population size exceeds a threshold, the Allee threshold A, upon which the population size rises to a limiting population size (N = K), referred to as carrying capacity. Full and open circles denote stable and unstable fixed points, respectively; arrows denote the flow of the dynamics. B. In contrast, a single-species population that undergoes regular logistic growth displays only one stable fixed point at the carrying capacity, while the extinct state is unstable. Additional mutualistic interactions, here tuned by a parameter α, can increase the effective carrying capacity K(α) without qualitatively changing the fixed point structure. https://doi.org/10.1371/journal.pcbi.1011899.g001 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 2 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities It is known that the impact of strong Allee effects can be qualitatively different to that of logistic growth in spatially structured populations. For example, metapopulations consisting of patches with local Allee effects show pushed instead of pulled waves in range expansions [14– 16], localized wave fronts [17], and pronounced patchiness [18]. However, it is currently unclear whether spatial structure can significantly alter ecosystem dynamics when mutualistic interactions are weak. Here, we show that when demographic fluctuations are taken into account, even the weak- est form of mutualism between species in a metacommunity can lead to a tipping point accom- panied by bistability and catastrophic shifts between coexistence and extinction. Our combined analytical and numerical methods reveal a strong Allee effect in the metacommunity despite the absence of a strong Allee effect in the single patch dynamics. This metacommunity- wide strong Allee effect leads to range of intermediate dispersal rates where species can avoid extinction when the mean population size overcomes a threshold value. Informed by our intui- tion regarding purely mutualistic interactions, we show that close to the shift from extinction to finite population sizes, metacommunities with random interactions undergo selection for mutualistic interactions. We further apply our analyses to metacommunities in which interac- tions between species increase their dispersal instead of their growth rate and find a similar emergent tipping point. Our results give insights into the role of demographic stochasticity and dispersal in metacommunities and highlight the emergence of tipping points and cata- strophic shifts even when absent under well-mixed conditions. Mathematical approach In the following, we consider S species that live in a metacommunity of P coupled communi- ties (patches), where P is assumed to be large. The dynamics of the population size Nx,i of spe- cies i 2 {1, . . .S} on patch x 2 {1, . . ., P} is modeled by the following set of generalized Lotka- Volterra equations: @tNx;iðtÞ ¼ rNx;i 1 (cid:0) Nx;i K þlð �N i (cid:0) Nx;iÞ þ þ a K p ! XS Nx;j j;j6¼i ffiffiffiffiffiffiffi Nx;i Zx;i : ð1Þ The first term in Eq (1) describes growth of a species’ population at a growth rate r > 0, which saturates at a carrying capacity K due to self-limiting interactions. The interaction parameter α > 0 denotes the strength of mutualistic interactions between species. Assuming a constant interaction strength α between all species allows an analytic mean-field description; the results of this analysis will yield important intuitions when we later allow variations in the species’ inter-species interactions. The second term in Eq (1) takes into account dispersal, where we assumed, as a simple spatial approach, that all patches are connected through dispersal with a species-independent dispersal rate λ and �N i denotes the abundance of species i averaged over all P patches. The last term in Eq (1) reflects demographic fluctuations due to random births and deaths of individuals within a population, where ηx,i denotes uncorrelated noise with zero mean and unit variance. The square-root dependence of demographic noise on the density ensures that the expected population size variance is proportional to the expected number of birth or death events occurring in a given population [19, 20] (setting the amplitude of noise to unity amounts to setting the unit of time to be about one generation time). Considering only the deterministic patch dynamics without migration, i.e. the the first line of Eq (1), all spe- cies display the same population dynamics. When we further assume that mutualistic interac- tions are weaker than self-limiting interactions, i.e. α < (S − 1)−1, (to ensure that population PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 3 / 14 PLOS COMPUTATIONAL BIOLOGY Tipping points emerge from weak mutualism in metacommunities sizes do not diverge) all species have an unstable state at zero population size and a stable state with abundance N* given by N∗ ¼ K=½1 (cid:0) aðS (cid:0) 1Þ� : ð2Þ The fixed point structure of each species on an isolated patch thus resembles regular logistic growth with a carrying capacity N* that increases with α, as illustrated in Fig 1B. Specifically, this means that population growth on an isolated patch does not exhibit a strong Allee effect nor bistability. In the following, we will solve Eq (1) numerically and employ mean-field analy- ses to study the effect of demographic fluctuations and dispersal and show how these can, nev- ertheless, generate bistability and an abrupt shift in the population size. Results Weak mutualism generates a tipping point in a metacommunity For a clearer presentation of our results, in the following we fix r, K, and α (with α � (S − 1)−1), and vary the dispersal rate λ for different numbers of species S. First, we discuss our numerical solutions of Eq (1) assuming small average abundances hNi = (SP)−1 ∑x,i Nx,i as initial condition (for details on the numerical solution, see S1 Text, Sec. S1). The impact of growth, demographic fluctuations and dispersal on single species has been extensively studied for short-range dis- persal in the context of spreading processes and transport in random media within the theory of directed percolation [21, 22] and for global dispersal in metapopulations [23–26]. From these earlier studies we expect that for S = 1, increasing the dispersal rate leads to a continuous transi- tion from a phase of zero population size (absorbing phase) to a phase of non-zero population sizes (active phase). Indeed, when numerically solving the dynamics Eq (1) with global dispersal and for only one species (S = 1), we find that for zero and small dispersal rates λ the species eventually goes extinct, i.e. the population sizes on all patches are zero (without return). In con- trast, for λ above a critical threshold value λc, the average population size after the final time step of our numerical solution is positive and increases continuously with λ (see triangles in Fig 2A) while being bounded from above by the deterministic carrying capacity N* given by Eq (2). Interestingly, when increasing the number of species at constant mutualistic interaction strength α > 0, the average abundance as a function of the dispersal rate λ undergoes a sudden jump at λc from zero to positive values. Discontinuous transitions are a telltale sign of subcriti- cal bifurcations and bistability close to the transition [27]. To investigate the possibility of multi- ple stable solutions, we repeat our numerical solutions for larger initial average abundances, and, indeed, find bistability close to the threshold dispersal rate λc for larger S (see circles in Fig 2A). To get deeper insights into the exact form of the bifurcation, we employ a mean-field approximation of Eq (1), that has been recently presented to approximate the stationary abundance distribution for metacommunities with competitive interactions [23]. This mean-field analysis allow us to consider metacommunities with an infinite number of patches and thereby circumvent possible finite size effects of our numerical analysis. In short, by expressing the interaction term through the species-averaged abundance on a patch defined as ^N x ¼ S(cid:0) 1 field parameters, we can map the dynamics Eq (1) to the solvable problem of a Brownian par- ticle in a fixed potential. Due to the statistical equivalence of all populations for all species and patches, we can demand that �N i and ^N x are both equal in equilibrium (in the limit of an infinite number of species and patches). This eventually allows us to derive an analytic expression for the abundance distribution as a function of the mean species abundance iNx;i and treating the mean-fields ^N x and �N i as deterministic mean- P PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 4 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities Fig 2. Weak mutualism generates a tipping point. A. Starting at small and large initial population sizes (triangles and circles, respectively), the mean deme population size hNi in our numerical solutions settles at a positive value or decays to zero. The long term steady state values are in very good agreement with our mean-field solution (lines). Solid and dashed lines denote stable and unstable manifolds, respectively. Colors denote different numbers S of species as indicated. The analytical result for λc is shown as black dotted vertical line and the S–dependent tipping point dispersal rates λt are indicated by vertical dotted lines of corresponding color. The deterministic steady states N, given by Eq (2), are indicated by stars on the right next to the plot. B. Numerical and analytic solutions for the abundance distribution P[N] (circles and lines, respectively) for S = 100 with λ < λc (λ = 10−4, green open circles and solid line) and λ > λc (λ = 10−1, green full circles and dashed line) as well as for S = 1 with λ closely above λc (λ = 10−2.4, blue). C. Small changes in the species number, i.e. through perturbations, can lead to a collapse of the metacommunity (as indicated by the arrow), λ = 0.001. Parameters: r = 0.3, K = 10, α = 0.005, P = 500. https://doi.org/10.1371/journal.pcbi.1011899.g002 hNi � �N i ¼ ^N x and the control parameters r, K, α, S, and λ, which can be solved self-consis- tently (for a detailed derivation see [23] and S1 Text, Sec. S2). In agreement with our numerical solution, our analytic mean-field approximation predicts bistability between the dispersal rate λc (for an analytic expression see [23] and S1 Text, Sec. S2.1) and a lower dispersal rate λt. The rate λt marks a point where a small decrease of the dis- persal rate causes an abrupt shift from positive population size to extinction, often referred to as tipping point [7] (see vertical dotted lines in Fig 2A). In addition, our analytic solution reveals an unstable branch marking the threshold mean abundance as a function of the dis- persal rate the metacommunity must exceed to reach a non-zero mean abundance and avoid extinction (see dashed lines in Fig 2A). We also obtain the complete abundance distribution P½N� in analytical form, which has main contributions from extinct species (N = 0) for small λ and a maximum close to N (see Fig 2B, for details see S1 Text, Sec. S2). In summary, our numerical and analytical results strongly suggest that despite the lack of bistability in the deterministic and well-mixed population dynamics, the metacommunity dis- plays a tipping point accompanied by a regime of bistability. The identified bifurcation pre- dicts a discontinuous transition from extinct to colonized as the migration rate surpasses λc, and from colonized to extinct when the dispersal rate is lowered towards the tipping point λt. The population goes through a hysteresis loop if the migration parameter cycles between a value below λt and above λc. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 5 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities The emergence of bistability for low migration rates can be rationalized in two steps. First, note that when the species’ patch population sizes are sufficiently small (hNi�1), the probabil- ity for two or more species to encounter each other on the same patch becomes small. This decrease in encounters is significant because interactions between species, which are nonlinear, depend on these encounters. Thus, in situations where population density is low, these species- to-species interactions can be reasonably ignored. Consequently, the extinct state is stable if the species’ dispersal rates are below the critical threshold for individual species survival (λc), irrespective of the number of species involved. The question of bistability then boils down to the question of whether, for λ < λc, there is another stable state—besides the extinct state. Such a stable state is possible if the species interactions allow the population to survive. Intuitively, this is possible at high enough densities so that the mutualistic interactions, as reflected by the positive contribution to the growth term in Eq (1), compensate for the noise-driven decay. In our phase diagram this occurs if the densities lie above the dashed line in Fig 2A (see S1 Text, Sec. S2.2, and S2 Fig). We find that the range of bistability increases the more species interact through mutualism (see Fig 2C). As a consequence, this suggests that perturbations that decrease the number of species, even if only by a few species, may shift the metacommunity into a regime where even- tually all species go extinct (see arrow in Fig 2C). While we find that demographic fluctuations combined with spatial structure can result in bistability, previous studies suggest that demographic fluctuations can also produce the oppo- site effect on population transitions: a single species metapopulation, which displays bistability under deterministic and well-mixed conditions (imposed through an explicit strong Allee effect on every patch), may experience a gradual transition when dispersal and stochasticity are taken into account (see [28–30]). In S1 Text, Sec. S3, we investigate this prior observation by means of our mean-field approach applied to a metapopulation model of a single species that shows bistabilty on every patch. Consistently with the above mentioned studies, we find that when increasing the dispersal rate, the system undergoes a continuous transition from extinction to positive population sizes. While these results show that stochasticity can generally smoothen discontinuous transitions, our study on mutualistic metacommunities highlights the interesting and somewhat opposite role of stochasticity when it occurs in combination with species interactions: here, it can lead to discontinuous transitions even when they are absent in the deterministic dynamics. Spatial structure selects for metacommunities with an excess of mutualistic interactions The dramatic change from a smooth to a discontinuous transition close to the threshold dis- persal rate λc suggests several implications for more general species interactions. Since in con- trast to species with competitive interactions, species with mutualistic interactions are able to reach large finite abundances already well below λc, we hypothesize that in a metacommunity with random interactions, mutualism may play an important role in community assembly, at least close to λc. To test this idea, we generalize the metacommunity dynamics Eq (1) and assume random interactions αi,j between species i and j. The generalized metacommunity dynamics reads @tNx;iðtÞ ¼ riNx;i 1 (cid:0) Nx;i K þ þlð �N i (cid:0) Nx;iÞ þ PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 XS j;j6¼i p ai;j K ffiffiffiffiffiffiffi Nx;i ! Nx;j Z ; ð3Þ 6 / 14 PLOS COMPUTATIONAL BIOLOGY Tipping points emerge from weak mutualism in metacommunities where we again assumed global dispersal between patches. For a clearer distinction between mutualistic and competitive interactions between species, we choose interactions to be sym- metric, i.e. αj,i = αi,j. Motivated by earlier work on well-mixed communities [31–36], we draw ffiffiffi αi,j from a Gaussian distribution with mean ^a and standard deviation s= . Hence, it is possi- S ble to choose ^a and σ in a way that interactions between some species i and j are mutualistic (i.e. αi,j > 0) while interactions are on average competitive (^a < 0). A detailed analytical understanding of the spatially structured community assembly with random interactions is beyond the scope of our mean-field analysis, but can be obtained using the replica method [37]. In the following we use numerical solutions of Eq (3) to show that, below the critical threshold dispersal rate λc, communities survive that are enriched in mutualistic interactions, even though the interactions among the initial species pool are on average competitive. p First, we solve the dynamics Eq (3) numerically for fixed r, K, S, σ, negative mean interac- tion ^a (i.e. an on average competitive interaction between species), and varying λ. Here, we will focus on relatively small interaction differences, specifically s≲1, where previous studies suggest that, under well-mixed conditions, a community approaches a unique equilibrium state [33–35]. For larger σ, we expect the system to be multistable due to interspecies interac- tions alone (see also S1 Text, Sec. S4)–an effect that may obscure the emergence of bistability due to demographic noise and dispersal, which is our main focus here. Similar to our results for purely mutualistic interactions (Fig 2A), we find positive population sizes already for dis- persal rates below λc (see Fig 3A). Furthermore, we observe bistability, i.e. for dispersal rates below λc the metacommunity approaches positive mean population sizes when the initial pop- ulation size is sufficiently large while it goes extinct otherwise. We comment that in contrast to our model of equal mutualistic interactions, Eq (1), where all species either go extinct or sur- vive together, differences in the species’ interactions generally lead to extinctions of some spe- cies in our numerical solution while the rest of the species persist. In the regime where species survive, the final mean population size (see different colors in Fig 3A) and the number of spe- cies that go extinct (see S1 Text, Sec. S4) depend on the chosen set of random interactions and generally increase with the dispersal rate. While we focused on purely symmetric interactions, based on [37] we expect these results to hold even in the presence of moderate asymmetric contributions in the interaction coefficients. Fig 3. Tipping point for metacommunities with random interactions. A. Numerical solutions of Eq (3) for initially low (triangles) and high (circles) mean population sizes for three different sets of random interactions (denoted by three different colors) suggest bistability and hysteresis between the tipping point (gray dotted vertical line) and close to λc (black dotted vertical line). Gray solid line shows mean-field solution for identical interactions (ai;j ¼ ^a) B. Distributions of mean interactions with other species for all three sets of random interactions shown in A for λ = 10−2.8, λ = 10−2.2, λ = 10−1.5 from dark to light, respectively (λc � 10−2.6). Remaining Parameters: s ¼ 0:5; r ¼ 0:3; S ¼ 100; P ¼ 500; ^a ¼ (cid:0) 0:01 (on average competitive interactions). https://doi.org/10.1371/journal.pcbi.1011899.g003 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 7 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities In order to investigate the role of mutualistic interactions especially in the regime of dis- persal rates around λc, we calculate the average interaction Ii ≔ h(αi,j/K)Nx,jii, where the aver- age is taken over all patches x and co-surviving species j at the last time point of our numerical solution (see Fig 3B). In line with our intuition from purely mutualistic interactions, we find that for dispersal rates below λc, all species that survive form on average mutualistic interac- tions with their fellow surviving species. (for λ < λc, the distribution of I, P½I�, in Fig 3B has only contributions from positive I). When we increase the dispersal rate beyond λc, more species survive and, in particular, also species with on average competitive interactions coexist. We were also interested how the communities of remaining survivors of our numerical solution compare to our mean-field model with species-independent interaction coefficient. To this end, we calculate the mean- field solution with the number of species S and the species-independent interaction coefficient α being equal to the number of surviving species, Ssurv and their average interaction coefficient, ^asurv, respectively, that we obtained from our numeric simulation of the metacommunity with random interactions and different dispersal rates. Similar to Fig 2, we can now plot the steady states of the metacommunity based on the mean-field solution with the parameters S = Ssurv and a ¼ ^asurv as a function of the dispersal rate λ. As shown in S1 Text, Sec. S4, and S5 Fig, the resulting bifurcation again shows a tipping point at some l∗ t , marking the onset of a stable state with positive population size. Interestingly, the dispersal rate l∗ t of this tipping point is close to the dispersal rate of the underlying numerical solution. Heuristically, this suggests that in the assembly process of a community with random, predominantly competitive interactions, spe- cies will die out and adapt their population sizes until the remaining community reaches a state which is just viable for a given dispersal rate (i.e. so that the tipping point dispersal rate l∗ t is close to the dispersal rate λ). As a result, random metacommunities with predominantly competitive interactions self-organize to a state very close to a tipping point, where survival of the metacommunity may be very sensitive towards perturbations in its parameters, e.g. the dis- persal rate. Tipping point through density-dependent dispersal So far, we have incorporated mutualism through direct interactions between species, such that interactions effectively increase the species growth [see Eq (1) and Eq (3)]. In the following, we investigate interactions between species through their dispersal, i.e. interactions that increase the species’ dispersal rates. There is indeed evidence in many species [38–42] that emigration rates from crowded areas tend to be elevated to avoid competition for resources. Since this effectively results in a dis- persal rate that increases with the abundance of other populations, we refer to this scenario as interactions through density-dependent dispersal in the following. When assuming that, as a first approximation, emigration from a patch increases linearly with the number of individuals of other species already present, the dispersal term in Eq (1) can be written as " XP y l P 1 þ b XS j6¼i ! Ny;j Ny;i (cid:0) 1 þ b ! # Nx;j Nx;i ; XS j6¼i ð4Þ where we assumed a constant baseline dispersal rate λ between every patch and a linear increase of dispersal with population size with a factor β. Next, we solve Eq (1) for fixed r, K, and S > 1 with α = 0, i.e. without direct mutualistic interactions, and the dispersal term Eq (4) with β > 0 numerically and discuss the respective mean-field approximation (for details on the mean-field description, see S1 Text, Sec. S2.3). Similarly to direct mutualistic interactions PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 8 / 14 PLOS COMPUTATIONAL BIOLOGY Tipping points emerge from weak mutualism in metacommunities Fig 4. Density-dependent dispersal generates tipping point. A. The mean population size in our numerical solutions can reach different values when starting at small and large initial population sizes (triangles and circles, respectively) in agreement with our mean-field solution (solid and dashed lines denote stable and unstable manifolds, respectively). The black dotted vertical line marks λc, the colored dotted vertical lines lines indicate λt based on numerical solutions. The inset illustrates density-dependent dispersal, where emigration increases with the abundance of individuals from other species on a patch. B. Mean-field solution of the mean abundance predicts a catastrophic shift as function of the species number, λ = 0.001. Remaining parameters: r = 0.3, K = 10, β = 0.02, P = 500. https://doi.org/10.1371/journal.pcbi.1011899.g004 between species, we find that when varying the baseline dispersal rate λ, the average abundance of the metacommunity undergoes a sudden shift at λc (see Fig 4A). As for direct mutualistic interactions, the regime of dispersal rates λ that shows bistability grows for increasing S see Fig 4B). We thus conclude that interactions that increase species’ dispersal with the population sizes on a patch result in a similar phenomenology as interactions that increase the growth of the species’ populations on a patch, including catastrophic shifts and a metacommunity-wide strong Allee effect. Intuitively, bistability can emerge from density-dependent dispersal since for λ < λc, suffi- ciently large population sizes can elevate the realized dispersal rate above the threshold value λc (this is in agreement with our observations, compare S1 Text, Sec. S2.3, and S2 Fig]). We comment that the observed mutualism and tipping point that result from density- dependent dispersal persist even in the presence of additional direct competition between spe- cies, albeit the regime of bistability shrinks with increasing strength of direct competition (see S1 Text, Sec. S2.3, and S3 Fig). Discussion While it is well-established that strong (obligate) mutualism, which leads to strong local Allee effects, can create complex behaviors in spatially structured systems, our findings reveal that even the weakest form of mutualism among species on a patch can have profound impacts. These effects would be overlooked in a conventional, well-mixed deterministic framework. Specifically, weak mutualism can greatly influence the stability of a metacommunity, leading to a metacommunity-wide strong Allee effect, characterized by tipping points and hysteresis. As a result, abrupt transitions between species coexistence and complete metacommunity col- lapse can be induced by slight variations in factors like dispersal rate, the number of interacting species, or the strength of mutualistic interactions, for instance, due to environmental changes. Previous research has demonstrated that mutualistic interactions enhancing population growth across different patches in a metapopulation [29] and memory effects in a metacom- munity [43] can lead to Allee thresholds along with critical tipping points. Our study adds to this body of work by illustrating that even a basic form of mild mutualistic interactions can induce thresholds at the metacommunity level. It is important to note that the pronounced Allee effect observed in our metacommunity is not a predefined part of our model, unlike PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 9 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities previous metapopulation models that explicitly include Allee thresholds [44–46], but rather a property that emerges from our metacommunity model. The appearance of bistability in our stochastic model contrasts with the typical smoothing effect of demographic noise ([28, 30] and S1 Text, Sec. S3) and illuminates how demographic fluctuations can act as a catalyst for abrupt transitions when they interact with interspecies relationships. In communities with random interactions we found that limiting dispersal tends to favor the mutualistically interacting part of the community. Previous studies suggest, that under well-mixed conditions community assembly selects against competitive and enriches mutual (beneficial) interactions [47]. In our metacommunity model, survival poses a much stronger requirement: for dispersal rates below λc, only communities with on average mutualistic inter- actions can survive if they are large enough. Our results for random interactions suggest various interesting directions for future studies that depart from our assumption of small symmetric differences in the interaction coefficients. These include metacommunities with larger differences in the interaction coefficients, which have been suggested to yield multistability even in a well-mixed scenario [34], as well as meta- communities with asymmetric interaction statistics, which can lead to fluctuating steady state and even chaos in a deterministic metacommunity [48, 49]. The relevance of our main results for more complex models advocates further study of simplified models that can help to under- stand isolated features of complex ecological systems while offering a more feasible mathemati- cal approach and interpretation. We found qualitatively similar effects in a system where interactions between species act on the dispersal rate of species in such a way that emigration from a patch increases with the pop- ulation size of other species present. From a technical point of view, the similarity of both types of interaction comes from the fact that an increase in both growth rate and dispersal rate should have a positive effect on a species persistence. In a more ecological context, dispersal that increases with the densities of other species on a patch can arise from competition between species and can thus be interpreted as effectively competitive interaction. Following this interpretation, our results suggest that competition that causes avoidance of species on a patch can eventually lead to a mutualistic effect between species on the metacommunity level, suggesting a more general view of mutualism in metacommunities. Methods Numerical solution of the metacommunity dynamics To numerically solve the metacommunity dynamics described by Eqs (1), (3) and (4), we employed a numerical update scheme where for every time step Δt we first calculate the deter- ministic contributions (i.e. growth, competition and dispersal of every species on every patch) based on a Euler forward method. After that, demographic fluctuations are implemented by drawing the updated abundances from a Poisson distribution, which ensures the right statistics for the stochastic contributions. All calculations were performed in Python [50] and the results were evaluated using Mathematica [51] (the Python code developed for this study is available at https://github.com/Hallatscheklab/Self-Consistent-Metapopulations). For a more detailed description of the numerical methods see S1 Text, Sec. S1. Mean-field approach Complementary to the numerical solution, we employed a mean-field theory where the spe- cies-averaged and patch-averaged abundances are approximated by their mean-field values ^N PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 10 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities and �N , respectively. As detailed in the main text and S1 Text, Sec. S2, this mean-field approxi- mation allows us to derive the equilibrium species abundance distribution of (1), P, as a func- tion of the mean-fields ^N and �N . Finally, we numerically calculate ^N and �N by demanding self-consistency, i.e. ^N ¼ �N ¼ hNiP, where hNiP denotes the mean abundance based on the distribution P (all calculations were performed using Mathematica [51]). Applying this mean-field analysis to Eq (1) (see S1 Text, Sec. S2.1), to metacommunities with a density- dependent dispersal given by Eq (4) (see S1 Text, Sec. S2.3), and the observed parameters from numerical solutions with random interactions (see S1 Text, Sec. S4), we can derive the bifurca- tions, i.e. the mean population size as a function of the dispersal rate λ, discussed in Figs 2 and 4, and in the context of random interactions, respectively. Supporting information S1 Text. Supporting information. Detailed explanations of the numerical approach and ana- lytical mean-field analyses including explanations of the emergent bistability and extended analyses of metacommunities with random interactions as well as metacommunities with den- sity-dependent dispersal with additional direct competitive interactions. (PDF) S1 Fig. Self-consistency condition in the mean-field approximation. (EPS) S2 Fig. Species interactions can raise the species growth and dispersal parameters and thereby enable survival. (EPS) S3 Fig. Moderate direct competition does not alter the emergence of a tipping point in a metacommunity with density-dependent dispersal. (EPS) S4 Fig. Smooth and discontinuous transitions in metapopulations with explicit strong Allee effect. (EPS) S5 Fig. Bistability and self-organized tipping points in random metacommunities. (EPS) Acknowledgments While completing this work, we became aware of Giulia Lorenzana, Giulio Biroli and Ada Altieri working on metapopulations with random interactions, Eq (3), using replica methods. We thank them for stimulating discussions about our complimentary approaches. We thank current and former members of the Hallatschek lab for helpful comments and discussions. This research was supported by the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231 using NERSC award BER- ERCAP0024898. Author Contributions Conceptualization: Jonas Denk, Oskar Hallatschek. Data curation: Jonas Denk. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 11 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities Funding acquisition: Oskar Hallatschek. Investigation: Jonas Denk, Oskar Hallatschek. Methodology: Jonas Denk, Oskar Hallatschek. Project administration: Oskar Hallatschek. Resources: Oskar Hallatschek. Software: Jonas Denk. Supervision: Oskar Hallatschek. Visualization: Jonas Denk. Writing – original draft: Jonas Denk, Oskar Hallatschek. Writing – review & editing: Jonas Denk, Oskar Hallatschek. References 1. Sung J, Kim S, Cabatbat JJT, Jang S, Jin YS, Jung GY, et al. Global metabolic interaction network of the human gut microbiota for context-specific community-scale analysis. Nature communications. 2017; 8(1):15393. https://doi.org/10.1038/ncomms15393 PMID: 28585563 2. Venturelli OS, Carr AV, Fisher G, Hsu RH, Lau R, Bowen BP, et al. Deciphering microbial interactions in synthetic human gut microbiome communities. Molecular systems biology. 2018; 14(6):e8157. https:// doi.org/10.15252/msb.20178157 PMID: 29930200 3. Van Hoek MJ, Merks RM. Emergence of microbial diversity due to cross-feeding interactions in a spatial model of gut microbial metabolism. BMC systems biology. 2017; 11(1):1–18. 4. Dai L, Vorselen D, Korolev KS, Gore J. Generic indicators for loss of resilience before a tipping point leading to population collapse. Science. 2012; 336(6085):1175–1177. https://doi.org/10.1126/science. 1219805 PMID: 22654061 5. Courchamp F, Clutton-Brock T, Grenfell B. Inverse density dependence and the Allee effect. Trends in ecology & evolution. 1999; 14(10):405–410. https://doi.org/10.1016/S0169-5347(99)01683-3 PMID: 10481205 6. Stephens PA, Sutherland WJ. Consequences of the Allee effect for behaviour, ecology and conserva- tion. Trends in ecology & evolution. 1999; 14(10):401–405. https://doi.org/10.1016/S0169-5347(99) 01684-5 PMID: 10481204 7. Scheffer M, Carpenter SR, Dakos V, van Nes EH. Generic indicators of ecological resilience: inferring the chance of a critical transition. Annual Review of Ecology, Evolution, and Systematics. 2015; 46(1): 145–167. https://doi.org/10.1146/annurev-ecolsys-112414-054242 8. Dakos V, Bascompte J. Critical slowing down as early warning for the onset of collapse in mutualistic communities. Proceedings of the National Academy of Sciences. 2014; 111(49):17546–17551. https:// doi.org/10.1073/pnas.1406326111 PMID: 25422412 9. Hoek TA, Axelrod K, Biancalani T, Yurtsev EA, Liu J, Gore J. Resource availability modulates the coop- erative and competitive nature of a microbial cross-feeding mutualism. PLoS biology. 2016; 14(8): e1002540. https://doi.org/10.1371/journal.pbio.1002540 PMID: 27557335 10. Vet S, Gelens L, Gonze D. Mutualistic cross-feeding in microbial systems generates bistability via an Allee effect. Scientific reports. 2020; 10(1):1–12. https://doi.org/10.1038/s41598-020-63772-4 PMID: 32385386 11. Kramer AM, Sarnelle O, Knapp RA. Allee effect limits colonization success of sexually reproducing zoo- plankton. Ecology. 2008; 89(10):2760–2769. https://doi.org/10.1890/07-1505.1 PMID: 18959313 12. Hackney EE, McGraw JB. Experimental demonstration of an Allee effect in American ginseng. Conser- vation Biology. 2001; 15(1):129–136. https://doi.org/10.1111/j.1523-1739.2001.98546.x 13. Molnar PK, Derocher AE, Lewis MA, Taylor MK. Modelling the mating system of polar bears: a mecha- nistic approach to the Allee effect. Proceedings of the Royal Society B: Biological Sciences. 2008; 275(1631):217–226. https://doi.org/10.1098/rspb.2007.1307 PMID: 18029307 14. Hallatschek O, Nelson DR. Gene surfing in expanding populations. Theoretical population biology. 2008; 73(1):158–170. https://doi.org/10.1016/j.tpb.2007.08.008 PMID: 17963807 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 12 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities 15. Birzu G, Hallatschek O, Korolev KS. Fluctuations uncover a distinct class of traveling waves. Proceed- ings of the National Academy of Sciences. 2018; 115(16):E3645–E3654. https://doi.org/10.1073/pnas. 1715737115 PMID: 29610340 16. Gandhi SR, Yurtsev EA, Korolev KS, Gore J. Range expansions transition from pulled to pushed waves as growth becomes more cooperative in an experimental microbial population. Proceedings of the National Academy of Sciences. 2016; 113(25):6922–6927. https://doi.org/10.1073/pnas.1521056113 17. Rulands S, Klu¨ nder B, Frey E. Stability of localized wave fronts in bistable systems. Physical Review Letters. 2013; 110(3):038102. https://doi.org/10.1103/PhysRevLett.110.038102 PMID: 23373954 18. Ratzke C, Gore J. Self-organized patchiness facilitates survival in a cooperatively growing Bacillus sub- tilis population. Nature microbiology. 2016; 1(5):1–5. https://doi.org/10.1038/nmicrobiol.2016.22 19. Allen LJ, Allen EJ. A comparison of three different stochastic population models with regard to persis- tence time. Theoretical Population Biology. 2003; 64(4):439–449. https://doi.org/10.1016/S0040-5809 (03)00104-7 PMID: 14630481 20. Van Kampen NG. Stochastic processes in physics and chemistry. vol. 1. Elsevier; 1992. 21. Hinrichsen H. Non-equilibrium critical phenomena and phase transitions into absorbing states. Advances in Physics. 2000; 49(7):815–958. https://doi.org/10.1080/00018730050198152 22. O´ dor G. Universality classes in nonequilibrium lattice systems. Reviews of modern physics. 2004; 76(3):663. https://doi.org/10.1103/RevModPhys.76.663 23. Denk J, Hallatschek O. Self-consistent dispersal puts tight constraints on the spatiotemporal organiza- tion of species-rich metacommunities. Proceedings of the National Academy of Sciences. 2022; 119(26):e2200390119. https://doi.org/10.1073/pnas.2200390119 PMID: 35727977 24. Nachman G. Effects of demographic parameters on metapopulation size and persistence: an analytical stochastic model. Oikos. 2000; 91(1):51–65. https://doi.org/10.1034/j.1600-0706.2000.910105.x 25. Eriksson A, Elı´as-Wolff F, Mehlig B. Metapopulation dynamics on the brink of extinction. Theoretical population biology. 2013; 83:101–122. https://doi.org/10.1016/j.tpb.2012.08.001 PMID: 23047064 26. Casagrandi R, Gatto M. A persistence criterion for metapopulations. Theoretical population biology. 2002; 61(2):115–125. https://doi.org/10.1006/tpbi.2001.1558 PMID: 11969384 27. Strogatz SH. Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engi- neering. CRC press; 2018. 28. Villa Martı´n P, Bonachela JA, Levin SA, Muñoz MA. Eluding catastrophic shifts. Proceedings of the National Academy of Sciences. 2015; 112(15):E1828–E1836. https://doi.org/10.1073/pnas. 1414708112 PMID: 25825772 29. Sardanye´ s J, Piñero J, Sole´ R. Habitat loss-induced tipping points in metapopulations with facilitation. Population Ecology. 2019; 61(4):436–449. https://doi.org/10.1002/1438-390X.12020 30. Weissmann H, Shnerb NM. Stochastic desertification. EPL (Europhysics Letters). 2014; 106(2):28004. https://doi.org/10.1209/0295-5075/106/28004 31. May RM. Will a Large Complex System be Stable? Nature. 1972; 238:413–414. https://doi.org/10.1038/ 238413a0 PMID: 4559589 32. May RM. Stability and complexity in model ecosystems. Princeton university press; 1974. 33. Bunin G. Ecological communities with Lotka-Volterra dynamics. Physical Review E. 2017; 95(4):1–8. https://doi.org/10.1103/PhysRevE.95.042414 PMID: 28505745 34. Biroli G, Bunin G, Cammarota C. Marginally stable equilibria in critical ecosystems. New Journal of Physics. 2018; 20(8). https://doi.org/10.1088/1367-2630/aada58 35. Galla T. Dynamically evolved community size and stability of random Lotka-Volterra ecosystemsA. Epl. 2018; 123(4):1–13. https://doi.org/10.1209/0295-5075/123/48004 36. Altieri A, Roy F, Cammarota C, Biroli G. Properties of equilibria and glassy phases of the random lotka- volterra model with demographic noise. Physical Review Letters. 2021; 126(25):258301. https://doi.org/ 10.1103/PhysRevLett.126.258301 PMID: 34241496 37. Lorenzana GG, Altieri A, Biroli G. Interactions and migration rescuing ecological diversity, arXiv preprint (2023), arXiv:2309.09900. 38. Simpson SJ, McCaffery AR, Ha¨ gele BF. A behavioural analysis of phase change in the desert locust. Biological reviews. 1999; 74(4):461–480. https://doi.org/10.1111/j.1469-185X.1999.tb00038.x 39. Anstey ML, Rogers SM, Ott SR, Burrows M, Simpson SJ. Serotonin mediates behavioral gregarization underlying swarm formation in desert locusts. science. 2009; 323(5914):627–630. https://doi.org/10. 1126/science.1165939 PMID: 19179529 40. Nowicki P, Vrabec V. Evidence for positive density-dependent emigration in butterfly metapopulations. Oecologia. 2011; 167(3):657–665. https://doi.org/10.1007/s00442-011-2025-x PMID: 21625981 PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 13 / 14 PLOS COMPUTATIONAL BIOLOGYTipping points emerge from weak mutualism in metacommunities 41. De Meester N, Bonte D. Information use and density-dependent emigration in an agrobiont spider. Behavioral Ecology. 2010; 21(5):992–998. https://doi.org/10.1093/beheco/arq088 42. Lamb JS, Satge´ YG, Jodice PG. Influence of density-dependent competition on foraging and migratory behavior of a subtropical colonial seabird. Ecology and Evolution. 2017; 7(16):6469–6481. https://doi. org/10.1002/ece3.3216 PMID: 28861249 43. Miller ZR, Allesina S. Metapopulations with habitat modification. Proceedings of the National Academy of Sciences. 2021; 118(49):e2109896118. https://doi.org/10.1073/pnas.2109896118 PMID: 34857638 44. Amarasekare P. Allee Effects in Metapopulation Dynamics. The American Naturalist. 1998; 152(2): 298–302. https://doi.org/10.1086/286169 PMID: 18811393 45. Sato K. Allee threshold and extinction threshold for spatially explicit metapopulation dynamics with Allee effects. Population ecology. 2009; 51:411–418. https://doi.org/10.1007/s10144-009-0156-2 46. Hui C, Li Z. Distribution patterns of metapopulation determined by Allee effects. Population Ecology. 2004; 46:55–63. https://doi.org/10.1007/s10144-004-0171-2 47. Bunin G. Interaction patterns and diversity in assembled ecological communities; 2016. 48. Pearce MT, Agarwala A, Fisher DS. Stabilization of extensive fine-scale diversity by ecologically driven spatiotemporal chaos. Proceedings of the National Academy of Sciences. 2020; 117(25):14572– 14583. https://doi.org/10.1073/pnas.1915313117 PMID: 32518107 49. Roy F, Barbier M, Biroli G, Bunin G. Complex interactions can create persistent fluctuations in high- diversity ecosystems. PLoS computational biology. 2020; 16(5):e1007827. https://doi.org/10.1371/ journal.pcbi.1007827 PMID: 32413026 50. Van Rossum G, Drake Jr FL. Python reference manual. Centrum voor Wiskunde en Informatica Amsterdam; 1995. 51. Wolfram Research. Mathematica, Version 12.3.1, Champaign, IL; 2021. PLOS Computational Biology \| https://doi.org/10.1371/journal.pcbi.1011899 March 5, 2024 14 / 14 PLOS COMPUTATIONAL BIOLOGY
10.1126_science.adf4197	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Science. Author manuscript; available in PMC 2023 September 05. Published in final edited form as: Science. 2023 July 21; 381(6655): 319–324. doi:10.1126/science.adf4197. Reorientation of INO80 on hexasomes reveals basis for mechanistic versatility Hao Wu1,†, Elise N. Muñoz1,2,†, Laura J. Hsieh1, Un Seng Chio1, Muryam A. Gourdet1,2,, Geeta J. Narlikar1,, Yifan Cheng1,3,* 1Department of Biochemistry and Biophysics, University of California San Francisco, San Francisco, CA 94158, USA 2Tetrad Graduate Program, University of California San Francisco, San Francisco, CA 94158, USA 3Howard Hughes Medical Institute, University of California San Francisco, San Francisco, CA 94158, USA Abstract Unlike other chromatin remodelers, INO80 preferentially mobilizes hexasomes, which can form during transcription. Why INO80 prefers hexasomes over nucleosomes remains unclear. Here, we report structures of S. cerevisiae INO80 bound to a hexasome or a nucleosome. INO80 binds the two substrates in substantially different orientations. On a hexasome, INO80 places its ATPase subunit, Ino80, at superhelical location (SHL)-2, across from SHL-6/−7 as previously seen on nucleosomes. Our results suggest that INO80 action on hexasomes resembles action by other remodelers on nucleosomes, such that Ino80 is maximally active near SHL-2. The SHL-2 position also plays a critical role for nucleosome remodeling by INO80. Overall, the mechanistic adaptations used by INO80 for preferential hexasome sliding imply that sub-nucleosomal particles play considerable regulatory roles. One-Sentence Summary This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. *Correspondence: [email protected], [email protected], and [email protected]. †These authors contributed equally to this work Author contributions: H.W., M.A.G., L.J.H and E.N.M. purified INO80, M.A.G., L.J.H and E.N.M. purified hexasomes and nucleosomes, H.W. performed cryo-EM study, E.N.M. performed and quantified all biochemical experiments, U.S.C. generated the H2A R81A mutant, M.A.G., G.J.N. and Y.C. conceived and oversaw the project. All authors participated in interpretation and discussion of the results and writing of the manuscript. Competing interests: Y.C. is scientific advisory board member of ShuiMu BioSciences Ltd. Data and materials availability: For the core INO80 of the INO80-Hexasome complex (class 1, class 2 and class 3) and the core INO80 of the INO80-Nucleosome complex (class 1 and class 2), the coordinates are deposited in the Protein Data Bank with the accession codes 8ETS, 8ETU, 8ETW, 8EU9, and 8EUF; the cryo-EM density maps are deposited in the Electron Microscopy Data Bank (EMDB) with the accession codes EMD-28597, EMD-28599, EMD-28601, EMD-28609, and EMD-28613. For the hexasome of the INO80-Hexasome complex (class 1, class 2 and class 3) and the nucleosome of the INO80-Nucleosome complex (class 1 and class 2), the coordinates are deposited in the Protein Data Bank with the accession codes 8ETT, 8ETV, 8EU2, 8EUE, and 8EUJ; the cryo-EM density maps are deposited in the Electron Microscopy Data Bank (EMDB) with the accession codes EMD-28598, EMD-28600, EMD-28602, EMD-28612, and EMD-28614. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 2 Structural studies reveal that INO80 binds hexasomes and nucleosomes in dramatically different orientations, providing mechanistic insight into how hexasomes are regulated. In eukaryotes, central nuclear processes such as gene expression, DNA replication and DNA repair are coordinated with dynamic changes in chromatin states (1-3). ATP-dependent chromatin remodeling enzymes play essential roles in catalyzing such changes. These enzymes are broadly categorized into four major families: SWI/SNF, ISWI, CHD, and INO80 (4, 5). Each of them contains a core remodeling ATPase subunit and several auxiliary subunits that regulate the core ATPase. It has typically been presumed that the preferred substrate of these enzymes is a nucleosome, the smallest unit of chromatin, containing ~147 bp of DNA wrapped around an octamer of histone proteins (6). Consistent with this assumption, between them, these four classes slide the histone octamer, exchange histone variants, and transfer entire octamers (5, 7). The INO80 complex has been shown to play roles in regulating transcription, DNA replication and DNA repair (8-11). How INO80’s biochemical activities relate to its diverse biological roles is not well understood. Unlike remodelers from other families, whose core ATPase subunits bind the nucleosome near superhelical location (SHL)-2, Ino80, the core ATPase subunit of the INO80 complex, binds nucleosomes near SHL-6/−7 (fig. S1A) (12-14). It has been speculated that this key difference in nucleosome engagement reflects a fundamentally different remodeling mechanism (15, 16). Indeed, we showed that the preferred substrate of the S. cerevisiae INO80 complex is not a nucleosome but a hexasome, which is a sub-nucleosomal particle that lacks a histone H2A-H2B dimer (17). Hexasomes are generated during transcription and may also be formed during DNA replication and repair (18-21). Further, INO80’s activity on nucleosomes is more dependent on flanking DNA length than on hexasomes (17, 22). These results suggested that INO80 has the versatility to act on hexasomes or nucleosomes based on the density of nucleosomes and hexasomes at a given locus. Yet, fundamental mechanistic questions remain. It is not clear how INO80 can act on both nucleosomes and hexasomes, which differ substantially in their structures. Additionally, why INO80 has different flanking DNA length dependencies on hexasomes versus nucleosomes is unclear. Here, we report cryogenic-electron microscopy (cryo-EM) structures of endogenously purified S. cerevisiae INO80 bound to a hexasome and a nucleosome. We find that INO80 binds hexasomes and nucleosomes in opposite orientations, with Ino80 binding near SHL-2 on hexasomes and near −6/−7 on nucleosomes. The location of the Arp8 module suggests how flanking DNA length differentially regulates nucleosome and hexasome sliding. DNA gaps near SHL-2 inhibit sliding of both substrates by INO80. Together, our findings provide mechanistic insights into how INO80 slides both hexasomes and nucleosomes. Structures of the INO80-hexasome and -nucleosome complexes To visualize how INO80 binds to a hexasome or a nucleosome, we prepared hexasomes and nucleosomes on the same DNA templates containing the 147 bp 601 nucleosome positioning sequence with 80 bp of additional DNA as described previously (+80H and +80N, with definition explained in Fig. 1A, and fig. S1, A and B, and Supplementary Text) (17, 23, 24). Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 3 Complexes were formed by mixing hexasomes or nucleosomes with endogenously purified S. cerevisiae INO80 without adding nucleotide (fig. S1, C to H). We determined cryo-EM structures of the INO80-hexasome complex in three different conformational snapshots (Fig. 1, B and C, and figs. S2 to S6). The overall shape of INO80 is similar within these structures and also to previously published structures of the nucleosome in complex with human (12) and Chaetomium thermophilum (14) INO80. We group subunits of the INO80 complex into four modules: Rvb module (Rvb1/Rvb2), Arp8 module (Arp8/Arp4/Actin/Ies4 and Taf14), Ino80 module (Ino80/Ies2) and Arp5 module (Arp5/Ies6). The Ino80 protein consists of three major regions: the N-terminal domain (NTD), the HSA region (Ino80HSA) and the ATPase domain (Ino80ATPase). Detailed descriptions of these modules in our structures are in the Supplementary Text. While the INO80 architecture appears similar to that in the INO80-nucleosome structures, a major difference is that it is rotated ~180° on a hexasome compared to a nucleosome (Fig. 1, B to E). We identified two primary interactions between INO80 and the hexasome: Ino80ATPase binds the hexasome near SHL-3 (class 1), −2.5 (class 2) and −2 (class 3), and the Arp5/Ies6 module binds near SHL+1, +1.5 and +2 (fig. S6, A and B) respectively. Class 3 is the predominant INO80-hexasome class. All Ino80ATPase locations on hexasomes are different than on nucleosomes, which are near SHL-6 or SHL-7 (12-14). However, the Ino80 orientation on hexasomes is consistent with structures of other major chromatin remodelers on nucleosomes such as S. cerevisiae ISW1 (25-27), Chd1 (28-30), RSC (31-33), Snf2 (34) and in particular the SWR1 complex (35), which is from the same sub-family as the INO80 complex. In these structures the ATPase domains interact with nucleosomes near either SHL+2 or SHL-2 (Fig. 1E). Loss of an H2A-H2B dimer in a hexasome causes an additional ~35 bp of DNA to unwrap from the histone core (free DNA) (Fig. 1A, and fig. S1B). Comparison of our hexasome structures with an unbound hexasome (PDB: 6ZHY, (36)) reveals different levels of further DNA unwrapping. In class 1, the hexasome is almost identical with the unbound hexasome, without detectable additional DNA unwrapping. The level of DNA unwrapping increases as the Ino80ATPase binding position changes from SHL-3 (class 1) to SHL-2 (class 3) (Fig. 2, and fig. S6C). For comparison, we also determined structures of S. cerevisiae INO80 bound to a nucleosome and captured two conformational snapshots (class 1 and 2) from the same dataset (figs. S7 to S9, Supplementary Text). Ino80ATPase in class 1 is located near SHL-7, similar to its location in the human INO80-nucleosome structure (12), while in class 2, it binds near SHL-6, similar to the C. thermophilum structure (14) (fig. S9, A and C). The Arp5/Ies6 module interacts with the nucleosome near SHL-3 and SHL-2 (fig. S9D), respectively. These observations are also consistent with previous findings showing that nucleosomal DNA between SHL-7 and −6 is protected by INO80 (13). Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 4 The SHL-2 position plays a critical role in nucleosome and hexasome sliding We observe that Ino80ATPase engages the hexasome predominantly near SHL −2. These results raise the possibility that Ino80ATPase acts near SHL-2 when sliding hexasomes. In contrast, consistent with prior findings (12, 14), we observe that Ino80ATPase engages the nucleosome near two positions, SHL-7 and −6. Also as previously proposed, our findings are consistent with the possibility that Ino80ATPase acts near SHL-6 when sliding nucleosomes (13). A commonly used assay to identify the DNA location from where the ATPase domain of a remodeler acts to translocate DNA is to place a single nucleotide gap at the proposed site of action and test if the gap inhibits DNA translocation (37-39). Therefore, to directly test the importance of the SHL-6 and SHL-2 locations, we assembled nucleosomes and hexasomes with single base gaps near SHL-2 or SHL-6 and measured INO80 activity using a gel-based sliding assay (Fig. 3A). We found that a gap at SHL-6 inhibits INO80’s sliding activity on nucleosomes by ~ 200-fold but so did a gap at SHL-2 (Fig. 3, B to G). In contrast, a gap at SHL-6 did not inhibit INO80’s sliding activity on hexasomes, but a gap at SHL-2 slowed hexasomes sliding by ~ 2000-fold (Fig. 3, B to G). These results are consistent with Ino80ATPase acting near SHL-2 when sliding hexasomes and raise new questions about why both the SHL-2 and SHL-6 locations are critical for nucleosome sliding by INO80. We describe possible explanations in the Discussion. The role of the Arp8 module in flanking DNA length dependence S. cerevisiae INO80 slides +40 nucleosomes ~100-fold slower than +80 nucleosomes (17, 22). However, sliding hexasomes is less flanking DNA dependent. Our structures suggest that the Arp8 module requires ~40 bp of DNA for appropriate engagement. In class 1 of the INO80-hexasome structure, Arp8 engages with the ~35 bp of DNA unwrapped from removal of the H2A-H2B dimer and an additional ~5 bp of flanking DNA. In class 3 of the INO80-hexasome structure, the Arp8 module engages entirely with ~40 bp of unwrapped DNA that now includes additional DNA unwrapped relative to the unbound hexasome (Fig. 4). In contrast, in class 2 of the INO80-nucleosome structure, the Arp8 module engages entirely with flanking DNA consistent with previous findings (40) (Fig. 4). Our structural data with hexasomes along with the previous data with nucleosomes suggest that 40 bp may be the minimum amount of DNA needed for the Arp8 module to bind and that proper Arp8 module engagement is essential for maximal remodeling activity (40). Altered interactions by the Arp5 module To understand why Ino80 may not bind a nucleosome directly near SHL-2, we compared interactions made by Arp5/Ies6 in hexasomes versus nucleosomes (Supplementary Text). When INO80 binds to a hexasome, the Arp5/Ies6 regions used in the context of a nucleosome are repurposed for different interactions. Modeling the missing H2A-H2B dimer into our INO80-hexasome structure reveals steric clashes of the Arp5 module with the entry side proximal H2A-H2B dimer and with part of the DNA that wraps around the Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 5 H2A-H2B dimer near SHL-2 (fig. S11). These clashes could be avoided if the H2A-H2B dimer is sufficiently dislodged. To test for this possibility, we inhibited dimer dislodgement by introducing a site-specific disulfide crosslink between the two H2A molecules (N38C) (41) or promoted dimer dislodgement by using an H2A mutant (R81A) that destabilizes the H2A-H2B/H3-H4 interface (42) (fig. S12, A, B and H). The disulfide crosslink did not inhibit nucleosome sliding while the H2A mutant did not promote nucleosome sliding (fig. S12, C to G), indicating that complete dimer dislodgement is not necessary for INO80- mediated nucleosome sliding. In the absence of dimer dislodgement, another way to avoid these clashes could be by substantial rearrangement of the Arp5 module together with subtle rearrangements of the H2A-H2B dimer (fig. S9E). DISCUSSION Implications of the INO80-hexasome structure for nucleosome sliding by INO80 The major conformation of the INO80-hexasome complex (class 3) has Ino80ATPase near SHL-2 and approximately ~15bp of unwrapped DNA from the entry site in addition to the ~35bp of DNA that is unwrapped from removal of an H2A-H2B dimer. The placement of Ino80ATPase near SHL-2 is consistent with how the ATPase subunits of other remodelers bind the nucleosome. Together with our prior finding that hexasomes are remodeled faster than nucleosomes, these results suggest that the class 3 structure represents the sliding- competent conformation of INO80 on hexasomes (Fig. 5A, and fig. S13A). In contrast, the states of INO80 bound to a nucleosome have Ino80ATPase bound near either SHL-6 or −7 consistent with previous findings. These differences raise the question of whether the INO80-nucleosome structures represent sliding-competent conformations or whether a rearrangement of Ino80ATPase to SHL-2 is necessary to achieve efficient nucleosome sliding. Previous crosslinking studies have shown that detachment of nucleosomal DNA from H2A-H2B close to the entry site occurs during INO80 remodeling (13). Our data show that progressively more DNA is unwrapped as Ino80ATPase binds closer to SHL-2 on hexasomes (Fig. 2, and fig. S6C). Together these results suggest that DNA unwrapping is coupled to Ino80ATPase accessing its most sliding-competent state. Foot-printing studies have shown that while binding of INO80 to nucleosomes mainly protects nucleosomal DNA from SHL-5 to SHL-6 and near SHL-3, there is modest but detectable protection near SHL-2 (13). Nicks and gaps between SHL-7 and SHL −2 have been shown to inhibit nucleosome sliding to different extents (13, 43). Here we show that site-specific gaps near SHL-2 or SHL-6 substantially inhibit INO80’s sliding of nucleosomes (by ~200 fold). DNA gaps are commonly used to identify the site of action of the ATPase domain of remodelers (37-39). We therefore speculate that INO80 initially binds the nucleosome with Ino80ATPase near SHL-6/−7, and this is followed by an ATP-dependent rotation around the nucleosome to position Ino80ATPase near SHL-2 from where Ino80ATPase then translocates nucleosomal DNA (Fig. 5B, and fig. S14A). A gap at SHL-6 would then inhibit ATP- dependent movement of Ino80ATPase on the nucleosome while the gap at SHL-2 would inhibit translocation of nucleosomal DNA by INO80 relative to the histone octamer (fig. S14). Single-molecule FRET studies have identified an ATP-dependent pause phase prior to ATP-dependent nucleosome sliding (22). The pause could represent the reorientation Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 6 of Ino80ATPase from SHL-6/−7 towards SHL-2 and add a step that slows remodeling of nucleosomes compared to hexasomes. Simply placing the INO80 complex as is on nucleosomes with the Ino80ATPase near SHL-2 results in steric clashes of the Arp5 module with the nucleosome (fig. S11). While partial H2A-H2B dimer dislodgment, as previously proposed, could avoid such clashes (17), our biochemical data here indicate that dimer dislodgement is not essential for nucleosome sliding by INO80 (fig. S12). Thus more structural studies are needed to understand how INO80 might rotate around a nucleosome. Alternatively, a gap near SHL-2 may affect the action of the Arp5 module. For such a scenario we speculate that Ino80ATPase translocates DNA near SHL-6 and effective translocation also requires action of the Arp5 module near SHL-2 as previously proposed (12, 14). A gap at SHL-6 would then inhibit translocation of nucleosomal DNA by Ino80ATPase and a gap at SHL-2 would inhibit productive engagement of the Arp5 module (fig. S15). Clearly distinguishing between these two models will require substantial additional structural analysis of INO80 remodeling intermediates on nucleosomes. Implications for hexasome sliding by INO80 Our structures provide a view into how INO80 engages a hexasome. In the predominant INO80-hexasome structure, Ino80ATPase binds near SHL-2. A site-specific gap at SHL-2 substantially inhibits INO80’s sliding of hexasomes (~2000 fold) while a gap near SHL-6 does not have a major effect. We therefore hypothesize that Ino80ATPase bound at SHL-2 on a hexasome represents the active structure. Compared to the subtle changes at SHL-2 observed when other remodelers bind nucleosomes (16), the 15 bp of unwrapped DNA (up to SHL-2.5) in class 3 substantially loosens histone DNA interactions and thus may allow more ready translocation from SHL-2. We further propose that the new contacts made by the Arp5/Ies6 module with the exposed H3-H4 surface provide an anchor allowing the Ino80 motor to efficiently pump DNA through the hexasome. These findings also explain the differential effects of the Arp5/Ies6 module on hexasome versus nucleosome sliding (17). The location of the Arp8 module is also different on hexasomes than on nucleosomes. On nucleosomes the Arp8 module binds ~ 40 bp entirely on the flanking DNA (Fig. 4). In the most prevalent INO80-hexasome state (class 3), the Arp8 module is bound entirely to the unwrapped DNA, substantially reducing the need to bind flanking DNA (Fig. 4). These different binding modes of the Arp8 module could explain why hexasome sliding by INO80 is less dependent on flanking DNA length compared to nucleosome sliding. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Acknowledgements Cryo-EM facility at UCSF is managed by Dr. David Bulkley and Mr. Glenn Gilbert. Computation at Cheng laboratory is supported by Mr. Matthiew Harrington and Dr. Junrui Li. We thank Dr. Lvqin Zheng for advice on model building, Dr. Zanlin Yu for providing GO grids, Julia Tretyakova for expressing and purifying histones, Upneet Kaur for providing INO80 WT enzyme, and members of Narlikar and Cheng laboratories for helpful discussions. We thank Dr. Carl Wu and Dr. Anand Ranjan for sharing unpublished data showing inhibition of Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Funding: INO80 activity on nucleosomes containing a gap near SHL-2. We also thank Dr. Sebastian Diendl and Dr. Anton Sabantsev for detailed advice on generating DNAs with site-specific base gaps. Page 7 This work is supported by grants from National Institute of Health (R35GM140847 to Y.C., R35GM127020 to G.J.N., F31GM136187 to M.A.G., F31GM142271 to E.N.M., F32GM137463 to U.S.C) and by an American Cancer Society—Roaring Fork Valley Research Fund Postdoctoral Fellowship (PF-18-155-01-DMC to L.J.H.). Equipment at UCSF cryo-EM facility was partially supported by National Institutes of Health (NIH) grants (S10OD020054, S10OD021741, and S10OD025881). Y.C. is an Investigator of Howard Hughes Medical Institute. References and Notes 1. Bar-Ziv R, Voichek Y, Barkai N, Chromatin dynamics during DNA replication. Genome Res 26, 1245–1256 (2016). doi:10.1101/gr.201244.115 [PubMed: 27225843] 2. Ehrenhofer-Murray AE, Chromatin dynamics at DNA replication, transcription and repair. Eur J Biochem 271, 2335–2349 (2004). doi:10.1111/j.1432-1033.2004.04162.x [PubMed: 15182349] 3. Hubner MR, Spector DL, Chromatin dynamics. Annu Rev Biophys 39, 471–489 (2010). doi:10.1146/annurev.biophys.093008.131348 [PubMed: 20462379] 4. Clapier CR, Cairns BR, The biology of chromatin remodeling complexes. Annu Rev Biochem 78, 273–304 (2009). doi:10.1146/annurev.biochem.77.062706.153223 [PubMed: 19355820] 5. Zhou CY, Johnson SL, Gamarra NI, Narlikar GJ, Mechanisms of ATP-Dependent Chromatin Remodeling Motors. Annu Rev Biophys 45, 153–181 (2016). doi:10.1146/annurev- biophys-051013-022819 [PubMed: 27391925] 6. Luger K, Mader AW, Richmond RK, Sargent DF, Richmond TJ, Crystal structure of the nucleosome core particle at 2.8 A resolution. Nature 389, 251–260 (1997). doi:10.1038/38444 [PubMed: 9305837] 7. Clapier CR, Iwasa J, Cairns BR, Peterson CL, Mechanisms of action and regulation of ATP-dependent chromatin-remodelling complexes. Nat Rev Mol Cell Biol 18, 407–422 (2017). doi:10.1038/nrm.2017.26 [PubMed: 28512350] 8. Bao Y, Shen X, INO80 subfamily of chromatin remodeling complexes. Mutat Res 618, 18–29 (2007). doi:10.1016/j.mrfmmm.2006.10.006 [PubMed: 17316710] 9. Morrison AJ, Shen X, Chromatin remodelling beyond transcription: the INO80 and SWR1 complexes. Nat Rev Mol Cell Biol 10, 373–384 (2009). doi:10.1038/nrm2693 [PubMed: 19424290] 10. Poli J, Gasser SM, Papamichos-Chronakis M, The INO80 remodeller in transcription, replication and repair. Philos Trans R Soc Lond B Biol Sci 372, (2017). doi:10.1098/rstb.2016.0290 11. Shen X, Mizuguchi G, Hamiche A, Wu C, A chromatin remodelling complex involved in transcription and DNA processing. Nature 406, 541–544 (2000). doi:10.1038/35020123 [PubMed: 10952318] 12. Ayala R et al. , Structure and regulation of the human INO80-nucleosome complex. Nature 556, 391–395 (2018). doi:10.1038/s41586-018-0021-6 [PubMed: 29643506] 13. Brahma S et al. , INO80 exchanges H2A.Z for H2A by translocating on DNA proximal to histone dimers. Nat Commun 8, 15616 (2017). doi:10.1038/ncomms15616 [PubMed: 28604691] 14. Eustermann S et al. , Structural basis for ATP-dependent chromatin remodelling by the INO80 complex. Nature 556, 386–390 (2018). doi:10.1038/s41586-018-0029-y [PubMed: 29643509] 15. Markert J, Luger K, Nucleosomes Meet Their Remodeler Match. Trends Biochem Sci 46, 41–50 (2021). doi:10.1016/j.tibs.2020.08.010 [PubMed: 32917506] 16. Yan L, Chen Z, A Unifying Mechanism of DNA Translocation Underlying Chromatin Remodeling. Trends Biochem Sci 45, 217–227 (2020). doi:10.1016/j.tibs.2019.09.002 [PubMed: 31623923] 17. Hsieh LJ et al. , A hexasome is the preferred substrate for the INO80 chromatin remodeling complex, allowing versatility of function. Mol Cell 82, 2098–2112 e2094 (2022). doi:10.1016/ j.molcel.2022.04.026 [PubMed: 35597239] 18. Henikoff S, Mechanisms of Nucleosome Dynamics In Vivo. Cold Spring Harb Perspect Med 6, (2016). doi:10.1101/cshperspect.a026666 Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 8 19. Kireeva ML et al. , Nucleosome remodeling induced by RNA polymerase II: loss of the H2A/H2B dimer during transcription. Mol Cell 9, 541–552 (2002). doi:10.1016/s1097-2765(02)00472-0 [PubMed: 11931762] 20. Kulaeva OI, Hsieh FK, Studitsky VM, RNA polymerase complexes cooperate to relieve the nucleosomal barrier and evict histones. Proc Natl Acad Sci U S A 107, 11325–11330 (2010). doi:10.1073/pnas.1001148107 [PubMed: 20534568] 21. Ramachandran S, Ahmad K, Henikoff S, Transcription and Remodeling Produce Asymmetrically Unwrapped Nucleosomal Intermediates. Mol Cell 68, 1038–1053 e1034 (2017). doi:10.1016/ j.molcel.2017.11.015 [PubMed: 29225036] 22. Zhou CY et al. , The Yeast INO80 Complex Operates as a Tunable DNA Length-Sensitive Switch to Regulate Nucleosome Sliding. Mol Cell 69, 677–688 e679 (2018). doi:10.1016/ j.molcel.2018.01.028 [PubMed: 29452642] 23. Levendosky RF, Bowman GD, Asymmetry between the two acidic patches dictates the direction of nucleosome sliding by the ISWI chromatin remodeler. Elife 8, (2019). doi:10.7554/eLife.45472 24. Levendosky RF, Sabantsev A, Deindl S, Bowman GD, The Chd1 chromatin remodeler shifts hexasomes unidirectionally. Elife 5, (2016). doi:10.7554/eLife.21356 25. Armache JP et al. , Cryo-EM structures of remodeler-nucleosome intermediates suggest allosteric control through the nucleosome. Elife 8, (2019). doi:10.7554/eLife.46057 26. Chittori S, Hong J, Bai Y, Subramaniam S, Structure of the primed state of the ATPase domain of chromatin remodeling factor ISWI bound to the nucleosome. Nucleic Acids Res 47, 9400–9409 (2019). doi:10.1093/nar/gkz670 [PubMed: 31402386] 27. Yan L, Wu H, Li X, Gao N, Chen Z, Structures of the ISWI-nucleosome complex reveal a conserved mechanism of chromatin remodeling. Nat Struct Mol Biol 26, 258–266 (2019). doi:10.1038/s41594-019-0199-9 [PubMed: 30872815] 28. Farnung L, Vos SM, Wigge C, Cramer P, Nucleosome-Chd1 structure and implications for chromatin remodelling. Nature 550, 539–542 (2017). doi:10.1038/nature24046 [PubMed: 29019976] 29. Nodelman IM et al. , Nucleosome recognition and DNA distortion by the Chd1 remodeler in a nucleotide-free state. Nat Struct Mol Biol 29, 121–129 (2022). doi:10.1038/s41594-021-00719-x [PubMed: 35173352] 30. Sundaramoorthy R et al. , Structure of the chromatin remodelling enzyme Chd1 bound to a ubiquitinylated nucleosome. Elife 7, (2018). doi:10.7554/eLife.35720 31. Patel AB et al. , Architecture of the chromatin remodeler RSC and insights into its nucleosome engagement. Elife 8, (2019). doi:10.7554/eLife.54449 32. Wagner FR et al. , Structure of SWI/SNF chromatin remodeller RSC bound to a nucleosome. Nature 579, 448–451 (2020). doi:10.1038/s41586-020-2088-0 [PubMed: 32188943] 33. Ye Y et al. , Structure of the RSC complex bound to the nucleosome. Science 366, 838–843 (2019). doi:10.1126/science.aay0033 [PubMed: 31672915] 34. Liu X, Li M, Xia X, Li X, Chen Z, Mechanism of chromatin remodelling revealed by the Snf2-nucleosome structure. Nature 544, 440–445 (2017). doi:10.1038/nature22036 [PubMed: 28424519] 35. Willhoft O et al. , Structure and dynamics of the yeast SWR1-nucleosome complex. Science 362, (2018). doi:10.1126/science.aat7716 36. Lehmann LC et al. , Mechanistic Insights into Regulation of the ALC1 Remodeler by the Nucleosome Acidic Patch. Cell Rep 33, 108529 (2020). doi:10.1016/j.celrep.2020.108529 [PubMed: 33357431] 37. Saha A, Wittmeyer J, Cairns BR, Chromatin remodeling through directional DNA translocation from an internal nucleosomal site. Nat Struct Mol Biol 12, 747–755 (2005). doi:10.1038/nsmb973 [PubMed: 16086025] 38. Schwanbeck R, Xiao H, Wu C, Spatial contacts and nucleosome step movements induced by the NURF chromatin remodeling complex. J Biol Chem 279, 39933–39941 (2004). doi:10.1074/ jbc.M406060200 [PubMed: 15262970] Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 9 39. Zofall M, Persinger J, Kassabov SR, Bartholomew B, Chromatin remodeling by ISW2 and SWI/SNF requires DNA translocation inside the nucleosome. Nat Struct Mol Biol 13, 339–346 (2006). doi:10.1038/nsmb1071 [PubMed: 16518397] 40. Kunert F et al. , Structural mechanism of extranucleosomal DNA readout by the INO80 complex. Sci Adv 8, eadd3189 (2022). doi:10.1126/sciadv.add3189 [PubMed: 36490333] 41. Frouws TD, Barth PD, Richmond TJ, Site-Specific Disulfide Crosslinked Nucleosomes with Enhanced Stability. J Mol Biol 430, 45–57 (2018). doi:10.1016/j.jmb.2017.10.029 [PubMed: 29113904] 42. Lehmann K et al. , Effects of charge-modifying mutations in histone H2A alpha3-domain on nucleosome stability assessed by single-pair FRET and MD simulations. Sci Rep 7, 13303 (2017). doi:10.1038/s41598-017-13416-x [PubMed: 29038501] 43. Mueller-Planitz F, Klinker H, Becker PB, Nucleosome sliding mechanisms: new twists in a looped history. Nat Struct Mol Biol 20, 1026–1032 (2013). doi:10.1038/nsmb.2648 [PubMed: 24008565] 44. Shen X, Preparation and analysis of the INO80 complex. Methods Enzymol 377, 401–412 (2004). doi:10.1016/S0076-6879(03)77026-8 [PubMed: 14979041] 45. Dyer PN et al. , Reconstitution of nucleosome core particles from recombinant histones and DNA. Methods Enzymol 375, 23–44 (2004). doi:10.1016/s0076-6879(03)75002-2 [PubMed: 14870657] 46. Luger K, Rechsteiner TJ, Richmond TJ, Expression and purification of recombinant histones and nucleosome reconstitution. Methods Mol Biol 119, 1–16 (1999). doi:10.1385/1-59259-681-9:1 [PubMed: 10804500] 47. Deindl S et al. , ISWI remodelers slide nucleosomes with coordinated multi-base-pair entry steps and single-base-pair exit steps. Cell 152, 442–452 (2013). doi:10.1016/j.cell.2012.12.040 [PubMed: 23374341] 48. Gamarra N, Narlikar GJ, Histone dynamics play a critical role in SNF2h-mediated nucleosome sliding. Nat Struct Mol Biol 28, 548–551 (2021). doi:10.1038/s41594-021-00620-7 [PubMed: 34226739] 49. Gkikopoulos T et al. , A role for Snf2-related nucleosome-spacing enzymes in genome- wide nucleosome organization. Science 333, 1758–1760 (2011). doi:10.1126/science.1206097 [PubMed: 21940898] 50. Rhee HS, Bataille AR, Zhang L, Pugh BF, Subnucleosomal structures and nucleosome asymmetry across a genome. Cell 159, 1377–1388 (2014). doi:10.1016/j.cell.2014.10.054 [PubMed: 25480300] 51. Palovcak E et al. , A simple and robust procedure for preparing graphene-oxide cryo-EM grids. J Struct Biol 204, 80–84 (2018). doi:10.1016/j.jsb.2018.07.007 [PubMed: 30017701] 52. Wang F et al. , Amino and PEG-amino graphene oxide grids enrich and protect samples for high- resolution single particle cryo-electron microscopy. J Struct Biol 209, 107437 (2020). doi:10.1016/ j.jsb.2019.107437 [PubMed: 31866389] 53. Ohi M, Li Y, Cheng Y, Walz T, Negative Staining and Image Classification - Powerful Tools in Modern Electron Microscopy. Biol Proced Online 6, 23–34 (2004). doi:10.1251/bpo70 [PubMed: 15103397] 54. Mastronarde DN, Automated electron microscope tomography using robust prediction of specimen movements. Journal of Structural Biology 152, 36–51 (2005). doi:10.1016/j.jsb.2005.07.007 [PubMed: 16182563] 55. Zheng SQ et al. , MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat Methods 14, 331–332 (2017). doi:10.1038/nmeth.4193 [PubMed: 28250466] 56. Punjani A, Rubinstein JL, Fleet DJ, Brubaker MA, cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat Methods 14, 290–296 (2017). doi:10.1038/nmeth.4169 [PubMed: 28165473] 57. Scheres SH, RELION: implementation of a Bayesian approach to cryo-EM structure determination. J Struct Biol 180, 519–530 (2012). doi:10.1016/j.jsb.2012.09.006 [PubMed: 23000701] 58. Pettersen EF et al. , UCSF Chimera--a visualization system for exploratory research and analysis. J Comput Chem 25, 1605–1612 (2004). doi:10.1002/jcc.20084 [PubMed: 15264254] Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 10 59. Grant T, Rohou A, Grigorieff N, cisTEM, user-friendly software for single-particle image processing. Elife 7, (2018). doi:10.7554/eLife.35383 60. Sanchez-Garcia R et al. , DeepEMhancer: a deep learning solution for cryo-EM volume post- processing. Commun Biol 4, 874 (2021). doi:10.1038/s42003-021-02399-1 [PubMed: 34267316] 61. Rosenthal PB, Henderson R, Optimal determination of particle orientation, absolute hand, and contrast loss in single-particle electron cryomicroscopy. J Mol Biol 333, 721–745 (2003). doi:10.1016/j.jmb.2003.07.013 [PubMed: 14568533] 62. Dang S et al. , Cryo-EM structures of the TMEM16A calcium-activated chloride channel. Nature 552, 426–429 (2017). doi:10.1038/nature25024 [PubMed: 29236684] 63. Jumper J et al. , Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). doi:10.1038/s41586-021-03819-2 [PubMed: 34265844] 64. Emsley P, Cowtan K, Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60, 2126–2132 (2004). doi:10.1107/S0907444904019158 [PubMed: 15572765] 65. Afonine PV et al. , Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 68, 352–367 (2012). doi:10.1107/S0907444912001308 [PubMed: 22505256] 66. Davey CA, Sargent DF, Luger K, Maeder AW, Richmond TJ, Solvent mediated interactions in the structure of the nucleosome core particle at 1.9 a resolution. J Mol Biol 319, 1097–1113 (2002). doi:10.1016/S0022-2836(02)00386-8 [PubMed: 12079350] 67. Farnung L, Ochmann M, Garg G, Vos SM, Cramer P, Structure of a backtracked hexasomal intermediate of nucleosome transcription. Mol Cell 82, 3126–3134 e3127 (2022). doi:10.1016/ j.molcel.2022.06.027 [PubMed: 35858621] 68. Mayanagi K et al. , Structural visualization of key steps in nucleosome reorganization by human FACT. Sci Rep 9, 10183 (2019). doi:10.1038/s41598-019-46617-7 [PubMed: 31308435] 69. Brahma S, Ngubo M, Paul S, Udugama M, Bartholomew B, The Arp8 and Arp4 module acts as a DNA sensor controlling INO80 chromatin remodeling. Nat Commun 9, 3309 (2018). doi:10.1038/ s41467-018-05710-7 [PubMed: 30120252] 70. Knoll KR et al. , The nuclear actin-containing Arp8 module is a linker DNA sensor driving INO80 chromatin remodeling. Nat Struct Mol Biol 25, 823–832 (2018). doi:10.1038/s41594-018-0115-8 [PubMed: 30177756] 71. Oberbeckmann E et al. , Ruler elements in chromatin remodelers set nucleosome array spacing and phasing. Nat Commun 12, 3232 (2021). doi:10.1038/s41467-021-23015-0 [PubMed: 34050140] Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 11 Fig. 1. Structure of the INO80-hexasome complex reveals large rotation. (A) Cartoon illustration of a +X Nucleosome and a +X Hexasome. H2A-H2B dimer proximal to the flanking DNA (entry side dimer): cyan; H3-H4: light gray; 601 DNA: dark gray; flanking DNA: orange; additional free (unwrapped) DNA: cyan; super helical locations: yellow dots; DNA from the bottom gyre: dotted line. (B) Two different views of cryo-EM density map of the INO80-hexasome complex (class 3). (C) Atomic model of the INO80-hexasome complex (class 3), viewed in the same orientation as the map is viewed in (B). (D) Cryo-EM density map of Chaetomium thermophilum INO80-nucleosome complex (EMDB: 4277 (14)) displayed with its nucleosome dyad and H3-H4 tetramer aligned with that of the hexasome in the right panel of (B). Note that INO80 on a hexasome rotates ~180° from where it sits on a nucleosome when keeping the nucleosome/hexasome dyad Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 12 and H3-H4 aligned. (E) Structural comparisons of INO80-nucleosome complex (left), SWR- nucleosome complex (middle) and INO80-hexasome complex (right), with nucleosome/ hexasome dyad and H3-H4 aligned. Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 13 Fig. 2. Conformational snapshots of INO80-hexasome complexes. Comparison of DNA from each INO80-hexasome class (blue) with DNA from an unbound hexasome (PDB: 6ZHY, gray), showing degree of DNA unwrapping (upper row) and binding locations of Ino80ATPase and Arp5 (bottom row). Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 14 Fig. 3. Inhibition of DNA translocation at specific SHL sites influence nucleosome and hexasome sliding by INO80. (A) Cartoon illustration of a +80 Nucleosome (left) and a +80 Hexasome (right) with approximate locations of site-specific single base gaps indicated. Colors are the same as in Fig. 1A. (B-C) Example gels and time courses of native gel-based remodeling assays of WT INO80 on +80 nucleosomes with no gap, gap near SHL-2, and gap near SHL-6. (D-E) Example gels and time courses of native gel-based remodeling assays of WT INO80 on +80 hexasomes with no gap, gap near SHL-2, and gap near SHL-6. (F-G) Average observed rate constants of INO80 sliding activity. kobs (min−1): +80N: 1.551 ± 0.1846; +80N Gap @ SHL-2: 0.005995 ± 0.001054; +80N Gap @ SHL-6: 0.006497 ± 0.0007117; +80H: 1.01 ± 0.1668; +80H Gap @ SHL-2: 0.000379 ± 0.0002849; +80H Gap @ SHL-6: 1.213 ± 0.2209. Data represent the mean ± SEM for three technical replicates performed under single-turnover conditions with saturating enzyme and ATP. Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 15 Fig. 4. The Arp8 module engages different regions of DNA in nucleosomes versus hexasomes. Overlay of atomic models of the hexasome (class 1 and class 3) and the nucleosome (class 2) with the Arp8 module (PDB: 8A5O), aligned by the H3-H4 tetramer. Science. Author manuscript; available in PMC 2023 September 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Wu et al. Page 16 Fig. 5. Model of INO80-induced hexasome and nucleosome sliding. (A) Hexasome sliding: the Ino80ATPase samples different positions between SHL-3 and SHL-2 but binds predominantly near SHL-2. The INO80 complex becomes sliding- competent when Ino80ATPase engages near SHL-2. (B) Nucleosome sliding: INO80 initially binds with Ino80ATPase at SHL-7 or −6. Upon ATP-hydrolysis, Ino80ATPase moves toward SHL-2 where INO80 becomes sliding-competent. Science. Author manuscript; available in PMC 2023 September 05.
10.1073_pnas.2300360120	RESEARCH ARTICLE \| MICROBIOLOGY OPEN ACCESS Structure-based design of a SARS-CoV-2 Omicron-specific inhibitor Kailu Yanga,b,c,d,e Tomas Kirchhausenf,g,h , Richard A. Pfuetznera,b,c,d,e, Luis Esquiviesa,b,c,d,e, , Chuchu Wanga,b,c,d,e, Alex J. B. Kreutzbergerf,g , and Axel T. Brungera,b,c,d,e,1 , K. Ian Whitea,b,c,d,e Edited by Adriaan Bax, NIH, Bethesda, MD; received January 7, 2023; accepted February 13, 2023 The Omicron variant of severe acute respiratory syndrome coronavirus 2 (SARS- CoV-2) introduced a relatively large number of mutations, including three mutations in the highly conserved heptad repeat 1 (HR1) region of the spike glycoprotein (S) critical for its membrane fusion activity. We show that one of these mutations, N969K induces a substantial displacement in the structure of the heptad repeat 2 (HR2) backbone in the HR1HR2 postfusion bundle. Due to this mutation, fusion-entry peptide inhibitors based on the Wuhan strain sequence are less efficacious. Here, we report an Omicron-specific peptide inhibitor designed based on the structure of the Omicron HR1HR2 postfusion bundle. Specifically, we inserted an additional residue in HR2 near the Omicron HR1 K969 residue to better accommodate the N969K mutation and relieve the distortion in the structure of the HR1HR2 postfusion bun- dle it introduced. The designed inhibitor recovers the loss of inhibition activity of the original longHR2_42 peptide with the Wuhan strain sequence against the Omicron variant in both a cell–cell fusion assay and a vesicular stomatitis virus (VSV)-SARS- CoV-2 chimera infection assay, suggesting that a similar approach could be used to combat future variants. From a mechanistic perspective, our work suggests the interactions in the extended region of HR2 may mediate the initial landing of HR2 onto HR1 during the transition of the S protein from the prehairpin intermediate to the postfusion state. SARS-CoV-2 \| Omicron \| membrane fusion \| inhibitor \| rational design Renewed threats from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) are caused by a series of variants, including Omicron. The first strain of Omicron, B.1.1.529 or BA.1, was identified in South Africa in late 2021, and since then a variety of subvariants have evolved, including BA.1.1, BA.2, BA.2.12.1, BA.4, and BA.5 (1). These Omicron subvariants have an increased ability to escape neutralizing antibodies. For example, the BA.2.12.1, BA.4, and BA.5 subvariants show increased evasion compared to BA.2 of plasma-derived neutralizing antibodies from individuals who were triple-vac- cinated or developed a BA.1 infection after vaccination (2), and the BA.2.75 variant shows enhanced evasion compared to BA.2 (3). New Omicron-specific vaccines are currently distributed although it has been suggested that they may not be much more effective than existing vaccines (4), and that they may be less efficacious against the most recent sub-var- iants BQ.1, BQ.1.1, and XBB.1.5 of Omicron (5–7). Moreover, infections in unvaccinated and immune-compromised individuals, breakthrough infections, and so-called long COVID are substantial public health concerns that warrant development of effective antiviral compounds in addition to existing vaccination regimens (8, 9). To reduce the public health threat of Omicron and future emerging variants, we and others have proposed development of antivirals targeting the relatively conserved heptad repeat 1 and 2 (HR1 and HR2) regions of the spike glycoprotein (S) that drive membrane fusion by the formation of the HR1HR2 postfusion bundle (10–14). Although the ~130 residues of the HR1 and HR2 regions are conserved in all prior SARS-CoV-2 variants, a few mutations have reached fixation with Omicron in the HR1 region (Q954H and N969K for all Omicron subvariants, and, additionally, L981F for BA.1). Interestingly, the mutational load of the HR1 and HR2 regions is lower than that of certain other regions of the S protein; for the Omicron variants, the receptor binding domain (RBD) has 5 to 8 times more mutations per site than the HR1 and HR2 regions relative to the original Wuhan strain. The higher mutation frequency in the RBD domain is probably driven by evolutionary escape from neutralizing antibodies (15). Previously, we reported an unmodified peptide with an extended sequence composed of 42 amino acids of the HR2 region, termed longHR2_42, that was designed to target these relatively more conserved HR1HR2 regions; this efficacious peptide inhibits infec- tions by the original Wuhan strain and several variants of SARS-CoV-2 with an IC50 of Significance The renewed threats from severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) by existing and emerging variants, including Omicron emerging sub-variants, require optimized development of vaccines and antiviral therapeutics. Vaccines and therapeutics targeting the critical viral spike glycoprotein (S) must accommodate the mutations in S. Even the highly conserved region that plays a critical role in membrane fusion exhibits up to three mutations in the HR1 region of Omicron variants. Here we show that one of these mutations causes a distortion in the structure of the SARS-CoV-2 postfusion structure, and we compensated for this structural change by structure-based design of an Omicron-specific peptide inhibitor that inhibits Omicron fusion and infection with low nanomolar activities. Author contributions: K.Y., C.W., A.J.B.K., T.K., and A.T.B. designed research; K.Y., C.W., A.J.B.K., R.A.P., and L.E. performed research; K.Y., C.W., A.J.B.K., K.I.W., T.K., and A.T.B. analyzed data; and K.Y., C.W., A.J.B.K., K.I.W., R.A.P., L.E., T.K., and A.T.B. wrote the paper. Competing interest statement: A US provisional patent application for aspects of this work has been filed. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2300360120/-/DCSupplemental. Published March 20, 2023. PNAS 2023 Vol. 120 No. 13 e2300360120 https://doi.org/10.1073/pnas.2300360120 1 of 7 ~1 nM (13). However, the inhibition activity of this peptide against the Omicron variant was around fourfold weaker in the vesicular stomatitis virus (VSV)-SARS-CoV-2 chimera infection assay and around tenfold weaker in the authentic SARS-CoV-2 infection assay. We hypothesized that this decreased inhibition activity is probably caused by the three mutations in the HR1 region—Q954H, N969K, and L981F—of Omicron BA.1. All three mutations are found at the HR1HR2 postfusion bundle interface and induce a backbone shift of HR2 (Fig. 1C) (14). Here, we show that the N969K mutation alone is sufficient to displace the HR2 backbone in a similar manner to the Omicron BA.1 HR1HR2 postfusion bundle with all three Omicron muta- tions. Based on this finding, we designed an optimized inhibitor peptide, named 42G, to accommodate the distortion of the HR1HR2 postfusion bundle caused by the N969K mutation. The structure of the complex between the Omicron HR1 and the 42G peptide confirms our design—the 42G peptide forms a bulge to accommodate the N969K mutation. Functionally, the 42G pep- tide shows more potent Omicron-specific inhibition in both cell- based fusion and VSV-SARS-CoV-2 chimera infection assays compared to our original inhibitory peptide derived from the Wuhan strain. Results The N969K Mutation Alone Induces a Shift of HR2 in the Omicron HR1HR2 Postfusion Bundle. Our structures of the HR1HR2 postfusion bundles of SARS-CoV-2 and its variants revealed a displacement of the HR2 backbone in the Omicron variant with the three mutations (N969K, Q954H, and L981F) in the HR1 region (PDB ID 7tik) (14). We thus determined the cryo-EM structure of the HR1HR2 bundle with the N969K mutation alone (Fig. 1A and SI Appendix, Fig. S1 and Table S1). The long sidechain of K969 is closely packed with F970 of a neighboring HR1 strand, which limits the possible rotamers of the K969 sidechain (Fig. 1B). The N969K mutant structure revealed a similar HR2 backbone displacement as observed in the Omicron HR1HR2 structure with all three mutations (Fig. 1C)—the average Cα atom rmsd for residues 1,162 to 1,170 is 0.7 Å between the N969K structure and the Omicron structure, 2.3 Å between the N969K structure and the Wuhan structure, and 2.4 Å between the Omicron variant structure and the Wuhan strain structure. This suggests that N969K alone is the major cause for the conformational change of HR2 in the postfusion bundle structure of Omicron. The comparison with the Omicron and Wuhan postfusion bundle structures illustrates structural rearrangement of HR2 to accommodate the long lysine sidechain at position 969 (Fig. 1C). The lysine would clash into the HR2 backbone if it were in the conformation of the Wuhan strain (Fig. 1D). Thus, the N969K mutation results in a major deflection of the HR2 backbone away from its Wuhan variant position to avoid steric clashes (Fig. 1C). Restoring the Interface between HR1 N969K and HR2. Our structures suggest that the inhibition activity against the Omicron variant can be improved if an inhibitory peptide better accommodates the HR1 N969K mutation than the previous wildtype HR2 peptides used. Since the N969K mutation causes the displacement of the HR2 backbone (Fig. 1C), a nearby complementary residue substitution to HR2 that only changes the sidechain is unlikely to restore the local geometry. Upon the inspection of the N696K mutant structure (Fig. 1), we predicted that the insertion of an additional residue in HR2 near the Omicron HR1 K969 residue might better accommodate the K969 mutation, thus promoting a more stable interface between HR1 and HR2. Starting with the previously reported longHR2_42 peptide (13) we chose to insert an additional residue between HR2 residues G1167 and D1168 which are proximal to the Omicron HR1 K969 sidechain (Fig. 2A), anticipating that this insertion would restore inhibition activity. We selected glycine due to its flexibility, hydrophilicity, and proximity to the naturally occurring glycine in HR2 (G1167) (peptide referred to as 42G). To confirm our prediction that 42G better accommodates the Omicron N969K mutation, we determined the cryo-EM structure of the complex of the Omicron HR1 and a slightly longer version of 42G, termed 42Gv2, to an overall resolution of 2.8 Å (Fig. 2B and SI Appendix, Fig. S1 and Table S1). We used the slightly longer A L981 C Wuhan Omicron N969K HR2 HR1 N969K Q954 D G1167 B HR1 F970 HR2 N969K HR1 Wuhan HR2 Clash N969K Omicron HR1 Fig. 1. The N969K mutation alone induces the backbone shift observed in the Omicron HR1HR2 postfusion bundle. (A and B) Cryo-EM map and model of the HR1HR2 postfusion bundle with the N969K mutation (PDB ID 8fa1, EMDB ID 28947, this study). Cyan: carbon atoms of the N969K structure. Blue: nitrogen atoms. Red: oxygen atoms. Only one HR2 protomer and its two neighboring HR1 protomers are shown for clarity, and the displayed region of the bundle is colored cyan in the lower right inset. The dashed vertical lines indicate the slab that is displayed in panel B. (C) Superposition of the Cα ribbons of the postfusion bundle structures of the Wuhan strain (yellow, PDB ID 8czi, EMDB ID 27098, published in ref. 13), of the Omicron variant (magenta, PDB ID 7tik, EMDB ID 25912, published in ref. 14), and of the N969K mutant (cyan, PDB ID 8fa1, EMDB ID 28947, this study), viewed in the same orientation as that of panel A. Only one HR2 protomer and one neighboring HR1 protomer are shown for clarity. Gray arrows connecting the Cα atoms of Wuhan strain and Omicron variant HR2 backbones illustrate the backbone shift. (D) N969K in Omicron HR1 would clash with G1167 in the Wuhan strain conformation of HR2 when superimposing the two postfusion bundles. The carbon atoms are colored the same as that in panels A and C, i.e., yellow for Wuhan and magenta for Omicron, while the nitrogen and oxygen atoms are colored as blue and red, respectively. The clashes are shown as disks (red) whose radii are proportional to the severity of the clashes. 2 of 7 https://doi.org/10.1073/pnas.2300360120 pnas.org Fig. 2. The glycine insertion of the 42G peptide alleviates the HR2 backbone shift induced by the N969K mutation in HR1. (A) Primary amino acid sequences of the longHR2_42, 42G, longHR2_45, and 42Gv2 peptides. Residue G1167 is in bold, and the inserted glycine residues is in bold and underlined. (B and C) Cryo-EM map and structure of the complex of Omicron HR1 and the 42Gv2 peptide (PDB ID 8fa2, EMDB ID 28948, this study). Green: carbon atoms. Blue: nitrogen atoms. Red: oxygen atoms. Only one 42Gv2 protomer and its two neighboring HR1 protomers are shown for clarity, and the displayed region of the bundle is colored green in the lower right inset. The dashed vertical lines indicate the slab that is displayed in panel C. (D) Close-up view around the mutated N969K residue and its interactions with F970 and the inserted glycine of 42G. The map and structure are rotated 30°, compared to that shown in panel A. Backbone atoms of 42G are shown, and the hydrogen bond between the Z nitrogen of N969K and the carbonyl oxygen of the inserted glycine residue is shown as a dashed line. (E) Superposition of the Cα ribbons of the postfusion HR1HR2 bundles of the Omicron HR1—42Gv2 complex shown in panel B, of the Wuhan strain, and of the Omicron variant. Only one 42Gv2 protomer and one neighboring HR1 protomer are shown for clarity. Gray arrows connecting the Cα atoms of Omicron HR2 backbone and of the 42Gv2 illustrate that the displaced backbone of Omicron is partially restored to the Wuhan strain conformation. (F) Graphic summary of the effect of the Omicron N969K mutation on the nearby HR2 region and accommodation of the mutation by the inserted glycine residue in the 42G peptide. 42Gv2 construct for structure determination since it has the same residue range as the longHR2_45 peptide (13) used for structure determination of the Wuhan, Omicron, and the N969K postfu- sion bundles (Fig. 1). The structure of the 42Gv2—Omicron HR1 complex maintains the overall six-helix-bundle architecture of the SARS-CoV-2 postfusion bundle and the packing between K969 and F970 (Fig. 2C). In addition, the structure features a bulge of the 42Gv2 backbone at the inserted glycine residue (Fig. 2D). This bulge results from the deviation of the 42Gv2 backbone around the long sidechain of the lysine residue of the mutated HR1 N969K, promoting closer overall packing with HR1; additionally, the terminal nitrogen of this lysine and the carbonyl oxygen of the inserted G form a hydrogen bond, presumably contributing addi- tional binding energy to the interface (Fig. 2D). Additionally, the insertion reduces the overall displacement of HR2 in this region, partially restoring the HR2 backbone to the original position in the Wuhan strain structure (Fig. 2 E and F and SI Appendix, Fig. S2 and Movie S1). The average Cα atom rmsd for HR2 res- idues 1,162 to 1,170 is reduced from 2.4 Å (between the Wuhan strain HR1HR2 structure and the Omicron HR1HR2 structure) to 1.5 Å (between the Wuhan strain HR1HR2 structure and the Omicron HR1—42Gv2 structure). Our structure of the Omicron HR1—42Gv2 complex shows that the glycine insertion restores the backbone interactions on both sides of the inserted glycine and Omicron HR1 (Fig. 2 E and F and SI Appendix, Fig. S2 and Movie S1). Thus, the N-terminal extension (residues 1,162 to 1,167), previously found to be critical for the peptide’s potent inhibition against infection (13), interacts with HR1 in a similar manner as observed in the Wuhan structure. Moreover, these results suggest that the interaction between the N-terminal exten- sion and HR1 is sequence-specific. Omicron-Specific Inhibition by 42G in a Cell–Cell Fusion Assay. To test whether the glycine insertion in the 42G peptide restores inhibition activity in a cell–cell membrane fusion assay using S protein and ACE2, we used the Omicron triple mutant (hereafter referred to as “Omicron3M”) that was used in our previous study (14). Omicron3M contains only the three mutations in the HR1 region, but not mutations in other regions of Omicron S. Compared to the original longHR2_42, the 42G peptide has an ~fivefold reduced inhibition activity against fusion with the Wuhan strain (Fig. 3A) but has an ~threefold enhanced inhibition activity against fusion with the Omicron3M, recovering the loss of inhibition activity of the original longHR2_42 peptide conferred by Omicron (Fig. 3B). At the same time, 42G still inhibits fusion with Wuhan strain S protein, albeit less so than the original longHR2_42 PNAS 2023 Vol. 120 No. 13 e2300360120 https://doi.org/10.1073/pnas.2300360120 3 of 7 A i n o s u f d e z i l a m r o N 120 100 80 60 40 20 0 ) l o r t n o c f o % ( Wuhan Inhibitor IC50 (nM) longHR2_42 1.3 ± 0.33 42G 7.0 ± 1.54 B i n o s u f d e z i l a m r o N 120 100 80 60 40 20 0 ) l o r t n o c f o % ( Omicron3M Inhibitor IC50 (nM) longHR2_42 4.7 ± 0.66 42G 1.7 ± 0.29 0 10-2 10-1 100 101 102 103 104 Peptide Concentration (nM) 0 10-2 10-1 100 101 102 103 104 Peptide Concentration (nM) Fig. 3. The optimized 42G peptide improves the inhibition of Omicron in a cell-cell membrane fusion assay. Inhibition activities by 42G against Wuhan (A) or Omicron3M (B) S protein. For 42G, the raw data points are plotted as black circles, while the error bars (SD), fitted curves, and vertical dashed lines at IC50 are plotted in blue color. For longHR2_42, the raw data are plotted as gray triangles with gray error bars (SD), while the fitted curves, and vertical dashed lines at IC50 are colored black. The data for longHR2_42 in A were previously published in ref. 13. Details about the number of repeats, calculation of means, fitting, and calculation of error of the fit are in the Materials and Methods. peptide. Thus, insertion of a glycine residue in the peptide enhances its Omicron-specific inhibition activity in our cell–cell fusion assay. Omicron-Specific Inhibition by 42G in a VSV-SARS-CoV-2 Chimera Infection Assay. We next tested the 42G peptide for variant-specific inhibition activity using a VSV-SARS-CoV-2 chimera infection assay, which has been previously shown to strongly recapitulate the entry route of natural SARS-CoV-2 (13, 16, 17). The chimeric virus was produced by replacing the glycoprotein of VSV with the S protein of Wuhan, Delta, or Omicron strains of SARS-CoV-2. The first round of VSV-SARS-CoV-2 infection was examined in VeroE6 cells overexpressing the transmembrane protease, serine 2 (TMPRSS2) (VeroE6+TMPRSS2) by expression of a soluble cytosolic eGFP reporter encoded in the virus 8 h after inoculation. Compared to the original longHR2_42, the 42G peptide displayed an ~fivefold enhanced inhibition activity against Omicron infection (Fig. 4C) and ~fourfold to sixfold decreased inhibition activities against Wuhan (Fig. 4A) and Delta strains infection (Fig. 4B). These results were consistent with the Omicron-specific enhanced inhibitory activity of 42G observed with the cell–cell fusion assay (Fig. 3). Discussion One striking difference between SARS-CoV-2 Omicron and prior variants is a marked increase in mutational load, including within the S protein. New mutations emerged both at antigenic interfaces under heavy selection pressure, such as the RBD, and at more distant but functionally critical sites, such as the S protein fusion core comprised of the HR1 and HR2 regions. These mutations within the fusion core of Omicron lead to major conformational changes of HR2 in the structure of the HR1HR2 postfusion bun- dle (14) (Fig. 1) and a decreased activity of a potent HR2-based inhibitor against infection (13) (Fig. 4). Here, we sought to recover the activity of the previously reported inhibitor through a rational, structure-based design. First, we iden- tified the Omicron N969K mutation as the primary cause of the structural change in the Omicron HR1HR2 postfusion bundle, the sidechain of which displaces the HR2 backbone in its N-terminal extended region (Fig. 1C). Then, we designed an opti- mized inhibitory peptide, 42G, to accommodate the long lysine sidechain by insertion of a highly flexible glycine residue in A 2 S S R P M T + 6 E o r e V C 2 S S R P M T + 6 E o r e V E 2 S S R P M T + 6 E o r e V : n a h u W 2 - V o C S R A S V S V - - : a t l - e D 2 - V o C S R A S V S V - : i n o r c m O 2 - V o C S R A S V S V - - Inhibitor IC50 (nM) longHR2_42 1.1 ± 0.02 42G 4.0 ± 0.49 Wuhan B s l l e C d e t c e f n I ) l o r t n o c f o % ( 100 80 60 40 20 0 50 µm 10-3 10-2 10-1 100 101 102 103 0 Inhibitor IC50 (nM) longHR2_42 0.9 ± 0.03 42G 5.2 ± 0.49 Delta ) l o r t n o c f o % ( 100 80 60 40 20 0 D s l l e C d e t c e f n I F s l l e C d e t c e f n I 10-3 10-2 10-1 100 101 102 103 0 Inhibitor IC50 (nM) longHR2_42 4.1 ± 0.33 42G 0.8 ± 0.08 Omicron ) l o r t n o c f o % ( 100 80 60 40 20 0 Untreated 1 nM 42G 5 nM 42G 20 nM 42G 0 10-3 10-2 10-1 100 101 102 103 Peptide Concentration (nM) Fig. 4. VSV-SARS-CoV-2 chimera infection assay confirms the Omicron-specific inhibition by the 42G peptide. Inhibition of VSV-SARS-CoV-2 Wuhan (A and B), VSV-SARS-CoV-2 Delta (C and D), or VSV-SARS-CoV-2 Omicron (E and F) infection by 42G in VeroE6+TMPRSS2 cells. Virus and peptide were incubated with cells for 1 h, washed, then fixed and imaged 8 h after initiation of infection allowing for one round of infection to occur. Images are maximum intensity projections of 20 µm z-planes taken with 1 µm spacing using a spinning disk-confocal microscopy (Materials and Methods). Expression of a soluble eGFP (green) reporter allowed for infected cells to be determined while cell outlines were obtained from WGA-Alexa647 stain applied immediately prior to fixation (A, C, and E). In panels B and D, and F, the plots are colored the same as Fig. 3. The data for longHR2_42 in B, D, and F were previously published in ref. 13. Details about number of repeats, calculation of means, fitting, and calculation of SE are in the Materials and Methods. 4 of 7 https://doi.org/10.1073/pnas.2300360120 pnas.org proximity to the lysine sidechain of Omicron HR1 (Fig. 2). The high-resolution structure of a slightly longer version of 42G, 42Gv2, bound to Omicron HR1 shows that the glycine insertion allows the HR2 backbone to bypass the lysine residue in the mutated HR1 N969K and remain closer to the Wuhan strain conformation of HR2 with minimal disruption of backbone geom- etry (Fig. 2 and SI Appendix, Fig. S2 and Movie S1). This glycine insertion allows for the formation of a bulge perpendicular to the bundle (Fig. 2D), permitting the HR2 residues on both sides of the mutated lysine residue (HR1 N969K) to adapt a conformation closer to that observed in the Wuhan strain (Fig. 2E). This improve- ment of 42G interactions with Omicron HR1 translates to increased inhibition activity for the Omicron variant as observed in both cell–cell fusion and VSV-SARS-CoV-2 chimera infection assays (Figs. 3 and 4). 42G still inhibits Wuhan and Delta strain infection, albeit less so than the original longHR2_42 peptide which was based on the Wuhan strain sequence. These observations suggest that sequence-specific interactions between HR2 (1,162 to 1,167) and HR1 (969 to 981) residues account for the improved inhibition activity of 42G against Omicron. Functionally, these interactions may mediate the initial landing of HR2 onto HR1 during the transition of the S protein from the prehairpin intermediate to the postfusion state (13), assuming the zippering of the HR1HR2 bundle begins from the N terminus of HR2. Given our observation that the HR1 N969K mutation destabi- lizes the postfusion conformation of S, it is perhaps surprising that all existing Omicron subvariants retain the N969K mutation. Recently, McCallium et al. (18) showed that the Omicron N969K and L981F mutations stabilize the prefusion conformation through interprotomer electrostatic interactions and intraprotomer hydro- phobic packing, respectively. However, the L981F mutation is prob- ably not required for tolerating the destabilization of the postfusion S by the N969K mutation, since several Omicron subvariants do not have the L981F mutation. It is unclear if the Omicron Q954H mutation has a functional role, although we previously showed that the hydrogen bond between the HR1 residue 954 and the HR2 residue S1175 is maintained for the Q954H mutation (14). In this study we used two different cell lines—HEK and Vero cells. The ACE2 expression level is very low in HEK cells (19); hence, their ectopic expression is required for efficient fusion. A previous study with a HEK cell–cell fusion assay (20) suggests that the S protein is activated by host cell proteases. Vero cells, which robustly express ACE2, are infected by SARS-CoV-2 upon proteolytic cleavage of the spike protein mediated by endosomal cathepsins (21). Ectopic expression of TMPRSS2 in Vero cells provides a complementary second proteolytic entry route. By using the above two cell systems, we were able to establish that the mode of HR2 peptide inhibition was the same regardless of the entry route used by the virus. While we focused our studies on the BA.1 subvariant, we expect that the Omicron-specific enhancement of the inhibition activity of 42G should also apply to other recently emerged Omicron BQ.1, BQ.1.1, BA.4.6, BF.7, BA.2.75.2, and XBB.1.5 subvariants (6, 7) since they all retain the N969K mutation. Our study shows that variant-specific inhibitors can be rationally designed in a rapid manner using structures of the postfusion bundles to examine the resistance caused by certain mutations. This will expand the tool- box for the development of variant-specific inhibitors to combat future variants and subvariants of SARS-CoV-2. We envision efficient delivery of peptide inhibitors to the nasopharyngeal cav- ities and lungs by inhalation in order to achieve targeted delivery. Thus, we anticipate that our nanomolar peptide inhibitors will be excellent leads for clinical drug development. Materials and Methods Structure Determination. The cryo-EM structures of the N969K HR1HR2 post- fusion bundle and that of the Omicron HR1—42Gv2 complex were determined following a molecular scaffolding method described previously (14). Briefly, the scaffolded complexes were generated by coexpressing the scaffolded HR1 and Small Ubiquitin-like Modifier (SUMO)-tagged HR2 peptides in E. coli BL21(DE3) using auto-inducing lysogeny broth (LB) medium (22), followed by nickel affinity chromatography and size exclusion chromatography (SEC) with a Superose 6 Increase 10/300 GL column in 25 mM Hepes-Na, pH 7.4, 150 mM NaCl, 0.5 mM EDTA, 0.5 mM Tris(2-carboxyethyl)phosphine (TCEP). The sample was concen- trated to 20 µM, supplemented with 0.05% Nonidet P-40, and plunge frozen on a Quantifoil 2/1 holey carbon grid using a Vitrobot Mark IV (Thermo Fisher). Movie stacks were recorded using a Titan Krios transmission electron microscope (Thermo Fisher) equipped with a K3 camera (Gatan) using the Serial-EM auto- mation software (23), at a nominal magnification of 130,000× and a pixel size of ~0.3 Å. Each movie stack contained ~40 frames with a total electron dose of ~55 e−/Å2. The data were processed using a combination of MotionCor2 (24), Gctf (25), EMAN2 (26), cryoSPARC (27), and RELION (28), as described previously (14). More details for data collection and processing are summarized in SI Appendix, Fig. S1 and Table S1. For model building, we first slightly improved the structure of Omicron HR1HR2 (PDB 7tik) by better fitting the sidechain of K969 into the correspond- ing density map (updated statistics are provided in Table S1, and the PDB entry has also been updated). We used this improved model of Omicron HR1HR2 as the initial template for the N969K and Omicron HR1—42Gv2 structures. The model near the glycine insertion was first built in Coot (29) and then refined by an automated structure refinement protocol with Rosetta (30). The structure was then subjected to real space refinement (global minimization, local grid search, adp) in PHENIX (31). Coot (29) was used for further fitting of sidechains and manual inspection. Peptide Synthesis and Characterization. The 42G peptide was synthesized by GenScript USA Inc. HPLC and liquid chromatography–mass spectrometry profiles are shown in SI Appendix, Fig. S3 (provided by the manufacturer). We performed SEC and size exclusion chromatography coupled with mul- ti-angle light scattering (SEC-MALS) to further characterize the 42G peptide in PBS buffer. Peptide powder was first dissolved in dimethyl sulfoxide (DMSO) to ~5 mg/mL. Subsequently, DMSO was exchanged to PBS buffer by three rounds of dilution and concentration. The dilution factor was ~15 for each round, and the centrifugal concentrator (Merck Millipore Ltd.) had a molecular weight cutoff at 3 kDa. The peptide solution was filtered by 0.22 µm PVDF membrane. SEC and SEC-MALS were performed in PBS buffer using a Superdex 75 10 300 GL column and a wtc-010S5 column (Wyatt Technology Corporation), respectively. The SEC and SEC-MALS profiles are shown in SI Appendix, Fig. S3. The concentration of the peptide stock solution was determined by absorption measurement at 205 nm using a Nanodrop instrument (Thermo Fisher). HEK Cell–Cell Fusion Assay. We optimized the cell-cell fusion assay (20) based on the α-complementation of E. coli β-galactosidase for comparing the inhibi- tory activity of different peptides with higher throughput. Suspension culture Expi293F cells (Thermo Fisher, Cat.# A14527), a clonal derivative of Human Embryonic Kidney (HEK) 293 cells were grown to a density of 1 ∼ 2 × 106 cells/ mL in FreeStyle 293 expression medium (Thermo Fisher, Cat.# 12338026) sup- plemented with 0.1 mg/mL penicillin-streptomycin antibiotics. The cells were then pelleted, resuspended in medium without antibiotics to a density of 1 × 106 cells/mL, and allowed to recover at 37 °C for 30 min. One group of cells was then cotransfected using polyethyleneimine (PEI, Sigma) (125 μg PEI/mL cells) with Wuhan strain full-length SARS-CoV2 S protein construct (12.5 μg DNA/mL cells) and the α-fragment of E. coli β-galactosidase construct (12.5 μg DNA/mL cells) to generate the S protein-expressing cells. Since HEK cells do not express TMPRSS2, it is likely that S is cleaved by host cell proteases that render S fusogenic upon binding to its receptor ACE2 (20). Using the same amount of PEI, the other group of cells was cotransfected with the full-length ACE2 (12.5 μg DNA/mL cells) con- struct and the ω-fragment of E. coli β-galactosidase construct (12.5 μg DNA/mL cells) to generate the ACE2 receptor-expressing cells. Note that the endogenous ACE2 level is very low in HEK cells (19), requiring ectopic expression for efficient fusion. As a negative control, two additional groups of cells were transfected with PNAS 2023 Vol. 120 No. 13 e2300360120 https://doi.org/10.1073/pnas.2300360120 5 of 7 either the α-fragment or the ω-fragment of E. coli β-galactosidase construct alone. After incubation of the cells in flasks at 37 °C for 24 h, the cells were pelleted. The S-expressing cells were resuspended in FreeStyle 293 expression medium supplemented with different concentrations of peptide (50 μL, 2 × 106 cells/mL), respectively. The ACE2-expressing cells and negative control cells were resus- pended in 50 μL fresh FreeStyle 293 expression medium (pH 7.4) to be 2 × 106 cells/mL S-expressing, and ACE2 cells or α-fragment and ω-fragment cells were then mixed in a 96-well plate (Greiner Bio-One) to initiate cell-cell fusion at 37 °C for 2 h. Fusion was arrested by adding 100 μL β-galactosidase substrate from the Gal-Screen reporter system (Invitrogen). The mixture was incubated at 37 °C in the dark for 1 h before recording luminescence using a Tecan Infinite M1000. Purification of VSV-SARS-CoV-2 Chimeras. Recombinant VSV chimeras with glycoprotein G replaced with the SARS-CoV-2 S protein with the sequence of the Wuhan-Hu-1 strain, D614G mutation in the Wuhan-Hu-1 strain, Delta strain, or Omicron strain of SARS-CoV-2 (VSV-SARS-CoV-2) and expressing a soluble eGFP infec- tion reporter were generated as described previously (16, 17, 21). VSV-SARS-CoV-2 was grown by infecting 12 to 18 150 mm dishes of MA104 cells at a multiplicity of infection (MOI) of 0.01. Supernatant was collected at 48 h post-infection. The super- natant was clarified by low-speed centrifugation at 1,000 × g for 10 min at 4 °C. Virus and extracellular particles were pelleted by centrifugation in a Ti45 fixed-angle rotor at 30,000 × g for 2 h at 4 °C. The pellet was resuspended in NTE buffer (100 mM NaCl, 10 mM Tris⋅HCl pH 7.4, 1 mM EDTA) at 4 °C. The resuspended pellet was layered on top of a 15% (v/v) sucrose-NTE solution and centrifuged in an SW55 swinging-bucket rotor at 110,000 × g for 2 h at 4 °C. The virus was resuspended in NTE overnight at 4 °C, then separated on a 15 to 45% (v/v) sucrose-NTE linear gradient by ultracentrifugation in an SW55 swinging-bucket rotor at 150,000 × g for 1.5 h at 4 °C. The predominant band in the lower one-third of the gradient was then extracted by side puncture of the centrifuge tube. Virus was then diluted in NTE and concentrated by ultracentrifugation in a Ti60 fixed-angle rotor at 115,000 × g for 2 h at 4 °C. The VSV-SARS-CoV-2 containing pellet was resuspended overnight in NTE in a volume of 0.5 mL and stored at 4 °C for subsequent experiments. VSV-SARS-CoV-2 Infection Assay. Glass slides (18 mm) were cleaned, over- layed with 3-mm polydimethylsiloxane (PDMS) wells and sterilized as previously described (13, 21). On the day prior to the experiment, VeroE6 cells overexpressing TMPRSS2 (Vero+TMPRSS2) were plated inside the PDMS wells on top of the glass slide; the slide was placed in a six-well plate and grown at 37 °C in the presence of 10% CO2 to achieve 70 to 80% confluence on the day of the experiment. Note that ACE2 expression is very high in VeroE6 cells (32). On the day of experiments, medium was removed, virus was diluted into media containing the desired con- centration of indicated peptide at a final VSV-SARS-CoV-2 concentration of 0.5 µg/ mL viral RNA (equivalent to an MOI of ∼0.5 infectious units [IFU] also determined in Vero+TMPRSS2 cells), and then immediately added to the desired PDMS well in a volume of 10 µL for 1 h at 37 °C in the presence of 10% CO2. Liquid was placed outside of the PDMS well to maintain humidity and reduce evaporation. Cells were then washed twice with medium to remove unbound virus and peptide inhibitor; the well were then filled with fresh medium. In all experiments, cells were kept at 37 °C with 10% CO2, and the medium was prewarmed to 37 °C. At 8 h post-infection, the medium was removed; cells were stained with 5 µg/mL wheat germ agglutin (WGA)-Alexa647 in PBS for 30 s at room temperature. Cells were then washed twice with PBS, fixed with 4% paraformaldehyde in PBS for 15 min, then washed three times with PBS and imaged using a spinning-disk confocal microscope. Imaging was done with a 40× oil objective and an EMCCD camera with a pixel size of 0.33 µm; volume data was obtained by acquiring 20 consecutive planes spaced 1 µm apart for every field of view (33). Cells were considered infected when they displayed a cytosolic eGFP fluorescence signal with a relative intensity at least 1.4 times that of uninfected cells. Example images are maximum-intensity projections highlighting the cell outline marked with the WGA-Alexa647 membrane label. 1. M. Dhawan et al., Omicron variant (B.1.1.529) and its sublineages: What do we know so far amid the emergence of recombinant variants of SARS-CoV-2? Biomed. Pharmacother. 154, 113522 (2022). Y. Cao et al., BA.2.12.1, BA.4 and BA.5 escape antibodies elicited by Omicron infection. Nature 608, 593–602 (2022). P. Qu et al., Evasion of neutralizing antibody responses by the SARS-CoV-2 BA.2.75 variant. Cell Host. Microbe. 30, 1526.e4 (2022), 10.1016/j.chom.2022.09.015. 2. 3. Statistics and Data Analysis. Data from the human embryonic kidney (HEK) cell-cell membrane fusion and VSV-SARS-CoV-2 infection assays from three inde- pendent biological replicates determined for each concentration of inhibitor used. For the HEK cell-cell membrane fusion assay, normalized fusion was calculated as (Luminescence(+inhibitor) − Luminescence(α&ω))/(Luminescence(+PBS) − Luminescence(α&ω)), where “+inhibitor” or “+PBS” refers to adding inhibitor or PBS to the mixture of the cells expressing the α-fragment of E. coli β-galactosidase and S, and the cells expressing the ω-fragment of E. coli β-galactosidase and ACE2, and “α&ω” refers to the mixture of the cells expressing the α-fragment only and the cells expressing the ω-fragment only. The infected cells counted in the VSV-SARS-CoV-2 infection assay were nor- malized by that of control HR2. After the normalization, the arithmetic means of the three replicates were used to fit the inhibition curves to obtain estimates of the IC50 values; the estimates were obtained by nonlinear regression of inhibitor concentration vs. response in GraphPad Prism version 9.1.0 for macOS (GraphPad Software, San Diego, CA, https://www.graphpad.com). The fitted model is Y = Bottom + (Top − Bottom)/(1 + (IC50/X)HillSlope), where Y is the extent of inhibition, X is the inhibitor concentration, Bottom and Top are the minimal and maximal inhibition. The SE of the IC50 estimation was calculated using OriginPro 9.1 (OriginLab Corporation). Figure Preparation. The figures of PDB structures and maps were made using UCSF Chimera (34) and PyMOL (The PyMOL Molecular Graphics System, Version 2.5, Schrödinger, LLC). Chains B, C, and E were chosen to display HR1 and HR2 of the Wuhan and Omicron structures. Chains A, C, and F were chosen to display HR1 and HR2 of the N969K and 42G structures. The data fitting of all inhibition assays was performed and plotted using GraphPad Prism version 9.1.0 for macOS (GraphPad Software, San Diego, CA, https://www.graphpad.com). Data, Materials, and Software Availability. The EM maps and corresponding structures reported here have been deposited in EMDB (Electron Microscopy Data Bank) and PDB with the following accession IDs: HR1HR2 N969K: EMDB 28947 (35), PDB 8fa1 (36); and Omicron HR1—42Gv2: EMDB 28948 (37), PDB 8fa2 (38). ACKNOWLEDGMENTS. We thank John Kuriyan, Subu Subramanian, Serena Muratcioglu, Timothy J. Eisen, Kendra Marcus, Catherine A. Doyle, and Sondra Schlessinger for stimulating discussions, Bing Chen for kindly providing plasmids and protocols for the cell-cell membrane fusion assay, the Sarafan ChEM-H High-Throughput Screening Knowledge Center for providing the Tecan microplate reader, and the Stanford Cryo-Electron Microscopy Center (cEMc) and the Stanford-SLAC Cryo-EM Center (S2C2) for support, as well as support by the Harvard Virology Program NIH training grant T32 AI07245 postdoctoral fellow- ship to A.J.B.K., NIH Maximizing Investigators’ Research Award GM130386 to T.K., NIH Grant AI163019 to T.K. (and Sean P.J. Whelan), funding from Danish Technical University to T.K., funding from Sana Biotechnology to T.K., and funding from HHMI to A.T.B. This article is subject to HHMI’s Open Access to Publications policy. HHMI laboratory heads have previously granted a nonexclusive CC BY 4.0 license to the public and a sublicensable license to HHMI in their research articles. Pursuant to those licenses, the accepted manuscript of this article can be made freely available under a CC BY 4.0 license immediately upon publication. Author affiliations: aHHMI, Stanford University, Stanford, CA 94305; bDepartment of Molecular and Cellular Physiology, Stanford University, Stanford, CA 94305; cDepartment of Neurology and Neurological Sciences, Stanford University, Stanford, CA 94305; dDepartment of Structural Biology, Stanford University, Stanford, CA 94305; eDepartment of Photon Science, Stanford University, Stanford, CA 94305; fProgram in Cellular and Molecular Medicine, Boston Children’s Hospital, Boston, MA 02115; gDepartment of Pediatrics, Harvard Medical School, Boston, MA 02115; and hDepartment of Cell Biology, Harvard Medical School, Boston, MA 02115 4. 5. 6. E. Callaway, New Omicron-specific vaccines offer similar protection to existing boosters. Nature 609, 232–233 (2022). P. Arora et al., Omicron sublineage BQ.1.1 resistance to monoclonal antibodies. Lancet Infect. Dis. 23, 22–23 (2022), 10.1016/S1473-3099(22)00733-2. P. Qu, Distinct neutralizing antibody escape of SARS-CoV-2 Omicron subvariants BQ.1, BQ.1.1, BA.4.6, BF.7 and BA.2.75.2. bioRxiv [Preprint] (2022), 10.1101/2022.10.19.512891 (Accessed 20 October 2022). 6 of 7 https://doi.org/10.1073/pnas.2300360120 pnas.org 7. 8. 9. C. Yue, Enhanced transmissibility of XBB.1.5 is contributed by both strong ACE2 binding and antibody evasion. bioRxiv [Preprint] (2023), 10.1101/2023.01.03.522427 (Accessed 5 January 2023). A. Ianevski et al., Mono- and combinational drug therapies for global viral pandemic preparedness. iScience 25, 104112 (2022). J. M. White et al., Drug combinations as a first line of defense against Coronaviruses and other emerging viruses. mBio 12, e0334721 (2021). 10. B. J. Bosch et al., Severe acute respiratory syndrome coronavirus (SARS-CoV) infection inhibition using spike protein heptad repeat-derived peptides. Proc. Natl. Acad. Sci. U.S.A. 101, 8455–8460 (2004). 11. S. Liu et al., Interaction between heptad repeat 1 and 2 regions in spike protein of SARS-associated coronavirus: implications for virus fusogenic mechanism and identification of fusion inhibitors. Lancet 363, 938–947 (2004). 12. S. C. Harrison, Viral membrane fusion. Virology 479–480, 498–507 (2015). 13. K. Yang et al., Nanomolar inhibition of SARS-CoV-2 infection by an unmodified peptide targeting the prehairpin intermediate of the spike protein. Proc. Natl. Acad. Sci. U.S.A. 119, e2210990119 (2022). 23. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of specimen movements. J. Struct. Biol. 152, 36–51 (2005). 24. S. Q. Zheng et al., MotionCor2: Anisotropic correction of beam-induced motion for improved cryo- electron microscopy. Nat. Methods 14, 331–332 (2017). 25. K. Zhang, Gctf: Real-time CTF determination and correction. J. Struct. Biol. 193, 1–12 (2016). 26. G. Tang et al., EMAN2: An extensible image processing suite for electron microscopy. J. Struct. Biol. 157, 38–46 (2007). 27. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: Algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017). 28. S. H. Scheres, RELION: Implementation of a Bayesian approach to cryo-EM structure determination. J. Struct. Biol. 180, 519–530 (2012). 29. P. Emsley, K. Cowtan, Coot: Model-building tools for molecular graphics. Acta. Crystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004). 30. R. Y. Wang et al., Automated structure refinement of macromolecular assemblies from cryo-EM maps using Rosetta. Elife 5, e17219 (2016). 31. P. D. Adams et al., PHENIX: A comprehensive Python-based system for macromolecular structure 14. K. Yang et al., Structural conservation among variants of the SARS-CoV-2 spike postfusion bundle. solution. Acta. Crystallogr. D Biol. Crystallogr. 66, 213–221 (2010). Proc. Natl. Acad. Sci. U.S.A. 119, e2119467119 (2022). 15. R. T. Eguia et al., A human coronavirus evolves antigenically to escape antibody immunity. PLoS Pathog. 17, e1009453 (2021). 16. J. B. Case et al., Neutralizing antibody and soluble ACE2 inhibition of a replication-competent VSV- SARS-CoV-2 and a clinical isolate of SARS-CoV-2. Cell Host. Microbe. 28, 475–485.e475 (2020). 17. A. J. B. Kreutzberger et al., SARS-CoV-2 requires acidic pH to infect cells. Proc. Natl. Acad. Sci. U.S.A. 119, e2209514119 (2022). 32. X. Ren et al., Analysis of ACE2 in polarized epithelial cells: Surface expression and function as receptor for severe acute respiratory syndrome-associated coronavirus. J. Gen. Virol. 87, 1691–1695 (2006). 33. E. Cocucci, F. Aguet, S. Boulant, T. Kirchhausen, The first five seconds in the life of a clathrin-coated pit. Cell 150, 495–507 (2012). 34. E. F. Pettersen et al., UCSF Chimera–a visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605–1612 (2004). 18. M. McCallum et al., Structural basis of SARS-CoV-2 Omicron immune evasion and receptor 35. K. Yang, A. T. Brunger, Cryo-EM structure of the SARS-CoV-2 HR1HR2 fusion core complex with engagement. Science 375, 864–868 (2022). 19. L. J. Partridge et al., ACE2-independent interaction of SARS-CoV-2 spike protein with human epithelial cells is inhibited by unfractionated Heparin. Cells 10, 1419 (2021). 20. Y. Cai et al., Distinct conformational states of SARS-CoV-2 spike protein. Science 369, 1586–1592 (2020). 21. A. J. B. Kreutzberger et al., Synergistic block of SARS-CoV-2 infection by combined drug inhibition of the host entry factors PIKfyve kinase and TMPRSS2 protease. J. Virol. 95, e0097521 (2021). 22. F. W. Studier, Protein production by auto-induction in high density shaking cultures. Protein. Expr. Purif. 41, 207–234 (2005). N969K mutation. Electron Microscopy Data Bank. https://www.emdataresource.org/EMD-28947. Deposited 25 November 2022. 36. K. Yang, A. T. Brunger, Cryo-EM structure of the SARS-CoV-2 HR1HR2 fusion core complex with N969K mutation. Protein Data Bank. https://www.rcsb.org/structure/8FA1. Deposited 25 November 2022. 37. K. Yang, A. T. Brunger, Cryo-EM structure of the SARS-CoV-2 Omicron HR1 bound with 42Gv2. Electron Microscopy Data Bank. https://www.emdataresource.org/EMD-28948. Deposited 25 November 2022. 38. K. Yang, A. T. Brunger, Cryo-EM structure of the SARS-CoV-2 Omicron HR1 bound with 42Gv2. Protein Data Bank. https://www.rcsb.org/structure/8FA2. Deposited 25 November 2022. PNAS 2023 Vol. 120 No. 13 e2300360120 https://doi.org/10.1073/pnas.2300360120 7 of 7
10.1093_gbe_evad119	GBE Rapid Evolution of Glycan Recognition Receptors Reveals an Axis of Host–Microbe Arms Races beyond Canonical Protein–Protein Interfaces Zoë A. Hilbert Ellen M. Leffler 1,2,, Paige E. Haffener1, Hannah J. Young1,2, Mara J.W. Schwiesow1,2, 1, and Nels C. Elde1,2, 1Department of Human Genetics, University of Utah, Salt Lake City, Utah, USA 2Howard Hughes Medical Institute, University of Utah School of Medicine, Salt Lake City, UT, USA Corresponding authors: E-mails: [email protected], [email protected]. Accepted: 23 June 2023 Abstract Detection of microbial pathogens is a primary function of many mammalian immune proteins. This is accomplished through the recognition of diverse microbial-produced macromolecules including proteins, nucleic acids, and carbohydrates. Pathogens subvert host defenses by rapidly changing these structures to avoid detection, placing strong selective pressures on host immune proteins that repeatedly adapt to remain effective. Signatures of rapid evolution have been identified in nu- merous immunity proteins involved in the detection of pathogenic protein substrates, but whether similar signals can be ob- served in host proteins engaged in interactions with other types of pathogen-derived molecules has received less attention. This focus on protein–protein interfaces has largely obscured the study of fungi as contributors to host–pathogen conflicts, despite their importance as a formidable class of vertebrate pathogens. Here, we provide evidence that mammalian immune receptors involved in the detection of microbial glycans have been subject to recurrent positive selection. We find that rapidly evolving sites in these genes cluster in key functional domains involved in carbohydrate recognition. Further, we identify con- vergent patterns of substitution and evidence for balancing selection in one particular gene, MelLec, which plays a critical role in controlling invasive fungal disease. Our results also highlight the power of evolutionary analyses to reveal uncharacterized interfaces of host–pathogen conflict by identifying genes, like CLEC12A, with strong signals of positive selection across mam- malian lineages. These results suggest that the realm of interfaces shaped by host–microbe conflicts extends beyond the world of host–viral protein–protein interactions and into the world of microbial glycans and fungi. Key words: host–pathogen interactions, evolutionary conflict, rapid evolution, balancing selection, pattern recognition receptor, microbial glycans. Significance The impact of host–pathogen conflicts in driving evolutionary innovation in mammalian immune proteins is well docu- mented; however, the role of nonprotein components of microbial pathogens in contributing to such evolutionary pro- cesses is not well understood. We identify widespread signals of adaptive evolution in mammalian immune receptors that engage largely with carbohydrate components that decorate the outer surfaces of diverse microbial pathogens, from viruses to fungi. Further, we demonstrate how interactions involving nonproteinaceous components of microbes have driven evolutionary change in mammalian genes across multiple timescales, including evidence for balancing se- lection in a fungal melanin receptor gene in many human populations. Collectively, these findings extend the realm of host–microbe evolutionary conflicts beyond traditionally studied protein–protein interfaces and demonstrate the im- pressively broad impact microbes have on the evolution of their animal hosts. © The Author(s) 2023. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 1 Hilbert et al. GBE Introduction Recognition of microbial pathogens by mammalian immune proteins is essential for activation of protective immune re- sponses and organismal survival. Pattern recognition recep- tors (PRRs) encompass a diverse group of host proteins which are integral in detecting microbial pathogens as for- eign invaders through recognition of unique molecular features (Medzhitov 2007; Kawai and Akira 2008; Vance et al. 2009; Tan et al. 2015). These pathogen-associated molecular patterns are similarly as diverse as the receptors that they engage with and range from proteins, like bac- terial flagellins, to nucleic acids, to complex carbohy- drates, or glycans. Microbial glycans are a defining feature of the cell walls of bacteria and fungi and decorate the outer membranes and surfaces of parasites, whereas glycosylation of coat and sur- face proteins is also well documented in many viruses (Nyame et al. 2004; Comstock and Kasper 2006; van Kooyk and Rabinovich 2008; Raman et al. 2016; Gow et al. 2017). Glycan-recognizing PRRs include, among others, a subset of the Toll-like receptors (TLRs) as well as many members of the calcium-binding C-type lectin receptor (CLR) family. Although the specific glycans recognized by some of these PRRs are known—such as Dectin1’s affinity for ß-glucans or TLR4’s for lipopolysaccharide—for many of these receptors, the exact molecular patterns on microbial surfaces required for recognition are unclear, as is the extent to which variation of these patterns among different microbial species might af- fect recognition (Poltorak et al. 1998; Brown and Gordon 2001; Herre et al. 2004; Park et al. 2009; Werling et al. 2009). Phylogenetic analysis of immune genes, including PRRs, has revealed them to be among the most rapidly evolving genes in mammalian genomes, reflecting the pace of evolu- tion needed to keep up with constantly shape-shifting patho- gens (George et al. 2011; Daugherty and Malik 2012; Rausell and Telenti 2014; Wang and Han 2021). Studies of rapidly evolving immune genes in mammals have largely focused on genes involved in interactions with pathogen-produced protein factors. Comparative analyses of recurrent rapid evo- lution (or positive selection) on the amino acid level frequently reveal the consequential interaction interfaces between host and pathogen proteins. Related experimental studies show how evolution on both sides of these interactions can have functional implications for both host and pathogen (Sawyer et al. 2005; Elde et al. 2009; Mitchell et al. 2012; Barber and Elde 2014; Tenthorey et al. 2020; Carey et al. 2021). These studies reveal the extent to which microbes can spur diversification and evolutionary innovation in the hosts they infect. However, detection of these host–pathogen “arms races” has so far been primarily limited to protein–protein in- terfaces involving viruses and bacteria, even though engage- ment between hosts and infectious microbes involves a wide variety of biological macromolecules and species. Fungi, in particular, represent a major class of human pathogens which are currently auspiciously absent from stud- ies of host–pathogen evolutionary conflict. Systemic fungal infections are associated with severe disease and high mortal- ity rates in human patients and the emergence of multidrug resistant strains has increased dramatically in recent years (Fisher et al. 2022). Beyond human patients, fungal infections pose a severe threat to the health of food crops, and fungal pathogens are currently responsible for massive declines in amphibian and hibernating bat populations world-wide (Fisher et al. 2020). Despite the importance of these patho- gens for the health of evolutionarily diverse organisms, our understanding of the role of host–fungal conflicts in shaping vertebrate immune defenses has been hampered by the rela- tive lack of known protein-based fungal virulence factors. As the first line of defense against recognition by host im- mune factors, diversification in microbial cell wall components and organization has been well documented in bacterial and fungal pathogens (Gow et al. 2017; Imperiali 2019). Further, molecular mimicry of host glycan structures, such as sialic acids, and hijacking of glycosylation pathways has been de- monstrated to be a common mechanism of immune evasion in numerous pathogenic bacteria and viruses (Comstock and Kasper 2006; Vigerust and Shepherd 2007; Carlin et al. 2009; Varki and Gagneux 2012; Raman et al. 2016). Although gly- can hijacking and mimicry in fungi is less well documented, re- ports of sialic acids and sialoglycoconjugates in the cell walls of several fungal species, including the pathogenic species Candida albicans and Cryptococcus neoformans, suggest that fungi may also use methods of molecular mimicry to evade host immune recognition (Rodrigues et al. 1997; Soares et al. 2000; Masuoka 2004). And in fungi, regulated secretion of exopolysaccharide “decoys” correlates with de- creased immune infiltration, suggesting these microbes have developed numerous strategies to prevent their recognition by host immune systems (Denham et al. 2018). Such evasion strategies among microbes suggest the po- tential for selective pressures to exist on immune receptors to be able to maintain the ability to recognize microbial gly- cans and initiate immune responses to control infection. In this study, we identify signatures of positive selection in a set of primarily glycan-recognizing PRRs across three dis- tinct mammalian lineages, suggesting that host–pathogen interfaces involving nonproteinaceous macromolecules may represent a new dimension of host–microbe arms races and can spur evolution in all species involved. Results Signatures of Rapid Evolution Are Pervasive Among Mammalian CLRs and Other Carbohydrate Recognition PRRs To assess whether host genes involved in microbial carbo- hydrate recognition are rapidly evolving in mammals, we 2 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 Rapid Evolution of Glycan Recognition Receptors GBE compiled a list of 26 relevant genes for analysis (fig. 1A and B and Supplementary Material online). These genes were selected based on annotated functions in the recogni- tion of microbial cell walls or other carbohydrate compo- nents of microbial cells. Genes were also prioritized for analysis based on documented expression patterns. Namely, genes expressed by immune cells or on mucosal surfaces were prioritized given their relevance for interac- tions with microbes and defense against infection. More than half of the selected PRR genes contain an an- notated C-type lectin domain (CTLD), including a number of CLR family members with a single CTLD (e.g., Dectin1/ CLEC7A, Langerin/CD207/CLEC4K, Mincle/CLEC4E) as well as the soluble CTLD-containing proteins (MBL2, SP-A, SP-D) and the multiple CTLD-containing mannose re- ceptors (MRC1 and MRC2). Beyond the CLRs and other CTLD-containing proteins, our list also included a putative chitin receptor (FIBCD1), complement receptor 3 (CD11B/ CD18), and TLRs (TLR2 and TLR4). Among this latter group, there have been previous reports of signatures of positive selection in the TLR genes as well as CD11B, which we were able to replicate in this study, while also extending analyses of selection in these genes to additional mamma- lian lineages (Wlasiuk and Nachman 2010; Areal et al. 2011; Liu et al. 2019; Boguslawski et al. 2020; Judd et al. 2021). Finally, we also included in our analyses the CTLDs of three conserved mammalian selectin genes: E-Selectin, L-Selectin, and P-Selectin. These CTLD containing proteins are expressed on a variety of different cell types and act to coordinate cell adhesion and leukocyte trafficking through recognition of “self”-produced carbohydrate li- gands or self-associated molecular patterns (SAMPs) (Varki 2011; Cummings et al. 2022). Given their important role in recognition of these SAMPs on leukocytes and other mammalian cells and no documented role in the recogni- tion of microbes, we hypothesized that the CTLDs from these Selectin genes would not be subject to the same evo- lutionary pressures as other candidate genes involved in dir- ect interactions with infectious microbes. For each of these genes, we obtained orthologous se- quences from publicly available databases for species with- in three distinct mammalian lineages: simian primates, mouse-like rodents (Myomorpha), and bats. Primates were chosen given their relevance to human health, where- as bats and rodents have been implicated as important re- servoirs for many microbial species with zoonotic potential, suggesting that such evolutionary analysis may reveal un- ique patterns of selection among PRRs across these three mammalian lineages (Han et al. 2015; Guth et al. 2022). The orthologous gene sequences within each lineage were aligned and each gene was assessed for signals of re- current positive selection using a combination of different analysis algorithms, including Phylogenetic Analysis by Maximum Likelihood (PAML) and Branch-Site Unrestricted in (BUSTED) for Episodic Diversification Test the Hypothesis Testing using Phylogenies (HyPhy) suite (Pond et al. 2005; Yang 2007; Murrell et al. 2015). Both algo- rithms use the calculation of the ratio of the nonsynon- ymous to synonymous substitution rates (dN/dS) and model fitting comparisons in order to make inferences about signatures of selection across genes and phylogenies. For genes or codons under purifying selection, nonsynon- ymous substitutions are selected against, leading to dN/ dS values less than 1. In contrast, positive selection—or ra- pid evolution—is characterized by the relative enrichment of nonsynonymous substitution rates, which can be identi- fied by elevated dN/dS values (>1) in these genes or at spe- cific codons within genes. Using the site models implemented in PAML along with BUSTED, we identified signatures of site-specific positive se- lection by at least one of the two algorithms (BUSTED P < 0.05 or PAML M7 vs. M8 likelihood-ratio test [LRT] P < 0.05) in nine (35%) of the primate PRRs (fig. 1 and supplementary data file 1, Supplementary Material online). This number was strikingly elevated among the rodent and bat lineages, with 16 (62%) and 21 (81%) genes under positive selection in these groups, respectively. Mapping these positively selected genes onto a phylogenetic tree of the CTLDs from the CLR-type PRRs revealed no clear pat- tern to the distribution of positive selection across this fam- ily of receptors (fig. 1B and supplementary fig. S1B, Supplementary Material online). Instead, rapid evolution seems pervasive across the entire family of CLRs that were analyzed. Through these approaches, we identified a core set of six PRRs predicted to be under positive selection by one or both algorithms in all mammalian lineages tested. These core genes include those, such as TLR4, with long-established roles in microbial recognition and previously defined li- gands. However, this core group, surprisingly, also includes the CLR gene CLEC12A, whose role in interactions with mi- crobes is still emerging, pointing to the possibility of as yet undefined, but important, roles for this CLR in microbial recognition. Beyond the shared signatures of positive selec- tion across lineages, these core rapidly evolving PRRs also tended to have a higher number of sites predicted to be un- der positive selection, with many of the rapidly evolving amino acid residues falling into functionally relevant re- gions of these receptors, namely the extracellular carbohydrate-binding domains. Outside of this core set of positively selected genes, we observed lineage-specific patterns of positive selection among the remaining PRRs. These different patterns of se- lection across the three mammalian lineages suggest the possibility that distinct populations of microbial species may have played a role in shaping the evolution of these mammalian receptors. Importantly, our analyses of the CTLDs of mammalian selectins revealed little evidence for Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 3 Hilbert et al. GBE FIG. 1.—Positive selection across mammalian carbohydrate recognition PRRs. (A) Positive selection analyses of 26 glycan PRRs in primates (left column), rodents (middle), and bats (right column). Colored boxes indicate whether evidence of positive selection was supported by PAML analyses only (medium blue) or by both PAML and BUSTED analyses (dark blue). Genes with no evidence for positive selection are represented by pale blue boxes. Statistical cutoffs were P < 0.05 for PAML M7 versus M8 likelihood ratio tests and for BUSTED analysis. (B) Patterns of positive selection mapped onto a phylogenetic tree of the human CTLD domains. Only genes from the gene set with CTLDs are represented. Colored circles represent evidence of positive selection in the primate (orange), rodent (purple), and/or bat (blue) lineages. Genes with black circles were not analyzed in this study because of unclear ortholog relationships across mammals but do have important roles in pathogen detection in mammals. Numbers indicate bootstrap values from phylogenetic tree construction using IQ-TREE. 4 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 ABRapid Evolution of Glycan Recognition Receptors GBE positive selection in these genes with high levels of conser- vation across lineages. This further underscores the role of microbial pathogen interactions in driving the evolutionary signatures we observe across this gene set of PRRs. Rapidly Evolving Codons in Mammalian Langerin (CD207) Correspond with Amino Acid Positions at Key Ligand Recognition Interfaces The set of PRRs under positive selection in all three of the tested mammalian lineages includes Langerin (CD207), a CLR expressed primarily by the Langerhans cells of the skin as well as other professional antigen presenting cells. Langerin has an established role in the activation of critical inflammatory responses following direct detection of di- verse microbial pathogens, including fungi, viruses, and bacteria (de Witte et al. 2007; de Jong et al. 2010; van der Vlist et al. 2011; van Dalen et al. 2019). In particular, Langerin has been shown to be able to recognize and bind to ß-glucans in Candida species as well as the skin-associated fungal species Malassezia furfur (de Jong et al. 2010). Bacterial recognition by Langerin has been ob- served for multiple species, including Staphylococcus aur- eus, a major cause of skin infections (Yang et al. 2015; van Dalen et al. 2019). In the context of both fungal and S. aureus infection, Langerin has been shown to play a role in regulating inflammatory Th17 responses (Sparber et al. 2018; van Dalen et al. 2019). Structural studies of hu- man Langerin have revealed it to have a canonical CLR fold, with a Glu-Pro-Asn (EPN) motif in the primary ligand bind- ing site, suggestive of a ligand preference for mannose and mannose-type carbohydrates (Tateno et al. 2010; Feinberg et al. 2011; Hanske et al. 2017). Interestingly, recent work examining the ligand-binding profiles of Langerin homo- logs from humans and mice identified distinct differences in the binding specificities for more complex bacterial- derived glycans among these homologs, despite conserva- tion of the EPN motif in the binding site (Hanske et al. 2017). This suggests that sequence variation in the Langerin CTLD may play an important role in modulating microbial recognition. To determine whether the signals of rapid evolution that we observe in Langerin across mammalian lineages might functionally correlate with differences in ligand preference, we first mapped the sites under positive selection in each lineage to the annotated protein domains (fig. 2A). A large proportion of positively selected sites in all three lineages mapped to the extracellular region of the protein, with many falling into the CTLD itself, including several overlap- ping amino acid positions which were predicted to be un- der positive selection in all three mammalian lineages. In addition to the PAML algorithm, we also used the HyPhy suite programs mixed effects model of evolution (MEME) and fast unbiased Bayesian approximation (FUBAR) to independently assess individual amino acid sites for elevated dN/dS values across the Langerin coding se- quence (Murrell et al. 2012, 2013). Although MEME, like PAML, assesses patterns of episodic selection occurring on at least one branch of the phylogeny, the FUBAR algo- rithm can be used to identify sites under pervasive positive selection across an entire phylogeny. These additional ana- lyses, thus, provide both confirmatory and complementary methods to PAML for assessing site-specific rapid evolution. Agreement between the three algorithms was high across all positively selected sites in Langerin (fig. 2B). In particular, amino acid positions 213 and 289, which were identified by PAML analyses in all three lineages, showed signatures of positive selection in the MEME and FUBAR analyses in both primates and bats. Similarly, multiple methods inde- pendently highlighted position 313 as rapidly evolving in bats and rodents, in agreement with the PAML analyses of primate sequences. Rapid evolution of other lineage-specific sites was also supported by all three analyses (fig. 2B). The convergence of these signatures of rapid evolution on the Langerin CTLD and these three residues (213, 289, and 313) across multiple mammalian lineages hints at pos- sible functional significance to amino acid changes at these positions. When mapped onto a crystal structure of the Langerin CTLD in complex with a mannose ligand and a co- ordinating calcium ion, we observed that many of the resi- dues under positive selection clustered around the ligand binding site (fig. 2C). This supports the hypothesis that vari- ation at these positions across mammalian Langerin homo- logs might result in differences in microbial glycan binding specificities. Furthermore, this suggests the possibility that the signals of rapid evolution we observe in mammalian Langerin homologs was driven by the selective pressure to maintain the ability to recognize specific microbial spe- cies through distinct microbial glycans on their surfaces and in their cell walls. Mapping Patterns of Substitution in an Invasive Aspergillosis Susceptibility Allele of MelLec (Melanin Lectin/CLEC1A) across Primates Unlike many CLRs, which can recognize similar ligands present on many different species of microbes, MelLec (also known as CLEC1A), was recently identified as being a highly specific receptor for 1,8-dihydroxynaphthalene (DHN)-melanin, a critical component of the cell walls of a relatively limited group of fungal species (Stappers et al. 2018). Included in these DHN-melanin-producing fungi are the human fungal pathogens Aspergillus fumigatus and the black yeasts, which account for significant morbidity and mortality in both immune- suppressed and immunocompetent patients worldwide (Brown et al. 2012; Seyedmousavi et al. 2014). Recognition of DHN-melanin in fungal cells via MelLec Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 5 Hilbert et al. GBE FIG. 2.—Diversification of Langerin (CD207) ligand-binding interfaces in all mammalian lineages. (A) Positively selected residues (triangles) predicted by PAML (Model 8, BEB > 0.9) cluster primarily in the extracellular portion of Langerin (CD207), with many in the CTLD. A number of positively selected sites in the CTLD are common across primates (orange triangles), rodents (purple triangles), and bats (blue triangles). (B) Agreement between different algorithms for identifying site-specific positive selection in Langerin of different mammalian groups. Listed residue numbers correspond to the position in the human Langerin sequence. Single letter residues correspond to the amino acid identity in human (primates, left), house mouse (rodents, middle), or black flying fox (bats, right) sequences. Bolded residues are those predicted to be under positive selection across all mammals by one or more tests. (C) Positively selected sites mapped onto a crystal structure of the human Langerin CTLD (gray, PDB:3p5d) in complex with a mannose ligand (yellow) and Ca2+ ion (magenta) (Feinberg et al. 2011). Positively selected sites in all three lineages (colored in green) along with several sites from rodent (blue) and bat (purple) analyses are shown with sidechains and surround the ligand binding site. has been demonstrated to be critical for the activation of an antifungal immune response and survival of systemic A. fumigatus infection in in vivo models. Notably, a com- mon human polymorphism causing a single amino acid change (Gly26Ala, rs2306894) has been identified in the cytoplasmic region of the MelLec protein. This Ala26 allele has been associated with higher probability of invasive Aspergillosis in transplant patients and has also been shown to result in decreased production of crit- ical cytokines in response to fungal stimulation in in vitro experiments (Stappers et al. 2018). Combined, these data support a role for MelLec in the immune responses to fun- gal infection in both mice and humans. Our PAML analyses revealed signatures of recurrent posi- tive selection in MelLec in both the primate and rodent lineages (fig. 1). Although significance by LRT varied for pri- mate analyses of MelLec depending on whether a species or gene tree was used in the analysis, manual inspection of the alignments revealed extensive sequence variation at PAML-identified sites across the primate MelLec orthologs (see Methods and supplementary data file 1, Supplementary Material online). This suggests that interac- tions between these mammalian groups and pathogenic fungi may have played a role in shaping amino acid diversi- fication in this PRR. Furthermore, the rapidly evolving amino acids within MelLec include several in the CTLD, consistent 6 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 ABCRapid Evolution of Glycan Recognition Receptors GBE with the potential for sequence variation to confer changes in ligand-binding affinity or specificity among different MelLec 1, homologs Supplementary Material online). (supplementary data file While mapping the positively selected sites in primate MelLec orthologs, we were surprised to find that at the site of the human polymorphism, Gly26, we observed a con- served alanine residue in all primates except humans and black-capped squirrel monkeys (Saimiri boliviensis bolivien- sis, fig. 3A). This suggests that Gly26 likely represents the de- rived human allele, while alanine is the ancestral allele among primates. Whether the alanine at position 26 in other primate homologs confers the same defects in cytokine pro- duction observed for the human allele is presently unknown. Although it is possible that sequence variation elsewhere in the primate MelLec homologs might compensate for the alanine at position 26, future experimental studies will be needed to assess how sequence variation at this and other sites contribute to function of the MelLec receptor. We next explored the distribution of these two MelLec al- leles in human populations. Across human populations in the 1000 Genomes Project (1KG) dataset, the frequency of the derived Gly26 allele varies widely, from only 0.11 in African (AFR) and 0.13 in European (EUR) populations to 0.65 in East Asian (EAS) populations (fig. 3C) (The 1000 Genomes Project Consortium 2015). Given the high fre- quency of the Gly26 allele in EAS populations, we turned to two additional resources to more comprehensively assess the distribution of this allele across Asia (GenomeAsia100K Consortium et al. 2019; Bergström et al. 2020). Using the Genome Asia 100K Browser and the Human Genome Diversity Project (HGDP), we observed that the Gly26 allele reached even higher frequencies in Oceanic (OC) and Southeast Asian (SAS) populations that were not repre- sented in the 1000 Genomes dataset. The Gly26 allele was fixed in the populations from Papua New Guinea in the HGDP, though the sample size was small (n = 17) and at an allele frequency (AF) of 0.77 in PNG in the Genome Asia 100K dataset (n = 70) (fig. 3C). The HGDP also revealed a high frequency of the Gly26 allele in multiple American (AMR) populations (e.g., AF = 1 in Colombian, AF = 0.94 in Karitiana and AF = 0.95 in Pima), which may reflect the shared ancestry between native American and Asian popula- tions. To quantify the allele frequency differences observed across these populations, we calculated pairwise FST be- tween EUR populations (with low Gly26 frequencies) and the OC, SAS, and AMR populations and tested for signifi- cance relative to other single nucleotide polymorphisms (SNPs) on chromosome 12 (supplementary data file 1, Supplementary Material online). FST was high between all tested populations, falling in the tail of the empirical distribu- tions, indicating an elevated signal of differentiation consist- ent with the allele frequency differences observed between these groups. The extreme population differentiation of the rs2306894 Gly26Ala SNP could reflect that this locus has been a target of selection in human populations. Both posi- tive and balancing selection can affect population differen- tiation and FST values. We first assessed whether rs2306894 or any other SNPs in MelLec showed signatures of local positive selection. Both searches of published scans for re- cent positive selection focusing on Asian populations as well as our own analysis of the Colombian population from the 1KG database using Relate showed no evidence for positive selection in MelLec in human populations (supplementary fig. S2, Supplementary Material online) (Voight et al. 2006; Liu et al. 2017; Speidel et al. 2019, 2021). Next, we calculated Tajima’s D in 1 kb windows across all of Chromosome 12 in each population from the HGDP and 1KG datasets. Notably, we observed elevated Tajima’s D values for the window containing MelLec and rs2306894 in the majority of the tested populations, with a significantly positive value in 31 of 62 populations as- sessed (empirical P < 0.05), suggestive of balancing selec- tion acting at this locus (fig. 3C, middle). To further confirm this, we ran BetaScan, a more sensitive method for detecting balancing selection, where high β(1) statistics are indicative of an excess of SNPs at similar frequencies, a key feature of genomic regions under balancing selection (Siewert and Voight 2017, 2020). The β(1) statistic was significantly elevated (empirical P < 0.05) for MelLec in all of the 1KG populations except for the AFR populations, further suggesting that this gene has been subject to balancing selection in many human populations (fig. 3C, bottom). is in perfect the selective signatures we It is important to note that while previous functional studies have focused solely on the Gly26Ala SNP, our ana- lyses revealed that this SNP linkage disequilibrium (LD) with a large number of other SNPs with- in MelLec (e.g., 42 SNPs in r2 = 1 with rs2306894 in EAS, spanning 8 kb) making it challenging to distinguish the tar- identify here get of (supplementary data file 1, Supplementary Material online). The vast majority of these SNPs fall into intronic regions and are documented eQTLs for MelLec in multiple tissues in the Genotype-Tissue Expression (GTEx) project (Lonsdale et al. 2013). Two of these SNPS in LD with rs2306894 fall within regulatory regions which could have direct regulatory ef- fects on expression of MelLec: rs2306893 in the 5′UTR and rs2277416 in a splice region. Future studies probing the effects of these SNPs on MelLec function may further our understanding of how they individually or collectively contribute to fungal disease and reveal a more nuanced un- derstanding of the target of the balancing selection signa- tures we observe. Beyond humans, we also noted that the black-capped squirrel monkey sequence from the NCBI GenBank data- base carried a valine at position 26, in contrast to the Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 7 Hilbert et al. GBE FIG. 3.—Single nucleotide polymorphisms in primate populations converge on a single site in Melanin Lectin (CLEC1A). (A) Patterns of conservation and variation at amino acid position 26 of MelLec across primates. Most primate species carry the ancestral alanine allele (orange highlighting), whereas single nucleotide polymorphisms in both humans (glycine, green highlighting) and squirrel monkeys (valine, pink highlighting) confer missense mutations. (B) Genotypes of 19 unrelated squirrel monkey gDNA samples from three S. boliviensis subspecies. The sex and the amino acid identity at position 26 for each individual are indicated, with heterozygous individuals indicated as carrying both Ala and Val amino acids (A/V in Black-capped and Peruvian squirrel monkeys). (C) (top) Geographic distribution of the glycine 26 allele (green) at SNP rs2306894 in human populations. Allele frequencies are shown for popula- tions from the 1KG Project and the HGDP. Individuals carrying the Ala26 allele (orange) have been previously shown to have higher risk of invasive fungal infections in stem-cell transplant patients (Stappers et al. 2018). (middle) Tajima’s D values for populations from the HGDP and 1KG and (bottom) β(1) for populations from the 1KG project showing evidence of balancing selection at the MelLec locus. For both plots, empirical P-value < 0.05, ** empirical P-value < 0.01. Population abbreviations are as follows: AMR, America; AFR, Africa; EUR, Europe; CSA, Central-South Asia; ME, Middle East; SAS, South Asia; EAS, East Asia; OC, Oceania. 8 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 ABCRapid Evolution of Glycan Recognition Receptors GBE alanine of all other primates (fig. 3A). To confirm this obser- vation and investigate the patterns of substitution at this position among squirrel monkey populations, we amplified the region surrounding this SNP from multiple genomic DNA (gDNA) samples from black-capped squirrel monkeys (S. boliviensis boliviensis) as well as two other closely related squirrel monkey subspecies: Peruvian squirrel monkeys (S. boliviensis peruvinsis) and Guianan squirrel monkeys (S. sciureus sciureus). In total, we genotyped 19 unrelated individuals from these three subspecies. Interestingly, the Guianan squirrel monkeys were universally homozygous for the ancestral Ala26 allele, whereas no individuals homo- zygous for this allele could be found in the other two sub- species (fig. 3B and supplementary fig. S3, Supplementary Material online). Among black-capped and Peruvian squir- rel monkeys, there was a mix of individuals homozygous for the derived Val26, as well as heterozygous individuals, again raising intriguing questions about the potential se- lective pressures that have shaped allele frequency distribu- tions in squirrel monkeys as in humans. To rule out the possibility that the lack of observed se- quence variation in other primates might be due to sam- pling bias of the publicly available sequences in GenBank, we also looked for variation at this locus among hominoid primates using data from the Great Ape Genome Project (Prado-Martinez et al. 2013). There was no evidence in these data for any sequence variation at amino acid pos- ition 26 in gorillas, bonobos, chimpanzees, or orangutans (supplementary data file 1, Supplementary Material online). Combined, these data strongly suggest that mutation of this locus has occurred independently in humans and squir- rel monkeys, perhaps due to similar evolutionary pressures in these species from fungi or other microbial species. Extensive Positive Selection across CLEC12A in Primates, Bats, and Rodents Portends an Unidentified Role in Microbial Recognition and Binding In addition to genes with well-established roles in immune responses to microbial pathogens, our analyses also re- vealed extensive positive selection occurring at sites within the CLEC12A gene, a more mysterious member of the CLR family of receptors. Originally identified as a receptor for uric acid, a marker of cell death, other reports have identi- fied roles for this receptor in the recognition of hemozoin produced by Plasmodium spp. during infection as well as in the regulation of antibacterial autophagy responses (Neumann et al. 2014; Begun et al. 2015; Raulf et al. 2019). Most recently, CLEC12A, has been shown to directly bind to a number of gut-resident bacteria and is required for the phagocytosis of these bacteria and subsequent modulation of microbiome community composition (Chiaro et al. 2023). Although the exact moiety that CLEC12A engages remains undefined, these data strongly suggest the possibility that CLEC12A is also capable of rec- ognizing molecular patterns found in the bacterial cell wall, including bacterial glycans. Given the breadth of the cur- rently known ligands and roles of CLEC12A and its expres- sion predominantly in myeloid cells, it is likely that the full scope and nature of the interfaces between CLEC12A and pathogenic microbes has not yet been revealed. Further supporting this idea, our phylogenetic analyses of CLEC12A revealed strong signals of positive selection on this gene across all mammalian lineages, suggestive of strong selection imposed on this gene by interactions with, perhaps, diverse pathogens (fig. 4). In fact, in both bats and primates, the gene-wide dN/dS calculated by PAML was >1 (supplementary data file 1, Supplementary Material online). CLEC12A was the only gene analyzed in this study for which this was true and supports the model that CLEC12A is evolving under remarkably strong positive selection in mammals. Although positively selected sites were distributed across the entire coding sequence of CLEC12A, a large number fall directly in the CTLD, a pattern which is most pronounced in primates (orange triangles, fig. 4A). Many of these sites were independently predicted to be rapidly evolving by PAML, MEME and FUBAR and tend to cluster in the same regions in all three mammalian groups, suggesting these may be regions important for the immune or ligand binding functions of the protein (supplementary data file 1, Supplementary Material online). Given the large number of sites under positive selection in the CTLD, no discernable patterns emerged from mapping these sites onto AlphaFold-predicted structures of CLEC12A CTLD homo- logs from different species that might hint at effects of se- quence diversification on ligand binding. Of note, however, was the fact that despite the primary sequence divergence across mammals, there were no significant differences in the AlphaFold-predicted structures of primate, rodent and bat homologs suggesting that more subtle modifications in structure may underlie any functional differences be- tween homologs (supplementary fig. S4, Supplementary Material online). To identify specific rapidly evolving branches in each mammalian lineage, we applied models implemented in PAML that allow calculation of dN/dS for each branch of a given phylogenetic tree (fig. 4B–D). This temporal view of the evolution of CLEC12A revealed extensive episodic positive selection across each of the mammalian phyloge- nies. Among the simian primates, all three major groups (Hominids, Old World, and New World Monkeys) contained branches with elevated dN/dS values, though these values were slightly higher among both the ancient and recent branches in the hominid and New World Monkey lineages (fig. 4B). Similar patterns can be seen in the rodent and bat phylogenies, where positive selection was also rampant (fig. 4C and D). Consistent with the elevated gene-wide Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 9 Hilbert et al. GBE FIG. 4.—Extensive positive selection in CLEC12A across mammals reveals a new host–pathogen battleground. (A) Diagram showing sites under positive selection in CLEC12A in primates (orange triangles), rodents (purple triangles) and bats (blue triangles). Indicated sites were predicted by PAML (Model 8, BEB > 0.9). Locations of the CTLD and transmembrane domain are indicated on the left. (B)–(D) dN/dS values for CLEC12A were calculated across the species phylogenies of primates (B), rodents (C), and bats (D) using PAML (free ratios, Model = 1 setting). Lineages with elevated dN/dS values (>1), suggestive of positive selection along that branch, are indicated with colored lines. Calculated dN/dS values are listed above each branch and for branches lacking either nonsynonymous or synonymous sites; ratios of the respective substitution numbers (N:S) are indicated. dN/dS value observed for bat CLEC12A (dN/dS = 1.2, supplementary data file 1, Supplementary Material online), especially high substitution rates were abundant across the bat phylogeny, and in particular among the new world leaf- nosed bats (Phyllostimidae), a group which includes the spear-nosed bats, Jamaican fruit bat and the Honduran yellow-shouldered bat (fig. 4D). Combined, the strength of the signals of rapid evolution that our analyses revealed 10 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 ABCDRapid Evolution of Glycan Recognition Receptors GBE in CLEC12A across multiple mammalian lineages, suggest it functions as an underappreciated but critical component in the arsenal of immune receptors that engage with micro- bial pathogens and play a role in immune defenses against infection. Although it is theoretically possible that the sig- nals we observe in CLEC12A have been driven by already identified ligands and interactions, we hypothesize that interactions between there are CLEC12A and other microbial species for which this se- quence variation will have functional implications. likely undiscovered Discussion Our study revealed widespread signatures of rapid evolu- tion across glycan-recognition PRRs in three major mamma- lian lineages: primates, rodents, and bats. Such strong signatures of positive selection are frequently associated with host–pathogen arms races, signifying the consequen- tial impacts on fitness associated with these interactions. We hypothesize that the evolutionary signatures we ob- serve among CLRs and related factors represent a new axis in these arms races where hosts keep pace with the nu- merous and well-studied evasive strategies microbes use to prevent detection of their immunogenic glycan-rich sur- faces. Consistent with this hypothesis, we found that posi- tive selection among these genes is often enriched in functionally significant portions of the protein, namely in the CTLDs which directly interact with glycans. In Langerin, this pattern was particularly clear, with a cluster of rapidly evolving residues falling directly surrounding the ligand binding pocket of the CTLD (fig. 2C). Positively selected sites in Langerin include amino acid position 313, which has previously been determined to contribute signifi- cantly to ligand binding, with mutations at this position re- sulting in a complete lack of recognition of certain simple carbohydrate ligands (Tateno et al. 2010). Across all the mammalian species we analyzed in this study, we observed eight different amino acids sampled at this position, a find- ing that strongly points to functional differences in ligand binding and specificity. The finding that the highly specific DHN-melanin binding MelLec receptor is rapidly evolving in both primates and rodents is particularly exciting. To date, studies of host–microbe evolutionary arms races have largely involved only interactions with viruses or bacteria; the role of eukary- otic pathogens, such as fungi, in shaping the evolution of mammalian host species has remained unexplored. Rapid evolution in MelLec across species when paired with the emerging patterns of substitution at a functionally import- ant site in both humans and squirrel monkeys strongly sug- gests that fungi can, in fact, play an important role in shaping the evolution of mammalian immune systems. Additionally, many of the other PRRs identified as rapidly evolving in this study also engage with fungal pathogens, suggesting that the breadth of host proteins shaped by in- teractions with pathogenic fungi may be extensive. Our population genetics analyses of the human MelLec Gly26Ala SNP further revealed strong population differenti- ation in the allele frequencies of this SNP along with signals of balancing selection within this locus in many human popu- lations. This raises several intriguing hypotheses: first, that dif- ferent association with fungal species across geographic regions might partially account for the allele frequency differ- ences observed across human populations. Other factors, such as lifestyle and/or dietary differences across human po- pulations could also play a role in driving the population dif- ferentiation we observe. Whether and how these different pressures shaped the distribution of these MelLec alleles in human populations remains a fascinating challenge to dis- sect. A second hypothesis that arises from our population genetic analysis of MelLec suggests that although the Gly26 allele appears to be protective under some circum- stances, there may be tradeoffs associated with changes at this position, reflected in the maintenance of the ancestral Ala26 allele in human populations and the signals of balan- cing selection we observe. Indeed, although MelLec is essen- tial for protection against invasive disease caused by fungal species like A. fumigatus, its function was shown to be detri- mental in in vivo models of asthma driven by the same fungal species suggesting that MelLec activity has a complex impact on establishing appropriate immune responses to fungi (Stappers et al. 2018; Tone et al. 2021). Whether and how mutation of position 26 (or other sites) within the MelLec lo- cus might contribute to these differing outcomes remains to be seen but may provide some insight into the signals of bal- ancing selection we observe in this gene. Previous analysis of carbohydrate-ligand binding in dif- ferent mammalian Langerin homologs led to the surprising finding that although specificity in ligand binding for simple carbohydrates was similar across different Langerin var- iants, dramatic differences were observed in the context of complex carbohydrates and intact bacterial cells (Hanske et al. 2017). These differences were identified des- pite high conservation in the solved crystal structures of the CTLDs from these homologs, suggesting that more subtle structural or sequence variation underlies variability in lig- and binding. Our analyses of the CLEC12A gene suggest this may be a general feature among these rapidly evolving CLRs. In CLEC12A, we observed extensive diversification of the primary sequence in all mammalian lineages analyzed, but very little change in the predicted structures of diverse variants of this protein (supplementary fig. S4, Supplementary Material online). This suggests that the CLR fold is highly ro- bust to sequence variation and underscores the need for fu- ture studies to parse the functional implications of the sequence variation we observe. Our results raise intriguing questions about the interac- tions that drive rapid evolution in glycan-recognition Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 11 Hilbert et al. GBE receptors and what the tradeoffs may be for interactions with other microbes. Many of these PRRs are nonspecific, involved in the recognition of many diverse glycan struc- tures found in multiple microbial species. This suggests that diversification of the carbohydrate recognition do- mains of these PRRs could have a profound impact on the recognition of numerous microbial species. Although this may make it challenging to identify the exact molecular changes or microbial species that have driven rapid evolu- tion in these glycan PRRs, this system represents a unique opportunity to study the tradeoffs associated with rapid evolution, a topic that has been largely ignored in pro- tein–protein arms races, where the focus has remained on 1:1 interactions between host proteins and highly specific pathogenic substrates. Recent advances in high-throughput profiling of host lectin interactions with complex microbial glycans when applied to these rapidly evolving PRRs will like- ly help to shed light on these questions of what drove these signals of evolution and what the consequences might be for specific microbial recognition (Stowell et al. 2014; Jégouzo et al. 2020). Finally, our phylogenetic screen identified extensive posi- tive selection among rodent and, in particular, bat glycan PRRs, where a striking 81% of the genes we analyzed were found to be rapidly evolving. This suggests that for these carbohydrate-recognition receptors, evolution has been driven by lineage-specific microbial communities, per- haps including both pathogenic and commensal species. Combined, our data reveal a new axis of evolutionary arms races—involving microbial glycan detection—and dramatically expand the realm of host–microbe interactions to include fungal pathogens with consequential influence on the evolution of eukaryotic biology. Materials and Methods Phylogenetic Analyses Candidate gene ortholog sequences were obtained from NCBI GenBank either through gene name searches or by BLAST searches using the Human ortholog sequence as query (see supplementary data file 1, Supplementary Material on- line for full list of accession numbers). Additional BLAST searches were carried out using alternate species as query to confirm that the same subsets of genes were being iden- tified through different searches. Orthologous relationships between genes were further confirmed by phylogenetic and synteny analysis and species were excluded from evolu- tionary analysis if clear orthology could not be established. Phylogenetic tree analysis of some of the more divergent genes, like CLEC12A, confirmed that orthologs of CLEC12A from all three mammalian groups cluster together on a single branch, removed from the other CLR genes (supplementary fig. S1A, Supplementary Material online). Sequences were obtained for all available simian primate species, Myomorpha species (minus Jaculus jaculus, for which we could not consistently find well-annotated ortho- logs), and the Chiroptera. Coding sequences were down- loaded and aligned using the Geneious Translation Align function with the MUSCLE algorithm option. Alignments were manually inspected and trimmed to remove gaps, ambiguous regions of the alignment and stop codons. Alignments were used to construct gene trees using IQ-TREE and the GTR + G + I model with 100 nonparametric bootstraps (Nguyen et al. 2015). Both gene trees and gen- erally accepted species phylogenies for each of the mam- malian groups were used for downstream evolutionary analyses. Alignments and trees used in analysis can be found in supplementary data file 2, Supplementary Material online. Data shown in figure 1 are based on analyses done with species trees, but all of the results of the analyses can be found in supplementary data file 1, Supplementary Material online. Unless otherwise noted, all computational analysis was performed using the University of Utah Center for High Performance Computing. Positive selection was assessed using the codeml func- tion of the PAML software package (v4.9) with the F3 × 4 codon frequency model (Yang 2007). Gene-wide dN/dS va- lues were calculated using model 0. To test whether a sub- set of amino acid sites were evolving under positive selection, we performed LRTs, comparing pairs of NSsites models including: M1 (neutral evolution) versus M2 (posi- tive selection) and M7 (neutral, beta distribution dN/dS ≤ 1) versus M8 (positive selection, beta distribution allowing for dN/dS > 1). For genes with statistical support for positive selection, specific amino acid positions were identified as being under positive selection based on having a Bayes Empirical Bayes (BEB) posterior probability of greater than 90% in the M8 model. For the free ratios analysis of CLEC12A, codeml Model 1, allowing variation of dN/dS across branches of the phylogeny, was run on the CLEC12A alignments with an unrooted species tree for each lineage. The BUSTED, MEME, and FUBAR programs from the HyPhy suite (version 2.5.41) were run through the com- mand line with the same input alignments and trees used for PAML analyses and default options (Pond et al. 2005; Murrell et al. 2012, 2013, 2015). Results were visualized using the HyPhy Vision web server. For several of the BUSTED analyses, we noticed that the algorithm found statistically significant support for positive selection in align- ments that had very high levels of conservation determined by other methods (e.g., Primate FIBCD1 and Dectin1). When we examined these results, we found that the signal was being driven entirely by codons containing multiple nu- cleotide substitutions, which has been a documented con- founding variable in branch-site models of rapid evolution (Venkat et al. 2018). For these anomalous results, we re-ran 12 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 Rapid Evolution of Glycan Recognition Receptors GBE the analyses without these multiply substituted sites and found that these genes were no longer predicted to be un- der positive selection by BUSTED (see “BUSTED P-value with MNMs removed” column in supplementary data file 1, Supplementary Material online). These re-runs are reflected in the results displayed in figure 1. Codon alignments of the Human CTLDs from each of the CLRs in the gene set were used as input to IQ-TREE for phylogenetic tree construction (fig. 1B) (Nguyen et al. 2015). The VT + G4 substitution model was selected as the best fit model by the ModelFinder function, and 100 nonparametric bootstrap replicates were performed (Kalyaanamoorthy et al. 2017). Some IQ-TREE analyses were performed with the IQ-Tree webserver (Trifinopoulos et al. 2016). CTLDs were identified based on annotated do- mains from UniProt and genes with multiple CTLDs (e.g., MRC1 and MRC2) were excluded. An alternate version of this tree built from an alignment of nine representative spe- cies spanning all three mammalian groups assessed is shown in supplementary figure S1B, Supplementary Material on- line. Species included were: Homo sapiens, Mucaca mulat- ta, S. boliviensis, Mus musculus, Microtus ochrogaster, Nannospalax galili, Myotis myotis, Pteropus alecto, and Rhinolophus sinicus. Tree topology varied only slightly across species and in this pan-species tree. MelLec Human Population Genetics Analyses To map the geographic distribution of the G26A poly- morphism (rs2306894) in Human MelLec (CLEC1A), sam- pling locations of 1KG on GRCh38 and HGDP populations were downloaded from the International Genome Sample Resource (Zheng-Bradley et al. 2017; Lowy-Gallego et al. 2019; Bergström et al. 2020). Chromosome 12 VCF files for HGDP and 1KG datasets were downloaded from their respective FTP sites (see Data Availability statement below). VCFtools was used to obtain the allele frequency at G26A for all populations, and the map was created using the R library ggmap (Danecek et al. 2011). Tajima’s D was calculated using VCFtools and β(1) statis- tics using BetaScan2 (Siewert and Voight 2020). The de- rived allele was obtained from ancestral FASTA files downloaded from Ensembl (see Data Availability statement below). Empirical P-values were calculated in R by compari- son with all other test statistic values on chr12 and plots were generated with ggplot2 (R Core Team 2022). Cowplot was used to combine the map, Tajima’s D, and β(1) plots. r2 was calculated between rs2306894 and SNPs within 100 kb in either direction to identify pairs in high linkage disequilibrium using VCFtools and plink2 (Chang et al. 2015). We also generated a population-specific chromo- some 12 VCF, using VCFtools, from the 1KG Colombian population to test for positive selection using Relate v1.1.8 and the add-on module for selection, which infers how quickly a mutation spread through the population based on genome-wide genealogies (Speidel et al. 2019, 2021). Squirrel Monkey gDNA MelLec Genotyping Squirrel monkey gDNA was originally isolated from blood samples kindly provided by the MD Anderson Squirrel Monkey Resource and Breeding Center in September 2015. The provided samples came from unrelated individuals and additional information including Sample IDs, sex and age of the animals can be found in supplementary figure S3, Supplementary Material online. One additional gDNA sample from S. sciureus sciureus was isolated from the AG05311 fibroblast cell line provided by the Coriell Institute. All gDNA samples have been stored at −20 °C. Primers MS_B17 and MS_B20 were designed to amp- lify a ∼500 bp fragment including the entirety of Exon 1 of MelLec (CLEC1A) which contains the polymorphic site (amino acid 26), along with flanking sequence. The black-capped squirrel monkey genome saiBol1was used as a reference for primer design. Polymerase chain reac- tions were performed using Phusion Flash polymerase and 50 ng of each gDNA sample from the squirrel mon- key individuals. PCR products were confirmed on a gel, purified with Exo-SAP and Sanger sequenced at the University of Utah Sequencing Core using primer MS_B19. Genotypes were called based on visualization of Sanger sequencing traces in Geneious. Primer se- quences are as follows: MS_B17 TCCATGAGAGGTGCAAACAG MS_B20 AGTTGTGGAAAGCGCACAG MS_B19 ACATGCTGTTTCCCTTCAGC Structural Modeling and Comparisons of CLEC12A CTLDs The structures of the CTLDs of nine mammalian CLEC12A orthologs were modeled using AlphaFold (v 2.1.2) (Jumper et al. 2021). The predicted structure with the high- est confidence (ranked_0.pdb) for each ortholog was com- pared with all other species using jFATCAT through the RCSB PDB Pairwise Structure Alignment tool (Prlić et al. 2010; Burley et al. 2018; Li et al. 2020). Alignments were performed using both the rigid and flexible alignment algo- rithms and results were identical between the two. RMSD values were plotted as a heatmap in R (supplementary fig. S4, Supplementary Material online). All ranked_0 predicted structures and CTLD sequences used for modeling can be found at: dx.doi.org/10.6084/m9.figshare.23535738. Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 13 Hilbert et al. GBE Supplementary Material • eQTL analysis available from GTEx: https://gtexportal.org/ Supplementary data are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/). Acknowledgments We thank members of the Elde lab for helpful discussions in the development of this project. We thank Stephen Goldstein for suggestions on tree-building and primate population genetics and Ian Boys for help with AlphaFold modeling. This work was supported by the National Institutes of Health (grant number R35 GM147709 to E.M.L, grant number R35 GM134936 to N.C.E., and grant number T32GM141848 to H.J.Y.); a Burroughs Wellcome Fund Investigator in the Pathogenesis of Infectious Disease Award to N.C.E.; and a postdoctoral fellowship from the Helen Hay Whitney Foundation to Z.A.H. Author Contributions Z.A.H. and N.C.E. designed the study and wrote the manu- script. Z.A.H. performed evolutionary analyses, structural modeling, and interpreted results. P.E.H. and E.M.L per- formed population genetics analyses on MelLec and inter- preted results. H.J.Y. performed BLAST searches and sequence alignments for phylogenetic analyses. M.J.W.S. performed squirrel monkey sample PCRs, sequencing, and data analysis. All authors reviewed and edited the manuscript. Data Availability NCBI accession numbers for all genes analyzed are provided in supplementary data file 1, Supplementary Material online. Alignments and trees used in positive selection analyses are provided in supplementary data file 2, Supplementary Material online. Genotypes for great ape species at the position of the rs2306894 human polymorphism were ob- tained from: https://www.biologiaevolutiva.org/greatape/ data.html. For analyses of MelLec in human populations, the following links were used to download or access the rele- vant datasets: • Sampling locations: https://www.internationalgenome. org/data-portal/population. • HGDP Chr12: https://ngs.sanger.ac.uk/production/hgdp/ hgdp_wgs.20190516/. • 1KG Chr12: http://ftp.1000genomes.ebi.ac.uk/vol1/ftp/ data_collections/1000G_2504_high_coverage/working/ 20201028_3202_phased/. • Ancestral FASTA files for GRCh38 (homo sapiens ances- tor GRCh38.tar.gz downloaded March 2023): https://ftp. ensembl.org/pub/current_fasta/ancestral_alleles/. home/snp/rs2306894. FST, Tajima’s D, β(1) statistics, and statistics from linkage disequilibrium analysis are provided in supplementary data file 1, Supplementary Material online. AlphaFold-modeled CLEC12A CTLD structures can be found on figshare at: dx.doi.org/10.6084/m9.figshare.23535738. Literature Cited The 1000 Genomes Project Consortium. 2015. A global reference for human genetic variation. Nature 526:68–74. Areal H, Abrantes J, Esteves PJ. 2011. Signatures of positive selection in Toll-like receptor (TLR) genes in mammals. BMC Evol Biol. 11:368. Barber MF, Elde NC. 2014. Escape from bacterial iron piracy through rapid evolution of transferrin. Science 346:1362–1366. Begun J, et al. 2015. Integrated genomics of Crohn’s disease risk vari- ant identifies a role for CLEC12A in antibacterial autophagy. Cell Rep. 11:1905–1918. Bergström A, et al. 2020. Insights into human genetic variation and population history from 929 diverse genomes. Science 367: eaay5012. Boguslawski KM, et al. 2020. Exploiting species specificity to under- stand the tropism of a human-specific toxin. Sci Adv. 6:eaax7515. Brown GD, et al. 2012. Hidden killers: human fungal infections. Sci Transl Med. 4(165):165rv13. Brown GD, Gordon S. 2001. A new receptor for β-glucans. Nature 413:36–37. Burley SK, et al. 2018. RCSB Protein Data Bank: biological macromol- ecular structures enabling research and education in fundamental biology, biomedicine, biotechnology and energy. Nucleic Acids Res. 47:D464–D474. Carey CM, Apple SE, Hilbert ZA, Kay MS, Elde NC. 2021. Diarrheal pathogens trigger rapid evolution of the guanylate cyclase-C sig- naling axis in bats. Cell Host Microbe. 29:1342–1350.e5. Carlin AF, et al. 2009. Molecular mimicry of host sialylated glycans al- lows a bacterial pathogen to engage neutrophil siglec-9 and dampen the innate immune response. Blood 113:3333–3336. Chang CC, et al. 2015. Second-generation PLINK: rising to the chal- lenge of larger and richer datasets. Gigascience 4:1–16. Chiaro TR, et al. 2023. Clec12a tempers inflammation while restricting expansion of a colitogenic commensal. Biorxiv. 2023.03.16.532997. doi:10.1101/2023.03.16.532997. Comstock LE, Kasper DL. 2006. Bacterial glycans: key mediators of di- verse host immune responses. Cell 126:847–850. Cummings RD, Chiffoleau E, van Kyook Y, McEver RP. 2022. Chapter 34: C-type lectins. In: Varki A et al., editors. Essentials of glycobiol- ogy [Internet]. 4th ed. Cold Spring Harbor, NY: Cold Spring Harbor Laboratory Press. doi:10.1101/glycobiology.4e.34. Danecek P, et al. 2011. The variant call format and VCFtools. Bioinformatics 27:2156–2158. Daugherty MD, Malik HS. 2012. Rules of engagement: molecular in- sights from host-virus arms races. Annu Rev Genet. 46:677–700. de Jong MAWP, et al. 2010. C-type lectin langerin is a β-glucan recep- tor on human langerhans cells that recognizes opportunistic and pathogenic fungi. Mol Immunol. 47:1216–1225. Denham ST, et al. 2018. Regulated release of cryptococcal polysac- charide drives virulence and suppresses immune cell infiltration into the central nervous system. Infect Immun. 86:e00662-17. de Witte L, et al. 2007. Langerin is a natural barrier to HIV-1 transmis- sion by Langerhans cells. Nat Med. 13:367–371. 14 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 Rapid Evolution of Glycan Recognition Receptors GBE Elde NC, Child SJ, Geballe AP, Malik HS. 2009. Protein kinase R reveals an evolutionary model for defeating viral mimicry. Nature 457: 485–489. Feinberg H, et al. 2011. Structural basis for langerin recognition of di- verse pathogen and mammalian glycans through a single binding site. J Mol Biol. 405:1027–1039. Fisher MC, et al. 2020. Threats posed by the fungal kingdom to hu- mans, wildlife, and agriculture. mBio 11: e00449–20. Fisher MC, et al. 2022. Tackling the emerging threat of antifungal re- sistance to human health. Nat Rev Microbiol. 20:557–571. GenomeAsia100K Consortium et al. 2019. The GenomeAsia 100K Project enables genetic discoveries across Asia. Nature 576: 106–111. George RD, et al. 2011. Trans genomic capture and sequencing of pri- mate exomes reveals new targets of positive selection. Genome Res. 21:1686–1694. Gow NAR, Latge J-P, Munro CA. 2017. The fungal cell wall: structure, biosynthesis, and function. Microbiol Spectr. 5:FUNK-0035-2016. Guth S, et al. 2022. Bats host the most virulent—but not the most dan- gerous—zoonotic viruses. Proc Natl Acad Sci. 119:e2113628119. Han BA, Schmidt JP, Bowden SE, Drake JM. 2015. Rodent reservoirs of future zoonotic diseases. Proc Natl Acad Sci. 112:7039–7044. Hanske J, et al. 2017. Bacterial polysaccharide specificity of the pattern recognition receptor langerin is highly species-dependent. J Biol Chem. 292:862–871. Herre J, Gordon S, Brown GD. 2004. Dectin-1 and its role in the recog- nition of β-glucans by macrophages. Mol Immunol. 40:869–876. Imperiali B. 2019. Bacterial carbohydrate diversity—a brave new world. Curr Opin Chem Biol. 53:1–8. Jégouzo SAF, et al. 2020. Mammalian lectin arrays for screening host– microbe interactions. J Biol Chem. 295:4541–4555. Judd EN, Gilchrist AR, Meyerson NR, Sawyer SL. 2021. Positive natural selection in primate genes of the type I interferon response. BMC Ecol Evol. 21:65. Jumper J, et al. 2021. Highly accurate protein structure prediction with AlphaFold. Nature 596:583–589. Kalyaanamoorthy S, Minh BQ, Wong TKF, von Haeseler A, Jermiin LS. 2017. Modelfinder: fast model selection for accurate phylogenetic estimates. Nat Methods. 14:587–589. Kawai T, Akira S. 2008. Toll-like receptor and RIG-1-like receptor sig- naling. Ann N Y Acad Sci. 1143:1–20. Li Z, Jaroszewski L, Iyer M, Sedova M, Godzik A. 2020. FATCAT 2.0: to- wards a better understanding of the structural diversity of proteins. Nucleic Acids Res. 48:W60–W64. Liu X, et al. 2017. Characterising private and shared signatures of positive selection in 37 Asian populations. Eur J Hum Genet. 25:499–508. Liu G, Huanxin Z, Zhao C, Honghai Z. 2019. Evolutionary history of the Toll-like receptor gene family across vertebrates. Genome Biol Evol. 12:3615–3634. Lonsdale J, et al. 2013. The genotype-tissue expression (GTEx) project. Nat Genet. 45:580–585. Lowy-Gallego E, et al. 2019. Variant calling on the GRCh38 assembly with the data from phase three of the 1000 genomes project. Wellcome Open Res. 4:50. Masuoka J. 2004. Surface glycans of Candida albicans and other pathogenic fungi: physiological roles, clinical uses, and experimen- tal challenges. Clin Microbiol Rev. 17:281–310. Medzhitov R. 2007. Recognition of microorganisms and activation of the immune response. Nature 449:819–826. Mitchell PS, et al. 2012. Evolution-guided identification of antiviral spe- cificity determinants in the broadly acting interferon-induced in- nate immunity factor MxA. Cell Host Microbe. 12:598–604. Murrell B, et al. 2012. Detecting individual sites subject to episodic di- versifying selection. PLoS Genet. 8:e1002764. Murrell B, et al. 2013. FUBAR: a fast, unconstrained Bayesian approxi- mation for inferring selection. Mol Biol Evol. 30:1196–1205. Murrell B, et al. 2015. Gene-wide identification of episodic selection. Mol Biol Evol. 32:1365–1371. Neumann K, et al. 2014. Clec12a is an inhibitory receptor for uric acid crystals that regulates inflammation in response to cell death. Immunity 40:389–399. Nguyen L-T, Schmidt HA, von Haeseler A, Minh BQ. 2015. IQ-TREE: a fast and effective stochastic algorithm for estimating maximum- likelihood phylogenies. Mol Biol Evol. 32:268–274. Nyame AK, Kawar ZS, Cummings RD. 2004. Antigenic glycans in para- sitic infections: implications for vaccines and diagnostics. Arch Biochem Biophys. 426:182–200. Park BS, et al. 2009. The structural basis of lipopolysaccharide recogni- tion by the TLR4–MD-2 complex. Nature 458:1191–1195. Poltorak A, et al. 1998. Defective LPS signaling in C3H/HeJ and C57BL/10ScCr mice: mutations in Tlr4 gene. Science 282: 2085–2088. Pond SLK, Frost SDW, Muse SV. 2005. Hyphy: hypothesis testing using phylogenies. Bioinformatics 21:676–679. Prado-Martinez J, et al. 2013. Great ape genetic diversity and popula- tion history. Nature 499:471–475. Prlić A, et al. 2010. Pre-calculated protein structure alignments at the RCSB PDB website. Bioinformatics 26:2983–2985. Raman R, Tharakaraman K, Sasisekharan V, Sasisekharan R. 2016. Glycan–protein interactions in viral pathogenesis. Curr Opin Struct Biol. 40:153–162. Raulf M-K, et al. 2019. The C-type lectin receptor CLEC12A recognizes plasmodial hemozoin and contributes to cerebral malaria develop- ment. Cell Rep. 28:30–38.e5. Rausell A, Telenti A. 2014. Genomics of host–pathogen interactions. Curr Opin Immunol. 30:32–38. R Core Team. 2022. R: a language and environment for statistical com- puting. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/. Rodrigues ML, et al. 1997. Identification of N-acetylneuraminic acid and its 9-O-acetylated derivative on the cell surface of Cryptococcus neoformans: influence on fungal phagocytosis. Infect Immun. 65:4937–4942. Sawyer SL, Wu LI, Emerman M, Malik HS. 2005. Positive selection of primate TRIM5α identifies a critical species-specific retroviral re- striction domain. Proc Natl Acad Sci. 102:2832–2837. Seyedmousavi S, et al. 2014. Black yeasts and their filamentous rela- tives: principles of pathogenesis and host defense. Clin Microbiol Rev. 27:527–542. Siewert KM, Voight BF. 2017. Detecting long-term balancing selection using allele frequency correlation. Mol Biol Evol. 34:2996–3005. Siewert KM, Voight BF. 2020. Betascan2: standardized statistics to de- tect balancing selection utilizing substitution data. Genome Biol Evol. 12:3873–3877. Soares RMA, et al. 2000. Identification of sialic acids on the cell surface of Candida albicans. Biochim Biophys Acta. 1474: 262–268. Sparber F, et al. 2018. Langerin+ DCs regulate innate IL-17 production in the oral mucosa during Candida albicans-mediated infection. PLoS Pathog. 14:e1007069. Speidel L, et al. 2021. Inferring population histories for ancient gen- omes using genome-wide genealogies. Mol Biol Evol. 38:3497. Speidel L, Forest M, Shi S, Myers SR. 2019. A method for genome-wide genealogy estimation for thousands of samples. Nat Genet. 51: 1321–1329. Stappers MHT, et al. 2018. Recognition of DHN-melanin by a C-type lectin receptor is required for immunity to Aspergillus. Nature 555:382–386. Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023 15 Hilbert et al. GBE Stowell SR, et al. 2014. Microbial glycan microarrays define key fea- tures of host-microbial interactions. Nat Chem Biol. 10:470–476. Tan X, Sun L, Chen J, Chen ZJ. 2015. Detection of microbial infections through innate immune sensing of nucleic acids. Annu Rev Microbiol. 72:447–478. Tateno H, et al. 2010. Dual specificity of Langerin to sulfated and man- nosylated glycans via a single C-type carbohydrate recognition do- main*. J Biol Chem. 285:6390–6400. Tenthorey JL, Young C, Sodeinde A, Emerman M, Malik HS. 2020. Mutational resilience of antiviral restriction favors primate TRIM5α in host-virus evolutionary arms races. Elife 9:e59988. Tone K, et al. 2021. Mellec exacerbates the pathogenesis of Aspergillus fumigatus-induced allergic inflammation in mice. Front Immunol. 12:675702. Trifinopoulos J, Nguyen L-T, von Haeseler A, Minh BQ. 2016. W-IQ-TREE: a fast online phylogenetic tool for maximum likelihood analysis. Nucleic Acids Res. 44:W232–W235. Vance RE, Isberg RR, Portnoy DA. 2009. Patterns of pathogenesis: discrimination of pathogenic and nonpathogenic microbes by the innate immune system. Cell Host Microbe. 6:10–21. Van Dalen R, et al. 2019. Langerhans cells sense Staphylococcus aur- eus wall teichoic acid through langerin to induce inflammatory re- sponses. mBio 10:e00330-19. van der Vlist M, et al. 2011. Human Langerhans cells capture measles virus through Langerin and present viral antigens to CD4+ T cells but are incapable of cross-presentation. Eur J Immunol. 41:2619–2631. Van Kooyk Y, Rabinovich GA. 2008. Protein-glycan interactions in the control of innate and adaptive immune responses. Nat Immunol. 9: 593–601. Varki A. 2011. Letter to the glyco-forum: since there are PAMPs and DAMPs, there must be SAMPs? Glycan “self-associated molecular patterns” dampen innate immunity, but pathogens can mimic them. Glycobiology 21:1121–1124. Varki A, Gagneux P. 2012. Multifarious roles of sialic acids in immunity. Ann N Y Acad Sci. 1253:16–36. Venkat A, Hahn MW, Thornton JW. 2018. Multinucleotide mutations cause false inferences of lineage-specific positive selection. Nat Ecol Evol. 2:1280–1288. Vigerust DJ, Shepherd VL. 2007. Virus glycosylation: role in virulence and immune interactions. Trends Microbiol. 15:211–218. Voight BF, Kudaravalli S, Wen X, Pritchard JK. 2006. A map of recent positive selection in the human genome. PLoS Biol. 4:e72. Wang W, Han G-Z. 2021. Pervasive positive selection on virus receptors driven by host-virus conflicts in mammals. J Virol. 95:e01029-21. Werling D, Jann OC, Offord V, Glass EJ, Coffey TJ. 2009. Variation mat- ters: TLR structure and species-specific pathogen recognition. Trends Immunol. 30:124–130. Wlasiuk G, Nachman MW. 2010. Adaptation and constraint at Toll-like receptors in primates. Mol Biol Evol. 27:2172–2186. Yang Z. 2007. PAML 4: phylogenetic analysis by maximum likelihood. Mol Biol Evol. 24:1586–1591. Yang K, et al. 2015. Host Langerin (CD207) is a receptor for Yersinia pestis phagocytosis and promotes dissemination. Immunol Cell Biol. 93:815–824. Zheng-Bradley X, et al. 2017. Alignment of 1000 Genomes Project reads to reference assembly GRCh38. Gigascience 6:1–8. Associate editor: Dr. George Zhang 16 Genome Biol. Evol. 15(7) https://doi.org/10.1093/gbe/evad119 Advance Access publication 30 June 2023
10.1093_gbe_evad099	GBE Automatic Differentiation is no Panacea for Phylogenetic Gradient Computation Mathieu Fourment Marc A. Suchard 1, Christiaan J. Swanepoel2,3, Jared G. Galloway4, Xiang Ji5, Karthik Gangavarapu6, 4,9,10,11,* 6,7,8,, and Frederick A. Matsen IV 1Australian Institute for Microbiology and Infection, University of Technology Sydney, Ultimo, NSW, Australia 2Centre for Computational Evolution, The University of Auckland, Auckland, New Zealand 3School of Computer Science, The University of Auckland, Auckland, New Zealand 4Public Health Sciences Division, Fred Hutchinson Cancer Research Center, Seattle, Washington, USA 5Department of Mathematics, Tulane University, New Orleans, Louisiana, USA 6Department of Human Genetics, University of California, Los Angeles, California, USA 7Department of Computational Medicine, University of California, Los Angeles, California, USA 8Department of Biostatistics, University of California, Los Angeles, California, USA 9Department of Statistics, University of Washington, Seattle, Washington, USA 10Department of Genome Sciences, University of Washington, Seattle, Washington, USA 11Howard Hughes Medical Institute, Fred Hutchinson Cancer Research Center, Seattle, Washington, USA Corresponding authors: E-mails: [email protected]; [email protected]. Accepted: 25 May 2023 Abstract Gradients of probabilistic model likelihoods with respect to their parameters are essential for modern computational statistics and machine learning. These calculations are readily available for arbitrary models via “automatic differentiation” implemented in general-purpose machine-learning libraries such as TensorFlow and PyTorch. Although these libraries are highly optimized, it is not clear if their general-purpose nature will limit their algorithmic complexity or implementation speed for the phylogenetic case compared to phylogenetics-specific code. In this paper, we compare six gradient implementations of the phylogenetic like- lihood functions, in isolation and also as part of a variational inference procedure. We find that although automatic differen- tiation can scale approximately linearly in tree size, it is much slower than the carefully implemented gradient calculation for tree likelihood and ratio transformation operations. We conclude that a mixed approach combining phylogenetic libraries with ma- chine learning libraries will provide the optimal combination of speed and model flexibility moving forward. Key words: phylogenetics, Bayesian inference, variational inference, gradient. Significance Bayesian phylogenetic analysis plays an essential role in understanding how organisms evolve, and is widely used as a tool for genomic surveillance and epidemiology studies. The classical Markov chain Monte Carlo algorithm is the engine of most Bayesian phylogenetic software, however, it becomes impractical when dealing with large datasets. To address this issue, more efficient methods leverage gradient information, albeit at the cost of increased computational demands. Here we present a benchmark comparing the efficiency of automatic differentiation implemented in general-purpose libraries against analytical gradients implemented in specialized phylogenetic tools. Our findings indicate that imple- menting analytical gradients for the computationally intensive components of the phylogenetic model significantly en- hances the efficiency of the inference algorithm. © The Author(s) 2023. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 1 Fourment et al. GBE Introduction Gradients (i.e. multidimensional derivatives) of probabilistic model likelihoods with respect to their unknown para- meters are essential for modern computational statistics and machine learning. For example, gradient-based Hamiltonian Monte Carlo (HMC) (Neal 2011), implemented in the Stan statistical framework (Carpenter et al. 2017), is a cornerstone of the modern Bayesian statistical toolbox. Variational Bayesian (VB) inference algorithms (Blei et al. 2017), which use gradients to improve fit of a variational distribution to the posterior, are another key modern tech- nique. In the more general setting of machine learning, gra- dients are used to train predictive models such as deep neural networks. Although gradients have been considered for a long time in phylogenetics (Schadt et al. 1998; Kenney and Gu 2012), they are now becoming of central importance to en- able faster approaches to Bayesian phylogenetic analysis. Bayesian methods have gained popularity among phylo- genetic practitioners due to their ability to integrate mul- tiple data sources, including ecological factors (Lemey et al. 2020) and clinical outcomes (Bedford et al. 2014) into a single analysis. A drawback of these methods is scal- ability, as it is well known that Bayesian phylogenetic packages, such as BEAST (Suchard et al. 2018), struggle with datasets containing thousands of sequences with moderately complex models. Bayesian phylogenetic ana- lysis typically uses classical Markov chain Monte Carlo (MCMC) and therefore does not need to calculate compu- tationally intensive gradients. In order to go beyond classical MCMC, recent research has developed HMC (Fisher et al. 2021) and Variational Bayes phylogenetic analysis (Dang and Kishino 2019; Fourment and Darling 2019; Zhang and Matsen 2019; Liu et al. 2021; Moretti et al. 2021; Ki and Terhorst 2022; Koptagel et al. 2022; Zhang and Matsen 2022). These methods require fast and efficient gradient calcula- tion algorithms to give viable alternatives to MCMC. Correspondingly, recent work has developed fast algo- rithms and implementations of phylogenetic likelihood gra- dient calculation (Ji et al. 2020) in the BEAGLE (Ayres et al. 2019) library. Outside of phylogenetics, gradient-based analysis has also exploded in popularity, in part driven by easy to use software libraries that provide gradients via automatic dif- ferentiation (AD). AD libraries “record” function composi- tions, have gradients on hand for component functions, and combine these simple gradients together via the chain rule (see Margossian 2019 for a review). This work has, re- markably, been extended to many computable operations that are not obviously differentiable such as dynamic con- trol flow and unbounded iteration (Yu et al. 2018). These libraries, exemplified by TensorFlow (Abadi 2016) and PyTorch (Paszke et al. 2019), are often developed by large dedicated teams of professional programmers. The combination of these various advances raises a num- ber of questions. Can we rely on AD exclusively in phyloge- netics, and avoid calculating gradients using hand-crafted algorithms? How do AD algorithms scale when presented with interdependent calculations on a tree? Does perform- ance depend on the package used? In this paper, we address these questions by performing the first benchmark analysis of AD versus carefully imple- mented gradient algorithms in compiled languages. We find that AD algorithms vary widely in performance de- pending on the backend library, the dataset size and the model/function under consideration. All of these AD imple- mentations are categorically slower than libraries designed specifically for phylogenetics; we do, however, find that they appear to scale roughly linearly in tree size. Moving forward, these results suggest an architecture in which core phylogenetic likelihood and branch-length transform- ation calculations are performed in specialized libraries, whereas rich models are formulated, and differentiated, in a machine learning library such as PyTorch or TensorFlow. Results Overview of Benchmarking Setup To coherently describe our results, we first provide a suc- cinct overview of the phylogenetic and machine learning packages that we will benchmark as well as the computa- tional tasks involved. We benchmark two packages where the core algorithm implementation is specialized to phylogenetics: BEAGLE (Ji et al. 2020), wrapped by our Python-interface C++ library bito, as well as physher (Fourment and Holmes 2014). The bito library also efficiently implements gradients of the ratio transformation, following (Ji et al. 2021), for uncon- strained node-height optimization. We compare these to the most popular AD libraries available, namely TensorFlow (Abadi 2016), PyTorch (Paszke et al. 2019), JAX (Bradbury et al. 2018), and Stan (Carpenter et al. 2017). These are leveraged in phylogenetics via treeflow, torchtree, phylojax, and phylostan (Fourment and Darling 2019), respectively. When using AD, these programs make use of reverse-mode automatic differentiation. Every pro- gram uses double precision unless specified otherwise. We divide the benchmarking into two flavors: a “micro-” and “macro-” benchmark. The macrobenchmark is meant to mimic running an actual inference algorithm, though stripped down to reduce the burden of implementing a complex model in each framework. Specifically, we infer parameters of a constant size coalescent process, strict clock, as well as node heights under a typical continuous- time Markov chain (CTMC) model for character substitution 2 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 Phylogenetic Gradient Benchmark GBE along an unknown phylogeny. Every implementation uses the automatic differentiation variational inference (ADVI) framework (Kucukelbir et al. 2017) to maximize the evi- dence lower bound (ELBO) over 5000 iterations. A priori we assume the CTMC substitution rate is exponentially dis- tributed with mean 0.001 and we use the Jeffrey’s prior for the unknown population size parameter. The microbenchmark, on the other hand, is meant to identify which parts of a phylogenetic model are the most computationally expensive in the context of gradient-based inference. This involves evaluating likelihoods and functions used in phylogenetic analysis and calculating their gradient (1) the phylogenetic likelihood, (2) the coalescent likeli- hood, (3) node-height transform, and (4) the determinant of the Jacobian of the node-height transform. Specifically, these tasks are: 1. Phylogenetic likelihood: the likelihood of observing an alignment under the Jukes–Cantor substitution mod- el (Jukes and Cantor 1969) is efficiently calculated using the pruning algorithm (Felsenstein 1981) requiring O(N) operations where N is the number of taxa. In this bench- mark, the derivatives are taken with respect to the branch lengths. Although a naive implementation of the gradient calculation requires O(N2) calculations, ef- ficient implementations (Fourment and Holmes 2014; Ji et al. 2020) necessitate only O(N) operations. We also benchmark the tree likelihood using the GTR substi- tution model. The gradient with respect to the GTR parameters is calculated analytically in physher while bito utilizes finite differences. Analytical gradients of the tree likelihood require O(N) operations for each of the eight free parameters while numerical gradients require two evaluations of the tree likelihood per parameter. 2. Coalescent likelihood: the likelihood of observing a phylogeny is calculated using the constant size popula- tion coalescent model (Kingman 1982). The gradient with respect to the node heights and the population size parameter requires O(N) time. 3. Node-height transform: Node ages of time trees need to be reparameterized in order to perform uncon- strained optimization (Fourment and Holmes 2014; Ji et al. 2021). Evaluating this function requires a single preorder traversal and requires O(N) operations. 4. Determinant of the Jacobian of the node-height transform: The transformation of the node ages re- quires an adjustment to the joint density through the in- clusion of the determinant of the Jacobian of the transform (Fourment and Darling 2019). The Jacobian is triangular and the determinant is therefore straight- forward to compute. Although calculating its gradient analytically is not trivial, requiring O(N2) calculations, re- cent work (Ji et al. 2021) proposed an O(N) algorithm. The derivatives are taken with respect to the node heights. AD Implementations Vary Widely in Performance, and Custom Gradients are Far Faster We find that on the macrobenchmark, AD implementations vary widely in their speed (fig. 1). This is remarkable given that these are highly optimized libraries doing the same flavor of operations. Specifically, both just-in-time (JIT) FIG. 1.—Speed of implementations for 5000 iterations of variational time-tree inference with a strict clock. See supplementary figure S1, Supplementary Material online for results without phylojax. Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 3 Fourment et al. GBE compiled JAX and compiled TensorFlow use XLA as a back- end, although they have strikingly different performance. (We note that this is now a known issue with JAX https:// github.com/google/jax/issues/10197.) Specifically, JAX was the only package that clearly scales quadratically in the number of tips. Moreover, PyTorch was several times faster than TensorFlow for our tasks of interest, which was surprising to us because of PyTorch uses a dynamic computation graph. Results for phylojax with datasets larger than 750 sequences are not reported as they ex- ceeded the maximum allocated computation time. None of these AD libraries approach the speed of hand- coded phylogenetic gradients. The BEAGLE gradients wrapped in bito and gradients computed in physher show comparable performance, which are at least eight times the speed of the fastest AD implementation (supplementary fig. S2, Supplementary Material online). As expected, memory usage of the pure C program physher is the smallest, while torchtree is less memory heavy than treeflow and phylostan’s memory usage in- creases significantly more rapidly (supplementary fig. S3, Supplementary Material online). It is worth noting that bito noticeably decreases the memory usage of torchtree. Overall using a specialized library for the tree likelihood within a Python program greatly improves the performance of a program making use of gradient-based optimization (e.g. ADVI, HMC) while incurring a small performance and memory cost compared to a fully C-based tool. Relative Performance of AD Depends on the Task To break down our inferential task into its components, we then performed a “microbenchmark” divided into the in- gredients needed for doing gradient-based inference (fig. 2 and supplementary fig. S4, Supplementary Material online). See Methods for a precise description of the indi- vidual tasks. Across tasks, we see the following shared fea- tures. The specialized phylogenetic packages (bito/ BEAGLE and physher) perform similarly to one another and are consistently faster than the AD packages, except for the Jacobian task. As expected, the tree likelihood is the computational and memory bottleneck (fig. 2 and supplementary fig. S3, Supplementary Material online) in phylogenetic models and efficient gradient calculation are warranted. TensorFlow-based treeflow was the slowest implementation across the board after excluding JAX. FIG. 2.—Speed of implementations for the gradient of various tasks needed for inference. See text for description of the tasks. JAX is excluded from this plot due to slow performance stretching the y-axis; see supplementary figure S5, Supplementary Material online for JAX. See supplementary figure S6, Supplementary Material online for function evaluations. 4 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 Phylogenetic Gradient Benchmark GBE Table 1 Code Availability and Version Number of Each Phylogenetic Program. Version Identifiers Correspond to Git Tags. Program bito Availability Version https://github.com/ autodiff-benchmark phylojax https://github.com/4ment/ v1.0.1 phylovi/bito phylostan https://github.com/4ment/ v1.0.5 phylojax phylostan online). In contrast, enabling JIT in torchtree showed no improvement and was not included in the results. The calculation of the tree likelihood and its gradient were sig- nificantly slower using single precision for datasets larger than 500 sequences. This is because torchtree, like most phylogenetic programs, rescales partial likelihood vec- tors in order to avoid underflow; using single precision re- quires more rescaling operations. physher https://github.com/4ment/ v2.0.0 Discussion physher torchtree https://github.com/4ment/ gradient-benchmark torchtree torchtree-bito https://github.com/4ment/ gradient-benchmark treeflow torchtree-bito https://github.com/ christiaanjs/treeflow autodiff-benchmark The AD programs also performed significantly worse in the node-height transform and tree likelihood tasks. Function calls in python are notoriously more expensive than in C and C++, potentially explaining the decrease in performance for algorithms involving a tree traversal. In addition, the tree likelihood implementations in BEAGLE and physher are highly optimized with SSE vectorization (Ayres et al. 2019) and manual loop unrolling. The calculations of the coalescent function and its gradi- ent were slightly faster in physher than in torchtree, although the difference was slight. The ratio transform has nontrivial computational expense—comparable to the phylogenetic likelihood gradient—in AD packages; how- ever, specialized algorithms for calculating these gradients scale much better. Interestingly, for large datasets, torchtree outperforms the specialized phylogenetic packages for the Jacobian ratio transform gradient calcula- tion. Since this is the fastest task, the overall execution time is not, however, significantly impacted. The phylogenetic gradient is approximately linear for packages other than JAX (supplementary fig. S7, Supplementary Material online), although the specialized phylogenetic packages are about 10 times faster. For the GTR calculation, we actually compare two flavors of evalu- ation: finite differences for bito and analytic gradients for physher. As expected, bito is increasingly faster than physher as the datasets increase in size. With the exception of the tree likelihood, JAX’s JIT cap- abilities greatly improved the performance of the algo- rithms in the microbenchmark (supplementary fig. S8, Supplementary Material online). Analytically calculating the gradient of the tree likelihood considerably improved the running time of phylojax pointing at implementa- tion issues in the gradient function in JAX for this type of al- gorithm (supplementary fig. S8, Supplementary Material We have found that, although AD packages provide unri- valed flexibility for model development and flexible like- lihood formulation, they cannot compete with carefully implemented gradients in compiled languages. Furthermore, they do differ between each other significantly in computa- tion time and memory usage for phylogenetic tasks. Our results motivate the design of bito: leverage specia- lized algorithms for phylogenetic gradients and ratio trans- forms, but wrap them in a way that invites model flexibility. In this paper, we have focused on two functionalities of bito: first as a wrapper for the high-performance BEAGLE library, and second, as a fast means of computing the ratio transforms. This is our first publication using this library, which will be the computational core of our future work on Bayesian phylogenetic inference via optimization. We will defer a more comprehensive description of bito to future work. Our results also motivate us to focus our future model developments using the PyTorch library, which shows the best performance as well as ease of use. Our study has the following limitations. First, these librar- ies are developing quickly and they may gain substantially in efficiency in future versions. Second, these results concern CPU computation only. Future work, including development of phylogenetic gradients using graphics processing units (GPUs), will evaluate the promise of GPUs for gradient-based inference. However, we note that initial results using GPUs for AD packages did not lead to a significant speedup. Methods Data To evaluate the performance of each implementation, we reused parts of the validation workflow introduced by Sagulenko et al. (2018). The data in this workflow consist of a collection of influenza A datasets ranging from 20 to 2000 sequences sampled from 2011 to 2013. Our bench- mark is built on top of this pipeline and makes use of a reproducible Nextflow (Di Tommaso et al. 2017) pipeline. Software Benchmarked torchtree is a Python-based tool that leverages the Pytorch library to calculate gradients using reverse mode AD. Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 5 Fourment et al. GBE torchtree-bito is a torchtree plugin that offers an interface to the bito library (https://github.com/phylovi/ bito). Within bito, analytical derivatives with respect to the branch lengths are calculated through the BEAGLE library (Ayres et al. 2019; Ji et al. 2020) while the gradient with re- spect to the GTR substitution model parameters are calculated numerically using finite differences. bito and BEAGLE do not provide analytical derivatives of the coalescent function, hence no results are shown in figure 2 and supplementary figures S4–S7, Supplementary Material online. physher is a C program that allows one to approximate distributions using ADVI (Fourment et al. 2020), while every derivative is calculated analytically. The derivatives with re- spect to the branch lengths are efficiently calculated using a linear-time algorithm developed independently of Ji et al. (2020). The gradient of the Jacobian transform is efficiently calculated using the method proposed by Ji et al. (2021). phylostan is a Python-based program (Fourment and Darling 2019) that generates phylogenetic models that are compatible with the Stan package. phylojax is a Python-based tool that leverages the JAX library to calculate gradients using reverse mode AD. treeflow is a Python-based tool that leverages the TensorFlow library to calculate gradients using reverse mode AD. treeflow’s implementation of the phylogenetic likeli- hood uses TensorFlow’s TensorArray construct (Yu et al. 2018), a data structure which represents a collection of arrays. Each array can only be written once in a computation, and read many times. Using these data structure to implement the dynamic programming steps of the pruning algorithm po- tentially allows for more scalable gradient computations. Computational Infrastructure The automated workflow was run using the Fred Hutchinson gizmo scientific computing infrastructure. A single node with 36 (2 sockets by 18 cores) Intel ® Xeon Gold 6254 CPU @ 3.10GHz cores was used for all individual processes in the pipeline. A total of 48G RAM was allo- cated. The node was running on Ubuntu 18.04.5 LTS (Bionic Beaver) with Nextflow (version 22.04.3.5703) and Singularity (version 3.5.3) modules installed. Supplementary Material Supplementary data are available at Genome Biology and Evolution online. Acknowledgments We are grateful to Jonathan Terhorst for discussions concerning phylogenetic gradients in JAX. This work was supported through US National Institutes of Health grants AI162611 and AI153044. Scientific Computing Infrastructure at Fred Hutch was funded by ORIP grant S10OD028685. Computational facilities were provided by the UTS eResearch High-Performance Compute Facilities. Dr. Matsen is an Investigator of the Howard Hughes Medical Institute. Data Availability The Nextflow pipeline is available from https://github.com/ 4ment/gradient-benchmark. The versions of the programs used in this study are provided in table 1. Literature Cited Abadi M, et al. 2016. TensorFlow: a system for large-scale machine learning. In: Proceedings of the 12th USENIX Conference on Operating Systems Design and Implementation, OSDI’16; 2016 Nov; USA. USENIX Association. p. 265–283. Available from: https://dl.acm.org/doi/10.5555/3026877.3026899. Ayres DL, et al. 2019. BEAGLE 3: improved performance, scaling, and usability for a high-performance computing library for statistical phylogenetics. Syst Biol. 68(6):1052–1061. Bedford T, et al. 2014. Integrating influenza antigenic dynamics with molecular evolution. elife. 3:e01914. Blei DM, Kucukelbir A, McAuliffe JD. 2017. Variational inference: a re- view for statisticians. J Am Stat Assoc. 112(518):859–877. doi:10. 1080/01621459.2017.1285773 Bradbury J, et al. 2018. JAX: composable transformations of Python +NumPy programs. Available from: http://github.com/google/jax. Carpenter B, et al. 2017. Stan: a probabilistic programming language. J Stat Softw. 76(1):1–32. doi:10.18637/jss.v076.i01 Dang T, Kishino H. 2019. Stochastic variational inference for Bayesian phylogenetics: a case of CAT model. Mol Biol Evol. 36(4):825–833. doi:10.1093/molbev/msz020 Di Tommaso P, et al. 2017. Nextflow enables reproducible computa- tional workflows. Nat Biotechnol. 35(4):316–319. doi:10.1038/ nbt.3820 Felsenstein J. 1981. Evolutionary trees from DNA sequences: a max- imum likelihood approach. J Mol Evol. 17(6):368–376. Fisher AA, Ji X, Zhang Z, Lemey P, Suchard MA. 2021. Relaxed random walks at scale. Syst Biol. 70(2):258–267. doi:10.1093/sysbio/ syaa056 Fourment M, Darling AE. 2019. Evaluating probabilistic programming and fast variational Bayesian inference in phylogenetics. PeerJ. 7: e8272.doi:10.7717/peerj.8272 Fourment M, Holmes EC. 2014. Novel non-parametric models to esti- mate evolutionary rates and divergence times from heterochro- nous sequence data. BMC Evol Biol. 14:163. doi:10.1186/ s12862-014-0163-6 Fourment M, et al. 2020. 19 dubious ways to compute the marginal like- lihood of a phylogenetic tree topology. Syst Biol. 69(2):209–220. Ji X, et al. 2020. Gradients do grow on trees: a linear-time O(N)-dimensional gradient for statistical phylogenetics. Mol Biol Evol. 37(10):3047–3060. doi:10.1093/molbev/msaa130 Ji X, et al. 2021. Scalable Bayesian divergence time estimation with ra- tio transformations; October. Available from: http://arxiv.org/abs/ 2110.13298. Jukes TH, Cantor CR. 1969. Evolution of protein molecules. In: Mammalian protein metabolism. Vol. 3. New York: Academic Press. p. 21–132. Kenney T, Gu H. 2012. Hessian calculation for phylogenetic likelihood based on the pruning algorithm and its applications. Stat Appl Genet Mol Biol. 11(4):Article 14. doi:10.1515/1544-6115.1779 6 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 Phylogenetic Gradient Benchmark GBE Ki C, Terhorst J. 2022. Variational phylodynamic inference using pandemic-scale data. Mol Biol Evol. 39(8):msac154. doi:10.1093/ molbev/msac154 Kingman JFC. 1982. The coalescent. Stoch Process Appl. 13(3):235–248. Koptagel H, Kviman O, Melin H, Safinianaini N, Lagergren J. 2022. VaiPhy: a variational inference based algorithm for phylogeny, March. Available from: http://arxiv.org/abs/2203.01121. Kucukelbir A, Tran D, Ranganath R, Gelman A, Blei DM. 2017. Automatic differentiation variational inference. J Mach Learn Res. 18(1):430–474. Lemey P, et al. 2020. Accommodating individual travel history, global in phylogeography: a mobility, and unsampled diversity SARS-CoV-2 case study. bioRxiv. Liu X, Ogilvie HA, Nakhleh L. 2021. Variational inference using ap- proximate likelihood under the coalescent with recombination. Genome Res. 31(11):2107–2119. doi:10.1101/gr.273631.120 Margossian CC. 2019. A review of automatic differentiation and its ef- ficient implementation. Wiley Interdiscip Rev Data Min Knowl Discov. 9(4):e1305. Moretti AK, et al. 2021. Variational combinatorial sequential Monte In: Carlo methods Uncertainty in artificial intelligence. PMLR. p. 971–981. for Bayesian phylogenetic inference. Neal R. 2011. MCMC using Hamiltonian dynamics. In: Brooks S, Gelman A, Jones G, Meng XL, editors. Handbook of Markov chain Monte Carlo. Chapman & Hall/CRC Handbooks of Modern Statistical Methods. Taylor & Francis. Available from: http:// books.google.com/books?id=qfRsAIKZ4rIC. Paszke A, et al. 2019. PyTorch: an imperative style, high-performance deep learning library; December. Available from: https://papers. nips.cc/paper/2019/file/bdbca288fee7f92f2bfa9f7012727740- Paper.pdf. Sagulenko P, Puller V, Neher RA. 2018. TreeTime: maximum-likelihood phylodynamic analysis. Virus Evol. 4(1):vex042. Schadt EE, Sinsheimer JS, Lange K. 1998. Computational advances in maximum likelihood methods for molecular phylogeny. Genome Res. 8(3):222–233. doi:10.1101/gr.8.3.222 Suchard MA, et al. 2018. Bayesian phylogenetic and phylodynamic data integration using beast 1.10. Virus Evol. 4(1):vey016. Yu Y, et al. 2018. Dynamic control flow in large-scale machine learn- ing. In: Proceedings of the Thirteenth EuroSys Conference. New York: Association for Computing Machinery (ACM). p. 1–15. Zhang C, Matsen FA IV. 2019. Variational Bayesian phylogenetic infer- ence. In: International Conference on Learning Representations (ICLR). New Orleans: OpenReview.net. Available from: https:// openreview.net/pdf?id=SJVmjjR9FX. Zhang C, Matsen FA IV. 2022. A variational approach to Bayesian phylogenetic inference, April. Available from: http://arxiv.org/abs/ 2204.07747. Associate editor: Tom Williams Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad099 Advance Access publication 2 June 2023 7
10.1103_physrevd.103.104047	University of Southern Denmark Constraints on black-hole charges with the 2017 EHT observations of M87* Kocherlakota, Prashant; Rezzolla, Luciano; Falcke, Heino; Fromm, Christian M.; Kramer, Michael; Mizuno, Yosuke; Nathanail, Antonios; Olivares, Hector; Younsi, Ziri; Akiyama, Kazunori; Alberdi, Antxon; Alef, Walter; Algaba, Juan Carlos; Anantua, Richard; Asada, Keiichi; Azulay, Rebecca; Baczko, Anne-Kathrin; Ball, David; Balokovic, Mislav; Barrett, John; Benson, Bradford A.; Bintley, Dan; Blackburn, Lindy; Blundell, Raymond; Boland, Wilfred; Bouman, Katherine L.; Bower, Geoffrey C.; Boyce, Hope; Bremer, Michael; Brinkerink, Christiaan D.; Brissenden, Roger; Britzen, Silke; Broderick, Avery E.; Broguiere, Dominique; Bronzwaer, Thomas; Byun, Do-Young; Carlstrom, John E.; Chael, Andrew; Chan, Chi-kwan; Chatterjee, Shami; Chatterjee, Koushik; Chen, Yongjun; Chesler, Paul M.; Cho, Ilje; Christian, Pierre; Conway, John E.; Cordes, James M.; Crawford, Thomas M.; Crew, Geoffrey B.; Cruz- Osorio, Alejandro; Cui, Yuzhu; Davelaar, Jordy; Laurentis, Mariafelicia De; Deane, Roger; Dempsey, Jessica; Desvignes, Gregory; Doeleman, Sheperd S.; Eatough, Ralph P.; Farah, Joseph; Fish, Vincent L.; Fomalont, Ed; Fraga-Encinas, Raquel; Friberg, Per; Ford, H. Alyson; Fuentes, Antonio; Galison, Peter; Gammie, Charles F.; Garcia, Roberto; Gentaz, Olivier; Georgiev, Boris; Goddi, Ciriaco; Gold, Roman; Gomez, Jose L.; Gomez-Ruiz, Arturo I.; Gu, Minfeng; Gurwell, Mark; Hada, Kazuhiro; Haggard, Daryl; Hecht, Michael H.; Hesper, Ronald; Ho, Luis C.; Ho, Paul; Honma, Mareki; Huang, Chih-Wei L.; Huang, Lei; Hughes, David H.; Ikeda, Shiro; Inoue, Makoto; Issaoun, Sara; James, David J.; Jannuzi, Buell T.; Janssen, Michael; Jeter, Britton; Jiang, Wu; Jimenez-Rosales, Alejandra; Johnson, Michael D.; Jorstad, Svetlana; Jung, Taehyun; Karami, Mansour; Karuppusamy, Ramesh; Kawashima, Tomohisa; Keating, Garrett K.; Kettenis, Mark; Kim, Dong-Jin; Kim, Jae-Young; Kim, Jongsoo; Kim, Junhan; Kino, Motoki; Koay, Jun Yi; Kofuji, Yutaro; Koch, Patrick M.; Koyama, Shoko; Kramer, Carsten; Krichbaum, Thomas P.; Kuo, Cheng-Yu; Lauer, Tod R.; Lee, Sang-Sung; Levis, Aviad; Li, Yan-Rong; Li, Zhiyuan; Lindqvist, Michael; Lico, Rocco; Lindahl, Greg; Liu, Jun; Liu, Kuo; Liuzzo, Elisabetta; Lo, Wen-Ping; Lobanov, Andrei P.; Loinard, Laurent; Lonsdale, Colin; Lu, Ru-Sen; MacDonald, Nicholas R.; Mao, Jirong; Marchili, Nicola; Markoff, Sera; Marrone, Daniel P.; Marscher, Alan P.; Marti-Vidal, Ivan; Matsushita, Satoki; Matthews, Lynn D.; Medeiros, Lia; Menten, Karl M.; Mizuno, Izumi; Moran, James M.; Moriyama, Kotaro; Moscibrodzka, Monika; Muller, Cornelia; Musoke, Gibwa; Mejias, Alejandro Mus; Nagai, Hiroshi; Nagar, Neil M.; Nakamura, Masanori; Narayan, Ramesh; Narayanan, Gopal; Natarajan, Iniyan; Neilsen, Joseph; Neri, Roberto; Ni, Chunchong; Noutsos, Aristeidis; Nowak, Michael A.; Okino, Hiroki; Ortiz-Leon, Gisela N.; Oyama, Tomoaki; Ozel, Feryal; Palumbo, Daniel C. M.; Park, Jongho; Patel, Nimesh; Pen, Ue-Li; Pesce, Dominic W.; Pietu, Vincent; Plambeck, Richard; PopStefanija, Aleksandar; Porth, Oliver; Potzl, Felix M.; Prather, Ben; Preciado-Lopez, Jorge A.; Psaltis, Dimitrios; Pu, Hung-Yi; Ramakrishnan, Venkatessh; Rao, Ramprasad; Rawlings, Mark G.; Raymond, Alexander W.; Ricarte, Angelo; Ripperda, Bart; Roelofs, Freek; Rogers, Alan; Ros, Eduardo; Rose, Mel; Roshanineshat, Arash; Rottmann, Helge; Roy, Alan L.; Ruszczyk, Chet; Rygl, Kazi L. J.; Sanchez, Salvador; Sanchez-Arguelles, David; Sasada, Mahito; Savolainen, Tuomas; Schloerb, F. Peter; Schuster, Karl-Friedrich; Shao, Lijing; Shen, Zhiqiang; Small, Des; Sohn, Bong Won; SooHoo, Jason; Sun, He; Tazaki, Fumie; Tetarenko, Alexandra J.; Tiede, Paul; Tilanus, Remo P. J.; Titus, Michael; Toma, Kenji; Torne, Pablo; Trent, Tyler; Traianou, Efthalia; Trippe, Sascha; Bemmel, Ilse van; Langevelde, Huib Jan van; Rossum, Daniel R. van; Wagner, Jan; Ward- Thompson, Derek; Wardle, John; Weintroub, Jonathan; Wex, Norbert; Wharton, Robert; Wielgus, Maciek; Wong, George N.; Wu, Qingwen; Yoon, Doosoo; Young, Andre; Young, Ken; Yuan, Feng; Yuan, Ye-Fei; Zensus, J. Anton; Zhao, Guang-Yao; Zhao, Shan-Shan; Collaboration, The EHT Published in: Physical Review D DOI: 10.1103/PhysRevD.103.104047 Publication date: 2021 Document version: Final published version Document license: CC BY Citation for pulished version (APA): Kocherlakota, P., Rezzolla, L., Falcke, H., Fromm, C. M., Kramer, M., Mizuno, Y., Nathanail, A., Olivares, H., Younsi, Z., Akiyama, K., Alberdi, A., Alef, W., Algaba, J. C., Anantua, R., Asada, K., Azulay, R., Baczko, A.-K., Ball, D., Balokovic, M., ... Collaboration, T. EHT. (2021). Constraints on black-hole charges with the 2017 EHT observations of M87. Physical Review D, 103(10), Article 104047. https://doi.org/10.1103/PhysRevD.103.104047 Go to publication entry in University of Southern Denmark's Research Portal Terms of use This work is brought to you by the University of Southern Denmark. Unless otherwise specified it has been shared according to the terms for self-archiving. If no other license is stated, these terms apply: • You may download this work for personal use only. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying this open access version If you believe that this document breaches copyright please contact us providing details and we will investigate your claim. Please direct all enquiries to [email protected] Download date: 07. Mar. 2025 PHYSICAL REVIEW D 103, 104047 (2021) Constraints on black-hole charges with the 2017 EHT observations of M87 Prashant Kocherlakota ,1 Luciano Rezzolla,1–3 Heino Falcke,4 Christian M. Fromm,5,6,1 Michael Kramer,7 Yosuke Mizuno,8,9 Antonios Nathanail,9,10 H´ector Olivares,4 Ziri Younsi,11,9 Kazunori Akiyama,12,13,5 Antxon Alberdi,14 Walter Alef,7 Juan Carlos Algaba,15 Richard Anantua,5,6,16 Keiichi Asada,17 Rebecca Azulay,18,19,7 Anne-Kathrin Baczko,7 David Ball,20 Mislav Baloković,5,6 John Barrett,12 Bradford A. Benson,21,22 Dan Bintley,23 Lindy Blackburn,5,6 Raymond Blundell,6 Wilfred Boland,24 Katherine L. Bouman,5,6,25 Geoffrey C. Bower,26 Hope Boyce,27,28 Michael Bremer,29 Christiaan D. Brinkerink,4 Roger Brissenden,5,6 Silke Britzen,7 Avery E. Broderick,30–32 Dominique Broguiere,29 Thomas Bronzwaer,4 Do-Young Byun,33,34 John E. Carlstrom,35,22,36,37 Andrew Chael,38,39 Chi-kwan Chan,20,40 Shami Chatterjee,41 Koushik Chatterjee,42 Ming-Tang Chen,26 Yongjun Chen (陈永军),43,44 Paul M. Chesler,5 Ilje Cho,33,34 Pierre Christian,45 John E. Conway,46 James M. Cordes,41 Thomas M. Crawford,22,35 Geoffrey B. Crew,12 Alejandro Cruz-Osorio,9 Yuzhu Cui,47,48 Jordy Davelaar,49,16,4 Mariafelicia De Laurentis,50,9,51 Roger Deane,52–54 Jessica Dempsey,23 Gregory Desvignes,55 Sheperd S. Doeleman,5,6 Ralph P. Eatough,56,7 Joseph Farah,6,5,57 Vincent L. Fish,12 Ed Fomalont,58 Raquel Fraga-Encinas,4 Per Friberg,23 H. Alyson Ford,59 Antonio Fuentes,14 Peter Galison,5,60,61 Charles F. Gammie,62,63 Roberto García,29 Olivier Gentaz,29 Boris Georgiev,31,32 Ciriaco Goddi,4,64 Roman Gold,65,30 Jos´e L. Gómez,14 Arturo I. Gómez-Ruiz,66,67 Minfeng Gu (顾敏峰),43,68 Mark Gurwell,6 Kazuhiro Hada,47,48 Daryl Haggard,27,28 Michael H. Hecht,12 Ronald Hesper,69 Luis C. Ho (何子山),70,71 Paul Ho,17 Mareki Honma,47,48,72 Chih-Wei L. Huang,17 Lei Huang (黄磊),43,68 David H. Hughes,66 Shiro Ikeda,13,73–75 Makoto Inoue,17 Sara Issaoun,4 David J. James,5,6 Buell T. Jannuzi,20 Michael Janssen,7 Britton Jeter,31,32 Wu Jiang (江悟),43 Alejandra Jimenez-Rosales,4 Michael D. Johnson,5,6 Svetlana Jorstad,76,77 Taehyun Jung,33,34 Mansour Karami,30,31 Ramesh Karuppusamy,7 Tomohisa Kawashima,78 Garrett K. Keating,6 Mark Kettenis,79 Dong-Jin Kim,7 Jae-Young Kim,33,7 Jongsoo Kim,33 Junhan Kim,20,25 Motoki Kino,13,80 Jun Yi Koay,17 Yutaro Kofuji,47,72 Patrick M. Koch,17 Shoko Koyama,17 Carsten Kramer,29 Thomas P. Krichbaum,7 Cheng-Yu Kuo,81,17 Tod R. Lauer,82 Sang-Sung Lee,33 Aviad Levis,25 Yan-Rong Li (李彦荣),83 Zhiyuan Li (李志远),84,85 Michael Lindqvist,46 Rocco Lico,14,7 Greg Lindahl,6 Jun Liu (刘俊),7 Kuo Liu,7 Elisabetta Liuzzo,86 Wen-Ping Lo,17,87 Andrei P. Lobanov,7 Laurent Loinard,88,89 Colin Lonsdale,12 Ru-Sen Lu (路如森),43,44,7 Nicholas R. MacDonald,7 Jirong Mao (毛基荣),90–92 Nicola Marchili,86,7 Sera Markoff,42,93 Daniel P. Marrone,20 Alan P. Marscher,76 Iván Martí-Vidal,18,19 Satoki Matsushita,17 Lynn D. Matthews,12 Lia Medeiros,94,20 Karl M. Menten,7 Izumi Mizuno,23 James M. Moran,5,6 Kotaro Moriyama,12,47 Monika Moscibrodzka,4 Cornelia Müller,7,4 Gibwa Musoke,42,4 Alejandro Mus Mejías,18,19 Hiroshi Nagai,13,48 Neil M. Nagar,95 Masanori Nakamura,96,17 Ramesh Narayan,5,6 Gopal Narayanan,97 Iniyan Natarajan,54,52,98 Joseph Neilsen,99 Roberto Neri,29 Chunchong Ni,31,32 Aristeidis Noutsos,7 Michael A. Nowak,100 Hiroki Okino,47,72 Gisela N. Ortiz-León,7 Tomoaki Oyama,47 Feryal Özel,20 Daniel C. M. Palumbo,5,6 Jongho Park,17 Nimesh Patel,6 Ue-Li Pen,30,101–103 Dominic W. Pesce,5,6 Vincent Pi´etu,29 Richard Plambeck,104 Aleksandar PopStefanija,97 Oliver Porth,42,9 Felix M. Pötzl,7 Ben Prather,62 Jorge A. Preciado-López,30 Dimitrios Psaltis,20 Hung-Yi Pu,105,17,30 Venkatessh Ramakrishnan,95 Ramprasad Rao,26 Mark G. Rawlings,23 Alexander W. Raymond,5,6 Angelo Ricarte,5,6 Bart Ripperda,106,16 Freek Roelofs,4 Alan Rogers,12 Eduardo Ros,7 Mel Rose,20 Arash Roshanineshat,20 Helge Rottmann,7 Alan L. Roy,7 Chet Ruszczyk,12 Kazi L. J. Rygl,86 Salvador Sánchez,107 David Sánchez-Arguelles,66,67 Mahito Sasada,47,108 Tuomas Savolainen,109,110,7 F. Peter Schloerb,97 Karl-Friedrich Schuster,29 Lijing Shao,7,71 Zhiqiang Shen (沈志强),43,44 Des Small,79 Bong Won Sohn,33,34,111 Jason SooHoo,12 He Sun (孙赫),25 Fumie Tazaki,47 Alexandra J. Tetarenko,112 Paul Tiede,31,32 Remo P. J. Tilanus,4,64,113,20 Michael Titus,12 Kenji Toma,114,115 Pablo Torne,7,107 Tyler Trent,20 Efthalia Traianou,7 Sascha Trippe,116 Ilse van Bemmel,79 Huib Jan van Langevelde,79,117 Daniel R. van Rossum,4 Jan Wagner,7 Derek Ward-Thompson,118 John Wardle,119 Jonathan Weintroub,5,6 Norbert Wex,7 Robert Wharton,7 Maciek Wielgus,5,6 George N. Wong,62 Qingwen Wu (吴庆文),120 Doosoo Yoon,42 Andr´e Young,4 Ken Young,6 Feng Yuan (袁峰),43,68,121 Ye-Fei Yuan (袁业飞),122 J. Anton Zensus,7 Guang-Yao Zhao,14 and Shan-Shan Zhao43 (EHT Collaboration) 1Institut für Theoretische Physik, Goethe-Universität, Max-von-Laue-Strasse 1, 60438 Frankfurt, Germany 2Frankfurt Institute for Advanced Studies, Ruth-Moufang-Strasse 1, 60438 Frankfurt, Germany 3School of Mathematics, Trinity College, Dublin 2, Ireland 4Department of Astrophysics, Institute for Mathematics, Astrophysics and Particle Physics (IMAPP), Radboud University, P.O. Box 9010, 6500 GL Nijmegen, Netherlands 5Black Hole Initiative at Harvard University, 20 Garden Street, Cambridge, Massachusetts 02138, USA 2470-0010=2021=103(10)=104047(18) 104047-1 Published by the American Physical Society PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) 6Center for Astrophysics—Harvard & Smithsonian, 60 Garden Street, Cambridge, Massachusetts 02138, USA 7Max-Planck-Institut für Radioastronomie, Auf dem Hügel 69, D-53121 Bonn, Germany 8Tsung-Dao Lee Institute and School of Physics and Astronomy, Shanghai Jiao Tong University, Shanghai, 200240, China 9Institut für Theoretische Physik, Goethe-Universität Frankfurt, Max-von-Laue-Straße 1, D-60438 Frankfurt am Main, Germany 10Department of Physics, National and Kapodistrian University of Athens, Panepistimiopolis, GR 15783 Zografos, Greece 11Mullard Space Science Laboratory, University College London, Holmbury St. Mary, Dorking, Surrey, RH5 6NT, United Kingdom 12Massachusetts Institute of Technology Haystack Observatory, 99 Millstone Road, Westford, Massachusetts 01886, USA 13National Astronomical Observatory of Japan, 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan 14Instituto de Astrofísica de Andalucía-CSIC, Glorieta de la Astronomía s/n, E-18008 Granada, Spain 15Department of Physics, Faculty of Science, University of Malaya, 50603 Kuala Lumpur, Malaysia 16Center for Computational Astrophysics, Flatiron Institute, 162 Fifth Avenue, New York, New York 10010, USA 17Institute of Astronomy and Astrophysics, Academia Sinica, 11F of Astronomy-Mathematics Building, AS/NTU No. 1, Sec. 4, Roosevelt Rd, Taipei 10617, Taiwan, R.O.C. 18Departament d’Astronomia i Astrofísica, Universitat de Val`encia, C. Dr. Moliner 50, E-46100 Burjassot, Val`encia, Spain 19Observatori Astronòmic, Universitat de Val`encia, C. Catedrático Jos´e Beltrán 2, E-46980 Paterna, Val`encia, Spain 20Steward Observatory and Department of Astronomy, University of Arizona, 933 N. Cherry Avenue, Tucson, Arizona 85721, USA 21Fermi National Accelerator Laboratory, MS209, P.O. Box 500, Batavia, Illinois 60510, USA 22Department of Astronomy and Astrophysics, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 23East Asian Observatory, 660 N. A’ohoku Place, Hilo, Hawaii 96720, USA 24Nederlandse Onderzoekschool voor Astronomie (NOVA), PO Box 9513, 2300 RA Leiden, Netherlands 25California Institute of Technology, 1200 East California Boulevard, Pasadena, California 91125, USA 26Institute of Astronomy and Astrophysics, Academia Sinica, 645 N. A’ohoku Place, Hilo, Hawaii 96720, USA 27Department of Physics, McGill University, 3600 rue University, Montr´eal, Quebec City H3A 2T8, Canada 28McGill Space Institute, McGill University, 3550 rue University, Montr´eal, Quebec City H3A 2A7, Canada 29Institut de Radioastronomie Millim´etrique, 300 rue de la Piscine, F-38406 Saint Martin d’H`eres, France 30Perimeter Institute for Theoretical Physics, 31 Caroline Street North, Waterloo, Ontario, N2L 2Y5, Canada 31Department of Physics and Astronomy, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, N2L 3G1, Canada 32Waterloo Centre for Astrophysics, University of Waterloo, Waterloo, Ontario, N2L 3G1, Canada 33Korea Astronomy and Space Science Institute, Daedeok-daero 776, Yuseong-gu, Daejeon 34055, Republic of Korea 34University of Science and Technology, Gajeong-ro 217, Yuseong-gu, Daejeon 34113, Republic of Korea 35Kavli Institute for Cosmological Physics, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 36Department of Physics, University of Chicago, 5720 South Ellis Avenue, Chicago, Illinois 60637, USA 37Enrico Fermi Institute, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 38Princeton Center for Theoretical Science, Jadwin Hall, Princeton University, Princeton, New Jersey 08544, USA 39NASA Hubble Fellowship Program, Einstein Fellow 40Data Science Institute, University of Arizona, 1230 N. Cherry Avenue, Tucson, Arizona 85721, USA 41Cornell Center for Astrophysics and Planetary Science, Cornell University, Ithaca, New York 14853, USA 42Anton Pannekoek Institute for Astronomy, University of Amsterdam, Science Park 904, 1098 XH, Amsterdam, Netherlands 43Shanghai Astronomical Observatory, Chinese Academy of Sciences, 80 Nandan Road, Shanghai 200030, People’s Republic of China 104047-2 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) 44Key Laboratory of Radio Astronomy, Chinese Academy of Sciences, Nanjing 210008, People’s Republic of China 45Physics Department, Fairfield University, 1073 North Benson Road, Fairfield, Connecticut 06824, USA 46Department of Space, Earth and Environment, Chalmers University of Technology, Onsala Space Observatory, SE-43992 Onsala, Sweden 47Mizusawa VLBI Observatory, National Astronomical Observatory of Japan, 2-12 Hoshigaoka, Mizusawa, Oshu, Iwate 023-0861, Japan 48Department of Astronomical Science, The Graduate University for Advanced Studies (SOKENDAI), 2-21-1 Osawa, Mitaka, Tokyo 181-8588, Japan 49Department of Astronomy and Columbia Astrophysics Laboratory, Columbia University, 550 W 120th Street, New York, New York 10027, USA 50Dipartimento di Fisica “E. Pancini”, Universitá di Napoli “Federico II”, Compl. Univ. di Monte S. Angelo, Edificio G, Via Cinthia, I-80126, Napoli, Italy 51INFN Sez. di Napoli, Compl. Univ. di Monte S. Angelo, Edificio G, Via Cinthia, I-80126, Napoli, Italy 52Wits Centre for Astrophysics, University of the Witwatersrand, 1 Jan Smuts Avenue, Braamfontein, Johannesburg 2050, South Africa 53Department of Physics, University of Pretoria, Hatfield, Pretoria 0028, South Africa 54Centre for Radio Astronomy Techniques and Technologies, Department of Physics and Electronics, Rhodes University, Makhanda 6140, South Africa 55LESIA, Observatoire de Paris, Universit´e PSL, CNRS, Sorbonne Universit´e, Universit´e de Paris, 5 place Jules Janssen, 92195 Meudon, France 56National Astronomical Observatories, Chinese Academy of Sciences, 20A Datun Road, Chaoyang District, Beijing 100101, PR China 57University of Massachusetts Boston, 100 William T. Morrissey Boulevard, Boston, Massachusetts 02125, USA 58National Radio Astronomy Observatory, 520 Edgemont Road, Charlottesville, Virginia 22903, USA 59Steward Observatory and Department of Astronomy, University of Arizona, 933 North Cherry Avenue, Tucson, Arizona 85721, USA 60Department of History of Science, Harvard University, Cambridge, Massachusetts 02138, USA 61Department of Physics, Harvard University, Cambridge, Massachusetts 02138, USA 62Department of Physics, University of Illinois, 1110 West Green Street, Urbana, Illinois 61801, USA 63Department of Astronomy, University of Illinois at Urbana-Champaign, 1002 West Green Street, Urbana, Illinois 61801, USA 64Leiden Observatory—Allegro, Leiden University, P.O. Box 9513, 2300 RA Leiden, Netherlands 65CP3-Origins, University of Southern Denmark, Campusvej 55, DK-5230 Odense M, Denmark 66Instituto Nacional de Astrofísica, Óptica y Electrónica. Apartado Postal 51 y 216, 72000. Puebla Pue., M´exico 67Consejo Nacional de Ciencia y Tecnología, Av. Insurgentes Sur 1582, 03940, Ciudad de M´exico, M´exico 68Key Laboratory for Research in Galaxies and Cosmology, Chinese Academy of Sciences, Shanghai 200030, People’s Republic of China 69NOVA Sub-mm Instrumentation Group, Kapteyn Astronomical Institute, University of Groningen, Landleven 12, 9747 AD Groningen, Netherlands 70Department of Astronomy, School of Physics, Peking University, Beijing 100871, People’s Republic of China 71Kavli Institute for Astronomy and Astrophysics, Peking University, Beijing 100871, People’s Republic of China 72Department of Astronomy, Graduate School of Science, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan 73The Institute of Statistical Mathematics, 10-3 Midori-cho, Tachikawa, Tokyo, 190-8562, Japan 74Department of Statistical Science, The Graduate University for Advanced Studies (SOKENDAI), 10-3 Midori-cho, Tachikawa, Tokyo 190-8562, Japan 75Kavli Institute for the Physics and Mathematics of the Universe, The University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, 277-8583, Japan 76Institute for Astrophysical Research, Boston University, 725 Commonwealth New Jersey, Boston, Massachusetts 02215, USA 77Astronomical Institute, St. Petersburg University, Universitetskij pr., 28, Petrodvorets,198504 St.Petersburg, Russia 78Institute for Cosmic Ray Research, The University of Tokyo, 5-1-5 Kashiwanoha, Kashiwa, Chiba 277-8582, Japan 79Joint Institute for VLBI ERIC (JIVE), Oude Hoogeveensedijk 4, 7991 PD Dwingeloo, Netherlands 104047-3 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) 80Kogakuin University of Technology & Engineering, Academic Support Center, 2665-1 Nakano, Hachioji, Tokyo 192-0015, Japan 81Physics Department, National Sun Yat-Sen University, No. 70, Lien-Hai Rd, Kaosiung City 80424, Taiwan, R.O.C 82National Optical Astronomy Observatory, 950 N. Cherry Avenue, Tucson, Arizona 85719, USA 83Key Laboratory for Particle Astrophysics, Institute of High Energy Physics, Chinese Academy of Sciences, 19B Yuquan Road, Shijingshan District, Beijing, People’s Republic of China 84School of Astronomy and Space Science, Nanjing University, Nanjing 210023, People’s Republic of China 85Key Laboratory of Modern Astronomy and Astrophysics, Nanjing University, Nanjing 210023, People’s Republic of China 86Italian ALMA Regional Centre, INAF-Istituto di Radioastronomia, Via P. Gobetti 101, I-40129 Bologna, Italy 87Department of Physics, National Taiwan University, No.1, Sect.4, Roosevelt Road., Taipei 10617, Taiwan, R.O.C 88Instituto de Radioastronomía y Astrofísica, Universidad Nacional Autónoma de M´exico, Morelia 58089, M´exico 89Instituto de Astronomía, Universidad Nacional Autónoma de M´exico, CdMx 04510, M´exico 90Yunnan Observatories, Chinese Academy of Sciences, 650011 Kunming, Yunnan Province, People’s Republic of China 91Center for Astronomical Mega-Science, Chinese Academy of Sciences, 20A Datun Road, Chaoyang District, Beijing, 100012, People’s Republic of China 92Key Laboratory for the Structure and Evolution of Celestial Objects, Chinese Academy of Sciences, 650011 Kunming, People’s Republic of China 93Gravitation Astroparticle Physics Amsterdam (GRAPPA) Institute, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, Netherlands 94School of Natural Sciences, Institute for Advanced Study, 1 Einstein Drive, Princeton, New Jersey 08540, USA 95Astronomy Department, Universidad de Concepción, Casilla 160-C, Concepción, Chile 96National Institute of Technology, Hachinohe College, 16-1 Uwanotai, Tamonoki, Hachinohe City, Aomori 039-1192, Japan 97Department of Astronomy, University of Massachusetts, 01003, Amherst, Massachusetts, USA 98South African Radio Astronomy Observatory, Observatory 7925, Cape Town, South Africa 99Villanova University, Mendel Science Center Rm. 263B, 800 E Lancaster Avenue, Villanova Pennsylvania 19085 100Physics Department, Washington University CB 1105, St Louis, Missouri 63130, USA 101Canadian Institute for Theoretical Astrophysics, University of Toronto, 60 St. George Street, Toronto, Ontario M5S 3H8, Canada 102Dunlap Institute for Astronomy and Astrophysics, University of Toronto, 50 St. George Street, Toronto, Ontario M5S 3H4, Canada 103Canadian Institute for Advanced Research, 180 Dundas St West, Toronto, Ontario M5G 1Z8, Canada 104Radio Astronomy Laboratory, University of California, Berkeley, California 94720, USA 105Department of Physics, National Taiwan Normal University, No. 88, Sec. 4, Tingzhou Road, Taipei 116, Taiwan, R.O.C. 106Department of Astrophysical Sciences, Peyton Hall, Princeton University, Princeton, New Jersey 08544, USA 107Instituto de Radioastronomía Milim´etrica, IRAM, Avenida Divina Pastora 7, Local 20, E-18012, Granada, Spain 108Hiroshima Astrophysical Science Center, Hiroshima University, 1-3-1 Kagamiyama, Higashi-Hiroshima, Hiroshima 739-8526, Japan 109Aalto University Department of Electronics and Nanoengineering, PL 15500, FI-00076 Aalto, Finland 110Aalto University Metsähovi Radio Observatory, Metsähovintie 114, FI-02540 Kylmälä, Finland 111Department of Astronomy, Yonsei University, Yonsei-ro 50, Seodaemun-gu, 03722 Seoul, Republic of Korea 112East Asian Observatory, 660 North A’ohoku Place, Hilo, Hawaii 96720, USA 113Netherlands Organisation for Scientific Research (NWO), Postbus 93138, 2509 AC Den Haag, Netherlands 114Frontier Research Institute for Interdisciplinary Sciences, Tohoku University, Sendai 980-8578, Japan 115Astronomical Institute, Tohoku University, Sendai 980-8578, Japan 104047-4 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) 116Department of Physics and Astronomy, Seoul National University, Gwanak-gu, Seoul 08826, Republic of Korea 117Leiden Observatory, Leiden University, Postbus 2300, 9513 RA Leiden, Netherlands 118Jeremiah Horrocks Institute, University of Central Lancashire, Preston PR1 2HE, United Kingdom 119Physics Department, Brandeis University, 415 South Street, Waltham, Massachusetts 02453, USA 120School of Physics, Huazhong University of Science and Technology, Wuhan, Hubei, 430074, People’s Republic of China 121School of Astronomy and Space Sciences, University of Chinese Academy of Sciences, No. 19A Yuquan Road, Beijing 100049, People’s Republic of China 122Astronomy Department, University of Science and Technology of China, Hefei 230026, People’s Republic of China (Received 29 November 2020; accepted 21 April 2021; published 20 May 2021) Our understanding of strong gravity near supermassive compact objects has recently improved thanks to the measurements made by the Event Horizon Telescope (EHT). We use here the M87* shadow size to infer constraints on the physical charges of a large variety of nonrotating or rotating black holes. For example, we show that the quality of the measurements is already sufficient to rule out that M87* is a highly charged dilaton black hole. Similarly, when considering black holes with two physical and independent charges, we are able to exclude considerable regions of the space of parameters for the doubly-charged dilaton and the Sen black holes. DOI: 10.1103/PhysRevD.103.104047 I. INTRODUCTION General relativity (GR) was formulated to consistently account for the interaction of dynamical gravitational fields with matter and energy, the central idea of which is that the former manifests itself through modifications of spacetime geometry and is fully characterized by a metric tensor. While the physical axioms that GR is founded on are contained in the equivalence principle [1,2], the Einstein- Hilbert action further postulates the associated equations of motion involve no more than second-order derivatives of the metric tensor. that The strength of the gravitational field outside an object of mass M and characteristic size R, in geometrized units (G ¼ c ¼ 1), is related to its compactness C ≔ M=R, which is ∼10−6 for the Sun, and takes values ∼0.2–1 for compact objects such as neutron stars and black holes. Predictions from GR have been tested and validated by various solar-system experiments to very high precision [2,3], setting it on firm footing as the best-tested theory of classical gravity in the weak-field regime. It is important, however, to consider whether signatures of deviations from the Einstein-Hilbert action, e.g., due to higher derivative terms [4–6], could appear in measurements of phenomena occurring in strong-field regimes where C is large. tests are needed to assess whether generic Similarly, Published by the American Physical Society under the terms of license. the Creative Commons Attribution 4.0 International Further distribution of this work must maintain attribution to the author(s) and the published article’s title, journal citation, and DOI. violations of the equivalence principle occur in strong- fields due, e.g., to the presence of additional dynamical fields, such as scalar [7,8] or vector fields [9–13], that may fall off asymptotically. Agreement with the predictions of GR coming from observations of binary pulsars [14–16], and of the gravitational redshift [17] and geodetic orbit- precession [18] of the star S2 near our galaxy’s central supermassive compact object Sgr A⋆ by the GRAVITY collaboration, all indicate the success of GR in describing strong-field physics as well. In addition, with the gravita- tional-wave detections of coalescing binaries of compact objects by the LIGO/Virgo collaboration [19,20] and the first images of black holes produced by EHT, it is now possible to envision testing GR at the strongest field strengths possible. While the inferred size of the shadow from the recently obtained horizon-scale images of the supermassive com- pact object in M87 galaxy by the EHT collaboration [21– 26] was found to be consistent to within 17% for a 68% confidence interval of the size predicted from GR for a Schwarzschild black hole using the a priori known estimates for the mass and distance of M87* based on stellar dynamics [27], this measurement admits other possibilities, as do various weak-field tests [2,28]. Since the number of alternative theories to be tested using this measurement is large, a systematic study of the constraints set by a strong-field measurement is naturally more tractable within a theory-agnostic framework, and various such systems have recently been explored [29–36]. This approach allows for tests of a broad range of possibilities that may not be captured in the limited set of known [28], where solutions. This was exploited in Ref. 104047-5 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) constraints on two deformed metrics were obtained when determining how different M87* could be from a Kerr black hole while remaining consistent with the EHT measurements. However, because such parametric tests cannot be connected directly to an underlying property of the alter- native theory, here we use instead the EHT measurements to set bounds on the physical parameters, i.e., angular momentum, electric charge, scalar charge, etc.—and which we will generically refer to as “charges” (or hairs)—that various well-known black-hole solutions depend upon. Such analyses can be very instructive [37–51] since they can shed light on which underlying theories are promising candidates and which must be discarded or modified. At the same time, they may provide insight into the types of additional dynamical fields that may be necessary for a complete theoretical description of physical phenomena, and whether associated violations of the equivalence principle occur. More specifically, since the bending of light in the presence of curvature—either in static or in stationary spacetimes—is assured in any metric theory of gravity, and the presence of large amounts of mass in very small volumes can allow for the existence of a region where null geodesics move on spherical orbits, an examination of the characteristics of such photon regions, when they exist, is a useful first step. The projected asymptotic collection of the photons trajectories that are captured by the black hole —namely, all of the photon trajectories falling within the value of the impact parameter at the unstable circular orbit in the case of nonrotating black holes—will appear as a dark area to a distant observer and thus represents the “shadow” of the capturing compact object. This shadow— which can obviously be associated with black holes [52– 57], but also more exotic compact objects such as grav- astars [58,59] or naked singularities [60,61]—is determined entirely by the underlying spacetime metric. Therefore, the properties of the shadow—and at lowest order its size— represent valuable observables common to all metric theories of gravity, and can be used to test them for their agreement with EHT measurements. While the EHT measurement contains far more informa- tion related to the flow of magnetized plasma near M87, we will consider only the measurement of the size of the bright ring. Here we consider various spherically symmetric black- hole solutions, from GR that are either singular (see, e.g., [62]) or non-singular [63–65], and string theory [66–70]. Additionally, we also consider the Reissner-Nordström (RN) and the Janis-Newman-Winicour (JNW) [71] naked singu- larity solutions, the latter being a solution of the Einstein- Klein-Gordon system. Many of these solutions have been recently summarized in Ref. [36], where they were cast in a generalized expansion of static and spherically symmetric metrics. Since angular momentum plays a key role in astrophysical scenarios, we also consider various rotating black-hole solutions [72–75] which can be expressed in the Newman-Janis form [76] to facilitate straightforward ana- lytical computations. It is to be noted that this study is meant to be a proof of principle and that while the constraints we can set here are limited, the analytical procedure outlined below for this large class of metrics is general, so that as future observations become available, we expect the constraints that can be imposed following the approach proposed here to be much stronger. II. SPHERICAL NULL GEODESICS AND SHADOWS For all the static, spherically symmetric spacetimes we consider here, the definition of the shadow can be cast in rather general terms. In particular, for all the solutions considered, the line element expressed in areal-radial polar coordinates ðt; ˜r; θ; ϕÞ has the form1 ds2 ¼ gμνdxμdxν ¼ −fð˜rÞdt2 þ gð˜rÞ fð˜rÞ d˜r2 þ ˜r2dΩ2 2; ð1Þ and the photon region, which degenerates into a photon sphere, is located at ˜r ≕ ˜rps, which can be obtained by solving [28] ˜r − 2fð˜rÞ ∂ ˜rfð˜rÞ ¼ 0: ð2Þ The boundary of this photon sphere when observed from the frame of an asymptotic observer, due to gravitational lensing, appears to be a circle of size [28] ˜rsh ¼ p ˜rps ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ fð˜rps Þ : ð3Þ On the other hand, the class of Newman-Janis stationary, axisymmetric spacetimes we consider here [76], which are geodesically integrable (see, e.g., [55,77,78]), can be expressed in Boyer-Lindquist coordinates (t; r; θ; ϕ) as ds2 ¼ −fdt2 − 2asin2θð1 − fÞdtdϕ þ ½Σ þ a2sin2θð2 − fÞ(cid:2)sin2θdϕ2 þ Σ Δ dr2 þ Σdθ2; ð4Þ where f ¼ fðr; θÞ and ΔðrÞ ≔ Σðr; θÞfðr; θÞ þ a2 sin2 θ. In particular, these are and Σðr; θÞ ≔ r2 þ a2 cos2 θ 1We use the tilde on the radial coordinate of static spacetimes to distinguish it from the corresponding radial coordinate of axisymmetric spacetimes. 104047-6 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) the stationary generalizations obtained by employing the Newman-Janis algorithm [76]) for “seed” metrics of the form (1) with gð˜rÞ ¼ 1.2 The Lagrangian L for geodesic motion in the spacetime (4) is given as 2L ≔ gμν _xμ _xν, where an overdot represents a derivative with respect to the affine parameter, and 2L ¼ −1 for timelike geodesics and 2L ¼ 0 for null geodesics. The two Killing vectors ∂t and ∂ϕ yield two constants of motion −E ¼ −f_t − a sin2 θð1 − fÞ _ϕ; L ¼ −a sin2 θð1 − fÞ_t þ ½Σ þ a2 sin2 θð2 − fÞ(cid:2) sin2 θ _ϕ; ð5Þ in terms of which the geodesic equation for photons can be separated into Σ2 _r2 ¼ ðr2 þ a2 − aξÞ2 − ΔI ≕ RðrÞ; Σ2 _θ2 ¼ I − ða sin θ − ξ csc θÞ2 ≕ ΘðθÞ; ð6Þ ð7Þ where we have introduced first ξ ≔ L=E, and then I ≔ η þ ða − ξÞ2. Also, η is the Carter constant, and the existence of this fourth constant of motion is typically associated with the existence of an additional Killing-Yano tensor (see for example [56,80]). In particular, we are interested here in spherical null geodesics (SNGs), which satisfy _r ¼ 0 and ̈r ¼ 0 and are not necessarily planar; equivalently, SNGs can exist at locations where RðrÞ ¼ 0 and dRðrÞ=dr ¼ 0. Since these are only two equations in three variables (r, ξ, η), it is convenient, for reasons that will become evident below, to obtain the associated conserved quantities along such SNGs in terms of their radii r as (see also [81]), ξ η SNG SNG − ðrÞ ¼ r2 þ a2 a r2 a2ð∂rΔÞ2 ðrÞ ¼ 4rΔ a∂rΔ ; ½16a2Δ − ðr∂rΔ − 4ΔÞ2(cid:2): ð8Þ The condition that ΘðθÞ ≥ 0, which must necessarily hold as can be seen from Eq. (7), restricts the radial range for which SNGs exist, and it is evident that this range depends on θ. This region, which is filled by such SNGs, is called the photon region (see, e.g., Fig. 3.3 of [52]). The equality ΘðθÞ ¼ 0 determines the boundaries of the photon region, and the (disconnected) piece which lies in the exterior of the outermost horizon is of primary interest since its image, as seen by an asymptotic observer, is the shadow. We denote the inner and outer surfaces of this photon region by rp−ðθÞ and rpþðθÞ respectively, with the former (smaller) SNG corresponding to the location of a prograde photon orbit (i.e., ξ ðrp−Þ > 0), and the latter to a retrograde orbit. SNG It can be shown that all of the SNGs that are admitted in the photon region, for both the spherically symmetric and axisymmetric solutions considered here, are unstable to radial perturbations. In particular, for the stability of SNGs at a the stationary solutions, radius r ¼ rSNG with respect to radial perturbations is determined by the sign of ∂2 Þ > 0, SNGs at that radius are unstable. The expression for ∂2 rR reads rR, and when ∂2 rRðrSNG ∂2 rR ¼ 8r ð∂rΔÞ2 ½rð∂rΔÞ2 − 2rΔ∂2 rΔ þ 2Δ∂rΔ(cid:2): ð9Þ To determine the appearance of the photon region and the associated shadow, as seen by asymptotic observers, we can introduce the usual notion of celestial coordinates ðα; βÞ, which for any photon with constants of the motion ðξ; ηÞ can be obtained, for an asymptotic observer present at an inclination angle i with respect to the spin-axis of the compact object as in [82]. For photons on an SNG, we can set the conserved quantities (ξ, η) to the values given in Eq. (8) above to obtain [80,81] α sh ¼ −ξ SNG csc i; ð10Þ β sh ¼ (cid:3)ðη SNG þ a2cos2i − ξ2 SNGcot2iÞ1=2: ð11Þ p ﬃﬃﬃﬃﬃﬃﬃﬃﬃ Recognizing that β ¼ (cid:3) ΘðiÞ , it becomes clear that only the SNGs with ΘðiÞ ≥ 0 determine the apparent shadow shape. Since the photon region is not spherically symmetric in rotating spacetimes, the associated shadow is also not circular in general. It can be shown that the band of radii for which SNGs can exist narrows as we move away from the equatorial plane, and reduces to a single value at the pole, i.e., in the limit θ → π=2, we have rpþ ¼ rp− (see e.g., Fig. 3.3 of [52]). As a result, the parametric curve of the shadow boundary as seen by an asymptotic observer lying along the pole is perfectly circular, α2 ðrp(cid:3);π=2Þþ sh ξ2 SNG ðrp(cid:3);π=2Þ. þ β2 sh ¼ η SNG We can now define the characteristic areal-radius of the shadow curve as [83] 2Note that while the Sen solution can be obtained via the Newman-Janis algorithm [79], the starting point is the static EMd-1 metric written in a non-areal-radial coordinate ρ such that gttgρρ ¼ −1. rsh;A ≔ (cid:3) 2 π Z rpþ rp− 104047-7 (cid:4) 1=2 drβ ðrÞ∂rα sh ðrÞ sh : ð12Þ PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) III. SHADOW SIZE CONSTRAINTS FROM THE 2017 EHT OBSERVATIONS OF M87 g Measurements of stellar dynamics near M87* were previously used to produce a posterior distribution function of the angular gravitational radius θ ≔ M=D, where M is the mass of and D the distance to M87. The 2017 EHT observations of M87 can be similarly used to determine such a posterior [26]. These observations were used to determine the angular diameter ˆd of the bright emission ring that surrounds the shadow [26]. In Sec. 5.3 there, using synthetic images from general-relativistic magnetohydro- dynamics (GRMHD) simulations of accreting Kerr black holes for a wide range of physical scenarios, the scaling factor α ¼ ˆd=θ g was calibrated. For emission from the outermost boundary of the photon region of a Kerr black hole, α should lie in the range ≃9.6–10.4. The EHT measurement picks out a class of best-fit images (“top-set”) from the image library, with a mean value for α of 11.55 (for the “xs-ring” model) and 11.50 (for the “xs-ringauss” model), when using two different geometric crescent models for the images, implying that the geometric models were accounting for emission in the top- set GRMHD images that preferentially fell outside of the photon ring. Using the distribution of α for these top-set images then enabled the determination of the posterior in Þ for the EHT data. the angular gravitational radius Pobs It is to be noted that this posterior was also determined using direct GRMHD fitting, and image domain feature extraction procedures, as described in Sec. 9.2 there, and a high level of consistency was found across all measurement the fractional in Sec. 9.5 of methods. Finally, deviation in the angular gravitational radius δ was intro- duced as [26], ðθ g δ ≔ θ θ dyn g − 1; ð13Þ g and θ where θ dyn were used to denote the EHT and the stellar-dynamics inferences of the angular gravitational radius, respectively. The posterior on δ—as defined in Eq. (32) of [26]—was then obtained (see Fig. 21 there), and its width was found to be δ ¼ −0.01 (cid:3) 0.17, for a 68% credible interval. This agreement of the 2017 EHT meas- urement of the angular gravitational radius for M87* with a previously existing estimate for the same, at much larger distances, constitutes a validation of the null hypothesis of the EHT, and in particular that M87* can be described by the Kerr black-hole solution. Since the stellar dynamics measurements [27] are sensi- tive only to the monopole of the metric (i.e., the mass) due the distances to negligible spin-dependent effects at involved in that analysis, modeling M87* conservatively using the Schwarzschild solution is reasonable with their obtained posterior. Then, using the angular gravitational sh θ ¼ rsh=D as being θ radius estimate from stellar dynamics yields a prediction for the angular shadow radius θ ¼ p ﬃﬃﬃ sh 3 3 dyn. The 2017 EHT measurement, which includes spin-dependent effects as described above and which probes near-horizon scales, then determines the allowed ≈ shadow diameter spread in the p ﬃﬃﬃ ð1 (cid:3) 0.17Þθ 3 3 g, at 68% confidence levels [28]. Finally, since both angular estimates θ g make use of the same distance estimate to M87, it is possible to convert the 1-σ bounds on θ sh to bounds on the allowed shadow size for M87. sh and θ as, θ angular sh That is, independently of whether the underlying sol- ution be spherically symmetric (in which case we will consider ˜rsh) or axisymmetric ðrsh;AÞ, the shadow size of p ﬃﬃﬃ ð1 (cid:3) 0.17ÞM [28], i.e., 3 M87* must lie in the range 3 (see gray-shaded region in Fig. 2) 4.31M ≈ rsh;EHT- min ≤ ˜rsh; rsh;A ≤ rsh;EHT- max ≈ 6.08M; ð14Þ where we have introduced the maximum/minimum shadow radii rsh;EHT- max=rsh;EHT- min inferred by the EHT, at 68% confidence levels. Note that the bounds thus derived are consistent with compact objects that cast shadows that are both signifi- cantly smaller and larger than the minimum and maximum shadow sizes that a Kerr black hole could cast, which lie in the range, 4.83M − 5.20M (see, e.g., [28,84]). An important caveat here is that the EHT posterior distribution on θ g was obtained after a comparison with a large library of synthetic images built from GRMHD simulations of accreting Kerr black holes [25]. Ideally, a rigorous comparison with non-Kerr solutions would require a similar analysis and posterior distributions built from equivalent libraries obtained from GRMHD simulations of such non-Kerr solutions. Besides being computationally unfeasible, this approach is arguably not necessary in practice. For example, the recent comparative analysis of Ref. [50] has shown that the image libraries produced in this way would be very similar and essentially indistin- guishable, given the present quality of the observations. As a result, we adopt here the working assumption that the 1-σ uncertainty in the shadow angular size for non-Kerr solutions is very similar to that for Kerr black holes, and hence employ the constraints (14) for all of the solutions considered here. IV. NOTABLE PROPERTIES OF VARIOUS SPACETIMES As mentioned above, a rigorous comparison with non- Kerr black holes would require constructing a series of exhaustive libraries of synthetic images obtained from GRMHD simulations on such non-Kerr black holes. In turn, this would provide consistent posterior distributions 104047-8 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) TABLE I. Summary of properties of spacetimes used here. For easy access, we show whether the spacetime contains a rotating compact object or not, whether it contains a spacetime singularity, and what type of stationary nongravitational fields are present in the spacetime. Starred spacetimes contain naked singularities and daggers indicate a violation of the equivalence principle (see, e.g., [36]); In particular, these indicate violations of the weak equivalence principle due to a varying fine structure constant, a result of the coupling of the dilaton to the EM Lagrangian [36,89]. Spacetime KN [73] Kerr [72] RN [62] RN* [62] Schwarzschild [62] Rot. Bardeen [75] Bardeen [63] Rot. Hayward [75] Frolov [65] Hayward [64] JNW* [71] KS [66] Sen† [74] EMd-1† [67,68] EMd-2† [70] Rotation Singularity Spacetime content Yes Yes No No No Yes No Yes No No No No Yes No No Yes Yes Yes Yes Yes No No No No No Yes Yes Yes Yes Yes EM fields vacuum EM fields EM fields vacuum matter matter matter EM fields, matter matter scalar field vacuum EM, dilaton, axion fields EM, dilaton fields EM, EM, dilaton fields of angular gravitational radii for the various black holes and hence determine how δ varies across different non-Kerr black holes, e.g., for Sen black holes. Because this is computationally unfeasible—the construction of only the Kerr library has required the joint effort of several groups with the EHTC over a good fraction of a year—we briefly discuss below qualitative arguments to support our use of the bounds given in Eq. (14) above as an approximate, yet indicative, measure. theories, solutions from three considered here i.e., To this end, we summarize in Table I the relevant properties of the various solutions used here. First, we have types the underlying actions are either of [62–66,71,72,75], (a) Einstein-Hilbert-Maxwell-matter (b) Einstein-Hilbert-Maxwell-dilaton-axion [67,68,74], or (c) Einstein-Hilbert-Maxwell-Maxwell-dilaton [70]. This careful choice implies that the gravitational piece of the action is always given by the Einstein-Hilbert term and that matter is minimally coupled to gravity. As a result, the dynamical evolution of the accreting plasma is expected to be very similar to that in GR, as indeed found in Ref. [50]. Second, since a microphysical description that allows one to describe the interaction of the exotic matter present in some of the regular black-hole spacetimes used here [63,64]—which typically do not satisfy some form of the energy conditions [75,85]—with the ordinary matter is thus far lacking, it is reasonable to assume that the interaction between these two types of fluids is gravita- is done in standard tional only. This is indeed what numerical simulations, either in dynamical spacetimes (see, e.g., [86]), or in fixed ones [49,87]. Third, since the mass-energy in the matter and electromagnetic fields for the non-vacuum spacetimes used here is of the order of the mass of the central compact object M, while the total mass of the accreting plasma in the GRMHD simulations is only a tiny fraction of the same, it is reasonable to treat the spacetime geometry and the stationary fields as unaffected by the plasma. Fourth, we have also been careful not to use solutions from theories with modified electrodynamics (such as nonlinear electrodynamics). As a result, the electromagnetic Lagrangian in all of the theories consid- ered here is the Maxwell Lagrangian (see, e.g., the discussion in [36] and compare with [53]). This ensures that in these spacetimes light moves along the null geo- desics of the metric tensor (see, e.g., Sec. 4.3 of [62] and compare against Sec. 2 of [88]). Therefore, we are also assured that ray-tracing the radiation emitted from the accreting matter in these spacetimes can be handled similarly as in the Kerr spacetime. Finally, under the assumption that the dominant effects in determining the angular gravitational radii come from variations in the location of the photon region and in location of the inner edge of the accretion disk in these spacetimes, it is instructive to learn how these two physical quantities vary when changing physical charges, and, in particular, they are quantitatively comparable to the corresponding values for the Kerr spacetime. to demonstrate that For this purpose, we study the single-charge solutions used here and report in Fig. 1 the variation in the location of the photon spheres (left panel) and innermost stable circular orbit (ISCO) radii (right panel) as a function of the relevant physical charge (cf. left panel of Fig. 1 in the main text). Note that both the photon-sphere and the ISCO radii 104047-9 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) FIG. 1. Left: variation in the photon sphere radii for the single-charge nonrotating solutions as a function of the normalized physical charge. Right: The same as in the left panel but for the ISCO radii. We include also, for comparison, the variation in the Kerr equatorial prograde and retrograde photon sphere and ISCO radii in the left and right panels respectively. depend exclusively on the gtt component of the metric when expressed using an areal radial coordinate ˜r (see, e.g., [28,36]). To gauge the effect of spin, we also show the variation in the locations of the equatorial prograde and retrograde circular photon orbits and the ISCOs in the Kerr black-hole spacetime, expressed in terms of the Cartesian Kerr-Schild radial coordinate rCKS, which, in the equatorial plane, is related to the Boyer-Lindquist radial coordinate used elsewhere in this work r simply via [90] p ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ r2 þ a2 : rCKS ¼ ð15Þ It is apparent from Fig. 1 that the maximum deviation in the photon-sphere size from the Schwarzschild solution occurs for the EMd-1 black hole and is ≈75%, while the size of the prograde equatorial circular photon orbit for Kerr deviates by at most ≈50%. Similarly, the maximum variation in the ISCO size also occurs for the EMd-1 solution and is ≈73%, while the prograde equatorial ISCO for Kerr can differ by ≈66%. V. CHARGE CONSTRAINTS FROM THE EHT M87* OBSERVATIONS We first consider compact objects with a single “charge,” and report in the left panel of Fig. 2 the variation in the shadow radius for various spherically symmetric black hole solutions, as well as for the RN and JNW naked singular- ities.3 More specifically, we consider the black-hole 3While the electromagnetic and scalar charge parameters for the RN and JNW spacetimes are allowed to take values ¯q > 1 and 0 < ˆ¯ν ≔ 1 − ¯ν < 1 respectively, they do not cast shadows for ¯q > p [36] and and 0.5 ≤ ˆ¯ν < 1 (see, e.g., Sec. IV D of ﬃﬃﬃﬃﬃﬃﬃﬃ 9=8 references therein). ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 16=27 ≤ ; Frolov: 0 < ¯l ≤ solutions given by Reissner-Nordström (RN) [62], Bardeen [63,75], Hayward [64,91], Kazakov-Solodhukin (KS) [66], and also the asymptotically-flat Einstein- Maxwell-dilaton (EMd-1) with ϕ∞ ¼ 0 and α1 ¼ 1 [67,68,88] solution (see Sec. IV of [36] for further details on these solutions). For each of these solutions we vary the corresponding charge (in units of M) in the allowed ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ range, i.e., RN: 0 < ¯q ≤ 1; Bardeen: 0 < ¯qm 16=27 ; ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ p Hayward: 0 < ¯l ≤ 16=27 , p ﬃﬃﬃ 0 < ¯q ≤ 1; KS: 0 < ¯l; EMd-1: 0 < ¯q < 2 , but report the normalized value in the figure so that all curves are in a range between 0 and 1. The figure shows the variation in the shadow size of KS black holes over the parameter range p ﬃﬃﬃ 0 < ¯l < 2 . Note that the shadow radii tend to become smaller with increasing physical charge, but also that this is not universal behavior, since the KS black holes have increasing shadow radii (the singularity is smeared out on a surface for this solution, which increases in size with increasing ¯l). p p the same time, Overall, it is apparent that the regular Bardeen, Hayward, and Frolov black-hole solutions are compatible with the present constraints. At the Reissner- Nordström and Einstein-Maxwell-dilaton 1 black-hole solu- tions, for certain values of the physical charge, produce shadow radii that lie outside the 1-σ region allowed by the 2017 EHT observations, and we find that these solutions are now constrained to take values in, 0 < ¯q ≲ 0.90 and 0 < ¯q ≲ 0.95 respectively. Furthermore, the Reissner-Nordström naked singularity is entirely eliminated as a viable model for M87* and the Janis-Newman-Winicour naked singularity parameter space is restricted further by this measurement to 0 < ˆ¯ν ≲ 0.47. Finally, we also find that the KS black hole is also restricted to have charges in the range ¯l < 1.53. In addition, note that the nonrotating Einstein-Maxwell- dilaton 2 (EMd-2) solution [70]—which depends on two 104047-10 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) FIG. 2. Left: shadow radii ˜rsh for various spherically symmetric black-hole solutions, as well as for the JNW and RN naked singularities (marked with an asterisk), as a function of the physical charge normalized to its maximum value. The gray/red shaded regions refer to the areas that are 1-σ consistent/inconsistent with the 2017 EHT observations and highlight that the latter set constraints on the physical charges (see also Fig. 3 for the EMd-2 black hole). Right: shadow areal radii rsh;A as a function of the dimensionless spin a for four families of black-hole solutions when viewed on the equatorial plane (i ¼ π=2). Also in this case, the observations restrict the ranges of the physical charges of the Kerr-Newman and the Sen black holes (see also Fig. 3). independent charges—can also produce shadow radii that are incompatible with the EHT observations; we will discuss this further below. The two EMd black-hole solutions (1 and 2) correspond to fundamentally different field contents, as discussed in [70]. [75]. The data refers i ¼ π=2, and we find that radii are shown as a function of We report in the right panel of Fig. 2 the shadow areal radius rsh;A for a number of stationary black holes, such as Kerr [72], Kerr-Newman (KN) [73], Sen [74], and the rotating versions of the Bardeen and Hayward to an observer black holes inclination angle of the variation in the shadow size with spin at higher inclinations (of up to i ¼ π=100) is at most about 7.1% (for i ¼ π=2, this is 5%); of course, at zero-spin the shadow size does not change with inclination. The shadow areal the dimensionless spin of the black hole a ≔ J=M2, where J is its angular momentum, and for representative values of the additional parameters that characterize the solu- tions. Note that—similar to the angular momentum for a Kerr black hole—the role of an electric charge or the presence of a de Sitter core (as in the case of the Hayward black holes) is to reduce the apparent size of the shadow. Furthermore, on increasing the spin para- meter, we recover the typical the shadow becomes increasingly noncircular, as encoded, e.g., in the distortion parameter δ (see Appendix). Also in this case, while the regular rotating Bardeen and Hayward solutions are compatible with the present constraints set by the 2017 EHT observations, the Kerr-Newman and Sen families of black holes can produce shadow areal radii that lie outside of the 1-σ region allowed by the observations. sh defined in [57,83] trend that To further explore the constraints on the excluded regions for the Einstein-Maxwell-dilaton 2 and the Sen black holes, we report in Fig. 3 the relevant ranges for these two solutions. The Einstein-Maxwell-dilaton 2 black holes are nonrotating but have two physical charges expressed by p ﬃﬃﬃ the coefficients 0 < ¯qe < 2 , while the Sen black holes spin (a) and have an additional electro- magnetic charge ¯qm. Also in this case, the gray/red shaded regions refer to the areas that are consistent/inconsistent with the 2017 EHT observations. The figure shows rather easily that for these two black-hole families there are large and 0 < ¯qm < p ﬃﬃﬃ 2 FIG. 3. Constraints set by the 2017 EHT observations on the nonrotating Einstein-Maxwell-dilaton 2 and on the rotating Sen black holes. Also in this case, the gray/red shaded regions refer to the areas that are 1-σ consistent/inconsistent with the 2017 EHT observations). 104047-11 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) areas of the space of parameters that are excluded at the 1-σ level. Not surprisingly, these areas are those where the physical charges take their largest values and hence the corresponding black-hole solutions are furthest away from the corresponding Schwarzschild or Kerr solutions. The obvious prospect is of course that as the EHT increases the precision of its measurements, increasingly larger portions of the space of parameters of these black holes will be excluded. Furthermore, other solutions that are presently still compatible with the observations may see their corresponding physical charges restricted. VI. CONCLUSIONS As our understanding of gravity under extreme regimes improves, and as physical measurements of these regimes are now becoming available—either through the imaging of supermassive black holes or the detection of gravitational waves from stellar-mass black holes—we are finally in the position of setting some constraints to the large landscape of non-Kerr black holes that have been proposed over the years. We have used here the recent 2017 EHT observations of M87* to set constraints, at the 1-σ-level, on the physical charges—either electric, scalar, or angular momentum—of a large variety of static (nonrotating) or stationary (rotating) black holes. In this way, when considering nonrotating black holes with a single physical charge, we have been able to rule out, at 68% confidence levels, the possibility that M87* is a near-extremal Reissner-Nordström or Einstein-Maxwell- dilaton 1 black hole and that the corresponding physical charge must be in the range, RN: 0 < ¯q ≲ 0.90 and EMd-1: 0 < ¯q ≲ 0.95. We also find that it cannot be a Reissner- Nordström naked singularity or a JNW naked singularity with large scalar charge, i.e., only 0 < ˆ¯ν ≲ 0.47 is allowed. Similarly, when considering black holes with two physical charges (either nonrotating or rotating), we have been able to exclude, with 68% confidence, considerable regions of the space of parameters in the case of the Einstein- Maxwell-dilaton 2, Kerr-Newman and Sen black holes. Although the idea of setting such constraints is an old one (see, e.g., [29–36,51,54,55]), and while there have been recent important developments in the study of other possible observational signatures of such alternative sol- utions, such as in X-ray spectra of accreting black holes (see, e.g., [92]) and in gravitational waves [88,93–97], to the best of our knowledge, constraints of this type have not been set before for the spacetimes considered here. As a final remark, we note that while we have chosen only a few solutions that can be seen as deviations from the Schwarzschild/Kerr solutions since they share the same basic Einstein-Hilbert-Maxwell action of GR, the work largely as a proof-of-concept presented here is meant investigation and a methodological example of how to exploit observations and measurements that impact the photon region. While a certain degeneracy in the shadow size induced by mass and spin remains and is inevitable, when in the future the relative difference in the posterior for the angular gravitational radius for M87* can be pushed to ≲5%, we should be able to constrain its spin, when modeling it as a Kerr black hole. Furthermore, since this posterior implies a spread in the estimated mass, one can expect small changes in the exact values of the maximum allowed charges reported here. Hence, as future observa- tions—either in terms of black-hole imaging or of gravi- tational-wave detection—will become more precise and notwithstanding a poor measurement of the black-hole spin, the methodology presented here can be readily applied to set even tighter constraints on the physical charges of non-Einsteinian black holes. ACKNOWLEDGMENTS It is a pleasure to thank Enrico Barausse, Sebastian Völkel and Nicola Franchini for insightful discussions on alternative black holes. Support comes the ERC Synergy Grant “BlackHoleCam: Imaging the Event Horizon of Black Holes” (Grant No. 610058). During the completion of this work we have become aware of a related work by S. Völkel et al. [98], which deals with topics that partly overlap with those of this manuscript (i.e., EHT tests of the strong-field regime of GR). The authors of the present paper thank the following organizations and programs: the Academy of Finland (projects No. 274477, No. 284495, No. 312496, No. 315721); the Alexander von Humboldt Stiftung; Agencia Nacional de Investigación y Desarrollo (ANID), Chile via NCN19\_058 (TITANs), and Fondecyt 3190878; an Alfred P. Sloan Research Fellowship; Allegro, the European ALMA Regional Centre node in the Netherlands, the NL astronomy research network NOVA and the astronomy institutes of the University of Amsterdam, Leiden University and Radboud University; the Black Hole Initiative at Harvard University, through a grant (60477) from the John Templeton Foundation; the China Scholarship Council; Consejo Nacional de Ciencia y Tecnología (CONACYT, Mexico, projects No. U0004- F0003-272050, 246083, No. U0004-259839, No. No. M0037-279006, No. F0003-281692, No. 104497, No. 275201, No. 263356); the Delaney Family via the Delaney Family John A.Wheeler Chair at Perimeter Institute; Dirección General de Asuntos del Personal Acad´emico-Universidad Nacional Autónomade M´exico and (DGAPA-UNAM, No. the East IN112820); the European Asia Core Observatories Association; “BlackHoleCam: Research Council Synergy Grant Imaging the Event Horizon of Black Holes” (Grant No. 610058); the Generalitat Valenciana postdoctoral grant APOSTD/2018/177 and GenT Program (project No. CIDEGENT/2018/021); MICINN Research Project No. and Betty Moore Foundation (Grants No. GBMF-3561, the EACOA Fellowship of PID2019-108995GB-C22; the Gordon IN112417 projects No. 104047-12 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) at the and sezione iniziative 25120007); di Napoli, for Astronomy 18H01245, No. and Astrophysics the MPG and the CAS; No. GBMF-5278); the Istituto Nazionale di Fisica Nucleare (INFN) specifiche the International Max Planck Research TEONGRAV; School the Universities of Bonn and Cologne; Joint Princeton/ Flatiron Postdoctoral Joint Columbia/Flatiron Fellowships, research at the Flatiron Institute is supported the Japanese Government by the Simons Foundation; (Monbukagakusho: MEXT) Scholarship; Japan Society for the Promotion of Science (JSPS) Grant-in- Aid for JSPS Research Fellowship (JP17J08829); the Key Research Program of Frontier Sciences, Chinese Academy of Sciences (CAS, grants No. QYZDJ-SSW-SLH057, No. QYZDJSSW-SYS008, No. ZDBS-LY-SLH011); the Lever-hulme Trust Early Career Research Fellowship; the Max-Planck-Gesellschaft (MPG); the Max Planck Partner Group of the MEXT/JSPS KAKENHI (Grants No. 18KK0090, No. JP18K13594, No. JP18K03656, No. JP18H03721, No. 18K03709, No. the Malaysian Fundamental Research Grant Scheme (FRGS) FRGS/1/ 2019/STG02/UM/02/6; the MIT International Science and Technology Initiatives (MISTI) Funds; the Ministry of Science and Technology (MOST) of Taiwan (105-2112- M-001-025-MY3, 106-2112-M-001-011, 106-2119-M- 001-027, 107-2119-M-001-017, 107-2119-M-001-020, 107-2119-M-110-005,108-2112-M-001-048, 109- the National Aeronautics and Space 2124-M-001-005); Administration Investigator grant 80NSSC20K1567 and 80NSSC20K1567, NASA Astrophysics Grant 80NSSC20K0527, NASA NuSTAR Award No. No. Grant NASA No. NNX17AL82G, and Hubble Fellowship Grant No. HST-HF2-51431.001-A awarded by the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., for NASA, under contract NAS5-26555); the National Institute of Natural Sciences (NINS) of Japan; the National Key Research and Development Program of China (Grants No. 2016YFA0400704, No. 2016YFA0400702); the National Science Foundation (NSF, Grants No. AST- 0096454, No. AST-0352953, No. AST-0521233, No. AST-0705062, No. AST-0905844, No. AST- 0922984, No. AST-1126433, No. AST-1140030, No. DGE-1144085, No. AST-1207704, No. AST- 1207730, No. AST-1207752, No. MRI-1228509, No. OPP-1248097, No. AST-1310896, No. AST- 1337663, No. AST-1440254, No. AST-1555365, No. AST-1615796, No. AST-1715061, No. AST- 1716327, No. AST-1716536, No. OISE-1743747, No. AST-1816420, No. AST-1903847, No. AST- the Natural Science 1935980, No. AST-2034306); Foundation 11573051, of No. 11633006, No. 11650110427, No. 10625314, 80NSSC20K0645, Fermi Guest (Grants No. Program (NASA, Theory China and No. No. Fellowship 11721303, No. 11725312, No. No. 11933007, No. 11991052, No. 11991053); a fellowship of China the Postdoctoral Science Foundation (2020M671266); Natural Sciences and Engineering Research Council of Canada (NSERC, including a Discovery Grant and the NSERC Alexander Graham Bell Canada Graduate the National Research Scholarships-Doctoral Program); Foundation of Korea (the Global PhD Fellowship Grant: grants NRF- 2014H1A2A1018695, 2015-R1D1A1A01056807, 2015H1A2A1033752, No. Program: NRF- the Korea Research 2015H1D3A1066561, Basic Research Support Grant No. 2019R1F1A1059721); the Netherlands Organization for Scientific Research (NWO) VICI award (Grant the No. 639.043.513) and Spinoza Prize SPI 78-409; New Scientific Precision Radio Frontiers with Interferometry Fellowship awarded by the South African Radio Astronomy Observatory (SARAO), which is a facility of the National Research Foundation (NRF), an agency of the Department of Science and Innovation (DSI) of South Africa; the South African Research Chairs Initiative of the Department of Science and Innovation and National Research Foundation; the Onsala Space Observatory (OSO) national infrastructure, for the provi- sioning of its facilities/observational support (OSO receives funding through the Swedish Research Council under Grant No. 2017-00648) Institute for Theoretical Physics (research at Perimeter Institute is supported by the Government of Canada through the Department of and Economic Innovation, Science Development and by the Province of Ontario through the Ministry of Research, Innovation and Science); the Spanish Ministerio de Ciencia e Innovación (grants No. PGC2018-098915-B-C21, No. AYA2016-80889-P; No. PID2019-108995GB-C21, No. PGC2018-098915-B- C21); the State Agency for Research of the Spanish MCIU through the “Center of Excellence Severo Ochoa” award for the Instituto de Astrofísica de Andalucía (SEV-2017- 0709); the Toray Science Foundation; the Consejería de Economía, Conocimiento, Empresas y Universidad, Junta the Consejo de Andalucía (Grant No. P18-FR-1769), Superior (Grant No. 2019AEP112); the U.S. Department of Energy (USDOE) through the Los Alamos National Laboratory (operated by Triad National Security, LLC, for the National Nuclear Security Administration of the USDOE (Contract the European Union’s No. Horizon 2020 research and innovation program under grant agreement No. 730562 RadioNet; ALMA North America Development Fund; the Academia Sinica; Chandra TM6- 17006X; Chandra award DD7-18089X. This work used the Extreme Science and Engineering Discovery Environment (XSEDE), supported by NSF grant ACI-1548562, and CyVerse, supported by NSF Grants No. DBI-0735191, No. DBI-1265383, and No. DBI-1743442. XSEDE 89233218CNA000001); Investigaciones the Perimeter Científicas de 104047-13 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) (Canada), Ministry Stampede2 resource at TACC was allocated through TG- AST170024 and TG-AST080026N. XSEDE Jet-Stream resource at PTI and TACC was allocated through AST170028. The simulations were performed in part on the SuperMUC cluster at the LRZ in Garching, on the LOEWE cluster in CSC in Frankfurt, and on the HazelHen cluster at the HLRS inStuttgart. This research was enabled in part by support provided by Compute Ontario [99], Calcul Quebec [100] and Compute Canada [101]. We thank the staff at the participating observatories, correlation centers, and institutions for their enthusiastic support. This paper makes use of the following ALMA data: ADS/JAO.ALMA\#2016.1.01154.V. ALMA is a partner- ship of the European Southern Observatory (ESO; Europe, states), NSF, and National representing its member together with Institutes of Natural Sciences of Japan, National Research Council of Science and Technology (MOST; Taiwan), Academia Sinica Institute of Astronomy and Astro-physics(ASIAA; Taiwan), and Korea Astronomy and Space Science Institute (KASI; Republic of Korea), in cooperation with the Republic of Chile. The Joint ALMA Observatory is operated by ESO, Associated Universities, Inc. (AUI)/ NRAO, and the National Astronomical Observatory of Japan (NAOJ). The NRAO is a facility of the NSF operated under cooperative agreement by AUI. This paper has made use of the following APEX data: Project ID T-091.F-0006- 2013. APEX is a collaboration between the Max-Planck- Institut für Radioastronomie (Germany), ESO, and the Onsala Space Observatory (Sweden). The SMA is a joint project between the SAO and ASIAA and is funded by the Smithsonian Institution and the Academia Sinica. The JCMT is operated by the East Asian Observatory on behalf of the NAOJ, ASIAA, and KASI, as well as the Ministry of Finance of China, Chinese Academy of Sciences, and the National Key R&D Program (No. 2017YFA0402700) of China. Additional funding support for the JCMT is pro- vided by the Science and Technologies Facility Council (UK) and participating universities in the UK and Canada. The LMT is a project operated by the Instituto Nacional de Astrofísica, Óptica y Electrónica (Mexico) and the University of Massachusetts at Amherst (USA), with financial support from the Consejo Nacional de Ciencia y Tecnología and the National Science Foundation. The IRAM 30-m telescope on Pico Veleta, Spain is operated by IRAM and supported by CNRS (Centre National de la Recherche Scientifique, France), MPG (Max-Planck- Gesellschaft, Germany) and IGN (Instituto Geográfico Nacional, Spain). The SMT is operated by the Arizona Radio Observatory, a part of the Steward Observatory of the University of Arizona, with financial support of operations for from the State of Arizona and financial support instrumentation development from the NSF. The SPT is supported by the National Science Foundation through Grant No. PLR-1248097. Partial support is also provided by the NSF Physics Frontier Center Grant No. PHY- 1125897 to the Kavli Institute of Cosmological Physics at the University of Chicago, the Kavli Foundation and the Gordon and Betty Moore Foundation grant GBMF 947. The SPT hydrogen maser was provided on loan from the GLT, courtesy of ASIAA. The EHTC has received gen- erous donations of FPGA chips from Xilinx Inc., under the Xilinx University Program. The EHTC has benefited from technology shared under open-source license by the Collaboration for Astronomy Signal Processing and is Electronics Research (CASPER). The EHT project grateful to T4Science and Microsemi for their assistance with Hydrogen Masers. This research has made use of NASA’s Astrophysics Data System. We gratefully acknowledge the support provided by the extended staff of the ALMA, both from the inception of the ALMA Phasing Project through the observational campaigns of 2017 and 2018. We would like to thank A. Deller and W. Brisken for EHT-specific support with the use of DiFX. We acknowledge the significance that Maunakea, where the SMA and JCMT EHT stations are located, has for the indigenous Hawaiian people. Facilities: EHT, ALMA, APEX, IRAM:30 m, JCMT, LMT, SMA, ARO:SMT, SPT. Software: AIPS [102], ParselTongue [103], GNU Parallel [104], eht-imaging [105], Difmap [106], Numpy [107], Scipy [108], Pandas [109], Astropy [110,111], Jupyter [112], Matplotlib [113], THEMIS [114], DMC [115], polsolve [116], GPCAL [117]. APPENDIX: DISTORTION PARAMETERS Since the boundary of the shadow region is a closed curve as discussed above, one can define various characteristic features for a quantitative comparison [80,83]. Out of the many possible measures of distortion of this curve from a perfect circle discussed in Ref. [83], we use here the simplest one which was originally introduced in Ref. [80], namely αl;c − αl ðA1Þ δ sh ¼ ; rsh;c where rsh;c is the radius of the circumcircle passing through the two points (since the images here are symmetric about the α-axis) with coordinates ðα Þ, which are the rightmost and topmost points of the shadow curve, and is given as [80], r; 0Þ and ðα t; β t rsh;c ¼ ðα t Þ2 þ β2 − α r t j − α 2jα r t ; ðA2Þ with ðαl; 0Þ and ðαl;c; 0Þ the leftmost points of the shadow curve and of the circumcircle respectively (see Fig. 3 of [57]). In Fig. 4 we display the distortion parameter δ sh for the shadow curves of various rotating black holes, for an equatorial observer, as an additional simple comparable characteristic. We note also that the deviation of δ sh from 104047-14 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) As a concluding remark we note that the EHT bounds on the size of the shadow of M87, as discussed above and impose straightforward displayed in Eq. (14), do not bounds on its shape. In particular, we can see from Fig. 4 that the rotating Bardeen black hole with ¯qm ¼ 0.25 for high spins can be more distorted from a circle than a Kerr black hole but still be compatible with the EHT measurement (see Fig. 2). On the other hand, even though we are able to exclude Sen black holes with large electromagnetic charges (see, e.g., the Sen curve for ¯qm ¼ 1.25 in the right panel of Fig. 2) as viable models for M87, its shadow is less distorted from a circle than that of an extremal Kerr black hole (see Fig. 4). In other words, the examples just made highlight the importance of using the appropriate bounds on a sufficiently robust quantity when using the EHT measurement to test theories of gravity. Failing to do so may lead to incorrect bounds on the black- hole properties. For instance, Ref. [54] is able to set bounds on the parameter space of the uncharged, rotating Hayward black hole by imposing bounds on the maximum distortion of the shape of its shadow boundaries, albeit using a different measure for the distortion from a circle [see Eq. (58) there], whereas we have shown that this is not possible, upon using the bounds 4.31M − 6.08M for the size of their shadows (cf. right panel of Fig. 2). FIG. 4. Distortion parameter δsh for a number of stationary black holes observed on the equatorial plane (i ¼ π=2) with dimensionless spin a. Because for observers viewing the black hole from inclinations increasingly close to the pole, the shadow boundary appears increasingly circular, the distortions reported can be taken as upper limits. zero is insignificant for observer viewing angles that are close to the pole of the black hole, as anticipated (not displayed here). [1] R. H. Dicke, Republication of: The theoretical significance of experimental relativity, Gen. Relativ. Gravit. 51, 57 (2019). [2] C. M. Will, The confrontation between general relativity and experiment, Living Rev. Relativity 9, 3 (2006). [3] T. E. Collett, L. J. Oldham, R. J. Smith, M. W. Auger, K. B. Westfall, D. Bacon, R. C. Nichol, K. L. Masters, K. Koyama, and R. van den Bosch, A precise extra- galactic test of General Relativity, Science 360, 1342 (2018). [4] G. ’t Hooft and M. Veltman, One-loop divergencies in the theory of gravitation, Ann. Inst. Henri Poincare Sect. A 20, 69 (1974). [5] N. V. Krasnikov, Nonlocal gauge theories, Theor. Math. Phys. 73, 1184 (1987). [6] H. Lü, A. Perkins, C. N. Pope, and K. S. Stelle, Black Holes in Higher Derivative Gravity, Phys. Rev. Lett. 114, 171601 (2015). [7] J. Scherk and J. H. Schwarz, How to get masses from extra dimensions, Nucl. Phys. B153, 61 (1979). [8] M. B. Green, J. H. Schwarz, and E. Witten, Superstring Theory (Cambridge University Press, Cambridge, England, 1988). [9] E. Barausse, T. Jacobson, and T. P. Sotiriou, Black holes in Einstein-aether and Hoˇrava-Lifshitz gravity, Phys. Rev. D 83, 124043 (2011). [10] E. Barausse and T. P. Sotiriou, Black holes in Lorentz- violating gravity theories, Classical Quantum Gravity 30, 244010 (2013). [11] E. Barausse, T. P. Sotiriou, and I. Vega, Slowly rotating black holes in Einstein-æther theory, Phys. Rev. D 93, 044044 (2016). [12] O. Ramos and E. Barausse, Constraints on Hoˇrava gravity from binary black hole observations, Phys. Rev. D 99, 024034 (2019). [13] O. Sarbach, E. Barausse, and J. A. Preciado-López, Well- theory, posed Cauchy formulation for Einstein-æther Classical Quantum Gravity 36, 165007 (2019). [14] T. Damour and J. H. Taylor, Strong-field tests of relativistic gravity and binary pulsars, Phys. Rev. D 45, 1840 (1992). [15] N. Wex, Testing relativistic gravity with radio pulsars, arXiv:1402.5594. [16] N. Wex and M. Kramer, Gravity tests with radio pulsars, Universe 6, 156 (2020). [17] R. Abuter et al. (GRAVITY Collaboration), Detection of the gravitational redshift in the orbit of the star S2 near the Galactic centre massive black hole, Astron. Astrophys. 615, L15 (2018). [18] R. Abuter, A. Amorim, M. Bauböck, J. P. Berger, H. Bonnet, W. Brandner, V. Cardoso, Y. Cl´enet, P. T. de Zeeuw, and J. Dexter, Detection of the Schwarzschild precession in the orbit of the star S2 near the Galactic 104047-15 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) centre massive black hole, Astron. Astrophys. 636, L5 (2020). [19] B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari et al., Observation of Gravitational Waves from a Binary Black Hole Merger, Phys. Rev. Lett. 116, 061102 (2016). [20] B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Abernathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari et al., Astrophysical implications of the binary black hole merger GW150914, Astrophys. J. Lett. 818, L22 (2016). [21] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. I. The shadow of the super- massive black hole, Astrophys. J. Lett. 875, L1 (2019). [22] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. II. Array and instrumentation, Astrophys. J. Lett. 875, L2 (2019). [23] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. III. Data processing and cali- bration, Astrophys. J. Lett. 875, L3 (2019). [24] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. IV. Imaging the central super- massive black hole, Astrophys. J. Lett. 875, L4 (2019). [25] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. V. Physical origin of the asym- metric ring, Astrophys. J. Lett. 875, L5 (2019). [26] K. Akiyama, A. Alberdi, W. Alef, K. Asada, R. Azulay, A.-K. Baczko, D. Ball, M. Baloković, J. Barrett et al. (Event Horizon Telescope Collaboration), First M87 event horizon telescope results. VI. The shadow and mass of the central black hole, Astrophys. J. Lett. 875, L6 (2019). [27] K. Gebhardt, J. Adams, D. Richstone, T. R. Lauer, S. M. Faber, K. Gültekin, J. Murphy, and S. Tremaine, The black hole mass iN M87 from GEMINI/NIFS adaptive optics observations, Astrophys. J. 729, 119 (2011). [28] D. Psaltis, L. Medeiros, P. Christian, F. Ozel, and the EHT Collaboration, Gravitational Test beyond the First Post- Newtonian Order with the Shadow of the M87 Black Hole, Phys. Rev. Lett. 125, 141104 (2020). [29] T. Johannsen and D. Psaltis, Metric for rapidly spinning black holes suitable for strong-field tests of the no-hair theorem, Phys. Rev. D 83, 124015 (2011). [30] T. Johannsen, Systematic study of event horizons and pathologies of parametrically deformed Kerr spacetimes, Phys. Rev. D 87, 124017 (2013). [31] S. Vigeland, N. Yunes, and L. C. Stein, Bumpy black holes in alternative theories of gravity, Phys. Rev. D 83, 104027 (2011). [32] T. Johannsen, Regular black hole metric with three con- stants of motion, Phys. Rev. D 88, 044002 (2013). [33] L. Rezzolla and A. Zhidenko, New parametrization for spherically symmetric black holes in metric theories of gravity, Phys. Rev. D 90, 084009 (2014). [34] Z. Younsi, A. Zhidenko, L. Rezzolla, R. Konoplya, and Y. Mizuno, New method for shadow calculations: Applica- tion to parametrized axisymmetric black holes, Phys. Rev. D 94, 084025 (2016). [35] R. Konoplya, L. Rezzolla, and A. Zhidenko, General parametrization of axisymmetric black holes in metric theories of gravity, Phys. Rev. D 93, 064015 (2016). [36] P. Kocherlakota and L. Rezzolla, Accurate mapping of spherically symmetric black holes in a parametrized framework, Phys. Rev. D 102, 064058 (2020). [37] C. Bambi and K. Freese, Apparent shape of super-spinning black holes, Phys. Rev. D 79, 043002 (2009). [38] C. Bambi and N. Yoshida, Shape and position of the shadow in the δ ¼ 2 Tomimatsu–Sato spacetime, Classical Quantum Gravity 27, 205006 (2010). [39] L. Amarilla, E. F. Eiroa, and G. Giribet, Null geodesics and shadow of a rotating black hole in extended Chern-Simons modified gravity, Phys. Rev. D 81, 124045 (2010). [40] L. Amarilla and E. F. Eiroa, Shadow of a Kaluza-Klein rotating dilaton black hole, Phys. Rev. D 87, 044057 (2013). [41] S.-W. Wei and Y.-X. Liu, Observing the shadow of Einstein-Maxwell-Dilaton-Axion black hole, J. Cosmol. Astropart. Phys. 11 (2013) 063. [42] P. G. Nedkova, V. K. Tinchev, and S. S. Yazadjiev, Shadow of a rotating traversable wormhole, Phys. Rev. D 88, 124019 (2013). [43] U. Papnoi, F. Atamurotov, S. G. Ghosh, and B. Ahmedov, Shadow of five-dimensional rotating Myers-Perry black hole, Phys. Rev. D 90, 024073 (2014). [44] S.-W. Wei, P. Cheng, Y. Zhong, and X.-N. Zhou, Shadow of noncommutative geometry inspired black hole, J. Cosmol. Astropart. Phys. 08 (2015) 004. [45] M. Ghasemi-Nodehi, Z. Li, and C. Bambi, Shadows of CPR black holes and tests of the Kerr metric, Eur. Phys. J. C 75, 315 (2015). [46] F. Atamurotov, S. G. Ghosh, and B. Ahmedov, Horizon structure of rotating Einstein–Born–Infeld black holes and shadow, Eur. Phys. J. C 76, 273 (2016). [47] B. P. Singh and S. G. Ghosh, Shadow of Schwarzschild- Tangherlini black holes, arXiv:1707.07125. [48] M. Amir, B. P. Singh, and S. G. Ghosh, Shadows of rotating five-dimensional charged EMCS black holes, Eur. Phys. J. C 78, 399 (2018). [49] H. Olivares, Z. Younsi, C. M. Fromm, M. De Laurentis, O. Porth, Y. Mizuno, H. Falcke, M. Kramer, and L. Rezzolla, How to tell an accreting boson star from a black hole, Mon. Not. R. Astron. Soc. 497, 521 (2020). [50] Y. Mizuno, Z. Younsi, C. M. Fromm, O. Porth, M. De Laurentis, H. Olivares, H. Falcke, M. Kramer, and L. Rezzolla, The current ability to test theories of gravity with black hole shadows, Nat. Astron. 2, 585 (2018). [51] P. V. P. Cunha, C. A. R. Herdeiro, and E. Radu, Spontaneously Scalarized Kerr Black Holes in Extended 104047-16 CONSTRAINTS ON BLACK-HOLE CHARGES WITH THE 2017 … PHYS. REV. D 103, 104047 (2021) Scalar-Tensor–Gauss-Bonnet Gravity, Phys. Rev. Lett. 123, 011101 (2019). [52] A. Grenzebach, The Shadow of Black Holes (Springer International Publishing, New York, 2016). [53] Z. Stuchlík and J. Schee, Shadow of the regular Bardeen black holes and comparison of the motion of photons and neutrinos, Eur. Phys. J. C 79, 44 (2019). [54] R. Kumar, S. G. Ghosh, and A. Wang, Shadow cast and deflection of light by charged rotating regular black holes, Phys. Rev. D 100, 124024 (2019). [55] R. Kumar, A. Kumar, and S. G. Ghosh, Testing rotating regular metrics as candidates for astrophysical black holes, Astrophys. J. 896, 89 (2020). [56] K. Hioki and U. Miyamoto, Hidden symmetries, null geodesics, and photon capture in the Sen black hole, Phys. Rev. D 78, 044007 (2008). [57] A. Abdujabbarov, M. Amir, B. Ahmedov, and S. G. Ghosh, Shadow of rotating regular black holes, Phys. Rev. D 93, 104004 (2016). [58] P. O. Mazur and E. Mottola, Gravitational vacuum con- densate stars, Proc. Natl. Acad. Sci. U.S.A. 101, 9545 (2004). [59] C. B. M. H. Chirenti and L. Rezzolla, Ergoregion insta- bility in rotating gravastars, Phys. Rev. D 78, 084011 (2008). [60] R. Shaikh, P. Kocherlakota, R. Narayan, and P. S. Joshi, Shadows of spherically symmetric black holes and naked singularities, Mon. Not. R. Astron. Soc. 482, 52 (2019). [61] D. Dey, R. Shaikh, and P. S. Joshi, Shadow of nulllike and timelike naked singularities without photon spheres, Phys. Rev. D 103, 024015 (2021). [62] R. M. Wald, General Relativity (The University of Chicago Press, Chicago, 1984). [63] J. Bardeen, in Proceedings of GR5, Tbilisi, USSR (Georgia, 1968), p. 174. [64] S. A. Hayward, Formation and Evaporation of Nonsingular Black Holes, Phys. Rev. Lett. 96, 031103 (2006). [65] V. P. Frolov, Notes on nonsingular models of black holes, Phys. Rev. D 94, 104056 (2016). [66] D. I. Kazakov and S. N. Solodukhin, On quantum defor- mation of the Schwarzschild solution, Nucl. Phys. B429, 153 (1994). [67] G. W. Gibbons and K.-I. Maeda, Black holes and mem- branes in higher-dimensional theories with dilaton fields, Nucl. Phys. B298, 741 (1988). [68] D. Garfinkle, G. T. Horowitz, and A. Strominger, Charged black holes in string theory, Phys. Rev. D 43, 3140 (1991). [69] author A. García, D. Galtsov, and O. Kechkin, Class of Stationary Axisymmetric Solutions of the Einstein-Max- well-Dilaton-Axion Field Equations, Phys. Rev. Lett. 74, 1276 (1995). [70] R. Kallosh, A. Linde, T. Ortín, A. Peet, and A. van Proeyen, Supersymmetry as a cosmic censor, Phys. Rev. D 46, 5278 (1992). [71] A. I. Janis, E. T. Newman, and J. Winicour, Reality of the Schwarzschild Singularity, Phys. Rev. Lett. 20, 878 (1968). [72] R. P. Kerr, Gravitational Field of a Spinning Mass as an Example of Algebraically Special Metrics, Phys. Rev. Lett. 11, 237 (1963). [73] E. T. Newman, E. Couch, K. Chinnapared, A. Exton, A. Prakash, and R. Torrence, Metric of a rotating, charged mass, J. Math. Phys. (N.Y.) 6, 918 (1965). [74] A. Sen, Rotating charged black hole solution in heterotic string theory, Phys. Rev. Lett. 69, 1006 (1992). [75] C. Bambi and L. Modesto, Rotating regular black holes, Phys. Lett. B 721, 329 (2013). [76] E. T. Newman and A. I. Janis, Note on the Kerr spinning‐ particle metric, J. Math. Phys. (N.Y.) 6, 915 (1965). [77] B. Carter, Global structure of the Kerr family of gravita- tional fields, Phys. Rev. 174, 1559 (1968). [78] F. Astorga, J. F. Salazar, and T. Zannias, On the integra- bility of the geodesic flow on a Friedmann–Robertson– Walker spacetime, Phys. Scr. 93, 085205 (2018). [79] S. Yazadjiev, LETTER: Newman–Janis method and rotat- ing dilaton-axion black hole, Gen. Relativ. Gravit. 32, 2345 (2000). [80] K. Hioki and K.-I. Maeda, Measurement of the Kerr spin parameter by observation of a compact object’s shadow, Phys. Rev. D 80, 024042 (2009). [81] R. Shaikh, Black hole shadow in a general rotating spacetime obtained through Newman-Janis algorithm, Phys. Rev. D 100, 024028 (2019). [82] J. M. Bardeen, W. H. Press, and S. A. Teukolsky, Rotating black holes: Locally nonrotating frames, energy extraction, and scalar synchrotron radiation, Astrophys. J. 178, 347 (1972). [83] A. A. Abdujabbarov, L. Rezzolla, and B. J. Ahmedov, A coordinate-independent characterization of a black hole shadow, Mon. Not. R. Astron. Soc. 454, 2423 (2015). [84] T. Johannsen and D. Psaltis, Testing the no-hair theorem with observations in the electromagnetic spectrum. II. Black hole images, Astrophys. J. 718, 446 (2010). [85] M. E. Rodrigues and M. V. d. S. Silva, Bardeen regular black hole with an electric source, J. Cosmol. Astropart. Phys. 06 (2018) 025. [86] S. L. Liebling and C. Palenzuela, Dynamical boson stars, Living Rev. Relativity 15, 6 (2012). [87] Z. Meliani, P. Grandcl´ement, F. Casse, F. H. Vincent, O. Straub, and F. Dauvergne, GR-AMRVAC code applica- tions: accretion onto compact objects, boson stars versus black holes, Classical Quantum Gravity 33, 155010 (2016). [88] E. W. Hirschmann, L. Lehner, S. L. Liebling, and C. Palenzuela, Black hole dynamics in Einstein-Maxwell- dilaton theory, Phys. Rev. D 97, 064032 (2018). [89] J. Magueijo, New varying speed of light theories, Rep. Prog. Phys. 66, 2025 (2003). [90] D. L. Wiltshire, M. Visser, and S. M. Scott, The Kerr Spacetime: Rotating Black Holes in General Relativity (Cambridge University Press, Cambridge, England, 2009). [91] A. Held, R. Gold, and A. Eichhorn, Asymptotic safety casts its shadow, J. Cosmol. Astropart. Phys. 06 (2019) 029. [92] C. Bambi, Testing the Bardeen metric with the black hole candidate in Cygnus X-1, Phys. Lett. B 730, 59 (2014). [93] E. Barausse, N. Yunes, and K. Chamberlain, Theory- Agnostic Constraints on Black-Hole Dipole Radiation with Multiband Gravitational-Wave Astrophysics, Phys. Rev. Lett. 116, 241104 (2016). 104047-17 PRASHANT KOCHERLAKOTA et al. PHYS. REV. D 103, 104047 (2021) [94] R. Konoplya and A. Zhidenko, Detection of gravitational waves from black holes: Is there a window for alternative theories? Phys. Lett. B 756, 350 (2016). [95] F.-L. Juli´e, On the motion of hairy black holes in Einstein- Maxwell-dilaton theories, J. Cosmol. Astropart. Phys. 01 (2018) 026. [96] F.-L. Juli´e, Gravitational radiation from compact binary systems in Einstein-Maxwell-dilaton theories, J. Cosmol. Astropart. Phys. 10 (2018) 033. [97] H. M. Siahaan, Merger estimates for Kerr-Sen black holes, Phys. Rev. D 101, 064036 (2020). [98] S. H. Völkel, E. Barausse, N. Franchini, and A. E. Broderick, EHT tests of the strong-field regime of General Relativity, arXiv:2011.06812. [99] http://computeontario.ca. [100] http://www.calculquebec.ca. [101] http://www.computecanada.ca. [102] E. W. Greisen, AIPS, the VLA, and the VLBA, in In: Heck A. (eds) Information Handling in Astronomy - Historical Vistas, edited by A. Heck, Astrophysics and Space Science Library Vol. 285 (Springer, Dordrecht, 2003), https://doi .org/10.1007/0-306-48080-8_7. [103] M. Kettenis, H. J. van Langevelde, C. Reynolds, and B. Cotton, ParselTongue: AIPS Talking Python, in Astro- nomical Data Analysis Software and Systems XV, edited by C. Gabriel, C. Arviset, D. Ponz, and S. Enrique, Astro- nomical Society of the Pacific Conference Series Vol. 351 (2006), https://ui.adsabs.harvard.edu/abs/2006ASPC..351. .497K. [104] O. Tange, login: The USENIX Magazine 36, 42 (2011). [105] A. A. Chael, M. D. Johnson, R. Narayan, S. S. Doeleman, J. F. C. Wardle, and K. L. Bouman, High-resolution linear polarimetric imaging for the event horizon telescope, Astrophys. J. 829, 11 (2016). [106] M. Shepherd, https://ui.adsabs.harvard.edu/abs/2011ascl .soft03001S. [107] S. van der Walt, S. C. Colbert, and G. Varoquaux, The NumPy array: A structure for efficient numerical compu- tation, Comput. Sci. Eng. 13, 22 (2011). [108] E. Jones et al., SciPy: Open Source Scientific Tools for Python (2001), http://www.scipy.org/. [109] W. McKinney, Proc. IX Python in Science Conf., edited by S. van der Walt and J. Millman (2010). [110] T. P. Robitaille et al., Astropy: A community Python package for astronomy, Astron. Astrophys. 558, A33 (2013). [111] A. M. Price-Whelan (The Astropy Collaboration), The astropy project: Building an open-science project and status of the v2.0 core package, Astron. J. 156, 123 (2018). [112] T. Kluyver et al., Positioning and Power in Academic Publishing: Players, Agents and Agendas, edited by F. Loizides and B. Schmidt (IOS Press, 2016). [113] J. D. Hunter, Matplotlib: A 2D graphics environment, Comput. Sci. Eng. 9, 90 (2007). [114] A. E. Broderick, THEMIS: A parameter estimation frame- work for the event horizon telescope, Astrophys. J. 897, 139 (2020). [115] D. W. Pesce, A D-term modeling code (DMC) for simultaneous calibration and full-stokes imaging of very long baseline interferometric data, Astron. J. 161, 178 (2021). [116] I. Martí-Vidal, A. Mus, M. Janssen, P. de Vicente, and J. González, Polarization calibration techniques for the new-generation VLBI, Astron. Astrophys. 646, A52 (2021). [117] J. Park, D.-Y. Byun, K. Asada, and Y. Yun, GPCAL: A generalized calibration pipeline for instrumental polariza- tion in VLBI data, Astrophys. J. 906, 85 (2021). 104047-18
10.1073_pnas.2309151120	BRIEF REPORT \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS Innate immune responses yield tissue- specific bottlenecks that scale with pathogen dose , Katherine G. Daileya,b Karthik Hullahallia,b,1 , and Matthew K. Waldora,b,c,1 Edited by Lawrence Steinman, Stanford University, Stanford, CA; received June 14, 2023; accepted July 31, 2023 To cause infection, pathogens must overcome bottlenecks imposed by the host immune system. These bottlenecks restrict the inoculum and largely determine whether pathogen exposure results in disease. Infection bottlenecks therefore quantify the effectiveness of immune barriers. Here, using a model of Escherichia coli systemic infection, we identify bottlenecks that tighten or widen with higher inoculum sizes, revealing that the efficacy of innate immune responses can increase or decrease with pathogen dose. We term this concept “dose scaling”. During E. coli systemic infection, dose scaling is tissue specific, dependent on the lipopolysaccharide (LPS) receptor TLR4, and can be recapitulated by mimicking high doses with killed bacteria. Scaling therefore depends on sensing of pathogen molecules rather than interactions between the host and live bacteria. We propose that dose scaling quantitatively links innate immunity with infection bottle- necks and is a valuable framework for understanding how the inoculum size governs the outcome of pathogen exposure. bottlenecks \| dose–response \| systemic infection \| innate immunity The COVID- 19 pandemic has catalyzed renewed interest in the long- standing question of how the pathogen inoculum size impacts subsequent infection (1, 2). A critical param- eter that determines whether a given pathogen dose can establish infection is the bottle- neck, which represents host processes that eliminate inoculated microorganisms (PMID: 31992714 and PMID: 36757366) (3–8). Organisms that survive bottlenecks and give rise to the population at an infection site are known as the founding population (FP) (9). FP cannot be measured by enumerating colony- forming units (CFU) alone since total burden is governed by both infection bottlenecks and pathogen replication. We developed Sequence Tag Based Analysis of Microbial Populations in R (STAMPR), a methodology that quantifies FP with barcoded bacteria (10). The relationship between FP and dose measures the bottleneck (3, 4). The few studies that have measured bottlenecks with barcoded bacteria have focused on enteric infections and found that bottlenecks restrict fixed fractions of the inoculum, rather than fixed numbers (3, 4). Thus, barriers to enteric colonization, such as stomach acid and the microbiota, have similar fractional efficacies at low or high doses. However, in other infection models, bottlenecks may constrict or widen in response to changes in dose. For example, positive feedback in the immune system may tighten bottlenecks at higher doses, or bottlenecks may widen at higher doses if immune effectors are overwhelmed. We recently profiled the dynamics of systemic Escherichia coli infection in mice (11). Following intravenous (IV) inoculation, multiple interconnected components of the innate immune system, including the production of proinflammatory cytokines and infiltration of immune cells to tissues, are rapidly triggered in a manner dependent on TLR4. These factors impose bottlenecks and govern infection establishment; FP explicitly quantifies the collective efficacy of these innate immune constituents at a single dose. The dose–FP relationship (i.e., the bottleneck) thus quantifies innate immune potency across all doses. Here, we leveraged STAMPR to define the broader role of TLR4 in modulating the dynamics of E. coli systemic infection and provide a conceptual link between dose, bottlenecks, and innate immunity. Results and Discussion Dose- FP Curves Reveal Tissue- Specific Potency of Innate Immunity. TLR4 knockout mice (TLR4KO) mice and heterozygous littermates (TLR4Het) were intravenously inoculated with varying doses of barcoded E. coli. Five days postinfection (dpi), the lungs, liver, and spleen were harvested for bacterial enumeration and STAMPR analysis. All organs exhibited dose- dependent increases in CFU. At higher doses, abscesses appeared in the livers of TLR4Het animals, but not in TLR4KO animals (Fig. 1A, blue box). Consistent with our companion study (12), abscesses result aDepartment of Microbiology, Author affiliations: Har vard Medical School, Boston, MA 02115; bDivision of Infectious Disease, Brigham & Women’s Hospital, Boston, MA 02115; and cHHMI, Boston, MA 02115 Preprint Servers: https://doi.org/10.1101/2023.06.09.54 3079 Author contributions: K.H., K.G.D., and M.K.W. designed research; K.H. and K.G.D. performed research; M.K.W. contributed new reagents/analytic tools; K.H. and K.G.D. analyzed data; and K.H., K.G.D., and M.K.W. wrote the paper. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2309151120/- /DCSupplemental. Published September 5, 2023. PNAS 2023 Vol. 120 No. 37 e2309151120 https://doi.org/10.1073/pnas.2309151120 1 of 3 spleen is dependent on TLR4 since in its absence, the slope increases to ~1 (“neutral scaling” Fig. 1G). Neutral scaling was observed in the livers of TLR4Het animals. In the absence of TLR4 in the liver, the slope increased to >1; fractionally more bacteria survive at higher doses (fractionally fewer are killed). Thus, in the absence of TLR4, the hepatic innate immune response is fractionally less effective at larger doses, which we term “negative scaling”. In contrast to the liver and spleen, we observed neutral scaling in the lung independent of TLR4. These results reveal that in addition to controlling tissue- specific immune responses, TLR4 governs the extent to which the effi- cacy of these responses quantitatively scales with inoculum size; mice lacking TLR4 have disproportionately reduced potency of innate immune responses at higher doses. Notably, TLR4 not only controls pathogen burdens but also their diversity, which may be important for longer- term population- level outcomes such as transmission of drug resistance or the emergence of hypervirulent pathogen lineages. Dose Scaling Does Not Require Live Bacteria. We hypothesized that dose scaling is separable from the true number of inoculated organisms and is instead dependent on how the innate immune system senses and “interprets” the inoculum size. To test this idea, we spiked killed bacteria into a live inoculum to simulate some of the immune- stimulatory aspects of high doses. TLR4Het and TLR4KO animals were inoculated with 1.25 × 106 CFU live bacteria or the same quantity of live cells plus 5 × 107 formalin- killed E. coli (Fig. 2A). Both groups received identical amounts of live bacteria (mean log10 dose ± St.dev, control: 6.03 ± 0.14, spike- in: 6.13 ± 0.10). CFU (Fig. 2 B, D, and F) and FP (Fig. 2 C, E, and G) were calculated at 5 dpi. The potency of splenic immune responses was greater at higher doses (positive scaling) in a manner dependent on TLR4 (Fig. 1D). Consistent with increased innate immune efficacy at higher doses, spike- in of killed bacteria yielded threefold fewer live founders in TLR4Het spleens (Fig. 2E), and the decrease in FP was not observed in the absence of TLR4. In contrast, the livers of TLR4KO animals were scaled negatively (Fig. 1B). With spiked killed cells, TLR4KO livers had significantly greater CFU and fivefold more founders than control animals without spiked- in killed cells (Fig. 2C). No changes in CFU or FP were seen with spiked- in cells in the livers of TLR4Het animals, consistent with neutral scaling (Fig. 2C). Therefore, positive scaling in the spleen and negative scaling in the liver do not require live bacteria. The magnitude of changes in FP in response to killed cells is close to predictions from dose–FP curves (Fig. 1). By multiplying the fractional bottleneck (FP/Dose) at a 2 × 107 dose with an actual dose of 1.25 × 106, the TLR4Het spleen and TLR4KO liver are predicted to have threefold fewer and fourfold greater found- ers, respectively, relative to control animals without spiked- in cells. The effect of killed cells on CFU in the spleen is not sta- tistically significant despite a significant difference in FP since two mice had stochastic bacterial replication (Fig. 2D). However, the livers of TLR4KO mice had significantly greater CFU after spike- in (Fig. 2B). Thus, when bacteria can replicate substan- tially, FP rather than CFU is a more faithful measure of scaling since it is not confounded by bacterial expansion. Although the lungs exhibited neutral scaling, we observed a minor but statis- tically significant decrease in CFU and FP following spike- in (Fig. 2 F and G). These data show that dose scaling can be recapitulated without increasing the inoculum size. Furthermore, the two seemingly opposite effects of spiking in killed cells (higher FP in TLR4KO Fig. 1. Bottleneck–dose–response analysis for E. coli systemic infection. (A and B) CFU (A) and FP (Ns) (B) are shown for the liver as a function of dose. Each point represents an animal, and the blue box indicates animals with abscesses. C–F are identical to A and B but represent spleen and lung CFU (C and E) and FP (D and F). Best fit lines from linear regression in FP plots are shown with 95% confidence bands. (G) Dose–FP plots for different scaling patterns. Dotted lines represent a 0% bottleneck. With slopes greater than one, fractionally more bacteria survive to become founders at higher doses; innate immune responses are less effective at higher inoculum sizes and therefore “negatively scale” with dose. With slopes less than one, the immune response is “positively scaled” since fractionally fewer bacteria survive at higher doses. With a slope equal to 1, a fixed fraction of the inoculum survives host bottlenecks. from bacterial replication (low FP and high CFU). TLR4KO mice are resistant to abscess formation (Fig. 1A, outside blue box) but have higher FP (Fig. 1B), indicating that in the absence of TLR4, more bacteria from the inoculum survive infection bottlenecks. In the spleen, TLR4KO animals also had higher burdens than TLR4Het animals (Fig. 1C). FPs were also higher in TLR4KO spleens, revealing that TLR4 controls splenic bottlenecks (Fig. 1D). In contrast to both the liver and spleen, neither bacterial burden nor FP was influenced by TLR4 in the lungs (Fig. 1 E and F). Bottlenecks are quantified by the relationship between FP and dose (3, 4). Bottlenecks with a slope of 1 on dose–FP curves indicate that numerically more bacteria are eliminated at higher doses, but the fraction killed (and fraction surviving) is constant. TLR4KO animals had higher dose–FP slopes compared to litter- mate TLR4Het animals in the spleen [0.9 ± 0.07 (SE) vs. 0.56 ± 0.06, Fig. 1D] and liver (1.6 ± 0.17 vs. 1.1 ± 0.14, Fig. 1B). Although TLR4 influences the slopes in both organs by a similar magnitude, the specific numerical values reveal tissue- specific differences in the potency of innate immunity. The slope <1 in the TLR4Het spleen is an example of “positive scaling” (Fig. 1G), where innate immunity is more effective at higher inoculum sizes; fractionally fewer bacteria survive to become founders at larger doses (fractionally more are killed). Positive scaling in the 2 of 3 https://doi.org/10.1073/pnas.2309151120 pnas.org livers but lower FP in the TLR4Het spleen) result from a similar conceptual basis in dose scaling. The dependence of the effects of killed cell spike- in on TLR4 suggests that dose scaling results from quantitative changes in LPS- induced immune responses, rather than the actual pathogen dose. The increase in liver FP following spike- in only in the absence of TLR4 suggests that pathogen mol- ecules other than LPS yield negative scaling patterns. We speculate that the host pathways engaged by these other molecules are more easily exhausted in the absence of TLR4. Other pathogen and host factors that control immune responses, such as TLR5 (flagellin) or Nod2 (peptidoglycan), may benefit from analysis with the dose–FP paradigm to decipher how innate immunity scales with dose to regulate infection outcome. Concluding Remarks We describe the concept of dose scaling, which relates changes in inoculum size with quantitative changes in infection bottlenecks and the efficacy of innate immune responses. We show that dif- ferent pathogen doses can yield tighter or wider bottlenecks in a tissue- specific manner. Since scaling can in part be recapitulated with killed organisms, leveraging scaling may represent a frame- work to identify therapeutics that tighten infection bottlenecks. We hypothesize that positive and negative scaling arise from dif- ferences in the rate of induction of individual components of the innate immune response as a function of dose. These differences are likely influenced by the underlying immune cell composition of different tissues as well as tissue- specific gene expression pat- terns. Further studies of dose scaling will provide critical contex- tualization for natural infections, where inoculum sizes are not controlled. Materials and Methods Ns calculations were performed as previously described (4). Note that all liver CFU calculations are reported for ¼ of the liver to accurately compare Ns and CFU. Details of animal experiments and STAMPR analysis are provided in SI Appendix, Materials and Methods. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. STAMPR code is available at https://github.com/ hullahalli/stampr_rtisan (13). ACKNOWLEDGMENTS. This work is supported by NIH F31AI156949 (K.H.), NIH R01AI042347 (M.K.W.), and the Howard Hughes Medical Institute (M.K.W.). We are grateful to members of the Waldor lab for feedback on this manuscript. Fig. 2. Tissue- and TLR4- dependent responses to spike- in of killed bacteria. (A) Barcoded live bacteria or the same quantity of live cells plus 40- fold excess of formalin- fixed bacteria were IV inoculated into TLR4Het or TLR4KO littermates. (B and C) CFU (B) and FP (Ns) (C) of the liver are shown (lines represent medians). The blue box indicates animals with abscesses. (D–G) are identical to B and C but represent spleen and lung CFU (D and F) and FP (E and G). (H) Curves from Fig. 1 (Left to Right: liver, spleen, and lung) are schematized for reference. 1. 2. 3. 4. 5. 6. 7. L. G. Rubin, Bacterial colonization and infection resulting from multiplication of a single organism. Rev. Infect Dis. 9, 488–493 (1987). E. E. Bendall et al., Rapid transmission and tight bottlenecks constrain the evolution of highly transmissible SARS- CoV- 2 variants. Nat. Commun. 14, 1–7 (2023). A. N. Gillman, A. Mahmutovic, P. Abel zur Wiesch, S. Abel, The Infectious Dose Shapes Vibrio cholerae Within- Host Dynamics. mSystems 6, e00659–21 (2021). I. W. Campbell, K. Hullahalli, J. R. Turner, M. K. Waldor, Quantitative dose- response analysis untangles host bottlenecks to enteric infection. Nat. Commun. 14, 1–13 (2023). E. J. G. Pollitt, P. T. Szkuta, N. Burns, S. J. Foster, Staphylococcus aureus infection dynamics. PLoS Pathog. 14, e1007112. (2018). T. Zhang et al., Deciphering the landscape of host barriers to Listeria monocytogenes infection. Proc. Natl. Acad. Sci. U.S.A. 114, 6334–6339 (2017). K. E. R. Bachta et al., Systemic infection facilitates transmission of Pseudomonas aeruginosa in mice. Nat. Commun. 11, 543 (2020). 8. D. Hoces et al., Fitness advantage of Bacteroides thetaiotaomicron capsular polysaccharide in the mouse gut depends on the resident microbiota. Elife. 12, e81212 (2023). S. Abel, P. Abel zur Wiesch, B. M. Davis, M. K. Waldor, Analysis of Bottlenecks in Experimental Models of Infection. PLoS Pathog. 11, e1004823. (2015). 9. 10. K. Hullahalli, J. R. Pritchard, M. K. Waldor, Refined Quantification of Infection Bottlenecks and Pathogen Dissemination with STAMPR. mSystems 6, e00887–21 (2021). 11. K. Hullahalli, M. K. Waldor, Pathogen clonal expansion underlies multiorgan dissemination and organ- specific outcomes during murine systemic infection. Elife 10 (2021). 12. K. Hullahalli et al., Genetic and immune determinants of E. coli liver abscess formation. bioRxiv 543319 (2023). 13. K. Hullahalli, K. G. Dailey, M. K. Waldor, DoseScalingTLR4. stampr_rtisan. https://github.com/ hullahalli/stampr_rtisan. Deposited 17 August 2023. PNAS 2023 Vol. 120 No. 37 e2309151120 https://doi.org/10.1073/pnas.2309151120 3 of 3
10.1073_pnas.2301121120	RESEARCH ARTICLE \| BIOCHEMISTRY OPEN ACCESS Gβγ activates PIP2 hydrolysis by recruiting and orienting PLCβ on the membrane surface Maria E. Falzonea,b and Roderick MacKinnona,b,1 Edited by James Hurley, University of California, Berkeley, CA; received January 19, 2023; accepted April 6, 2023 Phospholipase C-βs (PLCβs) catalyze the hydrolysis of phosphatidylinositol 4, 5–bisphosphate (PIP2) into inositoltriphosphate (IP3) and diacylglycerol (DAG). PIP2 regulates the activity of many membrane proteins, while IP3 and DAG lead to increased intracellular Ca2+ levels and activate protein kinase C, respectively. PLCβs are regulated by G protein–coupled receptors through direct interaction with G𝛼q and G𝛽𝛾 and are aqueous-soluble enzymes that must bind to the cell membrane to act on their lipid sub- strate. This study addresses the mechanism by which G𝛽𝛾 activates PLCβ3. We show that ∼ 0.43 mol % ) PLCβ3 functions as a slow Michaelis–Menten enzyme ( kcat on membrane surfaces. We used membrane partitioning experiments to study the solution-membrane localization equilibrium of PLCβ3. Its partition coefficient is such that only a small quantity of PLCβ3 exists in the membrane in the absence of G𝛽𝛾 . When G𝛽𝛾 is present, equilibrium binding on the membrane surface increases PLCβ3 in the membrane, increasing Vmax in proportion. Atomic structures on membrane vesicle surfaces show that two G𝛽𝛾 anchor PLCβ3 with its catalytic site oriented toward the membrane surface. Taken together, the enzyme kinetic, membrane partitioning, and structural data show that G𝛽𝛾 activates PLCβ by increasing its concentration on the membrane surface and orienting its catalytic core to engage PIP2 . This principle of activation explains rapid stimulated catalysis with low background activity, which is essential to the biological pro- cesses mediated by PIP2, IP3, and DAG. ∼ 2 s−1, KM PLCβ \| Gβγ \| PIP2 \| GPCR signaling \| membrane recruitment Phospholipase C-β (PLCβ) enzymes cleave phosphatidylinositol 4,5-bisphosphate ( PIP2 ) into inositoltriphosphate ( IP3 ) and diacylglycerol ( DAG ) (1, 2). Their activity is controlled by G protein–coupled receptors (GPCRs) through direct interaction with G proteins (3–5). IP3 increases intracellular calcium, DAG activates protein kinase C, and levels of PIP2 regulate numerous ion channels. Therefore, the PLCβ enzymes under GPCR regu- lation are central to cellular signaling (Fig. 1A) (6–8). There are four PLCβs (1–4) in humans: PLC 𝛽4 is activated by G𝛼q, and PLCβ1–3 are activated by both G𝛼q and G𝛽𝛾. PLCβ2/3 are also activated by the small GTPases Rac1/2 (9–15). What do we know about PLCβs and their regulation by G proteins? PLCβs are cytoplasmic enzymes that must access the membrane where their substrate PIP2 resides in the inner leaflet. They contain a catalytic core, a proximal C-terminal domain (CTD) with autoinhibitory activity, and a distal CTD with structural homology to a bin-amphiphysin-Rvs domain important for membrane binding (3, 4). At the active site, an X–Y linker exerts additional autoinhibitory regulation by direct occlusion (9, 15–17). G𝛼q binds to the proximal and distal CTDs, displacing the autoinhibitory proximal CTD from the catalytic core and Rac1 binds to the PH domain of PLCβ2 (9, 18–21). Notably, in both cases the autoinhibitory X–Y linker still occludes the active site. Less is known about regulation of PLCβs by G𝛽𝛾 . Potential binding sites have been described, but no structures have been determined (3, 4). The focus of this study is regulation of PLCβ3 by G𝛽𝛾. The mechanism of PLCβ activation by G𝛽𝛾 is unknown. In vitro studies have con- cluded that locally concentrating PLCβ on the membrane is not the basis of activation and this still dominates thinking in the field (3, 4, 22–28). However, the requirement of the lipid group on G𝛽𝛾 to achieve activation and the demonstration that over expression of G proteins in cells increases PLCβ in the membrane fraction suggests that a localization mechanism needs revisiting (13, 29). Part of the challenge in char- acterizing PLCβ enzymes is precisely the membrane involvement. PLCβs reside in 3 dimensions (the cytoplasm) but catalyze on a two-dimensional surface (the membrane). Functional measurements must account for this and at the same time permit sufficient time resolution, unlike the standard radioactive assay used in the field until now. To overcome the challenge, we have developed new functional methods, including a rapid kinetic analysis of PLCβ3 enzyme activity that employs a direct read-out of PIP2 concentration as a function of time, a membrane partitioning assay to quantify Significance GPCRs are major mediators of transmembrane signal transduction, responding to a wide range of stimuli including hormones and neurotransmitters. Important targets of GPCR signaling, PLCβ enzymes catalyze the hydrolysis of PIP2 into IP3 and DAG, leading to increased intracellular Ca2+ levels and activation of PKC, respectively. PLCβs exhibit very low basal activity through multiple mechanisms of autoinhibition and are activated by both G𝛼q and G𝛽𝛾 . In this study, we demonstrate that G𝛽𝛾 activates PLCβ by recruiting it to the membrane where its substrate PIP2 resides and by orienting its active site. This activation mechanism permits robust and rapid activation of PLCβ upon GPCR stimulation in the setting of low background activity during GPCR quiescence. Author affiliations: aLaboratory of Molecular Neuro biology and Biophysics, The Rockefeller University, New York, NY 10065; and bHHMI, The Rockefeller University, New York, NY 10065 Preprint servers: Deposited as a preprint on bioRxiv. Author contributions: M.E.F. and R.M. designed research; M.E.F. performed research; M.E.F. and R.M. analyzed data; and M.E.F. and R.M. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2301121120//DCSupplemental. Published May 12, 2023. PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 1 of 11 membrane recruitment, and atomic structures on lipid mem- brane surfaces, to analyze the mechanism by which G𝛽𝛾 acti- vates PLCβs. Results To explain with accuracy our data analysis, we present a series of equations and their rationale. At least a qualitative under- standing of these equations is required to fully appreciate the meaning and wider significance of the data, and what it implies about the molecular mechanisms crucial for PLCβ3 function. Some of the analysis and associated equations are, to our knowl- edge, unfamiliar to biochemical analysis. In particular, when analyzing both the kinetics of PIP2 hydrolysis on a membrane surface and the equilibrium binding reaction between proteins on a membrane surface, we encountered the complex issue of processes occurring in 2 dimensions that involve components in 3 dimensions. We dealt with this issue in a particular way, which we describe thoroughly to stimulate debate and invite critique. We appreciate that many readers will want to grasp the biological implications of this work without getting bogged down by equations. For this reason, we have explained the meaning of each equation in words, which should be sufficient to understand the main conclusions of this work. Development of a Planar Lipid Bilayer Assay for PLCβ3 Function. We developed a detergent-free, planar lipid bilayer assay to measure PLCβ3 function using a PIP2-dependent ion channel to report its concentration over time (Fig. 1 B–D). Briefly, two chamber cups were connected in the vertical configuration by a ~250 𝜇m hole in a 100 𝜇m piece of Fluorinated ethylene propylene copolymer (30). A ground electrode was placed in the Cis chamber and a reference electrode in the Trans chamber (Fig. 1B). Lipids dispersed in decane were used to paint a bilayer over the hole separating the two chambers. We used a 2:1:1 mixture of 1, 2-dioleoyl-sn-glycero-3-phosphoethanolamine (DOPE): 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC): 1- palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS) lipids and included a predetermined mole fraction of long-chain PIP2 inside the membrane to set its starting concentration. Ion channels and G proteins were incorporated by proteolipid vesicle application to the bilayer, and the current from reconstituted ion channels was measured (30). We added PLCβ3 to the Cis chamber, which was subjected to continuous mixing to ensure homogeneity of the chamber. The PIP2-dependent, G protein-dependent inward rectifier K+ channel-2 (GIRK2, specified as GIRK) was used as a readout of PIP2 concentration. This channel is well characterized in vitro, strictly depends on PIP2 for channel opening, and is amenable to measuring large macroscopic currents using planar lipid bilayers (31, 32). Further, GIRK exhibits fast rates of association and disassociation of A B C D Fig. 1. Development of a planar lipid bilayer assay for PLCβ activity using a PIP2 dependent ion channel as readout. (A) Cartoon summary of G𝛼idependent signaling to PLCβ through G𝛽𝛾 . (B) Cartoon schematic of planar lipid bilayer setup used to measure PLCβ function. (C) Representative current decay upon PLCβ dependent depletion of PIP2 . (D) Representative current recovery upon reactivation of incorporated channels with shortchain C8PIP2. This experiment was carried out under subsaturating longchain PIP2 (1.0 mol % ) in the bilayer, which correlates to ~30% of maximal GIRK current. In D, saturating C8PIP2 was added, which leads to ~3× the amount of starting current. 2 of 11 https://doi.org/10.1073/pnas.2301121120 pnas.org PIP2 , which permits the measurement of PLCβ3 catalytic activity that is not filtered by a slow channel response (31). Experiments were carried out in the presence of symmetric MgCl2 to ensure blockage of channels with their PIP2 binding sites facing the Trans chamber, which is not accessible to PLCβ3 (Fig. 1B) (33). This ensures that when positive voltage is applied to the reference relative to the ground, the current is derived only from channels accessible to PLCβ3 added to the Cis chamber. GIRK also requires G𝛽𝛾 for channel activity. To separate the effects of G𝛽𝛾 on channel function and PLCβ3 activity we used the ALFA nanobody system (34) to tether soluble G𝛽𝛾 to GIRK (35). We tagged GIRK with the short ALFA peptide on the C-terminus and G𝛾 with the ALFA nanobody on the N-terminus in the back- ground of the C68S mutant, which prevents lipidation of G𝛾 . Nanobody-tagged G𝛾 assembled normally with G𝛽 and was able to bind to other effectors (35). Because the ALFA nanobody binds to the ALFA tag with ~30 pM affinity (34), at 30 nM concentration, the ALFA nanobody-tagged G𝛽𝛾 fully activates ALFA peptide-tagged GIRK. In addition, the nanobody-tagged G𝛽𝛾 does not activate PLCβ3 due to its lack of a lipid anchor (13, 29). Human PLCβ3 was used to establish our assay owing to its significant activation by both G𝛽𝛾 and G𝛼q (14, 36, 37). The addition of PLCβ3 to membranes already containing lipidated G𝛽𝛾 , following an equilibration period of about 2 s, led to a rapid current decay that was complete in ~20 s (Fig. 1C). Subsequent addition of 32 𝜇M C8PIP2, an aqueous-soluble, short chain ver- sion of PIP2, rescued the current to a maximum level (Fig. 1D)(31), indicating that the current decay was due to PIP2 depletion from the bilayer by the PLCβ3 enzyme. The PLCβ3 mediated current decay was slower than when C8PIP2 is rapidly removed by perfu- sion (31). Furthermore, the rate of PLCβ3-mediated current decay depends on the PLCβ3 concentration (SI Appendix, Fig. S1). These findings indicate that the decay measures the rate of PLCβ3 cata- lytic activity rather than PIP2 unbinding from the channel. No change in the current was observed following PLCβ3 addition in the absence of CaCl2 (2 mM EGTA), which is required for enzy- matic function (SI Appendix, Fig. S1A). Repetitions of these exper- iments yielded consistent results with very similar time courses of current decay. These observations indicate that we can measure PLCβ3 catalytic activity using this system and that the addition of PLCβ3 does not induce artifacts to the bilayer or to reconsti- tuted GIRK channels. Kinetic Analysis of PIP2 Hydrolysis by PLCβ3. The interfacial nature of PLCβ3 activity presents a challenge to the study of its function because PLCβ3 is a soluble enzyme that must associate with the membrane to carry out catalysis. To describe the reaction occurring at the two-dimensional membrane surface, which must account for the exchange of PLCβ3 with the three-dimensional water phase, we give concentrations as dimensionless mole fraction × 100 ( mf , expressed as mol % ) using square brackets, [quantity], unless specified as molar units using square brackets with subscript molar, [quantity]molar. Furthermore, to simplify expressions, we approximate mf within each solvent phase, water or lipid, as moles solute per moles solvent rather than moles solute per moles solvent plus solute. This approximation introduces into the kinetic analysis a maximum error in mf of 1.0 % for the PIP2 concentration in membranes and less than 1.0 % for all other components. For PIP2 in membranes, the initial mf is predetermined through the bilayer lipid composition. For PLCβ3, the mf in membranes is calculated from that in three-dimensional solution using its partition coefficient, which is described below. The measured current decays can be converted to PIP2 decays using the PIP2 concentration dependence of the channel, which we determined using titration experiments. Bilayers were formed with varying concentrations of long chain PIP2 from 0.1 to 4.0 mol % , GIRK-containing vesicles were fused, the current was measured, and water-soluble C8PIP2 (32 𝜇M) was added to the Cis chamber to activate the channels maximally (SI Appendix, Fig. S1 B and C) (31). The measured current was normalized to the maximally activated current, Imax , for each PIP2 concentration and fit to a modified Hill equation, Eq. 1, to determine values A, k, and r (Fig. 2A): I Imax = A [PIP2]r kr + [PIP2]r . [1] Eq. 1 is an empirical function whose utility is to convert GIRK current into PIP2 concentration. In subsequent experiments with PLCβ3, bilayers initially contain 1.0 mol % PIP2 , which corre- sponds to ~30% of the maximal current (Fig. 2A). The PLCβ3/Gβγ-dependent current decays were corrected by subtracting a constant current value representing nonspecific leak, then normalized to the starting PIP2 concentration (1.0 mol % ), and converted to PIP2 concentration decays using Eq. 1 with the predetermined values for k , A , and r (Fig. 2B). After an approxi- mately 2 s delay associated with mixing of PLCβ3, PIP2 decays contained two components: an initial, approximately linear com- ponent followed by a slower, approximately exponential compo- nent (Fig. 2C). The linear component is consistent with PLCβ3 catalysis occurring as a 0th order reaction, where the catalytic rate is independent of the PIP2 concentration, suggesting that at our starting concentration (1.0 mol % PIP2 ), the active site of PLCβ3 is nearly fully occupied by substrate ( PIP2 ). The second, expo- nential, component is consistent with the PIP2 concentration becoming limiting to catalysis, a first-order reaction, as the decay progresses and the concentration of PIP2 decreases. In the example shown, for illustrative purpose, we estimated the rate within six intervals along the decay curve, demarcated with different colored circles (Fig. 2C), by measuring the slope to approximate d [PIP2] within each interval, and then plotted the slope’s absolute value against the average PIP2 concentration for the corresponding interval (Fig. 2D). A Michaelis–Menten equation (Eq. 2, below) fit the data points with R2 ~ 0.99, indicating that PLCβ3 catalytic activity can be described by this kinetic rate equation (Fig. 2D). The graphical procedure described above and in Fig. 2 C and D was used as an example to place the PIP2 hydrolysis data into a familiar form of rate as a function of substrate concentration. For processing all data, we took a more direct approach to analyze the time-dependent decays within the Michaelis–Menten frame- work. Expressing the Michaelis–Menten rate equation as dt d [PIP2] dt = − Vmax KM [PIP2] + [PIP2] , [2] and integrating from t = 0, we obtain for the PIP2 concentration as a function of time [PIP2(t )] = KM ProductLog e ([PIP2(0)]−t Vmax) KM [PIP2(0)] , [3] KM where [PIP2(0)] is the PIP2 concentration at t = 0 and KM and Vmax are the Michaelis–Menten parameters. Eq. 3 derived here contains a well-known function called the Lambert W function or ProductLog function (38). It describes for the PIP2 concen- tration an initially linear decay followed by an exponential decay. Substituting Eq. 3 into Eq. 1, we obtain an expression for GIRK current decay as a function of time due to PIP2 hydrolysis, I Imax = C + A [PIP2(t )]r kr + [PIP2(t )]r , [4] PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 3 of 11 A C B D E Fig. 2. Extraction of values for kinetic parameters for PLCβ3 catalysis in the presence of lipidated G𝛽𝛾 from current decay curves. (A) PIP2 activation curve for GIRK varying the mole % of PIP2 in the bilayer and maximally activating with C8PIP2. Green diamonds are average values, open circles are values from each experiment, and error bars are SEM. Each point is from 3–5 experiments. The normalized current (I/Imax) is fit to a modified hill equation, Eq. 1 (dashed red curve). R2 = 0.994. (B) Demonstration of using the PIP2 activation curve (Right) to convert the current decay (Left) to PIP2 decay. Points on the normalized current decay are matched to mol % PIP2 and time. (C) Resulting PIP2 decay over time. Circles denote regions used for measuring the rates graphed in D. (D) Plot of d[PIP2]mf∕dt at regions demarcated in C vs [PIP2] fit to the Michaelis–Menten equation, Eq. 2. R2 = 0.993. (E) Direct fit (shown as red curve) of the normalized current decay with Vmax, KM, and C as free parameters (Eq. 4). The gray dashed line denotes where the fit starts, which excludes an initial equilibration period. R2 = 0.975. which permits direct fitting of the normalized current decays to estimate Vmax and KM (Fig. 2E). [PIP2(t )] in Eq. 4 is given by Eq. 3, and a third free parameter, C , accounts for the level of background leak in bilayer experiments; this is visible as the small residual current (typically ≤ 5% of the GIRK current) at long times in Figs. 2E and 3A. [PIP2(0)], the initial PIP2 concentra- tion, is specified by the bilayer composition and A , k, and r are predetermined through the fit of Eq. 1 to the data shown in Fig. 2A. Eq. 4 fits the current decay data accurately after ~2 s (Fig. 2E) and yields consistent results for Vmax (0.17 ± 0.02 mol % ∕ sec ) and KM (0.42±0.05 mol % ) across repeated experiments (Fig. 3C). The Role of Gβγ in the Function of PLCβ3. In the experiments described above, G𝛽𝛾 was added to the planar lipid bilayers by equilibrating lipid vesicles containing G𝛽𝛾 with the bilayer surface prior to the application of PLCβ3. When G𝛽𝛾 is not added to the bilayer, PLCβ3 produces a much slower current decay, as shown (Fig. 3A and SI Appendix, Fig. S1 D and E). Similarly, in the presence of 1 µM aqueous-soluble G𝛽𝛾 without a lipid anchor, which does not partition onto the membrane surface (31), PLCβ3 catalyzed current decay is also slow (Fig. 3 B and C). Seven experiments were carried out in the absence of G𝛽𝛾 and the rmsd between the current decay curves and Eq. 4 were minimized to yield Vmax (0.0026 ± 0.0007 mol % ∕ sec ) and KM (0.43 ± 0.05 mol % ) (Fig. 3C). Thus, G𝛽𝛾 in the membrane increases Vmax ~65- fold without affecting KM (Fig. 3C). Because PLCβ3 is soluble in aqueous solution but must localize to the membrane surface to catalyze PIP2 hydrolysis, we next examined whether G𝛽𝛾 in the membrane influences PLCβ3 mem- brane localization. As detailed by White and colleagues, protein association with membranes cannot be considered as a simple binding equilibrium due to the fluid nature of the membrane without discrete binding sites (39). Instead, membrane association must be treated as a partitioning process between two immiscible solvents, the membrane and the aqueous solution. The equilib- rium partition coefficient, Kx, is the ratio of the mole fraction of PLCβ3 in the membrane (subscript m) to that in aqueous solution (subscript w) (39), Kx = [PLC 𝛽3m] [PLC 𝛽3w] . [5] To determine the value of Kx , detergent-free liposomes were recon- stituted using 2DOPE:1POPC:1POPS lipids to match the lipid composition of the bilayer experiments, and H+ NMR was used to measure the lipid concentration at the end of the detergent removal process (SI Appendix, Fig. S2A). Large unilamellar vesicles (LUVs) were prepared from the reconstituted liposomes using freeze–thaw cycles and extrusion through a 200 nm membrane. The LUVs were incubated with PLCβ3 and pelleted using ultra- centrifugation to separate the membrane-bound and aqueous protein fractions. This method allows direct measurement of both the bound and free protein using fluorescently labeled PLCβ3, which facilitates determining the partition coefficient from each experiment individually (39). The membrane-associated fraction of PLCβ3, fraction partitioned ( Fp), is = Fp [PLC 𝛽3m] [L]molar [PLC 𝛽3m] [L]molar + [PLC 𝛽3w] [W ]molar = Kx [L]molar Kx [L]molar + [W ]molar . [6] [W ]molar , the molar concentration of water, is ~55 M and [L]molar , the molar concentration of lipid, is set for each experiment using a stock measured by NMR. Thus, Eq. 6 is a function of the single 4 of 11 https://doi.org/10.1073/pnas.2301121120 pnas.org A C E B D F mol % s Fig. 3. G𝛽𝛾 activates PLCβ3 by increasing its concentration at the membrane. (A) Comparison of normalized current decay in the presence (pink) and absence , C = −0.03 ± 0.0004, KM = 0.43 ± 0.0008 mol % , R2 = (gray) of lipidated G𝛽𝛾 fit to Eq. 4 (black curves). Results from the fit without G𝛽𝛾 : Vmax =0.0023 ± 0.6E6 , C = 0.0074 ± 5E5, KM = 0.37 ± 0.0006 mol % . (B) Normalized current decay in the presence of 1 𝜇M soluble G𝛽𝛾 fit to 0.992. With G𝛽𝛾 : Vmax = 0.22 ± 0.0001 Eq. 4 (red curve). R2 = 0.955. (C) Comparison of Vmax , KM , and kcat for PLCβ3 alone, with lipidated G𝛽𝛾 ( G𝛽𝛾 (l)) and with soluble G𝛽𝛾 (G𝛽𝛾 (s)). (D) Membrane partitioning curve for PLCβ3 alone (black) or in the presence of lipidated G𝛽𝛾 (pink) for 2DOPE:1POPC:1POPS LUVs with Fraction Partitioned ( Fp ) on the Y axis. Data for 0 G𝛽𝛾 were fit to Eq. 6 for Kx (dashed black curve) and data for +G𝛽𝛾 were fit to Eq. 7 to determine Keq (39). Error bars are range of mean from two experiments for each lipid concentration. R2 = 0.96 in the absence of G𝛽𝛾 and R2 = 0.95 in the presence of G𝛽𝛾 . (E) Cartoon representation of PLCβ3 activation by G𝛽𝛾 through membrane recruitment. G𝛽𝛾 significantly increases the membrane association of PLCβ, and accordingly [PLCβ]membrane, which amplifies PIP2 hydrolysis. [PLCβ]membrane was calculated from Eq. 8 using [PLCβw]=5.3E8 mol%, [Gβγ]=[Gtot]=0.34 mol%, and Kx and Keq, which were determined through the fits in panel D. (F) Calculated Michaelis–Menten curves (from Eq. 2) for PLCβ3 alone (black), in the presence of 1 𝜇M soluble G𝛽𝛾 (blue) or in the presence of lipidated G𝛽𝛾 (pink) using the values for KM and Vmax determined from our fits. mol % s free parameter, Kx , which we determine by fitting Eq. 6 to the partitioning data, yielding Kx ~2.9 ⋅ 104 (Fig. 3D, black curve). Partitioning experiments carried out with unlabeled PLCβ3 quan- tified using sodium dodecyl sulfate–polyacrylamide gel electro- phoresis (SDS-PAGE) analysis yielded a similar value of Kx ( ∼ 4 ⋅ 104 ) (SI Appendix, Fig. S2 B and C), confirming that the fluorescent label does not alter the partitioning behavior of PLCβ3. LUVs with the same lipid composition were also prepared con- taining G𝛽𝛾 , which is exclusively membrane bound, at a protein to lipid ratio of 1:5 (wt:wt), corresponding to 0.34 mol % , to match the concentration of G𝛽𝛾 in vesicles equilibrated with pla- nar lipid bilayers in the kinetic experiments. At this concentration of G𝛽𝛾 , we observe that PLCβ3 binds to vesicles much more readily than in the absence of G𝛽𝛾 (Fig. 3D). This observation is explicable if, when PLCβ3 partitions onto the membrane surface, it binds to G𝛽𝛾 . Writing the binding reaction on the membrane + G𝛽𝛾 ⇌ PLC 𝛽3 ⋅ G𝛽𝛾 , we have Keq = surface as PLC 𝛽3m [PLC 𝛽3m][G𝛽𝛾] (Fig. 3E). (Note that subscript m indicates [PLC 𝛽3 ⋅ G𝛽𝛾] PLC 𝛽3 on the membrane. Since G𝛽𝛾 only resides on the mem- brane, a subscript is not used for [G𝛽𝛾] and [PLC 𝛽3 ⋅ G𝛽𝛾] ). When equilibrium is reached, the membrane surface will contain a quantity of PLC 𝛽3 in the membrane that is not bound to G𝛽𝛾 , set by Kx and the aqueous solution concentration of PLC 𝛽3 , as well as a quantity of PLC 𝛽3 in the membrane that is bound to G𝛽𝛾 (i.e., PLC 𝛽3 ⋅ G𝛽𝛾) , set by the membrane concentrations of PLC 𝛽3 , G𝛽𝛾 and Keq . Therefore, in the presence of a total quantity of G𝛽𝛾 on the membrane, [Gtot] = [G𝛽𝛾] + [G𝛽𝛾⋅PLC 𝛽3] , the fraction of PLC 𝛽3 on the membrane surface, unbound plus bound to G𝛽𝛾 , is given by (SI Appendix 2) (+G𝛽𝛾) = Fp Kx [L] molar (f (x) + 2 [Gtot] [W ] + [W ] molar) f (x) (Kx [L] molar molar) , [7] 2}1∕2 , where p = Kx with f (x) = p + q + x + {4 q x + (p − q + x) [L]molar [Gtot] , q = Kx [PLCtot]molar , and x = Keq ([W ] + Kx [L]molar) . Because [PLCtot]molar (the molar concentration of PLC 𝛽3 ( PLC 𝛽3w and PLC 𝛽3m ) plus PLC 𝛽3 ⋅ G𝛽𝛾 ), [L]molar and [W ]molar (molar concentrations of lipid and water) and [Gtot] ( mf G𝛽𝛾 plus PLC 𝛽3 ⋅ G𝛽𝛾 in the membrane) are established in the experimental setup, and Kx is determined through partition molar PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 5 of 11 measurements in the absence of G𝛽𝛾 (Fig. 3D), the right-hand side of Eq. 7 contains a single free parameter, Keq , for the binding of PLC 𝛽3 to G𝛽𝛾 on the lipid membrane surface. The red dashed curve in Fig. 3D corresponds to Keq = 0.0090 mol % . It may seem at first surprising that the series of partitioning experiments in the presence of G𝛽𝛾 , with knowledge of Kx for PLC 𝛽3 in the absence of G𝛽𝛾 , uncovers the equilibrium reaction between PLC 𝛽3 and G𝛽𝛾 on the membrane surface. Nevertheless, the binding reaction is discernable by this approach, and the inescapable conclusion is that G𝛽𝛾 concentrates PLC 𝛽3 on the membrane surface (Fig. 3E). The PLC 𝛽3 -concentrating effect of G𝛽𝛾 has obvious implica- tions for interpreting the kinetic data reported above, which show that G𝛽𝛾 increases Vmax by a factor ~65, without affecting KM very much (Fig. 3C). From Eq. 2, Vmax is the asymptotic rate of PIP2 hydrolysis when [ PIP2 ] far exceeds KM . In this limit, the maximum rate of hydrolysis, Vmax , is given by the total membrane concentration of PLC 𝛽3 times kcat , the turnover rate of a PLC 𝛽3 ⋅ PIP2 complex. In the bilayer chamber used for the kinetic experiments, the volume of the aqueous solution is so large compared to the small area of the lipid bilayer that surface binding does not significantly alter [PLC 𝛽3w] . Under this condition, we have Vmax = Kx [PLC 𝛽3w] ( 1 + [Gtot] + Kx [PLC 𝛽3w] ) kcat, Keq [8] where Kx [PLC 𝛽3w] is the membrane concentration of PLC 𝛽3 ( [Gtot] = 0 ) and Kx [PLC 𝛽3w] in ( ) 1 + is the membrane concentration in its pres- the absence of G𝛽𝛾 [Gtot] + Kx [PLC 𝛽3w] Keq ) Keq 1 + is a mul- [Gtot] + Kx [PLC 𝛽3w] ( ence ( [Gtot] > 0 ). Thus, the term tiplier giving the fold-increase in total membrane PLC 𝛽3 concentration due to the presence of G𝛽𝛾 at concentration [Gtot] . When the known quantities are entered for our experimental con- ditions, this factor is ~33. In the kinetic experiments, we observed a 65-fold increase in Vmax in the presence of G𝛽𝛾 . Eq. 8 predicts a 33-fold increase through G𝛽𝛾′s ability to increase the local con- centration of PLC 𝛽3 on the membrane surface. A mere two-fold increase in kcat produced by G𝛽𝛾 binding to PLC 𝛽3 would account for the full enhancement of Vmax in the kinetic experi- ments (Fig. 3C). The important conclusion is that most of the increase in Vmax (within a factor of ~2) is explained by the ability of G𝛽𝛾 to concentrate PLC 𝛽3 on the membrane surface. Indeed, it seems very possible that the ~two-fold shortfall is accountable by the ability of G𝛽𝛾 to orient PLC 𝛽3 , in addition to concentrat- ing it. Using a conventional Michaelis–Menten plot, with the Vmax and KM values derived experimentally, we observe that at concen- trations in our assay, G𝛽𝛾 essentially switches the PLC 𝛽3 enzyme on (Fig. 3F), and this effect is due largely to the ability of G𝛽𝛾 to concentrate PLC 𝛽3 on the membrane surface. In summary, the kinetic studies show that PLC 𝛽3 catalyzes PIP2 hydrolysis with a substrate concentration dependence like that of a Michaelis–Menten enzyme (Fig. 2 C–E). We note that KM corresponds to the mid-range of known PIP2 concentrations in cell membranes (Figs. 2D and 3F) (40, 41). PLC 𝛽3 aqueous- membrane partition studies show that G𝛽𝛾 concentrates PLC 𝛽3 on the membrane surface, enough to account for most of the effect on Vmax (Fig. 3 C and D). To a smaller extent (~two- fold), G𝛽𝛾 augments Vmax through kcat (Fig. 3C). Next, we evaluate the structural underpinnings of these functional properties. Structural Studies of PLCβ3 in Aqueous Solution by Cryo-EM. We next determined the structure of PLCβ3 in aqueous solution using cryo-EM. The structure, consisting of the PLCβ3 catalytic core at 3.6 Å resolution, contained the PH domain, EF hands, X and Y domains, the C-terminal part of the X-Y linker, the C2 domain, and the active site with a Ca2+ ion bound (Fig. 4 A and B and SI Appendix, Fig. S3 and Table S1). The autoinhibitory Hα2′ element in the proximal CTD was also resolved, bound to the catalytic core between the Y domain and the C2 domain, as proposed by Lyon and colleagues (Fig. 4 A and B) (16, 21) but not the distal CTD. We also obtained several low-resolution reconstructions with varying levels of density corresponding to the catalytic core and distal CTD with differing arrangements between the two domains (SI Appendix, Fig. S3F). This observation suggests that the distal CTD is disordered rather than proteolyzed in our final reconstruction and that the two domains are flexible with respect to each other, as previously proposed (19). The catalytic core resolved by cryo-EM is very similar to the crystal structure with a Cα rmsd of 0.6 Å if the Hα2′ helix is excluded (Fig. 4C). We note that, as in the crystal structure, the autoinhibitory X–Y linker occludes the active site (Fig. 4C). We attempted to determine a structure of PLCβ3 in complex with G𝛽𝛾 in solution, in the presence or absence of detergent, without success. Furthermore, we were unable to detect the formation of a complex in solution by size-exclusion chromatography (SI Appendix, Fig. S3G). Structural Studies of PLCβ3 Associated with Liposomes. We next determined the structure of PLCβ3 bound to liposomes consisting of 2DOPE:1POPC:1POPS. PIP2 was omitted from these samples because it would have been degraded by PLCβ3 prior to grid preparation. We obtained a low-resolution reconstruction with the distal CTD associated with the membrane and the catalytic core located away from the membrane surface (Fig. 4C and SI Appendix, Fig. S4 and Table S1). Although the map was low resolution, previously determined structures fit into the density for each domain and all reconstructions showed the same orientation of the protein on the membrane surface (SI Appendix, Fig. S4). The interaction of the distal CTD with the membrane is consistent with previous reports of its involvement in membrane association (3). The position of the catalytic core indicates that significant rearrangements of PLCβ3 with respect to the membrane must be involved in activation because the active site is too far from the membrane to access PIP2 . Activating rearrangements could be mediated by interactions of lipid-anchored G proteins with the PLCβ3 catalytic core. The PLCβ3 · Gβγ Complex on Liposomes Reveals Two Gβγ Binding Sites. We reconstituted G𝛽𝛾 into liposomes consisting of 2DOPE:1POPC:1POPS at a protein to lipid ratio of 1:15 (wt:wt) and incubated the liposomes with purified PLCβ3 prior to grid preparation. We determined the structure of the PLCβ3· Gβγ complex to 3.5 Å and observed two Gβγs bound to the catalytic core of PLCβ3 (Fig. 5 A–C and SI Appendix, Fig. S5 and Table S1). The distal CTD is not resolved in our reconstructions, suggesting that it might adopt many different orientations on the plane of the membrane relative to the catalytic core, in agreement with previous studies showing that heterogeneity in the distal CTD increases upon G𝛽𝛾 binding (42). The catalytic core is very similar to our cryo- EM structure without membranes, with a Cα rmsd of 0.7 Å. Only small rearrangements occur at the G𝛽𝛾 binding sites (SI Appendix, Fig. S6A). Both autoinhibitory elements, the Hα2' and the X–Y linker, are engaged with the catalytic core (Fig. 5C) consistent with previous proposals that G𝛽𝛾 does not play a role in relieving this autoinhibition (15, 16, 21). 6 of 11 https://doi.org/10.1073/pnas.2301121120 pnas.org Fig. 4. Structures of PLCβ3 in solution and on vesicles without G𝛽𝛾 . (A) primary structure arrangement of PLCβ enzymes. Sections are colored by domain as in C. Domains in gray (CTD linker and Distal CTD) are not observed in our structures. pCTD is proximal CTD, of which only the Hα2′ is resolved. (B) Sharpened, masked map of PLCβ3 catalytic core obtained from a sample in solution without membranes or detergent. (C) Structural alignment of the catalytic core of PLCβ3 from the crystal structure of the fulllength protein bound to G𝛼q [colored in gray, PDBID: 4GNK, (19)] and the structure determined using cryoEM without membranes (colored by domain). Cα rmsd is 0.6 Å. Calcium ion from the cryoEM structure is shown as a yellow sphere, and the active site is denoted with an asterisk. The PH domain is pink, the EF hand repeats are blue, the C2 domain is light blue, the Y domain is green, the X domain is teal, and the X–Y linker and the Hα2’ are red. (D) Unsharpened reconstruction of PLCβ3 bound to lipid vesicles containing 2DOPE:1POPC:1POPS. PLCβ3 is colored in yellow and the membrane is colored in gray. One G𝛽𝛾 is bound to the PH domain and the first EF hand, referred to as G𝛽𝛾 1, and the other is bound to the remaining EF hands, referred to as G𝛽𝛾 2 (Fig. 5C and SI Appendix, Fig. S6 A and B). Both interfaces are extensive, with the G𝛽𝛾 1 interface burying ~800 Å2 and involving 34 residues, (16 from PLCβ3 and 18 from G𝛽𝛾 ) and the G𝛽𝛾 2 interface burying ~1,100 Å2 and involving 44 residues (21 from PLCβ3 and 23 from G𝛽𝛾 ) (Fig. 5 D and E and SI Appendix, Fig. S6B and Table S2). The G𝛽𝛾 1 interface is mostly composed of hydrophobic interactions, with three hydrogen bonds (Fig. 5F and SI Appendix, Fig. S6C), whereas the G𝛽𝛾 2 interface is mostly composed of electrostatic interactions, including 10 hydrogen bonds spanning the length of the interface (Fig. 5 G and H). Both interfaces involve the same region of G𝛽𝛾 that interacts with G𝛼 and several residues on Gβ shown to be important for PLCβ activation are involved (43) (Fig. 5 E and F). Specifically, L117 and W99 on Gβ 1 form hydro- phobic interactions with L40, I29, and V89 on PLCβ3 (Fig. 5F and SI Appendix, Table S2) (43). On Gβ 2, W99 forms a hydrogen bond with E294 on PLCβ3, W332 forms an anion-edge interaction with D227, M101 and L117 form hydrophobic inter- actions with P239 and F245 on PLCβ3, and D186 forms a hydro- gen bond with Y240 on PLCβ3 (Fig. 5 G and H and SI Appendix, Table S2) (43). We also determined the structure of the PLCβ3 · Gβγ complex using lipid nanodiscs. We reconstituted G𝛽𝛾 into nanodiscs formed using the MSP2N2 scaffold protein (44) and 2DOPE:1POPC: 1POPS lipids and incubated them with purified PLCβ3 prior to grid preparation. We observed only reconstructions with two Gβγs bound and determined the structure of the complex to 3.3 Å (SI Appendix, Fig. S7). The two G𝛽𝛾s are bound in the same loca- tions as was observed in liposomes with comparable interfaces (SI Appendix, Fig. S6D). A model for this structure aligns well to the model built using the lipid vesicle reconstruction with a Cα rmsd of 0.8 Å for all proteins (SI Appendix, Fig. S6D). These struc- tures suggest that the PLCβ3 · Gβγ complex depends on a mem- brane environment as we were unable to form a stable complex in solution with or without detergent, which highlights the importance of the membrane in Gβγ-dependent activation of PLCβ3. Gβγ Mediates Membrane Association and Orientation of the PLCβ3 Catalytic Core. Unmasked refinement of our final subset of particles from the liposome structure yielded a 3.8 Å reconstruction showing the PLCβ3 · Gβγ assembly and density from the membrane (Figs. 5B and 6A). The two Gβγs and the region of PLCβ3 between them are closely associated with the membrane and the remainder of the catalytic core, including the active site, tilts away from the membrane (Fig. 6A). Despite the tilting, the structure reveals significant rearrangement of the catalytic core with respect to the membrane compared to its position in the absence of G𝛽𝛾 , where it was separated from the membrane surface by a larger distance (Fig. 6A). Additional 2D and 3D classification without alignment revealed heterogeneity in the position of the PLCβ3 · Gβγ assembly with respect to the membrane (Fig. 6 B–D). 2D classes show large variation in the orientation of the catalytic core with respect to the membrane surface, with some classes showing the entire cat- alytic core engaged with the membrane (Fig. 6B). The 2D classes also reveal differences in membrane curvature originating from differences in liposome size, which do not seem to be correlated with the degree of membrane tilting (Fig. 6B). 3D classification revealed four reconstructions capturing different degrees of tilting of the catalytic core ranging from ~26° to ~36° (Fig. 6D). We note that in a locally planar membrane, as opposed to a curved vesicle membrane, the active site would be nearer the membrane surface in all classes, but the variability in orientation would presumably still exist. The protein components of these reconstructions are like in the original reconstruction, with no internal conforma- tional changes, indicating that the whole complex tilts on the membrane as a rigid body. The lack of conformational changes observed upon G𝛽𝛾 bind- ing and the catalytic core membrane association are consistent with our functional studies showing that activation by G𝛽𝛾 is largely mediated by increasing membrane partitioning. Our struc- tures suggest that the configuration of the two G𝛽𝛾 binding sites maintains the catalytic core at the membrane and increases the probability of productive engagement with PIP2 , potentially mediated by orientation of the catalytic core observed in our PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 7 of 11 A D F B C E G H Fig. 5. PLCβ3 · Gβγ complex on lipid vesicles and G𝛽𝛾 interfaces. (A) Example micrograph showing lipid vesicles with protein complexes. (B) Unsharpened map from nonuniform refinement showing the PLCβ3 · Gβγ complex on the vesicle surface. Both the inner and outer leaflets of the vesicle are shown. (C) Sharpened, masked map of the catalytic core of PLCβ3 in complex with two Gβγs on lipid vesicles containing 2DOPE:1POPC:1POPS. PLCβ3 is yellow, G𝛽 1 is dark teal, G𝛾 1 is light purple, G𝛽 2 is light blue, and G𝛾 2 is light pink. The autoinhibitory elements Hα2′ and the X–Y linker are colored in red. Coloring is the same throughout. DE: Surface representation of the PLCβGβγ 1 (D) or PLCβGβγ 2 (E) interfaces peeled apart to show extensive interactions. Residues on PLCβ3 that interact with G𝛽𝛾 1 or 2 are colored according to the corresponding G𝛽 coloring and residues on the G𝛽 s that interact with PLCβ3 are colored in yellow. Interface residues were determined using the ChimeraX interface feature using a buried surface area cutoff of 15 Å2. (F and G) Interactions of residues on G𝛽 that have been shown to be important for PLCβ activation with residues from PLCβ3 in the PLCβGβγ 1 interface (F) or the PLCβGβγ 2 interface (G) (43). All labeled interactions are < ~4 Å. Interacting residues are shown as sticks and colored by heteroatom. Interactions are denoted by black dashed lines. (H) Extensive hydrogen bond network in the PLCβGβγ 2 interface including both sidechain and backbone interactions. All labeled hydrogen bonds are between ~2.3 and ~3.8 Å. Interacting residues are shown as sticks and colored by heteroatom. Interactions are denoted by black dashed lines. reconstructions. Taken together, our kinetic, binding, and struc- tural studies lead us to conclude that G𝛽𝛾 activates PLC𝛽 mainly by bringing it to the membrane and orienting the catalytic core so that the active site can access the PIP2-containing surface (Fig. 7). Discussion This study aims to understand how a G protein, G𝛽𝛾 , activates the PLC 𝛽3 phospholipase enzyme. We developed and applied three new technical approaches to study this process. First, because kinetic analyses of PLC 𝛽 enzymes historically have been limited to relatively slow radioactivity-based or semiquantitative fluores- cence assays, we have developed a new higher resolution assay using a modified, calibrated PIP2-dependent ion channel to pro- vide a direct read out of membrane PIP2 concentration as a func- tion of time. This assay is employed in a reconstituted system in which all components are defined with respect to composition and concentration. Second, we have used a membrane-water par- tition assay to study a surface equilibrium reaction between two proteins ( PLC 𝛽3 and G𝛽𝛾 ) on membranes. Third, we have deter- mined structures of a protein complex ( PLC 𝛽3 and G𝛽𝛾 ) assem- bled on the surface of pure lipid vesicles. We also determined the structures using lipid nanodiscs; however, the lipid vesicles per- mitted structural analysis of the enzyme-G protein complex on lipid surfaces unperturbed by the scaffold proteins required to make nanodiscs. The membrane in our nanodisc reconstructions is poorly resolved and the complex appears to be associated at nonphysiological orientations; therefore, we cannot gain any infor- mation regarding the positioning of the complex on the membrane from those reconstructions. We list our essential findings. 1) PLC 𝛽3 catalyzes PIP2 hydrol- ysis in accordance with Michaelis–Menten enzyme kinetics. 2) G𝛽𝛾 modifies Vmax , leaving KM essentially unchanged. Under our experimental conditions, Vmax increases ~65-fold. 3) G𝛽𝛾 increases membrane partitioning of PLC 𝛽3 , an effect accountable through equilibrium complex formation between G𝛽𝛾 and PLC 𝛽3 on the membrane surface. Under our experimental conditions, partition- ing increases the membrane concentration of PLC 𝛽3 ~33-fold. 4) The G𝛽𝛾 -mediated increase in PLC 𝛽3 partitioning can account for most of the increase in Vmax , with a smaller, ~two-fold, effect on kcat . Thus, G𝛽𝛾 regulates PLC 𝛽3 mainly by concentrating it on the membrane. 5) Two G𝛽𝛾 proteins assemble to form a com- plex with PLC 𝛽3 on vesicle surfaces. One G𝛽𝛾 binds to the PH domain and one EF hand of PLC 𝛽3 , while the other binds to the remaining EF hands. Both G𝛽𝛾 orient their covalent lipid groups toward the membrane so that the PLC 𝛽3 catalytic core is firmly anchored on the membrane surface. 6) The PLC 𝛽3 ⋅ G𝛽𝛾 assem- bly holds the PLC 𝛽3 catalytic core with its active site, as if on the end of a stylus, poised to sample the membrane surface. Assemblies on lipid vesicles reveal multiple orientations of the catalytic core with respect to the surface. 8 of 11 https://doi.org/10.1073/pnas.2301121120 pnas.org Fig. 6. Tilting of the PLCβ3 · Gβγ complex with respect to the membrane. (A) Consensus unmasked refinement with density for the PLCβ3 · Gβγ complex and the membrane colored by protein. The membrane is gray, PLCβ3 is yellow, G𝛽 1 is dark cyan, and G𝛾 1 is light purple. The X–Y linker is colored red to highlight the active site. (B) 2D class averages of the final subset of particles determined without alignment showing side views of the complex on the membrane. Different membrane curvatures and positions of the complex with respect to the membrane are demonstrated. (C) 2D projections of 3D classes of the PLCβ3 · Gβγ complex on the membrane. (D) 3D reconstructions of four 3D classes with different positions of the complex on the membrane arranged by degree of tilting with the most tilted on the left and least tilted on the right. We described the formation of a complex between PLCβ and G𝛽𝛾 as a two-step process: first, partitioning of PLCβ from aque- ous solution into the membrane, and second, binding to G𝛽𝛾 on the membrane surface. We explicitly consider two steps rather than one in which PLCβ binds directly to G𝛽𝛾 for the following reasons. We measured partitioning of PLCβ into membranes with- out G𝛽𝛾 and measured the corresponding catalysis of PIP2 in the absence of G𝛽𝛾 . Thus, we know that PLCβ partitions onto the membrane surface without G𝛽𝛾 . Furthermore, we find that PLCβ and G𝛽𝛾 do not form a complex in the absence of a membrane, Fig. 7. G𝛽𝛾 activates PLC𝛽 by increasing its concentration at the membrane and orienting the catalytic core to engage PIP2 . Upon activation of a G𝛼icoupled receptor, GTP is exchanged for GDP in the G𝛼i subunit and free G𝛽𝛾 is released to bind PLCβ, which increases the concentration of PLCβ at the membrane and orients the active site for catalysis. The kcat is limited by the X–Y linker (shown in red), which occludes the active site and is only transiently displaced from the active site to allow catalysis. The distal CTD of PLCβ was omitted for clarity. PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 9 of 11 neither as evaluated by size exclusion chromatography (SI Appendix, Fig. S3G) nor on cryo-EM grids. It was also shown previously that G𝛽𝛾 does not activate PLCβ in the absence of membranes (17). Taken together, this set of findings support the conclusion that PLCβ partitioning is a required first step in the two-step process of PLC 𝛽3 ⋅ G𝛽𝛾 complex formation on membranes. We hypoth- esize that partitioning orients PLCβ with respect to G𝛽𝛾 , defines a local surface concentration, and thus permits a binding equilib- rium process that occurs in 2 dimensions, rather than in a three-dimensional aqueous phase. We modeled the second step, the equilibrium reaction between PLC 𝛽 and G𝛽𝛾 on the membrane surface, as bimolecular (1:1 sto- ichiometry) characterized by a single Keq . In our structural analysis, however, we discovered two binding sites for G𝛽𝛾 on PLC 𝛽3 . Additional binding data, using multiple concentrations of G𝛽𝛾, for example, might reveal two distinct binding constants and whether they interact with each other (i.e., behave cooperatively). Such a finding would be important because multiple binding sites could shape the PLC 𝛽3 activity response to GPCR stimulation. But for purposes of the present study, the binding model treating a single site is sufficient. This is because using a single site model when two sites exist introduces an uncertainty in how PLC 𝛽3 is distributed over G𝛽𝛾 , not how much PLC 𝛽3 is present in the membrane. The kinetics depend on how much PLC 𝛽3 is present, and this we have measured directly with experiment. The conclusion that G𝛽𝛾 concentrates PLC 𝛽3 on the mem- brane in our assay is unequivocal. To what extent do these con- clusions apply to cell membranes? From Eq. 8, we saw that the increase in membrane PLC 𝛽3 concentration due to the fraction bound to G𝛽𝛾 is proportional to total G𝛽𝛾 concentration, [Gtot] . In our assay, [Gtot] is 0.34 mol % , which corresponds to ~5,000 G𝛽𝛾∕𝜇m2 . In cells, we have previously estimated the concentra- tion of G𝛽𝛾 near GIRK2 channels in dopamine neurons during GABAB receptor activation at ~1,200 G𝛽𝛾∕𝜇m2 (32). Applying Eq. 8, this would produce an ~nine-fold increase in the membrane concentration of PLC 𝛽3 . This is an estimate with certain unknowns, especially the cytoplasmic concentration of PLC 𝛽3 ( [PLC 𝛽3w] ), but the result suggests that the conditions of our in vitro assay are applicable to cell membranes. Moreover, both G𝛽𝛾 and G𝛼q have been shown to increase membrane association of PLC 𝛽s in cells, consistent with our results (29). We note that our demonstration that G𝛽𝛾 increases membrane association of PLC 𝛽3 directly contradicts many previous biochem- ical studies and the current consensus in the field that G proteins do not increase the local concentration of PLCβs in the membrane (3, 4, 22–24, 26–28). We suspect that the use of detergent solu- bilized G𝛽𝛾 in past studies may have interfered with the control of its concentration on the membrane (22–24). While our results and mechanism contradict the notion that G proteins do not concentrate PLC 𝛽s on the membrane, they are consistent with many previous observations, some we list here. As stated above, studies with cells have led to the conclusion that G𝛽𝛾 and G𝛼q increase membrane association of PLC 𝛽s (29). The lipid anchor is required for the activation of PLCβs by the small GTPases and G𝛽𝛾 , and G proteins do not activate PLCβs in the absence of a membrane environment (9–11, 13, 15, 17, 29). The binding of Rac1 or G𝛼q do not induce conformational changes around the active site, suggesting that activation is not mediated by obvious allosteric changes (9, 18, 19). Likewise, we observe no change in the PLC 𝛽3 active site conformation when G𝛽𝛾 is bound, only that G𝛽𝛾 recruits PLC 𝛽3 to the membrane and ori- ents its active site. Several properties of the G𝛽𝛾 binding sites on PLC 𝛽3 offer explanations of past observations. First, it has been shown that G𝛽𝛾 and G𝛼q can activate PLC 𝛽3 simultaneously (36, 37, 45–49). We find here that the G𝛽𝛾 sites do not occlude the G𝛼q binding site (18, 19), and therefore both G proteins can in principle bind to PLC 𝛽3 at the same time and activate PLC 𝛽3 (3, 36, 37, 45). Second, several amino acids on G𝛽𝛾 that contact PLCβ3 in the structure were previously shown to play a role in binding to G𝛼 , PLCβ, and other effectors (Fig. 5 C and D) (43). Third, the PH domain was shown to play a role in G𝛽𝛾 binding and activation; however, based on our structures, G𝛽𝛾 binding does not require or induce rearrangement of the catalytic core as was previously proposed (50, 51). Fourth, Rac1 was also shown to bind to the PH domain of PLCβ2 (SI Appendix, Fig. S6D), and Rac1-activated PLCβ was shown to be additionally activated by G𝛽𝛾 , leading to a proposal that the two binding sites did not overlap (9, 10). Our structures show that Rac1 and G𝛽𝛾 do indeed share an interface within the PH domain (SI Appendix, Fig. S6D); however, the sec- ond G𝛽𝛾 binding site can explain the dual activation (9, 10). An intriguing aspect of PLC 𝛽 enzymes is that all wild-type structures show that the active site is occluded by the inhibitory X–Y linker. This includes complexes with G𝛼q, Rac1 and, now, G𝛽𝛾 (9, 16, 18, 19). It has been proposed that lipids are required to remove the X–Y linker to achieve catalysis (3, 16, 17). This must be true to some extent because unless the linker is dis- placed, even if only rarely, catalysis cannot occur. From our data, we put forth an alternative proposal that the active site is pre- dominantly autoinhibited, accounting for a small kcat , even in the presence of lipids. Consequently, in the absence of GPCR stimulation, the baseline partitioning of PLC 𝛽 enzyme from the cytoplasm to the membrane, determined by Kx and the cyto- plasmic concentration of PLC 𝛽 , will produce very little PIP2 hydrolysis. Only upon GPCR stimulation, when a large quantity of PLC 𝛽 partitions into the membrane, determined by Keq and the G𝛽𝛾 concentration generated by GPCR stimulation, is there enough PLC 𝛽 enzyme in the membrane, even though kcat remains low, to catalyze PIP2 hydrolysis. In other words, a small kcat combined with an ability to enact large changes in membrane enzyme concentration upon GPCR stimulation permits a strong signal when the system is stimulated and a minimal baseline when it is not. Materials and Methods Protein Expression, Purification, and Reconstitution. All proteins were purified according to previously established protocols using affinity chroma- tography and size exclusion chromatography. Detailed methods are described in SI Appendix, Materials and Methods: Protein Expression and Purification and Protein Reconstitution. PLCβ3 Functional Assay. PLCβ activity was measured using a planar lipid bilayer setup and a PIP2-dependent ion channel to report PIP2 concentration in the mem- brane over time. Detailed methods are described in SI Appendix, Materials and Methods: Bilayer Experiments and Analysis. Membrane Partitioning Experiments. Fluorescently labeled PLCβ3 was mixed with LUVs and pelleted. Protein in the pellet and supernatant were quantified using fluorescence. Detailed methods are described in SI Appendix, Materials and Methods: PLCβ3 Vesicle Partition Experiments. PLCβ3 Structure Determination. PLCβ3 was mixed with liposomes with or without Gβγ prior to sample vitrification. Cryo-EM data were collected using a Titan Krios with a Gatan K3 direct electron detector according to the parameters in SI Appendix, Table S1 and analyzed according to the procedures outlined in SI Appendix, Figs. S3–S5 and S7. Atomic models from previously determined structures were fit into our density maps, refined using PHENIX real-space refine (52), and manually adjusted. Detailed methods are described in SI Appendix, 10 of 11 https://doi.org/10.1073/pnas.2301121120 pnas.org Materials and Methods: Cryo-EM Sample Preparation and Data Collection, Cryo-EM Data Processing, and Model Building and Validation. Data, Materials, and Software Availability. Cryo-EM maps and atomic mod- els for all structures described in this work have been deposited to the Electron Microscopy Data Bank (EMDB) and the Protein Data Bank (PDB), respectively. Accession codes are as follows: PLCβ3 in solution-8EMV and EMD-28266, PLCβ3 in complex with Gβγ on vesicles-8EMW and EMD-28267, and PLCβ3 in complex with Gβγ on nanodiscs-8EMX and EMD-28268. ACKNOWLEDGMENTS. We thank Chen Zhao for developing and characterizing the ALFA nanobody-mediated tethering of G𝛽𝛾 to GIRK and for insightful discus- sions. We thank Venkata S. Mandala for assistance with protein reconstitution and NMR experiments. We thank Christoph A. Haselwandter for insightful discussion and comments on the manuscript. We thank Yi Chun Hsiung for assistance with tissue culture. We thank members of the MacKinnon lab, Jue Chen and members of her lab for helpful discussions. This work was supported by National Institute of General Medical Sciences (NIHF32GM142137 to M.E.F.). R.M. is an investigator in the Howard Hughes Medical Institute. We thank Rui Yan and Zhiheng Yu at the HHMI Janelia Cryo-EM Facility for help in microscope operation and data collection. We thank Mark Ebrahim, Johanna Sotiris, and Honkit Ng at the Evelyn Gruss Lipper Cryo-EM Resource Center of Rockefeller University for assistance with cryo-EM data collection. Some of this work was performed at the Simons Electron Microscopy Center and National Resource for Automated Molecular Microscopy located at the New York Structural Biology Center, supported by grants from the Simons Foundation (SF349247), NYSTAR (Empire State Development Division of Science, Technology and Innovation), and the NIH National Institute of General Medical Sciences (GM103310) with additional support from Agouron Institute (F00316) and NIH (OD019994). 2. 1. P. Kemp, G. Hübscher, J. Hawthorne, Phosphoinositides. 3. Enzymic hydrolysis of inositol-containing phospholipids. Biochem. J. 79, 193–200 (1961). R. Rodnight, Cerebral diphosphoinositide breakdown: Activation, complexity and distribution in animal (mainly nervous) tissues. Biochem. J. 63, 223–231 (1956). A. M. Lyon, J. J. G. Tesmer, Structural insights into phospholipase C-β function. Mol. Pharmacol. 84, 488–500 (2013). 4. G. Kadamur, E. M. Ross, Mammalian phospholipase C. Annu. Rev. Physiol. 75, 127–154 (2013). 5. M. J. Berridge, Inositol trisphosphate and diacylglycerol: Two interacting second messengers. Annu. 3. 6. 7. 8. Rev. Biochem. 56, 159–193 (1987). J. A. Poveda et al., Lipid modulation of ion channels through specific binding sites. Biochim. et Biophys. Acta 1838, 1560–1567 (2014). C. V. Robinson, T. Rohacs, S. B. Hansen, Tools for understanding nanoscale lipid regulation of ion channels. Trends Biochem. Sci. 44, 795–806 (2019). S. B. Hansen, Lipid agonism: The PIP2 paradigm of ligand-gated ion channels. Biochim. et Biophys. Acta 1851, 620–628 (2015). 9. M. R. Jezyk et al., Crystal structure of Rac1 bound to its effector phospholipase C-β2. Nat. Struct. Mol. Biol. 13, 1135–1140 (2006). 10. D. Illenberger et al., Rac2 regulation of phospholipase C-β2 activity and mode of membrane interactions in intact cells. J. Biol. Chem. 278, 8645–8652 (2003). 11. D. Illenberger et al., Stimulation of phospholipase C-β2 by the Rho GTPases Cdc42Hs and Rac1. EMBO J. 17, 6241–6249 (1998). 12. M. Camps et al., Isozyme-selective stimulation of phospholipase C-β2 by G protein βγ-subunits. Nature 360, 684–686 (1992). 13. A. Katz, D. Wu, M. I. Simon, Subunits βγ of heterotrimeric G protein activate β2 isoform of phospholipase C. Nature 360, 686–689 (1992). 14. A. V. Smrcka, P. C. Sternweis, Regulation of purified subtypes of phosphatidylinositol-specific phospholipase C β by G protein α and βγ subunits. J. Biol. Chem. 268, 9667–9674 (1993). 15. S. N. Hicks et al., General and versatile autoinhibition of PLC isozymes. Mol. Cell 31, 383–394 (2008). 16. A. M. Lyon, V. G. Taylor, J. J. G. Tesmer, Strike a pose: Gαq complexes at the membrane. Trends Pharmacol. Sci. 35, 23–30 (2014). 17. T. H. Charpentier et al., Membrane-induced allosteric control of phospholipase C-β isozymes. J. Biol. Chem. 289, 29545–29557 (2014). 18. G. L. Waldo et al., Kinetic scaffolding mediated by a phospholipase C–b and Gq signaling complex. Science 330, 974–980 (2010). 19. A. M. Lyon, S. Dutta, C. A. Boguth, G. Skiniotis, J. J. G. Tesmer, Full-length Gαq-phospholipase C-β3 structure reveals interfaces of the C-terminal coiled-coil domain. Nat. Struct. Mol. Biol. 20, 355–362 (2013). 20. A. M. Lyon, J. A. Begley, T. D. Manett, J. J. G. Tesmer, Molecular mechanisms of phospholipase C β3 autoinhibition. Structure 22, 1844–1854 (2014). 21. A. M. Lyon et al., An autoinhibitory helix in the C-terminal region of phospholipase C-β mediates Gαq activation. Nat. Struct. Mol. Biol. 18, 999–1005 (2011). 22. J. M. Jenco, K. P. Becker, A. J. Morris, Membrane-binding properties of phospholipase C-β1 and phospholipase C-β: Role of the C-terminus and effects of polyphosphoinositides, G-proteins and Ca2+. Biochem. J. 327, 431–437 (1997). 23. V. Romoser, R. Ball, A. V. Smrcka, Phospholipase C β2 association with phospholipid interfaces assessed by fluorescence resonance energy transfer. J. Biol. Chem. 271, 25071–25078 (1996). 24. L. W. Runnels, J. Jenco, A. Morris, S. Scarlata, Membrane binding of phospholipases C-β1 and C-β2 is independent of phosphatidylinositol 4,5-bisphosphate and the α and βγ subunits of G proteins. Biochemistry 35, 16824–16832 (1996). 25. S. Scarlata, Regulation of the lateral association of phospholipase Cβ2 and G protein subunits by lipid rafts. Biochemistry 41, 7092–7099 (2002). 26. E. E. Garland-Kuntz et al., Direct observation of conformational dynamics of the PH domain in phospholipases Cε and β may contribute to subfamily-specific roles in regulation. J. Biol. Chem. 293, 17477–17490 (2018). 27. K. Muralidharan, M. M. Van Camp, A. M. Lyon, Structure and regulation of phospholipase Cβ and ε at the membrane. Chem. Phys. Lipids 235, 105050 (2021). 28. B. N. Hudson, R. E. Jessup, K. K. Prahalad, A. M. Lyon, Gαq and the phospholipase Cβ3 X-Y linker regulate adsorption and activity on compressed lipid monolayers. Biochemistry 58, 3454–3467 (2019). 29. O. Gutman, C. Walliser, T. Piechulek, P. Gierschik, Y. I. Henis, Differential regulation of phospholipase C-β2 activity and membrane interaction by Gαq, Gβ1γ 2, and Rac2. J. Biol. Chem. 285, 3905–3915 (2010). 30. C. Miller, Integral membrane channels: Studies in model membranes. Physiol. Rev. 63, 1209–1242 (1983). 31. W. Wang, M. R. Whorton, R. MacKinnon, Quantitative analysis of mammalian GIRK2 channel regulation by G proteins, the signaling lipid PIP2 and Na+ in a reconstituted system. eLife 3, e03671 (2014). 32. W. Wang, K. K. Touhara, K. Weir, B. P. Bean, R. MacKinnon, Cooperative regulation by G proteins and Na+ of neuronal GIRK2 K+ channels. eLife 5, 1–15 (2016). 33. C. G. Nichols, S.-J. Lee, Polyamines and potassium channels: A 25-year romance. J. Biol. Chem. 293, 18779–18788 (2018). 34. H. Götzke et al., The ALFA-tag is a highly versatile tool for nanobody-based bioscience applications. Nat. Commun. 10, 1–12 (2019). 35. C. Zhao, R. MacKinnon, Structural and functional analyses of a GPCR-inhibited ion channel TRPM3. Neuron 1 (2022). 36. F. Philip, G. Kadamur, R. G. Silos, J. Woodson, E. M. Ross, Synergistic activation of phospholipase C-β3 by Gαq and Gβγ describes a simple two-state coincidence detector. Curr. Biol. 20, 1327–1335 (2010). 37. R. A. Rebres et al., Synergistic Ca2+ responses by Gαi- and Gαq-coupled G-protein-coupled receptors require a single PLCβ isoform that is sensitive to both Gβγ and Gαq. J. Biol. Chem. 286, 942–951 (2011). 38. M. Goličnik, On the Lambert W function and its utility in biochemical kinetics. Biochem. Eng. J. 63, 116–123 (2012). 39. S. H. White, W. C. Wimley, A. S. Ladokhin, K. Hristova, "[4] Protein folding in membranes: Determining energetics of peptide-bilayer interactions" in Methods in Enzymology (Academic Press, 1998), vol. 295, pp. 62–87. 40. S. McLaughlin, J. Wang, A. Gambhir, D. Murray, PIP2 and proteins: Interactions, organization, and information flow. Annu. Rev. Biophys. Biomol. Struct. 31, 151–175 (2002). 41. K. Mandal, Review of PIP2 in cellular dignaling, functions and diseases. IJMS 21, 8342 (2020). 42. I. J. Fisher, M. L. Jenkins, G. G. Tall, J. E. Burke, A. V. Smrcka, Activation of phospholipase C β by Gβγ and Gαq involves C-terminal rearrangement to release autoinhibition. Structure 28, 1–10 (2020). 43. C. E. Ford et al., Molecular basis for interactions of G protein βγ subunits with effectors. Science 280, 1271–1274 (1998). 44. Y. V. Grinkova, I. G. Denisov, S. G. Sligar, Engineering extended membrane scaffold proteins for self- assembly of soluble nanoscale lipid bilayers. Protein Eng. Des. Sel. 23, 843–848 (2010). 45. T. I. A. Roach et al., Signaling and cross-talk by C5a and UDP in macrophages selectively use PLCβ3 to regulate intracellular free calcium. J. Biol. Chem. 283, 17351–17361 (2008). 46. R. C. Carroll, A. D. Morielli, E. G. Peralta, Coincidence detection at the level of phospholipase C activation mediated by the m4 muscarinic acetylcholine receptor. Curr. Biol. 5, 536–544 (1995). 47. F. Okajima, K. Sato, K. Sho, Y. Kondo, Stimulation of adenosine receptor enhances α1 -adrenergic receptor-mediated activation of phospholipase C and Ca2+ mobilization in a pertussis toxin- sensitive manner in FRTL-5 thyroid cells. FEBS Lett. 248, 145–149 (1989). 48. B. H. Shah et al., Co-activation of Gi and Gq proteins exerts synergistic effect on human platelet aggregation through activation of phospholipase C and Ca 2+ signalling pathways. Exp. Mol. Med. 31, 42–46 (2009). 49. J. A. Ware, M. Smith, E. W. Salzman, Synergism of platelet-aggregating agents. Role of elevation of cytoplasmic calcium. J. Clin. Invest. 80, 267–271 (1987). 50. G. Kadamur, E. M. Ross, Intrinsic pleckstrin homology (PH) domain motion in phospholipase C-β exposes a Gβγ protein binding site. J. Biol. Chem. 291, 11394–11406 (2016). 51. G. Drin, D. Douguet, S. Scarlata, The Pleckstrin homology domain of phospholipase Cβ transmits enzymatic activation through modulation of the membrane-domain orientation. Biochemistry 45, 5712–5724 (2006). 52. P. V. Afonine, J. J. Headd, T. C. Terwilliger, P. D. Adams, PHENIX news. Comput. Crystallogr. Newsletter 4, 43–44 (2013). PNAS 2023 Vol. 120 No. 20 e2301121120 https://doi.org/10.1073/pnas.2301121120 11 of 11
10.1093_beheco_arad033	Behavioral Ecology The official journal of the ISBE International Society for Behavioral Ecology Behavioral Ecology (2023), 34(4), 642–652. https://doi.org/10.1093/beheco/arad033 Original article Conditional indirect genetic effects of caregivers on brood in the clonal raider ant Patrick K. Piekarski,a, Stephany Valdés-Rodríguez,a,b and Daniel J.C. Kronauera,b aLaboratory of Social Evolution and Behavior, The Rockefeller University, New York, NY 10065, USA and bHoward Hughes Medical Institute, New York, NY 10065, USA Received 28 November 2022; revised 22 March 2023; editorial decision 5 April 2023; accepted 15 April 2023 Caregivers shape the rearing environment of their young. Consequently, offspring traits are influenced by the genes of their care- givers via indirect genetic effects (IGEs). However, the extent to which IGEs are modulated by environmental factors, other than the genotype of social partners (i.e., intergenomic epistasis), remains an open question. Here we investigate how brood are influenced by the genotype of their caregivers in the clonal raider ant, Ooceraea biroi, a species in which the genotype, age and number of both caregivers and brood can be experimentally controlled. First, we used four clonal lines to establish colonies that differed only in the genotype of caregivers and measured effects on foraging activity, as well as IGEs on brood phenotypes. In a second experiment, we tested whether these IGEs are conditional on the age and number of caregivers. We found that caregiver genotype affected the feeding and foraging activity of colonies, and influenced the rate of development, survival, body size, and caste fate of brood. Caregiver genotype interacted with other factors to influence the rate of development and survival of brood, demonstrating that IGEs can be conditional. Thus, we provide an empirical example of phenotypes being influenced by IGE-by-environment interactions be- yond intergenomic epistasis, highlighting that IGEs of caregivers/parents are alterable by factors other than their brood’s/offspring’s genotype. Key words: behavior, gene–environment interactions, maternal effects, parental care, parent–offspring interactions, phenotypic plasticity. INTRODUCTION In social species, the phenotypes of individuals are influenced by the genotypes of their social partners through indirect genetic ef- fects (IGEs), where genes expressed in one individual influence the phenotype of another individual by shaping the environ- ment experienced by that individual. A classic example of IGEs is found in mammalian maternal care, where the genotype of the mother influences maternal performance, and consequently off- spring growth and survival (Peripato and Cheverud 2002). Since the social environment of a focal individual is influenced by the genotype of its social partners, traits influenced by IGEs have a heritable environmental component (Moore et al. 1997; Wolf 2003). Thus, IGEs can influence evolutionary responses to selec- tion, resulting in either a slowing or acceleration of phenotypic evolution (Moore et al. 1997; Wolf et al. 1998; McGlothlin et al. 2010). A complete picture of how any trait evolves requires know- ledge of the extent to which it is shaped by both direct genetic effects (DGEs) and IGEs. Address correspondence to P.K. Piekarski. E-mail: [email protected]; D.J.C. Kronauer. E-mail: [email protected]. The relationship between a trait’s expression and the envi- ronment (i.e., the reaction norm) can differ across genotypes. Genotypes may produce the same phenotype in one environ- ment but different phenotypes in another environment, and such conditional DGEs are indicative of genotype-specific re- action norms (i.e., gene-by-environment interactions). For ex- ample, in Drosophila pseudoobscura, genotypes with the fastest development at one temperature do not differ from other geno- types at other temperatures (Gupta and Lewontin 1982). If the phenomenon of gene-by-environment interactions extends to IGEs, then conditional IGEs should be ubiquitous in social spe- cies due to variability in how genotypes of social partners re- spond to environmental variation. Genotype-by-genotype (GxG) epistasis, wherein the magnitude and direction of a genetic ef- fect on an individual’s phenotype depends upon the genotypes of social partners, may be considered a specific type of IGE- by-environment interaction and has been documented across a broad range of taxa for a variety of traits (Linksvayer 2007; Linksvayer et al. 2009; Buttery et al. 2010; Marie-Orleach et al. 2017; Rode et al. 2017; Culumber et al. 2018; Jaffe et al. 2020; Rebar et al. 2020; Walsh et al. 2022). However, reports of IGE- by-environment interactions beyond GxG epistasis, such as IGEs that depend on abiotic environmental factors or physiology, © The Author(s) 2023. Published by Oxford University Press on behalf of the International Society for Behavioral Ecology. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Piekarski et al. • Context alters indirect genetic effects 643 remain sparse (Bailey and Desjonquères 2021; Fisher et al. 2021). A rare example comes from Drosophila, where the direction and magnitude of IGEs that males have on female locomotion dif- fers across male genotypes, but the difference across genotypes depends on the presence/absence of ethanol in the medium (i.e., male genotype interacts with the abiotic environment to influ- ence female locomotion) (Signor et al. 2017). In social species, a trait’s evolution is necessarily influenced by IGEs, but also by how these IGEs change in direction and magnitude across dif- ferent environments. Thus, investigations into how IGEs change across environments are needed. Where parental care has evolved, the phenotype and fitness of off- spring is heavily influenced by the genotype of their caregivers (Hunt and Simmons 2002; Mcadam et al. 2002; Peripato and Cheverud 2002; Wilson et al. 2005). Similarly, in eusocial insects where sib- ling workers cooperatively care for brood as allomothers, the geno- types of caregivers will have an indirect effect on the phenotype of brood by shaping the rearing environment (Linksvayer and Wade 2005). In honeybees, cross-foster experiments using low and high pollen-hoarding strains show that the larval rearing environment (i.e., whether being reared by low- or high-pollen hoarding workers) had an influence on the body mass, sucrose responsiveness and ovariole number of the resulting adults, implying IGEs (Pankiw et al. 2002; Linksvayer et al. 2009). IGEs of workers on brood have also been shown in Temnothorax ants, where the body mass of resulting adults is influenced by the species of worker rearing them (Linksvayer 2006, 2007). Given the strong dependence of brood on their caregivers in eusocial insects, caregiver–brood interactions offer a promising av- enue to explore if and how IGEs are tuned by context. Determining the environmental factors that shape caregiver IGEs on brood phenotypes is not straightforward in a typical eusocial insect, where age demographics, genetic heterogeneity among workers and brood, heterogeneity in sex and develop- mental stages of larvae, and variability in colony size confound the IGEs between caregivers and brood. All these factors can be experimentally controlled in the clonal raider ant, Ooceraea biroi, making this species a tractable study system to test what environ- mental factors modify the IGEs that caregivers have on the brood. Colonies of O. biroi are queen-less but contain two sub-castes of workers that differ in morphology and physiology (Ravary and Jaisson 2004). Intercastes have four to six ovarioles and tend to have vestigial eyes and a larger body size compared to regular workers, which have two ovarioles and lack vestigial eyes. O. biroi reproduces asexually, with all colony members being near- identical clones and (almost always) female (Kronauer et al. 2012). In the presence of larvae, workers forage and their ovaries are inactive (Ravary and Jaisson 2002; Ravary et al. 2006; Chandra et al. 2018). Coincident with the end of larval development and the onset of metamorphosis, workers stop foraging and reactivate their ovaries. Thus, colonies alternate between a reproductive phase, in which larvae are absent and adults synchronously lay eggs, and a brood care phase, in which adults tend to larvae and forage for food. This ultimately produces age-matched cohorts of brood that all eclose within a few days of each other (Ravary and Jaisson 2002, 2004), permitting experimental control over the age of both caregiving adults and brood. For example, age-matched caregivers of varying genotypes can all be supplied brood of the same age and genotype, allowing experimenters to investigate how genetic variability among caregivers influences brood growth and development (i.e., IGEs). A previous cross-fostering experiment using two different clonal lineages (i.e., lines) showed a significant effect of caregiver geno- type, brood genotype, and a GxG interaction on the proportion of brood that developed into intercastes (Teseo et al. 2014). A more recent study identified a larger range of IGEs and DGEs on both caregiver and brood phenotypes by conducting developmental, morphological, and behavioral tracking analyses in a full-factorial cross-foster experiment using four lines (Jud et al. 2022). Both care- giver and brood genotype influenced length of larval development and the proportion of time that caregivers spent in the nest, while a GxG interaction influenced the body size and intercaste propor- tions of brood (Jud et al. 2022). However, to date, it has not been formally tested if any of the previously reported IGEs of workers on brood are modified by environmental factors beyond GxG epis- tasis in any eusocial insect. More broadly, there is a paucity of studies that investigate how IGEs vary across environments, in- cluding IGEs of (allo)parents on offspring. Thus, we first tested the effects of caregiver genotype on brood phenotypes, as well as on the foraging and feeding activity of colonies. Then, in a second ex- periment, we experimentally manipulated the age and number of caregivers to test if caregiver IGEs are modifiable by physiological and/or environmental factors. We found that caregiver effects on brood phenotypes were altered by context, showcasing an empir- ical example of IGE-by-environment interactions related to (allo) parental care. MATERIALS AND METHODS Animal maintenance and genotypes used Stock colonies were maintained in climate-controlled rooms, at 24 ± 1 °C and ≥60% relative humidity in sealed plastic con- tainers with a plaster of Paris floor. Stocks were fed Monday, Wednesday, and Friday during the brood care phase with frozen Solenopsis invicta larvae and pupae. For Experiment 1, we collected recently eclosed adults (i.e., callows) from stocks C17 (line A), STC6 (line B), BG9 (line D), and BG14 (line M) to serve as care- givers, and eggs from stock STC6 (line B) to serve as the focal brood. We used young adult workers (31–35 days old) as care- givers because young worker ants primarily engage in nursing brood but can also perform tasks regularly performed by older workers (Robinson 1992; Ulrich et al. 2021). For Experiment 2, we collected recently eclosed callows from stocks C16 (line A) and STC6 (line B) to serve as caregivers, and eggs and/or first instar larvae from two different stocks of STC6 (line B) to serve as the focal brood. Stock C16 and C17 (line A) are lineages originating from Okinawa, Japan; all STC6 stocks (line B) originated from St. Croix in the U.S. Virgin Islands (Kronauer et al. 2012). BG9 (line D) and BG14 (line M) were collected in Bangladesh (Trible et al. 2020). Lines A and B are globally invasive lines nested within a clade of lines native to Bangladesh, which includes lines D and M. Phylogenetic analysis using five loci (cytochrome oxidase I, cy- tochrome oxidase II, wingless, elongation factor 1α, and long wavelength rhodopsin) shows that all four lines are closely related, with a re- ported genetic distance between any pair below 0.02 (Trible et al. 2020). Due to their close relationship, we did not expect major incompatibilities between the adults and larvae of the different lines. At the same time, previous work has shown that line A and B caregivers differ in foraging activity (Ulrich et al. 2021) and in the proportion of intercastes they rear (Teseo et al. 2014). We 644 Behavioral Ecology therefore expected that even closely related lines would show dif- ferences in phenotype. Experiment 1: Caregiver genotype effects on feeding, foraging and brood traits Experimental design and setup We generated eight experimental colonies for each of four lines (A, B, D, and M) that were similar in every aspect except for the gen- otype of caregivers—the reproductive physiology, age and number of caregivers, the age, number and genotype of brood, and diet were controlled (Figure 1a; Supplementary Table S1). Because we sourced all brood from a single stock colony to control for any ma- ternal effects, variability generated in brood phenotypes between treatments must stem from genetic (or perhaps, non-genetic, envi- ronmentally induced) differences between caregivers. To minimize non-genetic differences between caregivers, we sourced them from stock colonies maintained in a climate-controlled environment and isolated them shortly after eclosion prior to the experiment (thereby partly controlling pre- and post-imaginal environmental variability). Experimental colonies were initially composed of 30 regular callow workers (3–7 days old) in 50 mm diameter Petri dishes with a plaster of Paris floor (see Supplementary Figure S1 for a sche- matic of the experimental setup). Each colony was fed fire ant brood every 48 h, until eggs were laid. All colonies were main- tained at 24 ± 1 °C. Each colony was transferred to a raiding arena (Figure 1b) 22 days later, at which point colonies had eggs that were two to four days from hatching. Larvae inhibit worker ovarian activity (Ravary et al. 2006; Chandra et al. 2018), so we performed the brood swap when all colonies had second to third instar larvae, ensuring that caregivers did not lay eggs when rearing the focal cohort of brood. This also provided care- givers 6 days to settle in their new housing before swapping in the focal cohort of brood. On the day we performed the brood swap, we recorded the number of larvae that had been produced by the caregivers in each colony, fed each colony three fire ant worker pupae, and adjusted all colonies to have exactly 25 reg- ular workers. This resulted in eight experimental colonies per line, with each colony containing 50 eggs that were 10 ± 1 days old and sourced from a single line B stock colony, and 25 caregivers of a single line that were approximately the same age across all four lines (Figure 1a). Once all colonies had ≥50% first instar larvae, we fed the col- onies fire ant worker pupae daily by putting food directly into the brood chamber. For the first five feeding events, we provided col- onies with six fire ant worker pupae. For days six to nine, we fed colonies with 12 fire ant worker pupae. Then, colonies were fed 18 fire ant worker pupae until ≥50% of remaining larvae had become prepupae. After the sixth day, prior to feeding, we began recording the number of uneaten fire ant pupae from the previous feeding event. For the next 8 days, we assigned a daily feeding score for each colony on a scale from zero to six, based on the number of food items eaten (e.g., score of six if at least 16 of 18 pupae were (a) (b) Genotype Age (days) A ~32 B ~35 D ~32 M ~31 # of adults # of eggs 25 50 25 50 25 50 25 50 Genotype B B B B n = 8 n = 8 n = 8 n = 8 (d ) a b ac bc d e s o l c e d o o r b f o n o i t r o p o r P 1.0 0.8 0.6 0.4 0.2 0.0 a a a a (e) 0.15 s e t s a c r e t n i f o n o i t r o p o r P 0.10 0.05 0.00 a ab b b A ab B a D b M b (c) e a p u p e r p % 0 5 ≥ o t s y a D (f ) 16.0 15.5 15.0 14.5 14.0 13.5 13.0 ) m m ( e z i s y d o B 1.9 1.8 1.7 1.6 1.5 1.4 A B D M A B D M A B D M Figure 1. (a) Experimental design to test the effects of caregiver genotype on the growth and development of brood. (b) Raiding arenas used to house the experimental colonies, with a small nest chamber and larger foraging chamber connected by a narrow tunnel; image from (Kronauer 2020). (c) Broods’ duration of larval development, (d) survival, (e) intercaste proportions, and (f) body size in response to caregiver genotype (x-axis). For c-e, each point represents a colony. For f, each point represents an individual ant. Letters indicate significant differences; bars represent 95% CIs of the mean. Piekarski et al. • Context alters indirect genetic effects 645 eaten, five if 13 to 15 pupae were eaten, etc.). During the entire brood care phase, we also video recorded each colony for four hours per day just prior to feeding. Video recordings were analyzed using the tracking software anTraX (Gal et al. 2020). We estimated the daily foraging activity of all colonies by quantifying the av- erage number of ants in the foraging chamber at any time and the total distance travelled by ants in the foraging chamber (detailed given in Supplementary Material). For each colony we recorded the date that the remaining brood were ≥50% first instar larvae, ≥50% prepupae, and ≥50% eclosed as adults. We also recorded the number of brood surviving to eclosion. The proportion of intercaste brood corresponds to the proportion with vestigial eyes (Supplementary Figure S2). Individuals from the focal brood cohort were harvested 5 to 7 days after they had eclosed as adults. To estimate body size, we measured and summed the length of the head, thorax, and first gastral segment from a lateral view (Supplementary Figure S2). We measured the body sizes of nine random caregivers from two colonies per line (18 caregivers per line), and 12–13 of the resulting focal adults without vestigial eyes from each colony. Colonies 5, 7, and 8 of line B, and colony 1 of line M had low survival, and we were only able to measure body size for one, seven, seven, and 11 individuals, respectively. Since brood that developed vestigial eyes were a priori excluded from the body size analysis, our tests for differences across caregiver genotypes is more conservative. Images were taken with a Leica Z16 APO microscope equipped with a Leica DFC450 camera using the Leica Application Suite version 4.12.0 software (Leica Microsystems, Switzerland). Lengths of body segments were meas- ured using ImageJ (Schindelin et al. 2012). Statistical analyses We used a one-way ANOVA with post hoc Tukey’s tests to statisti- cally test for differences in length of larval development and brood survival across conditions. We did the same to compare the average body size of the regular workers used as caregivers across condi- tions—all caregivers belonging to the same line came from the same stock colony, so there are no random effects due to colony. To assess whether caregiver genotype influenced colony feeding ac- tivity over the course of the brood care phase, we summed the daily feeding scores of each colony to calculate a cumulative feeding score and then performed a one-way ANOVA. We conducted a re- peated measures ANOVA on both metrics of daily foraging activity after fourth root transformation to test if caregiver genotypes differ in their foraging activities for each day. To compare the cumulative foraging activity over the entire brood care phase between caregiver genotypes, the distances travelled per day were summed and then square root transformed to produce values that met the assump- tions of ANOVA. The above analyses were conducted in GraphPad PRISM 8. Our body size measurements of reared brood corre- spond to individual ants, some of which share the same colony, and so we applied a linear mixed model (LMM) that treated caregiver genotype as a fixed effect and colony as a random effect. LMM ana- lyses were done in R 4.1.2 (R Core Team 2022) using the package lme4 (Bates et al. 2015). To test if caregiver genotype influenced the propensity of brood to develop as intercastes, a generalized linear mixed model (GLMM) with binomial error and logit-link was used with caregiver genotype being treated as a fixed effect, and colony identity as a random effect. This was implemented in R using the glmer function. Significance of model terms was assessed using the function Anova in the package car (Fox and Weisberg 2019), and emmeans (Lenth 2022) was used to conduct post hoc Tukey’s tests to correct for multiple comparisons. Assumptions of statistical models were validated using the function simulateResiduals in the package DHARMa (Hartig 2022). Experiment 2: Conditionality of caregiver genotype effects on brood traits Experimental design and setup We performed a second experiment to test whether the effects of caregiver genotype on brood phenotypes (i.e., length of larval de- velopment, survival and intercaste proportions) are dependent on other physiological or socio-environmental factors (i.e., caregiver age and colony size). First, we established 10 colonies with 50 reg- ular worker callows (colony size = 50) and five colonies with 25 regular worker callows (colony size = 25) for both lines A and B in Petri dishes. Caregivers of lines A and B came from stocks that began the brood care phase on the exact same day. A subset of five colonies with 50 caregivers per line were given 50 first instar larvae of line B (STC6) immediately, at which point the caregivers were 5–7 days old (young A50 and young B50 condition; Supplementary Table S2). For the remaining colonies, once their larvae began hatching, we swapped in 50 first instar larvae from a line B (STC6) stock colony (old A25, old B25, old A50, old B50 conditions; Supplementary Table S2), at which point caregivers were approx- imately one month old. Colonies were maintained at 24 ± 1 °C and fed every 48 h. Using the same approach as in experiment 1, we recorded the length of larval development, the total number of individuals reaching eclosion, and the proportion of brood that de- veloped as intercastes. Statistical analyses To test what factors influenced length of larval development and brood survival, we performed linear regression (package lme4) on a model incorporating all main effects and possible interactions in our experimental design: ~caregiver genotype + age + colony size + caregiver genotype: colony size + caregiver genotype: age. Length of larval development was reciprocal transformed (Y=1/Y) to meet the assumption of normally distributed residuals. However, the residuals showed heteroskedasticity. Thus, we estimated heteroskedasticity-consistent standard errors (i.e., Huber-White ro- bust standard errors) using the function vcovHC in the package sand- wich (Zeileis et al. 2020), applying the HC4 estimator (Cribari-Neto 2004), to make our interval estimates and hypothesis testing valid. For brood survival, the assumptions of normally distributed and homoscedastic residuals were met. To test what factors influenced the propensity of brood to develop as intercastes, a GLMM with binomial error and logit-link was used with all factors and inter- actions (as shown above) being treated as fixed effects, and colony identity as a random effect. This was implemented in R using the glmer function. Assumptions of all models were validated using the function simulateResiduals in the package DHARMa (Hartig 2022). Significance of model terms was assessed using the function Anova in the package car (Fox and Weisberg 2019), and emmeans (Lenth 2022) was used to conduct post hoc tests while correcting for multiple comparisons with the Benjamini-Hochberg method. We used the function emmip of package emmeans to create interaction plots of es- timated marginal means based on the linear models. Lastly, for each colony comprised of 25 old caregivers (i.e., old A25 and old B25), we measured the body size of five to six hap- hazardly selected adult individuals from the cohort they had raised. To test if caregiver genotypes influenced the final body size of their 646 Behavioral Ecology brood, we applied a LMM with caregiver genotype as the fixed ef- fect and colony as a random effect. RESULTS Experiment 1: Caregiver genotype effects on brood traits Caregiver genotype had a significant effect on the length of larval development (ANOVA: F3,28 = 4.94, P = 0.007; Figure 1c). Length of larval development differed between brood reared by line A and D caregivers, and line A and M caregivers (Tukey’s HSD: A vs. D, μd = 0.75 days, P = 0.024; A vs. M, μd = 0.88 days, P = 0.007). Length of larval development for brood reared by line A caregivers was not significantly different from those reared by line B caregivers after correcting for multiple comparisons (Tukey’s HSD: A vs. B, P = 0.126). Caregiver genotype also influenced the time to eclo- sion of brood (ANOVA: F3,28 = 14.91, P < 0.0001; Supplementary Figure S3). Brood reared by line A caregivers took longer to eclose compared to all other lines (Tukey’s HSD: A vs. B, μd = 1.44 days, P < 0.0001; A vs. D, μd = 1.44 days, P < 0.0001; A vs. M, μd = 1.19 days, P < 0.001), with no differences between line B, D, and M caregivers. Brood survival to eclosion was influenced by caregiver genotype (ANOVA: F3,28 = 15.97, P < 0.0001; Figure 1d). Line A caregivers increased the survival of brood relative to lines B and M, and D caregivers improved survival relative to B caregivers (Tukey’s HSD: A vs. B, P < 0.0001; A vs. M, P < 0.001; B vs. D, P = 0.002). Regarding the proportion of larvae that developed into intercastes, caregiver genotype had no significant effect (GLMM: χ²(3) = 1.27, P = 0.7361; Figure 1e). However, very few intercastes were produced in this experiment, resulting in low statistical power. Nevertheless, there was a trend for line A caregivers to produce more intercastes than line B caregivers (Figure 1e). Caregiver genotype influenced the average body size of brood (LMM: χ²(3) = 12.89, P = 0.005; Figure 1f). Brood reared by line B caregivers attained smaller adult body sizes on average compared to those reared by line D or M caregivers (Tukey’s HSD: B vs. D, P = 0.0245; B vs. M, P = 0.0160), but not line A caregivers (Tukey’s HSD: A vs. B, P = 0.2510). When adding the number of eclosed siblings as a fixed effect to the LMM, it had no significant effect on the body size of brood (χ²(1) = 0.766, P = 0.381), and neither did length of larval development (χ²(1) = 0.056, P = 0.813). Experiment 1: Body size, feeding, and foraging across caregiver genotypes Caregivers of different genotypes differed in their body size on av- erage (ANOVA: F3,68 = 33.31, P < 0.0001; Figure 2a). Line A and D caregivers were larger than line B and M caregivers on average (Tukey’s HSD: A vs. B, P < 0.0001; A vs. M, P < 0.0001; B vs. D, P < 0.0001; D vs. M, P < 0.0001; A vs. D, P = 0.509; B vs. M, P = 0.469). Caregiver genotype influenced the cumulative feeding score of colonies during the brood care phase (ANOVA: F3,28 = 16.59, P < 0.0001; Figure 2b). Colonies with line B caregivers ate less food relative to all other lines (Tukey’s HSD: A vs. B, P < 0.0001; B vs. D, P < 0.0001; B vs. M; P = 0.0001). Furthermore, a time course plot comparing daily feeding scores across the four genotypes shows that colonies with line B caregivers consumed less food than those with line A, D, and M caregivers on many days, while colonies with A, D, and M caregivers had similar feeding activity (Supplementary Figure S4a). Caregiver genotype influenced the total distance travelled by ants in the foraging chamber during the brood care phase (ANOVA: F3,28 = 30.72, P < 0.0001; Supplementary Figure S4b). Caregivers of line D were most active (Tukey’s HSD: A vs. D, P < 0.0001; B vs. D, P < 0.001; D vs. M; P < 0.0001), and line B caregivers were more active than line A and M caregivers (Tukey’s HSD: A vs. B, P = 0.016; B vs. M, P = 0.003). Caregivers of line A and M showed no difference in the total distance travelled during the brood care phase (Tukey’s HSD: A vs. M, P = 0.911). When looking at pat- terns of daily foraging activity, both metrics for estimating foraging activity show that line D caregivers were more active than line A and M on most days, while line B caregivers were more active than lines A and M, specifically on days eight and nine, which coincides with when larvae became fourth instars (Figure 2c; Supplementary Figure S4c; Supplementary Tables S3 and S4). In addition to differences in the average body size of regular workers (Figure 2a), we also found differences between the four lines in fecundity, reproductive maturation, and readiness to lay eggs after larvae became prepupae (Supplementary Figures S5 and S6). Experiment 2: Conditionality of caregiver IGEs Caregiver genotype (LM: F1,24 = 7.68, P = 0.011), age (F1,24 = 52.14, P = 1.843-07), an interaction between caregiver genotype and age (F1,24 = 5.11, P = 0.033), and an interaction between care- giver genotype and colony size (F1,24 = 21.44, P = 1.062-04) had a significant effect on the length of larval development. Interaction plots of the estimated marginal means (and 95% CIs) show that caregiver genotype and age interacted (Figure 3a), and caregiver genotype and colony size interacted (Figure 3b) to influence length of larval development. For colonies with 50 caregivers, larvae reared by young line B caregivers had longer development com- pared to those reared by old B caregivers (post hoc test: P < 0.0001; Supplementary Table S5), but larvae reared by young A caregivers did not differ from those reared by old A caregivers (post hoc test: P = 0.645; Supplementary Table S5; Figure 3a). When comparing old A caregivers at colony sizes of 25 and 50, larger colony size was associated with longer larval development (post hoc test: P < 0.001; Supplementary Table S5), but not when comparing old B caregivers (post hoc test: P = 0.060—trend in opposite direction; Supplementary Table S5; Figure 3b). Also, there was a difference in length of larval development when comparing A and B care- givers at colony sizes of 50 (both young and old), but not at colony sizes of 25 (post hoc tests: young A50 vs. young B50, P = 0.013; old A50 vs. old B50, P = 0.0001; old A25 vs. old B25, P = 0.497; Supplementary Table S5), meaning a caregiver genotype effect was detected only at colony sizes of 50 (Figure 3b). Caregiver genotype (LM: F1,24 = 5.22, P = 0.032) and its interac- tion with colony size (F1,24 = 18.96, P = 2.143-04) had a significant effect on brood survival. Interaction plots of the estimated marginal means (and 95% CIs) show that caregiver genotype and colony size interacted to influence brood survival (Figure 3c). Line A caregivers had higher brood survival at colony sizes of 25 compared to 50 (post hoc test: P < 0.001; Supplementary Table S6), but colony size had no effect for line B caregivers (post hoc test: P = 0.096—trend in opposite direction; Supplementary Table S6; Figure 3c). Also, line B caregivers outperformed line A caregivers at colony sizes of 50 (post hoc test: P < 0.001; Supplementary Table S6), but line A caregivers outperformed line B caregivers at colony sizes of 25 (post hoc test: P = 0.034; Supplementary Table S6). Thus, the direction of the care- giver genotype effect depended on colony size (Figure 3c). Piekarski et al. • Context alters indirect genetic effects 647 (a) ) m m ( e z i s y d o B (c) a b a a a b a b 1.8 1.7 1.6 1.5 1.4 (b) e r o c s g n i d e e f e v i t a l u m u C 50 45 40 35 30 25 A B D M A B D M Line A Line B Line D Line M 3rd instar 4th instar 225 200 175 150 125 100 75 50 25 0 y a d r e p ) m ( d e l l e v a r t e c n a t s i d l a t o T 0 2 4 6 8 Day 10 12 14 Figure 2. (a) Body size of caregivers used in experimental colonies across all four genotypes. (b) Cumulative feeding scores of colonies in response to caregiver genotype. (c) Total distance travelled per day by ants in the foraging chamber (a proxy for daily foraging activity of colonies) across the four caregiver genotypes. The approximate time that larvae transitioned into third and fourth instars is denoted. Letters show significant differences (a, b); bars represent 95% CIs of the mean. Only caregiver genotype had a significant effect on the proportion of brood that developed into intercastes (GLMM: χ²(1) = 25.17, P = 5.24-07). The caregiver genotype effect on intercaste proportions was not conditional on context (Figure 3d; Supplementary Figure S7). All three comparisons between A and B caregivers show that a higher proportion of brood developed into intercastes when reared by line A caregivers (post hoc tests: young A50 vs. young B50, P = 0.004; old A50 vs. old B50, P = 0.042; old A25 vs. old B25, P = 0.019; Supplementary Table S7). When adding the number of eclosed brood as a fixed effect to the GLMM, it had no significant effect on the proportion of intercaste brood (χ²(1) = 2.02, P = 0.155). Lastly, a LMM shows that brood reared by line A caregivers attained a larger body size on average compared to those reared by line B caregivers (t = 5.28, P = 0.001), even when excluding intercastes from the analysis (t = 4.88, P = 0.002; Supplementary Figure S8). DISCUSSION Our results show that caregiver genotype has a significant ef- fect on the foraging and feeding activity of colonies, and on the growth and development of brood. We recover significant and robust IGEs of caregivers on length of larval development, sur- vival, body size, and (sub)caste development. We found that care- giver genotype interacts with other factors (age and colony size) to influence length of larval development, and that differences in brood survival between caregiver genotypes can also be context de- pendent. Altogether, the results show that caregivers have profound and sweeping IGEs on brood phenotypes in O. biroi, and that these IGEs can be conditional. Indirect genetic effects versus indirect environmental effects Parental effects can be genetic, where genetic variability between parents generates variability in offspring phenotype (i.e., IGEs), or non-genetic, where variability in the phenotype of parents that was environmentally induced generates variability in offspring phenotype (i.e., indirect environmental effects—also known as transgenerational plasticity) (Mousseau and Fox 1998; Weaver et al. 2004; Danchin et al. 2011; Tariel et al. 2020). Our experimental 648 Behavioral Ecology (a) 0.06 0.05 0.04 0.03 ) ] s y a d [ t n e m p o l e v e d l a v r a l f o h t g n e l ( / 1 (b) 50 caregivers old caregivers Caregiver genotype A B short r e t s a f s l o w e r long (c) 40 30 20 10 d e s o l c e d o o r b f o r e b m u N (d ) old & young caregivers 50 caregivers 25 caregivers d o o r b e t s a c r e t n i f o n o i t r o p o r P 0.4 0.2 0.0 old caregiver age young 50 25 # of caregivers 50 25 # of caregivers old young caregiver age old Figure 3. Interaction plots of the estimated marginal means for (a, b) length of larval development (reciprocal transformed), (c) brood survival, and (d) proportion of intercaste brood reared across conditions. Experimental colonies were comprised of either 25 or 50 caregivers of line A (red) or line B (blue). For colony sizes of 50 caregivers, caregivers were either ~1 month old or ~1 week old (i.e., old vs. young). For colony sizes of 25, caregivers were ~1 month old. Error bars represent 95% CIs of the mean. Violin plots are shown in Supplementary Figure S9. design reduces phenotypic variability across caregiver genotypes due to different environmental backgrounds. Reproducibility of caregiver genotype effects on the same brood phenotypes across studies (Teseo et al. 2014; Jud et al. 2022), and across experiments (this study), provides good evidence that these effects have some ge- netic basis. Thus, these effects can be classified as IGEs. Prenatal (i.e., before egg hatching) maternal effects on caste fate have been demonstrated in some ants. In Pogonomyrmex seed har- vester ants, only eggs laid by queens that were exposed to cold and were at least 2 years of age have the potential to develop into fu- ture queens (Schwander et al. 2008; Libbrecht et al. 2013). Prenatal maternal effects on brood development and body size have also re- cently been shown in honeybees (Wei et al. 2019). The contribution of non-genetic maternal effects in determining body size has not been formally investigated in the clonal raider ant. However, pre- natal maternal effects do not contribute to differences in the body size and intercaste proportions of brood between caregiver geno- types in our experiments because all eggs within an experiment were sourced from the same line B stock colony, and thus the size and quality of eggs given to all caregivers were on average equal. There can also be postnatal (i.e., after egg hatching) maternal effects on offspring body size, in which larger caregivers provision offspring more (Hunt and Simmons 2000; Kindsvater et al. 2012; Rollinson and Rowe 2016). In Nicrophorus burying beetles, there is transgenerational plasticity of adult body size, such that, among genetically similar mothers that have been experimentally manipu- lated to be either small or large, large mothers rear their brood to larger body sizes mostly via postnatal maternal effects (Steiger 2013). In contrast, the size of workers does not affect the size of brood in bumblebees (Cnaani and Hefetz 1994). In our study, line M caregivers were smaller than line D caregivers, and of similar size to line B caregivers (Figure 2a), yet they reared brood to the same body size on average as line D caregivers and to a larger body size than line B caregivers (Figure 1f). This suggests that the body size differences of caregivers across genotypes, whether genetically based or not, do not explain the differences in the average body size of brood they reared. Possible mechanisms of caregiver effects on brood size and caste Our results demonstrate that caregiver genotype influences the body size and caste fate of brood in O. biroi through postnatal ef- fects, but the mechanistic basis of this effect remains unknown. One hypothesis is that differences in foraging activity translate to differ- ences in larval growth. Consistent with previous studies, we find that workers of different genotypes differ in their levels of foraging activity (Ulrich et al. 2021; Jud et al. 2022). However, the differ- ences in foraging activity between different genotypes of caregivers do not correlate with differences in the body size of brood. Line B caregivers showed higher foraging activity than lines A and M, but lower foraging activity than line D (Figure 2c; Supplementary Figure S3b), yet brood reared by line B were on average smaller than brood reared by lines D and M (Figure 1f). Another hypothesis is that differences in the permissiveness of larval cannibalism across caregiver genotypes leads to differences in the average body sizes of brood. However, adding brood survival as a fixed effect in the linear models of both experiments showed that it had no significant ef- fect on the body size or intercaste proportions of brood. Therefore, there is no evidence that differences in levels of larval cannibalism contribute to explaining the variation in body size and intercaste proportions, consistent with observations in other studies (Lecoutey et al. 2011; Teseo et al. 2014; Jud et al. 2022). Many behavioral differences across caregiver genotypes could generate variation in the average body size or intercaste Piekarski et al. • Context alters indirect genetic effects 649 proportions of brood in O. biroi. For example, caregivers influence caste fate by biting larvae in some ant species (Brian 1973; Penick and Liebig 2012). To determine if such, or other, adult-larva interactions are explanatory, high magnification video recordings of how caregivers of different genotypes interact with larvae are needed. Another possibility is that caregiver genotype influences the quantity of nutrition received by brood. We found that colo- nies with line B caregivers ate significantly less food compared to all other genotypes (Figure 2b) and reared brood to smaller body sizes on average (Figure 1f). Differences in feeding activity of col- onies may represent a DGE on caregiver food intake, an IGE on brood food intake, or a combination of both. Since we only meas- ured colony-level feeding activity, we are unable to discern how food was distributed amongst adults and larvae. It is possible that certain genotypes of caregivers are more voracious, perhaps due to a higher metabolic rate or satiation threshold, and the brood received the same amount of food across caregiver genotypes. Alternatively, perhaps caregivers of different genotypes exhibit differences in the production of pheromones that act on either larvae or workers to ultimately influence body size or caste ratios. In several ant species, queen pheromone prevents sexualization of brood by affecting worker rearing behavior and also elicits workers to execute sexual brood (Vargo and Fletcher 1986; Edwards 1991; Vargo and Passera 1991; Oliveira et al. 2020). Accordingly, a fertility signal analogous to the queen pheromone in other ants has been suggested to modulate intercaste proportions in O. biroi (Lecoutey et al. 2011). More experiments are needed to discover the mechanisms that contribute to worker regulation of body size and caste fate of brood in O. biroi. IGE-by-environment interactions on brood development and survival Caregiver genotype had a significant effect on length of larval development in both experiments, suggesting that genotypes differ in how they interact with their brood. This finding is cor- roborated by a recent study, which found that both caregiver and brood genotypes influenced the length of larval development (Jud et al. 2022). A shortcoming of our study is that we varied caregiver genotype but not brood genotype, meaning that the ob- served effects on brood phenotypes could be caused by either the matching status between partner genotypes (i.e., B\|B = match vs. A\|B = mismatch) or caregiver genotype. However, larvae reared by line A caregivers develop longer than when reared by line B caregivers regardless of brood genotype (Jud et al. 2022). Thus, the brood’s length of larval development is influenced by an IGE rather than a mismatch-of-genotypes effect (i.e., a GxG interac- tion). Furthermore, even if the caregiver genotype effect was de- pendent on the genotype of the brood (i.e., a GxG interaction), this translates to an IGE that is dependent on the genotype of the social partner and may also be modifiable by additional envi- ronmental factors. We found that length of larval development was influenced by an interaction between caregiver genotype and age, and caregiver genotype and colony size (Figure 3a,b), suggesting that the magni- tude of the IGE depended on the age and number of caregivers. For example, old line A and B caregivers showed differences in their brood’s length of larval development at colony sizes of 50, but not 25, indicating the IGE is conditional on colony size (Figure 3b). Altogether, our results highlight that the rate of develop- ment of brood depends on caregiver genotype, and that this IGE is modifiable by other physiological and/or environmental fac- tors. Although it is not surprising to find that caregiver IGEs on the length of larval development are context-dependent, to our knowledge, this is the first study to demonstrate such an IGE-by- environment interaction in a eusocial insect. In the alpine silver ant Formica selysi, brood reared by workers from monogynous colonies have better survival compared to those reared by workers from polygynous colonies (Purcell and Chapuisat 2012), and the workers-to-larvae ratio affects brood survival and body size (Purcell et al. 2012). We recovered differ- ences in brood survival between line A and B caregivers (Figure 1d), and evidence that the direction and magnitude of this IGE is context-dependent (Figure 3c). Two previous studies that cross- fostered lines A and B found no effect of caregiver genotype on brood survival (Teseo et al. 2014; Jud et al. 2022). In the most re- cent study (Jud et al. 2022), colonies were comprised of 8 adults and 7 larvae, and in the first study (Teseo et al. 2014), colonies were comprised of 50 adults, but the number of larvae was not standardized (ranging from ~15 to 78 larvae). We found that line B caregivers improved brood survival relative to line A caregivers in colonies of 50 adults and 50 larvae, while line A caregivers showed better performance relative to line B caregivers in colo- nies of 25 adults and 50 larvae in both experiments. For some un- known reason, colonies with 50 one-month-old line A caregivers and 50 larvae had strikingly poor brood survival (Figure 3c). This implies that an interaction between colony size and caregiver gen- otype influences brood survival, where line A caregivers perform better at smaller colony sizes or a lower workers-to-larvae ratio. Thus, the inconsistency of caregiver genotype effects on brood survival across studies could be due to each study having different colony sizes and workers-to-larvae ratios. Our results suggest that, in O. biroi, the effect of caregiver genotype on brood survival is conditional on additional properties of the social environment (i.e., another IGE-by-environment interaction). Outlook on IGE-by-environment interactions In the presence of IGE-by-environment interactions, it is expected that the magnitude and direction of IGEs will depend on the spe- cific environmental conditions experienced. Therefore, detection of IGEs on a trait may differ across studies. Some IGEs may be more robust and consistent across contexts, such as the effect of caregiver genotype on intercaste proportions in O. biroi (Teseo et al. 2014; Figure 3d). However, even for this trait, our first experiment did not detect a significant IGE (Figure 1e), suggesting some un- known factor dampened the effect size. A recent study examined the relative impact of DGEs and IGEs on brain gene expression in recently eclosed O. biroi adults when cross-fostering lines B and M (Kay et al. 2022). They found that 0 genes were differentially expressed in recently eclosed adults due to IGEs, but over 1000 were differentially expressed due to DGEs. Although surprising, it is important to point out that the transcriptomic data is limited to adult brain tissue one or eight days after eclosion, so the extent that caregiver IGEs influence gene expression of brood in other tis- sues and during pre-imaginal development is unknown. Also, IGEs on larval phenotypes (e.g., length of development) may not neces- sarily translate to differences in brain gene expression later in adult- hood. Furthermore, the experiment was performed under a single set of experimental conditions (i.e., colony sizes of nine adults and nine larvae), and IGEs on gene expression may also be context dependent. 650 Behavioral Ecology The magnitude and direction of IGEs are dependent on factors other than the social partner’s genotype (i.e., GxG epistasis), yet there are not many examples reported in the literature. A condi- tional IGE has been well documented in the red imported fire ant, Solenopsis invicta, where colony social organization is under strong genetic control by a supergene haplotype, referred to as the Social b (Sb) haplotype (Wang et al. 2013; Yan et al. 2020). Monogynous colonies contain only SB/SB queens and workers, while polygynous colonies have multiple SB/Sb queens and both SB/SB and SB/Sb workers. SB/SB workers normally kill SB/Sb and supernumerary queens, but when the frequency of SB/Sb workers in the colony surpasses a threshold of 5–10%, SB/SB workers tolerate multiple SB/Sb queens and kill SB/SB queens (Ross and Keller 2002). Thus, the IGE of SB/Sb workers on SB/SB worker behavior is conditional on the proportion of heterozygous workers. In other animals, IGEs contribute more to the heritable variance of growth rate in com- petitive (restricted feeding) relative to non-competitive (ad libitum feeding) contexts, suggesting IGEs may be stronger when resources are limited (in pigs: Piles et al. 2017; in shrimp: Luan et al. 2020). In flies, male IGEs on female locomotion are dependent on abiotic factors (Signor et al. 2017). A study on red squirrels found indirect effects on neighbor’s parturition dates at high population densities and not at low densities but could not conclude if these indirect effects had a genetic basis (Fisher et al. 2019). Our study provides an example of conditional IGEs relevant to interactions between caregivers and their young. ACKNOWLEDGEMENTS We thank Leonora Olivos-Cisneros for help with ant maintenance, and the entire Kronauer lab for helpful discussion and feedback. We are grateful to Vikram Chandra and Asaf Gal for assistance with automated behavioral tracking. We also thank Yuko Ulrich and Tim Linksvayer for providing valu- able feedback on previous versions of the manuscript. Lastly, we thank both anonymous reviewers for providing constructive comments that improved the manuscript. This is Clonal Raider Ant Project paper #25. AUTHOR CONTRIBUTIONS Patrick Piekarski (Conceptualization-Equal, Data curation-Lead, Formal analysis-Lead, Investigation-Lead, Methodology-Equal, Validation-Equal, Visualization-Lead, Writing – original draft-Lead, Writing – review & editing-Equal), Stephany Valdés-Rodríguez (Investigation-Supporting, Resources-Supporting, Writing – review & editing-Supporting), Daniel Kronauer acquisition-Lead, Methodology-Equal, Resources-Lead, Supervision-Lead, Validation- Equal, Writing – original draft-Supporting, Writing – review & editing-Equal) (Conceptualization-Equal, Funding CONFLICT OF INTEREST The authors declare that they have no conflict of interest in this study. DATA AVAILABILITY Analyses reported in this article can be reproduced using the data provided by Piekarski et al. (2023). CONCLUSIONS Handling Editor: Robin Tinghitella The significance of IGE-by-environment interactions in shaping traits has been largely omitted in the social insect literature. Here we found that brood phenotypes are influenced by IGEs stemming from their caregivers, and that these IGEs can be altered by the physiological and/or environmental context (i.e., the number and age of caregivers). The mechanisms by which O. biroi caregivers regulate their brood’s rate of development, survival, body size and caste fate remain unknown. However, by identifying caregiver genotypes that produce different rearing environments that ulti- mately generate differences in these brood traits, we can now begin to reveal such mechanisms and their genetic basis. Understanding the intimate coevolution of (allo)parent and off- spring phenotypes requires insights into the IGEs that both have on each other at the level of physiology, behavior, and gene expression, as well as how these IGEs are modified by other environmental fac- tors. We stress the importance of extending a gene-by-environment framework to future research on IGEs, where measurement of IGEs across various conditions may be necessary to draw inferences about how traits shaped by social interactions evolve. SUPPLEMENTARY MATERIAL Supplementary material can be found at Behavioral Ecology online. FUNDING This work was supported by the National Institute of General Medical Sciences of the National Institutes of Health (R35GM127007 to D.J.C.K.). The content is solely the responsibility of the authors and does not neces- sarily represent the official views of the National Institutes of Health. This work was also supported by a Faculty Scholar Award from the Howard Hughes Medical Institute to D.J.C.K. D.J.C.K. is an investigator of the Howard Hughes Medical Institute. REFERENCES Bailey NW, Desjonquères C. 2021. The indirect genetic effect interaction coefficient ψ: theoretically essential and empirically neglected. J Hered. 113:79–90. Bates D, Mächler M, Bolker B, Walker S. 2015. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67:1–48. Brian MV. 1973. Caste control through worker attack in the ant Myrmica. Insectes Soc. 20:87–102. Buttery NJ, Thompson CRL, Wolf JB. 2010. Complex genotype inter- actions influence social fitness during the developmental phase of the so- cial amoeba Dictyostelium discoideum. J Evol Biol. 23:1664–1671. Chandra V, Fetter-Pruneda I, Oxley PR, Ritger AL, McKenzie SK, Libbrecht R, Kronauer DJC. 2018. Social regulation of insulin signaling and the evolution of eusociality in ants. Science. 361:398–402. Cnaani J, Hefetz A. 1994. The effect of workers size frequency distribution on colony development in Bombus terrestris. Insectes Soc. 41:301–307. Cribari-Neto F. 2004. Asymptotic inference under heteroskedasticity of un- known form. Comput Stat Data Anal. 45:215–233. Culumber ZW, Kraft B, Lemakos V, Hoffner E, Travis J, Hughes KA. 2018. GxG epistasis in growth and condition and the maintenance of genetic polymorphism in Gambusia holbrooki. Evolution. 72:1146–1154. Danchin E, Charmantier A, Champagne FA, Mesoudi A, Pujol B, Blanchet S. 2011. Beyond DNA: integrating inclusive inheritance into an extended theory of evolution. Nat Rev Genet. 12:475–486. Edwards JP. 1991. Caste regulation in the pharaoh’s ant Monomorium pharaonis: recognition and cannibalism of sexual brood by workers. Physiol Entomol. 16:263–271. Fisher DN, Kilgour RJ, Siracusa ER, Foote JR, Hobson EA, Montiglio P-O, Saltz JB, Wey TW, Wice EW. 2021. Anticipated effects of abi- otic environmental change on intraspecific social interactions. Biol Rev. 96:2661–2693. Fisher DN, Wilson AJ, Boutin S, Dantzer B, Lane JE, Coltman DW, Gorrell JC, McAdam AG. 2019. Social effects of territorial neighbours on the timing of spring breeding in North American red squirrels. J Evol Biol. 32:559–571. Fox J, Weisberg S. 2019. An R companion to applied regression. Thousand Oaks (CA): Sage. Piekarski et al. • Context alters indirect genetic effects 651 Gal A, Saragosti J, Kronauer DJC. 2020. anTraX, a software package for high-throughput video tracking of color-tagged insects. eLife. 9:e58145. Gupta AP, Lewontin RC. 1982. A study of reaction norms in natural popu- lations of Drosophila pseudoobscura. Evolution. 36:934–948. Hartig F. 2022. DHARMa: residual diagnostics for hierarchical (multi- mevel/mixed) regression models. R package version 0.4.5. Hunt J, Simmons LW. 2000. Maternal and paternal effects on offspring phenotype in the dung beetle Onthophagus taurus. Evolution. 54:936–941. Hunt J, Simmons LW. 2002. The genetics of maternal care: Direct and in- direct genetic effects on phenotype in the dung beetle Onthophagus taurus. Proc Natl Acad Sci USA. 99:6828–6832. Jaffe A, Burns MP, Saltz JB. 2020. Genotype-by-genotype epistasis for ex- ploratory behaviour in D. simulans. Proc R Soc B. 287:20200057. Jud SL, Knebel D, Ulrich Y. 2022. Intergenerational genotypic interactions drive collective behavioural cycles in a social insect. Proc R Soc B. 289:20221273. Kay T, Alciatore G, La Mendola C, Reuter M, Ulrich Y, Keller L. 2022. A complete absence of indirect genetic effects on brain gene expression in a highly social context. Mol Ecol. 31:5602–5607. Kindsvater HK, Rosenthal GG, Alonzo SH. 2012. Maternal size and age shape offspring size in a live-bearing fish, Xiphophorus birchmanni. PLoS One. 7:e48473. Kronauer DJC. 2020. Army ants: nature’s ultimate social hunters. Cambridge (MA): Harvard University Press. Kronauer DJC, Pierce NE, Keller L. 2012. Asexual reproduction in introduced and native populations of the ant Cerapachys biroi. Mol Ecol. 21:5221–5235. Lecoutey E, Châline N, Jaisson P. 2011. Clonal ant societies exhibit fertility- dependent shifts in caste ratios. Behav Ecol. 22:108–113. Lenth RV. 2022. emmeans: estimated marginal means, aka least-squares means. R package version 1.7.2. Libbrecht R, Corona M, Wende F, Azevedo DO, Serrão JE, Keller L. 2013. Interplay between insulin signaling, juvenile hormone, and vitellogenin regulates maternal effects on polyphenism in ants. Proc Natl Acad Sci USA. 110:11050–11055. Linksvayer TA. 2006. Direct, maternal, and sibsocial genetic effects on indi- vidual and colony traits in an ant. Evolution. 60:2552–2561. Linksvayer TA. 2007. Ant species differences determined by epistasis be- tween brood and worker genomes. PLoS One. 2:e994. Linksvayer TA, Fondrk MK, Page RE Jr. 2009. Honeybee social regulatory networks are shaped by colony-level selection. Am Nat. 173:E99–E107. Linksvayer TA, Wade M. 2005. The evolutionary origin and elaboration of sociality in the aculeate hymenoptera: maternal effects, sib-social effects, and heterochrony. Q Rev Biol. 80:317–336. Luan S, Qiang G, Cao B, Luo K, Meng X, Chen B, Kong J. 2020. Feed competition reduces heritable variation for body weight in Litopenaeus vannamei. Genet Sel Evol. 52:45. Marie-Orleach L, Vogt-Burri N, Mouginot P, Schlatter A, Vizoso DB, Bailey NW, Schärer L. 2017. Indirect genetic effects and sexual conflicts: Partner genotype influences multiple morphological and behavioral re- productive traits in a flatworm. Evolution. 71:1232–1245. Mcadam AG, Boutin S, Réale D, Berteaux D. 2002. Maternal effects and the potential for evolution in a natural population of animals. Evolution. 56:846–851. McGlothlin JW, Moore AJ, Wolf JB, Brodie ED III. 2010. Interacting phenotypes and the evolutionary process. III. Social evolution. Evolution. 64:2558–2574. Moore AJ, Brodie ED III, Wolf JB. 1997. Interacting phenotypes and the evolutionary process: I. Direct and indirect genetic effects of social inter- actions. Evolution. 51:1352–1362. Mousseau TA, Fox CW. 1998. The adaptive significance of maternal effects. Trends Ecol Evol. 13:403–407. Oliveira RC, Warson J, Sillam-Dussès D, Herrera-Malaver B, Verstrepen K, Millar JG, Wenseleers T. 2020. Identification of a queen pheromone mediating the rearing of adult sexuals in the pharaoh ant Monomorium pharaonis. Biol Lett. 16:20200348. Pankiw T, Tarpy DR, Page RE Jr. 2002. Genotype and rearing environ- ment affect honeybee perception and foraging behaviour. Anim Behav. 64:663–672. Penick CA, Liebig J. 2012. Regulation of queen development through worker aggression in a predatory ant. Behav Ecol. 23:992–998. Peripato AC, Cheverud JM. 2002. Genetic influences on maternal care. Am Nat. 160:S173173–S17S185. Piekarski PK, Valdés-Rodríguez S, Kronauer DJC. 2023. Conditional indi- rect genetic effects of caregivers on brood in the clonal raider ant. Behav Ecol. doi:10.5061/dryad.wh70rxwsh Piles M, David I, Ramon J, Canario L, Rafel O, Pascual M, Ragab M, Sánchez JP. 2017. Interaction of direct and social genetic effects with feeding regime in growing rabbits. Genet Sel Evol. 49:58. Purcell J, Brütsch T, Chapuisat M. 2012. Effects of the social environment on the survival and fungal resistance of ant brood. Behav Ecol Sociobiol. 66:467–474. Purcell J, Chapuisat M. 2012. The influence of social structure on brood survival and development in a socially polymorphic ant: insights from a cross-fostering experiment. J Evol Biol. 25:2288–2297. R Core Team. 2022. R: a language and environment for statistical com- puting. Vienna, Austria: R Foundation for Statistical Computing. https:// www.R-project.org/. Ravary F, Jahyny B, Jaisson P. 2006. Brood stimulation controls the phasic reproductive cycle of the parthenogenetic ant Cerapachys biroi. Insectes Soc. 53:20–26. Ravary F, Jaisson P. 2002. The reproductive cycle of thelytokous colonies of Cerapachys biroi Forel (Formicidae, Cerapachyinae). Insectes Soc. 49:114–119. Ravary F, Jaisson P. 2004. Absence of individual sterility in thelytokous colonies of the ant Cerapachys biroi Forel (Formicidae, Cerapachyinae). Insectes Soc. 51:67–73. Rebar D, Bailey NW, Jarrett BJM, Kilner RM. 2020. An evolutionary switch from sibling rivalry to sibling cooperation, caused by a sustained loss of parental care. Proc Natl Acad Sci USA. 117:2544–2550. Robinson GE. 1992. Regulation of division of labor in insect societies. Annu Rev Entomol. 37:637–665. Rode NO, Soroye P, Kassen R, Rundle HD. 2017. Air-borne genotype by genotype indirect genetic effects are substantial in the filamentous fungus Aspergillus nidulans. Heredity. 119:1–7. Rollinson N, Rowe L. 2016. The positive correlation between maternal size and offspring size: fitting pieces of a life-history puzzle. Biol Rev Camb Philos Soc. 91:1134–1148. Ross K, Keller L. 2002. Experimental conversion of colony social orga- nization by manipulation of worker genotype composition in fire ants (Solenopsis invicta). Behav Ecol Sociobiol. 51:287–295. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, et al. 2012. Fiji: an open-source platform for biological-image analysis. Nat Methods. 9:676–682. Schwander T, Humbert JY, Brent CS, Cahan SH, Chapuis L, Renai E, Keller L. 2008. Maternal effect on female caste determination in a social insect. Curr Biol. 18:265–269. Signor SA, Abbasi M, Marjoram P, Nuzhdin SV. 2017. Social effects for locomotion vary between environments in Drosophila melanogaster females. Evolution. 71:1765–1775. Steiger S. 2013. Bigger mothers are better mothers: disentangling size-related prenatal and postnatal maternal effects. Proc R Soc B. 280:20131225. Tariel J, Plénet S, Luquet E. 2020. Transgenerational plasticity in the con- text of predator-prey interactions. Front. Ecol. Evol. 8. Teseo S, Châline N, Jaisson P, Kronauer DJC. 2014. Epistasis between adults and larvae underlies caste fate and fitness in a clonal ant. Nat Commun. 5:3363. Trible W, McKenzie SK, Kronauer DJC. 2020. Globally invasive popu- lations of the clonal raider ant are derived from Bangladesh. Biol Lett. 16:20200105. Ulrich Y, Kawakatsu M, Tokita CK, Saragosti J, Chandra V, Tarnita CE, Kronauer DJC. 2021. Response thresholds alone cannot explain empirical patterns of division of labor in social insects. PLoS Biol. 19:e3001269. Vargo EL, Fletcher DJC. 1986. Evidence of pheromonal queen control over the production of male and female sexuals in the fire ant, Solenopsis invicta. J Comp Physiol A. 159:741–749. Vargo EL, Passera L. 1991. Pheromonal and behavioral queen control over the production of gynes in the Argentine ant Iridomyrmex humilis (Mayr). Behav Ecol Sociobiol. 28:161–169. Walsh JT, Garonski A, Jackan C, Linksvayer TA. 2022. The collective be- havior of ant groups depends on group genotypic composition. J Hered. 113:102–108. Wang J, Wurm Y, Nipitwattanaphon M, Riba-Grognuz O, Huang Y-C, Shoemaker D, Keller L. 2013. A Y-like social chromosome causes alter- native colony organization in fire ants. Nature. 493:664–668. Weaver ICG, Cervoni N, Champagne FA, D’Alessio AC, Sharma S, Seckl JR, Dymov S, Szyf M, Meaney MJ. 2004. Epigenetic programming by maternal behavior. Nat Neurosci. 7:847–854. Wei H, He XJ, Liao CH, Wu XB, Jiang WJ, Zhang B, Zhou LB, Zhang LZ, Barron AB, Zeng ZJ. 2019. A maternal effect on queen production in honeybees. Curr Biol. 29:2208–2213.e3. 652 Behavioral Ecology Wilson AJ, Coltman DW, Pemberton JM, Overall ADJ, Byrne KA, Kruuk LEB. 2005. Maternal genetic effects set the potential for evolution in a free-living vertebrate population. J Evol Biol. 18:405–414. Wolf JB. 2003. Genetic architecture and evolutionary constraint when the environment contains genes. Proc Natl Acad Sci USA. 100:4655–4660. Wolf JB, Brodie ED III, Cheverud JM, Moore AJ, Wade MJ. 1998. Evolutionary consequences of indirect genetic effects. Trends Ecol Evol. 13:64–69. Yan Z, Martin SH, Gotzek D, Arsenault SV, Duchen P, Helleu Q , Riba- Grognuz O, Hunt BG, Salamin N, Shoemaker D, et al. 2020. Evolution of a supergene that regulates a trans-species social polymorphism. Nat Ecol Evol. 4:240–249. Zeileis A, Köll S, Graham N. 2020. Various versatile variances: an object-o- riented implementation of clustered covariances in R. J. Stat. Softw. 95:1–36.
10.1073_pnas.2301852120	INAUGURAL ARTICLE \| BIOPHYSICS AND COMPUTATIONAL BIOLOGY OPEN ACCESS Quantification of gallium cryo-FIB milling damage in biological lamellae Bronwyn A. Lucasa,1,2,3 and Nikolaus Grigorieff a,b,1 This contribution is part of the special series of Inaugural Articles by members of the National Academy of Sciences elected in 2021. Contributed by Nikolaus Grigorieff; received February 1, 2023; accepted April 20, 2023; reviewed by Jürgen M. Plitzko and Elizabeth Villa Cryogenic electron microscopy (cryo-EM) can reveal the molecular details of biologi- cal processes in their native, cellular environment at atomic resolution. However, few cells are sufficiently thin to permit imaging with cryo-EM. Thinning of frozen cells to <500 nm lamellae by focused-ion-beam (FIB) milling has enabled visualization of cellular structures with cryo-EM. FIB milling represents a significant advance over prior approaches because of its ease of use, scalability, and lack of large-scale sam- ple distortions. However, the amount of damage it causes to a thinned cell section has not yet been determined. We recently described an approach for detecting and identifying single molecules in cryo-EM images of cells using 2D template matching (2DTM). 2DTM is sensitive to small differences between a molecular model (tem- plate) and the detected structure (target). Here, we use 2DTM to demonstrate that under the standard conditions used for machining lamellae of biological samples, FIB milling introduces a layer of variable damage that extends to a depth of 60 nm from each lamella surface. This layer of damage limits the recovery of information for in situ structural biology. We find that the mechanism of FIB milling damage is distinct from radiation damage during cryo-EM imaging. By accounting for both electron scattering and FIB milling damage, we estimate that FIB milling damage with current protocols will negate the potential improvements from lamella thinning beyond 90 nm. electron cryomicroscopy \| template matching \| ribosome \| focused-ion-beam milling Cryogenic electron microscopy (cryo-EM) has enabled visualization of purified macro- molecular complexes at atomic resolution (1, 2). A more complete understanding of molecular function requires visualizing their location, structure, and interactions in the native cellular environment. The internal architecture of cells can be preserved with high fidelity by vitrification allowing for the visualization of molecules at high resolution directly in the cell (in situ) with cryo-EM (3). However, with few exceptions, cells are too thick to be electron transparent and therefore require thinning. Cryo-EM of vitreous sections (CEMOVIS) is one solution to generating thin slices of high-pressure frozen cells using a cryo-ultramicrotome (4). However, the process requires a skilled user, is difficult to automate, and introduces compression artifacts, which together have limited the widespread utility of this approach (5). Focused-ion-beam (FIB) milling is a technique in common use in materials science that has been adapted to produce thin cell sections for in situ cryo-EM under cryogenic conditions (6–8). In place of a physical ultramicrotome, a focused beam of ions, typically produced from a gallium liquid metal ion source (LMIS) or plasma, is used to sputter material above and below a thin section of the cell known as a lamella (8). FIB milling has higher throughput relative to CEMOVIS because of its ease of use, commercial avail- ability, and computational control allowing for automation of lamella production (9–11). As a result, cryo-FIB milling for lamella preparation of cells has recently seen widespread adoption and is now the predominant method for preparing cells for in situ cryo-EM (12). It has been demonstrated recently that it is possible to generate near-atomic resolution reconstructions by averaging subtomograms from vitreously frozen cells (13, 14). These successes highlight the need for a more quantitative understanding of potential sample damage introduced during FIB milling that could limit both the resolution of in situ reconstructions and the ability to accurately localize molecules in cells. Organic materials are particularly sensitive to radiation damage upon interaction with high-energy particles. Simulations of the stopping range in matter (SRIM) of ions in a glancing incidence beam at 30 keV, the typical conditions for cryo-lamella preparation for transmission electron microscopy (TEM), will implant Ga+ ions in frozen cells to a depth Significance The molecular mechanisms of biological macromolecules and their assemblies are often studied using purified material. However, the composition, conformation, and function of most macromolecules depend on their cellular context, which must be studied inside cells. Focused- ion-beam (FIB) milling enables cryogenic electron microscopy to visualize macromolecules in cells at near atomic resolution by generating thin sections of frozen cells. However, the extent of FIB milling damage to frozen cells is unknown. Here, we show that Ga+ FIB milling introduces damage to a depth of ~60 nm from each lamella surface, leading to a loss of recoverable information of up to 20% in 100 nm samples. FIB milling with Ga+ therefore presents both an opportunity and an obstacle for structural cell biology. Author contributions: B.A.L. and N.G. designed research; B.A.L. performed research; B.A.L. contributed new reagents/analytic tools; B.A.L. analyzed data; B.A.L. and N.G. interpretated results; and B.A.L. and N.G. wrote the paper. Reviewers: J.M.P., Max-Planck-Institut fur Biochemie; and E.V., University of California San Diego. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. 2Present address: Division of Biochemistry, Biophysics and Structural Biology, Department of Molecular and Cell Biology, University of California Berkeley, Berkeley, CA 94720. 3Present address: Center for Computational Biology, University of California Berkeley, Berkeley, CA 94720. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2301852120/-/DCSupplemental. Published May 22, 2023. PNAS 2023 Vol. 120 No. 23 e2301852120 https://doi.org/10.1073/pnas.2301852120 1 of 9 of 20 to 30 nm (7, 15). After accounting for removal of ~10 nm of material due to the concurrent milling action, the implantation zone is anticipated to be ~5 to 20 nm from the lamella surface (7). Cascading atomic collisions between Ga+ ions and sample atoms as the Ga+ ions imbed in the sample will introduce additional damage to an unknown depth from each lamella surface (16). Such damage introduced during FIB milling would decrease the usable volume of a lamella and could limit the resolution of in situ–determined structures. In a recent study (17), subtomogram averaging was used to generate high-resolution reconstructions of ribosomes taken from varying distances from the argon plasma FIB–milled lamella surface. To assess the argon plasma FIB damage, the B-factors affecting these reconstructions were analyzed, showing five-fold higher B-factors near the surface compared to 60 nm into the lamella. However, the B-factor analysis did not separate the contribution of the subtomo- gram alignment errors to the overall B-factors, thereby likely over- estimating the extent of FIB damage. In our study, we set out to quantify the degree and depth of FIB damage caused by the more commonly used Ga+ LMIS. We have recently described an approach, 2D template matching (2DTM) (18), to locate molecular assemblies in three dimensions with high precision in 2D cryo-EM images of unmilled cells (19, 20) and FIB-milled lamellae (21). Cross-correlation of a high-resolution template generated from a molecular model with a cryo-EM image produces a 2DTM signal-to-noise ratio (SNR) that reflects the similarity between the template and the individual target molecules in the image (18–21). In the present study, we apply 2DTM to quantify target integ- rity within FIB-milled lamellae at single-molecule resolution. We find that Ga+ FIB milling appreciably reduces target integrity to a depth of ~60 nm from the lamella surface. We find that the nature of FIB milling damage is distinct from electron radiation damage, consistent with interatomic collisions, rather than elec- tronic interactions, being primarily responsible for the damage. By comparing signal loss due to FIB milling damage to signal loss in thick samples due to inelastic electron scattering and molecular overlap, we show that recovery of structural information in 100 nm lamellae is reduced by ~20%. Results FIB Milling Introduces a Layer of Reduced Structural Integrity. A 2DTM template represents an ideal, undamaged model of the molecule to be detected. Any damage introduced during FIB milling will therefore decrease the correlation with the undamaged template, leading to a lower 2DTM SNR. Ribosomes are present at high density and relatively evenly distributed in the cytoplasm of the yeast Saccharomyces cerevisiae (21) and therefore present an ideal 2DTM target to quantify differences in target integrity. We prepared FIB-milled lamellae of S. cerevisiae cells of thickness varying from 120 nm to 260 nm (Fig. 1 A and B). In 30 images of the yeast cytoplasm from four lamellae, we located 11,030 large ribosomal subunits (LSUs) using 2DTM (Fig. 1 C and D and SI Appendix, Fig. S1 A and B). We estimated the z-coordinate of each LSU relative to the image defocus plane with 2 nm precision (Fig. 1 E and F, Materials and Methods). We found that the LSUs were located in a slab oriented at an angle of ~6 to 11° relative to the defocus plane, consistent with the milling angle relative to the grid surface (Fig. 1 C–F). The 2DTM SNRs of LSUs were noticeably lower at the edge of the lamellae than at the center and did not correlate with defocus (Fig. 1 E and F), indicating that this is unlikely to be the result of defocus estimation error. We used the tilt axis and angle estimated from the contrast transfer function (CTF) fit (21, 22), which indicates the pretilt of the lamella introduced during milling to adjust the coordinate frame to reflect the position of each LSU relative to the lamella center (Fig. 1 G and H). On average, the 2DTM SNRs were higher in the center and lower toward the surface in all lamellae examined (Fig. 1 G and H and SI Appendix, Fig. S2). The maximum 2DTM SNR decreased with increasing lamella thickness (SI Appendix, Fig. S1B) as observed previously (18–21). However, we observed a different 2DTM SNR profile as a function of z-coordinate in regions of different thicknesses. The 2DTM SNRs in ≤ ∼ 150-nm-thick lamellae increased toward the center of the lamella (e.g.: Fig. 1G), while in ≥ ∼ 150-nm-thick lamellae, they reached a plateau (e.g.: Fig. 1H). This is consistent with decreased structural integrity of LSUs close to each lamella surface. Quantification of the Damage Profile Reveals Damage up to ~60 nm from Each Lamella Surface. To assess the depth of the damage, we focused on images of 200 nm lamellae because we were able to detect targets throughout most of the volume, and both the number and 2DTM SNRs of the detected targets reached a plateau in the center, indicating that there is a zone of minimal damage. In seven images of 200 nm lamellae, we calculated the mean 2DTM SNR in bins of 10 nm from the lamella surface and divided this by the undamaged SNR ( SNRu ), defined as the mean 2DTM SNR of the targets between 90 and 100 nm from the lamella surface. Both the relative 2DTM SNR (Fig. 2A) and the number of LSUs detected (Fig. 2B) increased as a function of distance from the lamella surface. The lower number of detected LSUs at the lamella surface is likely a consequence of targets having a 2DTM SNR that falls below the chosen 2DTM SNR threshold of 7.85 at which we expect a single false positive per image (18). The low number of targets in the 10 nm bin prevented an accurate Gaussian fit (R2 = 0.8), and thus, this population was excluded from further analysis. In each of the bins >60 nm from the lamella surface (Fig. 2C), the distribution of 2DTM SNRs was Gaussian and not significantly different from the undamaged bin (t test P > 0.05, SI Appendix, Table S1). However, for each of the bins ≤ 60 nm from the lamella surface, the distribution shifts significantly (t test P < 0.0001, SI Appendix, Table S1) to the left, i.e., lower SNR values (Fig. 2C). This indicates that the structural similarity between target and template decreases closer to the lamella surface. We interpret this as a loss of target integrity due to FIB milling damage up to ~60 nm from the lamella surface. We found that the change in the mean 2DTM SNR at a par- ticular depth from the lamella surface ( d ), relative to, SNRu can be described by an exponential decay function: SN Rd SN Ru = 1 − Y 0 ⋅ e− [1] d k , where Y 0 and k are the fit and decay constants of our model. A least-squares fit gave values of Y 0 = 0.31 and k = 37.03 nm (R2 = 0.99) (Fig. 2D). Since SN Rd ∕SN Ru represents the remain- ing signal, the exponential model indicates a steep decline in dam- age in the first ~10 to 20 nm from the lamella surface, possibly explaining why few LSUs were detected in this range. The observed damage profile was absent in images of unmilled Mycoplasma pneumoniae cells, confirming that the observed pat- tern results from FIB milling and is not a result of error in the z-estimation in 2DTM (SI Appendix, Figs. S4 and S5). Mechanism of FIB Milling Damage. To characterize the mechanism of FIB milling damage, we compared its profile to the damage introduced by exposure to electrons during cryo-EM imaging. 2 of 9 https://doi.org/10.1073/pnas.2301852120 pnas.org A 120 nm lamella B 200 nm lamella C E 1500 1000 500 0 ) Å ( Z -500 -1000 -1500 G R N S M T D 2 20 18 16 14 12 10 8 369 LSU 1.4 1.2 1.0 0.8 2222222 0000000000000000000000000000 2000 2222000000000000 0020 22 00 0022 00 00 44444400004040040400000000000000000000 4000 000000000000 4400000000000 0000000000 440000000 00000000 040 00 6 6000 8000 corrected Y (Å) R e a l t i v e 2 D T M S N R D F 2000 1000 732 LSU ) Å ( Z 0 00000000000 00000000000000000000060000000660000066000006666666000000000000000000000000000000000000000000000000000000000000000000000000000066666666666666666666 6000 6 00000 8000 corrected Y (Å) -1000 -2000 H R N S M T D 2 20 18 16 14 12 10 8 1.4 1.2 1.0 0.8 R e a l t i v e 2 D T M S N R -1250 -1000 -750 -500 -250 0 250 500 750 1000 1250 -1250 -1000 -750 -500 -250 0 250 500 750 1000 1250 Lamella Z (Å) Lamella Z (Å) Fig. 1. Visualization of yeast cytoplasmic ribosomes in 3D with 2DTM. (A) An electron micrograph of the yeast cytoplasm in a 120-nm region of a lamella. Scale bars in (A and B) represent 50 nm. (B) As in (A), showing a 200-nm lamella. (C) Significant LSUs located in 3D in the image in (A) with 2DTM. (D) As in (C), showing the results for the image in (B). (E) Scatterplot showing a side view of the LSUs in (A). The color coding indicates the 2DTM SNR of each significant detection relative to the mean 2DTM SNR in each image. The z-coordinate represents the position of each target relative to the microscope defocus plane. (F) As in (E), showing the results from the image in (B). (G) Scatterplot showing the 2DTM SNR of each detected LSU in the image in (A), as a function of z-coordinate relative to the center of the lamella. (H) As in (G), showing the z-coordinate relative to the center of the lamella of each LSU detected in the image shown in (B). Cryo-EM imaging causes radiation damage, introducing differences between the template and the target structure that are more pronounced at high spatial frequencies (23). To measure radiation damage, we generated a series of images with different exposures by including different numbers of frames from the original movie in the summed image. Using the locations and orientations identified with 2DTM using a high-resolution template as above, we sought to calculate the contribution of different spatial frequencies to the 2DTM SNR. To achieve this, we generated a series of low-pass filtered templates with a sharp cutoff at different spatial frequencies and calculated the change in the 2DTM SNR of each identified LSU relative to the original high-resolution template as a function of electron exposure relative to a 20 electrons/Å2 exposure (Fig. 3 A and B). We find that the 2DTM SNR of templates low-pass filtered PNAS 2023 Vol. 120 No. 23 e2301852120 https://doi.org/10.1073/pnas.2301852120 3 of 9 A u R N S / R N S 1.4 1.2 1.0 0.8 0.6 C 100 y c n e u q e r F 80 60 40 20 0 500 400 300 200 100 0 0 20 40 60 Depth (nm) 80 100 B s t e g r a t d e t c e t e d f o r e b m u N D 10 20 30 40 50 60 70 80 90 100 Depth (nm) 60 50 40 30 20 10 u R N S / R N S 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.6 0.8 1.0 SNR/SNRu 1.2 1.4 0 20 40 60 Depth (nm) 80 100 Fig. 2. The number and 2DTM SNR values of detected LSUs increase as a function of distance from the lamella surface. (A) Boxplot showing the 2DTM SNR of LSUs at the indicated lamella depths, relative to the undamaged SNR ( SNRu ) in each image from 200 nm lamellae. Boxes represent the interquartile range (IQR), middle lines indicate the median, whiskers represent 1.5× IQR, and dots represent values outside of this range. (B) Scatterplot showing the number of detected targets in the indicated z-coordinate bins. (C) Gaussian fits to the distribution of 2DTM SNRs for LSUs identified in z-coordinate bins of 10 nm. Red indicates populations with means significantly different from the mean in the center of the lamella. Blue indicates that the mean in a bin is not significantly different from the mean in the lamella center. Fitting statistics are indicated in SI Appendix, Table S1. (D) Scatterplot showing the mean change in 2DTM SNR relative to SNRu at the indicated depths relative to the lamella surface estimated from the Gaussian fits in (C). The line shows the exponential fit (R2 = 0.99). Error bars indicate the SD from the Gaussian fit. to between 1/10 and 1/7 Å−1 increases with increasing exposure. The 2DTM SNRs of templates low-pass filtered with a cutoff at higher resolutions begin to decrease with increasing exposure (Fig. 3 A and B). Templates filtered to 1/5 Å−1 have a maximum 2DTM SNR at 32 electrons/Å2, while templates filtered to 1/3 Å−1 have a maximum 2DTM SNR at 28 electrons/Å2 (Fig. 3B). To estimate the extent of FIB milling damage on different spatial frequencies, we binned detected targets by lamella depth and calcu- lated SNRd ∕SN Ru . We found that for templates filtered to < 1/5 Å−1, SNRd ∕SN Ru fluctuated for targets detected further from the lamella center. This is likely due to differences in the defocus position that result in some of the targets having weak contrast (contrast transfer function close to zero) and therefore not contributing meaningful signal at different spatial frequencies relative to targets in the center of the lamella. For templates filtered to > 1/5 Å−1, the profile was similar between the different bins and approximately constant across spatial frequencies (Fig. 3 C and D). This is consistent with a model in which the FIB-damaged targets have effectively lost a fraction of their structure, compared to undamaged targets, possibly due to displacement of a subset of atoms by colliding ions. Radiation damage of nucleic acids has been well documented with one of the most labile bonds being the phosphodiester bond in the nucleic acid backbone (24) (Fig. 3E). We observed an accel- erated loss of signal from phosphorous atoms relative to the aver- age loss of signal for the whole template as a function of electron exposure (Fig. 3F). This is consistent with the phosphorous atoms being more mobile due to breakage of phosphodiester linkages in response to electron exposure. We did not observe a consistent difference in the accelerated loss of signal from phosphorous in the lamella z-coordinate groups (Fig. 3F). This indicates that the mechanism for FIB milling damage is distinct from the radiation damage observed during cryo-EM imaging. Sample Thickness Limits 2DTM SNR More Than FIB Milling Damage. Above we report that using the most common protocol for cryo-lamella generation by LMIS Ga+ FIB milling introduces a variable layer of damage up to 60 nm from each lamella surface. Lamellae for cryo-EM and electron cryotomography (cryo-ET) are typically milled to 100 to 300 nm, meaning that the damaged layer comprises 50 to 100% of the volume. Thicker lamellae will have a lower proportion of damaged particles. However, thicker lamellae will also suffer from signal loss due to the increased loss of electrons due to inelastic scattering and scattering outside the aperture, as well as the increased number of other molecules in the sample contributing to the background in the images. For a target inside a cell, the loss of 2DTM SNR with increasing thickness has been estimated as (19): SN Rt SN R0 = e−t ∕𝜆SNR , [2] where t denotes the sample thickness, SN R0 is the 2DTM SNR in the limit of an infinitesimally thin sample, and the decay constant 𝜆SNR = 426 nm. Optimal milling conditions for high-resolution imaging of FIB-milled lamellae will therefore need to strike a balance between lamella thickness and FIB damage. To assess the relative impact of these two factors on target detec- tion with 2DTM, we plotted the proportional loss in signal due to 4 of 9 https://doi.org/10.1073/pnas.2301852120 pnas.org A 1.08 Exposure (e/Å2) 20 B e 0 2 R N S / R N S M T D 2 22 24 26 28 30 32 34 36 0.0 0.1 0.2 0.3 0.4 0.5 Low-pass Filter (1/Å) Depth (nm) D 1.06 e 0 2 1.04 1.02 1.00 0.98 1.05 1.00 0.95 0.90 0.85 0.80 R N S / R N S M T D 2 C u R N S R N S / E 0.0 0.1 0.2 0.3 0.4 0.5 Low-pass Filter 1/Å F R N S f o n o i t r o p o r P s u o r o h p s o h P m o r f 0.04 0.03 0.02 0.01 0.00 15 1.08 1.06 1.04 1.02 1.00 0.98 1.05 1.00 0.95 0.90 0.85 0.80 2.12 Å 3 Å 5 Å 6.25 Å 10 Å 25 35 20 Exposure (electrons/Å2) 30 40 2.12 Å 3 Å 5 Å 6.25 Å 10 Å 0 20 40 60 Depth (nm) 80 100 Depth (nm) u R N S R N S / 20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100 0 10 20 30 40 Exposure (electrons/Å2) Fig. 3. The mechanism of FIB milling damage is distinct from radiation damage during cryo-EM imaging. (A) Plot showing the change in 2DTM SNR with the template low-pass filtered to the indicated spatial frequency in images collected with the indicated number of electrons/Å2 relative to images collected with 20 electrons/Å2. (B) Plot showing the change in 2DTM SNR as a function of electron exposure of templates low-pass filtered to the indicated spatial frequency. (C) As in (A), showing the change in the 2DTM SNR in the indicated lamella z-coordinate bins relative to the SNR in the undamaged bin ( SNRu ). (D) Plot showing the change in 2DTM SNR for templates low-pass filtered to the indicated spatial frequencies as a function of lamella z-coordinate bins. (E) Diagram showing a segment of an RNA strand of two nucleotides. The blue circle designates the phosphate; the two red arrows indicate the location of the backbone phosphodiester bonds. (F) Plot showing the relative contribution of template phosphorous atoms to the 2DTM SNR relative to the full-length template at the indicated exposure without dose weighting, calculated using Eq. 6. electrons lost in the image and background (Fig. 4, red curve). We can estimate the average loss of 2DTM SNR, SN Rd ∕SN Ru , due to FIB milling damage from the product of the loss (Eq. 1) from both surfaces: SN Rd SN Ru 1 − ed −t ∕k ( 1 − e−d ∕k 𝛿d . ) 1 t ∫ [3] = ) 0 ( ⋅ t Combining these two sources of signal loss gives the expected overall 2DTM SNR as a function of sample thickness (Fig. 4, black curve): SN Rd SN Ru = (∫ t 0 ( 1 − e−d ∕k ⋅ ) 1 − ed −t ∕k t ( 𝛿d ) e−t ∕𝜆SNR ) [4] . This model predicts that in samples thicker than ~90 nm, the relative loss in the signal due to the loss of electrons contributing to the image, as well as molecular overlap, is greater than the relative change due to FIB milling damage (Fig. 4A). In lamellae thinner than 90 nm, however, FIB milling damage will dominate and negate any benefit from further thinning. The difference between the expected signal loss given by Eq. 4) and signal loss solely from lost electrons and molecular overlap represents the potential gain if FIB milling damage could be avoided. Without FIB damage, the potential improvement in 2DTM SNR would be between ~10% in 200 nm lamellae and ~20% in 100 nm lamellae (Fig. 4). The model in Eq. 4) ignores the variable degree of damage expected to occur across LSUs that we used as probes to measure damage and that have a radius of ~15 nm. However, the resulting error in the measured damage constant k (Eq. 1) is PNAS 2023 Vol. 120 No. 23 e2301852120 https://doi.org/10.1073/pnas.2301852120 5 of 9 R N S M T D 2 e v i t a l e R 1.0 0.8 0.6 0.4 0 50 FIB damage Electron scattering Expected relative signal 100 150 Lamella thickness (nm) 200 250 300 Fig. 4. Signal loss due to increased inelastic and multiple electron scattering in thicker samples outweighs the effect of FIB damage on LSU 2DTM SNRs. Plot showing the expected signal recovery in lamellae of indicated thickness as a function of signal loss due to electron scattering (red curve), FIB damage (blue curve), and their product (black curve). expected to be small since k (~37 nm) significantly exceeds the LSU radius, and hence, the variable damage can be approximated by an average damage uniformly distributed across the target. We also expect that the number of detected targets will be reduced by FIB milling damage. The number of detected LSUs was variable across lamellae, likely due to biological differences in local ribosome concentration. In undamaged parts of a subset of 200-nm-thick lamellae, we identified ~425 LSU in z-coordinate intervals of 10 nm. If this density were maintained throughout the lamella, we would expect to detect ~40% more targets in these lamellae. We conclude that FIB damage reduces the number and integrity of detected targets but that signal loss due to electrons lost to the image, as well as background from overlapping molecules, is a greater limiting factor for target detection and characterization with 2DTM than FIB milling damage in lamellae thicker than ~90 nm. These data agree with other empirical observations that thinner lamellae are optimal for recovery of structural information and generation of high-resolution reconstructions. It may be possible to restore signal in images otherwise lost to inelastic scattering using Cc-correctors (25). This would be par- ticularly impactful for thick samples such as FIB-milled cellular lamellae. With the use of a Cc-corrector, the signal loss in thick samples would be reduced, and FIB milling damage may become the main limiting factor for in situ structural biology. Discussion Ga+ LMIS FIB milling is currently the preferred method for gen- erating thin, electron-transparent cell sections for in situ cryo-EM. We use 2DTM to evaluate the structural integrity of macromol- ecules in FIB-milled lamellae and provide evidence that FIB- milled lamellae have a region of structural damage to a depth of up to 60 nm from the lamella surface. By evaluating the relative similarity of a target molecule to a template model, 2DTM pro- vides a sensitive, highly position-specific, single-particle evaluation of sample integrity. 2DTM SNRs Provide a Readout of Sample Integrity and Image Quality. Changes to the 2DTM SNR provide a readout of the relative similarity of a target molecule to a given template. We have previously shown that relative 2DTM SNRs discriminate between molecular states and can reveal target identity (20, 21). In this study, we show that changes in 2DTM SNRs can also reflect damage introduced during FIB milling and radiation damage introduced during cryo-EM imaging. A previous attempt to measure FIB damage has relied on visual changes in the sample near the surface. These changes are difficult to quantify in terms of damage, and they could in part be caused by other mechanisms such as ice accumulation after milling (26). Argon plasma FIB damage has been assessed by comparing subtomogram averages of particles from different distances from the lamella surface and estimating their B-factors (17), which may overestimate the amount of damage due to unrelated contributions to the measured B-factors. The 2DTM SNR represents an alternative, more quantitative metric to assess sample integrity. 2DTM SNRs have also been used as a metric to assess image quality (27) and the fidelity of simulations (28). 2DTM, therefore, represents a sensitive, quantitative, and versatile method to meas- ure the dependence of data quality on sample preparation and data collection strategies, as well as new hardware technologies and image processing pipelines. Tool and method developers could use standard datasets and 2DTM to rapidly and quantitatively assess how any changes to a pipeline affect data quality. Estimating Errors in z-Coordinates and Thickness. The z-coordinates of each LSU were determined by modulating the template with a CTF corresponding to a range of defoci and identifying the defocus at which the cross-correlation with the 2D projection image was maximized (18). This quantification relies on an accurate estimate of defocus. The error in the z-coordinates determined this way was estimated to be about 60 Å (20). However, it is unlikely that these errors explain the observed decrease in 2DTM SNRs of LSUs near the edge of the lamellae because 1) the reduction in 2DTM SNRs correlates strongly with the z-coordinate within the lamella, and 2), we did not observe a consistent decrease in the number of detected LSUs (SI Appendix, Fig. S4A) or their 2DTM SNRs (SI Appendix, Fig. S4B) as a function of z-coordinate in images of unmilled M. pneumoniae cells (20). It remains possible that differences in cell density, residual motion (20) or differences in the size and resolution of the LSUs could contribute to the differences in the profile of 2DTM SNRs as a function of z-coordinate. In the future, it may be informative to examine the 2DTM SNRs of ribosomes and other complexes in other thin samples such as the extensions of mammalian cells. Undulations at the lamella surface caused by curtaining or other milling artifacts could contribute to the reduced number of ribo- somes detected near the lamella surface. We aimed to minimize the effect of curtaining in our analysis by calculating the lamella thickness in 120 × 120 pixel (127.2 × 127.2 Å) patches across an image and limiting our analysis to images with a thickness stand- ard deviation (SD) of less than 20 nm. The curtaining on the remaining lamellae cannot account for the reduced particle integ- rity toward the lamella surface. Possible Mechanisms of FIB Milling Damage. We find evidence for FIB milling damage consistent with an exponential decay of the amount of damage as a function of distance from the lamella surface, as measured by the 2DTM SNR. Unlike electron radiation damage, FIB damage 1) causes a reduction in the total signal and does not preferentially affect higher spatial frequencies contributing to the 2DTM SNR calculation, and 2), unlike electron beam radiation damage, it does not preferentially affect the phosphodiester bond in the RNA backbone. This suggests that different mechanisms are responsible for the damage caused by high-energy electrons and ions. 6 of 9 https://doi.org/10.1073/pnas.2301852120 pnas.org At the energy ranges used for FIB milling, the interactions between the bombarding ion and sample atoms can be modeled as a cascade of atom displacements resulting from the transfer of momentum from the incident Ga+ ions to the sample atoms (16). Atoms involved in the cascade will be displaced, while the position of other atoms will not change. This is consistent with our obser- vation that FIB damage decreases the LSU target signal overall without changing the relative contribution from different spatial frequencies. Further study is required to test this hypothesis and investigate the mechanism of FIB milling damage in more detail. SRIM simulations predict implantation of Ga+ up to ~25 nm into the sample (7, 15). This implies that the damage deeper in the sample is caused by secondary effects, possibly reflecting dis- placed sample atoms that were part of the collision cascade. We observe a different pattern of particles within 20 nm of the lamella surface (Fig. 3 C and D). One possible explanation is that implanted Ga+ ions cause additional damage. However, SRIM simulations cannot account for the full intensity profile of a Ga+ beam, and poorly match with experiment especially at low beam currents (29). Moreover, the use of a protective organoplatinum layer during FIB milling, as done in our experiments, will further change the effective profile of the beam acting on the sample (30). Further work is required to connect the quantification of particle integrity with the implantation of Ga+ ions during biological lamella preparation. Implications for Generating High-Resolution Reconstructions from FIB-Milled Samples. We have shown that particles on the edge of a lamella have reduced structural integrity relative to particles near the center of the lamella (Figs. 1 and 2). We found that FIB milling damage reduces the total 2DTM SNR. At distances >20 nm from the lamella surface the rate of signal loss is similar at different spatial frequencies, in contrast to radiation damage during cryo-EM imaging (Fig. 3). The practical implication of this finding is that particles >30 nm from the lamella surface can be included during subtomogram averaging without negatively affecting the resolution of the reconstruction, provided that they can be accurately aligned. We also predict that more particles will be required relative to unmilled samples. This is consistent with the observation that more particles <30 nm from the lamella surface are required to achieve the same resolution relative to >30 nm from the lamella surface from argon plasma FIB–milled lamellae (17). The depth at which particle quality is noticeably poorer is consistent between gallium and argon FIB–milled samples. This suggests that argon plasma FIB milling is not a solution to mitigate the damage introduced during gallium FIB milling. Due to the small number of particles detected within 10 nm of the lamella surface, these particles were not examined in more detail. Since ribosomes are ~25 nm in diameter, it is likely that these particles are more severely damaged compared to particles further away from the surface. 2DTM relies on high-resolution signal and therefore excludes more severely damaged particles that may be included using a low-resolution template matching approach, such as 3D template matching used typically to identify particles for subtomogram averaging. We therefore advise exclud- ing particles detected within 10 nm of the lamella surface. Alternate Methods for the Preparation of Thin Cell Sections. FIB damage reduces both the number of detected targets and the available signal per target. However, the damaged volume still contributes to the sample thickness, reducing the usable signal by 10 to 20% in lamellae of typical thicknesses (Fig. 4). Therefore, it would be advantageous to explore other strategies for cell thinning. Plasma FIBs allow different ions to be used for milling, and this may change the damage profile (31). Larger atoms such as xenon will have a higher sputtering yield and may result in reduced lamella damage, as has been demonstrated for milling of silicon samples (32, 33). The 2DTM-based approach described here provides a straightforward way to quantify the relative damaging effects of dif- ferent ion species by generating curves as shown in Fig. 4. CEMOVIS generates thin sections using a diamond knife rather than high-energy ions and would therefore not introduce radiation damage (4). It is unclear how the large-scale compression artifacts introduced by this method affect particle integrity (5). CEMOVIS has the additional benefit of being able to generate multiple sections per cell and thereby enable serial imaging of larger cell volumes. If the compression artifacts are unevenly dis- tributed throughout a section, leaving some regions undistorted, automation could make CEMOVIS a viable strategy for structural cell biology in the future. To retain the benefits of fast, reliable, high-throughput lamella generation with cryo-FIB milling, strategies to remove the damaged layer should be explored. In the Ga+ FIB, there are two properties that are easily tunable, the beam current, which affects the rate of ions to which the sample is exposed, and the energy of the ion beam. Lowering the current and the total exposure is unlikely to decrease the damage layer when milling thick samples because 1) there will be a minimum number of collisions required to sputter a sufficiently large volume to generate a lamella and 2) because the total exposure will greatly exceed the steady-state dose at which implantation of ions into the sample and sputtering are at equilibrium, such that any additional exposure will not cause additional damage. Consistently, we observe damage throughout the lamella and do not observe dra- matic increases in the damage layer close to the milling edge or when the organo-Pt layer is compromised relative to images collected fur- ther from the milling edge, which have been exposed to a lower dose (SI Appendix, Fig. S6). Alternately polishing the final ~50 nm from each lamella surface with a low energy (~5 kV) beam, which has the advantage of being easily implementable using the current configu- ration of most cryo-FIB-SEMs, would be expected to decrease the damage layer. Materials and Methods Yeast Culture and Freezing. S. cerevisiae strain BY4741 (ATCC) colonies were inoculated in 20 mL of yeast extract–peptone–dextrose (YPD) media, diluted 1/5, and grown overnight at 30 °C with shaking to mid-log phase. The cells were then diluted to 10,000 cells/mL, treated with 10 µg/mL cycloheximide (Sigma) for 10 min at 30 °C with shaking. 3 µL were applied to a 2/1 or 2/2 Quantifoil 200 mesh SiO2 Cu grid, allowed to rest for 15 s, back side blotted for 8 s at 27 °C, 95% humidity, and plunge-frozen in liquid ethane at –184 °C using a Leica EM GP2 plunger. Frozen grids were stored in liquid nitrogen until FIB milled. FIB Milling. Grids were transferred to an Aquilos 2 cryo-FIB/SEM, sputter coated with metallic Pt for 10 s and then coated with organo-Pt for 30 s and milled in a series of sequential milling steps using a 30 kV Ga+ LMIS beam using the follow- ing protocol: rough milling 1: 0.1 nA rough milling 2: 50 pA lamella polishing: 10 pA at a stage tilt of 15° (milling angle of 8°) or 18° (milling angle of 11°). Over and under tilt of 1° was used to generate lamellae of relatively consistent thickness during the 50 pA milling steps. No SEM imaging was performed after the milling started to prevent introducing additional damage. Cryo-EM Data Collection and Image Processing. Cryo-EM data were collected following the protocol described in ref. 21 using a Thermo Fisher Krios 300 kV electron microscope equipped with a Gatan K3 direct detector and Gatan energy filter with a slit width of 20 eV at a nominal magnification of 81,000× (pixel size of 1.06 Å2) and a 100-µm objective aperture. Movies were collected at an exposure rate of 1 e−/Å2/frame to a total dose of 50 e−/Å2 (dataset 1) with correlated double sampling using the microscope control software SerialEM (34). PNAS 2023 Vol. 120 No. 23 e2301852120 https://doi.org/10.1073/pnas.2301852120 7 of 9 Images were processed as described previously (21). Briefly, movie frames were aligned using the program unblur (35) in the cisTEM graphical user interface (GUI) (36) with or without dose weighting using the default param- eters where indicated in the text. Defocus, astigmatism, and sample pretilt were estimated using a modified version of CTFFIND4 (20, 22) in the cisTEM GUI (36). Images of the cytoplasm were identified visually for further analysis. Images visually containing organelles were excluded. Images of 3D densities and 2DTM results were prepared in ChimeraX (37). 2DTM. The atomic coordinates corresponding to the yeast LSU from the Protein Data Bank (PDB), code 6Q8Y (38) were used to generate a 3D volume using the cisTEM program simulate (28) and custom scripts as in ref. (21). 2DTM was performed using the program match_template (20) in the cisTEM GUI (36) using an in-plane search step of 1.5° and an out-of-plane search step of 2.5°. Significant targets were defined as described in ref. (20) and based on the significance criterion described in ref. (18). The coordinates were refined using the program refine_template (20) in rotational steps of 0.1° and a defocus range of 200 Å with a 20 Å step (2 nm z-precision). The template volume was placed in the identified locations and orientations using the program make_template_result (20) and visualized with ChimeraX (37). To generate the results in Fig. 3 A–D, we applied a series of sharp low-pass fil- ters in steps of 0.01 Å−1 to the template using the e2proc3d.py function in EMAN2 (39). We used the locations and orientations from the refined 2DTM search with the full-length template to recalculate the 2DTM SNR with each modified template using the program refine_template (20) by keeping the positions and orien- tations fixed. The normalized cross-correlation was determined by dividing the SNR calculated with each low-pass filtered template to the SNR of the full-length template for each target. Calculation of Pretilt and Coordinate Transform. We used Python scripts to extract the rotation angle and pretilt from the cisTEM (36) database gener- ated using the tilt-enabled version of the program CTFFIND4 (21, 22), perform a coordinate transform to convert the 2DTM coordinates to the lamella coordinate frame, and plot the 2DTM SNR as a function of lamella z-coordinate. Calculation of Sample Thickness and Depth. We estimated the lamella thickness per image by first summing the movie frames without dose weighting using the EMAN2 program, alignframes (39), and then calculating the average intensity of a sliding box of 120 × 120 pixels ( I ) relative to the same area of an image collected over vacuum ( Io ). We then used the mean free path for electron scattering ( 𝜆 ) of 283 nm (19) to estimate the local sample thickness ti using the Beer-Lambert law (40): The sample thickness was determined by taking the mean across the image. Only images with a SD of <20 nm across the image were included for estimation of the damage profile (Fig. 2B). The depth of each LSU relative to the lamella surface was calculated by assuming that the LSUs are evenly distributed in z and defining the median lamella z-coordinate as the lamella center (e.g.: Fig. 1 G and H and SI Appendix, Fig. S2). Measuring Change in Signal with Electron Exposure. We compared the change in the 2DTM SNR of each individual LSU as a function of electron exposure at different positions relative to the edge of the lamella in bins of 10 nm. We used the locations and orientations of LSUs identified in dose-filtered images exposed to 50 e−/Å2 to assess the correlation at the same locations and orientations in different numbers of unweighted frames corresponding to total exposures of 8-36 e−/Å2. To calculate the relative contribution of phosphorous to the 2DTM SNR, all phosphorous atoms in the PDB file were deleted, and a template was generated as described above without recentering so that it aligned with the full-length template. We used the locations and orientations from the refined 2DTM search with the full-length template for each exposure to calculate the 2DTM SNR with the template lacking phosphorous ( SNRΔP ) using the program refine_template (20) and keeping the positions and orientations fixed. The relative contribution of phosphorous atoms to the 2DTM SNR ( SNRP ) at each exposure was calculated using the following equation: SNRP = 1 − SNR ΔP SNRFL . [6] Data, Materials, and Software Availability. Cryo-EM images data have been deposited in Electron Microscopy Public Image Archive (EMPIAR) data- base with accession number EMPIAR-11544 (https://www.ebi.ac.uk/empiar/ EMPIAR-11544/) (41). ACKNOWLEDGMENTS. We thank Johannes Elferich, Ximena Zottig, and other members of the Grigorieff lab (University of Massachusetts Chan Medical School), Russo lab (Medical Research Council Laboratory of Molecular Biology), and de Marco lab (Monash University) for helpful discussions. We are also grateful for the use of and support from the cryo-EM facilities at Janelia Research Campus and UMass Chan Medical School. B.A.L. and N.G. gratefully acknowledge funding from the Chan Zuckerberg Initiative, grant #2021-234617 (5022). ti = − ln ( I Io )𝜆. [5] Author affiliations: aRNA Therapeutics Institute, University of Massachusetts Chan Medical School, Worcester, MA 01605; and bHHMI, University of Massachusetts Chan Medical School, Worcester, MA 01605 1. K. M. Yip, N. Fischer, E. Paknia, A. Chari, H. Stark, Atomic-resolution protein structure determination by cryo-EM. Nature 587, 157–161 (2020). 2. T. Nakane et al., Single-particle cryo-EM at atomic resolution. Nature 587, 152–156 (2020). 3. W. Baumeister, R. Grimm, J. Walz, Electron tomography of molecules and cells. Trends Cell Biol. 9, 4. 5. 81–85 (1999). A. Al-Amoudi, L. P. O. Norlen, J. Dubochet, Cryo-electron microscopy of vitreous sections of native biological cells and tissues. J. Struct. Biol. 148, 131–135 (2004). A. Al-Amoudi, D. Studer, J. Dubochet, Cutting artefacts and cutting process in vitreous sections for cryo-electron microscopy. J. Struct. Biol. 150, 109–121 (2005). 6. M. Marko, C. Hsieh, W. Moberlychan, C. A. Mannella, J. Frank, Focused ion beam milling of vitreous water: Prospects for an alternative to cryo-ultramicrotomy of frozen-hydrated biological samples. J. Microsc. 222, 42–47 (2006). 7. M. Marko, C. Hsieh, R. Schalek, J. Frank, C. Mannella, Focused-ion-beam thinning of frozen-hydrated 8. 9. biological specimens for cryo-electron microscopy. Nat. Methods 4, 215–217 (2007). A. Rigort et al., Focused ion beam micromachining of eukaryotic cells for cryoelectron tomography. Proc. Natl. Acad. Sci. U.S.A. 109, 4449–4454 (2012). G. Buckley et al., Automated cryo-lamella preparation for high-throughput in-situ structural biology. J. Struct. Biol. 210, 107488 (2020). 14. B. A. Himes, P. Zhang, emClarity: Software for high-resolution cryo-electron tomography and subtomogram averaging. Nat. Methods 15, 955–961 (2018). 15. Y. Fukuda, A. Leis, A. Rigort, “Preparation of vitrified cells for TEM by cryo-FIB microscopy” in Biological Field Emission Scanning Electron Microscopy (John Wiley & Sons, 2019), pp. 415–438 16. L. A. Giannuzzi, B. I. Prenitzer, B. W. Kempshall, “Ion - solid interactions” in Introduction to Focused Ion Beams (2005), pp. 13–52. 17. C. Berger et al., Plasma FIB milling for the determination of structures in situ. Nat. Commun. 114, 1–12 (2023). 18. J. P. Rickgauer, N. Grigorieff, W. Denk, Single-protein detection in crowded molecular environments in cryo-EM images. Elife 6, e25648 (2017). 19. J. P. Rickgauer, H. Choi, J. Lippincott-Schwartz, W. Denk, Label-free single-instance protein detection in vitrified cells. bioRxiv [Preprint] (2020). https://doi.org/10.1101/2020.04.22.053868 (Accessed 24 April 2020). 20. B. A. Lucas et al., Locating macromolecular assemblies in cells by 2D template matching with cisTEM. Elife 10, e68946 (2021). 21. B. A. Lucas, K. Zhang, S. Loerch, N. Grigorieff, In situ single particle classification reveals distinct 60S maturation intermediates in cells. Elife 11, e68946 (2022). 22. A. Rohou, N. Grigorieff, CTFFIND4: Fast and accurate defocus estimation from electron micrographs. 10. T. Zachs et al., Fully automated, sequential focused ion beam milling for cryo-electron tomography. J. Struct. Biol. 192, 216–221 (2015). Elife 9, e52286 (2020). 23. L. A. Baker, J. L. Rubinstein, Radiation damage in electron cryomicroscopy. Methods Enzymol. 481, 11. S. Klumpe et al., A modular platform for automated cryo-FIB workflows. Elife 10, 70506 (2021). 12. F. R. Wagner et al., Preparing samples from whole cells using focused-ion-beam milling for cryo- 371–388 (2010). 24. J. F. Ward, Molecular mechanisms of radiation-induced damage to nucleic acids. Adv. Radiat. Biol. 5, electron tomography. Nat. Protocols 615, 2041–2070 (2020). 181–239 (1975). 13. D. Tegunov, L. Xue, C. Dienemann, P. Cramer, J. Mahamid, Multi-particle cryo-EM refinement with M 25. J. L. Dickerson, C. J. Russo, Phase contrast imaging with inelastically scattered electrons from any visualizes ribosome-antibiotic complex at 3.5 Å in cells. Nat. Methods 18, 186–193 (2021). layer of a thick specimen. Ultramicroscopy 237, 113511 (2022). 8 of 9 https://doi.org/10.1073/pnas.2301852120 pnas.org 26. C. J. Russo, J. L. Dickerson, K. Naydenova, Cryomicroscopy in situ: What is the smallest 34. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of molecule that can be directly identified without labels in a cell? Faraday Discuss 240, 277–302 (2022). specimen movements. J. Struct. Biol. 152, 36–51 (2005). 35. T. Grant, N. Grigorieff, Measuring the optimal exposure for single particle cryo-EM using a 2.6 Å 27. J. Elferich, G. Schiroli, D. T. Scadden, N. Grigorieff, Defocus corrected large area cryo-EM (DeCo-LACE) reconstruction of rotavirus VP6. Elife 4, e06980 (2015). for label-free detection of molecules across entire cell sections. Elife 11, e80980 (2022). 28. B. Himes, N. Grigorieff, Cryo-TEM simulations of amorphous radiation-sensitive samples using 36. T. Grant, A. Rohou, N. Grigorieff, CisTEM, user-friendly software for single-particle image processing. Elife 7, e35383 (2018). multislice wave propagation. IUCrJ 8, 943–953 (2021). 37. E. F. Pettersen et al., UCSF ChimeraX: Structure visualization for researchers, educators, and 29. Y. Greenzweig, Y. Drezner, S. Tan, R. H. Livengood, A. Raveh, Current density profile characterization developers. Protein Sci. 30, 70 (2021). and analysis method for focused ion beam. Microelectron. Eng. 155, 19–24 (2016). 38. P. Tesina et al., Structure of the 80S ribosome–Xrn1 nuclease complex. Nat. Struct. Mol. Biol. 26, 30. M. Schaffer et al., Optimized cryo-focused ion beam sample preparation aimed at in situ structural 275–280 (2019). studies of membrane proteins. J. Struct. Biol 197, 73–82 (2017). 39. G. Tang et al., EMAN2: An extensible image processing suite for electron microscopy. J. Struct. Biol. 31. G. Sergey et al., Oxygen plasma focused ion beam scanning electron microscopy for biological 157, 38–46 (2007). samples. bioRxiv [Preprint] (2018). https://doi.org/10.1101/457820 (Accessed 31 October 2018). 40. W. J. Rice et al., Routine determination of ice thickness for cryo-EM grids. J. Struct. Biol. 204, 38 32. L. A. Giannuzzi, N. S. Smith, TEM specimen preparation with plasma FIB Xe + Ions Microsc. (2018). Microanal. 17, 2011 (2011). 33. R. Kelley et al., Xe + FIB milling and measurement of amorphous silicon damage. Microsc. Microanal. 19, 862–863 (2013). 41. A. L. Bronwyn, N. Grigorieff, Quantification of gallium cryo-FIB milling damage in biological lamella. Electron Microscopy Public Image Archive. https://www.ebi.ac.uk/empiar/EMPIAR-11544/. Deposited 7 May 2023. PNAS 2023 Vol. 120 No. 23 e2301852120 https://doi.org/10.1073/pnas.2301852120 9 of 9
10.1093_pnasnexus_pgad113	PNAS Nexus, 2023, 2, 1–13 https://doi.org/10.1093/pnasnexus/pgad113 Advance access publication 30 March 2023 Research Report Depicting pseudotime-lagged causality across single-cell trajectories for accurate gene-regulatory inference Caleb C. Reagor a,b,c,, Nicolas Velez-Angel a and A. J. Hudspeth a aHoward Hughes Medical Institute and Laboratory of Sensory Neuroscience, The Rockefeller University, New York, NY 10065, USA bTri-Institutional PhD Program in Computational Biology and Medicine, New York, NY 10065, USA cPresent address: 1230 York Avenue, Campus Box 314, New York, NY 10065, USA To whom correspondence should be addressed: Email: [email protected] Edited By: Shibu Yooseph Abstract Identifying the causal interactions in gene-regulatory networks requires an accurate understanding of the time-lagged relationships between transcription factors and their target genes. Here we describe DELAY (short for Depicting Lagged Causality), a convolutional neural network for the inference of gene-regulatory relationships across pseudotime-ordered single-cell trajectories. We show that combining supervised deep learning with joint probability matrices of pseudotime-lagged trajectories allows the network to overcome important limitations of ordinary Granger causality-based methods, for example, the inability to infer cyclic relationships such as feedback loops. Our network outperforms several common methods for inferring gene regulation and, when given partial ground- truth labels, predicts novel regulatory networks from single-cell RNA sequencing (scRNA-seq) and single-cell ATAC sequencing (scATAC-seq) data sets. To validate this approach, we used DELAY to identify important genes and modules in the regulatory network of auditory hair cells, as well as likely DNA-binding partners for two hair cell cofactors (Hist1h1c and Ccnd1) and a novel binding sequence for the hair cell-specific transcription factor Fiz1. We provide an easy-to-use implementation of DELAY under an open- source license at https://github.com/calebclayreagor/DELAY. Keywords: gene-regulatory inference, gene-regulatory networks, deep learning, single-cell omics, auditory hair cells Significance Statement The sequencing of genes expressed in single cells provides detailed information about the developmental programs that define the identities and states of various cell types, but few computational methods can use the dynamic information encoded in these repre- sentations to identify causal mechanisms. By exploiting advances in machine learning, we develop a deep neural network that learns from temporal features of gene regulation to identify direct regulatory interactions between transcription factors and their target genes. Our method provides mechanistic insights into the development and function of cells by generating high-confidence predic- tions for interactions in complex gene-regulatory networks. Introduction Single-cell sequencing technologies can provide detailed data for the investigation of heterogeneous populations of cells collected at specific times—so-called snapshots—during cellular differenti- ation or dynamic responses to stimulation (1). However, owing to inherent delays in molecular processes such as transcription and translation, static measurements from individual cells cannot reveal the causal interactions governing cells’ dynamic re- sponses to developmental and environmental cues (2–4). Because population-level heterogeneity in tissues often reflects the asyn- chronous progression of single cells through time-dependent processes, observed patterns of gene expression can nonetheless indicate the stages of development to which individual cells be- long (5). Many algorithms exploit these cell-to-cell differences to infer dynamic trajectories and reconstruct cells’ approximate temporal progressions along inferred lineages in pseudotime (6, 7). Several methods for gene-regulatory inference rely on pseudo- time in Granger causality tests, which try to determine whether new time series can add predictive power to inferred models of gene regulation (8, 9). However, Granger causality-based methods can be error-prone when genes display nonlinear or cyclic interac- tions or when the sampling rate is uneven or too low (9–12). Because pseudotime these problems, Granger causality-based methods often underperform model-free approaches that exploit pure statistical dependencies in gene expression data (9, 13, 14). trajectories exhibit By contrast, deep learning-based methods make no assump- tions about the temporal relationships or connectivity between genes in complex regulatory networks; instead, these data-driven Competing Interests: The authors declare no competing interest. Received: September 23, 2022. Revised: March 21, 2023. Accepted: March 23, 2023 © The Author(s) 2023. Published by Oxford University Press on behalf of National Academy of Sciences. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. 2 \| PNAS Nexus, 2023, Vol. 2, No. 4 approaches learn general features of regulatory interactions (15, 16). Here, we describe a deep learning-based method termed Depicting Lagged Causality (DELAY) that learns gene-regulatory interactions from discrete joint probability matrices of paired, pseudotime-lagged gene expression trajectories. Our data suggest that DELAY can address many shortcomings of current Granger causality-based methods and provide a useful, complementary approach to overcome common limitations in the inference of gene-regulatory networks from single-cell data. Results A convolutional neural network predicts pseudotime-lagged gene-regulatory relationships To predict gene-regulatory relationships from single-cell data, we developed a convolutional neural network based on Granger causality (17). The input to DELAY consisted of stacks of two-dimensional joint probability matrices for pairs of tran- scription factors, putative target genes, and highly correlated “neighbor” genes (18). We constructed the input matrices by aligning the gene expression trajectories of a transcription fac- tor A at several lagged positions in pseudotime relative to a target gene B to generate joint probability distributions from two-dimensional histograms of gene coexpression at each lag (Fig. 1A). The value of a given lag indicated the ordinal differ- ence between cells’ rank-ordered positions in pseudotime. Each input matrix consisted of the L2-normalized cell number counts from a two-dimensional gene coexpression histogram with 32 fixed-width bins in each dimension, spanning each gene’s minimum and maximum expression values. Although the marginal probability distributions for both A and B re- mained essentially unchanged at each lag except for cells lost at the leading and lagging edges of the shifted trajectories, realigning the gene expression in pseudotime altered key fea- tures of the resulting joint probability matrices. In other words, causally related genes share important pseudotime- lagged patterns of gene coexpression with nearby cells in single-cell trajectories. Using ground-truth labels from cell type–specific chromatin immunoprecipitation sequencing (ChIP-seq) data (14), we con- ducted supervised learning to train our neural network to pre- dict whether A directly regulates B. This procedure resembled a regression in a Granger causality test (17), in which values of a time series Y at timepoints yt are regressed against values from another time series X at timepoints xt, xt−1, xt−2, . . . , xt−T up to some maximum lag T to determine whether any time- lagged values of X add explanatory power to Y’s autoregres- sive model. DELAY likewise learned higher weights for gene coexpression matrices at specific pseudotime lags that indicated the true regulatory relationship between genes. After comparing several neural network architectures, we selected a six-layered convolutional network trained on pseudotime-aligned (T = 0) and five pseudotime-lagged (T ∈ {1, 2, 3, 4, 5}) gene coexpression matrices to predict direct gene-regulatory relationships (Fig. 1B). We also trained the network on lagged coexpression matrices of two highly corre- lated neighbor genes per gene, that is, the two transcription factors with the highest cross-correlation with A and B along the single-cell trajectory. Because the neighbor genes can pre- sent stronger alternatives to the primary hypothesis that A directly regulates B, including these matrices for A and B ver- sus their highly correlated transcription factors reduced false- positive predictions. DELAY outperforms several common methods of gene-regulatory inference We trained DELAY on gene expression data sets from human em- bryonic stem cells (hESCs) (19), mouse embryonic stem cells (mESCs) (20), and three lineages of mouse hematopoietic stem cells (mHSCs) (21). We generated separate training data sets for each of the hematopoietic lineages, and all data sets contained at least 400 cells per lineage (Table 1). Each trajectory was oriented according to known experimental timepoints or precursor cell types and lineages, and pseudotime values were inferred separ- ately with Slingshot (6) for each lineage (Fig. S1). We chose these data sets because the cell types and trajectories are well charac- terized and offer cell type–specific ChIP-seq data to generate ground-truth networks (14). Although the three hematopoietic data sets contained similar numbers of examples of true regula- tion and no interaction, both of the embryonic data sets were class-imbalanced and contained fewer examples of true regula- tion. To maximize the network’s generalizability, we trained DELAY simultaneously on all five gene expression data sets but validated it on each data set individually. We first generated ran- domly segregated 70–30% splits of all possible gene pair examples for each data set and then merged the 70% splits to create a combined training data set. In a separate analysis, we also cross- validated DELAY using inductive splits wherein specific transcrip- tion factors appeared in only the training or validation splits. After training DELAY on the combined 70% splits, the network outperformed eight of the most popular approaches for inferring gene-regulatory relationships, including six unsupervised meth- ods (8, 22–26 ) (Fig. 1C) and two deep convolutional neural net- works (18, 27) (Fig. 1D). We measured the performance of the six unsupervised methods across the combined training and valid- ation splits for each data set and then compared the results with the performance of the supervised methods on the held-out examples alone. With one exception, the deep learning-based methods outperformed all others according to the areas under both the precision–recall (PR) and receiver operating characteris- tic (ROC) curves (Fig. S2). Moreover, DELAY outperformed all the other methods according to both metrics, even though one of the deep learning-based methods, DeepDRIM, was trained on 5-fold as many neighbor gene matrices. Upon separate cross- validation, DELAY performed slightly worse on inductive splits but with a single exception outperformed all other methods on average (Fig. 1E). Across three of the training data sets, DELAY also outperformed a modified version of DeepDRIM that was trained on the same pseudotime-lagged input matrices as DELAY (Fig. S2). Together, these results suggest that DELAY out- performs other methods because it can better learn important gene-regulatory features from pseudotime-lagged gene coexpres- sion matrices. Transfer learning allows DELAY to predict novel gene-regulatory networks from new single-cell data sets To test whether DELAY generalizes to new data sets, we examined the human hepatocyte (hHep) gene-regulatory network using an additional data set with over 400 single cells and known ground- truth interactions from ChIP-seq data (14, 28). We inferred the network by the previous methods and found that tree- and mu- tual information-based methods performed slightly better than deep learning-based methods, which were not initially trained on the new data and consequently performed comparably to ran- dom predictors (Fig. 1F). To determine whether this lack of Reagor et al. \| 3 Fig. 1. Pseudotime-lagged causality allows accurate inference of gene-regulatory networks. A) Shifting the gene expression trajectory of transcription factor A forward in pseudotime with respect to that of target gene B generates a series of unique joint probability distributions. The resulting joint probability matrices contain gene coexpression signatures—in the upper-left, upper-right, and lower-right regions—that indicate A directly regulates B. B) The inverted architecture of DELAY is wide at the beginning of the network and progressively narrows to a one-dimensional vector followed by a linear classifier and sigmoid activation function to generate gene regulation probabilities. The network uses leaky ReLU activations and padded 3 × 3 convolutions throughout. C and D) DELAY outperforms eight of the most popular methods for inferring gene-regulatory interactions across several benchmark scRNA-seq data sets, including six unsupervised methods (C) and two supervised methods (D). E) All three supervised methods perform slightly worse upon cross-validation, but DELAY still outperforms all other methods on average, with the exception of PPCOR for a single metric. F and G) DELAY does not immediately generalize to a new testing data set (F) but outperforms all other methods if fine-tuned on a small percentage of the new data (G). Values for the area under the precision–recall curve (AUPRC) in C), D), F), and G) are normalized by the proportion of positive examples per data set (horizontal lines, E) and averaged across five model replicates for supervised methods. Boxes in E) show the first quartiles, means, and third quartiles for the per- transcription factor AUPRC values across the k = 5 validation folds. The statistical significance between DELAY and the next-best neural network was assessed using a one-sided Wilcoxon signed-rank test (P ≤ 0.05; P ≤ 0.01; *P ≤ 0.001). ABCEFGD4 \| PNAS Nexus, 2023, Vol. 2, No. 4 Table.1. Summary of the data sets used to train, test, and evaluate the neural network. Data sets Networks Cell type TC/S GEO Set hESC mESC mHSC-E mHSC-GM mHSC-L hHep mOHC (RNA) mOHC (ATAC) TC TC S S S TC TC S GSE75748 GSE98664 GSE81682 GSE81682 GSE81682 GSE81252 GSE137299 GSE157398 Train/Val Train/Val Train/Val Train/Val Train/Val Test Eval Eval Cells 758 421 1,071 889 847 425 1,563 391 TFs Targets Examples Edges Density 37 (4) 125 (17) 33 23 16 33 (3) 56 48 541 642 533 523 516 536 556 382 20,017 80,250 17,589 12,029 8,256 17,688 1,112 764* 3,166 20,580 9,592 6,587 4,026 6,171 166* 157* 15.82% 25.64% 54.53% 54.76% 48.76% 34.89% 14.93%* 20.55%* Columns cell type to cells describe the single-cell data sets used to train, test, and evaluate DELAY, and columns TFs to density describe the corresponding gene-regulatory networks. For three of the networks, differentially expressed transcription factors without target genes (in parentheses) were included only as target genes. Network density is equal to the number of edges from the ground-truth ChIP-seq data divided by the total number of examples. E, erythroid lineage; GM, granulocyte–monocyte lineage; L, lymphoid lineage; mOHC, mouse outer hair cell; S, snapshot; TC, time course; TF, transcription factor. Partial ground-truth network. generalizability arises from batch effects in the single-cell data, we used transfer learning to fine-tune the three deep learning- based methods on a small number of randomly segregated exam- ples from the new data set while validating on the rest of the examples (Fig. 1G). After training each network on only 1% of the new examples and validating it on the remaining 99%, the net- works’ performance matched or exceeded that of the unsuper- vised methods; training on up to 20% of the new examples yielded further performance increases. DELAY again outper- formed every other deep learning-based method. During fine- tuning, we also observed a monotonic increase in the value of the mean validation metric across PR and ROC curves, suggesting that the networks can learn useful features of gene regulation after longer training periods. Moreover, these results demonstrate that DELAY can accurately predict gene-regulatory relationships from new data sets with partially known ground-truth labels. single cells (Fig. 2C) or additional gene-dropout noise (Fig. 2D), we also discovered that the network was more sensitive to down- sampling of single cells than to gene expression losses in low- expressing cells alone. These results indicate that DELAY relies more heavily on highly expressing cells and therefore assigns lar- ger input weights for the first lagged input because on average that input contains stronger features of gene activation than other lags (Fig. 2E and F). To investigate this hypothesis, we used aug- mented input matrices with masked regions to show that DELAY relies heavily on the combined upper-left, upper-right, and lower-right regions (AON ∪ BON) across all lagged matrices (Fig. 2G). By performing a post hoc analysis of the statistical dis- persion across correctly inferred gene pair examples, we addition- ally found that DELAY generally performs better on transcription factors with stable gene expression along single-cell trajectories (Fig. S3). DELAY performs well with new input configurations, augmented matrices, and modified data sets Using different numbers of pseudotime-lagged gene coexpression matrices or neighbor gene matrices, as well as examples from modified data sets with fewer cells or additional gene dropouts, we next examined the performance of DELAY across various input configurations. We employed the original data sets to train new models on gene coexpression matrices of up to 10 pseudotime lags (Fig. 2A) or up to 10 neighbor genes, as well as on gene coex- pression matrices with varying dimensions and resolutions (Fig. S3). Training DELAY on at least one pseudotime-lagged ma- trix or with at least one neighbor gene greatly increased the net- work’s performance across all data sets. Although training DELAY on up to 10 pseudotime-lagged matrices resulted in com- parable or slightly better performance across all data sets, train- ing the network on more than two neighbor gene matrices per gene decreased performance in some instances. Adding channel masks for specific lagged matrices suggested that DELAY relies on redundancies across all available lagged inputs (Fig. S3). We also characterized DELAY’s performance on pseudotime- shuffled trajectories and observed a sharp decrease in perform- ance after reordering nearby cells in each trajectory (Fig. 2B). This result suggests that the network relies heavily on the specific ordering of adjacent cells in each trajectory. Upon examining DELAY’s performance on modified data sets containing fewer internal DELAY recognizes causal relationships in hierarchical and cyclic gene-regulatory interactions To further investigate how the selection of neighbor genes can alter DELAY’s representations and subsequent gene-regulatory inferences, we modified gene pair examples in the original 30% validation splits to exclude all neighbor genes C with known interactions across several classes of three-gene mo- tifs. First, we used DELAY to infer gene regulation for potential hierarchical and cyclic interactions across validation set exam- ples (Fig. 3A), including mutual interactions (MIs), feedback loops (FBLs), and three classes of feedforward loops (FFLs). Unlike ordin- ary Granger causality, DELAY performed well across cyclic inter- actions including 2-cycles (MIs) and 3-cycles (FBLs), which are ubiquitous among gene-regulatory networks (29). We next ex- cluded from the same validation set examples all input matrices for any neighbor gene of either A or B known from the ChIP-seq data to be involved in potential FBLs or FFLs, replacing the inputs for all omitted neighbors C with the matrices of other highly cor- related neighbor genes. Because we saw a significant decrease in performance upon excluding these neighbors C in most cases of shared and sequential gene regulation, but not for downstream targets, DELAY apparently recognizes the differences between in- put matrices of causally related genes and those of merely corre- lated genes. Moreover, these results suggest that although our network uses principles of Granger causality, it can achieve true Reagor et al. \| 5 DELAY performs best on zinc-finger proteins, bHLH factors, and other chromatin remodelers To explore how DELAY performs on classes of transcription fac- tors containing different types of DNA-binding domains, we ana- lyzed the enrichment of gene ontology (GO) terms across correctly inferred transcription factors (Fig. 3B). We found that DELAY per- forms best on zinc-finger proteins, bHLH factors, and other chro- matin remodelers. Although DELAY performs well on C2H2-type zinc-finger proteins in terms of the total number of correct predic- tions, we found that it performs better on other chromatin remod- elers and plant homeodomain (PHD) zinc-finger proteins by the overall term enrichment (Table S1). Interestingly, these results in- dicate that training the network on cell type–specific ChIP-seq data allows the network to identify regulatory relationships in- volving some non sequence-specific transcription factors and co- factors that nevertheless associate with specific targets at preferred chromatin conformations. Predicting the gene-regulatory network of auditory hair cells through multi-omic transfer learning We next sought to generate predictions for a cell type with com- plex but incompletely characterized gene-regulatory dynamics. We devised a pipeline for multi-omic transfer learning to infer the gene-regulatory network for developing mouse outer hair cells, the mechanical amplifiers of the inner ear. We used both gene expression (30) and chromatin accessibility (31) data sets to fine-tune our network. By first calculating the cell-by-gene acces- sibility scores across annotated genes, we were able to generate lagged input matrices for the single-cell ATAC sequencing (scATAC-seq) data set (Fig. 4A). Because the underlying gene ex- pression and gene accessibility distributions are both zero-inflated (8), the resulting coaccessibility matrices are qualita- tively similar to gene coexpression matrices. To determine whether pseudotime-lagged gene coaccessibility matrices contain features that indicate direct gene-regulatory re- lationships, we fine-tuned DELAY on the scATAC-seq matrices us- ing ground-truth examples collected from two cell type–specific data sets of Sox2 and Atoh1 target genes (32, 33). Seventy percent of the randomly segregated ground-truth examples were used for fine-tuning, and the remaining 30% were held out for validation (Fig. 4B). DELAY performed slightly worse by the metric of area under the PR curve than by area under the ROC curve, indicating that the network is better at discriminating false positives than selecting true positives. Although DELAY did not outperform all other deep learning-based methods, the network’s performance was comparable with previous training and validation on small fractions of the hHep single-cell RNA sequencing (scRNA-seq) data set, which suggests that gene coaccessibility matrices are also useful for inferring direct gene-regulatory interactions. We separately fine-tuned the network on all available Sox2 and Atoh1 ground-truth examples from an additional gene expression data set for mouse outer hair cells (30). With previously deter- mined hyperparameters for scRNA-seq data, training DELAY on two graphics processing units (GPUs) required 230 ± 1 min (mean ± SD) per model. We then compared the target gene rank correlations between the inferred transcription factor-only gene-regulatory networks to determine whether networks in- ferred from scRNA-seq data and scATAC-seq data were similar (Fig. 4C). Although we observed stronger correlations between predictions from data set-specific models than between average predictions across data sets, the target gene rank correlations Fig. 2. The neural network relies on cell order in pseudotime and gene expression strength. A) Training new models of DELAY on increasing numbers of pseudotime-lagged matrices gives the largest performance increase when using up to a single lag. B) Reordering single cells in pseudotime sharply decreases performance, suggesting that DELAY relies on the specific ordering of adjacent cells in each trajectory. C and D) The network is sensitive to random down-sampling of cells across data sets (C) but relatively more robust to induced, additional gene dropouts in weakly expressing cells (D), suggesting that DELAY relies heavily on highly expressing cells. E and F) The network learns larger input weights for lagged matrices of the first pseudotime lag (E), which also contain more cells in the combined “ON” region (AON ∪ BON; dotted outline, F) on average across training data sets (dotted line, E). The combined “ON” region is comprised of the upper-left “ON–OFF” quadrant (AON ∩ BOFF), upper-right “ON–ON” quadrant (AON ∩ BON), and lower-right “OFF–ON” quadrant (AOFF ∩ BON). G) Masking different regions of the input matrices shows that the network relies heavily on the combined “ON” region. The lines and shaded regions in A–E) show the average and full range of values across five model replicates, and the markers in G) show the average values across model replicates. The statistical significance in G) was assessed with a Kruskal–Wallis test (*P ≤ 0.001). causal inference for genes involved in several classes of hierarch- ical and cyclic regulatory interactions by avoiding several limiting assumptions of ordinary Granger causality. ABCDEFG6 \| PNAS Nexus, 2023, Vol. 2, No. 4 Fig. 3. DELAY uncovers causal, cyclic, and context-specific gene-regulatory relationships. A) The network’s performance when inferring putative transcription factor-target gene pairs across two- and three-gene motifs in the validation set suggests that DELAY can distinguish between different types of hierarchical and cyclic gene regulation. Upon exclusion of one or more neighbor genes C from the input features of the validation examples, the network’s performance declines significantly if C is a shared or sequential regulator of A and B, but not if C is a shared target. B) GO-term enrichment indicates that DELAY performs best on zinc-finger proteins (PHD-type, GATA-type, C2H2-type, NHR-type, and C5HC2-type), bHLH factors (Myc-type), and other chromatin remodelers (SNF2-related, chromodomain, bromodomain, HDA domain, and helicase). Markers and error bars in A) show the average values and full range of performance across five model replicates. The statistical significance in A) was assessed with a one-sided Wilcoxon signed-rank test (P ≤ 0.05) and in B) with a Fisher’s exact test (Padjusted ≤ 0.05). between the two data sets were highly variable, and the predic- tions of some transcription factors agreed better than others. Reasoning that the inferences with better agreement across data sets constituted the best predictions and highest confidence interactions, we derived the consensus hair cell gene-regulatory network from the average predictions across both data sets. The resulting network consisted of 347 predicted transcription factor–target interactions with gene regulation probabilities >0.5. DELAY identifies important genes, interactions, and modules in the hair cell gene-regulatory network We used hierarchical clustering to group transcription factors and target genes in the transcription factor-only gene-regulatory network for hair cells by similarities in their predicted targets and regulators, respectively (Fig. 4D). This procedure revealed two distinct developmental modules corresponding to prosen- sory genes such as Sox2, Id2, Hes1, and Prox1 and hair cell- specific genes such as Atoh1, Pou4f3, Gfi1, Lhx3, and Barhl1. We sought to validate the predicted interactions by comparing the locations of known transcription factor-binding sites (34, 35) to the locations of open-chromatin peaks (31) within 50 kb of target genes’ transcription start sites. Twenty-two of 28 predicted in- teractions were confirmed by the accessibility of transcription factor-binding sites. Of the six remaining interactions, three were instances of predicted target-gene regulation by transcrip- tional cofactors lacking true DNA-binding domains. We add- itionally identified 13 reported interactions that were not detected by DELAY (31, 36–54). The most notable feature of the inferred gene-regulatory network for hair cells is that Sox2 and Id2—in addition to their proteins’ well-known roles in regulating target genes and main- taining a prosensory cell fate—are themselves the targets of a wide variety of transcription factors including Sox proteins, homeobox factors, zinc-finger proteins, and Notch effectors. Eleven of the 22 interactions confirmed by our binding-site acces- sibility analysis represented direct regulation of Sox2 and Id2, in- cluding mutually activating interactions between Sox2 and Sox9, Sox11, Prox1, and Hes1, and mutual inhibition with Hist1h1c. One other study has suggested that Hes1 directly regulates Sox2 (55). Another notable feature of the inferred hair cell network is that the LIM-homeobox transcription factor Lhx3 regulates its own ex- pression as well as that of two other LIM-only transcription factors (Lmo4 and Lmo1). This result implies a role for Lhx3 in maintain- ing the later expression of these important early Sox2 inhibitors (31, 47, 49). Other key features of the network include Sox11 regu- lation by Hes1, Nfkbia regulation by Gfi1, and regulation of the spli- cing factor gene Srsf2 by the products of three different prosensory genes. Srsf2 was recently predicted to play a role in the splicing of an important deafness gene in humans (56). Discriminative motif analysis of target-gene enhancer sequences enables de novo discovery of DNA-binding motifs DELAY permits a complementary approach typical gene-regulatory inference workflows such as SCENIC (57) that use cis-regulatory sequences to identify and discard false-positive interactions resulting from indirect gene regulation (11, 30, 31). to ABReagor et al. \| 7 Fig. 4. DELAY accurately predicts gene-regulatory interactions in the auditory hair cell network through multi-omic transfer learning. A) An example of an empirical joint probability matrix from scATAC-seq trajectories. B) After fine-tuning the network on lagged scATAC-seq input matrices, DELAY performs comparably with other neural networks and to previous testing on hHep scRNA-Sseq data. C) Training DELAY separately on scRNA-seq and scATAC-seq data sets of hair cell development reveals a stronger correlation between predictions from model replicates than between data sets across the transcription factor-only network. D) Average gene regulation probabilities P across the hair cell gene-regulatory network accurately predict interactions between transcription factors (rows) and targets (columns), when comparing known binding sites to open-chromatin peaks in the target genes’ enhancer sequences. Hierarchical clustering with WPGMA reveals distinct gene modules for prosensory genes (Sox2, Id2, Hes1, and Prox1) and hair cell genes (Atoh1, Pou4f3, Gfi1, Lhx3, and Barhl1). Up- or down-regulation was deduced from the correlation in the gene expression data. Markers and error bars in B) show the average values and full range of performance across five model replicates for each neural network. Markers in C) show the median target gene rank correlations across comparisons. The statistical significance in C) was assessed with a Kruskal–Wallis test (*P ≤ 0.001). Because these methods rely on databases of known DNA-binding motifs, they necessarily overlook predictions for cofactors and pu- tative transcription factors with unknown binding sequences. Instead of using cis-regulatory sequences to refine our initial pre- dictions, we introduced enhancer sequence information post hoc to perform discriminative motif analysis within the enhancers of predicted targets in the hair cell network (Figs. 5A and S4). We identified enriched motifs that closely resembled known DNA-binding motifs for nine of 11 transcription factors with at least one predicted target gene in the transcription factor-only hair cell network (Table S2). We additionally sought to predict the mechanisms by which several identified hair cell cofactors accomplish sequence-specific gene regulation in the absence of true DNA-binding domains. Specifically, we compared sequences enriched in the enhancers of cofactors’ predicted targets to several databases of known DNA-binding motifs to identify the cofactors’ DNA-binding part- ners (so-called guilt by association). Through this analysis, we were able to determine that the most significantly enriched se- quence in the enhancers of the histone Hist1h1c’s predicted target genes closely matched known binding motifs for several SoxB2/A transcription factors (Fig. 5B). In addition, a sequence enriched in the enhancers of cyclin Ccnd1’s targets accorded with known motifs for the Pbx family of homeobox transcription factors (Fig. S5). These cofactors might form complexes with the identi- fied transcription factors to regulate their target genes in hair cells. As a final demonstration of DELAY’s high-confidence predic- tions, we used the fine-tuned model for the scRNA-seq data to generate regulatory predictions for transcription factors ex- pressed in at least 20% of developing hair cells (Table S3). After considering these additional transcription factors, we identified the putative C2H2 zinc-finger protein Fiz1 as a likely regulator of Sox2 and Hes6. Discriminative motif analysis of these genes’ en- hancer sequences (Fig. 5C) uncovered a likely Fiz1 consensus DNA-binding sequence (5′-CGCTGC-3′) similar to that of other Sox2 regulators from the Zic family of C2H2 zinc-finger proteins (58). Discussion Building upon several deep learning-based methods (18, 27), we have demonstrated that combining fully supervised deep learning with joint probability matrices of pseudotime-lagged single-cell trajectories can overcome certain limitations of current Granger causality-based methods of gene-regulatory inference (9, 12), ADBC8 \| PNAS Nexus, 2023, Vol. 2, No. 4 Fig. 5. Accurate target-gene predictions enable de novo discovery of DNA-binding motifs. A) Three examples of motifs enriched in the enhancer sequences of predicted target genes (bottom row) closely resemble known DNA-binding motifs (top row) for transcription factors with at least one predicted target gene in the transcription factor-only hair cell gene-regulatory network. B and C) Sequences for putative transcription factors or cofactors may represent novel DNA-binding motifs or indicate sequence-specific interactions through multi-protein complexes. Motifs for two hair cell genes closely match known DNA-binding sequences, suggesting that the linker histone Hist1h1c (B, top row) forms complexes with SoxB2/A factors (B, bottom row) and that the putative C2H2 zinc-finger protein Fiz1 (C, top row) recognizes motifs similar to Zic family C2H2 zinc-finger proteins (C, bottom row). The statistical significance of each motif alignment was estimated using the cumulative density function of all possible comparisons of known motifs across enriched sequences for A) or inferred motifs across all database motifs for B) and C) (P ≤ 0.01; *** P ≤ 0.001). such as their inability to infer cyclic regulatory motifs. Although our convolutional neural network DELAY remains sensitive to the ordering of single cells in pseudotime, the network can never- theless accurately infer direct gene-regulatory interactions from both time course and snapshot data sets, unlike many supervised methods that rely strongly on the number of available time course samples (15, 16). We suspect that DELAY is sensitive to the specific ordering of adjacent cells in trajectories because pseudotime in- ference methods such as Slingshot (6) infer lineages from min- imum spanning trees that directly depend on cell-to-cell similarities in gene expression values (7). We have presented a multi-omic paradigm for fine-tuning DELAY on both gene expression and chromatin accessibility data sets for the development of auditory hair cells. The network’s ABCaccurate predictions allow de novo inferences of known and unknown DNA-binding motifs, establishing DELAY as a comple- mentary approach to common methods of gene-regulatory infer- ence. Our method’s high-confidence predictions across the hair cell gene-regulatory network also support it as an attractive option for experimentalists with limited resources to predict true gene-regulatory relationships from complex, large-scale gene-regulatory networks while avoiding spurious, indirect inter- actions often introduced through unsupervised methods (14). We have additionally identified with high confidence interac- tions between cofactors and target genes in the regulatory net- work of hair cells that methods such as SCENIC (57) would have overlooked. Using “guilt by association,” we predict that the cyclin Ccnd1—a known transcriptional cofactor found at the enhancers of several Id genes during retinal development (59)—forms com- plexes with the Pbx family of homeobox transcription factors, of which at least one is a known target of Prox1 in the inner ear (60). Moreover, we predict that the retina-specific linker histone Hist1h1c (61) forms complexes with the SoxB2 family of transcrip- tion factors, which have known antagonistic effects on Sox2 ex- pression in hair cells (62). Finally, we predict that the putative hair cell-specific C2H2 zinc-finger protein Fiz1—which is also ex- pressed during retinal development (63)—is a regulator of Sox2 in hair cells and identified its likely DNA-binding sequence in the enhancers of predicted target genes. Believing that DELAY will be a valuable resource to the commu- nity, we have provided an easy-to-use and open-source imple- mentation of the algorithm as well as pre-trained model weights for subsequent fine-tuning on new single-cell data sets and partial ground-truth labels. Unlike other deep learning-based methods, our implementation of DELAY maximizes usability and prioritizes model flexibility. Although we chose to fine-tune DELAY on cell type–specific ChIP-seq data, future studies may choose to fine- tune DELAY on curated interactions from databases or from gain- or loss-of-function experiments, especially in the absence of ChIP-seq data (14). These additional interactions can also sup- plement smaller ground-truth data sets, such as those of Sox2 and Atoh1 target genes, to mitigate false-negative predictions. We be- lieve that the modest computational cost of training new models of DELAY will prove a worthwhile investment for future investiga- tions into complex gene-regulatory networks. Materials and methods Preparing single-cell RNA-seq data sets to train and test the convolutional neural network To train our convolutional neural network, we used scRNA-seq data sets from two time course studies of endodermal specifica- tion of hESCs (19) and mESCs (20) and an in vivo snapshot study of erythroid (E), granulocyte–monocyte (GM), and lymphoid (L) specification in mHSCs (21). We additionally employed a data set from a time course study of the differentiation of hHeps from induced pluripotent stem cells to test our network (28). For each of these studies, we collected the normalized gene expres- sion values, pseudotime values, and ground-truth networks from BEELINE’s supplementary data files (14). In BEELINE, Pratapa et al. collected the normalized gene expression values from the original studies or obtained the values themselves by log- transforming the transcripts per kilobase million. Moreover, those authors used Slingshot (6) to infer the pseudotime values separ- ately for each data set, orienting the inferred trajectories by known experimental timepoints or precursor cell types and Reagor et al. \| 9 lineages. Finally, Pratapa et al. selected genes with differential ex- pression across pseudotime by applying generalized additive models (GAMs) to compute gene variances and their associated P values. To train DELAY, we utilized their gene variances and cor- rected P values to select differentially expressed transcription fac- tors (Padjusted<0.01) and 500 additional genes with the highest variance for each data set. We also collected the ground-truth la- bels from BEELINE’s supplementary data files, which contained tables of transcription factor–target gene interactions curated from cell type–specific ChIP-seq experiments in the ENCODE (64), ChIP-Atlas (65), and ESCAPE (66) databases. Table 1 provides details for each data set including descriptions of the ground- truth networks. We also characterized our network on modified data sets with shuffled pseudotime, fewer cells, and additional gene-dropout noise. We shuffled pseudotime values across single-cell trajector- ies using np.random.normal in NumPy (v1.20.2) to select the indi- ces of single cells either leading or lagging successive cells in each trajectory at some length scale σ before swapping the cells’ posi- tions in pseudotime. We also used np.random.choice to select the indices of single cells to retain when down-sampling the num- ber of cells in a data set. Lastly, we induced additional gene- dropout noise in data sets by setting the gene expression values to 0 with a probability of p for the bottom p percent of cells ex- pressing each gene in a given data set. Generating examples of transcription factor– target gene pairs from single-cell data sets For each data set, we generated mini-batches of 512 transcription factor–target gene pair examples by first enumerating all possible combinations of differentially expressed transcription factors and potential target genes—including both transcription factors and highly varying genes. We then generated aligned and pseudotime- lagged gene coexpression matrices for up to five lags with the fol- lowing configurations as separate input channels: transcription factor–target, transcription factor–transcription factor, target– target, transcription factor–neighbor (for two neighbors), and tar- get–neighbor (for two neighbors). We concatenated the input channels for each example to form three-dimensional stacks of input matrices with dimensions 42 × 32 × 32 (channels × height × width). For each gene, we used np.correlate in NumPy to select the transcription factors with the highest absolute cross-correlation at a maximum offset of five pseudotime lags to use as neighbor genes. We trained five model replicates on unique, randomly seg- regated 70–30% splits of all possible gene pair examples generated with the random split function in PyTorch (v1.8.1). Each random split achieved a strict segregation of unique gene pair examples into either the training or validation sets, though gene pairs for the same transcription factor but different target genes appeared in both sets. For separate model cross-validation, we generated five inductive splits across each data set wherein specific tran- scription factors appeared in only the training (80%) or validation (20%) splits. For both analyses, training splits from individual data sets were subsequently merged to create combined training data sets. We later used the original randomly segregated validation splits to characterize our network on augmented coexpression matrices and matrices generated from modified single-cell data sets. To augment gene coexpression matrices, we masked with ze- ros either specific region across all input channels or entire input channels corresponding to specific pseudotime lags. 10 \| PNAS Nexus, 2023, Vol. 2, No. 4 Constructing a convolutional neural network to classify lagged gene coexpression matrices We designed a convolutional neural network based on an inverted VGGnet (67) that uses five convolutional layers to first expand the input to 1,024 channels and then successively halve the number of channels to 64 before classifying examples as either true regu- lation or no interaction with a fully connected linear layer. Each convolutional layer sums over the two-dimensional cross- correlations (⋆) of 3 × 3 kernels and input channels i to find the features for a given output channel j, as shown in Equation 1: output (Nk, Coutj ) = bias (Coutj ) + 􏽐Cin i=1 weights 3×3 (Cini , Coutj ) ⋆ input (Nk, Cini ), (1) in which the weights and bias are trainable parameters, N is the mini-batch size, C is the number of channels, Cout = Cin/2 for layers 2 through 5, and all convolutions are zero-padded at the edges of matrices. To preserve the gradient flow during training, we used leaky rectified linear units (ReLUs) with a negative slope of 0.2 (Equation 2) as our nonlinear activation function after each convolutional layer: Leaky ReLU (x) = max (0, x) + 0.2∗min (0, x). (2) As shown in Equation 3, we additionally used 2 × 2, unpadded max-pooling layers between convolutional layers to identify im- portant features and down-sample activation maps: 􏼓 􏼔 hin 2 , win 2 = stride 2×2 max m,n ∈ {0,1} 􏼕 input (Nk, Cinj , xm,n) , 􏼒 output Nk, Coutj , (3) in which x represents the 2 × 2 input windows and h, w are the height and width of the input channel j. After the final convolu- tional layer, we used global-average pooling (Equation 4) to reduce the remaining 64 feature maps j to a single, 64-dimensional vector x: output (Nk, Coutj ) = avg hin×win input (Nk, Cinj ). (4) We lastly used a fully connected linear layer (Equation 5) with a sigmoid activation function (Equation 6) to generate gene regula- tion probabilities: output (Nk) = bias FC + 􏽐64 i=1 weights FC (xi)∗input (Nk, xi), (5) Sigmoid (x) = 1 1 + exp(−x) , (6) where a probability of P > 0.5 indicates a true gene-regulatory interaction for the given gene pair example. Training and fine-tuning the network on pseudotime-lagged gene coexpression matrices We used PyTorch’s implementation of stochastic gradient descent (SGD) to optimize our network with respect to the class-weighted binary cross-entropy loss, summed across each mini-batch and scaled by the overall batch size 512, as shown in Equation 7: L(y, ˆy) = − wN 512 􏽘N n=1 yn∗log ˆyn + (1 − yn)∗log (1 − ˆyn), (7) in which y and ˆy are the target and predicted values (respectively), wN is the fraction of true examples in the mini-batch, and N is the size of the current mini-batch (which might be <512). Prior to optimization, we used He initialization (68) with uniform priors to set the weights for all convolutional and linear layers. We used a learning rate (LR) of 0.5 to train each model for up to 100 epochs, validating performance after each epoch and stopping training after 10 or more epochs without an improvement in the mean validation metric across PR and ROC curves. For fine-tuning, we trained each model for up to a maximum of 5,000 additional epochs due to the observed monotonic increase in the value of the mean validation metric. We occasionally reduced the LR to 0.25 or 0.1 if the training became unstable, and we tried several mini-batch sizes ≥24 for fine-tuning and separate model cross- validation. All training and testing was performed on two Nvidia RTX 8000 GPUs. Comparing DELAY to other top-performing gene-regulatory inference methods We compared the performance of our neural network to two other convolutional neural networks, as well as the top six best- performing methods identified in a previous benchmarking study. Using BEELINE, we inferred the gene-regulatory networks for all training and testing data sets with the tree-based methods GENIE3 (22) and GRNBoost2 (23), the mutual information-based method PIDC (24), and the partial correlation and regression- based methods PPCOR (25), SCODE (26), and SINCERITIES (8). We utilized the best parameter values identified in BEELINE for the partial correlation and regression-based methods. For the two deep learning-based methods [CNNC (27) and DeepDRIM (18)], we used the original studies to reconstruct the neural networks in PyTorch before training models on the same randomly segre- gated training examples as DELAY. In addition to training both neural networks on pseudotime-aligned gene coexpression matri- ces for primary transcription factor–target gene pairs, we also trained DeepDRIM on 10 neighbor gene matrices per gene as in the original study. Moreover, we separately trained DeepDRIM on five pseudotime-lagged matrices per gene pair with 10 neigh- bors across the same inductive splits as DELAY to compare the cross-validated performance between the networks. Analyzing enrichment of GO terms across correctly inferred gene pair examples We used Enrichr (69) (https://maayanlab.cloud/Enrichr/) to ana- lyze GO-term enrichment across correctly inferred transcription factors for terms related to InterPro DNA-binding domains (70). We used a Fisher’s exact test with Benjamini–Hochberg correction for multiple-hypothesis testing (Padjusted<0.05) to assess the statis- tical significance for each GO term. Preparing single-cell multi-omics data sets to infer the hair cell gene-regulatory network We used two single-cell data sets from developing mouse outer hair cells (30, 31) to predict the consensus hair cell gene- regulatory network. First, we collected the normalized gene expression values from the original scRNA-seq study (30) and then used reciprocal PCA to integrate single cells across four time- points of sensory epithelium development into a single data set in Seurat (71) (v3.1.4). Then, we used Slingshot (6) (v1.0.0) to infer pseudotime values across the outer hair cell trajectory and ap- plied LOESS regressions and GAMs to select the genes that were differentially expressed across both the sensory epithelium and pseudotime, respectively. We again selected all differentially ex- pressed transcription factors (P<0.01 after Bonferroni correction for multiple-hypothesis testing) and 500 additional genes with the highest variance and used the integrated gene expression as- say in Seurat to generate the corresponding gene coexpression matrices. In a separate analysis, we broadened our selection cri- teria to encompass all genes that were differentially expressed in at least 20% of the outer hair cells, regardless of their expression across the full sensory epithelium. For the scATAC-seq data set, we first collected the cell-by-peak chromatin accessibility data from the original study (31). Then, we used the function createGmatFromMat in SnapATAC (72) (v1.0.0) to calculate the cell-by-gene accessibility scores as the counts of each 5 kb bin per gene in the UCSC mouse genome mm10 (73) (TxDb.Mmusculus.UCSC.mm10.knownGene; v3.4.4) that contained at least one open-chromatin peak for a given cell. These values were then log-normalized. We again used Slingshot to infer pseudotime values across the outer hair cell trajectory and applied GAMs to the raw accessibility counts to select the dif- ferentially accessible genes across pseudotime. For the inferred scATAC-seq network, we selected all differentially accessible tran- scription factors and variable genes (Padjusted<0.01) that were also differentially expressed along the scRNA-seq trajectory. Because the scATAC-seq data set had fewer cells than the scRNA-seq data set, we used 24 fixed-width bins in each dimension to generate the corresponding gene coaccessibility matrices prior to fine-tuning. Discriminative motif analysis to discover de novo transcription factor-binding motifs We used the UCSC mouse genome mm10 and twoBitToFa to download the 100 kb enhancer sequences spanning 50 kb up- stream and downstream of all target genes’ transcription start sites in the transcription factor-only hair cell gene-regulatory net- work and then divided them into 100 bp fragments and sorted the resulting sequences into primary (predicted targets) and control (predicted no interaction) groups for each transcription factor with at least one target gene in the network. We then used STREME (74) (v5.4.1) to perform discriminative motif analysis with a P value threshold of 0.05 to identify enriched motifs in pre- dicted target genes’ enhancers, which we then compared with ei- ther the transcription factors’ known binding motifs from CIS-BP (34), or to all motifs from the JASPAR (75), UniPROBE (76) (mouse), and Jolma et al. (77) databases using TOMTOM (78) (v5.4.1) with an E value threshold of 10 for significant alignments. Acknowledgments The authors would like to acknowledge Junyue Cao, Viviana Risca, and Christina Leslie, as well as Adrian Jacobo, Agnik Dasgupta, Emily Atlas, and other members of the Laboratory of Sensory Neuroscience for helpful discussions and comments on the manuscript. Supplementary material Supplementary material is available at PNAS Nexus online. Funding Reagor et al. \| 11 Author contributions C.C.R. and N.V. conceived the study, and all authors designed the analysis. C.C.R. carried out all experiments and analysis and wrote the paper. A.J.H. supervised the project, and all the authors edited the paper. Data availability The processed experimental files for all single-cell data sets used in this study are available on Zenodo at https://doi.org/10.5281/ zenodo.7474099; Table 1 lists the Gene Expression Omnibus (GEO) accession numbers for each data set. The saved model weights for DELAY are available on Zenodo at https://doi.org/10. 5281/zenodo.7474115. All experimental logs from this study are available at https://tensorboard.dev/experiment/RBVBetLM RDiEvO7sBl452A. We have provided an open-source implementa- tion of DELAY in PyTorch with listed requirements and documen- tation at https://github.com/calebclayreagor/DELAY. References 1 Ocone A, Haghverdi L, Mueller NS, Theis FJ. 2015. Reconstructing gene regulatory dynamics from high-dimensional single-cell snapshot data. Bioinformatics 31:i89–i96. 2 Qian J, Dolled-Filhart M, Lin J, Yu H, Gerstein M. 2001. Beyond synexpression relationships: local clustering of time-shifted and inverted gene expression profiles identifies new, biologically relevant interactions. J Mol Biol. 314:1053–1066. Bar-Joseph Z, Gitter A, Simon I. 2012. Studying and modelling dy- namic biological processes using time-series gene expression data. Nat Rev Genet. 13:552–564. 3 4 Ding J, Sharon N, Bar-Joseph Z. 2022. Temporal modelling using single-cell transcriptomics. Nat Rev Genet. 23:355–368. https:// doi.org/10.1038/s41576-021-00444-7 5 Tritschler S, et al. 2019. Concepts and limitations for learning de- velopmental trajectories from single cell genomics. Development 146:dev170506. Street K, et al. 2018. Slingshot: cell lineage and pseudotime infer- ence for single-cell transcriptomics. BMC Genomics 19:477 . 6 7 Trapnell C, et al. 2014. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol. 32:381–386. Papili Gao N, Ud-Dean SMM, Gandrillon O, Gunawan R. 2018. SINCERITIES: inferring gene regulatory networks from time- stamped single cell transcriptional expression profiles. Bioinformatics 34:258–266. 8 9 Deshpande A, Chu L-F, Stewart R, Gitter A. 2022. Network infer- ence with Granger causality ensembles on single-cell transcrip- tomics. Cell Rep. 38:110333 . 10 Finkle JD, Wu JJ, Bagheri N. 2018. Windowed granger causal infer- ence strategy improves discovery of gene regulatory networks. Proc Natl Acad Sci. 115:2252–2257. 11 Michailidis G, d’Alché-Buc F. 2013. Autoregressive models for gene regulatory network inference: sparsity, stability and causal- ity issues. Math Biosci. 246:326–334. 12 Zhang Y, Chang X, Liu, X. 2021. Inference of gene regulatory net- works using pseudo-time series data. Bioinformatics. 37:btab099. https://doi.org/10.1093/bioinformatics/btab099. C.C.R. is supported by National Science Foundation Graduate Research Fellowship Grant No. 1946429. A.J.H. is an Investigator of Howard Hughes Medical Institute. 13 Qiu X, et al. 2020. Inferring causal gene regulatory networks from coupled single-cell expression dynamics using scribe. Cell Syst. 10:265–274.e11. 12 \| PNAS Nexus, 2023, Vol. 2, No. 4 14 Pratapa A, Jalihal AP, Law JN, Bharadwaj A, Murali TM. 2020. 35 Grant CE, Bailey TL, Noble WS. 2011. FIMO: scanning for occur- Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat Methods 17:147–154. 15 Yuan Y, Bar-Joseph Z. 2021. Deep learning of gene relationships from single cell time-course expression data. Brief Bioinform. 22: rences of a given motif. Bioinformatics 27:1017–1018. 36 Zheng JL, Shou J, Guillemot F, Kageyama R, Gao WQ. 2000. Hes1 is a negative regulator of inner ear hair cell differentiation. Development 127:4551–4560. bbab142. 37 Zine A, et al. 2001. Hes1 and Hes5 activities are required for the 16 Xu Y, Chen J, Lyu A, Cheung WK, Zhang, L. 2022. dynDeepDRIM: a normal development of the hair cells in the mammalian inner dynamic deep learning model to infer direct regulatory interac- ear. J Neurosci. 21:4712–4720. tions using time-course single-cell gene expression data. Brief 38 Ikeda R, Pak K, Chavez E, Ryan AF. 2015. Transcription factors Bioinform. bbac424. 17 Granger CWJ. 1969. Investigating causal relations by economet- ric models and cross-spectral methods. Econometrica 37:424–438. 18 Chen J, et al. 2021. DeepDRIM: a deep neural network to recon- struct cell-type-specific gene regulatory network using single- cell RNA-seq data. Brief Bioinform. 22:bbab325. 19 Chu L-F, et al. 2016. Single-cell RNA-seq reveals novel regulators of human embryonic stem cell differentiation to definitive endo- derm. Genome Biol. 17:173. 20 Hayashi T, et al. 2018. Single-cell full-length total RNA sequen- cing uncovers dynamics of recursive splicing and enhancer RNAs. Nat Commun. 9:619 . 21 Nestorowa S, et al. 2016. A single-cell resolution map of mouse hematopoietic stem and progenitor cell differentiation. Blood 128:e20–e31. 22 Huynh-Thu VA, Irrthum A, Wehenkel L, Geurts P. 2010. Inferring regulatory networks from expression data using tree-based methods. PLoS One 5:e12776. 23 Moerman T, et al. 2019. GRNBoost2 and Arboreto: efficient and scalable inference of gene regulatory networks. Bioinformatics 35:2159–2161. 24 Chan TE, Stumpf MPH, Babtie AC. 2017. Gene regulatory network inference from single-cell data using multivariate information measures. Cell Syst. 5:251–267.e3. 25 Kim S. 2015. Ppcor: an R package for a fast calculation to semi- partial correlation coefficients. Commun Stat Appl Methods 22: 665–674. 26 Matsumoto H, et al. 2017. SCODE: an efficient regulatory network inference algorithm from single-cell RNA-seq during differenti- ation. Bioinformatics 33:2314–2321. 27 Yuan Y, Bar-Joseph Z. 2019. Deep learning for inferring gene rela- tionships from single-cell expression data. Proc Natl Acad Sci. 116: 27151–27158. 28 Camp JG, et al. 2017. Multilineage communication regulates hu- man liver bud development from pluripotency. Nature 546: 533–538. with conserved binding sites near ATOH1 on the POU4F3 gene enhance the induction of cochlear hair cells. Mol Neurobiol. 51: 672–684. 39 Du X, et al. 2013. Regeneration of mammalian cochlear and ves- tibular hair cells through Hes1/Hes5 modulation with siRNA. Hear Res. 304:91–110. 40 Kirjavainen A, et al. 2008. Prox1 interacts with Atoh1 and Gfi1, and regulates cellular differentiation in the inner ear sensory epithelia. Dev Biol. 322:33–45. 41 Yu KS. et al. Development of the mouse and human cochlea at single cell resolution. 739680 Preprint at https://www.biorxiv. org/content/10.1101/739680v2 (2019). 42 Benito-Gonzalez A, Doetzlhofer A. 2014. Hey1 and Hey2 control the spatial and temporal pattern of mammalian auditory hair cell differentiation downstream of hedgehog signaling. J Neurosci. 34:12865–12876. 43 Doetzlhofer A, et al. 2009. Hey2 regulation by FGF provides a Notch-independent mechanism for maintaining pillar cell fate in the organ of Corti. Dev Cell. 16:58–69. 44 Kamaid A, Neves J, Giraldez F. 2010. Id gene regulation and func- tion in the prosensory domains of the chicken inner ear: a link be- tween Bmp signaling and Atoh1. J Neurosci. 30:11426–11434. 45 Yu R, Wang P, Chen X-W. 2020. The role of gfi1.2 in the develop- ment of zebrafish inner ear. Hear Res. 396:108055 . 46 Jones JM. 2006. Inhibitors of differentiation and DNA binding (Ids) regulate Math1 and hair cell formation during the development of the organ of Corti. J Neurosci. 26:550–558. 47 Hertzano R, et al. 2007. Lhx3, a LIM domain transcription factor, is regulated by Pou4f3 in the auditory but not in the vestibular sys- tem. Eur J Neurosci. 25:999–1005. 48 Hertzano R, et al. 2004. Transcription profiling of inner ears from Pou4f3 ddl/ddl identifies Gfi1 as a target of the Pou4f3 deafness gene. Hum Mol Genet. 13:2143–2153. 49 Deng M, et al. 2014. LMO4 Functions as a negative regulator of sensory organ formation in the mammalian cochlea. J Neurosci. 29 Alon U. 2007. Network motifs: theory and experimental ap- 34:10072–10077. proaches. Nat Rev Genet. 8:450–461. 30 Kolla L, et al. 2020. Characterization of the development of the 50 Bae S, Bessho Y, Hojo M, Kageyama R. 2000. The bHLH gene Hes6, an inhibitor of Hes1, promotes neuronal differentiation. mouse cochlear epithelium at the single cell level. Nat Commun. Development 127:2933–2943. 11:2389 . 51 Fior R, Henrique D. 2005. A novel hes5/hes6 circuitry of negative 31 Wang S, Lee MP, Jones S, Liu J, Waldhaus J. 2021. Mapping the regulation controls Notch activity during neurogenesis. Dev Biol. regulatory landscape of auditory hair cells from single-cell 281:318–333. multi-omics data. Genome Res. 31:1885–1899 https://doi.org/10. 52 Matern MS, et al. 2020. GFI1 Functions to repress neuronal gene 1101/gr.271080.120 expression in the developing inner ear hair cells. Development 32 Kwan KY, Shen J, Corey DP. 2015. C-MYC transcriptionally amp- 147:dev186015. lifies SOX2 target genes to regulate self-renewal in multipotent 53 Hou K, et al. 2019. A critical E-box in Barhl1 3′ enhancer is essen- otic progenitor cells. Stem Cell Rep. 4:47–60. tial for auditory hair cell differentiation. Cells 8:458. 33 Cai T, et al. 2015. Characterization of the transcriptome of nas- 54 Chonko KT, et al. 2013. Atoh1 directs hair cell differentiation and cent hair cells and identification of direct targets of the Atoh1 survival in the late embryonic mouse inner ear. Dev Biol. 381: transcription factor. J Neurosci. 35:5870–5883. 401–410. 34 Weirauch MT, et al. 2014. Determination and inference of eukary- 55 Waldhaus J, et al. 2012. Stemness of the organ of Corti relates to otic transcription factor sequence specificity. Cell 158:1431–1443. the epigenetic status of Sox2 enhancers. PLoS One 7:e36066. Reagor et al. \| 13 56 Booth KT, et al. 2020. Novel loss-of-function mutations in COCH cause autosomal recessive nonsyndromic hearing loss. Hum Genet. 139:1565–1574. 57 Aibar S, et al. 2017. SCENIC: single-cell regulatory network infer- ence and clustering. Nat Methods 14:1083–1086. 58 Marchal L, Luxardi G, Thomé V, Kodjabachian L. 2009. BMP inhib- ition initiates neural induction via FGF signaling and Zic genes. Proc. Natl. Acad. Sci. 106:17437–17442. 59 Bienvenu F, et al. 2010. Transcriptional role of cyclin D1 in devel- opment revealed by a genetic–proteomic screen. Nature 463: 374–378. 60 Luo Z, Zhang J, Qiao L, Lu F, Liu Z. 2021. Mapping genome-wide binding sites of Prox1 in mouse cochlea using the CUT&RUN ap- proach. Neurosci Bull. 37:1703–1707. 61 Popova EY, et al. 2013. Developmentally regulated linker histone H1c promotes heterochromatin condensation and mediates structural integrity of rod photoreceptors in mouse retina. J Biol Chem. 288:17895–17907. 62 Freeman SD, Daudet N. 2012. Artificial induction of Sox21 regu- lates sensory cell formation in the embryonic chicken inner ear. PLoS One 7:e46387. 63 Mali RS, et al. 2008. FIZ1 Is part of the regulatory protein complex on active photoreceptor-specific gene promoters in vivo. BMC Mol Biol. 9:87. 64 The ENCODE Project Consortium. 2012. An integrated encyclope- dia of DNA elements in the human genome. Nature 489:57–74. 65 Oki S, et al. 2018. ChIP-Atlas: a data-mining suite powered by full integration of public ChIP-seq data. EMBO Rep. 19:e46255. 66 Xu H, Ang Y-S, Sevilla A, Lemischka IR, Ma’ayan A. 2014. Construction and validation of a regulatory network for pluripo- tency and self-renewal of mouse embryonic stem cells. PLoS Comput Biol. 10:e1003777. 67 Simonyan K, Zisserman, A. 2015. Very deep convolutional net- works for large-scale image recognition. in. doi:arXiv:1409. 1556v6 68 He K, Zhang X, Ren S, Sun J. 2015. Delving deep into rectifiers: surpassing human-level performance on ImageNet classifica- tion. Preprint at http://arxiv.org/abs/1502.01852. 69 Kuleshov MV, et al. 2016. Enrichr: a comprehensive gene set en- richment analysis web server 2016 update. Nucleic Acids Res. 44: W90–W97. 70 Blum M, et al. 2021. The InterPro protein families and domains database: 20 years on. Nucleic Acids Res. 49:D344–D354. 71 Stuart T, et al. 2019. Comprehensive integration of single-cell data. Cell 177:1888–1902.e21. 72 Fang R, et al. 2021. Comprehensive analysis of single cell ATAC-seq data with SnapATAC. Nat Commun. 12:1337. 73 Navarro Gonzalez J, et al. 2021. The UCSC Genome Browser data- base: 2021 update. Nucleic Acids Res. 49:D1046–D1057. 74 Bailey TL. 2021. STREME: accurate and versatile sequence motif discovery. Bioinformatics 37:2834–2840. 75 Khan A, et al. 2018. JASPAR 2018: update of the open-access data- base of transcription factor binding profiles and its web frame- work. Nucleic Acids Res. 46:D260–D266. 76 Hume MA, Barrera LA, Gisselbrecht SS, Bulyk ML. 2015. UniPROBE, update 2015: new tools and content for the online database of protein-binding microarray data on protein–DNA in- teractions. Nucleic Acids Res. 43:D117–D122. 77 Jolma A, et al. 2013. DNA-binding specificities of human tran- scription factors. Cell 152:327–339. 78 Gupta S, Stamatoyannopoulos JA, Bailey TL, Noble W. 2007. Quantifying similarity between motifs. Genome Biol. 8:R24.
10.1073_pnas.2219624120	RESEARCH ARTICLE \| NEUROSCIENCE OPEN ACCESS Structural basis for severe pain caused by mutations in the S4-S5 linkers of voltage-gated sodium channel NaV1.7 Goragot Wisedchaisria , Tamer M. Gamal El-Dina, Ning Zhenga,b,1 , and William A. Catteralla,1 Contributed by William A. Catterall; received November 16, 2022; accepted February 24, 2023; reviewed by Eduardo Perozo, Peter C. Ruben, and Vladimir Yarov-Yarovoy Gain-of-function mutations in voltage-gated sodium channel NaV1.7 cause severe inherited pain syndromes, including inherited erythromelalgia (IEM). The structural basis of these disease mutations, however, remains elusive. Here, we focused on three mutations that all substitute threonine residues in the alpha-helical S4-S5 intracel- lular linker that connects the voltage sensor to the pore: NaV1.7/I234T, NaV1.7/ I848T, and NaV1.7/S241T in order of their positions in the amino acid sequence within the S4-S5 linkers. Introduction of these IEM mutations into the ancestral bacterial sodium channel NaVAb recapitulated the pathogenic gain-of-function of these mutants by inducing a negative shift in the voltage dependence of activation and slowing the kinetics of inactivation. Remarkably, our structural analysis reveals a common mechanism of action among the three mutations, in which the mutant threonine residues create new hydrogen bonds between the S4-S5 linker and the pore-lining S5 or S6 segment in the pore module. Because the S4-S5 linkers couple voltage sensor movements to pore opening, these newly formed hydrogen bonds would stabilize the activated state substantially and thereby promote the 8 to 18 mV negative shift in the voltage dependence of activation that is characteristic of the NaV1.7 IEM mutants. Our results provide key structural insights into how IEM mutations in the S4-S5 linkers may cause hyperexcitability of NaV1.7 and lead to severe pain in this debilitating disease. sodium channel \| action potential \| channelopathies \| pain \| sensory nerves Voltage-gated sodium (NaV) channels initiate and propagate action potentials in nerves and muscles (1, 2), and mutations in NaV channels lead to many diseases of hyperexcit- ability (3, 4). NaV1.7 is expressed mainly in peripheral somatic and visceral sensory neurons in dorsal root ganglia and in sympathetic ganglion neurons, where it serves as a threshold channel that sets the gain of nociceptors for generation of pain signals (4, 5). Its slow rate of closed-state inactivation allows smaller and slower depolarizations to activate the channel and thereby enhances responses to small nociceptive stimuli (4, 6). Neuropathic pain syndromes affect millions of people and cost billions of dollars in health care expenditure (7, 8). They arise from damage in peripheral nerves from injury, diabetes, autoimmune disease, and cancer chemotherapy. However, rare inherited or spo- radic pain syndromes are caused by mutations in NaV1.7. Inherited erythromelalgia (IEM), paroxysmal extreme pain disorder, and small fiber neuropathy are caused by heterozygous gain-of-function mutations in NaV1.7 (4). Patients with these syndromes have episodes of intense pain, redness, and swelling in different parts of the body. No targeted treatment for these pain syndromes is available, but some pharmacological interventions are effective in selected individuals (9). On the other hand, homozygous loss-of-function mutations caused by gene truncation or deletion of NaV1.7 produce Congenital Insensitivity to Pain (CIP), a rare disorder in which patients do not feel pain despite having injuries (10). As a result, NaV1.7 is an attractive drug target for novel pain treatments (8), but the devel- opment of effective analgesics targeting NaV1.7 has been fraught with difficulties (11). NaV1.7 belongs to the family of nine human NaV channel subtypes (NaV1.1-NaV1.9), which are expressed in different excitable tissues (12, 13). NaV channels contain 24 trans- membrane (TM) segments that form four homologous, but not identical, 6-TM domains (DI to DIV). The first four TM segments in each homologous 6-TM domain (S1 to S4) form the voltage-sensing module (VS), which is connected via the alpha-helical S4-S5 intracellular linker to the pore module (PM) formed by the S5 and S6 segments and the P loop between them. Structural studies of the homotetrameric bacterial sodium channel NaVAb showed that these four homologous domains come together to form a functional channel with an ion-conducting pore at its center (13, 14). Depolarization activates NaV channels according to the “sliding-helix model”, in which Arg or Lys gating charges in the S4 segment move outward toward the extracellular side of the membrane upon Significance Gain-of-function mutations in voltage-gated sodium channel NaV1.7 cause the severe inherited pain syndrome inherited erythromelalgia (IEM). We introduced three IEM mutations into the ancestral bacterial sodium channel NaVAb, which substitutes threonine residues in the S4-S5 linker connecting the voltage sensor to the pore. These mutants recapitulated the pathogenic negative shift in voltage dependence of activation. Structural analysis revealed a common mechanism, in which new hydrogen bonds are formed between the S4-S5 linker and the pore. Because the S4-S5 linkers couple voltage sensor movements to pore opening, these newly formed hydrogen bonds would stabilize the activated state and promote hyperexcitability. These results provide key structural insights into how IEM mutations in the S4-S5 linkers may cause severe pain in IEM. Author contributions: G.W., T.M.G.E.-D., N.Z., and W.A.C. designed research; G.W. and T.M.G.E.-D. performed research; G.W. and T.M.G.E.-D. analyzed data; and G.W., T.M.G.E.-D., N.Z., and W.A.C. wrote the paper. Reviewers: E.P., The University of Chicago; P.C.R., Simon Fraser University; and V.Y.-Y., University of California Davis. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2219624120/-/DCSupplemental. Published March 30, 2023. PNAS 2023 Vol. 120 No. 14 e2219624120 https://doi.org/10.1073/pnas.2219624120 1 of 9 Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.depolarization (15, 16). Outward S4 movement triggers a major conformational change in the S4-S5 linkers on the intracellular side of the membrane and induces the opening of the activation gate of the pore by bending and rotation of the intracellular ends of the S6 segments (17). Similar voltage-gating mechanisms have been proposed based on structural studies of the bacterial sodium channel NaChBac (16), the voltage-sensitive phosphatase CiVSP from the ancestral eukaryote Ciona (18), the two-pore-domain channel TPC from plants and mammals (19), the sea urchin hyperpolarization- and cyclic nucleotide-gated channel (20), the human proton channel (21), the human ether-a-go-go chan- nel (22), and the mammalian sodium channel NaV1.7 (23–25). IEM mutations shift the voltage dependence of activation to more negative membrane potentials, allowing smaller depolari- zations to trigger an action potential (4). These disease mutations cluster in DI and DII. However, the molecular and structural basis for the pathogenic effects of mutations that result in pain phenotypes in patients remains unclear. Here, we have uncovered the structural basis for the gain-of-function phenotypes caused by three IEM mutations that substitute threonine (Thr) residues in the S4-S5 linkers of NaV1.7. Although the structure of NaV1.7 has been determined by cryo-EM (26, 27), progress toward resolving the structure with disease mutations at high resolution has been limited by challenges in obtaining sufficient quantity of NaV1.7 protein (28), intrinsic mobility of the VSs, and the limited resolution of the VS afforded by cryo-EM. To overcome these limitations, we introduced these IEM mutations into NaVAb, a bacterial ancestor of mammalian NaV channels, confirmed their functional effects in negatively shifting the voltage dependence of channel activation, and determined their structures at high resolution by X-ray crystallography (14). The backbone structures of NaVAb and NaV1.7 are similar within <3 Å rmsd in their core transmembrane domains, which is identical within the limit of resolution of the cryo-EM structures (14, 26, 27). Key structural elements involved in the activation of human NaV1.7 are transferable to other sodium channels (14, 23, 24, 29). Each mutation is represented four times in the homote- trameric structure of NaVAb, facilitating structure determination at high resolution by X-ray crystallography. This approach has revealed a common pathogenic mechanism in which the hydroxyl groups of substituted Thr residues form hydrogen bonds that stabilize the activated state of the voltage sensor and thereby negatively shift the voltage dependence of activation. Our results unveil an unexpected structural and physiological basis for NaV1.7 channelopathy caused by these three IEM mutations in the S4-S5 linker at near-atomic resolution. Results Function of NaVAb with NaV1.7 IEM Mutations in the S4-S5 Linker. We surveyed IEM mutations in human NaV1.7 in the literature and selected three mutations that all substitute Thr residues in the S4-S5 linkers. These three IEM mutations are located in the alpha-helical S4-S5 linkers that connect the VS and PM: I234T and S241T in NaV1.7 DI and I848T in DII, which are equivalent to NaVAb I119T, V126T, and L123T, respectively (Fig. 1A). The positions of these mutations in the S4-S5 linkers are illustrated in the side views of NaV1.7 and NaVAb in Fig. 1 B and Fig. 1. IEM mutations in voltage sensors of NaV1.7. (A) Topology diagram of NaV1.7 with seven IEM mutations (red circles) in the VSDs and the S4-S5 linkers of DI and DII selected for this study. Equivalent IEM mutations were introduced into the homotetrameric bacterial channel NaVAb (cyan) for electrophysiological and structural characterization. (B) Structure of human NaV1.7 (PDB: 6J8I) with the locations of three IEM mutations mapped onto the model (red spheres). Each domain is colored as in A. (C) Structure of NaVAb (PDB: 3RVY) with the locations of three IEM mutations mapped onto the model (red spheres). For clarity, the IEM mutations are shown only in one subunit (teal). Due to the tetrameric nature of NaVAb, the mutations are also present in the remaining subunits (cyan). Residues are labeled with domain colors of their equivalent mutations in NaV1.7 as in B. (D) View of IEM mutations (red spheres) in NaV1.7 from the cytosolic side. (E) View of IEM mutations (red spheres) in NaVAb from the cytosolic side. The structure of NaV1.7 from panel D is superimposed (with the IEM mutations shown as pink spheres) with an rmsd of ~2.0 Å for Cα atoms in the S4-S5 linkers and PM. 2 of 9 https://doi.org/10.1073/pnas.2219624120 pnas.org Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.C, respectively. As viewed from the cytosolic side, the S4-S5 linkers of NaV1.7 surround the activation gate formed by the intracellular ends of the S6 segments (Fig. 1D). They are located in analogous positions in each of the four S4-S5 linkers in NaVAb (Fig. 1E). The genetic background, phenotypic properties and biophysical effects as determined in previous studies of these mutations in NaV1.7 are summarized in SI Appendix, Tables S1 and S2. We introduced each of these S4-S5 linker IEM mutations into the C-terminal-truncated construct NaVAbΔ28, which was then expressed and analyzed by whole-cell voltage clamp (30; Materials and Methods). All of these constructs conduct voltage-gated inward sodium currents that activate rapidly and inactivate more slowly than wild-type (WT) NaVAbΔ28 (Fig. 2 A–C). Their rates of activation are equal to or more rapid than NaVAb/WT in the critical voltage range from −50 mV to +10 mV, as illustrated in the plots of time-to-peak current vs. voltage in Fig. 2B. Similarly, these mutations increase the time constants for inactivation in the physiologically important voltage range of −60 to +10 mV (Fig. 2C), consistent with their hyperexcitability phenotype. In addition, the voltage depend- ence of activation (Va) for these mutants is negatively shifted to dif- ferent extents as expected for IEM mutations (Fig. 2D). Surprisingly, the conductance/voltage (G/V) curves for mutants I119T and V126T are biphasic (Fig. 2D). Peak conductance in these G/V exper- iments is reached in 1 ms or less, raising the possibility that the voltage-dependent activation process of NaVAb has not reached steady state in that short time. Therefore, we repeated steady-state activation experiments for these two mutants by depolarizing to a series of test voltages for 10 ms, hyperpolarizing to a fixed negative voltage (−180 mV), and measuring the peaks of the resulting inward tail currents as a measure of pore opening (Fig. 2E). The resulting tail current G/V curves are monophasic and show a clear negative shift of Va, consistent with the hypothesis that these mutants do indeed require 1 ms to 10 ms of depolarization to reach steady-state activation. We propose a structural basis for these unexpected find- ings in the Discussion. Overall, the rapid activation, slowed inacti- vation, and negatively shifted Va of these mutants are consistent with their hyperexcitability phenotype in native NaV1.7 in vitro in trans- fected cells and in vivo in affected individuals. Structures of NaVAb with IEM Mutations in S4-S5 Linkers. We solubilized, purified, and crystallized the mutant NaVAb proteins. All structures were determined for channels embedded in lipid bicelles by X-ray crystallography at a resolution ranging from 2.7 Å to 3.1 Å, which yielded near atomic clarity in these cases (SI Appendix, Table S3). All of these structures have their VSs in the activated state, as determined by the outward position of the gating charges in the S4 segment, while the PM is closed at the activation gate formed by the intracellular ends of the S6 segments. Because the IEM Fig. 2. Functional properties of NaVAb with IEM mutations. (A) Families of sodium currents. NaVAb/Δ28 WT and mutants were expressed in Hi5 insect cells and studied by whole-cell voltage clamp recording as described in Materials and Methods. (B) Time to reach peak current is plotted as a function of stimulus voltage as calculated from the results of panel A. (C) Mean time constants for inactivation calculated from the results of experiments similar to those in panel A as described in Materials and Methods. (D) Voltage dependence of activation of WT and the indicated IEM mutants in the S4-S5 linker. Current/voltage (I/V) relationships of the peak sodium currents were recorded in response to steps to voltages ranging from −160 mV to +60 mV in 10-mV increments from a holding potential of −160 mV or −180 mV. Conductance/voltage (G/V) relationships were calculated as described in Materials and Methods. (E) Voltage-dependent activation curves determined from peak tail current amplitudes measured at −180 mV. Short depolarizing pulses, 10 ms, were applied from a holding potential of −180 mV in 10-mV steps. Tail currents were normalized to the highest amplitudes. (Inset) Representative traces of inward sodium currents and tail currents for WT, I119T, and V126T. PNAS 2023 Vol. 120 No. 14 e2219624120 https://doi.org/10.1073/pnas.2219624120 3 of 9 Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.mutations primarily affect the activation process, we also generated structural models of the NaVAb IEM mutants in the resting state, using the disulfide-crosslinked resting-state structure of NaVAb as a template (17). This gave insight into the conformational transitions that are altered by the IEM mutations during activation of the VS. Notably, all three of these amino acid substitutions contain hydroxyl side chains of Thr residues that form new hydrogen bonds in our structures, which would stabilize the voltage sensor in its activated conformation. In the sections below, we consider these three IEM mutations in their order in the amino acid sequence of the S4-S5 linker of NaVAb: I119T, L123T, and V126T (Fig. 1C). NaVAb/I119T (NaV1.7/I234T). NaVAb/Ile119 is analogous to Ile234 in the S4-S5 linker in DI of NaV1.7, which is moderately conserved by replacement with an amino acid residue with a hydrophobic side chain of medium size in all four domains (SI Appendix, Fig. S1). NaVAb/I119T shows a surprisingly large shift of −37.5 mV in Va (Fig. 2 D and E and SI Appendix, Table S2). This shift of Va is greater than the shift of −17.9 mV for NaV1.7/I234T, which was the largest negative shift among all IEM mutations in NaV1.7 (31). The structure of NaVAb/I119T at 2.75-Å resolution (Fig. 3A) clearly indicates that the side chain of I119T forms a hydrogen bond with Ser132 on S5 of a neighboring subunit (referred to as Ser132′) (Fig. 3 A and B). Although the overall structure is similar to WT (Cα rmsd = 0.7 Å), the conformations of the hinge region between the S4-S5 linker and the S5 segment where S132′ is located, as well as the activation gate receptor region of the pore-lining S6 segment, differ from the WT structure significantly (Fig. 3C). In the S4-S5 linker/S5 hinge region, the Cα atoms of Fig. 3. Structure of NaVAb/I119T. (A) Crystal structure of NaVAb/I119T with the mutation shown. (B) Close-up view of the NaVAb/I119T structure with the VS in the activated state (teal and cyan for different subunits). Side chains of residues I119T (red), I217 (teal) and key side chains in S5 and S6 of a neighboring subunit (cyan) including Ser132′ and Asn211′ are shown as sticks with the σA-weighted 2FO-FC electron density map contoured at 1.0σ level overlaid (mesh). (C) Comparison of NaVAb/I119T structure with the structure of NaVAb WT (light orange). Side chain of I119T forms a hydrogen bond (black dashes) with side chain of Ser132′. Conformational changes in the C-terminal end of the S6 segments starting from Asn211′ in the mutant structure abolish the CHAPSO binding site present in the WT structure. (D) Homology model of NaVAb/I119T structure in the resting state. The van der Waals hemispheres are shown as dots. NaVAbΔ28 (PDB: 6MWA) was used for the comparison. Ile119 and Ser132′ are ~8 Å apart in the WT structure (Fig. 3C). This distance is shortened to ~7 Å in the I119T mutant structure due to the formation of a strong hydrogen bond (~2.5 Å) between the oxygen atoms of the I119T and Ser132′ side chains (Fig. 3C). Notably, the methyl group of the Thr side chain retains the hydro- phobic interaction of Ile119 with Val133 in S5, further strength- ening the interaction of these two side chains (Fig. 3B). In human NaV1.7, Asn857 in DII S5 is the equivalent residue to NaVAb Ser132′; therefore, the I234T mutation in the S4-S5 linker of DI of NaV1.7 is well positioned to form a hydrogen bond with the side chain of Asn857 in DII-S5 in the activated state, similar to NaVAb/Ser132′. This newly formed hydrogen bond would stabilize the conformation of the activated state and thereby contribute to the negative shift in Va. The S6 segment structure also exhibits a conformational change as it approaches the activation gate in NaVAb/I119T, which starts with a kink around the conserved Asn211 residue and propagates by a rotation that causes a larger shift of ~4 Å at the last residue of S6 (Ala220) (Fig. 3C). This structural change may be induced by the formation of the hydrogen bond between I119T and Ser132′ and may contribute to its stabilization of the activated state and its unusually large negative shift of activation. The conformational change observed here disrupts two binding sites for CHAPSO detergent molecules from the bicelles that normally wedge between the S4-S5 linker and the pore-lining S6 segment in the WT struc- ture, such that no electron density for the CHAPSO molecules was present in the mutant structure. Despite the conformational change in the activation gate, the pore structure of the mutant remains closed, with Ile217 forming a constriction site that seals the acti- vation gate of the pore (Fig. 3B), similar to the WT structure. The resting-state model of NaVAb/I119T suggests that the I119T side chain forms weaker van der Waals interactions with the outer surface of the pore module from the neighboring sub- unit. Notably, the I119T side chain makes van der Waals contacts with the conserved Asn211′ residue from S6 with a distance of ~4 Å (Fig. 3D), similar to that for the native Ile residue found in the resting state of NaVAb. Therefore, there is no evidence for a significant gain in stability by this mutation in the resting state. The gain of the hydrogen bond in the activated state from the I119T mutation, plus the conformational change induced in the S6 segment along with hydrogen bond formation, likely provides extra free energy that stabilizes the activated state and leads to channel hyperexcitability (Movie S1). NaVAb/L123T (NaV1.7/I848T). NaVAb/Leu123 is analogous to Ile848 in DII of NaV1.7, which is conservatively replaced by Leu in the other three domains of NaV1.7 (SI Appendix, Fig. S1). This conservation of Leu and Ile residues suggests an important role for a hydrophobic side chain at this position. Electrophysiological recordings of NaVAb/L123T show a shift of −15.5 mV in Va (Fig. 2D and SI Appendix, Table S2), similar to negative shifts in Va of −13.8 mV to −7.5 mV in previous studies of NaV1.7/I848T (32–34). The structure of NaVAb/L123T in the activated state at 2.7-Å resolution indicates that the substituted Thr side chain forms a hydrogen bond with the highly conserved Asn211 side chain in the S6 segment of a neighboring subunit (referred to as Asn211′) (Fig. 4 A and B). Asn211 is a conserved residue that is important for electromechanical coupling and critical for the gating process by bridging interactions between the S4-S5 linker and the S6 activation gate (17). In the WT structure, Asn211′ forms part of the wall of the ion-conducting pore with its side chain pointing toward the lumen, while also interacting with Val126 in the S4-S5 linker (Fig. 4C). In the L123T structure, the Asn211′ side chain adopts a different rotamer that points away from the lumen of the 4 of 9 https://doi.org/10.1073/pnas.2219624120 pnas.org Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.for NaV1.7/S241T (35). These results indicate that NaVAb/V126T is a valid structural model for this IEM mutation. The crystal structure of NaVAb/V126T at 3.1-Å resolution (Fig. 5A) suggests that the V126T side chain forms a long-range hydrogen bond with the side chain of the Asn211′ residue in the S6 segment of a neighboring subunit (Fig. 5B). The V126T side chain rotates almost 180° from the Val side chain in the WT structure, twisting around the axis of the Cα–Cβ bond to reorient the oxygen atom toward the Asn211′ side chain with a distance of ~3.4 Å while maintaining van der Waals interaction with the side chain of Ile216 on S6 (Fig. 5C). In the WT structure, Val126 interacts with Asn211′ through van der Waals interactions with a distance of ~4 Å. The V126T mutation changes the chemistry of the interaction to a hydrogen bond while preserving the volume of the side chain and the hydrophobic interaction with Ile216. The hydrogen bond between Val126T and Asn211, which is absent in the resting-state homology model (Fig. 5D and (17)), likely enhances activation by providing additional stability to the activated state and thereby leads to channel hyperexcitability (Movie S3). Modeling the Mechanism of Action of IEM Mutations in NaV1.7 Channels. NaV1.7 has four distinct domains rather than four identical subunits as in NaVAb. Similar mammalian NaV channels have complex voltage-dependent gating mechanisms, typically involving four resting states, four resting/inactivated states, one open state, one fast inactivated state, and one or more slow inactivated states (36). The available cryo-EM structures of NaV1.7 are most likely to be in slow inactivated states, considering that the protein has been in a depolarized environment for an Fig. 5. Structure of NaVAb/V126T. (A) Crystal structure of NaVAb/V126T with the mutation shown. (B) Close-up view of NaVAb/V126T structure with the VS in the activated state (teal and cyan for different subunits). Side chains of residues V126T (red), Ile216 (teal), and Asn211’ (cyan) are shown as sticks with the σA-weighted 2FO-FC electron density map contoured at 1.0σ level overlaid (mesh). (C) Comparison of NaVAb/V126T structure with the structure of NaVAb WT (light orange). Side chain of V126T rotates to form a hydrogen bond (black dashes) with side chain of Asn211’. (D) Homology model of NaVAb/ L123T structure in the resting state. The van der Waals hemispheres are shown as dots. Fig. 4. Structure of NaVAb/L123T. (A) Crystal structure of NaVAb/L123T with the mutation shown. (B) Close-up view of NaVAb/L123T structure with the VS in the activated state (teal and cyan for different subunits). Side chains of residues L123T (red) and Asn211′ (cyan) are shown as sticks with the σA- weighted 2FO-FC electron density map contoured at 1.0σ level overlaid (mesh). (C) Comparison of NaVAb/L123T structure with the structure of NaVAb WT (light orange). Side chain of N211′ makes a van der Waals contact with the side chain of V126 in the WT structure but rotates to form a hydrogen bond (black dashes) with side chain of L123T in the mutant structure. The van der Waals hemispheres are shown as dots. (D) Homology model of NaVAb/L123T structure in the resting state. pore toward the S4-S5 linker and forms a hydrogen bond with a distance of ~3.1 Å to the hydroxyl oxygen of the L123T side chain (Fig. 4 B and C). The hydrogen bond formed between these two residues from the S4-S5 linker and the S6′ segment would provide additional stabilization of the activated state and therefore signif- icantly shift voltage-dependent activation to more negative mem- brane potentials. Modeling NaVAb/L123T in the resting state suggests that the side chain of L123T makes van der Waals interactions with the conserved Asn211' side chain of an adjacent subunit (Fig. 4D). In contrast, the native Leu side chain in the NaVAb resting state makes an additional hydrophobic interaction with the conserved Phe207’ residue located one helical turn from Asn211′. This inter- action is absent in the L123T structure as the length of the side chain is reduced by the mutation, suggesting destabilization of the resting state by this mutation. In human NaV1.7, Phe1435 and Asn1439 in DIII S6 are found at the position equivalent to NaVAb/Phe207 and NaVAb/Asn211, respectively. Similarly, the I848T mutation likely abolishes the hydrophobic interaction with Phe1435 in the resting state, while creating a new hydrogen bond with Asn1439 in the activated state. Therefore, the loss of the hydrophobic interaction in the resting state and the gain of a hydrogen bond in the activated state both likely contribute to the large negative shift in Va (Movie S2). NaVAb/V126T (NaV1.7/S241T). NaVAb/Val126 is analogous to NaV1.7/Ser241. It is conserved in DI, DII, and DIV of NaV1.7 but is replaced by Ala in DIII (SI Appendix, Fig. S1). Electrophysiological recordings of NaVAb/V126T showed a shift of −12 mV in Va (Fig. 2 D and E and SI Appendix, Table S2), similar to the shift of −8.4 mV PNAS 2023 Vol. 120 No. 14 e2219624120 https://doi.org/10.1073/pnas.2219624120 5 of 9 Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.Fig. 6. Potential hydrogen bond formation between mutant IEM residues in the S4-S5 linker and the PM of NaV1.7. (A) NaV1.7/I234T. (B) NaV1.7/S241T. (C) NaV1.7/I848T. extended time (25, 26). Nevertheless, it is of interest to explore the potential functional relevance of the mechanism of action of the IEM mutations revealed here by modeling their impact on the available NaV1.7 structures (26) (Fig. 6). Of the three substituted Thr residues, we have studied, NaV1.7/I234T in the DI S4-S5 linker fits our proposed mechanism most closely. As in NaVAb, the mutant Thr residue is predicted to form a high-energy hydrogen bond with nearby Asn857 in the DII S5 segment of NaV1.7, while maintaining van der Waals interaction with Leu858 (Fig. 6A). Similarly, the nearby residue NaV1.7/S241T is predicted to form a lower energy hydrogen bond with Asn950 in DII S6 of NaV1.7 (Fig. 6B). In contrast, NaV1.7/I848T in the S4-S5 linker in DII is rotated away from the potential hydrogen-bonding residue, Asn1439 in DIII S6 in the available NaV1.7 structure (Fig. 6C). However, it is only one helical turn away from making this contact if the S4-S5 linker were to rotate counterclockwise by 60° to 90° from its position in the NaV1.7 inactivated state structure. Therefore, it is entirely possible that these two residues in NaV1.7 move near each other as the S4-S5 linker rotates during the activation process, similar to the small rotation of the S4-S5 linker in our gating model of NaVAb (Movie S2 and (17)). This subtle movement would allow hydrogen bond formation to stabilize the activated state of NaV1.7. Discussion Our study provides the first structural characterization of IEM mutations of human NaV1.7 channels and gives insight into the potential molecular basis for hyperexcitability in three sets of affected IEM families (SI Appendix, Table S1). We focused on the S4-S5 linkers of NaV1.7 where IEM mutations are clustered and coupling of voltage-dependent activation to pore opening takes place (15). By expressing these IEM mutations in the context of the ancestral sodium channel NaVAb, we were able to isolate suf- ficient quantity of protein in the homogeneous form to allow crystallization and high-resolution structure determination of each mutation in its native state with an activated VS. In addition, because NaVAb is the only sodium channel whose resting-state structure is known (17), we were able to use molecular modeling methods to define the effects of IEM mutations on the structures of both resting and activated states. By comparing the structural effects of these mutations in these two distinct states of the VS, a unified mechanism emerged for these three IEM mutants, which suggests a plausible structural basis for their hyperexcitability. IEM Mutations in the S4-S5 Linker Form Hydrogen Bonds with the Pore Module of NaVAb. The three IEM mutations we studied cause little or no change in the conformation of the S4-S5 linker. However, our structures indicate that these IEM mutations provide extra stability to the activated state by forming a new hydrogen bond with a residue from the PM of the adjacent subunit when the channel is in the activated state (Ser132 on S5 for I119T; Asn211 on S6 for L123T and V126T). All three of these mutated residues have been proposed to be important for electromechanical coupling and pore gating of the channel, based on the structure of NaVAb in the resting and activated states (17). These hydrogen bonds are well-positioned to stabilize the conformation of the VS in the activated state. In contrast, these mutations do not form a hydrogen bond with any residue in the resting state, in which the S4-S5 linker protrudes as an elbow into the cytosol (17). The fact that mutations of these residues cause hyperexcitability that leads to severe pain in IEM further highlights the critical role of these residues in channel gating. IEM Mutations Differentially Stabilize the Activated States of NaVAb and NaV1.7. When we introduced these IEM mutations into NaVAb, they caused negative shifts in Va by −12 mV to −37.5 mV, with the mutation equivalent to I234T causing the largest change. For all three of these IEM mutations, the negative shift of Va in NaVAb is significantly greater than that for the same mutation in NaV1.7 (SI Appendix, Table S2). In the most prominent example, Va for NaVAb/I119T is −37.5 mV compared to −17.9 mV for Va in NaV1.7/I234T (SI Appendix, Table S2). In comparison, the shift of Va in NaVAb/L123T is −15.5 mV compared to −10.4 mV (mean of n = 3 published studies; SI Appendix, Table S2) for Va in NaV1.7/I848T, whereas the shift of Va in NaVAb/V126T is −12 mV compared to −8.4 mV for Va in NaV1.7/S241T (SI Appendix, Table S2). It is likely that two distinct factors contribute to these differences in Va. First, the energy of hydrogen bonds diminishes with their length, and the bond lengths for NaVAb/I119T, NaVAb/ L123T, and NaVAb/V126T are approximately 2.5 Å, 3.1 Å, and 3.4 Å, respectively, providing a structural basis for their different shifts of Va based on their hydrogen bond length. Second, we consider it likely that the shifts of Va are larger in NaVAb because of conformational coupling among the four copies of the mutation in the tetramer, and this coupling energy may be greater for NaVAb/ I119T than for the other two mutants. IEM Mutations Are Well-Positioned to Form Hydrogen Bonds in NaV1.7. The I234T, S241T, and I848T mutations are located in the S4-S5 linkers in DI and DII of NaV1.7. Models of these mutations in the structure of NaV1.7 directly support the formation of hydrogen bonds in the activated state for two of these three mutations, and the third mutation is in position to form a potential new hydrogen bond with modest structural change during the activation process (Fig. 6). These newly formed hydrogen bonds could contribute up to 6 kcal/mol to stabilize the activated state in NaV1.7, which would make a substantial contribution to the free energy required for the 8 mV to 18 mV negative shifts in Va of these mutants in NaV1.7 (SI Appendix, Table S2). Stabilization of the activated state would 6 of 9 https://doi.org/10.1073/pnas.2219624120 pnas.org Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.lead to voltage-sensor trapping, similar to the effects of neurotoxins that negatively shift Va or block fast inactivation by binding with high affinity to the activated state of NaV channels (37, 38). Voltage sensor trapping in the activated state leads to prolonged trains of action potentials, similar to those induced by IEM mutations (4, 31). Interestingly, recent studies of phosphorylation of NaV1.7/I848T also showed a negative shift of −9.8 mV in Va mediated by Protein Kinase C (34), which would further enhance the negative shift of Va caused by this mutation that we have observed in NaVAb (ΔV1/2 of −15.5 mV). Thus, our high-resolution information on these focused structural changes induced by IEM mutations gives unprecedented insights into IEM pathogenesis at near-atomic resolution and provides a potential molecular template for mutation-specific therapy of this debilitating disease. Biphasic Kinetics of IEM Mutants Suggests Hydrogen Bond Formation During Activation. Mutations NaVAb/I119T and NaVAb/V126T give two-component Boltzmann fits of their activation curves, as if the voltage sensor activates in two steps (Fig. 2D). These results raise the possibility that the formation of hydrogen bonds with Ser132' and Asn211' is incomplete, resulting in a major fraction of the channels that open with negatively shifted voltage dependence stabilized by hydrogen bond formation and a minor fraction of channels that do not form hydrogen bonds successfully during the brief stimulus pulses. Although our stimulus pulses were 50 ms in duration, actual measurements of peak sodium currents were made at stimulus times less than 1 ms at the most positive stimulus potentials (Fig. 2 A–D). To test whether these short stimulus durations are responsible for the biphasic activation curves, we measured G/V curves from tail currents. In this protocol, cells were depolarized to different membrane potentials during a prepulse of 10 ms, and inward sodium currents through the channels that have opened were measured during a hyperpolarizing pulse to −180 mV (Fig. 2E). During this protocol, the VS have up to 10 ms to activate at all stimulus potentials. Remarkably, the biphasic activation curves were resolved to approximately single Boltzmann fits with this protocol (Fig. 2E), confirming that biphasic activation curves were caused by the short times available to reach peak current in the standard G/V curve protocol. This striking feature of our data provides further support for the importance of hydrogen bond formation to stabilize the activated state of the VS and cause a negative shift in Va and further suggests that these amino acid side chains can align and hydrogen bonds can form completely in less than 10 ms. Slow Inactivation of IEM Mutants in NaVAb. Mutations that cause IEM have variable effects on slow inactivation of NaV1.7 channels, suggesting that these effects are not central to the IEM phenotype but may modify it in individuals with specific mutations (4, 7). Slow inactivation of mammalian sodium channels involves conformational movements of the outer pore region (39–41). Bacterial NaV channels have a slow inactivation process that causes pore closure, similar to the slow inactivation of mammalian NaV channels, and analogous to conformational changes in transmembrane closure of the pore in the slow inactivation of the bacterial potassium channel KcsA (42–45). NaVAb has a complex three-phase mode of slow inactivation involving movements of the selectivity filter and outer pore as well as asymmetric closure of the intracellular activation gate formed by the S6 segments (30, 44). Consistent with previous studies, we found that the IEM mutations we studied here primarily slowed the rate of inactivation in the physiologically important membrane potential range from −60 mV to 0 mV (Fig. 2C). These effects on slow inactivation of IEM mutations expressed in NaVAb are consistent with their hyperexcitable phenotype in vivo. However, individual IEM mutations have a range of effects from little or no change to substantial delay of slow inactivation during trains of action potentials (4, 7); therefore, they probably do not contribute in a major way to the disease phenotype in most individuals with mutations in human NaV1.7 (46). Negatively shifted Va, stabilized by the formation of new hydrogen bonds, is most likely to be the underlying mechanism for hyperexcitability for the IEM mutations studied here. Materials and Methods Materials Availability. Further information and requests for resources and rea- gents should be directed to the corresponding author William A. Catterall (wcatt@ uw.edu). All reagents generated in this study are available from the corresponding author with a completed Materials Transfer Agreement. Microbe Strains. Escherichia coli GC10 was cultured at 37 °C in an LB medium supplemented with 100 mg/mL of ampicillin for plasmid DNA extraction. E. coli DH10Bac was cultured at 37 °C in LB medium supplemented with 50 mg/mL kanamycin sulfate, 7 mg/mL gentamicin, and 10 mg/mL tetracycline for bacmid production. Cell Lines. Sf9 (Spodoptera frugiperda) insect cells were maintained in Grace’s Insect Medium and supplemented with 8 to 10% fetal bovine serum (FBS) and penicillin/streptomycin at 27 °C and passaged at 80 to 95% confluence for bacu- lovirus production. Hi5 (Trichoplusia ni) insect cells were maintained and infected in Grace’s Insect Medium supplemented with 8 to 10% FBS and glutamine/peni- cillin/streptomycin at 27 °C for electrophysiology and protein expression. Mutagenesis and Baculovirus Production of NaVAb with IEM Mutations. The pFastBac-NaVAbΔ28 plasmid carrying a C-terminal 28-residue truncation (residues 1 to 239) of NaVAb gene (30) was used as a template for site-directed mutagenesis. To generate a desired plasmid of a NaVAb mutant, overlapping oligonucleotide prim- ers with a codon changed to a specific mutation were synthesized (Integrated DNA Technologies) and used for site-directed mutagenesis PCR with PfuUltra II Fusion HotStart DNA Polymerase (Agilent). The PCR products were treated with DpnI (New England Biolabs) and transformed into E. coli GC10 competent cells (Genesee Scientific). Plasmid DNAs were isolated from transformed colonies using the QIAprep Spin Miniprep Kit (Qiagen) according to the manufacturer’s protocol and sequenced to confirm the mutation. Plasmids containing NaVAbΔ28 with the IEM mutation were used to transform E. coli DH10Bac competent cells for bacmid production and bacu- loviruses were prepared with Sf9 insect cells using the Bac-to-Bac protocol according to the manufacturer (Life Technologies). Electrophysiology of NaVAb with IEM Mutations. Hi5 insect cells (Trichoplusia ni) were infected with baculovirus containing NaVAbΔ28 with an IEM mutation (30). After 24 to 48 h, whole-cell sodium currents were recorded using an Axopatch 200 amplifier (Molecular Devices) with glass micropipettes (1.5 to 2.5 MΩ) (30). The intracellular pipette solution contained (in mM): 35 NaCl, 105 CsF, 10 EGTA, 10 HEPES, pH 7.4 (adjusted with CsOH). The extracellular solution contained (in mM): 140 NaCl, 2 CaCl2, 2 MgCl2, 10 HEPES, pH 7.4 (adjusted with NaOH). Linear capacitance was subtracted and 80 to 90% of series resistance was com- pensated using internal amplifier circuitry. I/V relationships of the peak currents were recorded in response to steps to voltages ranging from −160 mV to +160 mV in 10-mV increments from a holding potential of −160 mV or −180 mV. In the case of voltage-dependent activation curves determined from peak tail current amplitudes measured at −180 mV, 10-ms depolarizing pulses were applied from a holding potential of −180 mV in 10-mV steps. Tail currents were normalized to the highest amplitudes. Pulses were generated and currents were recorded using Pulse software controlling an InstruTECH ITC-18 interface (HEKA). Data were analyzed using Igor Pro 8 (WaveMetrics). Analysis of Electrophysiological Data. Voltage-clamp data were analyzed using IGOR Pro 8 (WaveMetrics). Peak current at each voltage of the current family was plotted as a function of the stimulus voltage to visualize the current vs. voltage (I/V) relationship. For WT and mutant L123T, normalized conductance/voltage (G/V) curves were calculated from the I/V curves and fit with a simple one-component Boltzmann equation: 1/(1+exp(Va−Vm)/k) in which Vm is the stimulus potential, Va is PNAS 2023 Vol. 120 No. 14 e2219624120 https://doi.org/10.1073/pnas.2219624120 7 of 9 Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.the half-activation voltage, and k is a slope factor. Similarly, for WT, I119T, and V126T, G/V curves were determined from peak tail current amplitudes by fit to this simple one-component Boltzmann relationship. In contrast, biphasic G/V curves for I119T and V126T were calculated from I/V curves and fit with a two component Boltzmann equation: y = yo + A p Va1−Vm k1 1 + e [ + 1 − p 1 + e Va2−Vm k2 ] where y0 and A are the minimum and the maximum values taken by y. p is the fraction of the curve comprising first sigmoidal curve, (1 − p) is the fraction of the curve comprising the second sigmoidal curve, and Va1, Va2, k1, and k2 are the voltage midpoints and slope factors of the two phases. The results for the I119T mutation fit well to this two-component Boltzmann equation with Va1 = −127.6 mV, p = 0.76, k1 = 5.34 mV, Va2 = −64 mV, 1-p = 0.24, and k2 = 8.3 mV. However, the double-sigmoidal curve did not fit well for the second phase of activation of the V126T mutation, so only the first component was fitted with Va = −102 mV, k = 5.6 mV. The half-time (τ) for inactivation was measured from the peak of each current trace. A single exponential function was used to fit the inactivation kinetics of all three mutants. The data were presented as mean and SEM. Statistical significance was eval- uated with Student’s t test. Protein Expression and Purification of NaVAb with IEM Mutations. Third passage (P3) baculoviruses for NaVAbΔ28 with an IEM mutation were used to infect Hi5 cells in single-layer culture dishes (~12 dishes for each preparation), and the cells were incubated at 27 °C for ~72 h. The cells were harvested by centrifugation, and the pellets were resuspended in Buffer A (50 mM Tris HCl pH 7.5 and 200 mM NaCl) supplemented with 1 mM PMSF, 2× SigmaFast protease inhibitor cocktails (MilliporeSigma), benzamidine HCl, and DNase I. Cells were lysed by sonication and membranes were solubilized with 1% high-purity digitonin (MilliporeSigma) for 1 h at 4 °C with gentle mixing. The mixture was centrifuged at 15,000×g for 30 min at 4 °C, and the supernatant was incubated with Anti-FLAG-M2 affinity gel (MilliporeSigma) for 1 h at 4 °C with gentle mixing. The resins were washed with Buffer B (Buffer A supplemented with 0.12% digitonin) and bound protein was eluted with Buffer C (Buffer B supplemented with 200 μM FLAG peptide). Eluted protein was concentrated to 1 mL using Vivaspin20 100 kDa MWCO (Cytiva) and further purified with Superdex S200 size-exclusion chromatography (Cytiva) using 10 mM Tris HCl pH 7.5, 100 mM NaCl, and 0.12% digitonin as a column running buffer. Elution fractions were evaluated using SDS-PAGE, and peak fractions were combined and concentrated to final concentration of ~20 mg/ml. Protein concentrations were estimated using 1 A280 absorbance unit = 1 mg/mL on a NanoDrop spectrophotometer. Crystallization of NaVAb with IEM Mutations. NaVAb-bicelle complexes were prepared by mixing NaVAbΔ28 with an IEM mutation with 10% bicelle (7.5% w/v DMPC and 2.5% w/v CHAPSO) at 1:4 or 1:5 volume ratios to obtain final 3.0 to 3.5 mg/ml protein concentration. The complexes were screened for crystalli- zation conditions under 1.70 to 1.95 M ammonium sulfate and 0.1 M sodium citrate tribasic, pH 4.6 to 6.0 (Hampton Research). Crystals were cryo-protected by stepwise transfers to a series of cryo-protectant solutions containing 6 to 30% glucose (6% increments) with the same concentration of ammonium sulfate and sodium citrate in which the crystals were grown. X-ray Data Collection and Structure Determination of NaVAb with IEM Mutations. Crystals were tested for diffraction, and data were collected at Advanced Light Source beamline 8.2.1 and 8.2.2 Howard Hughes Medical Institute. Diffraction data were processed using the HKL2000 program (47) with anisotropic scaling and truncation due to anisotropic diffraction. Structures were solved by molecular replacement with PHASER (48) using the previously determined NaVAb structure PDB: 3RVY (14) or PDB: 6MW (30) as a search model and refined with REFMAC (49) in the CCP4 program suite (50). Manual model building and local real space refine- ment were carried out in COOT (51), followed by structure refinement in REFMAC. Subsequently structures were refined with Phenix.refine module and the final struc- tures were analyzed and validated using MolProbity in the Phenix program suite (52) (SI Appendix, Table S3). Modeling of NaVAb Structure with IEM Mutations in the Resting State. Homology models of NaVAb tetramer structure with the IEM mutations in the resting state were generated with Modeller 10.2 (53) using the disulfide-locked NaVAb/KAV/G94C/Q150C cryo-EM structure (PDB: 6P6W) (17) as a template. Structure figures and morph movies were made with Pymol (Schrodinger). Modeling of NaV1.7 Structure with IEM Mutations. Cryo-EM structure of human NaV1.7 (PDB: 7W9K) was used for modeling in Coot (51). Residues with IEM mutations were mutated to threonine and the most probable side chain rotamer without clashes was selected. Different rotamers of residues that poten- tially form hydrogen bond pairs with the IEM mutations were tested and a rotamer within a hydrogen bond distance without clashes was selected. Data, Materials, and Software Availability. The coordinates and the structure factors for the reported crystal structures have been deposited in the Protein Data Bank (PDB) under accession codes 8DIZ (54), 8DJ0 (55), and 8DJ1 (56) for NaVAbΔ28 I119T, L123T, and V126T, respectively. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. We thank the beamline staff at the Advanced Light Source (BL8.2.1 and BL8.2.2) for assistance during X-ray data collection and Dr. Jin Li (Department of Pharmacology, University of Washington) for the tech- nical and editorial support. This research was supported by NIH research grants R01 NS015751 (W.A.C.), R35 NS111573 (W.A.C.), and R01 HL112808 (N.Z. and W.A.C.) and by the Howard Hughes Medical Institute (N.Z.). The Berkeley Center for Structural Biology is supported in part by the Howard Hughes Medical Institute. The Advanced Light Source is a Department of Energy, Office of Science User Facility under Contract No. DE-AC02-05CH11231. This research also used resources of the Advanced Photon Source, a U.S. Department of Energy Office of Science user facility operated by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. Author affiliations: aDepartment of Pharmacology, University of Washington, Seattle, WA 98195; and bHHMI, University of Washington, Seattle, WA 98195 1. 2. A. L. Hodgkin, A. F. Huxley, A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117, 500–544 (1952). R. H. Adrian, W. K. Chandler, A. L. Hodgkin, Voltage clamp experiments in striated muscle fibres. J. Physiol. 208, 607–644 (1970). 9. P. Geha et al., Pharmacotherapy for pain in a family with inherited erythromelalgia guided by genomic analysis and functional profiling. JAMA Neurol 73, 659–667 (2016). 10. J. J. Cox et al., An SCN9A channelopathy causes congenital inability to experience pain. Nature 444, 894–898 (2006). 3. M. Mantegazza, S. Cestèle, W. A. Catterall, Sodium channelopathies of skeletal muscle and brain. 11. J. V. Mulcahy et al., Challenges and opportunities for therapeutics targeting the voltage-gated 4. Physiol. Rev. 101, 1633–1689 (2021). S. D. Dib-Hajj, Y. Yang, J. A. Black, S. G. Waxman, The Nav1.7 sodium channel from molecule to man. Nat. Rev. Neurosci. 14, 49–62 (2013). 6. 5. N. Klugbauer, L. Lacinova, V. Flockerzi, F. Hofmann, Structure and functional expression of a new member of the tetrodotoxin-sensitive voltage-activated sodium channel family from human neuroendocrine cells. EMBO J. 14, 1084–1090 (1995). T. R. Cummins, J. R. Howe, S. G. Waxman, Slow closed-state inactivation: A novel mechanism underlying ramp currents in cells expressing the hNE/PN1 sodium channel. J. Neurosci. 18, 9607–9619 (1998). S. G. Waxman et al., Sodium channel genes in pain-related disorders: Phenotype-genotype associations and recommendations for clinical use. Lancet Neurol. 13, 1152–1160 (2014). S. D. Dib-Hajj, S. G. Waxman, Translational pain research: Lessons from genetics and genomics. Sci. Transl. Med. 6, 249sr244 (2014). 7. 8. sodium channel Isoform NaV1.7. J. Med. Chem. 62, 8695–8710 (2019). 12. W. A. Catterall, From ionic currents to molecular mechanisms: The structure and function of voltage- gated sodium channels. Neuron 26, 13–25 (2000). 13. C. A. Ahern, J. Payandeh, F. Bosmans, B. Chanda, The hitchhiker’s guide to the voltage-gated sodium channel galaxy. J. Gen. Physiol. 147, 1–24 (2016). 14. J. Payandeh, T. Scheuer, N. Zheng, W. A. Catterall, The crystal structure of a voltage-gated sodium channel. Nature 475, 353–358 (2011). 15. W. A. Catterall, G. Wisedchaisri, N. Zheng, The chemical basis for electrical signaling. Nat. Chem. Biol. 13, 455–463 (2017). 16. V. Yarov-Yarovoy et al., Structural basis for gating charge movement in the voltage sensor of a sodium channel. Proc. Natl. Acad. Sci. U.S.A. 109, E93–E102 (2012). 17. G. Wisedchaisri et al., Resting-state structure and gating mechanism of a voltage-gated sodium channel. Cell 178, 993–1003.e1012 (2019). 8 of 9 https://doi.org/10.1073/pnas.2219624120 pnas.org Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.18. Q. Li et al., Structural mechanism of voltage-dependent gating in an isolated voltage-sensing 38. D. Jiang et al., Structural basis for voltage-sensor trapping of the cardiac sodium channel by a domain. Nat. Struct. Mol. Biol. 21, 244–252 (2014). deathstalker scorpion toxin. Nat. Commun. 12, 128 (2021). 19. M. S. Dickinson, A. Myasnikov, J. Eriksen, N. Poweleit, R. M. Stroud, Resting state structure of the 39. Y. Y. Vilin, N. Makita, A. L. George, P. C. Ruben, Structural determinants of slow inactivation in human hyperdepolarization activated two-pore channel 3. Proc. Natl. Acad. Sci. U.S.A. 117, 1988–1993 (2020). cardiac and skeletal muscle sodium channels. Biophys. J. 77, 1384–1393 (1999). 20. G. Dai, T. K. Aman, F. DiMaio, W. N. Zagotta, The HCN channel voltage sensor undergoes a large downward motion during hyperpolarization. Nat. Struct. Mol. Biol. 26, 686–694 (2019). 40. J. R. Balser et al., External pore residue mediates slow inactivation in μ1 rat skeletal muscle sodium channels. J. Physiol. 494, 431–442 (1996). 21. Q. Li et al., Resting state of the human proton channel dimer in a lipid bilayer. Proc. Natl. Acad. Sci. 41. B. H. Ong, G. F. Tomaselli, J. R. Balser, A structural rearrangement in the sodium channel pore linked U.S.A. 112, E5926–5935 (2015). to slow inactivation and use dependence. J. Gen. Physiol. 116, 653–662 (2000). 22. V. S. Mandala, R. MacKinnon, Voltage-sensor movements in the Eag Kv channel under an applied 42. E. Pavlov et al., The pore, not cytoplasmic domains, underlies inactivation in a prokaryotic sodium electric field. Proc. Natl. Acad. Sci. U.S.A. 119, e2214151119 (2022). channel. Biophys. J. 89, 232–242 (2005). 23. G. Wisedchaisri et al., Structural basis for high-affinity trapping of the NaV1.7 channel in its resting 43. Y. Y. Vilin, P. C. Ruben, Slow inactivation in voltage-gated sodium channels: Molecular substrates and state by a tarantula toxin. Mol. Cell 81, 38–48.e34 (2021). 24. H. Xu et al., Structural basis of NaV1.7 inhibition by a gating-modifier spider toxin. Cell 176, 702–715.e14 (2019). 25. G. Huang et al., Unwinding and spiral sliding of S4 and domain rotation of VSD during the electromechanical coupling in Nav1.7. Proc. Natl. Acad. Sci. U.S.A. 119, e2209164119 (2022). 26. H. Shen, D. Liu, K. Wu, J. Lei, N. Yan, Structures of human NaV1.7 channel in complex with auxiliary subunits and animal toxins. Science 363, 1303–1308 (2019). 27. G. Huang et al., High-resolution structures of human NaV1.7 reveal gating modulation through α-π helical transition of S6IV. Cell Rep. 39, 110735 (2022). 28. H. Shen, N. Yan, X. Pan, Structural determination of human NaV1.4 and NaV1.7 using single particle cryo-electron microscopy. Methods Enzymol. 653, 103–120 (2021). contributions to channelopathies. Cell Biochem. Biophys. 35, 171–190 (2001). 44. J. Payandeh, T. M. Gamal El-Din, T. Scheuer, N. Zheng, W. A. Catterall, Crystal structure of a voltage-gated sodium channel in two potentially inactivated states. Nature 486, 135–139 (2012). 45. J. Li, J. Ostmeyer, L. G. Cuello, E. Perozo, B. Roux, Rapid constriction of the selectivity filter underlies C-type inactivation in the KcsA potassium channel. J. Gen. Physiol. 150, 1408–1420 (2018). 46. A. Lampert, M. Eberhardt, S. G. Waxman, Altered sodium channel gating as molecular basis for pain: Contribution of activation, inactivation, and resurgent currents. Handb. Exp. Pharmacol. 221, 91–110 (2014). 47. Z. Otwinowski, W. Minor, Processing of X-ray diffraction data collected in oscillation mode. Methods 29. S. Ahuja et al., Structural basis of NaV1.7 inhibition by an isoform-selective small-molecule Enzymol. 276, 307–326 (1997). antagonist. Science 350, aac5464 (2015). 30. T. M. Gamal El-Din, M. J. Lenaeus, K. Ramanadane, N. Zheng, W. A. Catterall, Molecular dissection of multiphase inactivation of the bacterial sodium channel NaVAb. J. Gen. Physiol. 151, 174–185 (2019). 31. H. S. Ahn et al., A new NaV1.7 sodium channel mutation I234T in a child with severe pain. Eur. J. Pain 48. A. J. McCoy et al., Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007). 49. G. N. Murshudov et al., REFMAC5 for the refinement of macromolecular crystal structures. Acta Crystallogr. D Biol. Crystallogr. 67, 355–367 (2011). 50. M. D. Winn et al., Overview of the CCP4 suite and current developments. Acta Crystallogr. D Biol. 14, 944–950 (2010). Crystallogr. 67, 235–242 (2011). 32. T. R. Cummins, S. D. Dib-Hajj, S. G. Waxman, Electrophysiological properties of mutant NaV1.7 sodium channels in a painful inherited neuropathy. J. Neurosci. 24, 8232–8236 (2004). 33. M. T. Wu, P. Y. Huang, C. T. Yen, C. C. Chen, M. J. Lee, A novel SCN9A mutation responsible for primary erythromelalgia and is resistant to the treatment of sodium channel blockers. PLoS One 8, e55212 (2013). 51. P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66, 486–501 (2010). 52. D. Liebschner et al., Macromolecular structure determination using x-rays, neutrons, and electrons: Recent developments in Phenix. Acta Crystallogr. D Struct. Biol. 75, 861–877 (2019). 53. N. Eswar et al., Comparative protein modeling using modeller. Curr. Protoc. Bioinformatics 54, 34. C. M. Kerth, P. Hautvast, J. Körner, A. Lampert, J. E. Meents, Phosphorylation of a chronic pain 5.6.1–5.6.37 (2016). mutation in the voltage-gated sodium channel NaV1.7 increases voltage sensitivity. J. Biol. Chem. 296, 100227 (2021). 35. A. Lampert, S. D. Dib-Hajj, L. Tyrrell, S. G. Waxman, Size matters: Erythromelalgia mutation S241T in NaV1.7 alters channel gating. J. Biol. Chem. 281, 36029–36035 (2006). 36. J. F. Fohlmeister, Voltage gating by molecular subunits of Na+ and K+ ion channels: Higher- dimensional cubic kinetics, rate constants, and temperature. J. Neurophysiol. 113, 3759–3777 (2015). 37. S. Cestèle et al., Voltage sensor-trapping: Enhanced activation of sodium channels by beta-scorpion toxin bound to the S3–S4 loop in domain II. Neuron 21, 919–931 (1998). 54. G. Wisedchaisri, T. M. Gamal El-Din, N. Zheng, W. A. Catterall, Crystal structure of NaVAb I119T as a basis for the human NaV1.7 Inherited Erythromelalgia I234T mutation. Protein Data Bank. https:// www.rcsb.org/structure/8DIZ. Deposited 29 June 2022. 55. G. Wisedchaisri, T. M. Gamal El-Din, N. Zheng, W. A. Catterall, Crystal structure of NaVAb L123T as a basis for the human NaV1.7 Inherited Erythromelalgia I848T mutation. Protein Data Bank. https:// www.rcsb.org/structure/8DJ0. Deposited 29 June 2022. 56. G. Wisedchaisri, T. M. Gamal El-Din, N. Zheng, W. A. Catterall, Crystal structure of NaVAb V126T as a basis for the human NaV1.7 Inherited Erythromelalgia S241T mutation. Protein Data Bank. https:// www.rcsb.org/structure/8DJ1. Deposited 29 June 2022. PNAS 2023 Vol. 120 No. 14 e2219624120 https://doi.org/10.1073/pnas.2219624120 9 of 9 Downloaded from https://www.pnas.org by US DEPARTMENT OF ENERGY OFFICE OF SCI & TECH INFO on August 26, 2024 from IP address 192.107.175.1.
10.1073_pnas.2302191120	RESEARCH ARTICLE \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS Circular RNA vaccine induces potent T cell responses Laura Amayaa,b,1 , Zhijian Lid, Audrey Leec, Paul A. Wenderd,e , Bali Pulendranc,2, and Howard Y. Changa,f,2 , Lilit Grigoryanc,1 Contributed by Howard Y. Chang; received February 9, 2023; accepted April 14, 2023; reviewed by Daniel H. Kim and Antoni Ribas Circular RNAs (circRNAs) are a class of RNAs commonly found across eukaryotes and viruses, characterized by their resistance to exonuclease-mediated degradation. Their superior stability compared to linear RNAs, combined with previous work showing that engineered circRNAs serve as efficient protein translation templates, make circRNA a promising candidate for RNA medicine. Here, we systematically examine the adjuvant activity, route of administration, and antigen-specific immunity of circRNA vaccina- tion in mice. Potent circRNA adjuvant activity is associated with RNA uptake and activation of myeloid cells in the draining lymph nodes and transient cytokine release. Immunization of mice with engineered circRNA encoding a protein antigen delivered by a charge-altering releasable transporter induced innate activation of dendritic cells, robust antigen-specific CD8 T cell responses in lymph nodes and tissues, and strong antitumor efficacy as a therapeutic cancer vaccine. These results highlight the potential utility of circRNA vaccines for stimulating potent innate and T cell responses in tissues. circular RNA \| CD8 T cells \| vaccine Circular RNAs (circRNAs) are covalently closed single-stranded RNA molecules derived from back-splicing processes (1). Most eukaryotic circRNAs are noncoding RNAs with potential regulatory functions in gene expression (2). Among other biochemical functions, circRNAs interact with other noncoding RNAs and serve as microRNA sponges (3), protein scaffolds (4), canonical splicing competitors (5), and protein nuclear translocation triggers (6). However, a small portion of endogenous circRNAs possesses internal ribosome entry sites (IRESs) and can act as protein templates, inducing translation in a cap-independent manner (7–9). Large amounts of circRNAs can also be synthesized in vitro through an in vitro transcription reaction (IVT) that includes a DNA template and a phage RNA polymerase (10). Several efforts are being made to improve the circu- larization and translation efficiency of circRNAs (11, 12). Currently, up to 10 kb tran- scripts in length can circularize and express open-reading frames (13). Recent studies have also suggested a potential role of circRNAs as modulators of the immune system. Some endogenous circRNAs can inhibit protein kinase R and restrain innate immunity (14). Changes in circRNA abundance have been associated with the occurrence of autoimmune disease (15) and proposed as potential regulators of tumor immunity (16). The recognition of in vitro-transcribed circRNAs by pattern recognition receptors when delivered into mammalian cells has also been described (17). Differences in the structure and covalent modifications of cellular and pathogenic circRNA distinguish self from nonself (17–19). Chen et al. showed that RIG-I directly senses exogenous cir- cRNA and initiates an innate immune signaling cascade dependent on the absence of N6-methyladenosine (m6A) as a mark of self-identity (18). Due to their covalently closed loop structures and the absence of 5′ caps or 3′ poly A tails, circRNAs are more stable compared to linear RNA. CircRNAs are highly stable in blood (20) and may be released from cells via extracellular vesicles (21). These character- istics have drawn increasing attention to circRNAs for applications in clinical practice as diagnostic and prognostic biomarkers (22, 23), and more recently, as candidates for vac- cines. A circRNA-based vaccine was shown to induce broad-spectrum protection against SARS-CoV-2 in nonhuman primates (24). However, the mechanisms behind the in vivo recognition of circRNA, its translation, and activating signals leading to protective immu- nity remain poorly understood. Cytotoxic (CD8) T cells represent a major target of vaccination, as CD8 T cell–mediated protection has been shown to be important in both contexts of viral infections and tumor immunity. Heterologous viral vector regimens can uniquely elicit tissue-resident T cells at sites of infection compared to recombinant protein or mRNA vaccines (25, 26), but the former is difficult to scale and implement in humans. Thus, there is an unmet need to identify additional vaccine platforms to elicit T cell immunity. One challenge is that simultaneous antigen delivery and adjuvant to antigen-presenting cells are required to induce strong T cell immunity in vaccination settings (27). Here, we propose simplifying the formulation of RNA vaccines by using circRNA as both immunogen and adjuvant. Significance Circular RNAs (circRNAs) are a unique class of RNAs that are highly stable compared to linear mRNAs and can be engineered to provide durable protein expression. This study demonstrates that circRNA encoding antigenic protein sequences delivered by a charge-altering releasable transporter can effectively serve as both an adjuvant and an immunogen, inducing potent cellular immunity and leading to tumor clearance when used as a therapeutic vaccine. These results suggest engineered circRNAs for the development of vaccines and therapeutics. aCenter Author affiliations: for Personal Dynamic Regulomes, Stanford University, Stanford, CA 94305; bInstitute for Stem Cell Biology and Regenerative Medicine, Stanford University School of Medicine, Stanford, CA 94305; cInstitute for Immunity, Transplantation and Infection, Stanford University, Stanford, CA 94305; dDepartment of Chemistry, Stanford University, Stanford, CA 94305; eDepartment of Chemical and Systems Biology, Stanford University, Stanford, CA 94305; and fHHMI, Stanford University, Stanford, CA 94305 Author contributions: L.A., B.P., and H.Y.C. designed research; L.A., L.G., Z.L., and A.L. performed research; Z.L. and P.A.W. contributed new reagents/analytic tools; L.A., L.G., and B.P. analyzed data; and L.A. and H.Y.C. wrote the paper. Reviewers: D.H.K., University of California, Santa Cruz; and A.R., Ronald Reagan UCLA Medical Center. Competing interest statement: H.Y.C. is a co-founder of Accent Therapeutics, Boundless Bio, Cartography Biosciences, Orbital Therapeutics and an advisor of 10x Genomics, Arsenal Biosciences, Chroma Medicine, and Spring Discovery. P.A.W. is a co-founder of BryoLogyx and N1 Life and an advisor to BryoLogyx, N1 Life, Synaptogenix, Cytokinetics, Evonik, SuperTrans Medical, Ativo, and Vault Pharma. B.P. serves on the External Immunology Network of GSK and on the scientific advisory board of Medicago, Sanofi, EdJen, and Boehringer-Ingelheim. Stanford University has filed a patent on circular RNA technology on which L.A., B.P., and H.Y.C. are named as inventors. The remaining authors declare no competing interests. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1L.A. and L.G. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2302191120/-/DCSupplemental. Published May 8, 2023. PNAS 2023 Vol. 120 No. 20 e2302191120 https://doi.org/10.1073/pnas.2302191120 1 of 10 We previously found that in vitro-transcribed circRNAs, which are efficiently circularized via the self-splicing Td intron from T4 bacteriophage, are highly immunogenic in vivo (17, 18). Using circRNA as an adjuvant elicited potent B and T cell responses even when delivered without an RNA transfection reagent (18). However, it remains unclear whether exogenous circular RNA encoding antigen could induce antigen-specific immune responses. In this study, we explore the principles of how exogenous circR- NAs interact with the mammalian immune system. We investigate the innate immune responses induced by naked circRNA and cir- cRNA encapsulated in charge-altering releasable transporters (CARTs). Furthermore, we elaborate on the adjuvant properties of circRNA by examining T cell responses induced by different immu- nization routes and comparing it to commonly used vaccine adju- vants. Finally, we show that immunization with CART-encapsulated circRNA encoding antigen results in successful antigen presentation, induction of potent cellular immunity, and tumor clearance when used as a therapeutic cancer vaccine. Results CircRNA Acts as a Potent Vaccine Adjuvant in Multiple Immunization Routes When Combined with Soluble Protein. Adjuvants are essential to improve the effectiveness of vaccines by increasing the strength and duration of the immune response through different modes of action. We previously showed that exogenous circRNA help induce antigen-specific antibodies and T cells when delivered in vivo with soluble protein (18). Here, we further evaluate the potential of circRNA as a unique adjuvant in vaccine formulations. We systematically measured the acute and memory immune responses after vaccination with circRNA compared to commonly used adjuvants. The route of immunization can shape the immune response by determining which antigen-presenting cells (APCs) are activated and where the immune response is focused. Different routes may have different advantages and disadvantages depending on the type of vaccine. Therefore, we immunized C57BL/6 mice with in vitro-synthesized circRNA and chick Ovalbumin protein (referred as OVAp) and compared the magnitude of T cell and antibody responses with three delivery strategies: subcutaneous (s.c), intranasal (i.n.), and intravenous (i.v.). We measured the T cell and antibody responses in the spleen, draining lymph nodes (LNs), and lungs at day 7 and day 30 postboost (Fig. 1A). The immune responses observed with immunogenic circRNA [lacking m6A modification (18)] were compared to the results induced by com- mon vaccine adjuvants: AddaVax, a squalene-based oil-in-water nanoemulsion and effective subcutaneous adjuvant (28); and Poly(I:C), a synthetic dsRNA and effective intranasal adjuvant (29). At day 7 postboost, we observed that subcutaneous injection of circRNA+OVAp induced comparable T cell responses to AddaVax+OVAp in the lungs, spleens, and LNs (as measured by the frequency of MHC class I tetramer+ CD8 T cells) (Fig. 1B). Intranasal inoculation of circRNA+OVAp induced the highest T cell responses in the lungs at day 7 postboost, with the frequen- cies of tetramer-positive cells as high as 40% (Fig. 1B). Subcutaneous delivery of circRNA+OVAp induced a twofold higher frequency of antigen-specific CD8 T cells (~10%) com- pared to AddaVax+OVAp (~5%) at day 30 postboost (Fig. 1C), suggesting a potentially enhanced memory T cell induction by circRNA compared to AddaVax. In addition, strong lung memory CD8 T cell responses were observed with the intranasal and intra- venous delivery methods (Fig. 1C), with the induction of lung-resident memory CD8 T cell (CD69+CD103+) subsets (SI Appendix, Fig. S1 A and B). To measure the antibody responses following immunization, mice were bled 30 days postboost. We observed similar levels of anti-OVA IgG antibodies in serum among all delivery routes (Fig. 1D). However, only intranasal and intravenous immunization induced anti-OVA IgA antibodies in serum (SI Appendix, Fig. S1C). In addition, to compare circRNA and Poly(I:C) as intranasal adjuvants, mice were immunized with either adjuvant in combi- nation with soluble OVA protein. CircRNA and Poly(I:C) induced comparable frequencies of antigen-specific CD8 T cells in the lungs at 30 d postboost (Fig. 1E), including CD69+ and CD69+CD103+ antigen-specific tissue-resident memory T cells (TRM) (Fig. 1F). In addition, circRNA and Poly(I:C) induced similar levels of anti-OVA IgG and IgA antibodies (SI Appendix, Fig. S1 D and E). Taken together, our results suggest that circRNA can be used as a potent vaccine adjuvant in many routes of immu- nization and induce comparable responses to Poly(I:C) and AddaVax. In addition, we showed that mucosal immunization with circRNA as an adjuvant can induce potent resident memory CD8 T cell (TRM) responses. CircRNA Activates Innate Immune Cells When Injected into Mice. The innate immune system plays a fundamental role in programming the adaptive immune response's magnitude, quality, and durability (30). To understand the mechanisms behind the strong adaptive immune responses observed after vaccination with naked circRNA, we surveyed the innate immune compartment for marks of activation and circRNA recognition. We started by determining the biodistribution of circRNAs when delivered in vivo. We conjugated circRNA to the fluorophore AF488 and subcutaneously (s.c.) injected 25 μg of AF488-circRNA into C57BL/6 mice. Serum was analyzed by a Luminex panel of innate cytokines before and after circRNA immunization. In addition, innate immune cell subsets in the draining inguinal lymph nodes (iLNs) were analyzed via flow cytometry at 24 h following immunization (Fig. 2A). Innate cell activation was measured by the upregulation of the activation marker CD86 on each cell subset. Monocytes were defined as CD11b+Ly6C+ cells and dendritic cells as CD11c high MHC-II high cells, with DC subsets further subdivided into migratory CD103+ or CD11b+ DCs (mDC) and resident CD8a+ or CD11b+ DCs (rDC). Lymph node (LN) macrophages were identified as CD11b+Ly6CloF4/80+/- CD169+/- and plasmacytoid dendritic cells (pDCs) as CD11b- PDCA-1+ cells. Lastly, neutrophils were defined as CD11b+Ly6G+ and eosinophils as CD11b+Signlec-F+ (SI Appendix, Fig. S2A). At 24 h following s.c. injection, circRNA was detected in mono- cytes, dendritic cells, and several macrophage subsets in the drain- ing lymph nodes. The macrophage subsets taking up circRNAs included marginal cord macrophage (MCMs), marginal sinus macrophage (MSMs), and subcapsular sinus macrophages, (SSM), with MCMs and MSMs having the most significant uptake of circRNA (Fig. 2B). Even though no circRNA uptake was observed by B cells in mice, B cell frequencies in the iLNs significantly increased at 24 h post immunization (Fig. 2C), and B cell activa- tion, as measured by CD86 upregulation, increased as well (Fig. 2D). We observed a significant increase in the frequency (Fig. 2C) and activation (Fig. 2D) of monocytes in the iLNs, as well as increased activation of all macrophage and dendritic cell subsets compared to untreated controls (Fig. 2D). To examine the serum cytokine response to immunization with circRNA, sera from immunized mice were analyzed at 6- and 24-h postimmunization (SI Appendix, Fig. S2B). Significant production of chemokines: CCL5, CCL4, CCL3, CCL7, CXCL10, CCL2; and cytokines: IL-6, TNFa, IL-12; was observed, with a peak at 6 h after immunization, followed by a decrease at 24 h (Fig. 2E). 2 of 10 https://doi.org/10.1073/pnas.2302191120 pnas.org A Ovalbumin + Adjuvant Adjuvant: Delivery routes: Tissue collection: circRNA AddaVax Poly(IC) Intranasal ( i.n.) Subcutaneous (s.c.) Intravenous (i.v.) Spleen Lungs Lymph Nodes Serum Prime Day 0 Boost Day 21 Day 7 (Acute respone) Day 30 (Memory response) B Day 7 60 s l l **** ns ns C Day 30 40 s l l * * * Lung Spleen ** ns ns * ns ns e c T 8 D C + r e m a r t e T % 30 20 10 0 12 s 10 l l e c T 8 D C + r e m a r t e T % 8 6 4 2 0 30 s l l e c T 8 D C + r e m a r t e T 25 20 15 10 e c T 8 D C + r e m a r t e T % s l l e c T 8 D C + r e m a r t e T % s l l 40 20 0 6 5 4 3 2 1 0 6 5 4 3 2 1 0 Lung Spleen e c T 8 D C + r e m a r t e T % Draining Lymph nodes D E F * ns ns *** * ns ns ns ns ** ns ns ns 6 4 2 r e t i t 0 5 C I G g I a v O - i t n A 0 AddaVax s.c. Naive circRNA s.c. circRNA i.n. circRNA i.v. 3 2 1 0 s l l e c T 8 D C + r e m a r t e T % 80 60 40 20 + r e m a r t e T f o % Naive Poly(IC) i.n. circRNA i.n. Poly(IC) i.n. circRNA i.n. 0 CD69+CD103+ CD69+ Draining Lymph nodes % 5 0 AddaVax s.c. Naive circRNA s.c. circRNA i.n. circRNA i.v. AddaVax s.c. Naive circRNA s.c. circRNA i.n. Fig. 1. Adjuvant effect of circRNA by different routes of delivery. (A) Schematic representation of circRNA immunization strategy via different delivery routes and monitoring of immune responses. OVA protein (50 μg) and circRNA (25 μg), AddaVax (50 μL), or Poly(IC) (25 μg) was delivered by intranasal (i.n.), subcutaneous (s.c.), or intravenous (i.v.) injection. Serum, lung, lymph nodes, and spleen were analyzed at days 7 and 30 postboost. Percentage of OVA-specific T cell responses in lung, spleen, and lymph nodes after (B) 7 d or (C) 30 d postboost (n = 5, bars represent Min and Max). (D) Anti-OVA IgG antibodies in serum measured by ELISA at day 30 postboost after circRNA immunization by different delivery routes (n = 5, bars represent Min and Max). (E) Frequency of class I tetramer+ CD8 T cells at day 30 postboost of i.n. delivery of circRNA compared to Poly(IC) (n = 5, bars represent Min and Max). (F) Frequency of CD69+ and CD69+CD103+ CD8 TRM in the lungs at day 30 postboost (as percentage of antigen-specific CD8 T cells) (n = 5, bars represent Min and Max). One-way ANOVA was applied in B–E, and two-way ANOVA in F. P < 0.05, P < 0.01, P < 0.001, *P < 0.0001. Differences between groups were considered significant for P values < 0.05. ns, not significant. PNAS 2023 Vol. 120 No. 20 e2302191120 https://doi.org/10.1073/pnas.2302191120 3 of 10 Fig. 2. Biodistribution of circRNA and innate recognition. (A) Schematic representation of in vivo circRNA delivery and monitoring. Fifty micrograms of AF488- circRNA was delivered subcutaneously, and serum samples were collected 6 and 24 h after delivery. Draining lymph nodes were also analyzed after 24 h by flow cytometry. (B) Absolute fraction of fluorescently positive innate cell subsets that take up circRNA (n = 5, bars represent Min and Max). (C) Quantification of innate cell subsets proportions and (D) fluorescent intensity of activation marker CD86 in distinct innate immune cell subsets in lymph nodes, 24 h after s.c. delivery of fluorescently labeled circRNA (n = 5, bars represent Min and Max). (E) Time course analysis of cytokines in serum after circRNA delivery measured by Luminex (n = 5). Two-way ANOVA was applied in B–D. One-way ANOVA was applied in E. P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. BAFF and CCL11 showed a continued increase with a peak at 24 h after circRNA immunization (Fig. 2E). Taken together, our data suggest that naked circRNA is taken up by innate immune cells when injected into mice and induces subsequent activation of several innate immune cell types while promoting strong induc- tion of inflammatory cytokines. 4 of 10 https://doi.org/10.1073/pnas.2302191120 pnas.org CircRNA Induces Immune Activation of Dendritic Cells. Dendritic cells (DCs) are the most effective antigen-presenting cells, and their maturation indicates the acquisition of several properties, including antigen processing and presentation, migration, and T cell costimulation (31). After observing that dendritic cells are one of the primary innate cell subsets responsible for the recognition of circRNA when delivered in vivo, we next wanted to investigate whether the delivery of naked circRNA would influence the maturation and activation status of dendritic cells. MutuDCs, a murine DC cell line originated from splenic CD8α conventional DC tumors (32), were treated with circRNA or CpG oligodeoxynucleotides, short synthetic single-stranded DNA, known to induce dendritic cell maturation (33). CircRNA induced strong upregulation of costimulatory molecules such as MHC-II, MHC-I, CD80, CD86, and CD40 on MutuDCs, with higher levels of MHC-II and CD40 compared to CpG treatment. (SI Appendix, Fig. S3A). As innate immune signaling often promotes inflammatory cytokine gene expression to induce DC activation, we next examined cytokine gene expression in MutuDCs after incubation with circRNA. CirRNA treatment significantly induced the expression of proinflammatory cytokine genes IL-1β, TNFa, and IL-6, cytokines required for dendritic cell differentiation and maturation (34) (SI Appendix, Fig. S3B). CircRNA uptake in MutuDC cells also lead to a significant increase in the mRNA levels of the cytosolic RNA sensors RIG-I and MDA5. This observation in murine dendritic cells indicates that the recognition of circRNA by RIG-I is independent of the cell type, as this effect was previously described in HeLa cells after the transfection of foreign circRNA (17). Flow cytometry results confirmed the upregulation at the protein level of TNFa and IL-6 cytokines in culture supernatant from cells treated with circRNA. Increased levels of MCP1 were also observed. This monocyte chemoattractant protein-1 has the ability to drive the chemotaxis of myeloid and lymphoid cells (35) (SI Appendix, Fig. S3C). These results indicate that circRNA uptake leads to dendritic cell maturation, which increases the expression of costimulatory molecules and secretion of a wide variety of inflammatory cytokines and chemokines. Engineered circRNA-Encoding Protein Leads to Antigen Pres- entation. The observation that circRNA activates innate immunity prompted the investigation of the capacity of circRNA to subsequent- ly induce adaptive immune responses when used as an adjuvant and as an antigenic encoding sequence. We designed a circRNA-encoding chick Ovalbumin (hereafter named circOVA). To maximize circRNA translation, we used previously optimized elements for our circRNA design and transcription (12) (Fig. 3A). These elements include opti- mized RNA chemical modifications, 5′ and 3′ untranslated regions, a v O d e i f i r u P ) g n 0 1 ( d e t c e f s n a r t T 3 9 2 ) A V O c r i c g u 1 ( C CACC RTRR -TT circOVOO AVV Naive CD8 T cells isolation CFSE staining OTOO -TT 1 mouse 24 hours 3 days co-culture Tubulin 50kDa Ova 40kDa MutuDC Dendritic cell T cell (naïvïï e) MHC I TCR CD28 CD80/ 86 Flow cytyy ometry Proliferating cells t n u o C 1000 800 600 400 200 0 0 103 10400 105 106 CFSE Antigen processing Antigen presentation T-cell activation and proliferation Naive CART CART-circOVA circOVA B 160 125 90 70 50 38 30 25 15 E ) I F M ( 6 8 D C A Apt-eIF4G Ovalbumin HRV-B3 IRES circOVA 5′ PABP spacer Splicing scar HBA1 3′ UTR D 80 n o i t a r e f i l o r p 60 40 CART-circOVA SIINFEKL l l e c T % 20 0 0.1 1 10 100 1uM circOVA (ng) 20000 15000 10000 5000 0 -5000 l l e c B s l i h p o n s o E i s e t y c o n o M s l i h p o r t u e N M C M M S M M S S C D m C D p C D r Fig. 3. Circular RNA is translated and presented to the immune system. (A) Schematic representation of circOVA design and engineered components. Arrow indicates expected open-reading frame. (B) Detection of full-length Ovalbumin protein by western blot 24 h after transfection of 293T cells with circOVA. (C) Experimental design of proliferation assay used to measure antigen-specific T cell proliferation level of OT-I cells cocultured with MutuDC cells incubated with CART-circOVA. (D) CART-circOVA titration to determine the minimum amount required to induce antigen-specific T cell proliferation and peptide SINFEKL as positive control (n = 4, bars represent Min and Max). (E) CD86 expression of innate cell subsets 24 h after s.c. delivery of circRNA, circRNA delivered with CART (CART-circOVA), and CART alone (n = 5, bars represent Min and Max). Two-way ANOVA was applied P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Only differences between groups considered significant (P values < 0.05) are displayed. PNAS 2023 Vol. 120 No. 20 e2302191120 https://doi.org/10.1073/pnas.2302191120 5 of 10 internal ribosome entry sites (IRESs), and synthetic aptamers shown to increase circRNA translation over mRNA after a single transfection. We first validated the protein production activity of circOVA in 293T cells. Twenty-four hours after circOVA transfection, Ovalbumin protein was detected in cell lysate by immunoblot (Fig. 3B) and in supernatant by ELISA (210 pg OVA/μL of supernatant). intranasally We proceeded to validate the activation of adaptive immune responses after in vivo delivery of circRNA-encoding protein. The immunogenicity of circOVA alone was compared to soluble OVA protein combined with circRNA as an adjuvant (circRNA+ OVAp). Mice were immunized with either m6A-modified or -unmodified circOVA, and 30 d postboost, the lungs were analyzed for antigen-specific T cell responses. N6-methyladenosine (m6A) modification has been shown to pro- mote the translation of circRNAs (9); however, we have previously shown that m6A abrogates circRNA immunity (18). Indeed, naked delivery of m6A-modified circOVA did not induce any OVA-specific T cell responses (SI Appendix, Fig. S4A). A subset of animals in the unmodified circOVA group generated potent OVA-specific CD8 T cell responses, but many other animals did not (SI Appendix, Fig. S4A). We suspect that after in vivo delivery of naked circOVA, the amount of circRNA that enters antigen-presenting cells is sufficient to induce immune signaling activation; however, only a small proportion of circRNA might be readily accessible for translation. CircRNA Delivery with CART Improves circRNA Translation. We hypothesized that a delivery vehicle might be required to induce optimal immune responses in vivo with circRNA as both immunogen and adjuvant. We tested circRNA delivery using CARTs, a class of synthetic biodegradable materials shown to complex, protect, and efficiently deliver mRNA (36, 37) and circRNA (12) intracellularly, leading to highly efficient protein translation. An essential process for initiating cytotoxic immune responses is antigen presentation (38). To determine whether circRNA deliv- ered with CART (referred to as CART-circRNA) can be translated and processed for antigen presentation by dendritic cells, we tested the antigen presentation capacity of MutuDCs after transfection the capability of with CART-circOVA. We measured antigen-primed dendritic cells to induce antigen-specific T cell proliferation in vitro (Fig. 3C). Dose titration showed that only 0.1 ng circRNA is required to induce antigen-specific T cell pro- liferation when circRNA is complexed with CART (Fig. 3D). These observations indicate that the protein encoded by circRNA can be processed and presented to the immune system. We also asked whether the recognition of circRNA and conse- quent activation of the innate immune system differs between naked and CART-mediated circRNA delivery. Mice were s.c. immunized with CART alone, naked circRNA, or CART-circRNA. Innate cell frequencies, activation, and circOVA uptake were meas- ured in the inguinal LNs at 24 h following immunization. cir- cRNA delivered with CART resulted in a similar activation profile of innate immune subsets as previously observed with naked cir- cRNA (Fig. 3E). Immunization with CART alone induced some innate immunity, such as increased monocyte frequencies in the iLNs (SI Appendix, Fig. S5A), as well as increased CD86 expres- sion on mDCs (Fig. 3E). However, delivery of circRNA with CART did not significantly alter circRNA recognition by immune cells (SI Appendix, Fig. S5B) or the innate cell infiltration and activation in the iLNs (SI Appendix, Fig. S5A). Thus, CART does not impact the recognition of circRNA by the innate immune system and subsequent activation of dendritic cells and macrophages. CircRNA-Encoding Antigen Induces Strong T Cell Responses In Vivo. CARTs have been shown to work effectively in mice, have a high encapsulation efficiency, and are well tolerated and nonimmunogenic (36, 37, 39). We also previously showed robust and sustained protein production from circRNA delivered with CART after intraperitoneal injection (12). Thus, we selected to measure adaptive immune responses in mice after intraperitoneal delivery of circRNA-encoding protein. Three groups of mice were intraperitoneally immunized with either CART alone (vehicle-only control), CART-circOVA, and circRNA+OVAp at days 0 and 21. Antigen-specific CD8 T cell responses were assessed by MHC class I tetramer staining of the lung, spleen, and blood T cells at day 7 (7 d postprime) and day 42 (21 d postboost) (Fig. 4A). We observed that CART-circOVA induced potent CD8 T cell responses in the lung, spleen (Fig. 4 B and C), and blood (SI Appendix, Fig. S6B) at 7 d following a single immunization. Three weeks after the booster immunization (day 42), significant levels of CD8 T cell responses are observed in the CART-circOVA group in the spleen and lung (Fig. 4 B and C), yielding approxi- mately sixfold to eightfold greater frequency of OVA-specific CD8 T cells compared to circRNA+OVAp. Moreover, CART-circOVA immunization regimen induced both KLRG1+CD127- effector cells (short-lived effector cells or SLECs, ~20%) and KLRG1-CD127+ memory cells at day 42, with the latter T cell population being comprised heavily of effector memory T cells (TEM, ~35%) and a lower fraction of central memory (TCM) and resident memory (TRM) cells (SI Appendix, Fig. S6 A and C). In addition, to measure the antibody responses following immunization, mice were bled at days 0, 7, 21 (preboost), and 42 (21 d postboost). We observed that while circRNA+OVAp induces consistent and significantly higher anti-OVA IgG compared to CART-circOVA, anti-OVA antibodies were still detectable in the CART-circOVA group (Fig. 4D). The lack of consistent high-titer antibody responses may result from a reduced protein secretion after CART-circOVA immunization. Consistent with this notion, we were not able to detect OVA protein in blood 24 h after immu- nization. These results indicate that our immunization strategy with CART-circRNA leads to a T cell biased response and further optimization may be required to induce strong antibody responses to circOVA. We next sought to determine whether the combination of circRNA and CART could be generalized to additional vaccine candidates. Given that influenza virus is one of the most rapidly mutating viruses, strategies aimed at targeting the virus’s con- served regions have become crucial in the last few decades. In particular, CD8 T cells targeting conserved sites of the virus (such as nucleoprotein) have been shown to be protective against heterosubtypic influenza infection (40, 41), and T cell responses in humans have also been shown to correlate with protection (42). Therefore, we constructed a circRNA encoding the nucleo- protein (NP) sequence of the influenza virus (PR8 strain), referred to as circNP. Previous studies have shown that nucleoprotein-specific T cells are important in both homotypic and heterotypic protection against influenza virus infection (43–45), making NP-specific T cells a great target for a universal influenza vaccine. Intraperitoneal immunization with circNP + CART (CART-circNP) at days 0 and 21 resulted in the induction of nucleoprotein-specific T cell responses in the blood at day 7 postboost, as measured by class I tetramer staining of CD8 T cells (ASNENMETM epitope) (SI Appendix, Fig. S7 A and B). Our to encode pathogen-derived antigenic sequences can effectively induce antigen-specific CD8 T cell responses after immunization. that using circRNA indicate results 6 of 10 https://doi.org/10.1073/pnas.2302191120 pnas.org A CART-circOVA (i.p.) B Lung Spleen Prime Boost 0.03 % 0.52 % CART only Day 0 Day 7 Day 14 Day 42 Day 1 Day 21 Tissue collection: Serum: Day, 0, 7, 14, 21, and 42 Lung, Spleen: Day 7 and 42 Day 7 ns Day 42 ns ns ns ns ns C Lung Spleen 30 20 10 0 20 15 10 5 s l l e c T 8 D C + 4 4 D C + r e m a r t e T % 0 circRNA + OVAp CART-circOVA CART only circRNA + OVAp CART-circOVA CART only E B16-OVA inoculation CART-circOVA (i.p.) Day 22 Day 0 Day 4 Day 8 Tumor volume Tumor volume monitoring monitoring Bioluminescence-Based Tumor Quantification B16-OVA (s.c.) 10.6 % 5.44 % CART-circOVA 1.69 % 0.75 % circRNA+OVAp OVA Tetramer CART only circOVA + CART circRNA + OVAp 4 4 D C D 20 15 ) C U A ( 10 r e t i t G g I a v O - i t n A 5 2.0 1.5 l a t o T 1.0 0.5 0.0 Day 0 F ) 3 m m ( e m u o v l r o m u T 350 300 250 200 150 100 50 0 0 Day 7 Day 14 Day 21 Day 42 Control CART-circOVA 5 10 Days after tumor cell injection 15 20 Fig. 4. Circular RNA delivery in vivo activates T cell–specific responses. (A) Schematic representation of immunization strategy and monitoring of adaptive immune responses. Nine micrograms of circOVA was complexed with CART reagent and delivered intraperitoneally at days 0 and 21. Serum samples were collected weekly, and the spleen and lung were analyzed 7 d post prime and 21 d postboost. (B) Percentage of OVA-specific T cell responses in the lung and spleen at day 42 (representative sample). (C) Quantification of OVA-specific T cells in the lung and spleen at day 7 and day 42 (n = 5, bars represent Min and Max). (D) Time course analysis of anti-OVA IgG antibodies in serum measured by ELISA (n = 5, bars represent Min and Max). (E) Schematic representation of immunization strategy and monitoring of tumor volume after inoculation with B16-F10-OVA cells. Ten micrograms of circOVA was complexed with CART reagent and delivered intraperitoneally at days 4 and 8 after tumor cell inoculation. (F) Tumor volume monitoring over 22 d (n = 5, bars represent SEM). Results are representative of three independent experiments. Two-way ANOVA was applied in C and D. Repeated-measures ANOVA was applied in F. P < 0.05, P < 0.01, P < 0.001, ****P < 0.0001. Differences between groups were considered significant for P values < 0.05. ns, not significant. PNAS 2023 Vol. 120 No. 20 e2302191120 https://doi.org/10.1073/pnas.2302191120 7 of 10 Our data suggest that synthetic circular RNAs can encode both the antigen and adjuvant activity required for immuniza- tion. Likewise, the route, dose, and manner of circRNA delivery impact the programmed immune response's potency, consist- ency, and memory. CircRNA Vaccine Induces Antitumor Efficacy. Cancer vaccination aims to induce antigen-specific T cell–based cellular immunity capable of targeting and clearing tumor cells (46). The strong cytotoxic T cell responses observed systemically in tissue and blood after immunization with circOVA complexed with CART prompted us to further investigate circRNA as a cancer vaccine. We hypothesize that vaccine-induced OVA-specific T cells should eradicate OVA-expressing tumors. Moreover, the antitumor response should be systemic and be elicited by vaccination at a site distant from the tumor (i.e., abscopal effect). We tested the antitumor efficacy of the CART-circRNA vaccine in a therapeutic regime. C57BL/6 mice were randomly assigned into two groups: a control group (untreated) and a CART-circOVA (vaccine) group. Syngeneic B16-F10-OVA melanoma cells were inoculated subcutaneously on the backs of all mice. CART-circOVA formulations were injected intraperitoneally 4 and 8 d after tumor cell inoculation (Fig. 4E). The circRNA vaccine group showed a significant tumor growth inhibition compared to the untreated group (Fig. 4F). Bioluminescence imaging confirmed eradication of luciferase-labeled cancer cells (SI Appendix, Fig. S8 A and B). These results indicate that circRNA immunization could serve as an effective cancer immunotherapy to inhibit tumor growth in vivo. Discussion The potential of circRNA as a vaccine platform and gene delivery system has gained substantial interest since the speedy develop- ment and FDA approval of the mRNA vaccine against SARS- CoV-2 (24). CircRNA is an attractive platform as biomarkers and vectors for gene expression due to its superior durability and stability. However, the extent to which extracellular circRNAs may engage the innate and adaptive immune systems is poorly understood. In this study, we investigated the molecular and functional effects of circRNA recognition by the innate immune system. We elaborated on the potential of circRNA as an effective adjuvant when inoculated via various immunization routes. Our results showed that circRNAs, combined with protein antigens, can induce potent adaptive responses regardless of the immunization route. Notably, the mucosal immunization (intranasal delivery) of circRNA and OVAp induced potent lung-resident memory T cell responses. We observed the induction of cytosolic RNA sen- sors RIG-I and MDA5; inflammatory cytokines IL1-B, TNFa, and IL-6; and activation markers MHC-I and CD40 after cir- cRNA uptake by dendritic cells. This phenotypic profile suggests that the uptake of circRNA by APCs is sufficient to induce a dendritic cell activation and maturation that enables them to interact with antigen-specific T cells. In vivo delivery of fluores- cently labeled circRNA allowed us to characterize circRNA rec- ognition by the innate immune system. We showed specific circRNA response by monocytes, macrophages, and dendritic cells in draining lymph nodes after subcutaneous delivery of naked circRNA. Significant proliferation of B cell, monocyte, and macrophage populations was observed 24 h following immuni- zation with circRNA. In addition, all macrophages and dendritic cell subsets found in draining lymph nodes were shown to be activated based on the surface expression of CD86. Similarly, we observed a fast increase in proinflammatory cytokines and chemokines CCL5, CCL4, and CCL7, which act as chemoat- tractants of macrophages, T cell subsets, and DCs, among other immune cell types. In combination, these results highlight the potential of circRNA to induce proliferation and enhanced recruitment of specialized immune cell subsets, leading to the initiation of effective adaptive immunity. We were able to combine the dual roles of exogenous circRNAs as an adjuvant and a template for antigen expression, showing that circRNAs can serve as delivery systems and as immune potentia- tors. Our in vitro experiments in the MutuDC cell line showed that circOVA could be translated, and peptide sequences were efficiently loaded onto the MHC class I proteins for presentation to T cells, resulting in the activation and proliferation of antigen-specific T cells. To examine the capacity of circRNA to induce adaptive immunity in vivo, we immunized mice with cir- cOVA complexed with CART. We observed potent and persistent T cell responses in mice following immunization. Interestingly, antibody responses were not as effectively induced by the current CART-circOVA formulation, likely due to limited expression of cell-free soluble OVA protein vs. the intracellular pool. However, recent work has shown that T cell–inducing vaccines can provide protective immunity against simian–HIV (SHIV) even when sub- optimal antibody responses are present (25, 26). Nonetheless, future studies of the underlying mechanisms responsible for the induction of T cell responses without efficient induction of anti- body responses are warranted. And, induction of more potent antibody responses will be important for infectious disease appli- cations of circRNA vaccination. Taken together, we demonstrated the potential of circRNAs to generate an acute inflammatory environment that favors the gen- eration of potent cellular immunity. Combining the features of facile programmability, durable antigen expression, and safe administration, the use of circRNAs as a tool to program antigen-specific T cell response has the potential to advance prophylactic or therapeutic vaccines for infectious diseases and cancer immunotherapy. Materials and Methods CircRNA Design and In Vitro Transcription. CircRNA templates were syn- thesized by cloning DNA fragments into a custom entry vector which contains self-splicing introns, 5′ PABP spacer, HBA1 3′ UTR, and HRV-B3 IRES. CircRNA was synthesized using HiScribe T7 High-Yield RNA Synthesis Kit (NEB E2040S). IVT templates were PCR amplified (Q5 Hot Start High-Fidelity 2x Master Mix) and column purified (Zymo DNA Clean & Concentrator-100) prior to RNA synthesis as previously described (12). Briefly, 1 µg circRNA PCR-template was used per 20 µL IVT reaction. Reactions were incubated overnight at 37°C. The IVT templates were subsequently degraded with 2 µL DnaseI (NWB M0303S) for 20 min at 37°C. The remaining RNA was column purified and digested with 1U RnaseR per microgram of RNA for 60 min at 37°C. Samples were then column purified, quantified using a Nanodrop One spectrophotometer, and verified for complete digestion using an Agilent TapeStation. CircRNA was fluorescently labeled by incorporating 5% of Fluorescein-12-UTP (Sigma-Aldrich 11427857910) in the corresponding IVT reaction, or by posttranscriptional modification using Label IT Nucleic Acid Labeling Kit (Mirus Bio Cy3, Cy5, Fluorescein, or AF488). In all experiments, we used a mixture of unlabeled circRNA and fluorescently labeled circRNA at 20:1 ratio. Three circRNAs were produced: circOVA which encodes OVA protein, circNP which encodes influenza nucleoprotein, and circFOR that has a frame-shift sequence that interferes with protein translation. CircOVA and circNP were enhanced for translation by adding 5% m6A modifications (when specified) and 5% of 2′OMeC for in vivo delivery. Circular RNA elements and modifications are listed in SI Appendix, Table S1. Cell Lines. MutuDC cells were purchased from Applied Biological Materials Inc. (ABM T0528). The cells were maintained in IMDM-Glutamax (Gibco 31980) medium supplemented with 10% FBS, 1% penicillin-streptomycin, 10 mM Hepes 8 of 10 https://doi.org/10.1073/pnas.2302191120 pnas.org (Gibco 15630), and 50 μM β-mercaptoethanol (GIBCO 31350). B16 murine melanoma cell line and HEK293 cells were obtained from ATCC and cultured in DMEM medium supplemented with 10% FBS and 1% penicillin/streptomycin (Thermo Fisher). For routine subculture, 0.25% Trypsin-EDTA (Thermo Fisher) was used for cell dissociation. All cell lines were kept in culture at 37°C in a humidified incubator with 5% CO2 and regularly tested for Mycoplasma contamination (Lonza LT07-318). T Cell Proliferation Assay. OT-I CD8 T cells were purified from TCR-transgenic mice OT-I by negative selection using immunomagnetic beads (Miltenyi Biotech). For direct MHC-I antigen presentation assays, MutuDC lines were seeded at 10,000 cells per well in round-bottom 96-well plates. For MHC-I-restricted anti- gen presentation assays, MutuDC cells were incubated for 2 h with 1 nM SIINFEKL (OVA257–264, Sigma-Aldrich S7951), 1 mg/mL Ovalbumin protein (InvivoGen vac-pova), 1 μg circFOR, or 1 μg circOVA, in the presence or absence of 1 μM CpG (ODN 1585, InvivoGen). The cells were washed three times in medium and incubated with 50,000 purified CFSE-labeled OT-I CD8 T cells (CellTrace CFSE Cell Proliferation Kit, Invitrogen C34554). T cell proliferation was measured after 60 h of culture by flow cytometry analysis excluding doublets and dead cells. OT-I CD8 T cells were gated as CD8+ Vα2+ cells. Live dividing T cells were detected as low for cell proliferation dyes (CFSE low). MutuDC cells were similarly transfected with circOVA with or without CART reagent at the indicated concentrations. qRT-PCR Measurement of Immune Receptors. MutuDC cells were seeded as previously described and treated with 1 μM CpG or 1 μg circRNA in serum- free media. Twenty-four hours after treatment, total RNA was isolated from cells using TRIzol (Invitrogen, 15596018) and Direct-zol RNA Miniprep (Zymo Research, R2052) with on-column DNase I digestion, following the manufacturer’s instruc- tions. RT-qPCR analysis was performed in triplicate using Brilliant II SYBR Green qRT-PCR Master Mix (Agilent, 600825) and a LightCycler 480 (Roche). The relative RNA level was calculated by the ddCt method compared to B-Actin control. Primer sequences are listed in SI Appendix, Table S2. Flow Cytometry Analysis of Cytokines and Surface Receptors. MutuDC cells were seeded as previously described and treated with 1 μM CpG or 1 μg circRNA in serum-free media. Twenty-four hours after treatment, cell supernatant was collected and the cytokine levels were quantified using the cytometric bead array kit for mouse inflammatory cytokines (BD Cytometric Bead Array (CBA) Mouse Inflammation Kit). Similarly, cell suspensions were transferred to a v-bot- tom plate and washed twice with PBS, stained with Live/dead NIR fixable dye, and stained with anti-MHC-II (redFluor 710 Tonbo 80-5321-U025), anti-MCH-I (PE, eBioscience 12-5958-82), anti-CD86 (APCFire/750, BioLegene 105045), anti-CD40 (PerCP-eFluor 710, eBioscience 46-0401-80), and anti-CD80 (Pe-cy5, eBioscience 15-0801-82). After 30-min incubation on ice, the cells were then washed and analyzed by flow cytometry. circOVA Protein Measurements. 8 x105 293T cells were transfected with 5 μg circOVA using TransIT-mRNA transfection kit (Mirus MIR 2250), with 3 µL TransIT-mRNA reagent (Mirus Bio) per microgram of RNA. Twenty-four hours after transfection, the cells were collected and lysed to extract total proteins. Bioruptor sonication with RIPA buffer (150 mM NaCl, 1% Triton X-100, 0.5% sodium deox- ycholate, 0.1% SDS, 50 mM Tris, pH 8.0) was used to lyse the cells. Proteins were fractionated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS- PAGE), transferred to nitrocellulose membranes, blocked in phosphate-buffered saline containing 4% (wt/vol) nonfat milk for 1 h rocking at room temperature, and then incubated overnight at 4°C with the indicated primary antibody. OVA was validated with 1:500 Ovalbumin polyclonal antibody (Novus Biologicals, NB600-922-0.1 mg), and antialpha tubulin antibody was used as loading control (Abcam, ab7192). Secondary antibodies were incubated for 1 h at 1:15,000, in pairs depending on the primary antibody identities: IRDye 800CW Goat anti- Mouse IgG (Li-COR Biosciences, 926-32210) and IRDye 680RD Goat anti-Rabbit IgG (Li-COR Biosciences, 926-68071). Western blot detection and quantifica- tion was done using an Odyssey infrared imaging system (Li-COR Biosciences). Similarly, OVA concentrations in cell culture supernatants were measured 24 h transfection with Ovalbumin (OVA) ELISA Kit (Abbexa, abx259051). Mice and Immunizations. Wild-type C57BL/6J (000664) mice were purchased from Jackson Laboratories. The mice were matched for sex and aged between 8 and 14 wk. For immunization, the mice were injected intranasally with 30 μL circRNA (25 μg per mouse), intravenously with 100 μL circRNA (25 μg per mouse), subcutaneously at the base of the tail with 100 μL circRNA (25 or 50 μg per mouse when indicated), and intraperitoneally with 100 μL CART-circOVA (9 μg per mouse). When indicated, 50 ug Ovalbumin protein (InvivoGen vac-pova) was also delivered in combination with 25 μg Poly(I:C) (HMW VacciGrade, InvivoGen vac-pic) or 50 μL AddaVax (InvivoGen) as per the manufacturer’s instructions (each mouse should receive 50 μL AddaVax if performing subcutaneous injections). The approved institutional animal care and use committee (IACUC) protocols of Stanford University were followed when handling all the mice. Flow Cytometry Analysis of Innate Immune Subsets. Draining inguinal lymph nodes from immunized mice were collected and treated as previously described (47, 48). Briefly, the tissues were treated with collagenase type IV (Worthington) at a concentration of 1 mg/mL for 20 min at a temperature of 37 °C, and then passed through a 100-μm strainer to obtain a single-cell sus- pension. The resulting single-cell samples were stained with a range of mark- ers including Zombie UV (BUV496, BioLegend 423107), anti-Ly6C (BV780, BioLegend 128041), anti-Ly6G (APC-Cy7, BioLegend 127624), anti-CD19 (BUV395, BD 563557), anti-CD3 (BB700, BD742175), anti-MHCII (AF700, eBi- oscience 56-5321-82), anti-CD11b (BV650, BioLegend 101239), anti-CD11c (BV421, BioLegend 117330), anti-CD86 (A647, BioLegend 105020), anti-Siglec-F (PE-CF594, BD 562757), anti-CD45 (BV610, BioLegend 103140), anti-CD169 (PE-Cy7, BioLegend 142412), anti-PDCA-1 (BUV563, BD 749275), anti-CD8a (BUV805, BD 612898), anti-CD103 (PE, eBioscience 12-1031-82), anti-NK1.1 (BV510, BioLegend 108738), and anti-F4/80 (BUV737, BD 749283). The cells were analyzed using the BD FACSymphony analyzer located at the Stanford Shared Fluorescence-activated cell sorting (FACS) Facility. CD8 T Cell Flow Cytometry Analysis. Whole lung, spleen, or peripheral blood from immunized mice was collected after the indicated time points. The lung and spleen were digested with collagenase type IV (Worthington) at a concentration of 1 mg/mL for 20 min at a temperature of 37 °C, and then passed through a 100- μm strainer to obtain a single-cell suspension. Red blood cells were lysed before staining. Single-cell samples were then stained with Zombie Yellow (BUV570, BioLegend 423103), anti-CD3 (clone 145-2C11, BioLegend), anti-CD8α (clone 53-6.7, BioLegend), anti-CD4 (clone RM4-5, BioLegend), anti-CD44 (clone IM7, BioLegend), anti-CD45 (clone 30-F11, BioLegend), anti-CD69 (clone H1.2F3, BioLegend), anti-CD103 (clone 2E7, BioLegend), and Ova-specific tetramer (res- idues 257 to 264). The MHC class I tetramers used in this study (H-2K(b)-SIINFEKL and H-2D(b)-ASNENMETM) were kindly provided by the NIH Tetramer Core Facility. Cells were analyzed with an LSRII.UV analyzer at the Stanford Shared FACS Facility. Antibody ELISA. At the designated time points, serum was collected from the immunized mice. Ovalbumin (OVA) protein was procured from InvivoGen and used to coat high-binding 96-well plates at a concentration of 10 µg/mL in PBS. The plates were blocked with TBS containing 2% BSA and washed before add- ing serially diluted serum samples, which were then incubated at 37 °C for 1 h. After washing the wells three times with TBS and 0.05% Tween-20, secondary HRP-tagged goat anti-mouse IgG and IgA (SouthernBiotech, 1:5,000 dilution) was added, and the wells were incubated for another hour at 37 °C. The wells were washed three times before the addition of 3,3′,5,5′-tetramethylbenzidine substrate solution (Thermo Pierce). The reaction was stopped after 5 min with 0.16 M sulfuric acid. Finally, the optical density at 450 nm was measured with a SpectraMax Microplate Reader. The reciprocal EC50 and end point titers were calculated by GraphPad Prism. Luminex Assay. This assay was performed by the Human Immune Monitoring Center at Stanford University as previously described (47). Mouse 48 plex Procarta kit (Thermo-Fisher/Life Technologies) was used according to the manufacturer's instructions but with some modifications. The samples were added to a plate containing beads that were linked with antibodies and incubated overnight at 4°C with shaking. Then, a biotinylated detection antibody was added for 60 min at room temperature with shaking. After washing the plate, streptavidin-PE was added for 30 min at room temperature, and the plate was washed again. Reading buffer was added to the wells, and each sample was measured in duplicate. The plates were read using a Luminex 200 or a FM3D FlexMap instrument with a lower bound of 50 beads per sample per cytokine. Custom Assay Chex control beads were added to all wells. PNAS 2023 Vol. 120 No. 20 e2302191120 https://doi.org/10.1073/pnas.2302191120 9 of 10 CART Synthesis and circRNA Complex. O6-stat-N6:A9 CARTs (here referred simply as CART) consist of a block of on average 12 subunits made up of a statis- tical 1:1 mixture of oleyl- (O) and nonenyl-substituted (N) carbonate subunits fol- lowed by a block of on average 9 α-amino ester subunits (A). CircRNA formulation with CART was prepared as previously described (36, 37) CircRNA was diluted in PBS pH 5.5 and mixed with CART at 1:10 charge ratio immediately before in vitro transfection in MutuDC or intraperitoneal delivery into mice. 0.05 were considered significant. Sample sizes were not predetermined using statistical methods, and mice were randomly assigned to experimen- tal groups. Data collection and analysis were not blinded to the conditions of the experiments. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. Mouse Model of Subcutaneous Melanoma. B16-F10-OVA cells were harvested for injection in PBS at 1 × 106 cells/mL. One hundred microliters of cell suspension (1 × 105 cells/mouse) was subcutaneously injected into C57BL/6 mice. The mice were monitored daily for tumor incidence and growth. When palpable, tumors were measured every other day using digital calipers and measured in two dimensions. Tumor volume (V) was determined by using the formula V = L × W × D × 3.14/6. The mice were killed before the tumors became necrotic in the center. Statistical Analysis. Prism software version 9.2.0 was used to perform sta- tistical analyses. One-way and two-way ANOVA tests were used to compare more than two groups, and differences between groups with P-values below ACKNOWLEDGMENTS. This study was supported by NIH R35-CA209919 (H.Y.C.), NIH 5R01CA245533-03 (P.A.W.), Parker Institute for Cancer Immunotherapy (PICI) (H.Y.C., B.P.), Emerson Collective (H.Y.C., P.A.W.), the Stanford Center for Molecular Analysis and Design (Z.L.), and Stanford Bio-X (L.A.). We acknowledge the NIH (AI048638, U19 AI090023, and U19 AI057266), the Bill and Melinda Gates Foundation, the Soffer Fund endowment, and Open Philanthropy to B.P. for supporting this work in B.P.’s laboratory. We thank the technical support from the Transgenic, Knockout and Tumor Model Center, Human Immune Monitoring Center, and Veterinary Service Center at Stanford Medicine. FACS analysis was performed on an instrument in the Shared FACS Facility purchased by PICI. H.Y.C. is an investigator of the HHMI. 1. 2. 3. 4. L. S. Kristensen et al., The biogenesis, biology and characterization of circular RNAs. Nat. Rev. Genet. 20, 675–691 (2019). S. Memczak et al., Circular RNAs are a large class of animal RNAs with regulatory potency. Nature 495, 333–338 (2013). T. B. Hansen et al., Natural RNA circles function as efficient microRNA sponges. Nature 495, 384–388 (2013). A. Huang, H. Zheng, Z. Wu, M. Chen, Y. Huang, Circular RNA-protein interactions: Functions, mechanisms, and identification. Theranostics 10, 3503–3517 (2020). R. Ashwal-Fluss et al., circRNA biogenesis competes with pre-mRNA splicing. Mol. Cell 56, 55–66 (2014). 5. 6. Q. Yang et al., A circular RNA promotes tumorigenesis by inducing c-myc nuclear translocation. 27. D. J. Irvine, A. Aung, M. Silva, Controlling timing and location in vaccines. Adv. Drug. Deliv. Rev. 158, 91–115 (2020). 28. S. T. T. Schetters et al., Immunological dynamics after subcutaneous immunization with a squalene- based oil-in-water adjuvant. FASEB J. 34, 12406–12418 (2020). 29. T. Ichinohe et al., Synthetic double-stranded RNA poly(I:C) combined with mucosal vaccine protects against influenza virus infection. J. Virol. 79, 2910–2919 (2005). 30. B. Pulendran, R. Ahmed, Translating innate immunity into immunological memory: Implications for vaccine development. CELL 124, 849–863 (2006). 31. F. Sallusto, A. Lanzavecchia, The instructive role of dendritic cells on T-cell responses. Arthritis Res. 3, S127–S132 (2002). Cell Death Differ. 24, 1609–1620 (2017). 32. H. Acha-Orbea, Novel murine dendritic cell lines: A powerful auxiliary tool for dendritic cell. 7. N. R. Pamudurti et al., Translation of CircRNAs. Mol. Cell 66, 9–21.e7 (2017). 8. C.-K. Chen et al., Structured elements drive extensive circular RNA translation. Mol. Cell 81, 4300–4318.e13 (2021). Y. Yang et al., Extensive translation of circular RNAs driven by N6-methyladenosine. Cell Res. 27, 626–641 (2017). 9. 10. B. Beckert, B. Masquida, Synthesis of RNA by in vitro transcription. Methods Mol. Biol. 703, 29–41 (2011). Research 3, 1–25 (2012). 33. D. Askew, R. S. Chu, A. M. Krieg, C. V. Harding, CpG DNA induces maturation of dendritic cells with distinct effects on nascent and recycling MHC-II antigen-processing mechanisms. J. Immunol. 165, 6889–6895 (2000). 34. X. Wu, F. Xu, J. Liu, G. Wang, Comparative study of dendritic cells matured by using IL-1β, IL-6, TNF-α and prostaglandins E2 for different time span. Exp. Ther Med. 14, 1389–1394 (2017). 35. A. Cortez, E. Muxfeldt, Monocyte chemoattractant protein-1 and hypertension: An overview. 11. R. M. Meganck et al., Engineering highly efficient backsplicing and translation of synthetic circRNAs. Hipertens Riesgo Vasc. 39, 14–23 (2022). Mol. Ther. Nucleic Acids 23, 821–834 (2021). 36. C. J. McKinlay et al., Charge-altering releasable transporters (CARTs) for the delivery and release of 12. R. Chen et al., Engineering circular RNA for enhanced protein production. Nat. Biotechnol. 41, mRNA in living animals. Proc. Natl. Acad. Sci. U.S.A. 114, E448–E456 (2017). 262–272 (2023). 13. R. A. Wesselhoeft, P. S. Kowalski, D. G. Anderson, Engineering circular RNA for potent and stable translation in eukaryotic cells. Nat. Commun. 9, 2629 (2018). 37. C. J. McKinlay, N. L. Benner, O. A. Haabeth, R. M. Waymouth, P. A. Wender, Enhanced mRNA delivery into lymphocytes enabled by lipid-varied libraries of charge-altering releasable transporters. Proc. Natl. Acad. Sci. U.S.A. 115, E5859–E5866 (2018). 14. C.-X. Liu et al., Structure and degradation of circular RNAs regulate PKR activation in innate 38. O. P. Joffre, E. Segura, A. Savina, S. Amigorena, Cross-presentation by dendritic cells. Nat. Rev. immunity. Cell 177, 865–880.e21 (2019). Immunol. 12, 557–569 (2012). 15. X. Xia, X. Tang, S. Wang, Roles of CircRNAs in autoimmune diseases. Front Immunol. 10, 639 (2019). 16. Z. Feng et al., Functions and potential applications of circular RNAs in cancer stem cells. Front. Oncol. 9, 500 (2019). 17. Y. G. Chen et al., Sensing self and foreign circular RNAs by intron identity. Mol. Cell 67, 228–238.e5 (2017). 18. Y. G. Chen et al., N6-methyladenosine modification controls circular RNA immunity. Mol. Cell 76, 96–109.e9 (2019). 19. C.-X. Liu, L.-L. Chen, Expanded regulation of circular RNA translation. Mol. Cell 81, 4111–4113 (2021). 39. O. A. W. Haabeth et al., An mRNA SARS-CoV-2 vaccine employing charge-altering releasable transporters with a TLR-9 agonist induces neutralizing antibodies and T cell memory. ACS Cent Sci. 7, 1191–1204 (2021). 40. E. A. Hemann, S.-M. Kang, K. L. Legge, Protective CD8 T cell-mediated immunity against influenza A virus infection following influenza virus-like particle vaccination. J. Immunol. 191, 2486–2494 (2013). 41. K. D. Zens, J. K. Chen, D. L. Farber, Vaccine-generated lung tissue-resident memory T cells provide heterosubtypic protection to influenza infection. JCI Insight 1, e85832 (2016), 10.1172/jci. insight.85832. 20. Z. Zhang, T. Yang, J. Xiao, Circular RNAs: Promising biomarkers for human diseases. EBioMedicine 42. T. K. Tsang et al., Investigation of CD4 and CD8 T cell-mediated protection against influenza A virus 34, 267–274 (2018). in a cohort study. BMC Med. 20, 230–233 (2022). 21. E. Lasda, R. Parker, Circular RNAs co-precipitate with extracellular vesicles: A possible mechanism for 43. P. M. Taylor, B. A. Askonas, Influenza nucleoprotein-specific cytotoxic T-cell clones are protective circRNA Clearance. PLoS One 11, e0148407 (2016). in vivo. Immunology 58, 417–420 (1986). 22. M. Vausort et al., Myocardial infarction-associated circular RNA predicting left ventricular 44. A. J. McMichael, F. M. Gotch, G. R. Noble, P. A. Beare, Cytotoxic T-cell immunity to influenza. N. Engl. dysfunction. J. Am. Coll Cardiol. 68, 1247–1248 (2016). J. Med. 309, 13–17 (1983). 23. E. B. Hansen et al., The transcriptional landscape and biomarker potential of circular RNAs in 45. C. E. van de Sandt et al., Influenza B virus-specific CD8+ T-lymphocytes strongly cross-react with prostate cancer. Genome Med. 14, 8–16 (2022). viruses of the opposing influenza B lineage J. Gen. Virol. 96, 2061–2073 (2015). 24. L. Qu et al., Circular RNA vaccines against SARS-CoV-2 and emerging variants. Cell 185, 1728–1744. 46. P. L. Smith, K. Piadel, A. G. Dalgleish, Directing T-cell immune responses for cancer vaccination and e16 (2022). immunotherapy. Vaccines 9, 1392 (2021). 25. P. S. Arunachalam et al., T cell-inducing vaccine durably prevents mucosal SHIV infection even with 47. C. Li et al., Mechanisms of innate and adaptive immunity to the Pfizer-BioNTech BNT162b2 vaccine. lower neutralizing antibody titers. Nat. Med. 26, 932–940 (2020). Nat. Immunol. 23, 543–555 (2022). 26. C. Petitdemange et al., Vaccine induction of antibodies and tissue-resident CD8+ T cells enhances 48. L. Grigoryan et al., Adjuvanting a subunit SARS-CoV-2 vaccine with clinically relevant adjuvants protection against mucosal SHIV-infection in young macaques JCI Insight 4, 98 (2019). induces durable protection in mice. NPJ Vaccines 7, 14–55 (2022). 10 of 10 https://doi.org/10.1073/pnas.2302191120 pnas.org
10.1073_pnas.2221809120	RESEARCH ARTICLE \| MEDICAL SCIENCES OPEN ACCESS Proxalutamide reduces SARS- CoV- 2 infection and associated inflammatory response Yuanyuan Qiaoa,b,c,1 Abhijit Paroliaa, Tongchen Hea, Caleb Chenga, Xuhong Caoa, Rui Wanga Qianxiang Zhoug, Liandong Mag, Jonathan Z. Sextond,e,h,i,j, and Arul M. Chinnaiyana,b,c,k,l,2 , Charles J. Zhangd, Yuping Zhanga,b, Xia Jianga, Carla D. Prettoe, Sanjana Eyunnia, , Fengyun Sua, Stephanie J. Ellisona, Yini Wangf , Jesse W. Wotringd,1 , Yang Zhenga,1 , Jun Qinf, Honghua Yang, Contributed by Arul M. Chinnaiyan; received December 23, 2022; accepted June 12, 2023; reviewed by Thirumala- Devi Kanneganti and Amy Moran Early in the COVID- 19 pandemic, data suggested that males had a higher risk of devel- oping severe disease and that androgen deprivation therapy might be associated with protection. Combined with the fact that TMPRSS2 (transmembrane serine protease 2), a host entry factor for the SARS- CoV- 2 virus, was a well- known androgen- regulated gene, this led to an upsurge of research investigating androgen receptor (AR)- targeting drugs. Proxalutamide, an AR antagonist, was shown in initial clinical studies to benefit COVID- 19 patients; however, further validation is needed as one study was retracted. Due to continued interest in proxalutamide, which is in phase 3 trials, we examined its ability to impact SARS- CoV- 2 infection and downstream inflammatory responses. Proxalutamide exerted similar effects as enzalutamide, an AR antagonist prescribed for advanced prostate cancer, in decreasing AR signaling and expression of TMPRSS2 and angiotensin- converting enzyme 2 (ACE2), the SARS- CoV- 2 receptor. However, proxal- utamide led to degradation of AR protein, which was not observed with enzalutamide. Proxalutamide inhibited SARS- CoV- 2 infection with an IC50 value of 97 nM, compared to 281 nM for enzalutamide. Importantly, proxalutamide inhibited infection by mul- tiple SARS- CoV- 2 variants and synergized with remdesivir. Proxalutamide protected against cell death in response to tumor necrosis factor alpha and interferon gamma, and overall survival of mice was increased with proxalutamide treatment prior to cytokine exposure. Mechanistically, we found that proxalutamide increased levels of NRF2, an essential transcription factor that mediates antioxidant responses, and decreased lung inflammation. These data provide compelling evidence that proxalutamide can prevent SARS- CoV- 2 infection and cytokine- induced lung damage, suggesting that promising clinical data may emerge from ongoing phase 3 trials. proxalutamide \| SARS- CoV- 2 \| COVID- 19 \| androgen receptor \| cytokines Over 3 y have passed since the first documented cases of COVID- 19 arose from infec- tion by the severe acute respiratory syndrome coronavirus 2 (SARS- CoV- 2), yet many challenges remain worldwide in preventing and treating the disease (1). Robust vac- cination campaigns led to rapid development, testing, and deployment of several vaccines effective against infection and serious illness from the initial SARS- CoV- 2 genetic lineages (2–6). However, as the pandemic continued, waning vaccine protection and emergence of new variants have led to breakthrough infections, as well as many people now having been infected multiple times (5–9). Booster vaccines, including bivalent boosters effective against the highly transmissible omicron variant, have been developed in an effort to overcome these challenges (10). Oral antivirals such as mol- nupiravir and nirmatrelvir–ritonavir have been developed for high- risk individuals who contract COVID- 19, but these are also met with obstacles like potential recurrent infections or contraindications with other commonly prescribed drugs (11–14). Together, these challenges highlight the ongoing critical need for new therapeutics to combat SARS- CoV- 2. As it is the initial step in the viral life cycle, the entry process has been intensely studied to understand how to potentially block SARS- CoV- 2 infection (15). Early data during the pandemic showed that the spike (S) protein of SARS- CoV- 2 binds to host angiotensin- converting enzyme 2 (ACE2) receptors on the cell surface to initiate entry (16, 17). Cleavage of the spike protein by transmembrane serine protease 2 (TMPRSS2) facilitates fusion of the viral and cell membranes and cell entry (18, 19). With the presumed advantage that it will be difficult for the virus to mutate and evade host- directed drugs, multiple preclinical and clinical research efforts have since followed examining the efficacy of therapies directly targeting TMPRSS2 and ACE2, albeit with mixed results and several studies still ongoing (20–25). Significance Drugs that target androgen receptor (AR) signaling, including those that inhibit production of androgen ligands (degarelix) and those that bind to and directly block AR activity (enzalutamide), have been investigated in clinical trials for the treatment of COVID- 19 but failed to produce positive results. Another AR antagonist, proxalutamide, is in ongoing phase 3 studies for COVID- 19 after showing initial positive findings. Data from this study show that proxalutamide can inhibit infection of multiple variants of SARS- CoV- 2 in vitro. These data suggest that proxalutamide should continue to be investigated in clinical studies as a potential therapy for COVID- 19. Author contributions: Y.Q., J.W.W., Y. Zheng, J.Z.S., and A.M.C. designed research; Y.Q., J.W.W., Y. Zheng, C.J.Z., Y. Zhang, X.J., C.D.P., S.E., A.P., T.H., C.C., X.C., R.W., F.S., Y.W., J.Q., H.Y., Q.Z., L.M., and J.Z.S. performed research; Y.Q., J.W.W., Y. Zheng, C.J.Z., Y. Zhang, X.J., C.D.P., S.E., A.P., T.H., C.C., X.C., R.W., F.S., Y.W., J.Q., H.Y., Q.Z., L.M., J.Z.S., and A.M.C. analyzed data; and Y.Q., J.W.W., Y. Zheng, S.J.E., J.Z.S., and A.M.C. wrote the paper. Reviewers: T.- D.K., St. Jude Children’s Research Hospital; and A.M., Oregon Health and Science University. Competing interest statement: H.Y., Q.Z., and L.M. are part of Kintor Pharmaceutical Limited. The remaining authors declare no competing interests. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1Y.Q., J.W.W., and Y. Zheng contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221809120/- /DCSupplemental. Published July 17, 2023. PNAS 2023 Vol. 120 No. 30 e2221809120 https://doi.org/10.1073/pnas.2221809120 1 of 10 Since TMPRSS2 is a well- known androgen receptor (AR)- regulated gene, early hypotheses suggested that inhibition of AR activity could be a potential treatment strategy for COVID- 19 (26). As demo- graphic data became available, many reports also observed that males had higher incidences of severe SARS- CoV- 2 infections that required intensive care unit (ICU) admission or resulted in death (27–29). In further support of the initial hypothesis that AR activity may drive COVID- 19 pathogenesis, a retrospective study during the first months of the pandemic observed a reduced incidence of SARS- CoV- 2 infections in prostate cancer patients taking androgen deprivation therapy (ADT) compared to those not receiving ADT (30). Other small studies supported this observation and the prem- ise that anti- androgens could be protective against severe COVID- 19 (31, 32); however, these results quickly became debated as other studies found no association between ADT and SARS- CoV- 2 infec- tivity (33–35). These preliminary observations prompted a burst of basic science and clinical studies to attempt to elucidate the role of androgens in SARS- CoV- 2 infection and determine whether AR inhibitors could be viable treatment options for COVID- 19. Studies with AR antagonists prescribed for prostate cancer treatment (e.g., enzalu- tamide, apalutamide, and darolutamide) have since shown that SARS- CoV- 2 infectivity can be decreased in vitro in certain con- texts with these drugs (36–38). However, some randomized, con- trolled clinical trials of AR inhibition in COVID- 19 patients have not produced encouraging results. For instance, in the Hormonal Intervention for the Treatment in Veterans with COVID- 19 Requiring Hospitalization (HITCH) trial (NCT04397718) which tested degarelix, a gonadotropin- releasing hormone (GnRH) antag- onist that rapidly suppresses testosterone levels, in male veterans hospitalized with COVID- 19, no improvement in clinical out- come was observed compared to placebo (39). Similarly, the COVIDENZA trial (NCT04475601) found no improvement in outcome of COVID19- positive male or female patients who were randomized to treatment with enzalutamide vs. standard of care (40). In contrast, the AR antagonist proxalutamide was also tested as a possible treatment for COVID- 19 in randomized, controlled trials and showed encouraging positive benefits (41–43), but these findings were met with caution from the scientific community after a retraction statement was issued for one of the publications, citing concerns over randomization (44). Proxalutamide is cur- rently in additional phase 3 trials for COVID- 19 in both outpa- tient (NCT04870606 and NCT04869228) and hospital (NCT05009732) settings in different countries, including the United States. Proxalutamide was originally developed as an AR antagonist for advanced prostate cancer and is in ongoing phase 2 clinical trials for this indication as well (45–47). Our previous study found that AR antagonists (enzalutamide, apalutamide, and darolutamide) and degraders decreased TMPRSS2 and ACE2 expressions and were potent inhibitors of SARS- CoV- 2 infectivity in vitro (37). Given these data and the continued clinical interest surrounding proxalutamide in COVID- 19, we sought to test prox- alutamide for its ability to impact SARS- CoV- 2 infection. We find that proxalutamide inhibits cellular infection by multiple SARS- CoV- 2 variants and shows synergistic activity in vitro with remdesivir, an antiviral demonstrated to have clinical benefit in COVID- 19 patients (48, 49). Additionally, in vivo studies showed that prophylactic treatment with proxalutamide can improve over- all survival in mouse models of the TNFα (tumor necrosis factor alpha) and IFNγ (interferon gamma)- induced cytokine storm triggered by SARS- CoV- 2 infection (50), potentially occurring through increases in the nuclear factor erythroid 2- related factor 2 (NRF2) transcription factor responsible for mediating cellular antioxidant responses. Altogether, this study provides characteri- zation of proxalutamide in SARS- CoV- 2 infection models and provides data to possibly explain positive results that may emerge from clinical trials of proxalutamide for COVID- 19 treatment. Results Proxalutamide is an AR antagonist recently developed for castration- resistant prostate cancer (CRPC) (47), in comparison to enzalu- tamide which has been commonly prescribed for CRPC treatment for several years (51). To first compare the transcriptomic changes associated with proxalutamide and enzalutamide, RNA- sequencing (RNA- Seq) analysis was carried out in AR- positive prostate cancer Lymph Node Carcinoma of the Prostate (LNCaP) cells using either 20 µM proxalutamide or enzalutamide for 8 h of treatment. Gene set enrichment analysis was achieved by examining differ- entially expressed genes in either proxalutamide- or enzalutamide- treated cells compared to control. The normalized enrichment score results indicated that androgen responses were the top down- regulated hallmark in both proxalutamide- and enzalutamide- treated LNCaP cells (Fig. 1A). Gene set enrichment analysis on androgen responses further confirmed that proxalutamide signif- icantly down- regulated androgen- regulated genes that were sup- pressed by enzalutamide (Fig. 1B), suggesting proxalutamide suppresses AR signaling. In addition, the effect of proxalutamide on cell proliferation was examined in LNCaP cells and a castration- resistant variant of LNCaP called C4- 2B cells. In both LNCaP and C4- 2B cells, proxalutamide and enzalutamide treatment resulted in dose- dependent inhibition of cell proliferation in vitro, but growth inhibition was greater with proxalutamide treatment compared to enzalutamide at the same concentrations (Fig. 1 C and D). Importantly, we found that proxalutamide not only sup- pressed AR signaling but also decreased AR protein levels, which were not altered by enzalutamide treatment (Fig. 1E), indicating that proxalutamide possesses stronger inhibition of the AR sign- aling pathway than enzalutamide. Previously, we reported that enzalutamide can transcriptionally down- regulate SARS- CoV- 2 entry factors TMPRSS2 and ACE2 (37). Here, we found that proxalutamide had the same ability to decrease TMPRSS2 and ACE2 (Fig. 1F). Thus, we postulated that proxalutamide may block SARS- CoV- 2 infection. Employing a SARS- CoV- 2 bioassay platform, we have established an in vitro system with which to examine the various strains of authentic SARS- CoV- 2 viral infection (37, 52). In this system, cells were pretreated with the experimental compounds for 24 h prior to SARS- CoV- 2 infection for an additional 72 h (Fig. 2A). The results showed that proxalutamide decreased cellular infection by the WA1 strain of SARS- CoV- 2 in a dose- dependent manner with an IC50 value of 97 nM, whereas enzalutamide decreased infectivity with an IC50 value of 281 nM (Fig. 2B). Representative images of cellular infectivity by the WA1 strain of SARS- CoV- 2 in control- , proxalutamide- , or enzalutamide- treated conditions confirmed that decreased infection could be achieved by the AR antagonists prox- alutamide and enzalutamide in LNCaP cells (Fig. 2C). Since several variants of the SARS- CoV- 2 virus have emerged throughout the pandemic, we examined the effect of proxalutamide against infec- tion of multiple strains. The results indicated that proxalutamide possessed robust inhibitory effects in blocking SARS- CoV- 2 infec- tion by the most common strains, including WA1, alpha, delta, and omicron, with IC50 values of 69 nM, 48 nM, 98 nM, and 581 nM, respectively, in LNCaP cells (Fig. 2D). Furthermore, remdesivir is a Food and Drug Administration (FDA)- approved agent for treatment of SARS- CoV- 2 infection (48, 49). The combinatorial effect of proxalutamide or enzalutamide 2 of 10 https://doi.org/10.1073/pnas.2221809120 pnas.org A Enza * * Proxa * * * * * * * * * * * * * * * * * * * * * * * * * HALLMARK_INFLAMMATORY_RESPONSE HALLMARK_APOPTOSIS HALLMARK_CHOLESTEROL_HOMEOSTASIS HALLMARK_MTORC1_SIGNALING HALLMARK_MYOGENESIS HALLMARK_HEME_METABOLISM HALLMARK_PI3K_AKT_MTOR_SIGNALING HALLMARK_UNFOLDED_PROTEIN_RESPONSE HALLMARK_COMPLEMENT HALLMARK_TNFA_SIGNALING_VIA_NFKB NES 2 1 0 −1 −2 HALLMARK_P53_PATHWAY HALLMARK_HYPOXIA HALLMARK_INTERFERON_GAMMA_RESPONSE HALLMARK_ESTROGEN_RESPONSE_EARLY HALLMARK_G2M_CHECKPOINT HALLMARK_E2F_TARGETS HALLMARK_MYC_TARGETS_V1 HALLMARK_MYC_TARGETS_V2 HALLMARK_ANDROGEN_RESPONSE t n u o C l l e C 4×10 4 3×10 4 2×10 4 1×10 4 0 C E LNCaP Enzalutamide Proxalutamide 1 0 0 0 . 0 < p 0 24 48 72 96 144 0 24 48 72 96 120 144 120 LNCaP D t n u o C l l e C 7×10 4 6×10 4 5×10 4 4×10 4 3×10 4 2×10 4 1×10 4 0 30 µM 10 µM 3.33 µM 1.11 µM Ctrl 1 0 0 0 . 0 < p Hours Proxalutamide Enzalutamide 0 5 10 20 0 5 10 20 µM AR PSA GAPDH y t i s n e t n i d n a b H D P A G R A / 125 100 75 50 25 0 Proxalutamide Enzalutamide 3 8 2 0 . 0 = p 5 0 10 15 20 Concentration (μM) B HALLMARK_ANDROGEN_RESPONSE_ Proxalutamide 0.0 −0.2 −0.4 e r o c s t n e m h c i r n e NES = −1.56 p.adj = 0.009 −0.6 0 5000 rank 10000 HALLMARK_ANDROGEN_RESPONSE_Enzalutamide 0.00 e r o c s t n e m h c i r n e −0.25 −0.50 −0.75 NES = −1.99 p.adj = 0.007 0 5000 rank 10000 C4-2B Enzalutamide Proxalutamide 3 0 0 0 . 0 = p 30 µM 10 µM 3.33 µM 1.11 µM Ctrl 1 0 0 0 . 0 < p Hours 0 24 48 72 96 F 120 144 0 24 48 72 96 120 144 ACE2 TMPRSS2 p<0.0001 e g n a h c d o f l 1.25 1.00 0.75 A N R m e v i t l a e R 0.50 0.25 0.00 0.5(cid:31)M Ctrl 3(cid:31)M 1(cid:31)M 3(cid:31)M Enzalutamide Proxalutamide Proxalutamide ARD61 e g n a h c d o l f A N R m e v i t a e R l 1.25 1.00 0.75 0.50 0.25 0.00 1 0 0 0 . 0 < p 3 0 0 0 . 0 = p 1 0 0 0 . 0 < p 1 0 0 0 . 0 < p 0.5(cid:31)M Ctrl 1(cid:31)M 3(cid:31)M 3(cid:31)M Proxalutamide Enzalutamide Proxalutamide ARD61 Fig. 1. Proxalutamide is a recently developed AR antagonist that also down- regulates AR protein levels. (A) Hallmark of differential expressed gene signatures in proxalutamide (Proxa) and enzalutamide (Enza) treatment vs. control in LNCaP cells; the asterisk indicates a P value of less than 0.01. (B) Gene set enrichment of the androgen response pathway in proxalutamide- or enzalutamide- treated LNCaP cells. (C) Cell growth inhibition in enzalutamide- or proxalutamide- treated LNCaP cells. Ctrl, control. P values were calculated by the two- tailed unpaired t test between ctrl and 30 µM enzalutamide or proxalutamide (not between each dose). (D) Cell growth inhibition in enzalutamide- or proxalutamide- treated C4- 2B cells. Ctrl, control. P values were calculated by the two- tailed unpaired t test between ctrl and 30 µM enzalutamide or proxalutamide (not between each dose). (E) Immunoblotting of AR and PSA protein in LNCaP cells after treatment with various concentrations of proxalutamide and enzalutamide for 24 h. Quantification of band intensity of AR/GAPDH is shown on the right. P values were calculated by the two- tailed unpaired t test between 20 µM proxalutamide and enzalutamide. (F) Relative mRNA expression of ACE2 and TMPRSS2 in LNCaP cells after the indicated treatment. P values were calculated by the two- tailed unpaired t test between control and the indicated treatment. and remdesivir in preventing infection by the SARS- CoV- 2 alpha strain was examined in induced human alveolar cells (iAEC2) (Fig. 3A). The results indicated that proxalutamide had a strong synergistic effect with remdesivir in inhibition of alpha strain infection and achieved 100% protection against infection (Fig. 3B), with a synergy score of 14.516 (Fig. 3C). Similarly, the enzalutamide and remdesivir combination achieved synergy but with a slightly weaker synergistic effect than the proxalutamide and remdesivir PNAS 2023 Vol. 120 No. 30 e2221809120 https://doi.org/10.1073/pnas.2221809120 3 of 10 A B C D SARS-CoV2 Bioassay Day 0 Day 1 Day 2 Day 4 Seed LNCaP cells in 384 well plate Pre-incubate compounds for 24 hours Infect with SARS-CoV-2 virus Fix, Stain, Image Proxalutamide Enzalutamide % n o i t c e f n I 150 100 50 0 IC50: 97 nM 150 100 50 0 V i a b i l i t y % % n o i t c e f n I 150 100 50 0 IC50: 281 nM 10 -8 10 -7 Concentration (M) 10 -6 10 -5 10 -9 10 -8 10 -7 Concentration (M) 10 -6 150 100 50 0 V i a b i l i t y % 10 -5 3 µM 750 nM 188 nM 23 nM Viral Control 10 -9 i e d m a t u a x o r P l i e d m a t u a z n E l LNCaP 150 100 50 % n o i t c e f n I IC50 WA1: 69 nM IC50 Alpha: 48 nM IC50 Delta: 98 nM IC50 Omicron: 581 nM 0 10 -9 10 -8 10 -7 Proxalutamide [M] 10 -6 150 100 50 % v i a b i l i t y 0 10 -5 Viability WA1 strain Alpha strain Delta strain Omicron strain Fig. 2. Proxalutamide inhibits multiple strains of SARS- CoV- 2 infection in vitro. (A) Schematic illustration of the SARS- CoV- 2 bioassay. (B) Dose- dependent inhibition of SARS- CoV- 2 WA1 strain infection by proxalutamide and enzalutamide in LNCaP cells with IC50 values shown for each. Cell viability is also graphed. (C) Representative images of SARS- CoV- 2 WA1 strain infection after proxalutamide or enzalutamide treatment in LNCaP cells. (D) Dose- dependent inhibition of infection by multiple strains of SARS- CoV- 2 with proxalutamide treatment in LNCaP cells. combination (Fig. 3 E and F). Both proxalutamide or enzalutamide and remdesivir combination treatments had no detrimental effects on the viability of iAEC2 cells (Fig. 3 D and G). These results suggest that proxalutamide may have clinical utility in combination with current SARS- CoV- 2 treatments, such as remdesivir. SARS- CoV- 2- induced mortality is largely triggered by a cytokine storm that occurs in the pulmonary system and systemically (53). It has been reported that TNFα and INFγ can act synergistically to trigger inflammatory cell death in vitro and in vivo, which mimics the SARS- CoV- 2- induced cytokine shock syndrome (CSS) that 4 of 10 https://doi.org/10.1073/pnas.2221809120 pnas.org A B ) M n ( r i v i s e d m e R E ) M n ( r i v i s e d m e R Inhibition of SARS-CoV-2 infection 300 56.99 86.94 82.92 83.59 88.73 97.83 100.00 100 45.09 43.77 57.88 55.17 52.51 88.21 99.06 30 10 0 32.64 37.43 39.69 40.22 64.38 74.19 96.33 11.94 14.53 19.55 25.93 25.47 63.42 95.70 0.00 6.01 6.94 4.58 10.57 56.08 82.77 0 30 100 1000 300 Proxalutamide (nM) 3000 10000 C F Inhibition of SARS-CoV-2 infection 300 56.99 66.23 70.50 73.50 79.06 92.28 100 45.09 63.56 71.99 83.04 86.73 94.98 30 10 0 32.64 64.21 69.54 79.51 85.20 97.75 11.94 44.63 55.47 65.48 85.39 95.91 0.00 1.49 30.45 62.57 79.61 97.95 0 30 100 300 1000 3000 Enzalutamide (nM) D Viability 300 93.44 124.35 114.51 129.68 104.44 118.83 106.49 100 92.50 105.26 113.44 117.08 91.37 113.52 122.49 30 101.20 102.79 92.72 109.47 97.41 134.10 120.49 10 107.50 121.99 131.75 109.71 109.68 112.65 117.08 0 100.00 132.56 116.73 128.47 148.89 121.12 97.83 0 30 100 1000 300 Proxalutamide (nM) 3000 10000 G Viability 300 93.44 99.96 106.75 103.63 101.60 98.53 100 92.50 105.09 90.65 91.27 86.04 88.77 30 101.20 106.26 108.09 95.80 95.52 100.91 10 107.50 136.12 108.35 101.32 112.93 103.38 0 100.00 97.32 90.03 93.52 149.88 108.82 0 30 100 300 1000 3000 Enzalutamide (nM) ) M n ( r i v i s e d m e R ) M n ( i r i v s e d m e R Fig. 3. Proxalutamide and remdesivir combination exerts strong synergistic effect in blocking SARS- CoV- 2 infection in iAEC2. (A) Schematic illustration of the study design of the SARS- CoV- 2 bioassay on iAEC2 cells. (B) Combination matrix of proxalutamide and remdesivir in inhibition of SARS- CoV- 2 alpha strain infection. (C) Bliss synergy score of proxalutamide and remdesivir against SARS- CoV- 2 alpha strain infection. (D) Combination matrix of cell viability on proxalutamide and remdesivir. (E) Combination matrix of enzalutamide and remdesivir in inhibition of SARS- CoV- 2 alpha strain infection. (F) Bliss synergy score of enzalutamide and remdesivir against SARS- CoV- 2 alpha strain infection. (G) Combination matrix of cell viability on enzalutamide and remdesivir. occurs in COVID- 19 patients (50). Specifically, TNFα and INFγ induce a type of inflammatory cell death called PANoptosis, which is regulated by the PANoptosome and involves molecular compo- nents of pyroptosis, apoptosis, and necroptosis (50, 54). In an AR- positive lung cell line, H1437, we demonstrated that the com- bination of TNFα and INFγ induced maximal cell death compared to either cytokine alone (Fig. 4A). Interestingly, the cell death induced by combination treatment with TNFα and INFγ was atten- uated by proxalutamide and another AR antagonist darolutamide in a dose- dependent manner (Fig. 4B) but not by enzalutamide or apalutamide (SI Appendix, Fig. S1A). Additionally, the cell death triggered by TNFα and INFγ combination treatment was confirmed by elevated cleaved PARP (c- PARP) levels, which were dose dependently blocked by proxalutamide and darolutamide (Fig. 4C) but not enzalutamide or apalutamide (SI Appendix, Fig. S1B). Similarly, AR protein levels were down- regulated by proxalutamide and darolutamide (Fig. 4D) but not enzalutamide or apalutamide (SI Appendix, Fig. S1C). This suggests that AR antagonists such as proxalutamide or darolutamide may provide additional benefits in terms of reducing CSS in vivo. In normal mouse prostate organoids, we confirmed that proxalutamide inhibited murine AR signaling by decreasing androgen (dihydrotestosterone, DHT)- stimulated induc- tion of Fkbp5 and Psca target genes; additionally, proxalutamide decreased Ar mRNA levels (SI Appendix, Fig. S2A). These results prompted us to examine the in vivo efficacy of proxalutamide in preventing death in the TNFα and INFγ CSS model (50) in wild- type C57BL6 male mice. We tested two treatment regimens of proxalutamide prior to cytokine challenge with the TNFα and PNAS 2023 Vol. 120 No. 30 e2221809120 https://doi.org/10.1073/pnas.2221809120 5 of 10 A B C H1437 lung adenocarcinoma Ctrl TNFα IFNγ TNFα+IFNγ 1500 1000 500 e c n e u l f n o c / s l l e c I + P 0 0 24 48 72 Hours of treatment 1500 1000 500 e c n e u l f n o c / s l l e c I + P 0 0 TNFα+IFNγ DMSO Proxalutamide 5μM Proxalutamide 10μM Proxalutamide 20μM 24 48 72 Hours of treatment 1 0 0 0 0 < p . 3 0 0 0 . 0 = p Ctrl TNFα IFNγ TNFα+IFNγ TNFα+IFNγ DMSO Darolutamide 10μM Darolutamide 20μM 7 1 0 0 . 0 = p 24 48 72 Hours of treatment 1500 1000 500 e c n e u l f n o c / s l l e c I + P 0 0 D Darolutamide Proxalutamide TNFα+IFNγ + ++ - - - + ++ + ++ + ++ - - + + + + + Proxalutamide Darolutamide 10 20 10 20 c-PARP Vinculin µM µM AR Vinculin Fig. 4. Proxalutamide attenuates CSS–related cell death and mortality. (A) Real- time analysis of cell death in H1437 cells in vitro under control, TNFα, IFNγ, or combination treatment. Representative images of dead cells under the indicated conditions are shown on the Right. The P value was calculated by the two- tailed unpaired t test between control and TNFα/IFNγ combination treatment. (B) Real- time analysis of cell death in H1437 cells in vitro under TNFα and IFNγ combination and various concentrations of proxalutamide or darolutamide. P values were calculated by the two- tailed unpaired t test comparing dimethylsulfoxide (DMSO) control and 20 µM proxalutamide or darolutamide. (C) Immunoblotting of c- PARP and vinculin (loading control) in H1437 cells after treatment with 10 and 20 µM of proxalutamide or darolutamide with or without TNFα and IFNγ combination for 72 h. (D) Immunoblotting of AR and vinculin in H1437 after treatment with 10 and 20 µM of proxalutamide or darolutamide for 72 h. INFγ combination. The data showed that both proxalutamide treat- ment regimens reduced mortality induced by TNFα and INFγ (SI Appendix, Fig. S2 B and C). Histology evaluation of tissue dam- age triggered by TNFα and INFγ combination was examined in the small intestine and lung (SI Appendix, Fig. S2D). Compared with the PBS treated group, atrophy of the villi and an increase in inflam- matory cell infiltration in the lamina propria area of the intestine were observed post- TNFα and IFNγ treatment, which was largely alleviated with proxalutamide treatment. In addition, TNFα and IFNγ treatment induced interlobular septal thickening in the lungs of mice showing focal epithelial hyperplasia, and such effects were rescued by proxalutamide treatment. Thus, these results suggest that proxalutamide may reduce TNFα and IFNγ cytokine storm- induced cell death in vitro and in vivo. The NRF2 pathway is an important part of cellular defense through the production of antioxidants, which occurs via binding of the NRF2 transcription factor to antioxidant response elements in target genes (55–57). The upregulation of NRF2 has been reported to control inflammation in several studies (56–60). Here, we found that proxalutamide increases NRF2 transcriptional activity by enhancing NRF2 DNA binding in RAW264.7 and THP- 1 cells (Fig. 5A). In RAW264.7 cells, proxalutamide also up- regulated NRF2 protein expression in lipopolysaccharide (LPS)- stimulated conditions (Fig. 5B). In the in vitro CSS model triggered by TNFα and INFγ combination treatment, proxalutamide augmented NRF2 protein levels and decreased cell death in THP- 1 cells (Fig. 5 C and D). Apoptotic cell death triggered by TNFα and INFγ combination treatment was attenuated by proxalutamide (Fig. 5E). Next, we 6 of 10 https://doi.org/10.1073/pnas.2221809120 pnas.org A C TNFα IFNγ F G B Proxalutamide LPS RAW264.7 - 0.3 1 310 + + - - + 1.0 1.3 1.3 1.4 1.6 1.6 + + µM NRF2 GAPDH THP-1 (PMA induced macrophage) DMSO Proxalutamide + + + 1.0 2.5 0.4 0.8 3.1 3.5 3.0 3.3 + + + + + D THP-1 1500 TNFα+IFNγ Proxalutamide E e c n e u l f n o c / s l l e c I + P 1000 500 0 0 NRF2 Vinculin TNFα+IFNγ PBS Proxalutamide TNFα+IFNγ - - - + + - + + 2 0 0 0 . 0 = p 1 0 0 0 . 0 < p c-PARP GAPDH 24 48 72 Hours of treatment p<0.01 4 0 1 x F L A B n i r e b m u n l l e c l a t o T 700 600 500 400 300 200 100 0 1 0 0 0 . 0 < p 4 0 1 x r e b m u n s l i h p o r t u e N 500 400 300 200 100 0 1 0 0 0 . 0 < p 5 0 . 0 < p 1 0 . 0 < p 1 0 0 0 . 0 < p Normal-Vehicle Poly (I:C) Model-Vehicle Poly (I:C) Model-Dex+Roflumilast Poly (I:C) Model-Proxalutamide 20mg/kg Poly (I:C) Model-Proxalutamide 40mg/kg Fig. 5. Proxalutamide enhances NRF2 transcriptional activity and inhibits acute immune response in the poly (I:C)- induced lung injury animal model. (A) Proxalutamide increased NRF2 transcriptional activity in RAW264.7 and THP- 1 cells. (B) Immunoblotting of NRF2 protein in RAW264.7 cells with or without LPS stimulation and indicated concentration of proxalutamide. GAPDH serves as a loading control. (C) Immunoblotting of NRF2 protein in THP- 1 cells with TNFα, IFNγ, or combination, with or without 20 µM proxalutamide. Vinculin serves as a loading control. (D) Real- time analysis of cell death in THP- 1 cells in vitro treated with the indicated cytokines. P values were calculated by the two- tailed unpaired t test between the indicated groups. (E) Immunoblotting of c- PARP and GAPDH in THP- 1 cells after treatment with proxalutamide with or without TNFα and IFNγ combination for 72 h. (F) Schematic illustration of acute immune response in poly (I:C)- induced acute lung injury animal model. (G) Total cell number and neutrophil cell counts in the bronchoalveolar lavage fluid (BALF) under indicated treatment. P values were calculated by the two- tailed unpaired t test between the poly (I:C)- vehicle and indicated treatment. examined proxalutamide in an acute lung injury animal model trig- gered by poly(I:C), and combination dexamethasone and roflumilast treatment was used as a positive control (Fig. 5F). In this model, proxalutamide significantly reduced the total mononuclear cells and neutrophils in alveolar lavage fluids from poly(I:C)- induced animals (Fig. 5G). Together, our data show that proxalutamide up- regulates NRF2 protein levels and decreases inflammation in the lungs induced by poly(I:C), suggesting a possible benefit of proxalutamide against SARS- CoV- 2- associated inflammatory responses and mortality in COVID- 19 patients. Discussion Proxalutamide was initially developed as an AR antagonist that could potentially have efficacy in CRPC patients, including those that had developed resistance to existing AR- targeted therapies. PNAS 2023 Vol. 120 No. 30 e2221809120 https://doi.org/10.1073/pnas.2221809120 7 of 10 Results from phase 1 testing in CRPC patients showed that prox- alutamide was well tolerated, had a favorable pharmacokinetic profile, and exhibited antitumor activity in select patients (47). AR- targeting compounds became one of the initial groups of drugs to be pursued as potential COVID- 19 treatments for the myriad of reasons discussed in preceding sections. With phase 1 testing complete, proxalutamide was positioned to be tested in the setting of COVID- 19, along with other AR- targeted drugs that have been FDA- approved for prostate cancer for years, such as enzalutamide. Although positive results were reported for the initial COVID- 19 trials with proxalutamide, clarity is still needed as one of the stud- ies was retracted last year (41–44). Here, we performed several in vitro and in vivo assays assessing the activity of proxalutamide against SARS- CoV- 2 infection and inflammatory responses. We indeed demonstrate that proxalutamide decreases SARS- CoV- 2 infectivity in vitro, and the compound is active against several strains of the virus (WA1, alpha, delta, and omicron). Synergy can be obtained when proxalutamide is combined with remdesivir. Interestingly, proxalutamide also increases levels of the NRF2 transcription factor. It is well established that COVID- 19 can be associated with a cytokine storm, a hyperactivation of the immune system that can ultimately result in death (53). In this study, we employed two in vivo lines of experimentation to analyze the effect of proxaluta- mide on CSS and lung injury. Proxalutamide pretreatment in the TNFα/IFNγ model of CSS (50) results in a modest increase in overall survival (SI Appendix, Fig. S2 B and C), mirroring the atten- uation of in vitro cell death observed with proxalutamide in the H1437 and THP- 1 cell lines (Figs. 4B and 5D). Using poly(I:C) that induces inflammatory responses in the lung similar to viral infections (61), we observe that proxalutamide significantly decreases total cell and neutrophil levels in BALF (bronchoalveolar lavage fluid) (Fig. 5G). Altogether, results from these two in vivo models suggest that proxalutamide can decrease CSS responses and lung inflammation, but there are associated caveats to note. TNFα and IFNγ induce PANoptosis in mice that leads to CSS and death, which has been suggested to mimic severe COVID- 19 in patients (50). However, TNFα/IFNγ- induced death in mice occurs within hours, whereas death from acute respiratory distress syndrome (ARDS) in COVID- 19 patients happens over a much longer time (62). Additionally, studies have implicated alternative cytokines (e.g., IL- 6 and IL- 1) rather than just TNFα and IFNγ as the primary inducers of ARDS in COVID- 19 (63). In terms of the poly(I:C) model, it is prudent to also note that this is a model of lung injury, rather than lung epithelial cell death. Finally, these in vivo experiments are mod- els of the possible downstream effects of SARS- CoV- 2 and did not directly involve animal infection with the virus. It is interesting to note, however, that proxalutamide increases the DNA binding activ- ity and expression of Nrf2, and Nrf2 has been shown to be an essential factor for tempering the immune response and protecting against sepsis (64, 65). A recent study also shows that SARS- CoV- 2 can inhibit Nrf2 signaling through one of its nonstructural proteins (66). In line with our findings, Nrf2 agonists consequently inhibited SARS- CoV- 2 replication (66). Combined, the data in this study support the notion that proxal- utamide has antiviral activity against SARS- CoV- 2 and suggest that it could show positive clinical benefit in cases of COVID- 19, war- ranting further clinical exploration. In comparison, as mentioned above, clinical studies with degarelix (HITCH trial, NCT04397718) and enzalutamide (COVIDENZA trial, NCT04475601) did not find any improvements in clinical outcome with COVID- 19 (39, 40). There are a multitude of explanations that could account for these disparate findings from different AR- targeting drugs. Degarelix is a GnRH antagonist that prevents release of follicle- stimulating hormone and luteinizing hormone, thereby leading to suppression of testicular testosterone release and a decrease in AR activity at the level of ligand availability (67). In contrast, proxalutamide, like enzaluta- mide, binds directly to the ligand- binding domain of AR to block receptor activation (47, 68). As shown in Fig. 1, proxalutamide and enzalutamide exert similar effects in LNCaP prostate cancer cells—decreasing or activating similar signaling pathways, decreas- ing androgen signaling, and decreasing cell proliferation. Relevant to SARS- CoV- 2, both compounds decrease expression of host entry receptors ACE2 and TMPRSS2 (Fig. 1F). However, certain differences exist with these two compounds. For instance, a pre- clinical report on proxalutamide reported a 3.4- fold higher bind- ing affinity for AR compared to enzalutamide (47). As shown here and previously (47), proxalutamide can also decrease AR protein expression, while enzalutamide does not lead to AR deg- radation (Fig. 1E). In the SARS- CoV- 2 bioassays, proxalutamide exhibited increased potency in inhibiting infection compared to enzalutamide (IC50 of 97 nM for proxalutamide and 281 nM for enzalutamide, Fig. 2B) and a higher Bliss synergy score with remdesivir (14.516 and 11.685 for proxalutamide and enzalut- amide, respectively, Fig. 3). Furthermore, in the cell line models of cytokine- mediated death with combined TNFα and IFNγ treatment, addition of proxalutamide prevented cell death (Fig. 4B), whereas enzalutamide was without effect, even at the high dose of 20 µM (SI Appendix, Fig. S1A). These data show that although proxalutamide and enzalutamide are both AR antagonists, differences in their mechanisms of action exist. However, since both compounds decrease ACE2 and TMPRSS2 expression and ultimately prevent SARS- CoV- 2 infectivity in vitro (albeit with different IC50 values), further research is needed to define the precise mechanisms that could account for disparate clinical outcomes in COVID- 19 treatment. Several phase 3 clinical trials of proxalutamide treatment for COVID- 19, all sponsored by Kintor Pharmaceuticals, are ongoing in different countries, and these studies should provide more definitive answers as to its efficacy. One phase 3 randomized, placebo- controlled, multiregional clinical trial of outpatients with mild or moderate COVID- 19 (NCT04870606) primarily enrolled patients at centers across the United States (99%) (69). Efficacy data showed that prox- alutamide reduced the risk of hospitalization or death compared to placebo, and proxalutamide continued to show a positive safety profile (69). An additional outpatient clinical trial of males with mild to moderate COVID- 19 in Brazil is ongoing (NCT04869228), with the primary outcome being oxygen requirement at Day 28. Finally, NCT05009732 is an ongoing phase 3 trial of proxaluta- mide in hospitalized adults with COVID- 19 that has participating locations across several countries, including the United States, China, Philippines, and South Africa. The primary end point for this study is time to clinical deterioration (need for ICU care, mechanical ventilation, or mortality). The data presented in our report suggest that proxalutamide can markedly decrease SARS- CoV- 2 infectivity and associated inflammatory responses, which could result in positive clinical benefit, and results from the clinical studies above are eagerly awaited. Methods Cell Culture. LNCaP, RAW264.7, and THP- 1 cells were purchased from the American Type Culture Collection (ATCC) and cultured in 5% CO2 at 37 °C in medium as suggested by ATCC. iAEC2 cells [iPSC (SPC2 iPSC line, clone SPC2- ST- B2, Boston University) derived alveolar epithelial type 2 cells] were maintained as previously described (52). iAEC2 cells were also subcultured as previously described (70). Cell lines underwent genotype authentication and were confirmed to be negative for mycoplasma. 8 of 10 https://doi.org/10.1073/pnas.2221809120 pnas.org SARS- CoV- 2 Bioassay. SARS- CoV- 2 isolates USA- WA1/2020, hCoV- 19/USA/OR- OHSU- PHL00037/2021 (Lineage B.1.1.7; Alpha Variant), hCoV- 19/USA/MD- HP05285/2021 (Lineage B.1.617.2; Delta Variant), and hCoV- 19/USA/GA- EHC- 2811C/2021 (Lineage B.1.1.529; Omicron Variant) were obtained from BEI resources and propagated in VeroE6 cells (ATCC). Viral titers were established by TCID50 with the Reed and Muench method. LNCaP or iACE2 cells were plated in 384- well plates and treated with increas- ing concentrations of proxalutamide or enzalutamide for 24 h prior to SARS- CoV- 2 virus infection in a Biosafety Level 3 facility. Cells were then incubated for 48 h postinfection under culture conditions of 5% CO2 and 37°C. Assay plates were fixed, permeabilized, and labeled with antinucleocapsid SARS- CoV- 2 primary antibody (Antibodies Online, Cat. #: ABIN6952432) as previously described (52). The remaining of the assay pro- ceeded as previously described (70). Fluorescence Imaging and High- Content Analysis. A Thermo- Fisher CX5 high- content microscope with LED excitation (386/23 nm, 650/13 nm) at 10× magnification was used to image assay plates. Nine fields per well were imaged at a single Z- plane in these experiments. Imaging, processing, and normalization were performed as previously described (70, 71). Gene Expression Analysis. RNA was extracted from LNCaP cells treated with DMSO, 20 µM proxalutamide, or enzalutamide for 8 h using a Qiagen RNA extrac- tion kit. RNA quality was determined using a Bioanalyzer RNA Nano Chip. Poly- A selection was performed with Sera- Mag Oligo(dT)- Coated Magnetic Particles (38152103010150; GE Healthcare Life Sciences), and libraries were generated using a KAPA RNA HyperPrep kit (KK8541; Roche Sequencing Solutions). RNA- seq was performed on an Illumina HiSeq 2500. Reads were aligned with the Spliced Transcripts Alignment to a Reference mapper to the human reference genome gh38. Gene differential expression analysis was carried out with edgeR70. Mouse Prostate Organoid Culture. Whole mouse prostate was dissected from C57BL6J wild- type mice, and organoid culture was generated according to pre- vious publication (72). Mouse prostate organoids were treated with 5 µM or 10 µM proxalutamide or enzalutamide for 16 h prior to 10 nM DHT stimulation for 8 h. Total RNA was extracted from organoid culture using the miRNeasy mini kit (Qiagen), and cDNA was synthesized from 1 µg total RNA using the High- Capacity cDNA Reverse Transcription Kit (Applied Biosystems). qPCR was performed using fast SYBR green master mix on the QuantStudio Real- Time PCR Systems (Applied Biosystems). The SYBR green primer sequences are Fkbp5 forward: GATTGCCGAGATGTGGTGTTCG, Fkbp5 reverse: GGCTTCTCCAAAACCATAGCGTG; Psca for- ward: GCACAGTTGCTTTACATCGCGC, Psca reverse: ACAGGTCAGAGTAGCAGCACGT; and Ar forward: CCTTGGATGGAGAACTACTCCG, Ar reverse: TCCGTAGTGACAGCCAGAAGCT. Immunoblotting. For western blotting analysis, cells were harvested and lysed in Pierce RIPA buffer (Thermo Fisher) with added phosphatase (Millipore) and pro- tease (Roche) inhibitor cocktails. Protein quantification, sodium dodecyl- sulfate polyacrylamide gel electrophoresis, transfer, blocking, and antibody incubation were performed as described previously (73), and protein signals were detected with ECL Primer (Amersham) on a Li- Cor machine. Antibodies were used at dilu- tions recommended by the manufacturer and consisted of the following: AR (06- 680, Millipore), PSA (Dako), NRF2 (12721S, Cell Signaling Technology), and GAPDH (3683S, Cell Signaling Technology). Real- Time Imaging for Cell Death. The kinetics of cell death were determined using the IncuCyte ZOOM (Essen BioScience) live- cell automated system. H1437 or THP- 1 cells (1 × 105 cells/well) were seeded in 24- well tissue culture plates. Cells were treated with 50 ng/mL of human TNFα (Peprotech, AF- 300- 01A) and /or 100 ng/mL of human IFNγ (Peprotech, 300- 02) for the indicated time and stained with 1 µg/mL propidium iodide (PI) (Life Technologies, P3566) following the manufac- turer’s protocol. The plate was scanned, and fluorescent and phase- contrast images were acquired in real- time every 4 h. PI- positive dead cells are marked with a red mask for visualization. The image analysis, masking, and quantification of dead cells were done using the software package supplied with the IncuCyte imager. In Vivo TNFα and IFNγ- Induced Inflammatory Shock. C57BL6J mice were pur- chased from The Jackson Laboratory. Eight- to nine- week- old male C57BL6J mice were given vehicle or 40 mg/kg proxalutamide by oral gavage either 2 h or once daily for 5 d prior to cytokine injection. Cytokine combination of 10 μg TNFα (Preprotech, 315- 01A) and 20 μg IFNγ (Preprotech, 315- 05) was diluted in Dulbecco’s phosphate- buffered saline (PBS) and injected intraperitoneally. After cytokine injection, animals were under permanent observation, and survival was assessed every 30 min. Poly(I:C)- Induced Acute Lung Injury In Vivo Model. Six- to eight- week- old male BALB/c (Bagg Albino/c) mice were assigned to treatment groups by ran- domization in BioBook software to achieve similar group mean weight before treatment; 10 mice were allocated into each group. Group 1 was normal- vehicle; groups 2 to 5 were challenged with poly(I:C) with vehicle sodium carboxymethly cellulose (CMC- Na), 10 mg/kg dexamethasone and 20 mg/kg roflumilast com- bination, 20 mg/kg proxalutamide, or 40 mg/kg proxalutamide, respectively. Dexamethasone was dissolved in 0.5% CMC- Na to make a suspension at a final concentration of 1 mg/mL. Roflumilast was dissolved in 0.5% CMC- Na to make a suspension at a final concentration of 2 mg/mL. Mice were treated with vehicle, dexamethasone and roflumilast combination, or proxalutamide 16 h and 1 h prior to poly(I:C) injection and 6 h after poly(I:C) injection. Additional proxalutamide dose was given 18 h post poly(I:C) injection. Poly(I:C) solution was prepared to a 0.06% solution in sterile PBS freshly prepared where 1.8 mg poly(I:C) was dissolved in 3 mL PBS to make a suspension at a final concentration of 0.6 mg/ mL. Twenty- four hours post poly(I:C) injection, all mice were anesthetized with Zoletil (i.p., 25 to 50 mg/kg, containing 1 mg/mL Xylazine). Lungs were gently lavaged via the tracheal cannula with 0.5 mL PBS containing 1% fetal bovine serum (FBS), and the BALF was collected. Then, the lungs were gently lavaged with another 0.5 mL PBS containing 1% FBS. After lavage, the collected BALF was stored on ice. The total cell number in BALF was counted using a hemocytometer. After lavage by PBS, all mice were killed by exsanguination. Liquid Mass Spectrometry Quantification after TFRE (Transcription Factors Response Element) Enrichment. Mouse monocyte RAW264.7 cells (0, 2 h, 4 h, and 8 h) and human monocyte THP- 1 (0, 0.5 h, 2 h, and 6 h) were treated with 10 μM proxalutamide, respectively. Cells were collected and cocul- tured with TFRE- binding beads, and the beads were rotated and combined for 1.5 h at 4°C. After the combined TFRE beads were washed 3 times with NETN and 2 times with mass spectrometry (to remove the scale removing agent; if there were still bubbles, they were washed again with water). Then, 50 μL NH4HCO3 and 1.5 μg tyrosinase were added to the beads. The beads were hydrolyzed overnight, and the tube wall was lightly spritzed 1 to 2 times in the middle. Two hundred microliters of 50% acetonitrile + 0.1% formic acid was added to the suspension for 3 to 5 min, and then, the supernatant was transferred on a magnetic rack to a new Eppendorf tube; this was then repeated once. The supernatant was vacuum dried into peptide powder and stored at low temperature. Protein sequences were identified by liquid chromatography with tandem mass spectrometry. Statistical Analysis. Statistical analyses were performed by the two- tailed, unpaired t test, unless otherwise indicated in figure captions. Error bars indicate mean ± SEM. GraphPad Prism software (version 9) was used for statistical calcu- lations. No data were excluded from the analyses. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. Sequencing data are available through the National Center for Biotechnology Information Gene Expression Omnibus, accession num- ber GSE234805 (74). ACKNOWLEDGMENTS. All strains of SARS- CoV- 2 virus were obtained through the Biodefense and Emerging Infections Resources Repository of the National Institute of Allergy and Infectious Diseases that were deposited by the Centers for Disease Control and Prevention. J.Z.S. is supported by the National Institute of Diabetes and Kidney Diseases (R01DK120623). J.W.W. is supported by an American Foundation for Pharmaceutical Education regional award. A.M.C. is a Howard Hughes Medical Institute Investigator, A. Alfred Taubman Scholar, and American Cancer Society Professor. Author affiliations: aMichigan Center for Translational Pathology, University of Michigan, Ann Arbor, MI 48109; bDepartment of Pathology, University of Michigan, Ann Arbor, MI 48109; cRogel Cancer Center, University of Michigan, Ann Arbor, MI 48109; dDepartment of Medicinal Chemistry, College of Pharmacy, University of Michigan, Ann Arbor, MI 48109; eDepartment of Internal Medicine, University of Michigan, Ann Arbor, MI 48109; fState Key Laboratory of Proteomics, Beijing Proteome Research Center, National Center for Protein Sciences (Beijing), Beijing Institute of Lifeomics, Beijing 102206, China; gKintor Pharmaceutical Limited, Suzhou Industrial Park, Suzhuo 215123, China; hCenter for Drug Repurposing, University of Michigan, Ann Arbor, MI 48109; iMichigan Institute for Clinical and Health Research, University of Michigan, Ann Arbor, MI 48109; jDepartment of Pharmacology, University of Michigan, Ann Arbor, MI 48109; kHHMI, University of Michigan, Ann Arbor, MI 48109; and lDepartment of Urology, University of Michigan, Ann Arbor, MI 48109 PNAS 2023 Vol. 120 No. 30 e2221809120 https://doi.org/10.1073/pnas.2221809120 9 of 10 1. 2. 3. 4. 5. 6. 7. K. Koelle, M. A. Martin, R. Antia, B. Lopman, N. E. Dean, The changing epidemiology of SARS- CoV- 2. Science 375, 1116–1121 (2022). F. P. Polack et al., Safety and efficacy of the BNT162b2 mRNA Covid- 19 vaccine. N. Engl. J. Med. 383, 2603–2615 (2020). L. R. Baden et al., Efficacy and safety of the mRNA- 1273 SARS- CoV- 2 vaccine. N. Engl. J. Med. 384, 403–416 (2020). J. Sadoff et al., Safety and efficacy of single- dose Ad26.COV2.S vaccine against Covid- 19. N. Engl. J. Med. 384, 2187–2201 (2021). D. H. Barouch, Covid- 19 vaccines—Immunity, variants, boosters. N. Engl. J. Med. 387, 1011–1020 (2022). D. M. Altmann, R. J. Boyton, COVID- 19 vaccination: The road ahead. Science 375, 1127–1132 (2022). E. G. Levin et al., Waning immune humoral response to BNT162b2 Covid- 19 vaccine over 6 months. N. Engl. J. Med. 385, e84 (2021). 40. K. Welén et al., A phase 2 trial of the effect of antiandrogen therapy on COVID- 19 outcome: No evidence of benefit, supported by epidemiology and in vitro data. Eur. Urol. 81, 285–293 (2022). 41. F. A. Cadegiani et al., Proxalutamide significantly accelerates viral clearance and reduces time to clinical remission in patients with mild to moderate COVID- 19: Results from a randomized, double- blinded, placebo- controlled trial. Cureus 13, e13492 (2021). 42. F. A. Cadegiani et al., Final results of a randomized, placebo- controlled, two- arm, parallel clinical trial of proxalutamide for hospitalized COVID- 19 patients: A multiregional, joint analysis of the proxa- rescue androCoV trial. Cureus 13, e20691 (2021). 43. J. McCoy et al., Proxalutamide reduces the rate of hospitalization for COVID- 19 male outpatients: A randomized double- blinded placebo- controlled trial. Front. Med. (Lausanne) 8, 668698 (2021). 44. Frontiers Editorial Office, Retraction: Proxalutamide reduces the rate of hospitalization for COVID- 19 male outpatients: A randomized double- blinded placebo- controlled trial. Front. Med. (Lausanne) 9, 964099 (2022). 45. Y. Gu et al., Novel strategy of proxalutamide for the treatment of prostate cancer through 8. H. Chemaitelly et al., Waning of BNT162b2 vaccine protection against SARS- CoV- 2 infection in coordinated blockade of lipogenesis and androgen receptor axis. Int. J. Mol. Sci. 22, 13222 (2021). Qatar. N. Engl. J. Med. 385, e83 (2021). 9. H. Chemaitelly et al., Immune imprinting and protection against repeat reinfection with SARS- CoV- 2. N. Engl. J. Med. 387, 1716–1718 (2022), 10.1056/NEJMc2211055. 46. F. Qu et al., Metabolomic profiling to evaluate the efficacy of proxalutamide, a novel androgen receptor antagonist, in prostate cancer cells. Invest. New Drugs. 38, 1292–1302 (2020). 47. T. Zhou et al., Preclinical profile and phase I clinical trial of a novel androgen receptor antagonist 10. S. Chalkias et al., A bivalent omicron- containing booster vaccine against Covid- 19. N. Engl. J. Med. GT0918 in castration- resistant prostate cancer. Eur. J. Cancer 134, 29–40 (2020). 387, 1279–1291 (2022). 48. J. H. Beigel et al., Remdesivir for the treatment of Covid- 19—Final report. N. Engl. J. Med. 383, 11. J. M. Coulson, A. Adams, L. A. Gray, A. Evans, COVID- 19 "Rebound" associated with nirmatrelvir/ 1813–1826 (2020). ritonavir pre- hospital therapy. J. Infect. 85, 436–480 (2022). 49. R. L. Gottlieb et al., Early remdesivir to prevent progression to severe Covid- 19 in outpatients. N. 12. J. Hammond et al., Oral nirmatrelvir for high- risk, nonhospitalized adults with Covid- 19. N. Engl. J. Engl. J. Med. 386, 305–315 (2021). Med. 386, 1397–1408 (2022). 13. F. Kabinger et al., Mechanism of molnupiravir- induced SARS- CoV- 2 mutagenesis. Nat. Struct. Mol. Biol. 28, 740–746 (2021). 50. R. Karki et al., Synergism of TNF- α and IFN- γ triggers inflammatory cell death, tissue damage, and mortality in SARS- CoV- 2 infection and cytokine shock syndromes. Cell 184, 149–168.e17 (2021). 51. H. I. Scher et al., Increased survival with enzalutamide in prostate cancer after chemotherapy. N. 14. H. Ledford, Hundreds of COVID trials could provide a deluge of new drugs. Nature 603, 25–27 Engl. J. Med. 367, 1187–1197 (2012). (2022). 52. C. Mirabelli et al., Morphological cell profiling of SARS- CoV- 2 infection identifies drug repurposing 15. C. B. Jackson, M. Farzan, B. Chen, H. Choe, Mechanisms of SARS- CoV- 2 entry into cells. Nat. Rev. Mol. candidates for COVID- 19. Proc. Natl. Acad. Sci. U.S.A. 118, e2105815118 (2021). Cell Biol. 23, 3–20 (2022). 53. L. Yang et al., The signal pathways and treatment of cytokine storm in COVID- 19. Signal Transduct. 16. J. Lan et al., Structure of the SARS- CoV- 2 spike receptor- binding domain bound to the ACE2 receptor. Targeted Ther. 6, 255 (2021). Nature 581, 215–220 (2020). 17. J. Shang et al., Structural basis of receptor recognition by SARS- CoV- 2. Nature 581, 221–224 (2020). 18. M. Hoffmann et al., SARS- CoV- 2 cell entry depends on ACE2 and TMPRSS2 and is blocked by a 54. R. K. S. Malireddi et al., Inflammatory cell death, PANoptosis, mediated by cytokines in diverse cancer lineages inhibits tumor growth. Immunohorizons 5, 568–580 (2021). 55. Y. Hirotsu et al., Nrf2- MafG heterodimers contribute globally to antioxidant and metabolic networks. clinically proven protease inhibitor. Cell 181, 271–280 (2020). Nucleic Acids Res. 40, 10228–10239 (2012). 19. B. Meng et al., Altered TMPRSS2 usage by SARS- CoV- 2 Omicron impacts infectivity and fusogenicity. Nature 603, 706–714 (2022). 20. J. D. Gunst et al., Efficacy of the TMPRSS2 inhibitor camostat mesilate in patients hospitalized with Covid- 19- a double- blind randomized controlled trial. EClinicalMedicine 35, 100849 (2021). 21. R. H. Shoemaker et al., Development of an aerosol intervention for COVID- 19 disease: Tolerability of soluble ACE2 (APN01) administered via nebulizer. PLoS One 17, e0271066 (2022). 22. E. Tobback et al., Efficacy and safety of camostat mesylate in early COVID- 19 disease in an ambulatory setting: A randomized placebo- controlled phase II trial. Int. J. Infect. Dis. 122, 628–635 (2022). 56. X. L. Chen et al., Activation of Nrf2/ARE pathway protects endothelial cells from oxidant injury and inhibits inflammatory gene expression. Am. J. Physiol. Heart Circ. Physiol. 290, H1862–H1870 (2006). 57. S. Ruiz, P. E. Pergola, R. A. Zager, N. D. Vaziri, Targeting the transcription factor Nrf2 to ameliorate oxidative stress and inflammation in chronic kidney disease. Kidney Int. 83, 1029–1041 (2013). 58. E. Singh et al., Management of COVID- 19- induced cytokine storm by Keap1- Nrf2 system: A review. Inflammopharmacology 29, 1347–1355 (2021). 59. T. O. Khor et al., Nrf2- deficient mice have an increased susceptibility to dextran sulfate sodium- induced colitis. Cancer Res. 66, 11580–11584 (2006). 23. L. Wettstein, F. Kirchhoff, J. Münch, The transmembrane protease TMPRSS2 as a therapeutic target 60. E. H. Kobayashi et al., Nrf2 suppresses macrophage inflammatory response by blocking for COVID- 19 treatment. Int. J. Mol. Sci. 23, 1351 (2022). proinflammatory cytokine transcription. Nat. Commun. 7, 11624 (2016). 24. T. Shapira et al., A TMPRSS2 inhibitor acts as a pan- SARS- CoV- 2 prophylactic and therapeutic. Nature 61. N. C. Stowell et al., Long- term activation of TLR3 by Poly(I:C) induces inflammation and impairs lung 605, 340–348 (2022). function in mice. Respir. Res. 10, 43 (2009). 25. M. Mahoney et al., A novel class of TMPRSS2 inhibitors potently block SARS- CoV- 2 and MERS- CoV viral entry and protect human epithelial lung cells. Proc. Natl. Acad. Sci. U.S.A. 118, e2108728118 (2021). 62. C. Wu et al., Risk factors associated with acute respiratory distress syndrome and death in patients with Coronavirus disease 2019 pneumonia in Wuhan, China. JAMA Intern. Med. 180, 934–943 (2020). 26. K. H. Stopsack, L. A. Mucci, E. S. Antonarakis, P. S. Nelson, P. W. Kantoff, TMPRSS2 and COVID- 19: 63. S. Montazersaheb et al., COVID- 19 infection: An overview on cytokine storm and related Serendipity or opportunity for intervention? Cancer Discov. 10, 779–782 (2020). interventions. Virol. J. 19, 92 (2022). 27. J. M. Jin et al., Gender differences in patients with COVID- 19: Focus on severity and mortality. Front. Public Health 8, 152 (2020). 28. H. Peckham et al., Male sex identified by global COVID- 19 meta- analysis as a risk factor for death and ITU admission. Nat. Commun. 11, 6317 (2020). 64. Y. Wang et al., A small- molecule inhibitor of Keap1- Nrf2 interaction attenuates sepsis by selectively augmenting the antibacterial defence of macrophages at infection sites. EBioMedicine 90, 104480 (2023). 65. R. K. Thimmulappa et al., Nrf2 is a critical regulator of the innate immune response and survival 29. D. K. Twitchell et al., Examining male predominance of severe COVID- 19 outcomes: A systematic during experimental sepsis. J. Clin. Invest. 116, 984–995 (2006). review. Androg. Clin. Res. Ther. 3, 41–53 (2022). 30. M. Montopoli et al., Androgen- deprivation therapies for prostate cancer and risk of infection by SARS- CoV- 2: A population- based study (N = 4532). Ann. Oncol. 31, 1040–1045 (2020). 31. A. Goren et al., Anti- androgens may protect against severe COVID- 19 outcomes: Results from a prospective cohort study of 77 hospitalized men. J. Eur. Acad. Dermatol. Venereol. 35, e13–e15 (2021). 32. V. G. Patel et al., Does androgen deprivation therapy protect against severe complications from COVID- 19?. Ann. Oncol. 31, 1419–1420 (2020). 33. E. A. Klein et al., Androgen deprivation therapy in men with prostate cancer does not affect risk of infection with SARS- CoV- 2. J. Urol. 205, 441–443 (2021). 34. A. L. Schmidt et al., Association between androgen deprivation therapy and mortality among patients with prostate cancer and COVID- 19. JAMA Netw. Open 4, e2134330 (2021). 66. S. Zhang, J. Wang, L. Wang, S. Aliyari, G. Cheng, SARS- CoV- 2 virus NSP14 Impairs NRF2/HMOX1 activation by targeting Sirtuin 1. Cell Mol. Immunol. 19, 872–882 (2022). 67. N. D. Shore, Experience with degarelix in the treatment of prostate cancer. Ther. Adv. Urol. 5, 11–24 (2013). 68. C. Tran et al., Development of a second- generation antiandrogen for treatment of advanced prostate cancer. Science 324, 787–790 (2009). 69. Cision PR Newswire, Kintor Pharma’s Proxalutamide demonstrated reduction in hospitalization/ mortality for patients with mild to moderate COVID- 19 in phase III MRCT study. https://www. prnewswire.com/news- releases/kintor- pharmas- proxalutamide- demonstrated- reduction- in- hospitalizationmortality- for- patients- with- mild- to- moderate- covid- 19- in- phase- iii- mrct- study- 301518525.html (2022). 70. Y. Qiao et al., Antisense oligonucleotides to therapeutically target SARS- CoV- 2 infection. PLoS One 35. N. J. Shah et al., The impact of androgen deprivation therapy on COVID- 19 illness in men with 18, e0281281 (2023). prostate cancer. JNCI Cancer Spectr. 6, pkac035 (2022). 71. J. W. Wotring, R. Fursmidt, L. Ward, J. Z. Sexton, Evaluating the in vitro efficacy of bovine lactoferrin 36. D. A. Leach et al., The antiandrogen enzalutamide downregulates TMPRSS2 and reduces cellular products against SARS- CoV- 2 variants of concern. J. Dairy Sci. 105, 2791–2802 (2022). entry of SARS- CoV- 2 in human lung cells. Nat. Commun. 12, 4068 (2021). 72. J. Drost et al., Organoid culture systems for prostate epithelial and cancer tissue. Nat. Protoc. 11, 37. Y. Qiao et al., Targeting transcriptional regulation of SARS- CoV- 2 entry factors ACE2 and TMPRSS2. 347–358 (2016). Proc. Natl. Acad. Sci. U.S.A. 118, e2021450118 (2020). 73. Y. Qiao et al., Autophagy inhibition by targeting PIKfyve potentiates response to immune checkpoint 38. Q. Deng, R. U. Rasool, R. M. Russell, R. Natesan, I. A. Asangani, Targeting androgen regulation of TMPRSS2 and ACE2 as a therapeutic strategy to combat COVID- 19. iScience 24, 102254 (2021). 39. N. G. Nickols et al., Effect of androgen suppression on clinical outcomes in hospitalized men with COVID- 19: The HITCH randomized clinical trial. JAMA Netw. Open 5, e227852 (2022). blockade in prostate cancer. Nat. Cancer 2, 978–993 (2021). 74. Y. Qiao et al., Proxalutamide reduces SARS- CoV- 2 infection and associated inflammatory response. National Center for Biotechnology Information Gene Expression Omnibus. https://www.ncbi.nlm. nih.gov/geo/query/acc.cgi?acc=GSE234805. Deposited 13 June 2023. 10 of 10 https://doi.org/10.1073/pnas.2221809120 pnas.org
10.1073_pnas.2221064120	RESEARCH ARTICLE \| BIOPHYSICS AND COMPUTATIONAL BIOLOGY EVOLUTION OPEN ACCESS Passive endocytosis in model protocells Stephanie J. Zhanga,b,1 and Anna Wangc,d,e,2 , Palapuravan Aneesf,g, Yamuna Krishnanf,g,h , Lauren A. Lowec,d,e , Thomas G. Faii,2 , Jack W. Szostaka,b,g,j,2 , Edited by Gerald Joyce, Salk Institute for Biological Studies, La Jolla, CA; received January 2, 2023; accepted May 10, 2023 Semipermeable membranes are a key feature of all living organisms. While specialized membrane transporters in cells can import otherwise impermeable nutrients, the earliest cells would have lacked a mechanism to import nutrients rapidly under nutrient- rich circumstances. Using both experiments and simulations, we find that a process akin to passive endocytosis can be recreated in model primitive cells. Molecules that are too impermeable to be absorbed can be taken up in a matter of seconds in an endocytic vesicle. The internalized cargo can then be slowly released over hours, into the main lumen or putative cytoplasm. This work demonstrates a way by which primitive life could have broken the symmetry of passive permeation prior to the evolution of protein transporters. vesicles \| membranes \| lipids \| origin of life \| budding In extant life, membranes provide a selective barrier between a cell and its environment, which enables the inheritance of adaptive traits and ultimately leads to Darwinian evolu- tion (1, 2). Life itself may have emerged from self- replicating informational molecules spatially constrained by primitive membranes (1–3). Among various lipid candidates, fatty acids are particularly well suited as the components of primitive membranes owing to their prebiotic relevance (4, 5) and dynamic exchange properties (6, 7). The possible involvement of fatty acid vesicles in the origin of life has been further demonstrated by their ability to grow and divide without complex biochemical machinery (8–10) and to encapsulate RNA templates that are being nonenzymatically copied (11, 12). In nascent life, the absence of protein transporters implies that a protocell enveloped by lipid bilayers would have had to rely on passive diffusion for internalizing nutri- ents (13–15). Diffusive transport across the membrane is symmetric and is driven purely by the concentration gradient of solutes across the membrane. Despite this, the fluxes need not be the same. When the symmetry is broken, for example, by placing a nutrient sink within the cell, or by a changing external environment, there can be a net flux into or out of the protocell. However, under transient nutrient- rich circumstances, the low permeability of primitive membranes, needed to prevent the loss of the encapsulated cargo, also delays the efficient acquisition of nutrients. Hence, a protocell would have to reside within a pool of nutrients for hours to absorb useful levels of polar or charged mole- cules (12, 15). Whether physicochemical stimuli can fuel nondiffusive transport mecha- nisms remains an important question because such mechanisms could break the transmembrane symmetry in a way that could bias the inward nutrient flow, or expedite nutrient import. One type of transport that has both “active” and “passive” modes is endocytosis. In modern biology, the inward budding of lipid membranes can be active such as in receptor- mediated endocytosis, or passive as in fluid- phase endocytosis. Such a higher- order topological transformation has been previously demonstrated in phospholipid membranes via various pathways (16–18). However, the ability of model primitive membranes to endocytose, i.e., internalize cargo from the external milieu, has yet to be definitively shown. Model primitive membranes typically consist of lipids with dynamic properties distinct from those of phospholipids and thus the routes to engender shape changes could poten- tially differ (19). For instance, flip- flop of model primitive membrane lipids rapidly relaxes any curvature stress (20) that might otherwise help drive the shape transformation and help overcome the energy barrier required for a topological change (19). On the contrary, flip- flop may also be useful for enabling the membrane to adopt the extreme configurations required for inward or outward budding. Here, we demonstrate that primitive cell compartments composed of fatty acids can passively endocytose via a purely physicochemical process. A simultaneous reduction in volume and increase in surface area allows larger molecules to be imported via an inward bud into model protocells, mimicking the process of passive endocytosis in complex eukaryotes. This process takes only seconds, similar to modern forms of endocytosis. The Significance In contemporary life, a molecule’s permeability across cell membranes is tightly regulated by protein transporters. How did primitive cells obtain nutrients, prior to the advent of such transporters? Molecules can passively diffuse across membranes, but importing nutrients in this manner would require a primitive cell to reside in a pool of nutrients for hours if not days. If the membrane is too permeable, nutrients would leak out as soon as they enter. Worse still, leaky membranes would lose primordial genetic material. We present a physicochemical method for triggering passive endocytosis in model primitive cells. The import of nutrients into a stable internalized compartment enables the model primitive cell to gradually absorb the nutrients, thereby breaking the symmetry of passive permeation. Author contributions: S.J.Z., L.A.L., T.G.F., J.W.S., and A.W. designed research; S.J.Z., L.A.L., T.G.F., and A.W. performed research; P.A., Y.K., T.G.F., and J.W.S. contributed new reagents/analytic tools; S.J.Z., L.A.L., T.G.F., J.W.S., and A.W. analyzed data; and S.J.Z., L.A.L., P.A., Y.K., T.G.F., J.W.S., and A.W. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1Present address: Department of Pathology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115. 2To whom correspondence may be addressed. Email: [email protected], [email protected], or anna. [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221064120/- /DCSupplemental. Published June 5, 2023. PNAS 2023 Vol. 120 No. 24 e2221064120 https://doi.org/10.1073/pnas.2221064120 1 of 9 protocells continue to retain the internalized solutes even when removed from the nutrient source, and are thereby provided time to absorb the nutrients. Further, we construct a numerical model that captures such out- of- equilibrium shape changes. Taken together, we believe this marks an important step toward under- standing the development of solute internalization and transport in primitive cells. Results and Discussions Fatty Acid Vesicles Undergo Membrane Invagination and Internal Budding. Here, we study the shape transformations and subsequent topological transitions of fatty acid vesicles in response to volume and area perturbations. For a sphere to change shape, it must have an increase in surface area- to- volume ratio. In principle, this can be accomplished via two pathways: by reducing the volume or increasing the surface area. We explore shape changes obtained through combinations of these pathways. We use hyperosmotic shocks to draw water out of the vesicles and reduce the internal volume; the magnitude of the osmotic shock is denoted as a change in concentration of the salt (Na- bicine), ∆CV, from its starting point of 50 mM. We achieve membrane area growth by adding alkaline oleate micelles to a buffered solution of vesicles; the increase in total membrane lipid is denoted as a change in lipid concentration, ∆CA, relative to a starting lipid concentration of 0.5 mM. Evaluations of the vesicle radius before and after micelle addition suggest that the increase in surface area is no more than twofold under the conditions used in this work (SI Appendix). To observe shape changes in real time with fluorescence microscopy, we prepared oleic acid/oleate (C18:1) giant unilamellar vesicles (GUVs) encapsulating water- soluble and membrane- impermeable fluorescent dyes (either 0.5 mM HPTS 8- hydroxypyrene- 1,3,6- trisulfonic acid trisodium salt, or 1 mM calcein blue). These vesicles were then diluted tenfold into buffer that was identical in composition and pH to the original buffer (but lacked the lipids and dyes) to provide good contrast between encapsulated and unencapsulated dyes, enabling visualization of the dye encapsulated within vesicles. We previously reported that an increase in lipid concentration by micelle addition (∆CA = 1 mM) to oleate GUVs results in outward budding, with the GUVs transforming into a series of smaller vesicles in close proximity, and possibly still connected by tethers (Fig. 1A) (9). Addition of a higher concentration of micelles (∆CA > 1 mM) resulted in vesicle division. In terms of a putative geological setting for such an event, fluid flow from pre- cipitation, in pools, or from tectonic or volcanic events that drive hydrothermal processes, might have carried and mixed nutrients, mineral particulates, and lipid components. Such episodic events Fig. 1. Schematics and confocal micrographs of topological transformations of oleic acid/oleate GUVs following addition of oleate micelles or exposure to hyperosmotic shock. (A) Outward budding of GUVs was induced by micelle addition (stimulus B; ∆CA = 1 mM). The resulting smaller, closely spaced vesicles can be seen in the confocal micrographs. This is similar to the supplemental videos shown in our previous work (9). (B) GUVs could be induced to “endocytose” upon a hyperosmotic shock coupled with micelle addition (stimulus A; ∆CV = 100 mM, ∆CA = 1 mM. See also Movie S1). Confocal micrographs reveal that vesicles initially containing 0.05 M Na- bicine and labeled with 0.5 mM HPTS (green fluorescence) have compartments that appeared as dark voids. Scale bars represent 5 μm. 2 of 9 https://doi.org/10.1073/pnas.2221064120 pnas.org from fluid flow–induced geochemical processes might result in fluctuations in osmotic pressures and simultaneous release of nutrients and lipids. Because increasing the surface area and reduc- ing the internal volume would be predicted to increase the surface area- to- volume ratio, we expected that doing both simultaneously, denoted as stimulus A (∆CV = 100 mM, ∆CA = 1 mM), might also cause vesicle division. Instead, a series of different dramatic shape transformations occurred upon a simultaneous introduction of membrane material and an osmotic shock (Fig. 1B, SI Appendix, Fig. S1, and Movies S1–S8; please see Materials and Methods for exact experimental conditions). Specifically, GUVs appeared to elongate, then rapidly transform first into a stomatocyte via invagination and then into a sphere. We confirmed that the topological transformation was to a vesicle- in- vesicle structure rather than a torus by taking sectional image slices along the z- axis and reconstructing the volume in 3D (Movie S2). Further, we performed two distinct experiments to show that the internal vesicles are indeed fully separated from the external membrane. First, we added an aqueous dye, Alexa 594 hydrazide, that is exclusively aqueous and does not bind to mem- branes, to a final concentration of 10 μM. Alexa 594 hydrazide stayed completely outside of the GUV and did not enter the interior of the endocytosed vesicles, which demonstrates that the inner ves- icles are not connected to the outside (SI Appendix, Fig. S2). Second, we added the dye VoltairPM (21) (final concentration: 250 nM) which localizes to the fatty acid membrane by its POPE (1- palmitoyl- 2- oleoyl- sn- glycero- 3- phosphoethanolamine) moiety, but cannot flip- flop across the external membrane or reach any inner compart- ment owing to its conjugation to polar DNA oligonucleotides. The observation that only the GUV outer membrane leaflet became labeled excludes the possibility that the inner vesicle membranes are connected to the outer GUV membrane (SI Appendix, Fig. S3). Therefore, we can conclude that instead of budding outward (stim- ulus B; ∆CV = 0 mM, ∆CA = 1 mM Fig. 1B), the vesicles had bud- ded inward. We hypothesize that the complete budding of an internal vesicle was possible because both the generated lateral ten- sion from the insertion of lipids into the outer leaflet and the cur- vature stress from flip- flop not transporting lipids into the inner leaflet as rapidly as lipids inserting into the outer leaflet were sufficient to overcome the energies required for breaking the neck and completing the budding process (22). The exact mechanism is still uncertain, and should be the subject of future work. Protocells Take Up Cargo Passively Following Membrane Invagination. As previously mentioned, micelle addition by itself leads only to outward budding (8, 9). Interestingly, micelle addition (∆CA = 0.5 mM to 5 mM) accompanied by varying levels of osmotic shock (∆CV = 25 mM to 100 mM) consistently led to inward budding. By exploring this two- dimensional parameter space, we found that the efficiency of internal compartment formation increased with the magnitude of the surface area increase. As more micelles were added (∆CA = 0.5 mM to 5 mM), the fraction of vesicles with internal compartments increased, and this effect was evident across all magnitudes of osmotic shock (∆CV = 25 to 100 mM, Fig. 2. See also Table 1). For example, the population of vesicles with internal compartments increased from approximately 7.3 ± 1.3% (∆CA = 0.5 mM; ∆CV = 100 mM, total vesicles analyzed = 3,840) to approximately 88.5 ± 3.0% (∆CA = 5 mM; ∆CV = 100 mM, total vesicles analyzed = 1,169). We also found that the number of internal compartments could be tuned by varying the magnitude of the osmotic shock. At con- centrations of added micelles between 2.5 mM and 5 mM, increasing the osmotic shock increased both the number of vesicles with compartments and the occurrence of multicompartment vesicles. For example, at an added micelle concentration of 5 mM, the fraction of vesicles with internal compartments increased from approximately 70.1 ± 9.9% (at ∆CV = 25 mM, total vesicles ana- lyzed = 695) to approximately 88.5 ± 3.0% (at ∆CV = 100 mM, total vesicles analyzed = 1,169), while the fraction of multicom- partment vesicles increased threefold from approximately 12.3 ± 3.0% (∆CV = 25 mM, total vesicles analyzed = 695) to approxi- mately 36.2 ± 8.7% (∆CV = 100 mM, total vesicles analyzed = 1,169). Creating two- or multi- compartment vesicles required a higher concentration of added micelles for lower osmotic shocks: A non- negligible yield of two- compartment or multicompartment vesicles required ∆CA = 5 mM at a lower osmotic shock (∆CV = 25 mM), whereas it required only ∆CA = 2.5 mM at a higher osmotic shock Fig. 2. Number of compartments per vesicle as a function of osmotic shock and micelle addition. Conditions are summarized in Table 1. (A) Distributions of compartments per vesicle for osmotic shocks and micelle additions of different magnitudes [∆CV = 25 (i), 50 (ii), 100 (iii) mM Na- bicine, ∆CA = 0.5, 1, 2.5, 5 mM oleate]. The equivalents of oleate in the micellar solution added are defined with respect to the concentration of oleate/oleic acid in the starting vesicle suspension. (B) Representative confocal micrographs of vesicles with (i) single and (ii) multiple internal compartments. Scale bars represent 5 μm. Each condition consisted of at least five replicates, and error bars indicate the SD from the mean. The total number of analyzed vesicles for each condition is indicated in Table 1. PNAS 2023 Vol. 120 No. 24 e2221064120 https://doi.org/10.1073/pnas.2221064120 3 of 9 (∆CV = 100 mM). At lower concentrations of added micelles (∆CA = 0.5 mM to 1 mM), varying the magnitude of osmotic shock (∆CV = 25 mM to 100 mM) did not lead to significant changes in the number of internal compartments formed. Having determined that the inward budding was robust, we then tested whether this pathway allowed for the passage of large polar molecules into protocells prior to the evolution of complex protein machinery. We found that passive endocytosis allowed the long DNA–dye conjugate (5′- Cy5- C(10) A(18)- 3′) (Fig. 3A) to be imported from the external medium into the vesicle itself. Specifically, the Cy5- labeled DNA 28- mers (appearing magenta) were introduced along with the simultaneous osmotic shock and micelle addition (stimulus A) into oleic acid/oleate GUVs con- taining encapsulated calcein blue (appearing blue). The resultant formation of magenta interior compartments indicates that the Cy5- labeled DNA 28- mers were successfully transported to the lumen of the internalized vesicles and thereby internalized by the GUVs (Fig. 3 B, i and ii). The resultant vesicle- in- vesicle structures were not expected to preclude further inward budding events. We therefore tested whether repeated inward budding could be induced by successive rounds of simultaneous osmotic shock and micelle addition (stim- ulus A). Following the intake of Cy5- labeled DNA 28- mers into one set of internalized buds, a second simultaneous osmotic shock and micelle addition event was induced in the presence of the yellow aqueous compound fluorescein- 12- UTP in the exterior solution. This resulted in the UTP being internalized via a second set of buds: The blue vesicles contained both magenta and yellow inner com- partments (Fig. 3 B, iii). With both stimulus events being in the regime where two or more internal compartments are likely (Fig. 2), the resultant vesicles displayed a high internal volume fraction of internalized vesicles. The excess micelles introduced also led to de novo vesicle formation: Magenta vesicles containing yellow inner compartments are also occasionally observed. Some compartments also appear intermediate in color (e.g., orange), suggesting that some mixing between compartments might have occurred. This result is informative for two reasons. First, the presence of inner compartments that are of a different color from the external medium (after two cycles of micelle addition in the presence of different dyes) indicates that the shape transformation of the vesicles does not end in the stage of stomatocytes (SI Appendix, Fig. S1A). Rather, inward vesiculation was completed and a full topological transformation had occurred (SI Appendix, Fig. S1B). Without the final topological transformation, all the endocytic compartments would be of the same color as the external medium. Second, the presence of both yellow and magenta internal compartments is consistent with two separate rounds of endocytosis having taken place. This confirms that the simultaneous osmotic shock and micelle addition event can trigger repeated endocytic events. Release of Cargo from Internal Compartment into the Main Lumen. A substantial uptake of nutrients from the external solution by passive diffusion across the limiting membranes of protocells requires protocells to reside within the pool of Fig. 3. Induction of multiple sequential passive endocytosis events. (A) Schematic for induction of multiple passive endocytosis events induced by successive osmotic shocks coupled with micelle additions. (B) Confocal micrographs of fatty acid vesicles with various numbers and sizes of interior compartments. (i) Initial GUVs with encapsulated calcein blue (blue). (ii) Simultaneous addition of 1 μM 5′- Cy5- C(10) A(18)- 3′, ∆CV = 100 mM Na- bicine, and ∆CA = 2.5 mM oleate resulted in inner compartments containing encapsulated 5′- Cy5- C(10) A(18)- 3′ (magenta) derived from the external solution. (iii) A second round of passive endocytosis stimulated by the simultaneous addition of 5 μM Fluorescein- 12- UTP, ∆CV = 100 mM Na- bicine, and ∆CA = 2.5 mM oleate resulted in new inner compartments containing encapsulated Fluorescein- 12- UTP (yellow) derived from the external solution. Scale bars represent 5 μm. The introduction of multiple heterogeneous stimuli to a uniform starting solution of vesicles ultimately leads to the emergence of population diversity (see also Movie S12). 4 of 9 https://doi.org/10.1073/pnas.2221064120 pnas.org nutrients for hours. By contrast, passive endocytosis results in a net inward flow of cargo within seconds, although the cargo is still contained within a membrane. Releasing nutrients stored inside the endocytic compartments into the main lumen, i.e., putative cytoplasm, would allow them to interact with other components within the protocell and is the final critical step in this pathway for nutrient uptake. For example, this could enable dinucleotides to interact with replicating internal oligonucleotides. To test the ability of the internal endocytic compartment to release nutrients, we first used a relatively impermeable substrate, the fluorescein- labeled cyclic dinucleotide (cGAMPfluo), as a fluorescent tracer. cGAMPfluo was introduced along with a simultaneous osmotic shock and micelle addition (∆CV = 100 mM, ∆CA = 1 mM) to oleic acid/oleate GUVs (Fig. 4A). The vesicles were then diluted fivefold into a 200 mM glucose buffer that was identical in composition and pH to the original buffer but lacked the lipids and cGAMPfluo, to dilute the free cGAMPfluo in the external medium. We then took confocal microscopy images over time to measure the fluorescence inten- sities in the internalized compartments, the main lumen of the GUVs, and the external medium. The fluorescence intensity serves as a measure of the relative concentrations of encapsulated material. Tracking the fluorescence intensity of the internalized compart- ments relative to the external medium (Fig. 4A), which we consider an infinite bath, showed a decrease from 6.2 ± 1.0 to 3.9 ± 0.9 (N = 124) over 24 h. This suggests that the oligomer was slowly released by the internalized compartment, into the main lumen. Indeed, the fluorescence intensity of the internalized compartments relative to the main lumen decreased more than twofold, from the ratio of 36.3 ± 1.7 to 14.3 ± 1.1 over 24 h (Fig. 4B, N = 113). This indicates that the oligonucleotide was slowly released from the inter- nalized compartment into the main lumen. We found similar results for another representative molecule with low permeability to fatty acid membranes (23), HPTS (SI Appendix, Fig. S5). Overall, these results demonstrate that internalized compartments can be used as nutrient storage depots, which would slowly release contents to the main lumen or “cytoplasm” of a protocell. Overall, our studies show that protocells are capable of a primitive form of endocytosis con- sisting of three characteristic steps: membrane invagination and budding, uptake of soluble external cargo into the resultant inter- nalized compartments, and cargo release from the compartments into the main lumen. These constitute the three basic steps of all forms of modern endocytosis. Modeling Out- of- Equilibrium Topological Changes. To understand the reason for invagination, we model the vesicle as an inextensible elastic material with prescribed bending modulus, spontaneous curvature, and permeability. While the model cannot predict whether a full topological transformation can occur, it does enable the prediction of morphological changes preceding a potential topological transition. The surrounding fluid is considered as an aqueous solution that contains amphiphiles that incorporate into the vesicle membrane at a given rate. As shown in the study by Ruiz- Herrero et al. (19), the resulting behavior may be described in terms of two dimensionless parameters that govern the steady- state vesicle morphology. In addition, to capture the effect of osmotic shocks, we include the possibility of osmotic pressures that drive flows across the membrane. This is done by adding an appropriate normal force to the membrane surface corresponding to the van’t Hoff pressure p = ckBT , where c is the difference in osmotic concentrations across the membrane. After applying osmotic shocks in the presence of permeation and membrane growth, we observe that vesicles develop stomat- ocyte morphologies that progress spontaneously to develop inward buds (Fig. 5 A and B). The initial hyperosmotic shock causes the vesicles to shrink in volume while adding additional surface area. This increases the surface area- to- volume ratio over time. We use parameters listed in Table 2 and confirm that the timescales of A B y t i s n e t n i e v i t a e R l Na+ Na+ Na+ cGAMPfluo external medium compartment “cytoplasm” or main lumen 40 30 20 10 0 0 15 10 5 0 5 10 Time (hours) 15 20 25 l R e a t i v e i n t e n s i t y i n n e r c o m p a r t m e n t : e x t e r n a l m e d u m i ” m s a p o l t y c “ : t n e m t r a p m o c r e n n i Fig. 4. Internal compartments can slowly deliver nutrients of low permeability to the main lumen or “cytoplasm” of a protocell. (A) Schematic of fluorescein- labeled cyclic dinucleotide (cGAMPfluo), a relatively impermeable substrate, being encapsulated in the interior compartment via one hyperosmotic shock coupled with micelle addition (∆CV = 100 mM Na- bicine, ∆CA = 2.5 mM oleate, CcGAMPfluo = 0.1 mM). (B) The ratio of fluorescence intensity of the internal compartment vs. the “cytoplasm” decreases with time; similarly, the ratio of fluorescence intensity of the internal compartment vs the external medium decreases with time, indicating that the dinucleotide inside the internal compartment is slowly entering the cytoplasm. PNAS 2023 Vol. 120 No. 24 e2221064120 https://doi.org/10.1073/pnas.2221064120 5 of 9 C A B D E Fig. 5. Modeling of vesicle topological changes. For parameters used, see Table 2. (A) Side view and (B) cutaway of the final morphology of an initially spherical vesicle after 2 s of growth, with spontaneous formation of an incomplete inner bud, yet to be pinched off. (C) Final morphology of an initially spherical vesicle after 3 s of growth, for different values of the growth rate (γ) and osmolarity (Δc = cout − cin). Increasing the osmotic shock results in a higher effective growth rate, thereby reducing the threshold for vesiculation. (D) Comparison between simulations (gray) and experiments (green) showing inward vesiculation for a single compartment and (E) two compartments. Scale bars represent 5 μm. Further examples of inward vesiculation in simulations and experiments can be found in Movies S3–S11. volume loss and area increase are consistent with those of prior work (24). In terms of the morphological phase plane described in the study by Ruiz- Herrero et al. (19), the experimental parameter regime corresponds to invagination. We find that the added effect of osmotic pressures is parallel to a higher effective membrane growth rate and lowers the threshold for vesiculation (Fig. 5C). Our model recapitulates several of the behaviors observed exper- imentally, including the spontaneous onset of inward vesiculation on the correct timescale of seconds. In terms of the resulting vesicle morphology, we show that the resulting effect from a hyperosmotic shock corresponds to that of an increased effective membrane growth rate. This rationale is consistent with the progressively lower vesiculation threshold that is observed in experiments on vesicles subjected to increasing osmotic shocks (Fig. 2). Furthermore, we find that reducing the magnitude of the sponta- neous curvature and the bending modulus can lead to several internal invaginations forming (Fig. 5E and Movies S10 and S11). These results suggest that the addition of both micelles and an osmotic shock may contribute toward decreasing the bending modulus of the mem- brane. Indeed, this is consistent with our recent results showing that either a slight increase in pH or the addition of salt can decrease the bending modulus of fatty acid membranes (9). In general, the time series of modeled geometries (Movies S9–S11) correspond well to experimental video microscopy data showing the endocytosis process (Movies S3–S8 and Materials and Methods). The changes in shape shown in the movies and in the sequences of images illustrated in Fig. 5 D and E take place on the timescale of seconds for both experiments and simulations. The relative timescales of water efflux, which is expected to be rapid, and micelle- driven surface area growth (SI Appendix), which is slower, indicate that water efflux and a modest increase in total area are adequate to drive endocytosis within seconds. Several open questions remain. First, positive spontaneous curvature favors outward budding, whereas negative spontaneous curvature favors inward budding. This suggests that negative spontaneous curvature is important for passive endocytosis to occur, yet the mechanism of how negative spontaneous curvature would arise in our system remains unclear. One possibility is that the bilayer leaflets have an asymmetric interaction with water molecules or ions during simultaneous water efflux and membrane growth. Another hypothesis is that increased external salt bridges the carboxylate headgroups, either decreasing effective membrane area or perhaps changing effective lipid shape to favor negative curvature. Ultimately, while coarse- grained models are effective for predicting shape changes, molecular dynamics simulations of membrane packing might be necessary to reveal the underlying mechanism and cause (25). Second, although here we do explicitly solve for the hydrodynamics of the incompressible surrounding fluid, we have not studied the role of hydrodynamics in controlling the vesiculation behavior in detail. More specifically, systematically varying the mem- brane’s spontaneous curvature and the bending modulus would be an interesting direction for future study. Comparison to Passive Permeability. In light of our experimental demonstration of the release of both dinucleotides and HPTS from the endocytic compartment into the lumen, we also seek to further understand the utility of endocytosis as an inward- transport mechanism and estimate the time that is required for the vesicle’s lumen to obtain various nutrients by passive diffusion across the compartment membrane. Such time scales can be estimated from the net flux (denoted by J , with units of number of molecules crossing unit area per unit time). The spontaneous transport of molecules along their concentration gradient across membranes is described by a simple equation J = Ps ⋅ΔC , [1] where Ps is the permeability coefficient for the solute and ΔC (Cin − Cout) is the concentration difference across the membrane. The number of molecules that cross a given area per unit time can be determined by rearranging Eq. 1 and the definition of flux J = dN dt ⋅ 1 A , into dN = J ⋅dt ⋅A = Ps ⋅ΔC ⋅dt ⋅A. [2] [3] Membrane permeability coefficients ( Ps ) for short oligomers up to the size of trinucleotides crossing oleic acid membranes have been previously measured (12). We consider trinucleotides, which can enhance nonenzymatic primer extension. The permeability of the ≈ 0.21 ⋅ 10−12 oleic acid membrane to an average trinucleotide is Ps 6 of 9 https://doi.org/10.1073/pnas.2221064120 pnas.org Table 1. Summary of different stimuli used to trigger shape changes in vesicles Volume(Na- bicine); volume(micelles) (μL); total number of vesicles analyzed Osmotic shock [Na- bicine] (mM) 25 50 100 Micelles (relative to the fatty acid concentration) 1 equivalent (eqv.) 2 eqv. 2.7; 0.5; 4800 5.56; 0.5; 4395 2.7; 1; 4269 5.56; 1; 5897 5 eqv. 2.7; 2.5; 3786 10 eqv. 2.7; 5; 695 5.56; 2.5; 1730 5.56; 5; 1576 11.76; 0.5; 3840 11.76; 1; 2941 11.76; 2.5; 1697 11.76; 5; 1169 cm/s. In the presence of a 1 mM trinucleotide “pool”, the number of trinucleotides that enter a GUV with a diameter of 4 μm from the external medium can be estimated from Eq. 3, giving 0.06 trinucleotides per second. In other words, it takes approximately 16 s for a single trinucleotide to cross the oleic acid membranes when vesicles are surrounded by 1 mM trinucleotides. We assessed the flux for a range of permeabilities (SI Appendix, Fig. S6) spanning values for common nutrients ( 10−10 and 10−13 cm/s). We found that it takes anywhere between 0.03 and 33 s for a single molecule to cross 4- μm- diameter oleic acid vesicles. It is clear that a protocell would have to reside within such a pool of nutrients for minutes to hours to acquire significant quantities of nutrients by passive diffusion. This result is in stark contrast to passive endocytosis, which trans- ports a parcel of nutrients inwards within seconds. In this scenario, a dramatic release event is not necessary for the nutrients to reach the putative cytoplasm. Instead, the molecules in the inner com- partment can be slowly released into the main lumen of the proto- cell, where they can then interact with other components. For an inner compartment ¼ the diameter of a 4- μm- diameter GUV, 90% of molecules with permeability comparable to trinucleotides can be released into the lumen within 11 h (SI Appendix, Fig. S6A). This indicates that protocells that endocytose can absorb a substantial amount of nutrients from a pool despite interacting with it only briefly. Examples include a surface- immobilized protocell capturing nutrients from intermittently nutrient- rich streams of water, or a protocell capturing the released contents from a nearby burst pro- tocell that would otherwise diffuse away. This principle can be extended further to consider the transport of even larger oligomers. While permeabilities of longer oligomers through oleic acid membranes remain unknown, they are expected to have a lower permeability than trinucleotides. After approxi- mately 4.7 h (SI Appendix, Fig. S6B), 10% of the molecules with a permeability of 10−13 cm/s are released from the interior com- partment, with full release within 5 d. This result points to a remarkable function of passive endocytosis—a means of importing otherwise “membrane- impermeant” molecules. Conclusions We have shown that model primitive cell membranes are capable of invagination and inward vesiculation, leading to a complete top- ological transition to a vesicle- in- vesicle morphology. The number of internal compartments can be controlled by the rate of surface area growth, with increasingly strong osmotic shocks decreasing the rate of surface area growth necessary for vesiculation. We also reca- pitulated the main results and the relevant timescales in an out- of- equilibrium numerical model. We then found that such inward vesiculation events could lead to internalization of nutrient solutes including mononucleotides and oligonucleotides, drawing further parallels to endocytosis. Such processes could have helped primitive cells capture nutrients that are otherwise impermeable and could have also generated population diversity from a uniform starting solution of vesicles (Movie S12). Materials and Methods Materials. Oleic acid (C18:1) was purchased from NuChek Prep. DNA labeled with a fluorescent dye (5′- Cy5- C(10) A(18)- 3′) was synthesized by IDT. All other chemicals were purchased from Sigma- Aldrich and were used without any further purification. Data analysis was performed using ImageJ (version 1.53a), Python, and GraphPad Prism (version 8.4.0). Preparation of Micelles and GUVs. One hundred millimolars of oleate micelles was prepared by dissolving 50 μmol neat oleic acid oil in 1 equivalent of NaOH solution to a volume of 500 μL with Millipore water (18.2 MΩ · cm). Fatty acid GUVs were made by resuspending the micelle solution in buffer stock (1 M Na- bicine, pH 8.45), sucrose stock, and Millipore water to the final concentration of 5 mM oleic acid, 50 mM Na- bicine buffer, and 200 mM sucrose as described previously (9). Encapsulation of fluorescent dyes was achieved by mixing the solute with the resuspension buffer before adding the micelles. Vesicle Endocytosis. A 100 μL aliquot of the GUV suspension was carefully transferred to a 1.7 mL microcentrifuge tube (Fisher Scientific). These vesicles were subsequently diluted 1:9 into a buffer consisting of 50 mM Na- bicine at pH 8.45 and 200 mM glucose to a final oleic acid concentration of 0.5 mM, enabling good contrast of vesicles against the background when imaged under fluorescence microscopy. This diluted vesicle solution was then split into ten 100 μL aliquots. Vesicle endocytosis was initiated by adding micelles from a 100 mM stock solution and Na- bicine buffer from a 1 M stock solution to a 100 μL diluted aliquot, then mixing by inverting the tube for ∼5 s. The added volumes and con- centrations corresponding to each condition are summarized and listed in Table 1. Vesicle suspensions were allowed to equilibrate for at least 1 h before microscopy. Washing Endocytosed Vesicles. To dilute the free (unencapsulated) aqueous dye for imaging after passive endocytosis, the vesicle solution was diluted tenfold into a 200 mM glucose buffer that was identical in composition and pH to the original buffer, but lacked the lipids and dye. After centrifugation at 2,000 g for 30 s, the top 90% of volume was removed by pipetting. The remaining solution was then agitated to resuspend the vesicles. Sequential Endocytosis Experiments. One hundred and fifty microliters of the washed vesicle suspension was transferred into one well of a Nunc Lab- Tek II 8- well chambered coverslip (Thermo Scientific) and the sample allowed to settle for 10 min to form a GUV monolayer at the bottom of the chamber. The endocytosis triggers were then pipetted into the open well in the sequence described in Fig. 3. Specifically, the initial GUVs that encapsulated calcein blue (blue) were pre- pared following the protocol outlined in the “Preparation of Micelles and GUVs” section. After diluting the GUVs 1:9 into a buffer containing 50 mM Na- bicine at pH 8.45 and 200 mM glucose, to a final oleic acid concentration of 0.5 mM, the GUVs were allowed to settle for 10 min. Subsequently, the first endocytosis trigger was added to the well, resulting in a final concentration of 1 μM of 5′- Cy5- C(10) A(18)- 3′ along with ∆CV = 100 mM Na- bicine and ∆CA = 2.5 mM oleate. Inner compartments containing encapsulated 5′- Cy5- C(10) A(18)- 3′ (magenta) derived from the external solution were formed (see also Movie S1). After 1 h, a second round of passive endocytosis was initiated by pipetting in a second stimulus to a final concentration of 5 μM Fluorescein- 12- UTP, ∆CV = 100 mM Na- bicine and ∆CA = 2.5 mM oleate. The changes in Na- bicine and oleate concentration were relative to the end of the first endocytic event; the final Na- bicine concentration was 250 mM and the final oleate concentration was 5.5 mM. This second trig- ger led to the formation of new inner compartments containing encapsulated Fluorescein- 12- UTP (yellow) derived from the external solution. PNAS 2023 Vol. 120 No. 24 e2221064120 https://doi.org/10.1073/pnas.2221064120 7 of 9 Table 2. Parameters used in simulations of growing permeable vesicles under osmotic shock Values used for figures and movies Range tested 5 A and B 5D 5E S9 S10 S11 Initial diameter Water permeability Growth rate Viscosity Density Osmolarity Specific volume Spontaneous curvature Bending modulus Total time Symbol R Pf γ μ ρ Δc vw c0 Kb T Units μm cm/s s- 1 dyn/cm×s g/cm3 mM 1 14e- 3 0 to 2.5 0.01 1 10 to 100 cm3/mol 20 μm kBT s −0.2 to −1 2 to 100 1 to 3 0.02 2.5 0.75 0.1 0.02 0.007 50 −1 10 10 100 10 50 50 −0.66 −0.2 100 10 −1 10 −0.66 −0.4 3 2 images were collected using a Nikon Confocal Microscopy. Confocal A1R HD25 confocal laser scanning microscope equipped with LU- N4/N4S 4- laser unit. Vesicle Endocytosis Movies S3–S8. An aliquot of the GUV suspension encapsulating 1 mM HPTS was diluted 1:9 into a buffer consisting of 50 mM Na- bicine, pH 8.45, and 200 mM glucose to a final oleic acid concentration of 0.5 mM. One hundred and fifty microliters of the diluted suspension was transferred into one well of a Nunc Lab- Tek II 8- well chambered coverslip (Thermo Scientific) and the sample allowed to settle for 10 min to form a GUV monolayer at the bottom of the chamber. Two microliters of a solution containing both 500 mM Na- bicine and 25 mM oleate micelles was then pipetted into the well containing the diluted GUVs. Epifluorescence microscopy was performed using a Nikon TE2000- U inverted microscope. A blue LED was used to excite the sample using a CoolLED pE- 300ultra system, and images were captured using a pco.edge 4.2 sCMOS camera. Immersed Boundary Method Simulations. To simulate growing, permea- ble vesicles under osmotic shock, we used the immersed boundary method to simulate the coupled fluid–structure interaction of permeable vesicles growing in an incompressible aqueous solution (26). Given the complexity of possible changes in vesicle morphology, it was important to perform simulations in three dimensions, and for computational efficiency, we used the method described in the study by Fai et al. (27). For simplicity, we held the inner and outer osmolyte concentrations constant and did not account for solute exchange across the membrane (i.e., we set Ps = 0 in the notation of Sacerdote et al.) (24). Given that the rate of solute exchange is typically much lower than that of solvent, this approximation is expected to be reasonable for the relatively short timescales of interest here. Further details can be found in SI Appendix, Appendix I. Data, Materials, and Software Availability. All study data are included in the article, SI Appendix and/or the OSF repository (https://osf.io/r4zsp/?view_ only=6da09fcb508d4ab2b93c1d13ee406b18) (28). ACKNOWLEDGMENTS. Y.K. acknowledges the support from the Human Frontier of Science Program (RGP0032/2022 to Y.K.), DP1GM149751, 1R01NS112139- 01A1, and Ono Pharma Foundation. T.G.F. acknowledges the support from NSF MCB- 2213583 and DMS- 1913093. J.W.S. is an investigator of the Howard Hughes Medical Institute. This work was funded in part by a grant from the Simons Foundation (290363) to J.W.S. A.W. acknowledges support from the Australian Research Council (DE210100291) and the Human Frontier Science Program (RGP0029/2020 to A.W.). Author affiliations: aDepartment of Chemistry and Chemical Biology, Harvard University, Cambridge, MA 02138; bDepartment of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, MA 02114; cSchool of Chemistry, University of New South Wales Sydney, Bedegal Country, Sydney, NSW 2052, Australia; dAustralian Centre for Astrobiology, University of New South Wales Sydney, Bedegal Country, Sydney, NSW 2052, Australia; eARC Centre of Excellence in Synthetic Biology, University of New South Wales Sydney, Bedegal Country, Sydney, NSW 2052, Australia; fNeuroscience Institute, University of Chicago, Chicago, IL 60637; gDepartment of Chemistry, University of Chicago, Chicago, IL 60637; hInstitute of Biophysical Dynamics, University of Chicago, Chicago, IL 60637; iDepartment of Mathematics, Brandeis University, Waltham, MA 02453; and jHHMI, Massachusetts General Hospital, Boston, MA 02114 1. 2. 3. 4. 5. 6. 7. 8. 9. J. P. Schrum, T. F. Zhu, J. W. Szostak, The origins of cellular life. Cold Spring Harb. Perspect. Biol. 2, a002212 (2010). R. V. Sole, Evolution and self- assembly of protocells. Int. J. Biochem. Cell B 41, 274–284 (2009). A. J. Dzieciol, S. Mann, Designs for life: Protocell models in the laboratory. Chem. Soc. Rev. 41, 79–85 (2012). D. W. Deamer, R. M. Pashley, Amphiphilic components of the Murchison carbonaceous chondrite: Surface properties and membrane formation. Orig. Life Evol. Biosph. 19, 21–38 (1989). D. W. Deamer, G. L. Barchfeld, Encapsulation of macromolecules by lipid vesicles under simulated prebiotic conditions. J. Mol. Evol. 18, 203–206 (1982). S. S. Mansy, Model protocells from single- chain lipids. Int. J. Mol. Sci. 10, 835–843 (2009). A. Wang, J. W. Szostak, Lipid constituents of model protocell membranes. Emerg. Top Life Sci. 3, 537–542 (2019). T. F. Zhu, J. W. Szostak, Coupled growth and division of model protocell membranes. J. Am. Chem. Soc. 131, 5705–5713 (2009). J. T. Kindt, J. W. Szostak, A. Wang, Bulk self- assembly of giant, unilamellar vesicles. ACS Nano 14, 14627–14634 (2020). 10. E. Blöchliger, M. Blocher, P. Walde, P. L. Luisi, Matrix effect in the size distribution of fatty acid vesicles. J. Phys. Chem. B 102, 10383–10390 (1998). 11. K. Adamala, J. W. Szostak, Nonenzymatic template- directed RNA synthesis inside model protocells. Science 342, 1098–1100 (2013). 13. P. G. Barton, F. D. Gunstone, Hydrocarbon chain packing and molecular- motion in phospholipid bilayers formed from unsaturated lecithins. Synthesis and properties of 16 positional isomers of 1,2- dioctadecenoyl- sn- glycero- 3- phosphorylcholine. J. Biol. Chem. 250, 4470–4476 (1975). 14. M. C. Blok, E. C. M. Vanderneutkok, L. L. M. Vandeenen, J. Degier, Effect of chain- length and lipid phase- transitions on selective permeability properties of liposomes. Biochim. Biophys. Acta 406, 187–196 (1975). 15. S. S. Mansy, Membrane transport in primitive cells. Cold Spring Harb. Perspect. Biol. 2, a002188 (2010). 16. W. Zong et al., A fissionable artificial eukaryote- like cell model. J. Am. Chem. Soc. 139, 9955–9960 (2017). 17. R. Lipowsky, The conformation of membranes. Nature 349, 475–481 (1991). 18. P. Peterlin, V. Arrigler, K. Kogej, S. Svetina, P. Walde, Growth and shape transformations of giant phospholipid vesicles upon interaction with an aqueous oleic acid suspension. Chem. Phys. Lipids 159, 67–76 (2009). 19. T. Ruiz- Herrero, T. G. Fai, L. Mahadevan, Dynamics of growth and form in prebiotic vesicles. Phys. Rev. Lett. 123, 038102 (2019). 20. R. J. Bruckner, S. S. Mansy, A. Ricardo, L. Mahadevan, J. W. Szostak, Flip- flop- induced relaxation of bending energy: Implications for membrane remodeling. Biophys. J. 97, 3113–3122 (2009). 21. A. Saminathan et al., A DNA- based voltmeter for organelles. Nat. Nanotechnol. 16, 96–103 (2021). 22. R. Lipowsky, Remodeling of membrane compartments: Some consequences of membrane fluidity. 12. D. K. O’Flaherty et al., Copying of mixed- sequence RNA templates inside model protocells. J. Am. Biol. Chem. 395, 253–274 (2014). Chem. Soc. 140, 5171–5178 (2018). 23. S. M. Fujikawa, I. A. Chen, J. W. Szostak, Shrink- wrap vesicles. Langmuir 21, 12124–12129 (2005). 8 of 9 https://doi.org/10.1073/pnas.2221064120 pnas.org 24. M. G. Sacerdote, J. W. Szostak, Semipermeable lipid bilayers exhibit diastereoselectivity favoring ribose. Proc. Natl. Acad. Sci. U.S.A. 102, 6004–6008 (2005). 25. A. J. Markvoort et al., Vesicle deformation by draining: Geometrical and topological shape changes. J. Phys. Chem. B 113, 8731–8737 (2009). 26. C. H. Wu, T. G. Fai, P. J. Atzberger, C. S. Peskin, Simulation of osmotic swelling by the stochastic immersed boundary method. Siam. J. Sci. Comput. 37, 660–688 (2015). 27. T. G. Fai, B. E. Griffith, Y. Mori, C. S. Peskin, Immersed boundary method for variable viscosity and variable density problems using fast constant- coefficient linear solvers I: Numerical method and results. Siam. J. Sci. Comput. 35, 1132–1161 (2013). 28. S. J. Zhang, A. Wang, Passive endocytosis in model protocells. Open Science Framework. https://osf.io/r4zsp/?view_only=6da09fcb508d4ab2b93c1d13ee406b18. Deposited 23 May 2023. PNAS 2023 Vol. 120 No. 24 e2221064120 https://doi.org/10.1073/pnas.2221064120 9 of 9
10.1073_pnas.2304730120	RESEARCH ARTICLE \| GENETICS OPEN ACCESS split-intein Gal4 provides intersectional genetic labeling that is repressible by Gal80 Ben Ewen-Campena,1 and Norbert Perrimona,f,2 , Haojiang Luanb,1, Jun Xua,c,1, Rohit Singhd,1, Neha Joshia, Tanuj Thakkara, Bonnie Bergerd,e, Benjamin H. Whiteb , Contributed by Norbert Perrimon; received March 22, 2023; accepted May 10, 2023; reviewed by Fillip Port and Christopher J. Potter The split-Gal4 system allows for intersectional genetic labeling of highly specific cell types and tissues in Drosophila. However, the existing split-Gal4 system, unlike the standard Gal4 system, cannot be repressed by Gal80, and therefore cannot be controlled temporally. This lack of temporal control precludes split-Gal4 experiments in which a genetic manipulation must be restricted to specific timepoints. Here, we describe a split-Gal4 system based on a self-excising split-intein, which drives transgene expression as strongly as the current split-Gal4 system and Gal4 reagents, yet which is repressible by Gal80. We demonstrate the potent inducibility of “split-intein Gal4” in vivo using both fluorescent reporters and via reversible tumor induction in the gut. Further, we show that our split-intein Gal4 can be extended to the drug-inducible GeneSwitch system, providing an independent method for intersectional labeling with inducible control. We also show that the split-intein Gal4 system can be used to generate highly cell type–specific genetic drivers based on in silico predictions generated by single-cell RNAseq (scRNAseq) datasets, and we describe an algorithm (“Two Against Background” or TAB) to predict cluster-specific gene pairs across multiple tissue-specific scRNA datasets. We provide a plasmid toolkit to efficiently create split-intein Gal4 drivers based on either CRISPR knock-ins to target genes or using enhancer fragments. Altogether, the split-intein Gal4 system allows for the creation of highly specific intersectional genetic drivers that are inducible/repressible. Drosophila \| split-Gal4 \| intersectional genetics \| single-cell transcriptomics The ability to restrict transgene expression to specific, genetically defined cell types using binary expression systems such as Gal4/UAS, LexA/LexAOP, and QF/QUAS has pro- foundly transformed Drosophila research (1–3). In particular, the Gal4 system has been deployed extraordinarily effectively, with thousands of Gal4 drivers available in Drosophila resource centers. However, the lack of tissue- and cell-type specificity of many Gal4 drivers remains a drawback. This is especially true for certain areas of research. For example, studies of interorgan communication in which a Gal4-driven manipulation is performed in one tissue and the effects are measured in a distant tissue must take special care to avoid the confounding effects of Gal4 expression outside of the intended tissue (4). Similarly, many neurobiological studies require Gal4 expression to be limited to one, or very few, transcriptionally defined neuron, which is not generally possible using standard Gal4 drivers, even when driven by 2 to 3 kb genomic enhancer fragments (5–7). The split-Gal4 system was developed to overcome the issue of limited cell-type speci- ficity, by restricting transgene expression to those cells that coexpress two independent enhancers, a strategy termed “intersectional genetic labeling” (8, 9). In split-Gal4, the N-terminal 147 amino acids of Gal4, which includes its DNA-binding domain (Gal4DBD) (10) and its dimerization domain (11), is expressed under the control of one enhancer, while a potent transcriptional activator domain (AD) from either VP16 or p65 is expressed under the control of a second enhancer (Fig. 1) (8, 12). The Gal4DBD and the VP16/p65 activation domains are each flanked by a leucine zipper domain, which heterodimerize in any cell expressing both components, and reconstitute a functional Gal4-like transcription factor (8, 9). The split-Gal4 system has been successfully used to build thousands of exquisitely specific genetic drivers, especially in the Drosophila nervous system where split-Gal4 lines are now routinely utilized to drive expression in a single pair of neurons (6, 7), and in the adult gut, where thousands of split-Gal4 lines have been characterized (13, 14). The ability to create split-Gal4 lines that are specifically expressed in the same patterns as genes of interest using “trojan exons” or other knock-in strategies has further augmented the power of the split-Gal4 method (15). This capability has particular promise in permitting the construction of genetic driver lines that target transcriptionally distinct clusters identified via scRNAseq studies (16). For such clusters, the intersection of at least two genetic markers is typically necessary to uniquely identify specific clusters. Significance The split-Gal4 system allows Drosophila researchers to drive transgene expression with extraordinary cell type specificity. However, the existing split-Gal4 system cannot be controlled temporally, and therefore cannot be applied to many important areas of research. Here, we present a split-Gal4 system based on a self-excising split- intein, which is controllable by Gal80, as well as a related drug-inducible split GeneSwitch system. This approach can both leverage and inform single-cell RNAseq datasets, and we introduce an algorithm to identify pairs of genes that precisely and narrowly mark a desired cell cluster. Our split-intein Gal4 system will be of value to the Drosophila research community, and allow for the creation of highly specific genetic drivers that are also inducible/ repressible. Author contributions: B.E.-C., H.L., J.X., R.S., B.B., B.H.W., and N.P. designed research; B.E.-C., H.L., and J.X. performed research; B.E.-C., H.L., J.X., R.S., N.J., T.T., B.B., B.H.W., and N.P. contributed new reagents/analytic tools; B.E.-C., H.L., J.X., and R.S. analyzed data; and B.E.-C. wrote the paper. Reviewers: F.P., German Cancer Research Center; and C.J.P., Johns Hopkins University School of Medicine. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1B.E.-C., H.L., J.X., and R.S. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2304730120/-/DCSupplemental. Published June 5, 2023. PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 1 of 12 A Gal4 B Original split-Gal4 C split-intein Gal4 enhancer 1 Gal4 Gal80 l 4 a G enhancer 1 enhancer 2 Gal4DBD zip- p65 or VP16 zip+ leucine zipper binding Zip- Zip+ Gal80 no binding p65 D B D 4 a G l enhancer 1 Gal41-20 int-N Gal421-881 enhancer 2 int-C in vivo trans-splicing Gal80 intein intein l 4 a G seamless UAS GFP UAS GFP UAS GFP Fig. 1. Schematic comparison of Gal4, split-Gal4, and split-intein Gal4. (A) The Gal4 transcription factor binds to UAS sequences to drive transcription and can be repressed by the binding of Gal80. Gal4 is drawn here as a monomer, but functions as a dimer in vivo. (B) In the original split-Gal4 system, the Gal4DBD and a strong transcriptional activator (VP16 or p65) are each driven by separate enhancers and reconstituted in cells by leucine zipper domains. Gal80 cannot bind or repress the split-Gal4 complex. (C) In the split-intein Gal4 system, two fragments of the Gal4 protein, each flanked by a split-intein, are independently driven by separate enhancers and seamlessly transspliced to reconstitute a functional, wild-type Gal4 protein, which can be repressed by Gal80. But while the split-Gal4 system has effectively solved the prob- lem of restricting expression in anatomical space, the existing split-Gal4 system cannot be controlled in time. This is in contrast to standard Gal4 drivers, which can be temporally controlled using a temperature-sensitive variant of the Gal80 repressor (Gal80ts) (17). By shifting between a permissive temperature (18 °C), where Gal80ts represses Gal4 expression, and a restrictive temperature (29 °C), where Gal80ts is inactivated and Gal4 becomes active, researchers can restrict genetic manipulations to specific time peri- ods or developmental stages. By contrast to the standard Gal4 system, the split-Gal4 system is completely insensitive to the Gal80 repressor (9). This is because the region of Gal4 that is bound by Gal80, the C-terminal 30 amino acids (18), falls squarely within the Gal4 AD domain, which is replaced in existing split-Gal4 implementations with either VP16 or p65 in order to drive sufficiently high levels of expression (Figs. 1 and 2A). Thus, the Gal4DBD-VP16 or Gal4DBD-p65 protein complexes do not contain any binding site for Gal80 and therefore cannot be repressed (Fig. 1). There is thus a clear need for an intersectional labeling system that is repressible by Gal80ts or otherwise inducible. Here, we describe two such systems. The first we term “split-intein Gal4.” This system combines the enhanced spatial control offered by split-Gal4 with the ability to strictly limit genetic manipulations to specific periods of time using existing Gal80ts reagents. We demonstrate that split-intein Gal4 drives UAS transgenes at expression levels that are indistinguishable from existing split-Gal4 and Gal4 reagents and that it can be repressed by standard Gal80 reagents. The second system is a closely related drug-inducible GeneSwitch technique (“split-intein GeneSwitch”), which pro- vides an alternative means to induce intersectional genetic labeling using a drug rather than a temperature shift. Finally, we demon- strate that the split-intein Gal4 system can be effectively used with scRNAseq datasets to generate split-intein Gal4 driver lines. To facilitate production of such lines, we present an algorithm to select gene pairs with low levels of predicted coexpression outside the cluster of interest. Finally, we provide a simple cloning and transgenesis workflow that can be used to generate large numbers of split-intein-Gal4 lines, either via CRISPR-based knock-in or using enhancer fragments. Results Designing a Split-Gal4 Technique That Can Be Repressed by Gal80. We wished to create an inducible/repressible intersectional labeling technique that can be controlled temporally. We focused our efforts on modifying the highly effective split-Gal4 concept, to make a split system that did not rely on leucine zipper heterodimerization and that incorporated the native Gal4 activation domain, thus rendering it sensitive to the Gal80ts repressor. We devised two independent strategies, which we refer to as “split-intein Gal4” and “NanoTag split-Gal4.” Split-intein Gal4. Native split-inteins consist of N- and C-terminal peptides that are fused to proteins encoded at separate genomic loci. Upon translation, these peptides associate with one another, self-excise, and seamlessly trans-splice the two adjacent polypeptide chains to which they are fused (19). Split-inteins have been successfully exploited to generate split proteins used in other expression systems (20, 21), and we sought to use them here to reconstitute wild-type Gal4 from two functionally inert fragments. In a series of pilot experiments in S2 cells, we tested three dif- ferent cysteine residues to split Gal4 into two nonfunctional fragments and four different split-intein systems (see Materials and Methods for full description) and ultimately identified the most potent system, which we refer to as “split-intein Gal4.” In this system, the Gal4 protein is split into two fragments: an N-terminal 20 amino acid portion (Gal4N-int) and the remaining C-terminal 861 amino acids (Gal4C-int), each flanked by compo- nents of the highly active gp41-1 split-intein sequence (19) (Figs. 1 and 2A). When these two fragments are coexpressed in a cell, the 2 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org A split-intein Gal4 B Original split-Gal4 split-intein Gal4 Myo1A-T2A-Gal4DBD ∩ tub-VP16 Larval midgut enterocytes Myo1A-T2A-Gal4N-int ∩ tub-Gal4C-int Gal41-20 Gal421-881 int-N (88aa) int-C (38aa) in vivo trans-splicing intein + Gal41-881 DBD domain AD-1 AD-2 dimerization domain Gal80 binding site P F G E I P A D P F G E P F G E I P A D P F G E esg-T2A-Gal4DBD ∩ tub-VP16 esg-T2A-Gal4N-int ∩ tub-Gal4C-int Larval ISCs C esg-T2A-Gal4N-int ∩ tub-Gal4C-int Dl-T2A-Gal4C-int ∩ tub-Gal4N-int esg-T2A-Gal4N-int ∩ Dl-T2A-Gal4C-int Adult ISCs P F G E I P A D P F G E Fig. 2. Split-intein Gal4 system drives intersectional expression at levels indistinguishable from split-Gal4. (A) Schematic diagram of the split-intein Gal4 system. Gal4 and gp-41 are drawn to scale, illustrating the N-terminal DNA-binding domain (DBD) and dimerization domain, and the overlap of the second activation domain (AD-2) with the Gal80-binding site. (B) Components for original split-Gal4 (Left) or split-intein Gal4 (Right) were knocked into two gut cell-type markers: Myo1A, which labels enterocytes (ECs), and esg, which labels intestinal stem cells (ISCs). These knock-in lines were crossed to ubiquitously expressed tester lines to visualize their full expression pattern. (C) Intersectional labeling of midgut intestinal stem cells using esg ∩ Dl split-intein Gal4 knock-in lines. Brackets indicate expression in the anterior hindgut which is driven by Dl but not esg, which is absent in their intersection. Anterior is to the left. (Scale bars, 50 µm.) split-intein activity is predicted to reconstitute the full wild-type Gal4 protein, which should be repressible by Gal80 (Figs. 1 and 2A). Previous studies in Caenorhabditis elegans have demonstrated a related approach, in which a DNA-binding domain and an AD are transspliced via gp41-1 split-intein (20). However, in that approach, a VP64 AD is used instead of the native Gal4 domains, and thus this system is not repressible by Gal80. NanoTag split-Gal4. “NanoTags” are short epitope tags (<25 amino acids) that are recognized with very high affinity by single-domain nanobodies. Recently, two high-affinity NanoTags, 127D01 and VHH05, have been adapted for a variety of applications in vivo in Drosophila (22). We designed a split-Gal4 system based on the affinity of the 127D01 tag and its genetically encoded nanobody, Nb127D01. In a series of pilot experiments in S2R+ cells, we observed that Gal4DBD-Nb127D01 combined with Gal4AD- 1x127D01 drove only very weak expression. However, when we fused three Nb12701 nanobodies in tandem to a Gal4DBD domain (Gal4DBD-3xNb127D01), making it capable of rec- ruiting three Gal4-AD molecules to each Gal4DBD domain, we observed robust transgene expression. We refer to this combination as NanoTag split-Gal4. Both Split-Intein Gal4 and NanoTag Split-Gal4 Function in Drosophila Cell Culture. To test the transcriptional activation strength of each system in cell culture, we transiently transfected either split-intein components (Gal4N-int and Gal4C-int) or NanoTag Split-Gal4 components (Gal4DBD-3xNb12701 and Gal4AD- 1x127D01), all driven by a constitutive Actin5c promoter, into PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 3 of 12 S2R+ cells, along with a green fluorescent protein reporter (UAS:GFP). As positive controls, we transfected full-length Gal4 and standard split-Gal4 components, Zip-Gal4DBD and p65-Zip. Two days after transient transfection, we observed strong GFP expression for both split-intein Gal4 and NanoTag split-Gal4, at similar levels to Gal4 itself or to the existing split-Gal4 system (SI Appendix, Fig. S1). We tested whether split-intein Gal4 and NanoTag Gal4 were repressible by Gal80 in S2R+ cells by cotransfecting these com- ponents with pTub:Gal80. As expected, wild-type Gal4, but not the existing split-Gal4 system (Gal4DBD-p65), was strongly repressed by Gal80 (SI Appendix, Fig. S1). Strikingly, both split-intein Gal4 and NanoTag split-Gal4 exhibited strong repres- sion by Gal80, albeit slightly weaker than that observed for wild-type Gal4 (SI Appendix, Fig. S1). We conclude that both approaches offer robust transcriptional activation at levels similar to the existing split-Gal4 system, but have the critical advantage that they are also sensitive to the Gal80 repressor. While both approaches showed promise in S2R+ cells, the 3:1 stoichiometry of Gal4AD:Gal4DBD in the NanoTag system sug- gested that this approach might require higher levels of Gal80 than the split-intein Gal4 approach to achieve the same level of repression. Since Gal80 expression levels in vivo will generally vary from cell type to cell type for any given Gal80 line and high sen- sitivity is therefore desirable, we chose to focus on the split-intein Gal4 system for additional in vivo testing. To The Split-Intein Gal4 System Activates High Levels of Intersectional UAS-Driven Expression In Vivo. In order to be a broadly useful tool in vivo, split-intein Gal4 must meet three criteria. It must: 1) drive robust expression in vivo, at levels similar to existing split-Gal4 or Gal4 lines; 2) drive clean intersectional labeling that is not “leaky,” and includes only those cells expressing both Gal4N-int and Gal4C-int components; 3) be repressible using existing Gal80ts lines. characterize in vivo, we used split-intein Gal4 CRISPR-mediated knock-in transgenesis to insert split-intein components into various genes with well-characterized expression patterns. We first selected two genes expressed in specific cell types of the midgut: Myo1A (aka Myo31DF) which is expressed in entero- cytes (ECs) (23), and esg, which is expressed in intestinal stem cells (ISCs) (24). To permit direct comparison with the current split-Gal4 system, we also generated knock-ins of ZipGal4DBD into the same positions within esg and Myo1A. To create these knock-ins, we adapted the “drop-in” cloning method (25) to generate homology-driven repair (HDR) donor plasmids that would insert an in-frame T2A sequence, followed by the split-intein Gal4 or split-Gal4 component, into an early exon of the target gene. We also generated ubiquitously expressed split-intein Gal4 compo- nents, driven by the alphaTubulin48B promoter, to use as “tester” lines to visualize the complete expression pattern of each knock-in. We crossed Myo1A-T2A-Gal4N-int to the tub-Gal4C-int; UAS:2xEGFP (enhanced green fluorescent protein) tester line and observed cell type–specific expression in larval ECs, at levels statistically indistin- guishable from the original split-Gal4 system (mean pixel intensity measured in n = 4 larval guts; t(6) = 0.9325, P = 0.3871) (Fig. 2B). Similarly, esg-T2A-Gal4N-int ∩ tub-Gal4C-int (hereafter we follow the convention of using the ∩ symbol to indicate intersectional labeling) drove specific expression in ISCs at similar levels to the standard split-Gal4 system (n = 3 larval guts; t(4) = 2.22; P = 0.091) (Fig. 2B). These results indicate that the split-intein Gal4 system functions robustly in vivo. Expression in the midgut was specific for the two targeted cell types, indicating that the Gal4C-int fragment did not sup- port leaky expression. We then tested whether the split-intein Gal4 approach would successfully drive intersectional expression using two cell type– specific knock-in lines. As esg and Myo1A are not coexpressed in the gut, we knocked Gal4C-int into the Delta (Dl) gene, which is also expressed in ISCs (26). As expected, esg ∩ Dl expression was observed in adult ISCs (Fig. 2C). Importantly, the expression of Dl in the anterior hindgut was not observed in the esg ∩ Dl inter- section, providing additional evidence that the split-intein system is not leaky (Fig. 2C). Thus, the split-intein Gal4 system satisfies the first two criteria identified above: it drives expression at similar levels to the existing split-Gal4 system, and expression can only be detected in cells coexpressing both components. Split-Intein Gal4 Is Repressible by Gal80ts. The split-intein Gal4 system should seamlessly reconstitute wild-type Gal4, which, unlike the original split-Gal4 system, is repressible by Gal80 (Fig. 3A). To confirm this is the case, we generated larvae expressing both tub-Gal4N-int and tub-Gal4C-int as well as tub-Gal80ts and a UAS:2x-EGFP reporter. When tub-Gal4N-int ∩ tub-Gal4C-int, tub-Gal80ts > UAS:2xEGFP larva were grown at 18 °C, no EGFP could be detected, similar to standard tub-Gal4, tub-Gal80ts > UAS:2xEGFP larvae (Fig. 3 B, Left). However, when grown at 29 °C, strong EGFP expression was observed (Fig. 2 B, Right). We also confirmed that, in the absence of Gal80ts, the split-intein Gal4 system does indeed drive EGFP expression at 18 °C (SI Appendix, Fig. S2), indicating that the lack of EGFP expression at 18 °C is not the result of compro- mised split-intein trans-splicing, and is indeed due to Gal80 repression. These results also demonstrated that, like wild-type Gal4, split-intein Gal4 activity increases with temperature (27) (SI Appendix, Fig. S2). These results show that split-intein Gal4 is repressible by Gal80. To confirm that the Gal80 repression of split-intein Gal4 is sufficiently potent to fully repress strong, dominant phenotypes at 18°C, we turned to a widely used tumor model in the adult gut. When activated yki is expressed in ISCs using esg-Gal4, it generates severe tumor phenotypes in the adult gut (28–30). We used the split-intein Gal4 system to drive activated yki (UAS:yki3SA) in adult ISCs (esg-Gal4N-int ∩ tub-Gal4C-int), in the presence of tub-Gal80ts. Grown at 18 °C, these flies developed and eclosed normally, and displayed no EGFP expression or tumor growth in the ISCs, similar to the corresponding Gal4 system (Fig. 3 C, Left). However, after a 3-d temperature shift, we observed a dramatic tumor phenotype indistinguishable from those produced by esg-Gal4 (Fig. 3 C, Middle). Further, we could re-repress this phe- notype and the associated EGFP expression by shifting these flies back to 18 °C for 10 d (Fig. 3 C, Right). Altogether, these exper- iments confirm that split-intein Gal4 drives high levels of expres- sion, is not leaky, and is repressible by existing Gal80ts reagents. The gene-specific split-intein Gal4 drivers described above were generated using CRISPR-based knock-ins, in order to fully reca- pitulate the endogenous expression pattern of the target gene. However, many researchers may wish to create Gal80-sensitive split-intein Gal4 drivers using specific enhancer fragments, as has been done successfully for thousands of split-Gal4 lines that have been generated for the VT split-Gal4 collection (6, 7). To facilitate this approach, we modified the pBPZpGAL4DBD and pBP- p65ADZp destination vectors (12) to encode split-intein Gal4 components. These vectors are compatible with the well-established Gateway LR–based cloning workflow for generating enhancer fragment–driven split-Gal4 vectors (12, 31). To demonstrate the effectiveness of this approach, we selected a genomic fragment known to drive expression in the adult gut ISCs, VT024642 (13), 4 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org A 18ºC Gal80ts B tubGal80ts, UAS:2x-EGFP 18ºC 29ºC l 4 a G no expression Gal4 tub-Gal4 29ºC UAS GFP Gal80ts Gal80ts heat- inactivated l 4 a G split-intein Gal4 tub-Gal4N-int ∩ tub-Gal4C-int GFP Enhanced contrast UAS GFP Enhanced contrast C UAS:yki3SA, tubGal80ts, UAS:GFP 18ºC18ºC 29ºC (3 days) GFP GFP DAPI GFP GFP DAPI Return to 18ºC (10 days) GFP DAPI GFP Gal4 esg-Gal4 split-intein Gal4 esg-T2A-Gal4N-int ∩ tub-Gal4C-int Fig. 3. Split-intein Gal4 is repressible by Gal80. (A) Cartoon diagram of temperature-inducible expression using Gal80ts. (B) tub-Gal4N-int ∩ tub-Gal4C-int tub-Gal80ts > UAS:2xEGFP expression is repressed at 18 °C and highly active at 29 °C, as is tub-Gal4, tub-Gal80ts > UAS:2xEGFP. Note that the 18 °C images are shown with increased gain relative to the 29 °C images in order to visualize the presence of larva. (C) Recapitulation of a well-characterized adult stem cell tumor system using split-intein Gal4. esg-Gal4N-int ∩ tub-Gal4C-int, tub-Gal80ts > UAS:yki3SA in adult ISCs can be repressed throughout development and adult stages by tubGal80ts and reversibly activated using temperature shift to 29 °C. Anterior is up, scale bar in (C) is 50 µm. and cloned this fragment into our pBP-Gal4N-int destination vec- tor. As predicted, VT024642-Gal4N-int ∩ tub-Gal4C-int drove strong, specific expression in adult ISCs (SI Appendix, Fig. S3.) Split-Intein Gal4 Components Can Be Adapted to GeneSwitch for Drug-Inducible Intersectional Labeling. Having established that split-intein Gal4 is highly effective, we reasoned that this system should also be adaptable to the drug-inducible GeneSwitch system (32). In GeneSwitch, the Gal4DBD (the first 93 amino acids of the Gal4 protein) is fused to an RU486-sensitive ligand-binding domain (PR-LBD) and a p65 transcriptional AD. In the absence of RU486 (RU), the GeneSwitch complex is inactive, whereas in the presence of RU, the complex undergoes a conformational change that allows for the transcriptional activation of UAS-driven transgenes. We noted that the Gal4N-int fragment, compromising the first 20 amino acids of Gal4, could be compatible with the correspond- ing C-terminal region of GeneSwitch (Gal421-93-PR-LBD-p65 aka GeneSwitchC-int) flanked by a split-intein (Fig. 4A). In other words, the same Gal4N-int lines could be crossed to either a Gal4C-int for split-intein Gal4 expression, or to a GeneSwitchC-int line for split-intein GeneSwitch expression. To test this, we generated a transgenic line expressing split-intein-GeneSwitchC-int under the control of the tub promoter. We crossed tub-split-intein- GeneSwitchC-int; UAS:2xEGFP to esg-T2A-Gal4N-int and split the F1 adult flies into RU- and RU+ minus treatments for 6 d. In the absence of RU, we observed no GFP expression in the adult gut, whereas flies fed RU-containing food displayed strong and specific GFP in adult ISCs (Fig. 4 B, Top). In parallel, we crossed tub-split-intein-GeneSwitchC-int; UAS:2xEGFP to esg-T2A-Gal4N-int; UAS:yki3SA to test whether we could successfully regulate tumor growth via RU feeding. While RU- flies displayed no EGFP expression and wild-type gut morphology, RU+ flies displayed strong ISC tumor phenotypes resembling those observed using either Gal4 or split-intein Gal4 (Fig. 4 B, Bottom). Thus, the split-intein GeneSwitch system successfully combines intersec- tional genetic labeling with the RU- inducibility of GeneSwitch. Importantly, the GeneSwitch system has been shown to be leaky in some tissues, with detectable expression in the absence of RU (32–34). Given that split-intein GeneSwitch simply reconstitutes the existing GeneSwitch protein, we predicted this leaky expres- sion would also be the case with split-intein GeneSwitch. We tested PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 5 of 12 A B split-intein GeneSwitch enhancer 1 Gal41-20 int-N in vivo trans-splicing full length GeneSwitch Gal41-93 PR-LBD p65 enhancer 2 GeneSwitchC-term Gal421-93 PR-LBD p65 int-C + intein esg-T2A-Gal4N-int ∩ tub-split-intein-GeneSwitchC-int no RU +RU (200 µM, 6 days) GFP EGFP DAPI GFP EGFP DAPI UAS:2xEGFP UAS:yki3SA + UAS:2xEGFP Fig. 4. Drug-inducible intersectional labeling using split-intein GeneSwitch (A) Cartoon schematic of the split-intein GeneSwitch system, not drawn to scale. The same N-terminal fragment of Gal4 used in split-intein Gal4 can be combined with the C terminus of the GeneSwitch system, which includes amino acids 21 to 93 of Gal4, a progesterone ligand-binding domain (PR-LBD), and the p65 transcriptional activator. (B) Drug-inducible ISC tumor model using the esg-Gal4N-int ∩ tub-GeneSwitchC-int > UAS:yki3SA. Anterior is up. (Scale bar, 50 µm.) this by crossing the ubiquitously expressed tub-split-intein- GeneSwitchC-inttester line to three additional Gal4N-int lines: Myo1A-T2A-Gal4N-int, Dl-T2A-Gal4N-int, and tub-Gal4N-int. While we observed clean RU-dependent expression in adult ISCs with both esg and Dl (SI Appendix, Fig. S4 A and B), we observed leaky expres- sion in a portion of the adult gut using Myo1A (SI Appendix, Fig. S4C), as well in the larval gut when using the tub promoter (SI Appendix, Fig. S4D). Thus, as with existing GeneSwitch reagents, it will be important for researchers to carefully characterize the RU- and RU+ expression patterns for split-intein GeneSwitch lines. Mapping scRNA Clusters to Anatomy Using Split-Intein Gal4 Drivers. One particularly promising use of intersectional labeling techniques such as split-intein Gal4 is to characterize the many transcriptionally defined “clusters” of cells that are identified using scRNAseq. Single-cell and single-nuclei transcriptomic atlases are now available for many individual Drosophila tissues, as well as for the entire adult body (35). These atlases identify many different distinct cell types within a given tissue based on transcriptional similarity, many of which remain uncharacterized either anatomically or functionally. In most cases, a single genetic marker is insufficiently specific to label a cluster, and a minimum of two coexpressed genes are generally required to demarcate a cluster (36). Thus, intersectional genetic labeling approaches are a promising tool to interrogate hypotheses generated via scRNAseq. The promise of this approach has recently been piloted in the Drosophila optic lobe (16). To explore how the split-intein Gal4 system can leverage and inform scRNAseq datasets, we began with a recent atlas of the adult midgut, which identified 22 transcriptionally distinct cell types (37). To pick pairs of genes that uniquely mark scRNA clusters, we implemented a recently described gene selection algo- rithm, NS-Forest version 2.0 (36). NS-Forest v2 is a machine learning algorithm that estimates the minimum number of marker genes that can be used to uniquely define scRNAseq clusters. Using NS-Forest v2 to guide gene pair selection, we generated transgenic split-intein Gal4 lines to mark three of these clusters in vivo: 1) aEC-3, predicted to be a subset of anterior ECs, marked by Peritrophin-15a ∩ CG4830; 2) iron and copper cells, a func- tional analog of the human stomach located midway between the anterior and posterior of the midgut, marked by CG43774 ∩ thetaTry; and 3) pEC-1, predicted to be a subset of posterior ECs, marked by LManV ∩ ninaD (Fig. 5). We generated split-intein Gal4 lines for each of these three pairs and examined their expression using UAS:2xEGFP. In each case, the expression pattern conformed well with the predicted location of the cell cluster. Our aEC3 split-intein Gal4 line Peritrophin-15a ∩ CG4830 drove expression in a band of ECs anterior to the copper cells, corresponding to the “A3” region identified by (38) (Fig. 5A). This band of expression had a sharp posterior border, 6 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org A B C scRNA co-expression by cluster in vivo split-intein Gal4 expression pattern Peritrophin-15a CG4830 Anterior EC-3 i n o s s e r p x E m u S 1000 800 600 400 200 0 Peritrophin-15aGal4-N-int ∩ CG4830Gal4-C-int Cu-Fe cardia 500 µm GFP DAPI midgut- hindgut juncture Cluster: cardia ISC/EB AstA-EE NPF-EE AstC-EE dEC aEC1 aEC2 aEC3 aEC4 mEC Cu\|Fe LFC pEC1 pEC2 pEC3 EC-like d1 d2 fat un1 un2 CG43774 thetaTry Copper Iron cells CG43774Gal4-N-int ∩ thetaTryGal4-C-int i n o s s e r p x E m u S 2000 1500 1000 500 0 cardia ISC/EB AstA-EE NPF-EE AstC-EE dEC aEC1 aEC2 aEC3 aEC4 mEC Cu\|Fe LFC pEC1 pEC2 pEC3 EC-like d1 d2 fat un1 un2 LManV ninaD Posterior EC-1 LManVGal4-N-int ∩ ninaDGal4-C-int i n o s s e r p x E m u S 2000 1500 1000 500 0 cardia ISC/EB AstA-EE NPF-EE AstC-EE dEC aEC1 aEC2 aEC3 aEC4 mEC Cu\|Fe LFC pEC1 pEC2 pEC3 EC-like d1 d2 fat un1 un2 Fig. 5. Mapping scRNA clusters to anatomy using split-intein Gal4 lines based on NS-Forest v2 predictions. Based on a gut-specific scRNA dataset from Hung et al. (37), the NS-Forest v2 algorithm identified pairs of genes to mark specific clusters in the adult gut. Left: Summed expression in each of the 22 clusters is shown on the Left. Right: In vivo expression of split-intein Gal4 lines in adult guts. Brackets indicate the approximate position of the copper and iron cell (Cu–Fe) region. (A) Peritrophin-15a ∩ CG4830 drives expression in a band of enterocytes anterior to the copper cells. (B) CG43774 ∩ thetaTry drives expression in the copper and iron cells. (C) LManV ∩ ninaD drives expression in a posterior band of enterocytes. Anterior is to the left. but we also observed patchy expression extending anteriorly from the aEC3 cluster (Fig. 5A). This may reflect the in silico prediction of lower levels of coexpression in the aEC2 cluster, which is tran- scriptionally very similar to aEC3 (38) (Fig. 5A). The split-intein Gal4 line CG43774 ∩ thetaTry, predicted to express in iron and copper cells, drove expression in precisely this region (Fig. 5B), and the pEC-1 line LManV ∩ ninaD drove expression in a band of ECs posterior to the copper cells (Fig. 5C). Thus, using in silico predictions to guide gene selection, we were able to successfully use split-intein Gal4 to label anatomically distinct populations of cells along the anterior–posterior axis of the adult midgut, demon- strating the potential power to identify and functionally charac- terize cell types identified via scRNAseq studies. Developing an Algorithm to Pick Cluster-Specific Gene Pairs with Minimal Coexpression across the “Whole-Body” scRNA Dataset. Subsequent to the publication of the midgut-specific scRNAseq dataset described by (37) and utilized above, the Fly Cell Atlas Consortium published scRNAseq datasets covering many additional adult Drosophila tissues, as well as a whole-body dataset (35). As the set of potential off-target clusters now spans the whole body, it becomes more difficult to obtain gene pairs that mark a specific cluster. In particular, for each of the midgut- relevant gene pairs predicted by NS-Forest v2, we examined the in silico coexpression of these gene pairs across the entire body of Fly Cell Atlas dataset. For the large majority of gene pairs, we observed that the NS-Forest v2 gene pairs had a high degree of coexpression in multiple other tissues. This has practical implications: it is crucial to identify intersectional genetic drivers that are exclusively expressed in specific tissues, with no additional expression anywhere else in the body. We therefore sought to develop an algorithm that would spe- cifically identify cluster-specific gene pairs that maximize coex- pression in the cluster of interest, while minimizing additional coexpression in both the tissue of interest, as well as across other scRNAseq datasets from the same organism. To do so, we devel- oped a gene-selection algorithm that we call “Two Against Background” or TAB. The TAB algorithm is schematized in Fig. 6A. Briefly, TAB incorporates three features to guide gene pair selection: First, it incor- porates bulk RNAseq data, when available, to supplement scRNAseq expression estimates. Second, it emphasizes selecting genes with robust within-cluster expression profiles that are stable and not highly variable. Third, to calibrate the importance of cluster speci- ficity relative to these robustness considerations, it employs a hyper- parameter optimization approach by incorporating rankings by a researcher who is blinded to the parameters. As input, TAB requires both the cell cluster (e.g., “escort cells”) and the containing tissue (e.g., “ovary”). To ensure that the genes being selected are well expressed in the tissue, we crossreference the gene’s bulk scRNAseq expression levels in the corresponding organ. While the organ-level resolution is coarser than scRNAseq cell clusters, the higher quality of bulk expression data is a valuable corrective for noisy scRNAseq expression estimates. To quantify the specificity of any gene to the organ of interest, we use the Tau statistic. In addition, we require that the gene’s within-cluster PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 7 of 12 A Two Against Background (TAB) B Hemocytes Input - FCA cell cluster (“escort cells”) - Fly Organ (“ovary”) Fly Cell Atlas - Body “hemocytes” Ppn + kuz Fly Cell Atlas Bulk fly RNA-seq Per-gene statistical tests differential expression vs. rest expression dispersion in cluster : : organ specificity Gene-pair statistical tests All pairwise combinations - Wilcoxon test (avg. of both genes) - Co-presence in other FCA clusters Ranked set of gene pairs GFP L3 larvae GFP Adult body wall Ppn-Gal4N-int ∩ kuz-Gal4C-int GFP Hemese DAPI 98.8% overlap (n = 1076 cells) Larval body wall 100 µm GFP Hemese DAPI 50 µm C D E Cluster ID Fly Cell Atlas - Ovary In silico co-expression In vivo expression “choriogenic main body follicle cell St. 14” CG31928 + Pez CG31928-Gal4N-int ∩ Pez-Gal4C-int GFP DAPI Stage 14 follicle cells Corpus luteum Escort cells st.14 oocyte 100 µm “choriogenic main body follicle cell and corpus luteum” hdc + Nox hdc-Gal4N-int ∩ Nox-Gal4C-int corpus luteum 50 µm SKIP + CG42566 SKIP-Gal4N-int ∩ CG42566-Gal4C-int escort cells 50 µm additional expression “escort cell” Fig. 6. Characterization of split-intein Gal4 lines based on Two Against Background (TAB) algorithm predictions. (A) Schematic of the TAB algorithm to pick gene pairs that specifically mark scRNAseq clusters. (B) Hemocyte-specific split-intein Gal4 line based on TAB gene pairs. Top Left: Fly Cell Atlas 10× “Body” atlas, with hemocyte cluster indicated in blue. Top Right: In silico prediction of Ppn and kuz coexpression in the hemocyte cluster, with coexpression shown in yellow. Bottom: In vivo expression of the Ppn ∩ kuz split-intein Gal4 driver in larval and adult hemocytes. Hemocytes are costained with the H2 antibody against Hemese, a larval pan-hemocyte marker. (C–E) In silico predictions of coexpression in three different clusters from the FCA “Ovary” dataset and in vivo coexpression in the indicated cell types of the ovary. scRNAseq data are screenshots from the FCA data viewer for the “10×, stringent” datasets. expression be stable and not highly variable, i.e., the dispersion (variance/mean) of the gene expression in the cluster is limited. In the TAB algorithm, a candidate set of gene pairs is created for each cluster of interest using the intersection of three metrics: the Tau statistic, dispersion metrics, and t test of differentiation against all other clusters (Fig. 6A). From this candidate set, we evaluate all pairwise combinations of genes and select pairs that are effective at distinguishing the cluster of interest from others. 8 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org One of the metrics we consider is the number of other clusters where both the candidate genes are potential markers. We also introduce an additional metric: we construct a metagene as the average of the two genes and perform the Wilcoxon rank-sum test to assess differential expression of the metagene in this cluster against other clusters. To optimize hyperparameters, a subset of gene pair predictions was analyzed by a researcher who was blinded to the parameters and who used the FCA data visualizer to rank the specificity of each gene pair in the cluster of interest, and across multiple FCA datasets. The final score for the candidate gene pair is a weighted combination of these metrics, and we output a ranked list of choices from which we select final gene pairs. Our implementation of the TAB algorithm is publicly available at https://github.com/rs239/tab_gene_markers. To test the efficacy of the TAB algorithm, we used it to identify a “hemocyte” cluster from the FCA 10X “Body” dataset that uniquely coexpressed the genes Ppn and kuz (Fig. 6B). We then generated a pair of split-intein Gal4 lines designed to target the intersection of Ppn and kuz (Fig. 6B.) We observed specific expres- sion of Ppn ∩ kuz in larval and adult hemocytes, verified by stain- ing with the pan-hemocyte H2 antibody, which recognizes Hemese (Fig. 6B) (39). In larvae, 98.8% of Hemese+ cells were also GFP+ (n = 1,076 cells from two larvae), and we did not observe GFP+ cells that were Hemese negative. In adults, the H2 antibody did not appear to stain all of the morphologically iden- tifiable hemocytes in the adult, consistent with previous observa- tions (39). To test whether Ppn ∩ kuz drives expression in additional cell types, we examined expression in sagittally sec- tioned, decapitated adult flies, as well as in adult brains. In addi- tion to specific expression in circulating hemocytes, we observed strong expression in a band of epithelial cells within the cardia, also known as the proventriculus, a structure at the foregut–mid- gut juncture (SI Appendix, Fig. S5B) (40). Interestingly, several previous studies have identified hemocyte-like cells at this location in the larva, which express independent hemocyte markers and may play an immune function (41, 42). We confirmed that the pan-hemocyte marker hml-Gal4 is expressed in a subset of the cells at this anatomical position (SI Appendix, Fig. S5B). Thus, our split-intein Gal4 Ppn ∩ kuz line appears to be concordant with other hemocyte-specific markers and was not detected in other tissues. We next generated a series of split-intein Gal4 lines to mark specific clusters based on the tissue-specific FCA “Ovary” atlas while minimizing expression in any other cluster at the level of the whole body. We selected three clusters from the FCA ovary dataset (Fig. 6 C–E) and examined the expression of the resulting split-intein Gal4 lines in vivo. In all the three cases, we observed the predicted expression. CG31928 ∩ Pez predicted to express in follicle cells of stage 14 oocytes and drove GFP expression specif- ically in these cells (Fig. 6C). hdc ∩ Nox drove expression in the corpus luteum, a tissue composed of the follicle cells left behind after an egg is laid (43) (Fig. 6D). SKIP ∩ CG42566 drove the predicted expression in escort cells at the anterior tip of each ovariole, although we also observed expression in stalk cells between each germline cyst (Fig. 6E). To test whether these split-intein Gal4 lines drive expression in additional, nontargeted tissues, we examined expression in adult flies that had been sagitally sectioned, as well as in adult brains. Neither CG31928 ∩ Pez nor hdc ∩ Nox drove detectable EGFP expression in any other adult tissue outside the desired cell type (SI Appendix, Fig. S5 C and D). SKIP ∩ CG42566 drove expres- sion in a small number of neurons in the brain (SI Appendix, Fig. S5E), but was otherwise undetectable outside of the germaria, in which escort cells reside. Thus, these lines were highly specific to the targeted cell type. These pilot experiments demonstrate that the TAB algorithm will be a useful tool to generate highly specific genetic tools for clusters identified by scRNAseq datasets, whether at the level of individual scRNAseq datasets, or across multiple datasets or the whole organism. In addition to the NS-Forest v2 gene selection algorithm, a recent manuscript has presented an alternative gene selection algorithm for creating cell type–specific split-Gal4 lines in Drosophila based on scRNAseq datasets (16). Our TAB algorithm differs in several ways from the one described by Chen et al. TAB considers bulk tissue RNAseq datasets to enhance confidence in tissue-specific expression, and it also considers each member of the gene pair as equally important rather than implementing a greedy search based on a prespecified first gene. In addition, TAB uses well-established metrics of differential expression and gene specificity to create tests, rather than assuming a unimodal vs. bimodal gene expression model. Future in vivo experiments will be useful to empirically compare the relative utility of these and other approaches. One-Step Generation of Double-Knock-In Split-Intein Gal4 Lines Using Dual Drug Selection. One technical bottleneck in the production of split-intein Gal4 lines or split-Gal4 lines is the fact that two independent transgenic lines must be created for each desired genetic driver. We reasoned that it may be possible to generate split-intein Gal4 lines in a single step by simultaneously generating double knock-in lines. To test this approach, we adapted our knock-in vectors to contain drug selection markers that were recently characterized as transgenesis markers in Drosophila (44). In this approach, each knock-in is marked by a separate drug resistance gene, and double knock-in transformants are selected by rearing larvae on food containing both of the relevant drugs. We created a modified version of our Gal4N-int donor vector containing a resistance gene for blasticidin (BlastR), and a version of our Gal4C-int donor vector containing a resistance gene for G418 (G418R), and retained the fluorescent 3xP3-dsRed marker in both vectors. We used TAB to identify two pairs of genes which mark clusters from the FCA Gut atlas: CG13321 ∩ CG6484 to label “enterocyte of anterior adult midgut epithelium,” and CG14275 ∩ CG5404 to label “hindgut.” For each gene pair, we generated two separate drug resistance knock-in vectors, with one construct resistant to blasticidin and the other resistant to G418 (Fig. 7A). We injected a 1:1 mixture of these two vectors into nos-Cas9 embryos and mass-mated the resulting injected G0 flies to a bal- ancer stock, on food containing both blasticidin and G418 (Fig. 7A). Of the flies that survived, we selected flies with dsRed+ eyes and screened these by crossing to a UAS:2xEGFP reporter. For CG14275 ∩ CG5404, we only recovered a small number of flies after drug screening, zero of which were dsRed+. However, for CG13321 ∩ CG6484, of the 12 dsRed+ flies we screened, six (50%) drove GFP expression, indicating successful one-step cre- ation of double knock-ins in cis on chromosome 2R. These six lines drove strong EGFP expression throughout the anterior region of the midgut, from the posterior limit of the cardia to the anterior limit of the copper cells, as well as weaker, spotty expression in portions of the posterior midgut (Fig. 7B). Independently, we for CG13321-Gal4N-int and created CG6484-Gal4C-int using our standard HDR vectors and confirmed that these lines drove expression in an identical pattern. separate knock-ins We note that both of these genes are located on Chromosome 2R, indicating that it would be challenging to use standard recom- bination genetics to create a single chromosome containing both inserts, which further demonstrates the value of making a one-step double knock-in. However, we note an important caveat. Specifically, the double-drug selection protocol was not 100% PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 9 of 12 A One-step generation of “double knock-in” split-Gal4 lines using drug resistance markers B Predicted co-expression in: “enterocyte of anterior adult midgut epithelium” CG13321-T2A-Gal4N-int ∩ CG6484-T2A-Gal4C-int > UAS:2xEGFP GFP GFP DAPI A1-A3 Drug-resistant “drop-in” donor plasmids for CRISPR-HDR knock-in CG13321 homology arms CG6484 homology arms + hsp:BlastR 3xP3-dsRed T2A-Gal4N-int sgRNA hsp:G418R 3xP3-dsRed T2A-Gal4C-int sgRNA inject embryos with both constructs yv ;; nos::Cas9 Mass-mating to w ; Gla/CyO on blasticidin + G418 food Screen surviving flies for dsRed+ eyes Balance and screen with w ; Sp/CyO ; UAS:2xEGFP CG13321-T2A-Gal4N-int ∩ CG6484-T2A-Gal4C-int > UAS:2xEGFP GFP - n = 6 GFP + n = 6 Fig. 7. One-step generation of double-knock-in split-intein Gal4 line using drug selection markers. (A) Knock-in donor vectors containing drug resistance markers (BlastR or G418R) are coinjected into embryos expressing germline-restricted Cas9, and the offspring of these injected G0s are screened for double drug resistance. Of the surviving F1, dsRed+ flies are screened for the ability to drive UAS:EGFP expression. Pie chart indicates the proportion of dsRed+ flies that successfully drove EGFP in the predicted cells, and image shows an L3 larva expressing EGFP in a portion of the gut, anterior to the left. (B) Expression pattern of CG6484-Gal4N-int ∩ CG13321-Gal4C-int in the adult gut, with anterior enterocyte regions A1-A3 (38) indicated with a bracket. Anterior is up. effective, as some dsRed-negative flies were observed after the first round of mating. In addition, 50% of the dsRed+ flies did not drive EGFP expression, which could indicate either that the two knock-in events occurred in trans on homologous chromosomes and were thus not captured via our screening procedure, or that only a single knock-in occurred. Thus, while the double knock-in strategy can serve to quickly generate split-intein Gal4 lines, it will require addi- tional troubleshooting to be a reliable and scalable approach. Discussion In the original description of the split-Gal4 system, it was noted that the replacement of the native Gal4 AD with the VP16 activator represented a trade-off: VP16 drove much stronger expression, but rendered split-Gal4 insensitive to repression by Gal80 (8). Here, we present an alternative split-Gal4 system that obviates the need for this trade-off by generating full-length wild-type Gal4 protein from two nonfunctional fragments, using self-splicing split-inteins. The split-intein Gal4 system combines the exquisite cell type spec- ificity of split-Gal4 with the ability to temporally control Gal4 activity using existing Gal80ts reagents. This system drives clean and specific transgene expression at similar levels to the existing split-Gal4 and Gal4 systems and is repressible by Gal80ts. Similar to Gal4 lines, additional spatial restriction of split-intein Gal4 activ- ity should be possible using existing Gal80 lines. We believe that these advantages will make the split-intein Gal4 system a valuable addition to the toolkit available to the Drosophila research commu- nity for targeted transgene expression in specific cell types. Targeting of specific cell types should be further facilitated by the widespread availability of scRNAseq datasets. As demonstrated here, such datasets can be leveraged to create intersectional split-intein Gal4 tools based on the knowledge of cell type–specific gene expression. This approach should allow researchers to test hypotheses generated by scRNAseq atlases and to de-orphan clus- ters of unknown anatomy or function. It will also aid in the cre- ation of highly specific drivers for nearly all cell types and tissues in the fly and permit functional manipulations of these cell types with temporal control. To aid in the design of cell type–specific drivers, we have developed the TAB algorithm, which we believe will reduce the potential for coexpression outside a specific cluster of interest. Future characterization of the TAB algorithm across a variety of scRNAseq datasets will help further refine the cell type– specific tools available to the Drosophila research community. To facilitate the creation of split-intein Gal4 lines, we have generated a plasmid tool kit to create split-intein Gal4 lines, either enhancer driven or via knock-in. For knock-ins, we provide plas- mids for cloning via “long” homology arms (~1,000 bp), or via “drop-in cloning” using 200 bp fragments, which is what we use in this manuscript. These plasmids are diagrammed in SI Appendix, Fig. S6 and have been deposited in Addgene. 10 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org Materials and Methods Full methods are available in SI Appendix. Experimental Animals. Drosophila melanogaster stocks were maintained and crossed on standard laboratory cornmeal food, and experiments were conducted at 18 °C, 25 °C, or 29 °C as indicated in the text. All adult experiments were per- formed in females. The new transgenic lines created in this study are described in SI Appendix, Table S1, and all genotypes are provided in SI Appendix, Table S2. Optimization and Cloning of Split-Intein Gal4 and NanoTag Split-Gal4 Components. Design and cloning of components for both cell culture and in vivo, including “drop-in” cloning of knock-in vectors as well as promoter-driven constructs, is described in detail in Supplemental Index Materials and Methods. Testing Split-Intein Gal4 and NanoTag Split-Gal4 in S2R+ Cells. Drosophila S2R+ cells (DGRC, 150) were cultured at 25 °C, in Schneider’s media (Thermo Fisher Scientific, 21720–024) with 10% fetal bovine serum (Sigma, A3912) and 50 U/mL penicillin-streptomycin (Thermo Fisher Scientific, 15070–063). S2R+ cells were transfected using Effectene (Qiagen, 301427) following the manu- facturer’s instructions. Two hundred nanograms of plasmid DNA per well was transfected in 24-well plates. The cultured cells were imaged live 2 d after trans- fection on an InCell Analyzer 6,000 automated confocal fluorescence microscope (GE Healthcare Lifesciences). RU486 Treatment. RU486 (Cayman Chemical Company Cat. No. 10006317) was added to standard fly food at a final concentration of 200 µm. For larval experiments, eggs were laid directly onto RU-containing food. For adult gut experiments, eggs were laid on and developed on standard food, and adults were transferred to RU-containing food for the indicated time. Drug Selection for Double Knock-Ins. G418 (final concentration = 250 µg/mL) and blasticidin (final concentration = 45 µg/mL) were added to 25 mL of standard food in bottles and allowed to dry, uncovered, overnight in a fume-hood. Injected flies were mass-mated to balancer lines on drug food and flipped approximately every 3 d onto new drug-containing food. The surviving F1 offspring were screened for dsRed+ eyes, and any flies with dsRed+ eyes were then crossed to w; Sp/CyO; 2xEGFP to simultaneously balance and screen for double knock-ins of split-intein Gal4 components. Antibody Staining and Imaging. For sagittal sections of whole flies, decap- itated adult female flies were fixed overnight in 4% paraformaldehyde, then manually sectioned using a fine razorblade (Personna by AccuTec, Cat No. 74-0002). After antibody staining, bisected flies were placed in a drop of VECTASHIELD mounting media in a 35-mm, glass-bottom imaging µ-Dish (Ibidi, Cat. No. 81158). Tissues were dissected in PBS, fixed for 20 to 30 min in 4% paraformaldehyde, and stained using standard protocols. GFP was detected using either Alexa488-coupled anti-GFP (Invitrogen A21311, used at 1:400) or chicken anti-GFP (Aves Lab GFP1020, used at 1:2,000). Hemocytes were stained using the pan-hemocyte H2 antibody (39) (Gift of Andó lab, used at 1:100). Primary antibodies were detected with Alexa-488 or Alexa-555 coupled secondary antibodies (Molecular Probes). Confocal imaging was performed on either a Zeiss LSM 780 or Zeiss Axio Observer Z1 with a LSM980 Scan Head, with the “Tile Scan” feature for whole guts using system defaults. Whole-larva imaging was performed on a Zeiss AxioZoom microscope. Mean pixel intensity was measured using FIJI/ImageJ, based on maximum intensity projections, with GFP+ pixels selected as regions of interest. TAB Algorithm. The scripts for TAB implementation are available at https:// github.com/rs239/tab_gene_markers, and the algorithm is described in detail in Supplemental Index. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. We thank the Microscopy Resources on the North Quad core at Harvard Medical School for microscopy support. We thank Yifang Liu and Claire Hu for assistance in selecting gene pairs using NS-Forest v2 and Rich Binari for assistance with fly work. The Perrimon Lab received funding from 5P41GM132087 and 5R24OD026435. The White lab received funding from the Intramural Research Program of the National Institute of Mental Health (ZIAMH002800, B.H.W.). J.X. is supported by the start-up funding from Center for Excellence in Molecular Plant Sciences, CAS. R.S. and B.B. received funding from NIH 1R35GM141861. N.P. is an HHMI investigator. Author affiliations: aDepartment of Genetics, Blavatnik Institute, Harvard Medical School, Boston, MA 02115; bLaboratory of Molecular Biology, National Institute of Mental Health, NIH, Bethesda, MD 20892; cCAS Key Laboratory of Insect Developmental and Evolutionary Biology, CAS Center for Excellence in Molecular Plant Sciences, Chinese Academy of Sciences, Shanghai 200032, China; dComputer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139; eDepartment of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02143; and fHHMI, Boston, MA 02115 1. 2. 3. 4. 5. A. H. Brand, N. Perrimon, Targeted gene expression as a means of altering cell fates and generating dominant phenotypes. Development. 118, 401–415 (1993). S.-L. Lai, T. Lee, Genetic mosaic with dual binary transcriptional systems in Drosophila. Nat. Neurosci. 9, 703–709 (2006). C. J. Potter, B. Tasic, E. V. Russler, L. Liang, L. Luo, The Q system: A repressible binary system for transgene expression, lineage tracing, and mosaic analysis. Cell 141, 536–548 (2010). I. A. Droujinine, N. Perrimon, Interorgan communication pathways in physiology: Focus on Drosophila. Annu. Rev. Genet. 50, 539–570 (2015). A. Jenett et al., A GAL4-driver line resource for Drosophila neurobiology. Cell Rep. 2, 991–1001 (2012). 6. H. Dionne, K. L. Hibbard, A. Cavallaro, J.-C. Kao, G. M. Rubin, Genetic reagents for making split-GAL4 7. lines in Drosophila. Genetics 209, 31–35 (2018). L. Tirian, B. J. Dickson, The VT GAL4, LexA, and split-GAL4 driver line collections for targeted expression in the Drosophila nervous system. bioRxiv [Preprint] (2017). https://doi. org/10.1101/198648. (Accessed 3 January 2023). 16. Y.-C. D. Chen et al., Using single-cell RNA sequencing to generate cell-type-specific split-GAL4 reagents throughout development. bioRxiv [Preprint] (2023), https://doi. org/10.1101/2023.02.03.527019 (Accessed 2 April 2023). 17. S. E. McGuire, P. T. Le, A. J. Osborn, K. Matsumoto, R. L. Davis, Spatiotemporal rescue of memory dysfunction in Drosophila. Science. 302, 1765–1768 (2003). 18. J. Ma, M. Ptashne, The carboxy-terminal 30 amino acids of GAL4 are recognized by GAL80. Cell 50, 137–142 (1987). 19. P. Carvajal-Vallejos, R. Pallissé, H. D. Mootz, S. R. Schmidt, Unprecedented rates and efficiencies revealed for new natural split inteins from metagenomic sources*. J. Biol. Chem. 287, 28686–28696 (2012). 20. H. Wang, J. Liu, K. P. Yuet, A. J. Hill, P. W. Sternberg, Split cGAL, an intersectional strategy using a split intein for refined spatiotemporal transgene control in Caenorhabditis elegans. Proc. Natl. Acad. Sci. U.S.A. 115, 3900–3905 (2018). 21. H. Luan, A. Kuzin, W. F. Odenwald, B. H. White, Cre-assisted fine-mapping of neural circuits using orthogonal split inteins. Elife 9, e53041 (2020). 22. J. Xu et al., Protein visualization and manipulation in Drosophila through the use of epitope tags 8. H. Luan, N. C. Peabody, C. R. Vinson, B. H. White, Refined spatial manipulation of Neuronal function recognized by nanobodies. Elife 11, e74326 (2022). by combinatorial restriction of transgene expression. Neuron 52, 425–436 (2006). 23. H. Jiang et al., Cytokine/Jak/Stat signaling mediates regeneration and homeostasis in the 9. H. Luan, F. Diao, R. L. Scott, B. H. White, The Drosophila split Gal4 system for neural circuit mapping. Drosophila midgut. Cell 137, 1343–1355 (2009). Front. Neural Circuits 14, 603397 (2020). 24. C. A. Micchelli, N. Perrimon, Evidence that stem cells reside in the adult Drosophila midgut 10. L. Keegan, G. Gill, M. Ptashne, Separation of DNA binding from the transcription-activating function epithelium. Nature 439, 475–479 (2005). of a eukaryotic regulatory protein. Science 231, 699–704 (1986). 25. O. Kanca et al., An efficient CRISPR-based strategy to insert small and large fragments of DNA using 11. M. Carey, H. Kakidani, J. Leatherwood, F. Mostashari, M. Ptashne, An amino-terminal fragment of short homology arms. Elife 8, e51539 (2019). GAL4 binds DNA as a dimer. J. Mol. Biol. 209, 423–432 (1989). 26. B. Ohlstein, A. Spradling, Multipotent Drosophila intestinal stem cells specify daughter cell fates by 12. B. D. Pfeiffer et al., Refinement of tools for targeted gene expression in Drosophila. Genetics 186, differential notch signaling. Science 315, 988–992 (2007). 13. 735–755 (2010). I. S. Ariyapala et al., Identification of split-GAL4 drivers and enhancers that allow regional cell type manipulations of the Drosophila melanogaster Intestine. Genetics 216, 891–903 (2020). 14. J. M. Holsopple, K. R. Cook, E. M. Popodi, Identification of novel split-GAL4 drivers for the characterization of enteroendocrine cells in the Drosophila melanogaster midgut. G3 (Bethesda). 12, jkac102 (2022), 10.1093/g3journal/jkac102. 27. A. H. Brand, A. S. Manoukian, N. Perrimon, Chapter 33 ectopic expression in Drosophila. Methods Cell Biol., 635–654 (1994). 28. F. Ren et al., Hippo signaling regulates Drosophila intestine stem cell proliferation through multiple pathways. Proc. Natl. Acad. Sci. U.S.A. 107, 21064–21069 (2010). 29. P. Karpowicz, J. Perez, N. Perrimon, The Hippo tumor suppressor pathway regulates intestinal stem cell regeneration. Development. 137, 4135–4145 (2010). 15. F. Diao et al., Plug-and-play genetic access to Drosophila cell types using exchangeable exon 30. R. L. Shaw et al., The Hippo pathway regulates intestinal stem cell proliferation during Drosophila cassettes. Cell Rep. 10, 1410–1421 (2015). adult midgut regeneration. Development 137, 4147–4158 (2010). PNAS 2023 Vol. 120 No. 24 e2304730120 https://doi.org/10.1073/pnas.2304730120 11 of 12 31. B. D. Pfeiffer et al., Tools for neuroanatomy and neurogenetics in Drosophila. Proc. Natl. Acad. Sci. 38. A. Marianes, A. C. Spradling, Physiological and stem cell compartmentalization within the U.S.A. 105, 9715–9720 (2008). Drosophila midgut. Elife 2, e00886 (2013). 32. T. Osterwalder, K. S. Yoon, B. H. White, H. Keshishian, A conditional tissue-specific transgene 39. É. Kurucz et al., Definition of Drosophila hemocyte subsets by cell-type specific antigens. Acta Biol. expression system using inducible GAL4. Proc. Natl. Acad. Sci. U.S.A. 98, 12596–12601 (2001). 33. F. Scialo, A. Sriram, R. Stefanatos, A. Sanz, Practical recommendations for the use of the geneswitch Hung. 58, 95–111 (2007). 40. D. G. King, Cellular organization and peritrophic membrane formation in the cardia (Proventriculus) Gal4 system to knock-down genes in Drosophila melanogaster. PLoS One 11, e0161817 (2016). of Drosophila melanogaster. J. Morphol. 196, 253–282 (1988). 34. L. Poirier, A. Shane, J. Zheng, L. Seroude, Characterization of the Drosophila gene-switch system in 41. A. Zaidman-Rémy, J. C. Regan, A. S. Brandão, A. Jacinto, The Drosophila larva as a tool to study gut- aging studies: A cautionary tale. Aging Cell 7, 758–770 (2008). 35. H. Li et al., Fly cell Atlas: A single-nucleus transcriptomic atlas of the adult fruit fly. Science 375, eabk2432 (2022). 36. B. D. Aevermann et al., A machine learning method for the discovery of minimum marker gene combinations for cell-type identification from single-cell RNA sequencing. Genome Res. 31, 1767–1780 (2021). associated macrophages: PI3K regulates a discrete hemocyte population at the proventriculus. Dev. Comp. Immunol. 36, 638–647 (2012). 42. B. Charroux, J. Royet, Elimination of plasmatocytes by targeted apoptosis reveals their role in multiple aspects of the Drosophila immune response. Proc. Natl. Acad. Sci. U.S.A. 106, 9797–9802 (2009). 43. L. D. Deady, W. Shen, S. A. Mosure, A. C. Spradling, J. Sun, Matrix metalloproteinase 2 is required for ovulation and corpus luteum formation in Drosophila. PLoS Genet. 11, e1004989 (2015). 37. R.-J. Hung et al., A cell atlas of the adult Drosophila midgut. Proc. Natl. Acad. Sci. U.S.A. 117, 44. N. Matinyan et al., Multiplexed drug-based selection and counterselection genetic manipulations in 1514–1523 (2020). Drosophila. Cell Rep. 36, 109700 (2021). 12 of 12 https://doi.org/10.1073/pnas.2304730120 pnas.org
10.1073_pnas.2220537120	RESEARCH ARTICLE \| BIOPHYSICS AND COMPUTATIONAL BIOLOGY OPEN ACCESS Multiple RNA- and DNA- binding proteins exhibit direct transfer of polynucleotides with implications for target- site search Wayne O. Hemphilla,b,c , James A. Goodricha, and Thomas R. Cecha,b,c,1 , Regan Fenskea,b,c , Calvin K. Voonga Edited by Joseph Puglisi, Stanford University School of Medicine, Stanford, CA; received December 2, 2022; accepted May 9, 2023 We previously demonstrated that the polycomb repressive complex 2 chromatin–modifying enzyme can directly transfer between RNA and DNA without a free- enzyme intermedi- ate state. Simulations suggested that such a direct transfer mechanism may be generally necessary for RNA to recruit proteins to chromatin, but the prevalence of direct transfer capability is unknown. Herein, we used fluorescence polarization assays and observed direct transfer for several well- characterized nucleic acid–binding proteins: three- prime repair exo- nuclease 1, heterogeneous nuclear ribonucleoprotein U, Fem- 3- binding factor 2, and MS2 bacteriophage coat protein. For TREX1, the direct transfer mechanism was additionally observed in single- molecule assays, and the data suggest that direct transfer occurs through an unstable ternary intermediate with partially associated polynucleotides. Generally, direct transfer could allow many DNA- and RNA- binding proteins to conduct a one- dimensional search for their target sites. Furthermore, proteins that bind both RNA and DNA might be capable of readily translocating between those ligands. nucleic acid \| chromatin \| single- molecule \| exchange \| displacement A few well- characterized oligomeric proteins have been shown to exchange ligand species through highly unstable ternary complex intermediates (1–5). This mechanism of molec- ular exchange between protein and polynucleotide molecules has been previously termed “direct transfer,” “facilitated dissociation” (FD), “facilitated exchange,” and “monkey bar mechanism;” herein, we use “direct transfer” since it is the earliest term used to describe direct protein translocation between ligand molecules (1, 6–12). In concurrent studies (13), we demonstrated that direct transfer occurs between RNA and DNA for the chromatin- modifying enzyme polycomb repressive complex 2 (PRC2) (14); this phenom- enon could allow mutually antagonistic RNA and DNA binding to both positively and negatively regulate PRC2 activity depending on the transcriptional environment. The prevalence of direct transfer among RNA- and DNA- binding proteins is not known. Thus, a robust study of direct transfer with various nucleic acid–binding proteins (NBPs), includ- ing monomeric proteins, was warranted. NBPs with well- characterized biochemistry span a wide range of biological functions. Heterogeneous nuclear ribonucleoprotein U (hnRNP- U) is an RNA- and DNA- binding protein proposed to regulate chromatin structure and pre- mRNA processing (15). Recent work has also demonstrated that hnRNP- U has specificity for G- quadruplex (G4) RNA (16), much like PRC2 (17). Three- prime repair exonuclease 1 (TREX1) (18, 19) is a 3′- to- 5′ exonuclease (20) that degrades DNA to prevent aberrant nucleic acid sensing (21) and the resulting autoimmunity (22). In recent years, interest in TREX1 has risen due to its potential as a cancer immunotherapy target (23), and TREX1 activity on ss- versus dsDNA, the purpose of its homodimer structure, and the source of TREX1’s DNA substrates in vivo remain areas of active interest (24, 25). Fem- 3- binding factor 2 (FBF- 2) is a Pumilio Factor (PUF) family sequence–specific RNA- binding protein that binds the 3′- untranslated regions of mRNAs to inhibit expression of proteins necessary for meiotic entry during Caenorhabditis elegans germline development (26–28). MS2 coat protein (MS2- CP) forms the capsid of this Escherichia coli bacteriophage, and it also negatively regulates MS2 replicase expression (29, 30). These functions require MS2- CP binding specifically to a hairpin RNA (31–34). Here, we use biophysical assays to interrogate the prevalence, mechanism, and bio- physical requisites of direct transfer among these NBPs. Our findings indicate that direct transfer occurs when polynucleotides compete for shared protein contacts and partially associate to form an unstable ternary complex intermediate (Fig. 1 A–C). This supports direct transfer being a feature of many RNA- binding chromatin- associated proteins that may be generally required for their tunable regulation (13). Notably, prior work has suggested that direct transfer could allow for protein movement along DNA (11, 12, 35–37), which would allow many NBPs to more efficiently search for their gene targets Significance Classically, the lifetime of a protein–ligand complex is presumed to be an intrinsic property, unaffected by competitor molecules in free solution. By contrast, a few oligomeric nucleic acid–binding proteins have been observed to exchange competing ligands in their binding sites, and consequently their lifetimes decrease with competitor concentration. Our findings suggest that this “direct transfer” capability may be a more general property of nucleic acid– binding proteins. Thus, many DNA- and RNA- binding proteins could reduce the dimensionality of their search for their target sites by direct transfer to nucleosome DNA, instead of relying entirely on three- dimensional diffusion. Furthermore, direct transfer from nascent RNA to DNA may explain why so many DNA- binding proteins also bind RNA. Author contributions: W.O.H., C.K.V., J.A.G., and T.R.C. designed research; W.O.H., C.K.V., and R.F. performed research; W.O.H. contributed new reagents/analytic tools; W.O.H. and R.F. analyzed data; and W.O.H. and T.R.C. wrote the paper. Competing interest statement: T.R.C. declares consulting status for Storm Therapeutics, Eikon Therapeutics, and SomaLogic. W.O.H. declares the filing of U.S. Provisional Application No. 62/706,167 Trex1 Inhibitors and Uses Thereof. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2220537120/- /DCSupplemental. Published June 20, 2023. PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 1 of 11 C A B D Fig. 1. Models and reaction schemes for protein–polynucleotide binding competition. (A) Classic model of protein–polynucleotide binding competition. The initial complex fully dissociates at a rate independent of free competitor, then the resulting free protein (green shape) can be bound opportunistically by a competitor polynucleotide (red shape), which prevents rebinding of the original ligand (blue shape). (B) Dynamic models of protein–polynucleotide binding competition. Many nucleic acid–binding reactions involve flexible ligands (“Flexible Ligand”) and/or proteins (“Flexible Protein”) and extensive protein–ligand contacts, which might be better described by a dynamic model that allows the formation of short- lived ternary complexes and partial ligand interactions. As a result, a competitor (red shape) can influence how quickly protein (green shape) dissociates from an initially bound ligand (blue shape). We term these “direct transfer” models (purple), in contrast to a classic transfer model (green). (C) Reaction schemes for classic and direct transfer. The models in panel B can be more agnostically presented as a reaction scheme (Complete). Complete reaction scheme includes protein (E), competitor (D), ligand (P), and numerous rate constants (k), where conjugations of reactants are complexes, and asterisked complex components are partially associated with protein (SI Appendix, Eq. 1). If partially associated complexes are presumed to be highly transient, then the simplified reaction scheme can describe the concentrations of remaining reactants and stable complexes (SI Appendix, Eq. 2). This is the reaction scheme used to experimentally determine the associated rate constant values herein, and its nomenclature corresponds with panel D and Table 1; in the cases where competitor and ligand are the same, the P/D nomenclature for rate constants can be omitted. (D) Experimental strategy to measure direct transfer kinetics. 1) The minimum amount of protein required for saturated binding is mixed with a trace amount of fluorescently labeled oligonucleotide, then incubated until thermal and reaction equilibrium. 2) Various concentrations of unlabeled competitor oligonucleotide are added to the preformed complex to initiate reactions. 3) The time- course reactions are immediately monitored by fluorescence polarization (FP) in a microplate reader. Potential complexes with their polarization states are shown, and they are labeled with rate constants describing intercomplex transitions. Rate constants associated with a classic competition model are indicated by green boxes, and those additionally necessary for a direct transfer model are indicated by a purple box. 4) Polarization signals are normalized to the range in polarization signal across all competitor concentrations to give proportion of initial complex remaining. Normalized polarization signals are plotted versus time and fit with one- phase exponential decay regression. 5) The initial slopes of obs) are plotted versus competitor concentration and regressed with custom equations describing the classic competition and direct transfer the regressions (koff models to determine rate constant values. Models are compared with the Bayesian Information Criterion (BIC). 2 of 11 https://doi.org/10.1073/pnas.2220537120 pnas.org after initial chromatin association (38, 39). Similarly, RNA- binding proteins could be transferred intramolecularly and intermolecu- larly in search for optimal binding sites. In addition, NBP com- plexes in vivo can be prematurely displaced by other binders in the reaction environment, instead of being rate limited by intrinsic dissociation of the complex, as previously discussed for FD (6, 40). These implications substantially expand the possibilities for how NBPs find their respective binding partners and regulate their biological activity. Results Diverse Protein–Polynucleotide Interactions Exhibit Direct Transfer Kinetics. Based on prior mechanistic proposals for direct transfer (1, 6, 11) and our concurrent PRC2 direct transfer studies (13), we suspected that many NBPs may have direct transfer capability. This could derive from competition on many NBP surfaces being less rigid than posited by a classic model (Fig. 1A), and instead being dynamic enough to facilitate direct transfer reactions within a single binding site (Fig. 1B). In SI Appendix, Theoretical Background, we outline how the complete reaction scheme for such a direct transfer model can be well approximated by a simplified reaction scheme (Fig. 1C), which others have previously discussed (1). Furthermore, we demonstrate that the rate constants for this simplified reaction scheme can be accurately determined experimentally with FP–based competitive dissociation (FPCD) experiments and regression models (Fig. 1D and SI Appendix, Table S1). To assess the generality of direct transfer, we selected four well- characterized protein–nucleic acid interactions, including Table 1. Rate constants for protein–ligand interactions Protein Ligand Competitor T (°C) hnRNP- U r(G3A2)4[F] r(G3A2)4 \|\|r(G3A2)4 \| C1 Streptavidin FAM- Biotin Biotin TREX1 [F]d(N)5 d(N)5 \|\|d(N)5 \| C2 ds- d(N)60 TREX1R174A,K175A ds- [F]d(N)60 d(N)5 [F]d(N)5 ds- d(N)60 ds- [F]d(N)60 d(N)5 FBF- 2 [F]RNAPUF MS2- CP [F]RNA2MS2 RNAPUF \|\|RNAPUF \| C3 RNA1MS2 \|\|RNA1MS2 \| C3 25 4 25 37 25 4 25 4 25 4 25 4 25 4→25 25 4 hnRNP- U binding to G4 RNA (16), TREX1 (sequence inde- pendent) binding to short ssDNA (18, 19), FBF- 2 (sequence specific) binding to a short ssRNA sequence (26, 27), and MS2- CP (sequence specific) binding to its RNA hairpin motif (31, 32). We also tested streptavidin binding to biotin as a non- NBP control (41). These five protein–ligand systems were chosen for their col- lective diversity in structure, type of bonding between protein and ligand, ligand specificity, binding surface area, affinity, and stoi- chiometry. We used purified recombinant protein and FP exper- app for these interactions (Table 1 and iments to determine Kd SI Appendix, Fig. S1). Our binding affinity results were generally consistent with prior reports for TREX1 (42), hnRNP- U (16), MS2- CP (33), FBF- 2 (27), and streptavidin (41), including the observation that streptavidin and hnRNP- U were more appropri- ately described by a Hill regression model (SI Appendix, Fig. S1A). Streptavidin and MS2- CP have significantly lower Kd than can be accurately determined by our methodology, so their binding curves are likely limited by ligand concentration (Table 1). Then, we performed FPCD experiments (Fig. 1D) with ligand (fluoro- phore labeled) and competitor (unlabeled) polynucleotides of the same identity to determine which protein–ligand interactions, if obs) that was dependent any, had an apparent dissociation rate (koff on the competitor concentration (i.e., direct transfer), versus pla- teauing at high competitor concentrations (i.e., classic competi- tion). In other words, was the dissociation rate sufficiently described by one rate constant for classic dissociation (k−1), or did it require an additional rate constant for direct transfer (kθ)? Unexpectedly, our results (Fig. 2) revealed direct transfer- like dis- sociation kinetics for every NBP, with their second- order rate constants for direct transfer spanning a 40- fold range. We note KdP app (nM) 15 ± 4.8 n.d. †2.5 ± 0.16 n.d. 8.9 ± 3.8 n.d. n.d. 30 ± 3.9 n.d. 46 ± 16 n.d. 130 ± 28 n.d. 50 ± 9.2 n.d. †2.9 ± 0.62 n.d. k−1P (s−1) n.d. (≥3.3×10−2) 6.4 ± 0.42 (×10−4) ‡9.7 (×10−4) ¶1.3 ± 0.68 (×10−5) ‡,¶1.7 (×10−5) ‡,#7.0 (×10−3) 6.8 ± 0.93 (×10−3) ‡6.6 (×10−3) ‡6.7 (×10−3) ‡,#1.9 (×10−2) 1.3 ± 0.34 (×10−2) n.d. ‡,#7.1 (×10−3) n.d. 6.7 ± 2.6 (×10−3) 4.4 ± 0.42 (×10−3) 1.1 ± 0.10 (×10−3) 1.4 ± 2.0 (×10−3) ‡4.6 (×10−3) ‡2.1 (×10−3) kθD (M−1s−1) n.d. 39 ± 9.2 ‡35 §0 §0 ‡,#1,500 700 ± 84 ‡610 ‡1,200 ‡,§0 §0 n.d. ‡,#330 n.d. 280 ± 65 140 ± 55 42 ± 13 89 ± 51 ‡57 ‡46 HOP n.d. −0.98 −1.7 n/a n/a 0.84 −0.22 −0.38 −0.58 n/a n/a n.d. −1.4 n.d. −1.5 1.4 −1.7 −0.92 −3.3 −2.5 app), intrinsic dissociation rate constants (k−1P), direct transfer rate constants FP- based methodology (Fig. 1D) was used to determine the apparent equilibrium dissociation constants (KdP (kθD), and “hand- off” proficiency (HOP) scores for several protein–ligand interactions. HOP scores are defined in Theoretical Background (SI Appendix) and are a metric of the ratio kθD/k−1P, where positive or negative values indicate above- or below- average propensity for direct transfer, respectively. Values indicate mean ± SD for at least three independent experiments. Kd experiments were carried out in the absence of competitor and were fit with standard (non- Hill, nonquadratic) regression. Rate constant definitions are in Fig. 1D, ligand/competitor definitions are in SI Appendix, Table S2, and binding curves with standard versus Hill regression are in SI Appendix, Fig. S1A. n.d. = not determined; n/a = not applicable. *Dissociation completed during initiation- measurement delay (~90 s). †Experiment used [Prey] ≥ KdP ‡Value from single experiment. §Bayesian Information Criterion (BIC) favored classic competition; rate constant not applicable. ¶Early partial dissociation curves and manual baseline used for regression; it is possible that k- 1 < k–1 #Late partial dissociation curves used for regression; it is possible that k–1 > k- 1 \|\|Total polynucleotide concentration was kept constant by serially diluting competitor in a carrier nucleic acid; C1 = r(A)20, C2 = r(N)5, and C3 = r(A)10. app; it is possible that Kd < Kd app or kθ > kθ app or kθ < kθ app. app. app. PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 3 of 11 A B C D E hnRNP-U (4°C): r(G3A2)4[F] r(G3A2)4 F Streptavidin (25°C): FAM-Biotin Biotin TREX1 (4°C): [F]d(N)5 d(N)5 FBF-2 (25°C): [F]RNAPUF RNAPUF MS2-CP (25°C): [F]RNA2MS2 RNA1MS2 (koff obs) (green versus (Fig. 1D), and Fig. 2. All tested nucleic acid– binding proteins exhibit kinetics consistent with direct transfer. (A–E) For several protein–ligand interactions, FPCD experiments were performed and analyzed as des- the final cribed plots of apparent ligand dissociation rate competitor concentration are shown alongside the crystal structures of their inter- actions. Plots show best- fit regression of the data, with equations describing classic competition lines) versus direct transfer (purple lines). Graphs are from representative experiments of at least 3, and error bars are mean ± SD across four reaction replicates. Crystal structures show proteins as gray cartoons and nucleic acid ligands as orange cartoons/sticks. Structures in panels B–E are PDB 2IZJ, 2OA8, 3V74, and 2C51, respectively; no crystal structure exists for hnRNP- U. The regression values corresponding app and are in Table 1, ligand Kd equilibrium competition data are in SI Appendix, Fig. S1, raw data examples are in SI Appendix, Fig. S2, carrier polynucleotide control experiments are in SI Appendix, Fig. S3, helpful nomenclature definitions are in Fig. 1D, and ligand/competitor definitions are in SI Appendix, Table S2. (F) The predicted proportion of protein trans- locations between polyn ucleotides proceeding through a direct transfer versus classic pathway (Fig. 1B) at various competitor effective molarities (“Comp.”). Data are for the inter- actions in panels A–E. Dashed gray line indicates equivalent flux through classic and direct transfer pathways. that the direct transfer rate constants are all several orders of mag- nitude lower than typical association rate constants, which we attribute to partial association of the competitor (to initiate direct transfer) being rate limited by infrequent fluctuation of the bound ligand to a partial association state. Only the streptavidin–biotin interaction had an apparent dissociation rate that plateaued at high competitor concentrations, consistent with classic competi- tion. The equilibrium data from these competition reactions are provided in SI Appendix, Fig. S1B. Since it was unexpected that these diverse protein–nucleic acid interactions would all exhibit direct transfer, we considered pos- sible artifactual explanations for the data. First, the assumptions 4 of 11 https://doi.org/10.1073/pnas.2220537120 pnas.org of our regression and analysis approach (SI Appendix, Theoretical Background and Methods) include exponential dissociation for reactions, so nonexponential dissociation of our complexes might obs calculation errors that mimic direct transfer kinet- produce koff ics. However, our raw data clearly exhibited exponential disso- ciation (SI Appendix, Fig. S2). Next, it seemed possible that obs across competitor concentrations could result variations in koff from nonspecific polynucleotide concentration–dependent fac- tors, such as polynucleotide aggregation or perturbation of the ionic environment of the reactions. To test this, we repeated the experiments of Fig. 2 using a nonbinding carrier polynucleotide to keep total competitor polynucleotide concentration constant while varying the ratio of binding to nonbinding competitor. We found that all protein–polynucleotide interactions still exhib- ited direct transfer kinetics (SI Appendix, Fig. S3). Collectively, our findings suggest that direct transfer might be widespread among NBPs. Relative flux through classic versus direct transfer pathways (Fig. 1B), as a function of competitor concentration, is described by SI Appendix, Eq. 10.4; this suggests that direct transfer signif- icantly affects our NBPs only at low- micromolar or higher effec- tive concentrations of competitor (Fig. 2F). Relative pathway flux is directly related to the ratio kθ/k−1 for a protein–ligand interac- tion, and these ratios for our NBPs fall within the range of those for previously interrogated direct transfer proteins such as SSB, CAP, and recA (1–3). Single- Molecule (SM) Experiments Support Direct Transfer for TREX1. As an orthogonal test for direct transfer (Fig. 1B), we employed total internal reflection fluorescence (TIRF) microscopy- based SM experiments (43). We chose the TREX1 + 5- mer ssDNA interaction for these experiments because it was well behaved. Although TREX1 is a dimer, the two DNA- binding sites cannot be spanned by DNA molecules of the length tested, and the protein does not exhibit cooperative binding for short ssDNA ligands (24, 42) (SI Appendix, Fig. S1A). Prior crystal structures indicate that ligands bound to separate protomers should have their 5′- fluorophores separated by ~8 nm with opposing orientations, suggesting that Förster/fluorescence resonance energy transfer (FRET) efficiency between these ligands may be limited (44–46). Unlabeled TREX1 was conjugated to a microscope slide, fluorophore- labeled oligonucleotide was flowed onto the slide with or without an excess of unlabeled oligonucleotide, and millisecond- interval movies were recorded to track individual particle- binding events (e.g., Fig. 3A). Distributions of residence times across all individual binding events (Fig. 3B) were used to obs for each movie (Fig. 3C), and then the average calculate koff obs values under each reaction condition were used to approx- koff imate k−1P and kθD (Fig. 3C). The SM results (k−1 ≈ 4.0 × 10−2 s−1, kθ ≈ 9,800 M−1s−1) had rate constant values about five- fold higher than those of our corresponding FP data (k−1 ≈ 0.7 × 10−2 s−1, kθ ≈ 1,500 M−1s−1), but they were performed under different buffer conditions. Importantly, the kθ/k−1 ratios between SM and FP results were similar (2.4 × 105 M−1 versus 2.1 × 105 M−1), and the competitor dependence of residence times was quite evident (Figs. 2 and 3B). Photobleaching control experiments indicated that only 5- 20% of these apparent dissociation events were attrib- utable to photobleaching (Materials and Methods). These findings demonstrate that the observation of direct transfer kinetics for TREX1 is not restricted to FP- based methodology. To directly observe transfer events, we modified the initial SM experiments to simultaneously track the TREX1- binding states of 5- mer ssDNA labeled with two different fluorophores. Across two such independent experiments, we identified n = 36 (10 nM of each fluorescently labeled ligand) or n = 34 (1 nM each ligand) apparent direct transfer events among 453 (10 nM each ligand) or 781 (1 nM each ligand) binding events (Fig. 3D); such modest proportions (7.9% or 4.4%, respectively) of direct transfer versus classic binding events were expected under the low ligand con- centrations necessary for these experiments. We declare the limi- tation that we cannot statistically associate the relative frequency of these observed ligand exchange events with increasing free ligand concentration, since higher fluorophore concentrations cannot be used in these SM experiments. We observed no instances in these experiments of more than two ligand molecules stably binding a TREX1 homodimer, suggesting that two ligands do not stably bind a single TREX1 protomer. To further interrogate this, we also recorded movies under conditions to detect FRET, and we were unable to detect any stable FRET signals (red/acceptor channel) on the same slides and fields of view as the colocalization experiments despite comparable particle numbers in the green/ donor channel (raw data available; see Data, Materials, and Software Availability). Collectively, these findings suggest that TREX1 protomers can be directly transferred between ligands, likely via a short- lived (i.e., <150 ms) ternary intermediate. Nonmultimeric Interactions Can Exhibit Direct Transfer Kinetics. Previous reports of direct transfer have concerned multimeric complexes with higher protomer–polynucleotide binding ratios, where ternary intermediates can be achieved by separate protomers binding separate ligands (1–3, 7). However, existing crystal structures of TREX1 (44), FBF- 2 (27), and MS2- CP (47) bound to ligand reveal that only one active site contributes to binding of each ligand, suggesting that their direct transfer must be occurring within a single protomer binding site. To confirm this implication, we performed FP- based stoichiometry experiments (SI Appendix, Supplemental Methods) to determine the number of ligand molecules bound to each functional unit of our proteins. Our results for all protein–polynucleotide interactions (SI Appendix, Fig. S4) were consistent with the stoichiometry expected from their crystal structures. These findings indicate that direct transfer can occur for protein–polynucleotide interactions within a single protomer binding site. The Direct Transfer Mechanism Is Revealed in TREX1 Ligand Competition. Direct transfer was ubiquitous among our tested NBPs (Fig. 2) with virtually identical ligand and competitor molecules, implicating shared protein contacts in the direct transfer mechanism. Consequently, we suspected that NBPs might accommodate direct transfer due to intrinsically dynamic protein– polynucleotide binding interfaces (Fig. 1B). This model for the direct transfer mechanism suggests that its kinetics are likely influenced by the proportion of shared versus unshared protein contacts for competing ligands. TREX1 has well- characterized properties that make it suitable for interrogating this hypothesis. Existing crystal structures demonstrate that the 3′- termini of short ssDNA (44) and lengthier dsDNA (48) are positioned identically in TREX1’s nucleotide- binding pocket (Fig. 4A), and biochemical studies indicate that lengthy dsDNA (>10 bp) has significant additional interactions with TREX1’s flexible binding loop (Fig. 4A) that do not occur for short ssDNA (24, 49). Thus, we designed two different ligands for TREX1 (Fig. 4A): a ssDNA 5- mer, [F]d(N)5, which should only bind in TREX1’s nucleotide- binding pocket, and a 60- bp dsDNA, ds- [F]d(N)60, which should bind in TREX1’s nucleotide- binding pocket with an additional “foothold” on its flexible binding loop. We anticipated that the dsDNA ligand’s foothold on TREX1’s flexible binding loop would prevent efficient direct transfer of TREX1 from the dsDNA ligand PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 5 of 11 C A B D 0.47 ∆AU ∆ 0.40 ∆AU -0.44 ∆AU ∆ -0.72 ∆AU 0.77 ∆AU 0.57 ∆AU 0.55 ∆AU ∆ ∆ Time (s) Time (s) Fig. 3. Single- molecule experiments support direct transfer of two DNA molecules on TREX1. (A–C) Single- molecule TIRF- microscopy experiments used FLAG- tagged TREX1 conjugated to the microscope slide, Cy5- labeled ligand DNA, and unlabeled competitor DNA. (A) Representative particle trace of TREX1- binding event. (B) Distributions of single- molecule residence times. Slides were prepared with labeled ligand only (blue), or labeled and unlabeled ligands (red), in two independent days of experiments (A and B). On each day and for each condition, at least four replicate movies each were collected with two different power/ exposure settings (0.2 s and 0.5 s). All data for each condition were regressed with an exponential probability density function, which is shown normalized alongside a normalized histogram (top graph). Total event numbers recorded for each condition are also shown (n). Box- and- whisker plots (bottom graph) of the distribution of single- particle residence times are shown for each contributing dataset for each condition, which are comprised of at least four replicate movies. On box- and- whisker plots, narrowed box centers indicate median, boxes define innerquartile range, and line segments define total range; top- to- bottom, the plots represent 157, 114, 320, 234, 185, 198, 274, and 290 data points. (C) Effect of competitor on apparent TREX1–ssDNA dissociation rate. Apparent dissociation obs) were calculated for each contributing data set in panel B via regression with an exponential probability density function and are plotted as mean ± rates (koff obs) as described in Materials and Methods. SD. The dissociation (k−1) and direct transfer (kθ) rate constants were calculated from apparent dissociation rates (koff (D) Observing direct transfer of ligands on TREX1. SM- TIRF colocalization experiments were carried out using FLAG- tagged TREX1 conjugated to the microscope slide and a mixture of Cy5- (red) and Cy3- labeled (green) ligands. Slides were prepared with 5 to 20 pM of conjugated protein and 1 to 10 nM of each ligand, then respective binding states were simultaneously monitored via dual red and green channel excitation and imaging. Example traces of observed ligand transfer events are shown, taken from two different days of experiments. Cartoons illustrate respective binding states for TREX1 homodimer (blue), Cy3- labeled ligand (green), and Cy5- labeled ligand (red). Solid lines indicate signal, dashed horizontal lines indicate average signal per state, and corresponding changes in average signal between states (ΔAU) are shown at top between corresponding cartoons. to the ssDNA ligand, while direct transfer of ssDNA to dsDNA should occur. We confirmed competitive binding of the two DNAs to TREX1 via FP (Fig. 4B). Next, we tested the proposed binding interactions for the two ligands. Protein–ligand interactions in the TREX1’s nucleotide- binding pocket occur primarily through its two divalent metal ions and hydrogen bonding with 3′- terminal nucleotides (44), while interactions with its flexible binding loop are reportedly via ionic bonding with the DNA backbone (24). We measured TREX1 binding to both ligands by FP. TREX1’s binding affinity for the ssDNA ligand was sig- nificantly reduced by divalent metal ion chelation and minimally 6 of 11 https://doi.org/10.1073/pnas.2220537120 pnas.org A B E G C D F Classic Compe(cid:21)(cid:21)on Model Direct Transfer Model Classic Compe(cid:22)(cid:22)on Model Direct Transfer Model [Compe(cid:22)tor] (µM) [Compe(cid:22)tor] (µM) Classic Compe(cid:21)(cid:21)on Model Classic Compe(cid:21)(cid:21)on Model Direct Transfer Model Direeeecttttttttttt TTTTTTTraaaaaaaaaaaaaaaaaaaannnnnnnnnnnnnnnnnnssssssssssssssssssffffffffffffeeeeeeeeeeeeeeeeeeerrrrrrrrrrrrrr MMMMMMMMMMMMMMMMMMMMMoooooooooooooooooooooooddddddddddddddddddeeeeeeeeeeeeeeeeeelllllllllll Classic Compe(cid:22)(cid:22)on Model Direct Transfer Model Classic Compe(cid:22)(cid:22)on Model Direct Transfer Model [Compe(cid:22)tor] (µM) [Compe(cid:22)tor] (µM) Fig. 4. Two competing TREX1 ligands exhibit different magnitudes of direct transfer. (A) TREX1 binds long dsDNA with extra protein–polynucleotide contacts. TREX1 (gray cartoon; ref. 24) binds exposed 3′- hydroxyls of DNA (orange cartoon) in its nucleotide- binding pocket (blue circle) with the help of divalent metal ions (black spheres). TREX1 makes additional contacts with long dsDNA through its flexible binding loop (dark red cartoon), particularly residues R174 and K175 (red sticks, labeled). We designed a 5- nt ssDNA that exclusively binds TREX1’s nucleotide- binding pocket ([F]d(N)5), and a 60- bp dsDNA having significant additional interactions with TREX1’s flexible binding loop (ds- [F]d(N)60). (24). (B) Two TREX1 ligands exhibit competitive binding. Fluorescence polarization was measured for labeled 60- bp dsDNA alone (5 nM ds- [F]d(N)60), TREX1 with labeled 60- bp dsDNA (+100 nM TREX1), and TREX1 with labeled 60- bp dsDNA and unlabeled 5- nt app for two ligands binding ssDNA (+100 nM TREX1 +10 µM d(N)5). Plot bars indicate mean ± SD from 8 reaction replicates. (C) Effects of buffer conditions on Kd to TREX1. Graphs are composites of three experiments with at least two reaction replicates each, where error bars indicate mean ± SD, and Kd values are mean app for each ligand. Standard binding experiments with TREX1 (solid ± SD across the replicate experiments. (D) Effects of flexible binding loop mutations on Kd lines) and TREX1R174A,K17A (dashed lines) were performed for the indicated ligands and binding buffers. Number of replicates and error bars as in (C). (E–G) Direct transfer for the TREX1 ds[F]d(N)60 and d(N)5 ligands is tunable. FPCD experiments (Fig. 1D) were performed for the indicated substrates, and the data fit with equations for classic competition (green line) and direct transfer (purple line). Graphs are from representative experiments (n ≥ 3 for panels E and F, n = 1 for panel G), where error bars are mean ± SD across four technical replicates. reduced by higher salt concentrations, while its affinity for the dsDNA ligand exhibited the opposite trend (Fig. 4C), supporting the binding interactions proposed for these two ligands. To fur- ther validate the binding interactions, we introduced the R174A/ K175A suppression- of- function mutations (49) (Fig. 4A) to TREX1’s flexible binding loop and repeated these binding exper- iments. The mutant had affinity for the ssDNA ligand compa- rable to that of the WT protein with no change in salt sensitivity, PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 7 of 11 but it had significantly reduced affinity for the dsDNA ligand with ablated salt sensitivity (Fig. 4D). Collectively, these findings confirm that our two TREX1 ligands are competitive binders, that the ssDNA ligand primarily interacts with TREX1’s nucleotide- binding pocket, and that the dsDNA ligand has a substantial additional foothold on TREX1’s flexible binding loop. To test whether the dsDNA ligand was resistant to displace- ment by ssDNA, we performed FPCD experiments (TREX1 dsDNAssDNA transfer, Fig. 1D). Our results (Fig. 4E and Table 1) demonstrated a competitor concentration- independent dissociation rate at high competitor concentrations. This was consistent with classic competition, indicating that direct trans- fer was ablated (Fig. 2C). This is our only example (within this particular study) of a protein–polynucleotide interaction with- out detectable direct transfer under our experimental condi- tions. Repeating these experiments with the TREX1 flexible binding loop mutant, which should perturb the dsDNA ligand’s foothold, restored direct transfer kinetics (Fig. 4F and Table 1), consistent with our proposed mechanism that the flexible loop acts as an additional foothold for dsDNA to prevent its dis- placement. Finally, we expected that testing both proteins with ssDNA as a ligand and dsDNA as a competitor would exhibit direct transfer, since the ssDNA has no unique foothold to mitigate direct transfer by the dsDNA. Indeed, our results (Fig. 4G) were consistent with direct transfer kinetics for both the wild- type and mutant TREX1 enzymes. These findings sup- port our proposed direct transfer mechanism (Fig. 1B). Fast Intrinsic Dissociation Is Correlated with Rapid Direct Transfer. Our model for the direct transfer mechanism (Fig. 1B) suggests that direct transfer (kθ) and dissociation (k- 1) rate constants should be correlated (SI Appendix, Eq. 6.3 and Theoretical Background). To test this implication, we compiled rate constant data from these studies (Table 1), concurrent studies (13), and applicable published studies (1–3) and plotted kθ as a function of k−1 (Fig. 5). The results reveal a striking correlation between these rate constants (R2 = 0.91), and this correlation was maintained within only this study (R2 = 0.75) and only published studies (R2 = 0.89). We emphasize that, as noted above (SI Appendix, Eq. 10.4 and Fig. 2F), an NBP’s location relative to the trendline in a graph shown in Fig. 5 is related to how quickly increasing ligand concentrations lead to increasing flux through a direct transfer versus classic pathway (Fig. 1B). Consequently, kθ/k−1 may be informative of relative direct transfer efficiency, and we propose a hand- off proficiency (HOP) score as a metric (SI Appendix, Theoretical Background); an HOP score of zero indicates a HOP that is average for the tested proteins. Taken together, these findings support the direct transfer mechanism (Fig. 1B) and its consistency among numerous independent studies. Discussion Our surveyed NBPs included RNA- and DNA- binders, single- and double- stranded polynucleotides, disordered and folded pol- ynucleotide structures, variable protein–polynucleotide contact surface areas, and sequence- independent versus sequence- specific binders. Notably, direct transfer occurred for virtually every NBP we studied, although the direct transfer rate constants varied among these interactions (Fig. 5 and Table 1). This implies that direct transfer could be a common capability among NBPs, though definitive conclusions will require further experimental testing. Importantly, we note that direct transfer can occur between Protein TREX1 FBF-2 hnRNP-U MS2-CP PRC2 PRC2 SSB CAP Ligand d(N)5 RNAPUF r(G3A2)4 RNA2MS2 Compe(cid:17)tor Source d(N)5 RNAPUF r(G3A2)4 RNA1MS2 These Studies These Studies These Studies These Studies G-quad RNA G-quad RNA Concurrent Studies 60-bp dsDNA 60-bp dsDNA Concurrent Studies d(T)69 d(T)70 Kozlov & Lohman, 2002 lac promoter-operator (203-bp) sheared E. coli DNA † Fried & Crothers, 1984 recA (slow) recA (fast) M13 DNA M13 DNA poly(dT) Menetski & Kowalczykowski, 1987 poly(dT) Menetski & Kowalczykowski, 1987 A B C D E F G H I J Fig. 5. Direct transfer kinetics correlate with intrinsic dissociation kinetics. Direct transfer rate constants (kθ) are plotted as a function of intrinsic dissociation rate constants (k−1), where rate constant nomenclature follows Fig. 1D. Each letter in the graph corresponds to a NBP direct transfer reaction, whose identities are listed in the table below. Regression of the data (blue dashed line) was performed on linear axes with a linear 0- intercept model. Data sourced from these studies refer to the lowest temperature isothermal competition experiments with no carrier polynucleotide, found in Table 1 and Fig. 2, and other data are sourced as described in SI Appendix, Supplemental Methods. Ligand/competitor definitions for these studies are in SI Appendix, Table S2.† Rate constant values were not explicitly stated in source publication; values were calculated from other source data as described in Materials and Methods. identical polynucleotides (i.e., self- competition), so even NBPs with a single preferred ligand can be affected by the consequences of direct transfer. Requirements for Direct Transfer. Direct transfer has been extensively studied for oligomeric NBPs (1–5, 50, 51) and one with separate flexible DNA- binding domains (11). For such proteins, it is intuitive how an incoming polynucleotide could gain a “foothold” on one of several unbound protomers to facilitate its displacement of a second polynucleotide prebound to a different protomer. Our findings (Fig. 2 and SI Appendix, Fig. S4) expand the relevance of direct transfer, suggesting that it may be common among RNA- and DNA- binding proteins that interact with a polynucleotide through only one binding site. Our working hypothesis is that direct transfer stems from dynamic protein–ligand contacts in a protein- binding site, as shown in Fig. 1B. This implies that direct transfer fundamentally relies on 1) shared protein contacts between competing polynucleotides, 2) the capacity for partial protein–polynucleotide interactions, and 3) the capacity to form transient ternary complexes. In the case of TREX1, our SM experiments showed multiple examples of directly correlated binding of one ssDNA and release of another (Fig. 3D) without stable FRET signaling, consistent with a short- lived 8 of 11 https://doi.org/10.1073/pnas.2220537120 pnas.org (e.g., <150 ms) ternary complex. Furthermore, the magnitude of direct transfer (as reflected in kθ) should be directly related to the extent of protein–polynucleotide contacts, the proportion of protein contacts shared by two polynucleotides, and the flexibility of the protein–polynucleotide interface(s), and it should be inversely related to the strength of the protein–ligand interaction. The last of these proposals is supported by the correlation between the k−1 and kθ rate constants in Fig. 5. The degree to which each of these properties contributes to direct transfer is an open question. Since streptavidin–biotin epit- omizes extremes of these properties and does not exhibit direct transfer (Fig. 2B), we expect that there are thresholds for each of these properties, past which direct transfer cannot meaningfully occur. However, we tested high- affinity interactions with RNA hairpins (Fig. 2E) and 3 to 4 nt of DNA (Fig. 2C), which have very limited protein- binding interfaces, and these still exhibited direct transfer to some degree. Finally, while direct transfer was absent for streptavidin–biotin, we note that it may still occur for other small- molecule binders (12). Our findings suggest requisite properties for direct transfer and provide an initial framework for their quantitative boundaries, but they compel further studies with other protein–polynucleotide interactions to better define the biophysical requirements of direct transfer. Considerations for Methodologies to Assess Dissociation Rate Constants and/or Mean Lifetimes. Recent studies (6–9) of FD, where proteins compete one- another off gene targets through the same mechanism that polynucleotides compete one- another off proteins in direct transfer, have implications that are echoed for direct transfer due to their shared kinetic mechanism. Notably, methodologies that infer dissociation rates through “chase” of preformed complexes with an excess of competitor, such as the FP- based experiments used here (52, 53), should consider that the competitor may not be a neutral participant in ligand dissociation. Making measurements across competitor concentrations may confirm classic competition, but if direct transfer is instead revealed, then extrapolation is required to derive the intrinsic dissociation rate (Fig. 1D). Alternatively, methods that do not rely on competitors to prevent reassociation could be considered, such as surface plasmon resonance (SPR) (54). SPR removes dissociated ligand from the reaction via constant buffer flow across a binding surface, which should limit direct transfer. Like “chase” experiments, SM approaches that leave dissociated ligand in the reaction can have their measurements of mean complex lifetime skewed by direct transfer events causing premature complex dissociation, though the effects should be negligible under typical ligand concentrations (Fig. 2F). Approaches to determine the equilibrium dissociation constant (Kd) are unlikely to have their measurements affected by direct transfer, since ligands that prematurely dissociate during direct transfer events are likely replaced by another ligand, keeping the total concentration of complex constant. This has been discussed in greater depth for FD (6), and the conclusions are generally applicable. However, we note that the steric considerations of protein–protein competition on a ligand surface are not necessarily the same for ligand–ligand competition on a protein surface. Biological Relevance of Direct Transfer. Since the potential for direct transfer between polynucleotides might be an intrinsic property of many NBPs, it is natural to speculate on the potential for flux through a direct transfer versus classic pathway in vivo. Our Fig. 5 plot indicates that NBPs have an average kθ/k−1 ≈ 105 M−1, which suggests that flux through a direct transfer versus classic pathway becomes comparable around low- micromolar concentrations of free ligand. One context where this could be relevant is in a cell nucleus, where the concentration of RNA is ~1 mM in terms of ribonucleotides (~50 µM of 20- nt RNA segments) (55) and the DNA concentration is ~20 mM in terms of deoxyribonucleotides (~170 µM of 60- bp dsDNA segments) (56). Alternatively, this threshold could be achieved by modulating local effective molarity, for example via biological condensates or liquid–liquid phase- separated granules. Finally, for large NBP ligands (e.g., chromatin), it is possible that neighboring intramolecular DNA- binding sites could have high enough effective molarities, by virtue of their restricted spatial proximity, for intramolecular direct transfer to occur with comparable scale to classic dissociation (57). Similarly, nascent RNA transcripts proximal to NBP target genes could have sufficiently increased effective molarity for RNA–DNA direct transfer (13). In these cases, a noteworthy biological implication is that the mean lifetime of an NBP–ligand complex may not be a fixed value, determined by the dissociation rate constant, but may instead be modulated by the concentration of other ligands in its local environment. This concept of tunable protein turnover on binding sites has been previously discussed for direct transfer- like phenomena (6, 38). In the case of intramolecular direct transfer (e.g., on chromatin), such behavior could allow proteins to conduct a more efficient “search” for their target sites (2, 38, 40). Similarly, the case of RNA–DNA direct transfer could be a strategy for RNA- binding chromatin- associated proteins to be recruited to or evicted from their gene targets (13). While these implications are intriguing, it is premature to predict which instances of direct transfer will be biologically relevant. We acknowledge that just because direct transfer can be measured in vitro does not mean that it represents a mechanism integral to the protein’s function in vivo. In terms of both efficiency and magnitude, TREX1 exhibits average direct transfer (Fig. 5), but there is no readily appar- ent role for direct transfer in promoting TREX1’s degradation of DNA to prevent nucleic acid sensing (18). However, TREX1 is purportedly capable of “sliding” along DNA for efficient substrate binding and exonucleolytic degradation (24), and direct transfer could be an alternative or additional way to reduce the dimension- ality of an otherwise 3D search process (12, 35). At minimum, the observation that TREX1’s flexible loop can drastically modulate its direct transfer kinetics (Fig. 4 E and F) demonstrates the flexible loop’s critical contribution to TREX1 dsDNA–binding affinity, con- sistent with recent studies (24). In contrast to TREX1, our concur- rent studies of PRC2 (13) suggest that direct transfer may explain some of PRC2’s biological function: RNA- mediated regulation of PRC2’s histone methyltransferase activity on nucleosomes. Thus, direct transfer has an intuitive biological role for PRC2. Finally, what in vivo approaches might be used to test the bio- logical role of a protein’s direct transfer? Ideally, one would specif- ically ablate the capacity for a protein to perform direct transfer between two ligands without affecting their independent binding, presumably through mutation of the protein or polynucleotide sequences, and then compare phenotypes. However, since direct transfer appears to be mediated by protein contacts shared between ligand species (Fig. 1B), modulating direct transfer might neces- sarily perturb the ligands’ independent binding. In that case, the mutant phenotype could not be specifically attributed to direct transfer versus affinity perturbations. Similarly, since unique pro- tein contacts between competing ligands act as “footholds” that should affect direct transfer, trying to modify shared versus unshared protein contacts to achieve similar binding affinities with discrepant direct transfer kinetics could prove problematic. However, we look forward to future advances that may allow direct in vivo interrogation of direct transfer biology. PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 9 of 11 Materials and Methods Streptavidin was purchased commercially; other proteins were expressed in bacteria and purified via affinity chromatography (16, 27, 33, 58). Oligonucleotides were purchased from IDT (Coralville, IA). FP- based Kd experiments were performed by titrating protein with 5 nM ligand (no competitor); binding curves were regressed with standard (nonquadratic) and Hill equations. FPCD experiments (Fig. 1D) were performed by titrating competitor into preformed protein–ligand complex and then monitoring complex dissociation over time; apparent dissociation rates for each com- petitor concentration were determined via single exponential regression, off- rate versus competitor data were fit with custom regression models for classic and direct transfer to determine rate constants, and the superior model was identified via the Bayesian Information Criterion (BIC) (59). For SM- TIRF, slides were prepared by con- jugating FLAG- tagged TREX1 to PEG- biotin slides via biotin- labeled α- FLAG antibody and streptavidin (43). Single- label experiments were performed with 1 nM ligand and 0 or 10 µM competitor flowed onto the slide together. Dual- label experiments were performed with 1 or 10 nM of each ligand flowed onto the slide together and utilized alternating colocalization and FRET imaging conditions on the same slides and fields of view. Detailed equations, methodology, and theoretical background are provided in SI Appendix. Data, Materials, and Software Availability. GitHub hosts the FPalyze (60) and SMBalyze (61) R packages. The custom scripts referenced in these meth- ods are available on GitHub (62). For the SM experiments, raw movie files and SMBalyze output files have been uploaded to Zenodo (63), with an embargo that expires in October 2023. Our pMALcPP vector and pMALcPP/MS2- CP plasmids will be deposited to Addgene. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. W.O.H. was supported by the NIH (F32- GM147934). T.R.C. is an investigator of the Howard Hughes Medical Institute. We thank Otto Kletzien (Batey lab, University of Colorado Boulder) for the generous gift of recom- binant hnRNP- U protein and Chen Qiu (Hall lab, NIEHS) for the generous gift of FBF- 2 protein. We also thank Olke Uhlenbeck, Halley Steiner, Deborah Wuttke, and members of the Cech lab (University of Colorado Boulder) for stimulating discussion and feedback concerning these studies. Author affiliations: aDepartment of Biochemistry, University of Colorado Boulder, Boulder, CO 80309; bBioFrontiers Institute, University of Colorado Boulder, Boulder, CO 80309; and cHHMI, University of Colorado Boulder, Boulder, CO 80309 1. A. G. Kozlov, T. M. Lohman, Kinetic mechanism of direct transfer of Escherichia coli SSB tetramers between single- stranded DNA molecules. Biochemistry 41, 11611–11627 (2002). 2. M. G. Fried, D. M. Crothers, Kinetics and mechanism in the reaction of gene regulatory proteins with 3. 4. 5. 6. 7. 8. 9. DNA. J. Mol. Biol. 172, 263–282 (1984). J. P. Menetski, S. C. Kowalczykowski, Transfer of recA protein from one polynucleotide to another. Kinetic evidence for a ternary intermediate during the transfer reaction. J. Biol. Chem. 262, 2085–2092 (1987). T. Ruusala, D. M. Crothers, Sliding and intermolecular transfer of the lac repressor: Kinetic perturbation of a reaction intermediate by a distant DNA sequence. Proc. Natl. Acad. Sci. U.S.A. 89, 4903–4907 (1992). J. Roca, A. Santiago- Frangos, S. A. Woodson, Diversity of bacterial small RNAs drives competitive strategies for a mutual chaperone. Nat. Commun. 13, 2449 (2022). A. Erbaş, J. F. Marko, How do DNA- bound proteins leave their binding sites? The role of facilitated dissociation Curr. Opin. Chem. Biol. 53, 118–124 (2019). C. E. Sing, M. Olvera de la Cruz, J. F. Marko, Multiple- binding- site mechanism explains concentration- dependent unbinding rates of DNA- binding proteins. Nucleic Acids Res. 42, 3783–3791 (2014). J. S. Graham, R. C. Johnson, J. F. Marko, Concentration- dependent exchange accelerates turnover of proteins bound to double- stranded DNA. Nucleic Acids Res. 39, 2249–2259 (2011). R. I. Kamar et al., Facilitated dissociation of transcription factors from single DNA binding sites. Proc. Natl. Acad. Sci. U.S.A. 114, E3251–E3257 (2017). 27. C. Qiu, A crystal structure of a collaborative RNA regulatory complex reveals mechanisms to refine target specificity. Elife 8, e48968 (2019). 28. C. Merritt, G. Seydoux, The Puf RNA- binding proteins FBF- 1 and FBF- 2 inhibit the expression of synaptonemal complex proteins in germline stem cells. Development 137, 1787–1798 (2010). 29. D. S. Peabody, The RNA binding site of bacteriophage MS2 coat protein. EMBO J. 12, 595–600 (1993). 30. R. Koning et al., Visualization by cryo- electron microscopy of genomic RNA that binds to the protein capsid inside bacteriophage MS2. J. Mol. Biol. 332, 415–422 (2003). 31. H. E. Johansson, L. Liljas, O. C. Uhlenbeck, RNA recognition by the MS2 phage coat protein. Semin. Virol. 8, 176–185 (1997). 32. H. E. Johansson et al., A thermodynamic analysis of the sequence- specific binding of RNA by bacteriophage MS2 coat protein. Proc. Natl. Acad. Sci. U.S.A. 95, 9244–9249 (1998). 33. K. A. LeCuyer, L. S. Behlen, O. C. Uhlenbeck, Mutants of the bacteriophage MS2 coat protein that alter its cooperative binding to RNA. Biochemistry 34, 10600–10606 (1995). 34. K. A. LeCuyer, L. S. Behlen, O. C. Uhlenbeck, Mutagenesis of a stacking contact in the MS2 coat protein- RNA complex. EMBO J. 15, 6847–6853 (1996). 35. P. H. von Hippel, A. Revzin, C. A. Gross, A. C. Wang, “Interaction of lac repressor with non- specific DNA binding sites” in Protein- Ligand Interactions (Walter de Gruyter and Co, Berlin, 1975), pp. 270–288. 36. J. Iwahara, M. Zweckstetter, G. M. Clore, NMR structural and kinetic characterization of a homeodomain diffusing and hopping on nonspecific DNA. Proc. Natl. Acad. Sci. U.S.A. 103, 15062–15067 (2006). 10. R. D. Giuntoli et al., DNA- segment- facilitated dissociation of Fis and NHP6A from DNA detected via 37. M. Doucleff, G. M. Clore, Global jumping and domain- specific intersegment transfer between single- molecule mechanical response. J. Mol. Biol. 427, 3123–3136 (2015). 11. J. Rudolph, J. Mahadevan, P. Dyer, K. Luger, Poly(ADP- ribose) polymerase 1 searches DNA via a ‘monkey bar’ mechanism. Elife 7, e37818 (2018). DNA cognate sites of the multidomain transcription factor Oct- 1. Proc. Natl. Acad. Sci. U.S.A. 105, 13871–13876 (2008). 38. O. G. Berg, R. B. Winter, P. H. von Hippel, Diffusion- driven mechanisms of protein translocation on 12. J. L. Bresloff, D. M. Crothers, DNA- ethidium reaction kinetics: Demonstration of direct ligand transfer nucleic acids. 1. Models and theory. Biochemistry 20, 6929–6948 (1981). between DNA binding sites. J. Mol. Biol. 95, 103–123 (1975). 39. L. Mirny et al., How a protein searches for its site on DNA: The mechanism of facilitated diffusion. J. 13. W. O. Hemphill, R. Fenske, A. R. Gooding, T. R. Cech, PRC2 direct transfer from G- quadruplex RNA to dsDNA has implications for RNA- binding chromatin modifiers. Proceedings of the National Academy of Sciences [In Press] (2023) https://www.biorxiv.org/content/10.1101/2022.11.30.518601v1. 14. C. Davidovich, T. R. Cech, The recruitment of chromatin modifiers by long noncoding RNAs: Lessons from PRC2. RNA 21, 2007–2022 (2015). Phys. Math. Theor. 42, 434013 (2009). 40. S. E. Halford, J. F. Marko, How do site- specific DNA- binding proteins find their targets? Nucleic Acids Res. 32, 3040–3052 (2004). 41. N. Michael Green, “[5] Avidin and streptavidin” in Avidin- Biotin Technology, M. Wilchek, E. A. Bayer, Eds. (Methods in Enzymology, Academic Press, 1990), pp. 51–67. 15. M. Marenda, E. Lazarova, N. Gilbert, The role of SAF- A/hnRNP U in regulating chromatin structure. 42. U. de Silva, “Structural and biochemical studies of trex–three prime repair exonucleases,” PhD Curr. Opin. Genet. Dev. 72, 38–44 (2022). 16. O. Kletzien, “Promiscuous RNA binding by the tandem RGG/RG domains of hnRNP U,” PhD dissertation, University of Colorado at Boulder, Boulder, CO (2022). dissertation, Wake Forest University Graduate School of Arts and Sciences, Winston- Salem, NC USA (2007). 43. E. Ly, J. A. Goodrich, J. F. Kugel, Monitoring transcriptional activity by RNA polymerase II in vitro 17. X. Wang et al., Targeting of polycomb repressive complex 2 to RNA by short repeats of consecutive using single molecule co- localization. Methods 159–160, 45–50 (2019). guanines. Mol. Cell 65, 1056–1067.e5 (2017). 18. S. R. Simpson, W. O. Hemphill, T. Hudson, F. W. Perrino, TREX1–Apex predator of cytosolic DNA metabolism. DNA Repair 94, 102894 (2020). 19. W. O. Hemphill “Biochemical, structural, & computational studies of TREX1 exonuclease activity,” PhD dissertation, Wake Forest University Graduate School of Arts and Sciences, Winston- Salem, NC USA (2021). 20. D. J. Mazur, F. W. Perrino, Excision of 3′ termini by the Trex1 and TREX2 3′→5′ exonucleases characterization of the recombinant proteins. J. Biol. Chem. 276, 17022–17029 (2001). 21. A. Ablasser et al., TREX1 deficiency triggers cell- autonomous immunity in a cGAS- dependent 44. U. de Silva et al., The crystal structure of TREX1 explains the 3′ nucleotide specificity and reveals a polyproline II helix for protein partnering. J. Biol. Chem. 282, 10537–10543 (2007). 45. S. L. Bailey, S. Harvey, F. W. Perrino, T. Hollis, Defects in DNA degradation revealed in crystal structures of TREX1 exonuclease mutations linked to autoimmune disease. DNA Repair 11, 65–73 (2012). 46. G. A. Jones, D. S. Bradshaw, Resonance energy transfer: From fundamental theory to recent applications. Front. Phys. 7, 100 (2019). 47. E. Grahn et al., Structural basis of pyrimidine specificity in the MS2 RNA hairpin- coat- protein complex. RNA 7, 1616–1627 (2001). manner. J. Immunol. 192, 5993–5997 (2014). 48. J. L. Grieves et al., Exonuclease TREX1 degrades double- stranded DNA to prevent spontaneous 22. G. I. Rice, M. P. Rodero, Y. J. Crow, Human disease phenotypes associated with mutations in TREX1. J. lupus- like inflammatory disease. Proc. Natl. Acad. Sci. U.S.A. 112, 5117–5122 (2015). Clin. Immunol. 35, 235–243 (2015). 23. W. O. Hemphill et al., TREX1 as a novel immunotherapeutic target. Front. Immunol. 12, 660184 (2021). 24. W. O. Hemphill, T. Hollis, F. R. Salsbury, F. W. Perrino, TREX1’s homodimer structure has evolved for double- stranded DNA degradation. bioRxiv [Preprint] (2022). https://www.biorxiv.org/ content/10.1101/2022.02.25.481063v2. (Accessed 17 May 2023). 25. J. Maciejowski et al., APOBEC3- dependent kataegis and TREX1- driven chromothripsis during telomere crisis. Nat. Genet. 52, 884–890 (2020). 49. J. M. Fye, C. D. Orebaugh, S. R. Coffin, T. Hollis, F. W. Perrino, Dominant mutations of the TREX1 exonuclease gene in lupus and aicardi- goutières syndrome. J. Biol. Chem. 286, 32373–32382 (2011). 50. R. J. Schneider, J. G. Wetmur, Kinetics of transfer of Escherichia coli single strand DNA binding protein between single- stranded DNA molecules. Biochemistry 21, 608–615 (1982). 51. M. Sawadogo, Multiple forms of the human gene- specific transcription factor USF. II. DNA binding properties and transcriptional activity of the purified HeLa USF. J. Biol. Chem. 263, 11994–12001 (1988). 26. C. Qiu et al., Distinct RNA- binding modules in a single PUF protein cooperate to determine RNA 52. X. Wang et al., Molecular analysis of PRC2 recruitment to DNA in chromatin and its inhibition by specificity. Nucleic Acids Res. 47, 8770–8784 (2019). RNA. Nat. Struct. Mol. Biol. 24, 1028–1038 (2017). 10 of 11 https://doi.org/10.1073/pnas.2220537120 pnas.org 53. Y. Long et al., RNA is essential for PRC2 chromatin occupancy and function in human 58. W. O. Hemphill, F. W. Perrino, Measuring TREX1 and TREX2 exonuclease activities. Methods Enzymol. pluripotent stem cells. Nat. Genet. 52, 931–938 (2020), 10.1038/s41588- 020- 0662- x (10 August 2020). 54. P. Pattnaik, Surface plasmon resonance. Appl. Biochem. Biotechnol. 126, 79–92 (2005). 55. A. Khong, R. Parker, The landscape of eukaryotic mRNPs. RNA 26, 229–239 (2020). 56. J. F. Gillooly, A. Hein, R. Damiani, Nuclear DNA content varies with cell size across human cell types. Cold Spring Harb. Perspect. Biol. 7, a019091 (2015). 57. L. Ringrose, S. Chabanis, P.- O. Angrand, C. Woodroofe, A. F. Stewart, Quantitative comparison of DNA looping in vitro and in vivo: Chromatin increases effective DNA flexibility at short distances. EMBO J. 18, 6630–6641 (1999). 625, 109–133 (2019). 59. Y. Sakamoto, M. Ishiguro, G. Kitagawa, Akaike Information Criterion Statistics (D. Reidel Publishing Company, ed. 3, 1986). 60. W. Hemphill, FPalyze. Github. https://github.com/whemphil/FPalyze. Deposited 16 May 2023. 61. W. Hemphill, SMBalyze. Github. https://github.com/whemphil/SMBalyze. Deposited 16 May 2023. 62. W. Hemphill, Direct- Transfer_Manuscript. Github. https://github.com/whemphil/Direct- Transfer_ Manuscript. Deposited 16 May 2023. 63. W. Hemphill et al., TIRF microscopy data files. Zenodo. https://zenodo.org/record/7838123#. ZGaLZOzML0o. Deposited 30 November 2022. PNAS 2023 Vol. 120 No. 26 e2220537120 https://doi.org/10.1073/pnas.2220537120 11 of 11
10.1073_pnas.2221163120	RESEARCH ARTICLE \| EVOLUTION OPEN ACCESS Marginal specificity in protein interactions constrains evolution of a paralogous family Dia A. Ghosea , Kaitlyn E. Przydziala, Emily M. Mahoneya , and Michael T. Lauba,d,1 , Amy E. Keatinga,b,c Edited by William DeGrado, University of California San Francisco, San Francisco, CA; received December 15, 2022; accepted March 24, 2023 The evolution of novel functions in biology relies heavily on gene duplication and diver- gence, creating large paralogous protein families. Selective pressure to avoid detrimental cross-talk often results in paralogs that exhibit exquisite specificity for their interaction partners. But how robust or sensitive is this specificity to mutation? Here, using deep mutational scanning, we demonstrate that a paralogous family of bacterial signaling proteins exhibits marginal specificity, such that many individual substitutions give rise to substantial cross-talk between normally insulated pathways. Our results indicate that sequence space is locally crowded despite overall sparseness, and we provide evidence that this crowding has constrained the evolution of bacterial signaling proteins. These findings underscore how evolution selects for “good enough” rather than optimized phenotypes, leading to restrictions on the subsequent evolution of paralogs. protein evolution \| signal transduction \| paralogous proteins \| gene duplication \| protein-protein interactions The process of gene duplication and divergence fuels the evolution of proteins with new functions (1). This fundamental mechanism has created large paralogous protein families within all clades of life (2, 3). However, the expansion of these protein families presents a challenge when members are required to bind distinct interaction partners (4–8). Given their highly similar structures and sequences, how do the individual members of such families maintain different interaction specificities? And, do paralogs constrain each other’s evolution? Answers to these questions lie in the nature of the sequence space relevant to such paralogous families. This sequence space is defined by the set of residues governing the interaction specificity of paralogs and their binding partners. In sequence space, each paralog must reside within a specific “niche,” defined here as the set of sequences capable of interacting with its binding partner(s). A given paralog may also have to avoid the niches of other proteins within this space to maintain interaction specificity. How much constraint is posed by other paralogs depends on the size, distribution, and extent of overlap of niches within sequence space. Prior work demonstrated that the sequence space of some paralogous protein families is sparsely occupied, with ample room for new members, based on the observation that new, synthetic proteins could be readily discovered or introduced without cross-talk to existing systems (9–12). However, the overall distribution of niches for extant paralogs in sequence space is not known, and there are two general possibilities. First, niches could be widely distributed throughout sequence space. Due to either selection pressure (13–18) or the drift of sequences over evolutionary time, individual niches may have moved far apart in space. This would result in “robust specificity” in the sense that cross-talk between paralogs would require multiple substitutions (Fig. 1 A, Top). Alternatively, niches for extant paralogs could be clustered and partially overlapping (Fig. 1 A, Bottom), creating crowded local regions of sequence space despite overall sparsity. This could result in “marginal specificity” such that individual substitutions could, in principle, produce cross-talk. Such marginal specificity is akin to the well-documented marginal stability of proteins in which proteins are often just above a threshold stability level needed for folding (19–23). This marginal stability arises because evolution does not select for additional stability once a protein can stably fold. A consequence, or reflection, of marginal stability is that individual substitutions can lead to unfolding. Marginal specificity could similarly arise if recently duplicated paralogs are only under selective pressure to separate in sequence space just enough to prevent unwanted cross-talk, with no pressure to further diverge and enhance the robustness of specificity. To distinguish between these models of specificity, we investigated two-component sig- naling pathways, the most prevalent form of signal transduction system in bacteria, with most species encoding dozens of paralogous pathways (24). The typical pathway consists of a histidine kinase (HK) that detects a signal, autophosphorylates, and then transfers a phosphoryl group to a cooperonic, cognate response regulator (RR). The phosphorylated RR elicits a cellular output, often by regulating gene expression (25). The HK also Significance Large paralogous protein families are found throughout biology, the product of extensive gene duplication. To execute different functions inside cells, paralogs often acquire different specificities, interacting with desired, cognate partners and avoiding cross-talk with noncognate proteins. But how robust is this interaction specificity to mutation? Can individual mutations lead to cross-talk or do paralogs diverge enough to provide a mutational “buffer” against cross-talk? To address these questions in the context of a family of bacterial signaling proteins, we built mutant libraries that produce all single substitutions of the kinase EnvZ and then screened for cross-talk to noncognate proteins. Strikingly, we find that many substitutions can produce cross-talk, meaning that these pathways typically exhibit only “marginal specificity” and demonstrate that this restricts their evolvability. This manuscript was deposited as a biorxiv preprint at: https://doi.org/10.1101/2023.02.18.529082 (2023). Author contributions: D.A.G., A.E.K., and M.T.L. designed research; D.A.G., K.E.P., and E.M.M. performed research; D.A.G. contributed new reagents/analytic tools; D.A.G. analyzed data; and D.A.G., A.E.K., and M.T.L. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221163120/-/DCSupplemental. Published April 25, 2023. PNAS 2023 Vol. 120 No. 18 e2221163120 https://doi.org/10.1073/pnas.2221163120 1 of 10 dephosphorylates its cognate RR in the absence of signal (26). There is typically very high structural and sequence similarity between paralogs in the domains responsible for HK–RR interactions: the dimerization and histidine phosphotransfer (DHp) domain of the HK and the receiver domain of the RR (27). However, interactions between cognate HK–RR pairs are highly specific in vivo and in vitro, with little cross-talk between noncognate partners (28). HK–RR specificity is determined primarily by a subset of residues that strongly coevolve (Fig. 1B and SI Appendix, Fig. S1A) (29, 30). These residues are found at the HK–RR interface and, when col- lectively swapped from one system to another, are often sufficient to rewire interaction specificity (20, 29). Introducing multiple sub- stitutions at these key residues can produce cross-talk between noncognate proteins that is severely detrimental to cellular fitness in certain conditions (4). However, how likely individual substitu- tions are to produce cross-talk has not been systematically probed. Thus, whether two-component signaling pathways exhibit marginal or robust specificity is not yet known. Results A High-Throughput Method for Assessing Cross-Talk between Signaling Pathways. To assess paralog specificity and determine how crowded sequence space is locally, we focused on the Escherichia coli two-component signaling systems EnvZ–OmpR, RstB–RstA, and CpxA–CpxR (Fig. 1A). These three systems are widespread in β- and γ-proteobacteria, likely resulting from two ancient duplication and divergence events that occurred ~2 billion years ago in their common ancestor (31) (SI Appendix, Fig. S1 B and C). To examine the occupancy of sequence space immediately surrounding EnvZ, we sought to measure the ability of variants harboring each possible single substitution in the DHp domain to activate the cognate regulator OmpR and the noncognate regulators RstA and CpxR. To monitor activation, we generated strains in which a green fluorescent protein (GFP) reporter is expressed from a known OmpR-, RstA-, or CpxR-regulated promoter (32–34) (Fig. 1C and SI Appendix, Fig. S2A). Because the native signal for EnvZ is not known, we deleted envZ in each strain and introduced taz, which encodes a chimeric receptor containing the aspartate-sensing domain of the chemoreceptor Tar fused to the cytoplasmic, signaling domains of EnvZ (33, 35, 36) (Fig. 1C). Taz drives robust (~14-fold), signal-dependent induction of our OmpR reporter, but not the RstA or CpxR reporters, reflecting the limited cross-talk between the three wild- type pathways (SI Appendix, Fig. S2 B and C). For simplicity, we refer to the wild-type Taz construct as EnvZ. To assess cross- phosphorylation of the noncognate regulators RstA and CpxR, we deleted rstB and cpxA from each strain to prevent the phosphatase activity of these kinases from counteracting any phosphorylation of RstA or CpxR by EnvZ (33) (SI Appendix, Fig. S2A). A A e st spe ecificity robust specificity y robust speecificity C library: 1,140 EnvZ DHp substitutions (19 substitutions x 60 positions) RstA RstB x x x CpxR CpxA mmargim marginal specificity requires multiple mutations - cross-talk requires multiple mutations OmpR EnvZ RstB RstA CpxR CpxA OmpR EnvZ PompC 600 t n u o c 400 200 ... P gfp x + aspartate (ON) P Tar-EnvZ chimera (DHp mutant library) OmpR Pasr P gfp RstA P gfp CpxR PcpxP B HK RR 0 -2 -1 0 1 2 3 -2 -1 0 1 2 3 -2 -1 0 1 2 3 log (GFP fluorescence (a.u.)) - cross-talk via a single mutation s-talk via a single mutation + deep sequence sequence of a given HK set of HK sequences compatible with a given RR y c n e u q e r f variant 1 variant 1 variant 1 infer signaling from 1,140 EnvZ DHp variants to OmpR, RstA, CpxR GFP bin Fig. 1. Assessing the density of local sequence space. (A) Sequence space diagram in which rectangles represent the space of all possible HK sequences, dots represent extant paralogs in a species (HKs in E. coli), and gray spheres represent the set of HK sequences that interact with a given RR. HK paralogs must interact with their cognate RR, i.e., be within the niche of that RR, but avoid the niches of noncognate RRs. Top: robust specificity model, where niches are generally well separated in sequence space. Bottom: marginal specificity model, where niches are separated just enough to avoid cross-talk, but often overlap such that local sequence space can be crowded. Insets show the arrangement of niches for OmpR, RstA, and CpxR. In the robust specificity model, multiple mutations (depicted as arrows) to an HK such as EnvZ are required to introduce cross-talk; in the marginal specificity model, a single mutation may introduce cross-talk. (B) Model of an HK–RR complex. RR chains (light blue) from PDB ID 3DGE are positioned relative to the structure of EnvZ (5B1N, deep blue) using the DHp domains in each structure for alignment. HK positions that coevolve with positions in the RR are shown as red spheres. (C) A library of single-substitution EnvZ variants was transformed into three GFP reporter strains that read out activation of OmpR (PompC), RstA (Pasr), or CpxR (PcpxP). The resulting populations showed a distribution of GFP fluorescence and were sorted into eight bins based on GFP level. Populations from each bin were deep sequenced, the frequency of variants in each bin was calculated, and profiles were fit to Gaussians to extract the peak fluorescence of each variant. Error bars in frequency profiles indicate SD from three replicates. 2 of 10 https://doi.org/10.1073/pnas.2221163120 pnas.org To focus our investigation on the local sequence space imme- diately surrounding EnvZ, we performed deep mutational scan- ning (37, 38). We constructed a library of all 1,140 single substitutions in the 60 amino acid DHp domain of EnvZ and then used a high-throughput screening approach, Sort-seq (9), to measure interaction with OmpR, RstA, and CpxR (Fig. 1C). We transformed the library into each reporter strain and then grew cells in the presence or absence of signal (aspartate) for 3 h before sorting cells into bins based on their fluorescence. The plasmids encoding envZ in cells from each bin in each condition were deep sequenced to determine the frequency of each variant, with frequency profiles fit to a Gaussian to extract the peak flu- orescence values (9) (Fig. 1C and SI Appendix, Fig. S2 D–J). To validate these values, 30 variants spanning the range of output fluorescence values for each regulator in both conditions were measured individually using flow cytometry, and the median flu- orescence was compared to the values obtained from Sort-seq. For each reporter, there was a high correlation (R2 > 0.8) between the values obtained from Sort-seq and flow cytometry (Fig. 2A). We also purified a selection of 14 EnvZ variants and used 32P-based phosphotransfer assays to demonstrate that the activities toward the regulators seen in vivo were recapitulated in vitro (SI Appendix, Fig. S3). Deep Mutational Scanning Reveals the Marginal Specificity of Paralogs. To visualize our deep mutational scanning data for transfer to the cognate regulator OmpR, we generated a heatmap displaying the fluorescence levels in the presence of inducer (+signal) for each possible substitution at each position in the DHp domain of EnvZ (Fig. 2B). Most (81%) substitutions retained levels similar (within 5-fold) to wild-type EnvZ, indicating that they retain kinase activity (Fig. 2B). When visualizing fold-induction value (fluorescence +/− signal, Fig. 2C), 76% of substitutions eliminated or reduced the fold-induction relative to the wild-type (Fig. 2D). 101 100 A y r t e m o t y c w o l f B OmpR RstA CpxR 101 102 103 sort-seq r2=0.84 104 C - aspartate (OFF) + aspartate (ON) Tar-EnvZ chimera (DHp mutant library) P P PompC OmpR gfp P OmpR PompC gfp GFP α1 EnvZ α2 230 260 LADDRTLLMAGVSHDLRTPLTRIRLATEMMSEQDGYLAESINKDIEECNAIIEQFIDYLR 270 S LA EM 250 240 280 KD TR 290 D E P G A V I L M W F Y T S C Q N D E H K R P G A V I L M W F Y T S C Q N D E H K R P G A V I L M W F Y T S C Q N D E H K R = WT residue l o g ( f l u o r e s c e n c e + s g n a l ) i l o g ( f o d l i n d u c t i o n ) 0 -1 -2 -3 -4 3 2 1 0 l o g ( f l u o r e s c e n c e - s g n a l ) i 2 1 0 -1 Fig. 2. Sort-seq reveals the landscape of mutational tolerance in EnvZ-OmpR signaling. (A) Correlations between fluorescence values obtained by Sort-seq and by individual-clone flow cytometry. Error bars indicate SD from two independent biological replicates. Individual Pearson’s coefficients for the three reporters were r2 = 0.91 (OmpR), 0.92 (RstA), and 0.83 (CpxR). (B) Heatmap of OmpR reporter data with columns representing positions along the EnvZ DHp sequence; yellow highlights indicate coevolving residues. Rows indicate specific amino acids introduced at each position. Dots mark wild-type residues. Color-coded values represent log10(fluorescence) of each variant in the +signal condition. Wild-type EnvZ is set to white (blue represents increases in fluorescence, red shows decreases). (C) Diagram illustrating induction for the OmpR reporter. In low aspartate conditions, wild-type EnvZ is a phosphatase, removing phosphoryl groups from OmpR and leading to low GFP levels. In high aspartate conditions, wild-type EnvZ is a kinase, phosphorylating OmpR and driving high GFP production. (D) Same as (B) but with purple color indicating the log10(fold induction) value for each variant at each position. (E) Same as (B) but for the −signal condition. Wild-type EnvZ is set to white (blue represents increases in fluorescence, red represents decreases). Many variants have increased fluorescence in this condition suggesting that they have reduced phosphatase activity and are constitutively active. PNAS 2023 Vol. 120 No. 18 e2221163120 https://doi.org/10.1073/pnas.2221163120 3 of 10 A clear exception was within the loop region connecting the α1 and α2 helices of the DHp where a wide range of substitutions was tolerated. The loss of signal responsiveness for many variants may result from reduced phosphatase activity in the absence of a signal, producing a constitutively active kinase (Fig. 2E). To quantify cross-talk from each EnvZ variant to CpxR, we generated a heatmap showing the increase in fluorescence of the CpxR reporter in the +signal condition relative to that of wild-type EnvZ, which was set to 0 (Fig. 3A). Increases in fluo- rescence represent increased kinase activity, which could disrupt signaling fidelity and constitute detrimental cross-talk (Fig. 3A and SI Appendix, Fig. S4 A–C). No single substitution produced an EnvZ variant with signal-responsive activity toward CpxR, possibly due to reduced in vivo phosphatase activity. Although the majority of substitutions in EnvZ did not increase cross-talk to CpxR, a small number of substitutions showed fluorescence values increased as much as 30-fold relative to wild-type EnvZ. This level of activation was similar to that of a chimera of the Tar sensor domain fused to the cytoplasmic signaling domains of CpxA, the cognate HK of CpxR (SI Appendix, Fig. S4D). The cross-talk-inducing substitutions occurred primarily at the coev- olving positions previously shown to be important for HK–RR interaction specificity (29, 30). For instance, at Ala255, Glu257, and Asp273, multiple substitutions with dissimilar biochemical characteristics caused cross-talk, suggesting that the native resi- dues at these positions serve as negative design elements that prevent cross-talk to CpxR. At Ser269, only two substitutions, the positively charged residues Arg and Lys, caused substantial increases in cross-talk, suggesting that these residues may promote an interaction with CpxR that the wild-type Ser residue cannot. Notably, the corresponding residue of CpxA is Arg, consistent with positive charge at this position facilitating interaction with CpxR. At other coevolving positions, including Thr250, Arg251, Leu254, and Met258, no substitutions substantially increased α2 280 B 290 + aspartate (ON) A α1 EnvZ 240 250 260 270 230 P G A V I L M W F Y T S C Q N D E H K R EM EMM EnvZ LADDRTLLMAGVSHDLRTPLTRIRLATEMMSEQDGYLAESINKDIEECNAIIEQFIDYLR AL AL CpxA MMTSQQRLLSDISHELRTPLTRLQLGTALLRRRESKELERIETEAQRLDSMINDLLVMSR KD KD TEA TE TR TR TRL TR LA LA LG LG S S R R C P G A V I L M W F Y T S C Q N D E H K R EnvZ LADDRTLLMAGVSHDLRTPLTRIRLATEMMSEQDGYLAESINKDIEECNAIIEQFIDYLR RstB LIASKKQLIDGIAHELRTPLVRLRYRLEMSDNLSAAESQALNRDISQLEALIEELLTYAR RLAT LA YR RYRL LTR TR VR LVR EMM EM EM EM KDI KD RD RDI SI S AL AL EnvZ variant P P CpxR CpxR PcpxP gfp WT EnvZ P gfp PcpxP GFP D OmpR 1.50 1.25 1.00 0.75 0.50 0.25 0 l o g ( i n c r e a s e r e l . t o W T ) = WT residue RstA 699 203 3 29 21 CpxR WT EnvZ EnvZ library RstA reporter CpxR reporter PhoP reporter E 00 600 t n u o c 400 200 0 -2 -1 0 1 2 3 -2 -1 0 1 2 3 -2 -1 0 1 2 3 log (GFP fluorescence (a.u.)) Fig. 3. EnvZ exhibits marginal specificity, reflecting a crowded local sequence space. (A) Top: Heatmap of CpxR reporter data where values represent increases in GFP fluorescence relative to wild-type EnvZ in the +signal condition. Dots mark wild-type residues. EnvZ and CpxA DHp sequences are shown below. Yellow highlighted positions mark coevolving residues. Values were normalized by increases in fluorescence toward OmpR, which may reflect nonspecific effects of a substitution on expression level or kinase activity that increase activity toward all RRs. Bottom: Logo transformation of heatmap where the height of a letter represents the increase in fluorescence relative to wild-type for that amino acid substitution at that position. Letters are stacked in ranked order. (B) Diagram illustrating fluorescence measurements for CpxR reporter. Wild-type EnvZ shows low levels of kinase activity, leading to low GFP levels. EnvZ variants (as indicated by the star) may show increased kinase activity, driving high GFP production. (C) Same as (A) except for RstA reporter data. (D) Overlap of single-substitution EnvZ variants with activity toward different RRs. The OmpR set contains variants with kinase activity for OmpR comparable to wild-type EnvZ (within 5-fold). The RstA and CpxR sets contain variants with ≥5-fold increases in activity toward RstA or CpxR relative to wild-type EnvZ. (E) Histograms of GFP fluorescence distributions for the RstA, CpxR, and PhoP reporters. Red populations are for wild-type EnvZ, blue populations are transformations of the single mutant library. Blue library populations for RstA and CpxR reporters are replicated from Fig. 1C for comparison to PhoP. 4 of 10 https://doi.org/10.1073/pnas.2221163120 pnas.org cross-talk. These residues are each identical or biochemically sim- ilar in EnvZ and CpxA, consistent with them not being involved in insulating these two pathways. We also assessed cross-talk to RstA (Fig. 3C). The strongest cross-talk-inducing substitutions again tended to occur at the coevolving residues, with similar patterns seen as with CpxR. At some coevolving positions in which EnvZ differs significantly from the corresponding residue of RstB, such as Thr250, Ala255, and Ser269, multiple biochemically distinct substitutions led to sub- stantial cross-talk, indicating that these residues act as negative design elements with respect to RstA. At Leu254, only aromatic residues caused cross-talk suggesting that they form specific con- tacts with RstA that enhance its interaction with EnvZ; notably, RstB features a Tyr at this position. In a strikingly different pattern than we observed for CpxR, there were also a large number of substitutions at noncoevolving positions that caused cross-talk, which are discussed below. In total, there were 21, 206, and 29 substitutions that increased cross-talk more than 5-fold toward CpxR, RstA, or both, respec- tively (Fig. 3D). Similar patterns and relative numbers of variants were seen at thresholds of 3- and 10-fold, indicating that our results are robust to the precise threshold used (SI Appendix, Fig. S4 E–G). The observation that many individual substitutions can readily produce cross-talk indicates that EnvZ exhibits mar- ginal, rather than robust, specificity with respect to CpxR and RstA. Considering both CpxR and RstA, we found that EnvZ variants containing the corresponding residue of the respective cognate kinase (CpxA and RstB, respectively) were more likely to exhibit cross-talk relative to other substitutions (P = 0.004, odds ratio = 2.62, Fisher’s exact test; SI Appendix, Fig. S5A). However, there were still a large number of EnvZ substitutions that did not resemble the corresponding residue on the other kinases but still caused cross-talk. Together, these findings demonstrate that mutat- ing the EnvZ sequence to mimic RstB or CpxA is not the only way to generate cross-talk to RstA or CpxR (SI Appendix, Fig. S5B). We also found many substitutions that decreased the activity of EnvZ toward either or both noncognate regulators, without decreasing activity toward the cognate regulator OmpR (SI Appendix, Fig. S6 A–C). We likely observed this only because EnvZ is overexpressed in our assay; at native levels, EnvZ shows no detectable activity toward RstA and CpxR (SI Appendix, Fig. S6 D and E). However, this finding does suggest that cross-talk between these pathways has not been eliminated and instead has only been reduced to such a level that it has no effect on fitness. The notion that interactions with noncognate proteins could be reduced further by many different single substitutions emphasizes that only marginal specificity has been selected for between these systems. We hypothesized that the marginal specificity of EnvZ–OmpR, RstBA, and CpxAR reflects their phylogenetic history as closely related paralogs. Duplication events that led to the emergence of these three systems were likely followed by changes in specificity sufficient to insulate these pathways, such that they could carry out distinct functions, but leaving them close in sequence space. In contrast, less closely related pathways are likely further apart in sequence space such that specificity is more robust. To test this idea, we transformed the library of EnvZ variants into a reporter strain for the more distantly related regulator PhoP, for which the cognate kinase is PhoQ (SI Appendix, Fig. S1C), and performed flow cytometry. Unlike with the RstA and CpxR reporter strains, there was no subpopulation of cells showing significantly increased fluorescence relative to wild-type EnvZ, indicating that no or very few single substitutions in EnvZ cause substantial cross-talk to the distantly related PhoP (Fig. 3E). The Extent of Marginal Specificity Reflects the Evolutionary History of Paralogs. Collectively, our results suggest that the occupancy of sequence space reflects the evolutionary history of paralogs and that cross-talk is most likely to occur between more closely related systems. Because EnvZ–OmpR is more closely related to RstBA than to CpxAR (SI Appendix, Fig. S1C), this model may explain why many more single substitutions can cross-talk to RstA than CpxR (Fig. 3C). To assess the spatial distribution of substitutions that induced cross-talk, we mapped these substitutions onto a modeled structure of the EnvZ–OmpR complex in the phosphatase state, i.e., the state in which an HK is thought to promote dephosphorylation of its phosphorylated cognate RR (Fig. 4A). This is the state most commonly seen in two-component complex structures (39–41), likely due to its rigidity facilitating crystallization. Substitutions that increased cross-talk to CpxR in our assay were generally on the surface of the HK dimer at the interface with the RR, which we refer to here as the primary interface. As noted above, these cross-talking substitutions largely involved residues known to coevolve in HK–RR pairs (29, 30) (Fig. 3A). In contrast, the substitutions that caused cross-talk to RstA mapped all over the DHp domain, including at positions distal to the interface and even some within the core of the dimer (Fig. 4A). We also modeled EnvZ–OmpR in the suspected kinase, or phosphotransfer, state, using the single complex that has been solved (42). In this structure, the N- and C-terminal portions of the DHp domain, which form the upper part of the dimeric four-helix bundle, reside closer to the β4–α4 loop of the RR (Fig. 4B). Interactions at this secondary interface may explain the cross-talk behavior of some substitutions distal to the primary interface of both states. For example, the substitution D286V in EnvZ caused a ~50-fold decrease in activity toward OmpR but a ~5-fold increase in activity toward RstA (Fig. 4C). Examining EnvZ–OmpR modeled onto the kinase-state structure, Asp286 is positioned such that it can form a salt bridge with Lys83 on the β4–α4 loop of OmpR (Fig. 4B). Substituting this Asp with a Val eliminates this favorable interaction, which likely explains the decreased transfer from EnvZ(D286V) to OmpR. The residue corresponding to Lys83 in RstA is Leu (L79), possibly explaining why a hydrophobic Val in place of Asp286 in EnvZ is more favora- ble for this interaction than the charged Asp (Fig. 4B). Although additional interactions at the secondary interface may explain some of our results, they are unlikely to explain the effects of cross-talk–inducing substitutions within the DHp core. Such substitutions presumably impact the primary or secondary inter- face allosterically to enhance interaction with the noncognate RstA. Importantly, almost no substitutions in the DHp core or at positions away from the primary interface created cross-talk with the less closely related CpxR (Fig. 4A). Thus, we hypothesized that the primary interfaces of EnvZ–OmpR and RstB–RstA have diverged enough to reduce cross-talk between wild-type EnvZ and RstA but are still sufficiently compatible that a single substitution at a distal site can result in cross-talk. By contrast, the primary interface of CpxA–CpxR is far enough diverged from EnvZ– OmpR that only substitutions at this interface can yield large enough effects to produce cross-talk (Fig. 4D). This hypothesis predicts that increasing the compatibility of the primary interface between EnvZ and CpxR should increase the propensity of distal substitutions in EnvZ to cause cross-talk (Fig. 4E). To test this idea, we sought to substitute interfacial residues in CpxR with those found in OmpR and measure whether there is epistasis between these substitutions and distal substitu- tions in EnvZ. We focused on two coevolving interface positions at the primary interface of CpxR that differ significantly from the PNAS 2023 Vol. 120 No. 18 e2221163120 https://doi.org/10.1073/pnas.2221163120 5 of 10 A CpxR RstA 0 E F B HK RR D286 K83 C EnvZ(D286V) mpR O RstA CpxR V286 L79 WT OmpR D286V RstA 10 1 0.1 P F G n i e g n a h c l d o f ) Z v n E T W o t e v i t l a e r ( D wild-type EnvZ 1o interface variant 2o interface variant RstA OmpR CpxR G distal substitution substitutions causing cross-talk 12 distal substitution CpxR EnvZ OmpR-like substitution RstA EnvZ CpxR-like substitution distal EnvZ substitutions H distal EnvZ substitutions R234W N278L F284T Y287V R234W N278L F284T Y287V 10 P F G n i e g n a h c l d o f ) Z v n E T W o t e v i t a e r ( l 1 0.5 P F G n i e g n a h c l d o f ) Z v n E T W o t e v i t a e r ( l 10 1 0.7 CpxR CpxR(E22R) CpxR(L23Y) RstA RstA(A22E) RstA(Y23L) Fig. 4. The degree of marginal specificity reflects phylogenetic relatedness between paralogs. (A) Number of substitutions causing ≥5-fold cross-talk to either CpxR (green) or RstA (orange) at each position in EnvZ is color-coded and mapped onto the model complex structure in the phosphatase state. (B) Active phosphotransfer state structure with HK in deep blue and RR in light blue (PDB: 5IUL). Insets show wild-type EnvZ and OmpR, or EnvZ(D286V) and RstA residues modeled onto this structure. Side chains were placed in the most preferred rotamers for interaction. (C) Fold change in fluorescence +signal for EnvZ(D286V) relative to wild-type EnvZ for each reporter strain. n = 3 biological replicates. (D) Model in which faded protein pairs represent unsuccessful interaction and full-color pairs represent successful interaction. RstA and CpxR have diverged sufficiently from OmpR to prevent cross-talk with wild-type EnvZ. However, RstA retains enough compatibility that substitutions at either the primary or secondary interface (yellow stars) can produce cross-talk. By contrast, CpxR has diverged such that its primary interface is fundamentally incompatible with EnvZ, and only substitutions that suppress incompatibility at this interface create cross-talk. (E) Model in which distal substitutions do not create cross-talk to wild-type CpxR but substitutions at the CpxR primary interface that increase its compatibility with EnvZ can increase sensitivity to distal substitutions. (F) Fold changes in +signal fluorescence relative to wild-type EnvZ for four distal single substitutions in EnvZ, against wild-type CpxR and CpxR variants with OmpR-like primary interface substitutions: E22R and L23Y. n = 3 biological replicates. (G) Model in which distal substitutions in EnvZ can produce cross-talk to RstA, but substitutions at the RstA primary interface that reduce its compatibility with EnvZ decrease its sensitivity to these distal substitutions. (H) Fold changes in +signal fluorescence relative to wild-type EnvZ for four distal single substitutions in EnvZ, against wild-type RstA and RstA variants with CpxR-like primary interface substitutions: A22E and Y23L. n = 3 biological replicates. corresponding residues in both OmpR and RstA: Glu22 and Leu23 (SI Appendix, Fig. S7A). We substituted each of these res- idues individually with the corresponding residue found in OmpR, making CpxR variants E22R and L23Y, and then tested these variants for interaction with wild-type EnvZ and a selection of EnvZ variants with substitutions (R234W, N278L, F284T, and Y287V) that are distal to the primary interface and caused cross-talk to RstA but not CpxR (SI Appendix, Fig. S7 B and C). For each EnvZ variant, there was significantly more cross-talk to the CpxR primary interface variants than to wild-type CpxR (Fig. 4F and SI Appendix, Fig. S7D). This positive epistasis between interfacial residues of CpxR and interface-distal residues 6 of 10 https://doi.org/10.1073/pnas.2221163120 pnas.org of EnvZ suggests that improved compatibility between EnvZ and CpxR at the primary interface increases the susceptibility of CpxR to cross-talk induced by single substitutions elsewhere in EnvZ. Our model also predicts that decreasing the compatibility of the primary interface between EnvZ and RstA could have the opposite effect, decreasing the propensity of interface-distal sub- stitutions in EnvZ to cause cross-talk (Fig. 4G). To test this pre- diction, we substituted the same primary interface positions in RstA to match those of CpxR, A22E, and Y23L, and then tested these RstA variants for interaction with wild-type EnvZ and the same EnvZ variants as above. In each case, cross-talk to RstA caused by distal substitutions in EnvZ was significantly reduced for both RstA A22E and Y23L (Fig. 4H and SI Appendix, Fig. S7 E and F). This negative epistasis between interfacial residues of RstA and interface-distal residues of EnvZ further supports our model that the propensity for cross-talk between noncognate pro- teins depends on latent compatibility between the proteins at the primary interface, which reflects the evolutionary history of the paralogs. Avoiding Cross-Talk Is a Pervasive Selective Pressure Shaping Sequence Space Occupancy. We conclude that EnvZ exhibits marginal specificity, reflecting a crowded local region of sequence space. This further suggests that the specificity of extant two- component signaling paralogs can be fragile, easily disrupted by single substitutions throughout the kinase. We sought to assess whether this marginal specificity has broadly affected EnvZ evolution. First, we used HMMER to identify and align a set of 5,751 EnvZ orthologs from a wide range of proteobacteria. We then calculated the frequencies at which residues that caused cross-talk from E. coli EnvZ to either CpxR or RstA appear at the equivalent position in EnvZ orthologs from other species that also have RstBA and CpxAR. These cross-talk-inducing residues were found less frequently than residues that do not cause cross-talk (P = 1.2 × 10−4, D = 0.15, Kolmogorov–Smirnov test, Fig. 5A). We also found that a higher proportion of cross-talk–inducing residues were completely absent at the equivalent positions in EnvZ orthologs (P = 6.2 × 10−4, odds ratio = 0.15, Fisher’s exact test, SI Appendix, Fig. S8A). These observations suggest that even averaged across a large number of sequence backgrounds, the substitutions that we found to cause cross-talk may have been selected against, leading to their lower prevalence among EnvZ orthologs. To test whether the differences seen were specific to EnvZ, we aligned ortholog sequences of three other HKs, PhoR, YehU, and BarA, which are increasingly distantly related to EnvZ (SI Appendix, Fig. S1B). For YehU and BarA, there was no significant difference between the frequencies of the two classes of residues at the equiv- alent positions (P = 0.80, D = 0.043 for YehU, P = 0.18, D = 0.074 for BarA, Kolmogorov–Smirnov test, SI Appendix, Fig. S8 B and C) or the proportion of residues that were absent (P = 0.68, odds ratio = 0.94 for YehU, P = 0.17, odds ratio = 0.82 for BarA, Fisher’s exact test, SI Appendix, Fig. S8 D and E). For PhoR, there was a significant difference between the two categories, although it was smaller than the difference observed for EnvZ (P = 0.0016, D = 0.13, Kolmogorov–Smirnov test, P = 0.003, odds ratio = 0.12, Fisher’s exact test, SI Appendix, Fig. S8 F and G). These results suggest that more closely related kinases share some of the same sequence features and selective pressures faced by EnvZ, but these pressures do not apply to more distantly related kinases. Although many γ-proteobacteria, like E. coli, have EnvZ– OmpR, RstBA, and CpxAR, several species have lost one or more of these systems (Fig. 5B). We wondered if losing RstBA or CpxAR relaxes the selection pressure against cross-talking mutations and allows drift of EnvZ into the regions of sequence space they pre- viously occupied. Indeed, we found that cross-talk–inducing sub- stitutions are seen at higher frequencies in EnvZ orthologs from species that have lost RstBA or CpxAR (P = 0.048, D = 0.11, Kolmogorov–Smirnov test, SI Appendix, Fig. S8H). Additionally, we found that species that have lost RstBA and CpxAR were more likely to have duplicated EnvZ–OmpR (P = 0.025, odds ratio = 0.77 for CpxAR loss, P = 1.5 × 10−6, odds ratio = 0.49 for RstBA loss, Fisher’s exact test, Fig. 5C). This finding suggests that the presence of these systems, particularly the most closely related RstBA, limits the sequence space available to EnvZ and thus con- strains the ability of EnvZ–OmpR to duplicate and establish a new system (Fig. 5D). We further predicted that in species lacking RstBA and CpxAR in which EnvZ–OmpR had duplicated, the EnvZ par- alogs may now occupy sequence space made available by the loss of the other systems (Fig. 5D). To test this prediction, we chose four species, distantly related to each other and to E. coli, in which RstBA and CpxAR were independently lost and EnvZ had been duplicated (Fig. 5B). Each of the resulting EnvZ homologs had residues that caused cross-talk in the context of E. coli EnvZ (SI Appendix, Fig. S8I). These residues occurred at both primary and secondary interface positions, as well as in core residues of the DHp domain. We cloned and expressed each homolog in our reporter strains for E. coli OmpR, RstA, and CpxR and then measured GFP expression relative to that seen with E. coli EnvZ. For each species considered, one or both EnvZ homologs showed high levels of cross-talk to E. coli RstA, CpxR, or both (Fig. 5E). This result strongly supports the notion that without RstBA and CpxAR, EnvZ duplicates commonly enter the sequence space freed up by the loss of these other systems. This finding further demonstrates how the presence of closely related paralogs, and the consequent marginal specificity, has constrained EnvZ evolution. Discussion Our findings demonstrate that the distribution of niches in sequence space of paralogous two-component signaling systems is not globally optimized for specificity or selected for robustness to mutation. Although the requirement for only a marginal level of specificity during the early establishment of duplicates may facilitate their evolution, it comes at the cost of future constraint on the emergence of additional duplications. Over time, due to drift and movement catalyzed by subsequent duplications, paral- ogous systems can continue to move apart in sequence space such that more distantly related systems are robustly insulated. However, systems like EnvZ–OmpR, CpxA–CpxR, and RstB–RstA, with ~2 billion years of divergence continue to exhibit marginal spec- ificity. Thus, this drift is likely slower than the rate at which addi- tional duplication events occur, such that the marginal specificity of existing paralogs will constrain the evolution of new duplicates when they emerge. Examples of single substitutions in proteins creating nonspecific interactions have also been found in other unrelated paralogous families, in both bacteria and eukaryotes (5, 7, 43). These anec- dotal examples suggest that the principle of marginal specificity may generally apply during evolution. In many families, paralogs may share functional binding partners and not require all inter- actions to be fully specific. However, the marginal specificity prin- ciple could apply in any case where there is a cost incurred by a nonspecific interaction. Such cases are likely to occur between most paralogs whose functions are nonredundant and involve protein–protein interactions. PNAS 2023 Vol. 120 No. 18 e2221163120 https://doi.org/10.1073/pnas.2221163120 7 of 10 A s t n u o c 140 120 100 80 60 40 20 0 D p = 1.2 x 10-4 B present absent EnvZ-OmpR duplication CpxRA RstAB sp272627 sp279238 sp266809 sp1860096 sp1134510 sp441620 sp155892 sp570156 sp43658 sp1348114 sp1763536 sp1265503 sp265726 sp670 sp675816 sp990268 sp80852 sp1109412 sp1197719 sp511145 sp1691903 sp1076550 sp1560201 sp1628855 sp1288385 sp595494 sp1331007 sp1113895 sp717774 sp1561206 sp400668 sp1123519 sp1294143 sp1856685 sp1148509 sp1960827 sp1111735 sp1632857 sp1797492 sp1121035 sp1304883 sp511 sp1851544 sp762376 sp1121480 sp1736471 sp1707785 sp342113 sp1770053 sp1267562 Caulobacter crescentus Escherichia coli Tolumonas auensis Betaproteobacterium sp. Burkholderia sp. JS23 (cid:31) (cid:30) (cid:29) p = 1.5 x 10-6 p = 0.025 C i s e c e p s f o n o i t c a r f 1.0 0.8 0.6 0.4 0.2 0 no RstBA/ CpxAR only CpxAR only RstBA duplication no duplication no cross-talk cross-talk 0.1 scenario A CCpxCpxpxRRRRC RstRstRstRstAAARstA CCpxRCCC RstARstAststAR tA CpxA RstB loss of RstBA and CpxAR RstA CCpxCpxpxRRRRC RstRstRstRstAAARststA CpxR EnvZ OmpR EnvZ OmpR CCpxCpxpxRRRRC RsstARststststAAAAAAtAAARstRstA RsRstA OmpR2 EnvZ OmpR EnvZ OmpR CCpxpxpxRRRRC tAsts AAAAsttttAAAAAAAsts A RstARst OmpR2 CCpxCpxpxRRRRC tAsts AAAAsttttAAAAAAAsts A RstARst OmpR2 OmpR1 OmpR1 EnvZ2 E Z1 EnvZ1 EnvZ1 scenario B BB OmpR1 OmpR1 EnvZ2 EnvZ1 EnvZ1 EnvZ1 OmpR1 OmpR1 EnvZ2 En Z1 EnvZ1 EnvZ1 Caulobacter EnvZ1 Caulobacter EnvZ2 Burkholderia EnvZ1 um sp Betaproteobacterium sp. EnvZ1 ToTT lumonas Tolumonas EnvZ1 Tolumonas EnvZ2 E P F G n i e g n a h c l d o f ) Z v n E i l o c . E o t e v i t a e r ( l 100 10 1 0.2 OmpR RstA CpxR Fig. 5. A densely occupied local sequence space constrains the evolution of two-component systems. (A) Violin plots show the distributions of counts of single substitutions found at the equivalent position in 1,019 EnvZ orthologs from species that also have RstBA and CpxAR. Counts are shown for two categories of substitution: those which do produce cross-talk to either RstA or CpxR and those which do not (see Fig. 3, P = 1.2 × 10−4, Kolmogorov–Smirnov test). The inner box shows the quartiles and the whiskers show the range except for outliers. (B) Tree of a subset of species indicating whether they have RstBA and CpxAR and whether they have EnvZ–OmpR duplications. Closed and open squares indicate presence and absence, respectively. (C) Fraction of species that either do not have RstBA or CpxAR, have only CpxAR, or have only RstBA, that have duplications of EnvZ–OmpR (P = 0.025 for CpxAR loss, P = 1.5 × 10−6 for RstBA loss, Fisher’s exact test). (D) Sequence space diagram illustrating how species that have lost RstBA and CpxAR may relax the selection pressure against entering regions of sequence space that would previously have resulted in cross-talk, also freeing up sequence space for EnvZ–OmpR duplications. Blue arrow indicates a single mutation in EnvZ that would be tolerated only in species that have lost CpxR and RstA. This could occur before duplication (scenario A) or after (scenario B). (E) Fold changes in fluorescence relative to E. coli EnvZ for six EnvZ homologs from four distantly related species, for E. coli OmpR, RstA, and CpxR reporters. n = 3 biological replicates. Our results demonstrate how the nature of evolution, in only selecting for “good enough,” rather than fully optimized systems, can result in small margins in specificity and constrain the subse- quent evolvability of a paralogous family. This principle also applies to protein stability (21, 44), abundance (45), and localization and assembly properties (46). In each case, a large proportion of sub- stitutions can disrupt the relevant property, suggesting that robust- ness has not evolved in these traits. The fragility of these properties has important implications for disease pathogenesis. For example, single substitutions that disrupt the stability and abundance of tumor suppressor proteins are implicated in cancer (45), and single substitutions that affect assembly properties of proteins can drive hemoglobinopathies such as sickle cell anemia. The same appears to be true of specificity, where single “network-attacking” substi- tutions that alter the specificity profile of human kinases are thought to disrupt cellular signaling networks and contribute to cancer progression (47). In addition to shedding light on the fundamental mechanisms of evolution and their consequences for paralogous proteins, our findings also have implications for protein design and directed evolution methods. Attempts to build new signaling systems while avoiding detrimental cross-talk with existing cellular systems may 8 of 10 https://doi.org/10.1073/pnas.2221163120 pnas.org work better if employing randomization or mutagenesis of mul- tiple residues, allowing “jumps” into sparsely occupied regions of sequence space, rather than methods that traverse crowded local sequence spaces by moving one mutation at a time. Overall, we demonstrate an example of marginal specificity in protein inter- actions that has implications for the evolvability of paralogous proteins, in both natural and synthetic settings. Methods Bacterial Strains and Media. E. coli strains were grown in M9 medium (1× M9 salts, 100 μM CaCl2, 0.2% glucose, 2 mM MgSO4, with or without 5 mM aspartate). When indicated, antibiotics were used at the following concentra- tions: carbenicillin 50 μg/mL, kanamycin 50 μg/mL, spectinomycin 50 μg/mL, and chloramphenicol 32 μg/mL. The base strain for all studies was E. coli strain ML1803 (Yale BW28357 ΔenvZ, SI Appendix, Table S1) (32). The OmpR reporter strain contained a p15a/cmR plasmid containing PompC-gfp (32) (SI Appendix, Table S2 and Dataset S1). The RstA, CpxR, and PhoP reporter strains each contained additional deletions: ΔackA-pta (removes a pathway that generates acetyl phos- phate, which can phosphorylate RRs in the absence of a HK) and ΔrstB, ΔcpxA, or ΔphoQ (to prevent the cognate HKs of RstA, CpxR, or PhoP, respectively, from phosphorylating or dephosphorylating them, SI Appendix, Table S1), and the same reporter plasmid but with Pasr-gfp, PcpxP-gfp, or PmgrB-gfp, respectively (SI Appendix, Table S2 and Datasets S2–S4). All libraries were cloned onto a low-copy pSC101/specR plasmid in which Taz variant expression was driven by a constitutive Plpp promoter (SI Appendix, Table S2 and Dataset S5). Characterization of individual, specific Taz variants was done using the same plasmid. EnvZ point mutations or homolog sequences were introduced using Gibson assembly using primers DG001-066 (Dataset S6). For the experiments involving mutations in the RRs, genomic mutations were made. Deletions discussed above and genomic mutations were made using sacB-kanR cloning. The loci in the relevant reporter strains were first replaced with the sacB- kanR locus using recombination (48), and selected using kanamycin resistance. The sacB-kanR loci were then replaced using DNA fragments that corresponded to either the sequence of clean deletions, or genes with the relevant mutations, and selected using negative selection on sucrose. The relevant region of the genome was ampli- fied by PCR and sequenced to confirm that the deletions/mutations were correct. Flow Cytometry Characterization. To induce Taz, cells were grown to early exponential phase [optical density at 600 nm (OD600) of about 0.2] in M9 before adding aspartate to a final concentration of 5 mM. Cells were grown for 3 h and diluted 1:40 into phosphate buffered saline (PBS) with 0.5 g/L kanamycin, and fluorescence was measured on a Miltenyi MACSQuant VYB. In each cytometry experiment, three colonies of each strain were grown and induced independently and 30,000 cells were measured per replicate. FlowJo was used to analyze the data, gating on single live cells and extracting the median of the GFP distribution. Design and Assembly of the Taz Library. A comprehensive single-mutant library was constructed using oligonucleotide-directed mutagenesis of the EnvZ DHp domain. To mutate each position in the Taz DHp (positions 230 to 289), two complementary 30-nucleotide primers (one sense, one antisense) were synthe- sized that introduce an NNS codon at the targeted position (primers DG067-186, Dataset S6). N is a mixture of A, T, C, and G, and S is a mixture of G and C. This mutagenesis strategy results in 32 possible codons, which cover all 20 amino acids. One round of PCR was carried out with one reaction containing the anti- sense NNS primer and DG187, a primer located 105 bp upstream of the 5′ end of the DHp domain, containing a SacI restriction site, and a second reaction con- taining the sense NNS primer and DG188, a primer flanking the 3′ end of the taz gene, containing a SalI restriction site. A second PCR round using both first-round products and both flanking primers produced the full-length double-stranded product. All reactions yielded a band of the correct size on an agarose gel, which was extracted and purified (Zymo). PCR product concentrations were quantified (NanoDrop), pooled in equimolar ratios, digested with SacI and SalI, and ligated into the pSC101/specR expression vector. Each ligation was dialyzed on Millipore VSWP 0.025-μm membrane filters for 60 min and the entire volume was elec- troporated into 20 μL Invitrogen One Shot TOP10 Electrocomp E. coli to yield ~106 total transformants. Plasmids from these transformants were then purified by miniprep (Zymo), dialyzed, and electroporated into the experimental strains, yielding ~109 transformants for each strain. Sort-seq. For each of three replicates, 1 mL of overnight culture of the library was washed with M9 and inoculated into 50 mL of M9. Cells were grown to OD600 = 0.2, and each culture was split into two: Aspartate was added to a final concentration of 5 mM to one of the cultures. After 3 h, cells were diluted (1:30) into PBS containing 320 μg/mL chloramphenicol and cells were placed on ice for sorting. Cells were sorted into bins based on GFP expression on a BD FACS (flourescence activated cell sorting) Aria machine. Single live cells were isolated using the gating strategy in SI Appendix, Fig. S2J. The FITC (fluorescein isothiocyanate) voltage was adjusted so that the population spanned the range of fluorescence the machine could detect. A live histogram of FITC fluorescence was drawn and gates were spaced evenly along the log10(GFP) axis. For each library replicate, both the on and off cultures were sorted into eight separate bins, generating 48 total bins. Up to ~2 million cells were sorted into bins per replicate (SI Appendix, Fig. S2 G–I). Sorted cells were added to 2× YT medium containing chloramphenicol and spectinomycin and then grown overnight. There is a possibility of enrichment or deenrichment of variants during the overnight growth due to differences in fitness; however, we have not observed such differences during growth in rich media previously. In addition, variants would be expected to be enriched or deenriched at proportionally the same rates in each bin sorted. Illumina Sample Preparation. After FACS, plasmids were purified (Zymo) from overnight cultures representing each bin from each library replicate. Two PCR reac- tions were performed, both using KAPA HiFi, to add Illumina sequencing adaptors and barcodes. First, DHp domain sequences were amplified for 12 cycles (95 °C for 30 s, 68 °C for 15 s, and 72 °C for 30 s) with Illumina inner amplification primers (primers DG189-198, Dataset S6). Second, purified PCR product from the first reac- tion was amplified in a second PCR with barcoding primers (primers DG199-224, Dataset S6) for nine cycles (95 °C for 30 s, 68 °C for 15 s, and 72 °C for 30 s). Final products were quantified (NanoDrop), normalized, combined, and sequenced on an Illumina NextSeq. For each bin, 1 to 2 million reads were collected. Illumina Data Processing. Sort-seq data processing was carried out as previ- ously described (9). The frequency of each variant in each bin was calculated by taking the fraction of reads in a given bin that corresponded to a given sequence, normalized by the fraction of cells in that given bin. For each variant, the mean frequencies in each bin across three replicates and SD were used to fit Gaussian functions to each distribution [in log10(GFP units)], from both the on and off sorts (SciPy optimize package). Fold-induction values were calculated as the ratio of the means between the induced and uninduced states: μon/μoff. To assess cross-talk to CpxR and RstA, fluorescence values in the presence of inducer were normalized against increases in fluorescence toward OmpR, which may reflect nonspecific effects of a substitution on expression level or kinase activity that increase activity toward all three RRs. Gaussian fit means for each variant for each reporter can be found in Dataset S7 (OmpR), Dataset S8 (RstA), and Dataset S9 (CpxR). Purification of Two-Component Signaling Proteins and In Vitro Phospho- transfer Assays. Expression and purification of EnvZ variants and RRs, and phos- photransfer experiments, were carried out as previously described (29, 30, 49). RRs were fused to a His6–Trx tag, and the cytoplasmic region of EnvZ (residues 222 to 451) was fused to a His6–MBP (maltose-binding protein) tag, expressed in BL21(DE3) cells and purified on a Ni2+-NTA column. For phosphotransfer reac- tions, the kinase was autophosphorylated for 1 h at 30 °C with [γ-32P]ATP (Perkin Elmer) before being combined with RRs at a 1:4 ratio (10-μL reactions contained 1 μM EnvZ and 4 μM RR). Reactions were stopped at the times noted by adding 4× Laemmli buffer with 8% 2-mercaptoethanol. The control lane had this solution added after the autophosphorylation but prior to addition of RR. HKs and RRs were separated by SDS-PAGE, gels were incubated with phosphor screens and imaged using a Typhoon imager (GE Healthcare) at 50-mm resolution. A representative image of two independent experiments is shown in SI Appendix, Fig. S3. Identification and Assembly of Ortholog Sequences and Trees. An align- ment of HKs was built by constructing a Hidden Markov Model profile from an alignment of DHp and CA domains of E. coli HKs (29) and searching the ProGenomes2.0 database (50) with this profile (51). The phylogenetic tree shown in SI Appendix, Fig. S1C was constructed from this alignment using FastTree (52) and pruned to display only E. coli systems (Newick file for this tree in Dataset S10). EnvZ, RstB, and CpxA homologs from the ProGenomes2.0 database were PNAS 2023 Vol. 120 No. 18 e2221163120 https://doi.org/10.1073/pnas.2221163120 9 of 10 identified and aligned using HMMER; specifically, jckhmmer was used to itera- tively search the database for matches to the three kinase domain sequences. A phylogenetic tree was constructed using FastTree, and orthologs were classified based on clade identity. The resulting collection of sequences was further filtered by reciprocal HMMER to confirm that the best hit for each sequence in the E. coli genome was the correct paralog out of EnvZ, RstB, and CpxA. Each sequence maintained its species ID, allowing species with or without the relevant paralogs to be identified. The same process allowed YehU, BarA, and PhoR homologs to be identified, aligned, and filtered (fasta files for ortholog alignments are in Datasets S11–S16). The species tree in SI Appendix, Fig. S1B and Fig. 5B was obtained by pruning the tree constructed in ref. 49 (Newick file in Dataset S17). Progenomes 2.0 species IDs for species tested in Fig. 5E are 155892 (Caulobacter), 1770053 (Burkholderia), 1797492 (Betaproteobacterium sp.), and 595494 (Tolumonas). Statistical Calculations. Two-sided Kolmogorov–Smirnov tests were used to deter- mine whether there was a significant difference in distribution between counts of substitutions that do or do not cross-talk in multiple sequence alignments (MSAs) of HK orthologs. Two-sided Fisher’s exact tests were used to determine whether there was a significant difference between presence/absence within the MSA of substitutions that do or do not cross-talk. Enrichment in SI Appendix, Fig. S8H is cal- culated as counts in the first category of species, with counts in the second category subtracted, after scaling for the number of species in each category. 2. 3. 1. G. C. Conant, K. H. Wolfe, Turning a hobby into a job: How duplicated genes find new functions. Nat. Rev. Genet. 9, 938–950 (2008). G. Schreiber, A. E. Keating, Protein binding specificity versus promiscuity. Curr. Opin. Struct. Biol. 21, 50–61 (2011). S. Itzkovitz, T. Tlusty, U. Alon, Coding limits on the number of transcription factors. BMC Genomics 7, 239 (2006). E. J. Capra, B. S. Perchuk, J. M. Skerker, M. T. Laub, Adaptive mutations that prevent crosstalk enable the expansion of paralogous signaling protein families. Cell 150, 222–232 (2012). A. Zarrinpar, S.-H. Park, W. A. Lim, Optimization of specificity in a cellular protein interaction network by negative selection. Nature 426, 676 (2003). J. Alexander et al., Spatial exclusivity combined with positive and negative selection of phosphorylation motifs is the basis for context-dependent mitotic signaling. Sci. Signal. 4, ra42–ra42 (2011). 7. N. H. Shah, M. Löbel, A. Weiss, J. Kuriyan, Fine-tuning of substrate preferences of the Src-family 4. 5. 6. kinase Lck revealed through a high-throughput specificity screen. Elife 7, e35190 (2018), https:// elifesciences.org/articles/35190. J. M. Nicoludis et al., Interaction specificity of clustered protocadherins inferred from sequence covariation and structural analysis. Proc. Natl. Acad. Sci. U.S.A. 116, 17825–17830 (2019). C. J. McClune, A. Alvarez-Buylla, C. A. Voigt, M. T. Laub, Engineering orthogonal signalling pathways reveals the sparse occupancy of sequence space. Nature 574, 702–706 (2019). 8. 9. 10. Z. Chen et al., Programmable design of orthogonal protein heterodimers. Nature 565, 106–111 (2019). enrichment = count(species without RstBA or CpxAR) − count(species with RstBA and CpxAR) #species without RstBA or CpxAR #species with RstBA and CpxAR × . Data, Materials, and Software Availability. Python scripts for analysis are avail- able at https://github.com/d-ghose/laub (53). Datasets generated during this study have been deposited in the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA). Raw reads can be found under BioProject ID PRJNA902002 (54). All other data are included in the manuscript and/or SI Appendix. ACKNOWLEDGMENTS. We thank I. Nocedal for help with bioinformatic code and datasets. We thank S. Srikant, I. Frumkin, D. Saxton, and S. Swanson for helpful discussions. M.T.L. is an Investigator of the Howard Hughes Medical Institute. This work was also supported by AFO MURI FA9550-22-1-0316 to M.T.L. Author affiliations: aDepartment of Biology, Massachusetts Institute of Technology, Cambridge, MA 02139; bDepartment of Biological Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139; cKoch Center for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, MA 02139; and dHHMI, Massachusetts Institute of Technology, Cambridge, MA 02139 28. J. M. Skerker, M. S. Prasol, B. S. Perchuk, E. G. Biondi, M. T. Laub, Two-component signal transduction pathways regulating growth and cell cycle progression in a bacterium: A system-level analysis. PLOS Biol. 3, e334 (2005). 29. J. M. Skerker et al., Rewiring the specificity of two-component signal transduction systems. Cell 133, 1043–1054 (2008). 30. E. J. Capra et al., Systematic dissection and trajectory-scanning mutagenesis of the molecular interface that ensures specificity of two-component signaling pathways. PLoS Genet. 6, e1001220 (2010). 31. F. U. Battistuzzi, A. Feijao, S. B. Hedges, A genomic timescale of prokaryote evolution: Insights into the origin of methanogenesis, phototrophy, and the colonization of land. BMC Evol. Biol. 4, 44 (2004). 32. A. Siryaporn, M. Goulian, Cross-talk suppression between the CpxA–CpxR and EnvZ–OmpR two- component systems in E. coli. Mol. Microbiol. 70, 494–506 (2008). 33. E. S. Groban, E. J. Clarke, H. M. Salis, S. M. Miller, C. A. Voigt, Kinetic buffering of cross talk between bacterial two-component sensors. J. Mol. Biol. 390, 380–393 (2009). 34. H. Ogasawara et al., Genomic SELEX search for target promoters under the control of the PhoQP- RstBA signal relay cascade. J. Bacteriol. 189, 4791–4799 (2007). 35. Y. Yang, M. Inouye, Requirement of both kinase and phosphatase activities of an Escherichia coli receptor (Taz1) for ligand-dependent signal transduction. J. Mol. Biol. 231, 335–342 (1993). 36. T. Jin, M. Inouye, Ligand binding to the receptor domain regulates the ratio of kinase to phosphatase activities of the signaling domain of the hybrid Escherichia coli transmembrane receptor, Taz1. J. Mol. Biol. 232, 484–492 (1993). 37. C. L. Araya, D. M. Fowler, Deep mutational scanning: Assessing protein function on a massive scale. 11. R. Netzer et al., Ultrahigh specificity in a network of computationally designed protein-interaction Trends Biotechnol. 29, 435–442 (2011). pairs. Nat. Commun. 9, 1–13 (2018). 38. D. M. Fowler, S. Fields, Deep mutational scanning: A new style of protein science. Nat. Methods 11, 12. G. Grigoryan, A. W. Reinke, A. E. Keating, Design of protein-interaction specificity gives selective 801–807 (2014). bZIP-binding peptides. Nature 458, 859–864 (2009). 39. P. Casino, V. Rubio, A. Marina, Structural insight into partner specificity and phosphoryl transfer in 13. E. van Nimwegen, J. P. Crutchfield, M. Huynen, Neutral evolution of mutational robustness. Proc. two-component signal transduction. Cell 139, 325–336 (2009). Natl. Acad. Sci. U.S.A. 96, 9716–9720 (1999). 40. A. I. Podgornaia, P. Casino, A. Marina, M. T. Laub, Structural basis of a rationally rewired protein- 14. C. O. Wilke, C. Adami, Evolution of mutational robustness. Mutat. Res. Mol. Mech. Mutagen. 522, protein interface critical to bacterial signaling. Structure 21, 1636–1647 (2013). 3–11 (2003). 15. F. M. Codoñer, J.-A. Darós, R. V. Solé, S. F. Elena, The fittest versus the flattest: Experimental confirmation of the quasispecies effect with subviral pathogens. PLOS Pathog. 2, e136 (2006). 16. J. Zheng, N. Guo, A. Wagner, Selection enhances protein evolvability by increasing mutational robustness and foldability. Science 370, eabb5962 (2020). 41. S. Yamada et al., Structure of PAS-linked histidine kinase and the response regulator complex. Structure 17, 1333–1344 (2009). 42. F. Trajtenberg et al., Regulation of signaling directionality revealed by 3D snapshots of a kinase:regulator complex in action. Elife 5, e21422 (2016), https://elifesciences.org/articles/21422. 43. T.-L.V. Lite et al., Uncovering the basis of protein-protein interaction specificity with a combinatorially 17. J. D. Bloom et al., Evolution favors protein mutational robustness in sufficiently large populations. complete library. eLife 9, e60924 (2020). BMC Biol. 5, 29 (2007). 18. C. O. Wilke, J. L. Wang, C. Ofria, R. E. Lenski, C. Adami, Evolution of digital organisms at high mutation rates leads to survival of the flattest. Nature 412, 331–333 (2001). 44. C. L. Araya et al., A fundamental protein property, thermodynamic stability, revealed solely from large- scale measurements of protein function. Proc. Natl. Acad. Sci. U.S.A. 109, 16858–16863 (2012). 45. K. A. Matreyek et al., Multiplex assessment of protein variant abundance by massively parallel 19. M. J. Harms, J. W. Thornton, Evolutionary biochemistry: Revealing the historical and physical causes sequencing. Nat. Genet. 50, 874–882 (2018). of protein properties. Nat. Rev. Genet. 14, 559–571 (2013). 20. J. D. Bloom, A. Raval, C. O. Wilke, Thermodynamics of neutral protein evolution. Genetics 175, 255–266 (2007). 21. C. N. Pace, Conformational stability of globular proteins. Trends Biochem. Sci. 15, 14–17 (1990). 22. P. D. Williams, D. D. Pollock, R. A. Goldstein, Functionality and the evolution of marginal stability in proteins: Inferences from lattice simulations. Evol. Bioinforma. Online 2, 91–101 (2007). 23. J. D. Bloom, S. T. Labthavikul, C. R. Otey, F. H. Arnold, Protein stability promotes evolvability. Proc. 49. Natl. Acad. Sci. U.S.A. 103, 5869–5874 (2006). 46. H. Garcia Seisdedos, T. Levin, G. Shapira, S. Freud, E. D. Levy, Mutant libraries reveal negative design shielding proteins from supramolecular self-assembly and relocalization in cells. Proc. Natl. Acad. Sci. U.S.A. 119, e2101117119 (2022). 47. P. Creixell et al., Kinome-wide decoding of network-attacking mutations rewiring cancer signaling. Cell 163, 202–217 (2015). 48. K. A. Datsenko, B. L. Wanner, One-step inactivation of chromosomal genes in Escherichia coli K-12 using PCR products. Proc. Natl. Acad. Sci. U.S.A. 97, 6640–6645 (2000). I. Nocedal, M. T. Laub, Ancestral reconstruction of duplicated signaling proteins reveals the evolution of signaling specificity. eLife 11, e77346 (2022). 24. M. T. Laub, M. Goulian, Specificity in two-component signal transduction pathways. Annu. Rev. 50. D. R. Mende et al., proGenomes2: An improved database for accurate and consistent habitat, taxonomic Genet. 41, 121–145 (2007). 25. E. J. Capra, M. T. Laub, Evolution of two-component signal transduction systems. Annu. Rev. Microbiol. 66, 325–347 (2012). 26. T. N. Huynh, C. E. Noriega, V. Stewart, Conserved mechanism for sensor phosphatase control of two-component signaling revealed in the nitrate sensor NarX. Proc. Natl. Acad. Sci. U.S.A. 107, 21140–21145 (2010). and functional annotations of prokaryotic genomes. Nucleic Acids Res. 48, D621–D625 (2020). 51. S. R. Eddy, Accelerated profile HMM searches. PLOS Comput. Biol. 7, e1002195 (2011). 52. M. N. Price, P. S. Dehal, A. P. Arkin, Fasttree: Computing large minimum evolution trees with profiles instead of a distance matrix. Mol. Biol. Evol. 26, 1641–1650 (2009). 53. D. A. Ghose, Marginal specificity code. Github. https://github.com/d-ghose/laub. Deposited 10 November 2022. 27. E. Alm, K. Huang, A. Arkin, The evolution of two-component systems in bacteria reveals different 54. D. A. Ghose, Deep mutational scanning of E. coli protein EnvZ. NCBI SRA. https://www.ncbi.nlm.nih. strategies for niche adaptation. PLOS Comput. Biol. 2, e143 (2006). gov/bioproject?term=PRJNA902002. Deposited 15 November 2022. 10 of 10 https://doi.org/10.1073/pnas.2221163120 pnas.org
10.1073_pnas.2217885120	RESEARCH ARTICLE \| MEDICAL SCIENCES OPEN ACCESS Chromophore supply modulates cone function and survival in retinitis pigmentosa mouse models , Gayle B. Collinf, Vladimir J. Kefalovd,1,2 Yunlu Xuea,b,c,d,1 , Xiaomei Suna, Sean K. Wangb,c,e , and Constance L. Cepkob,c,e,1 Contributed by Constance L. Cepko; received October 19, 2022; accepted May 3, 2023; reviewed by Jeannie Chen and Gordon L. Fain Retinitis pigmentosa (RP) is an ocular disease characterized by the loss of night vision, followed by the loss of daylight vision. Daylight vision is initiated in the retina by cone photoreceptors, which are gradually lost in RP, often as bystanders in a disease process that initiates in their neighboring rod photoreceptors. Using physiological assays, we investigated the timing of cone electroretinogram (ERG) decline in RP mouse models. A correlation between the time of loss of the cone ERG and the loss of rods was found. To investigate a potential role of the visual chromophore supply in this loss, mouse mutants with alterations in the regeneration of the retinal chromophore, 11-cis retinal, were exam- ined. Reducing chromophore supply via mutations in Rlbp1 or Rpe65 resulted in greater cone function and survival in a RP mouse model. Conversely, overexpression of Rpe65 and Lrat, genes that can drive the regeneration of the chromophore, led to greater cone degeneration. These data suggest that abnormally high chromophore supply to cones upon the loss of rods is toxic to cones, and that a potential therapy in at least some forms of RP is to slow the turnover and/or reduce the level of visual chromophore in the retina. retinitis pigmentosa \| visual cycle \| Alström syndrome \| cone photoreceptors \| retina Retinitis pigmentosa (RP) is one of the most common inherited retinal degenerative diseases (IRDs), affecting 1 in 4,000 people worldwide (1). It is characterized by the loss of night vision, followed by the loss of color and daylight vision. The identification of many disease genes, which now number ~100 (https://sph.uth.edu/retnet/), revealed the molecular basis for the initial loss of night vision. Many of the disease genes are expressed only in rods, the photoreceptor type that detects light in dim light conditions. Cone photoreceptors, the photoreceptor type active in brighter light conditions, are then affected secondarily, generally after most of the rods in their neighborhood have died. The cause(s) of secondary cone death in RP is not clear, though studies have suggested that trophic factor loss, oxidative stress, metabolic changes, and immune responses contribute to the loss of cone function and survival (see review ref. 2). Support for these hypotheses comes from therapies that address some of these problems, as they have led to increased cone survival (3–17). Such therapies also have led to a retention of cone-mediated vision, as reflected in measurements made by the opto- motor behavioral test. This test measures an animal’s response to moving bars of different sizes to determine its ability to detect different spatial frequencies. This is a sensitive assay that can show positive results even when only a small fraction of light-responsive cones remains. Most studies utilizing gene therapy to prolong cone survival have not directly assayed cone physiology (10, 18). Similarly, studies of untreated RP mouse strains have not char- acterized cone physiology over time until recently with one RP strain (19), and several studies that did carry out such assessments have left some open questions that we wished to address. For example, following gene therapy aimed at improving cone survival and function in RP mice, electrophysiology measurements were made at a stage before all of the rods were gone (3). Any observed improvement in cone function could thus be due to effects on rods rather than direct effects on cones. Our own experiments aimed at improving cone survival often led to improved optomotor responses. However, we were unable to show improved cone function using electroretinograms (ERGs), a more direct assay of cone physiology (11, 14). To address the issues mentioned above, we set out to characterize the cone ERG during photoreceptor degeneration in RP mouse models and to determine treatments that might improve it. We confirmed that the cone ERG became undetectable at a stage that correlated with the near-complete loss of rods. We further investigated whether this might be due to the loss of rod function, but found that it was due to the loss of the rods themselves. Notably, suppressing the recycling of the visual chromophore for phototransduction was found to preserve the cone ERG, improve cone opsin expression and localization, and extend cone survival after the rods died in RP mice. This leads to a model wherein RP cones are poisoned from excessive chromophore that reaches a toxic threshold when the last few rods die. Significance Retinitis pigmentosa (RP) is a blinding disease affecting 2 million people worldwide. The night vision of RP patients is affected first due to the expression of a disease gene in rod photoreceptors. This loss of rod-mediated night vision is followed by a bystander effect on cone photoreceptors, which results in loss of daylight vision. The mechanism of the secondary cone degeneration is unclear. We used genetically modified mice and electrophysiology to explore this question. We found that slowing down the recycling of vitamin A derivatives for the regeneration of opsin, a GPCR, could protect RP cone function and extend their survival. These studies reveal a unique pathway for RP cone degeneration and suggest potential therapies that could benefit RP patients. Author contributions: Y.X., S.K.W., V.J.K., and C.L.C. designed research; Y.X., X.S., and S.K.W. performed research; G.B.C. contributed new reagents/analytic tools; Y.X., X.S., S.K.W., G.B.C., V.J.K., and C.L.C. analyzed data; and Y.X., S.K.W., V.J.K., and C.L.C. wrote the paper. Reviewers: J.C., University of Southern California; and G.L.F., University of California, Los Angeles. Competing interest statement: C.L.C. is a consultant for Biogen. We have a provisional patent for lowering retinoids for diseases affecting vision. Research support from Spark Therapeutics. C.L.C. is on the scientific advisory board of the Insitute of Ophthalmology, Basel and on Genesight. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected], cepko@ genetics.med.harvard.edu. [email protected], or 2Present address: Department of Ophthalmology, Gavin Herbert Eye Institute, University of California, Irvine, CA 92697. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2217885120/-/DCSupplemental. Published May 30, 2023. PNAS 2023 Vol. 120 No. 23 e2217885120 https://doi.org/10.1073/pnas.2217885120 1 of 10 Results ERG Measurements of the RP Cone Pathway during the Period of Rod Death. We first examined the literature characterizing the ERG response in several commonly used RP mouse strains. As the cone ERG a-wave, which measures cone function, is too small to measure, even in wild-type eyes, studies of the cone ERG a-wave during the disease process in vivo have not been reported. Due to this difficulty, we focused on the cone ERG of the b-wave, which originates from cone ON-bipolar cells that are immediately downstream of cone photoreceptors. We examined this response in several mouse models of RP. The Rhotm1Phm knockout model, also known as Rho−/−, has a complete loss of rhodopsin. In this strain, ERG b-wave signals were detectable until postnatal day 80 to 90 (P80 to P90) (20, 21), the age when rods are almost completely gone (17, 22). We concluded that these ERG responses must have originated from the cone pathway, as rod phototransduction in Rho−/− mice should be absent due to the lack of rhodopsin. In the rd10 strain, which carries a missense mutation in the Pde6b gene that is critical for rod phototransduction, a cone ERG b-wave was detected at P18-30 but lost by P50-63 when rods were no longer present (23). In addition, no cone ERG response was reported in the untreated rd1 strain (23, 24), which carries a null mutation in the Pde6b gene, leading to fast degeneration of rods. Absence of an ERG in this strain was likely due to the complete loss of rods by ~1 mo of age, when the ERG was first measured (24). Together, these results suggest that the loss of cone ERG signals might be correlated with the loss of RP rods. To further examine whether the absence of a cone ERG cor- related with rod loss in the rd1 retina, we first tested cone func- tion at P15 in rd1, a stage where 2-3 rows of rod nuclei would have been present in the outer nuclear layer (ONL) (Fig. 1A). ERG responses were observed in dark-adapted P15 rd1 eyes (Fig. 1 B and C). In contrast, the ERG recordings of rd1 mice at P21, a stage where almost all rods would have been lost (17, 23, 25), showed barely detectable signals (Fig. 1 A–C). Because the waveform and kinetics of the P15 rd1 dark–adapted ERG looked similar to a pure cone pathway response (i.e., no a-wave, emergence of the b-wave ~0.1 cd s/m2), we wondered whether PDE6B deletion alone abolished phototransduction in rods, leaving only the signals from cones. In keeping with this, ERG responses have been observed in Rho−/− mice, where rod phototransduction is absent but cones are functional and persist until the rods die (20). Fig. 1. Characterization of histology and electroretinography (ERG) of RP eyes. (A) Images from immunohistochemistry carried out on postnatal day 39 (P39) wildtype (WT), P15 rd1, P21 rd1, P60 Rho-/-, and P90 Rho-/- retinal cross sections stained with DAPI (cyan), and ARR3 (red). ONL: outer nuclear layer; INL: inner nuclear layer. N = 5. (B) Representative dark-adapted ERG traces from P15 and P21 rd1 eyes. Flash intensities eliciting traces are labeled on the right side in cd s/m2. (C) Ensemble-averaged dark-adapted ERG b-wave amplitude from P15 and P21 rd1 eyes (same as in B). (D) Representative dark-adapted ERG traces from P15 and P21 Gnat1−/− eyes. (E) Ensemble-averaged dark-adapted ERG b-wave amplitude from P15 and P21 Gnat1−/− eyes (same as in D). (F) Representative dark-adapted ERG traces from P60 and P90 Rho−/− eyes. (G) Ensemble-averaged dark-adapted ERG b-wave amplitude from P60 and P90 Rho−/− eyes (same as in F). Error bar: SEM. NS: not significant; P > 0.05, P < 0.01, P < 0.001, **P < or <<0.0001. The number in the round brackets “()” indicates the number of eyes within each group. 2 of 10 https://doi.org/10.1073/pnas.2217885120 pnas.org To further examine whether the loss of rods or the absence of rod phototransduction was responsible for the differences in the rd1 cone ERG at P15 vs. P21, we applied the same ERG protocol with transducin-α subunit-deficient mice (Gnat1tm1Clma, also referred as Gnat1−/−). These mice have a normal number and morphology of rods, but do not respond to light (26). Robust cone ERG signals were observed in Gnat1−/− eyes at both P15 and P21 (Fig. 1 D and E), confirming that the loss of rod phototransduction is not the cause of RP cone ERG loss. This finding is also consistent with the observation of persistent ERG responses in adult Gnat1−/− eyes in multiple previous studies (26–33). To further investigate the correlation between RP rod number and RP cone ERG, ERG measurements were made using Rho−/− mice at flash intensities spanning 4 log-units, which cover a brighter range than the previous longitudinal studies conducted at only one flash intensity (20, 21). In the retinas of Rho−/− mice, ~3 to 4 rows of rod nuclei were present in the ONL at P60, while all rods were gone by P90 (Fig. 1A). ERG measurements at these ages showed a similar correlation between the presence of rod nuclei and the cone ERG (Fig. 1 F and G), consistent with the findings of the previous study (20, 21). Although we wished to run the same ERG protocol on rd10 mice, it was difficult to obtain reproducible data. Our protocol measures the dark-adapted cone responses without background light for strains that do not possess functional rods, such as Rho−/−, rd1, and Gnat1−/−. In rd10, rod phototransduction is active, while rod degeneration is geographically uneven and dependent upon light, which can vary within the mouse room. We thus did not carry out this ERG protocol in this strain. In summary, the rd1, Rho−/−, and Gnat1−/− results show that the disappearance of the cone ERG correlates with the loss of rods in the ONL, but not with the loss of rod phototransduction (see schematics in SI Appendix, Fig. S1). RP Cone ERG Signals Are Retained in Mice with Rlbp1 Deficiency. As the function of cones, not merely their survival, is critical for daylight vision, we asked how changes in the retina accompanying rod death might affect cone function. We postulated that the amount of the visual chromophore, and its derivatives, which might be buffered by rods, could affect cone function. In wild-type retinas, rods and cones might compete for the regenerated chromophore from the retinal pigmented epithelium (RPE). Following rod death, the abundant supply and/or accelerated turnover of retinoids for cones, especially retinaldehyde (i.e., retinal), might be harmful to cones. If true, slowing down the visual cycle, the process through which the visual chromophore is recycled, might preserve the function of RP cones. CRALBP is a carrier protein for 11-cis retinoids and is expressed in both RPE and Müller glial cells (34). We previously showed that CRALBP, encoded by the Rlbp1 gene, is important for the chromophore supply to cones, and that the deletion of Rlbp1 slows the dark adaptation of cones in mice (35). To investigate whether this gene might play a role in modulating the cone function of RP strains, we crossed the Rlbp1−/− strain (also known as Rlbp1tm1Jsa) with the rd1 strain. Rlbp1−/−;rd1 double homozygous mice did not have any improvement in rod survival, as reflected by ONL thickness at P21 (Fig. 2A). However, when tested for cone function, ERG signals were observed at P20, P30, and P40. Age-matched Rlbp1+/+;rd1 controls, which were derived from the same founders as those of Rlbp1−/−;rd1 mice, had almost flat or noisy cone ERGs at these ages (Fig. 2 B–G). In addition, Rlbp1−/−;Rho−/− double homozygous mice were generated and tested at P90. Improvement in ERG waveforms was observed in these mice, compared to the Rlbp1+/+;Rho−/− controls derived from the same founders (Fig. 2 H and I). Cone Opsin Expression in Rlbp1-Deficient rd1 Mice. We next examined cone opsin expression during disease progression in Rlbp1-deficient mice. Antisera for the OPN1MW and OPN1SW proteins were combined and used to stain M-opsin and S-opsin together (SI Appendix, Fig. S2). The central retina of P14 rd1 mice exhibited cone opsin “dots” resembling the degenerating cone outer segments (Fig. 3A), while the P21 rd1 cones lacked such punctate structures and the cone opsins were mislocalized to the cell bodies (Fig. 3B). Similar to the P14 Rlbp1+/+;rd1 mice, P14 Rlbp1−/−;rd1 double homozygous cones had cone opsin–enriched dot structures (Fig. 3A), but unlike the P21 Rlbp1+/+ rd1;control, P21 Rlbp1−/−;rd1 preserved more of these structures, as seen in both flat-mounts and cross-sections (Fig. 3 B and C). Cone Survival in Rlbp1-Deficient rd1 Mice. To test whether RP cone survival would be improved with Rlbp1 deficiency, the number of cones in Rlbp1−/−;rd1 retinal flat-mounts was assayed using cone arrestin (ARR3) antibody staining. We first looked at P21, around the beginning of cone death, and observed ARR3 staining throughout the retina (Fig. 4A). Compared to the P21 retina, P40 ARR3 staining was dramatically decreased in the central retina of both Rlbp1−/−;rd1 and Rlbp1+/+;rd1 mice, suggesting severe cone degeneration and death from P21 to P40 (Fig. 4B), as has been reported for the rd1 strain (36). Nonetheless, when the number of ARR3+ cells in P40 rd1 retinas was quantified, there were significantly more cones (ARR3+ cells) in the central retina of Rlbp1−/− rd1 mice relative to the Rlbp1+/+ rd1 mice (Fig. 4C). The Homozygous L450M Mutation in Rpe65 Promotes Retention of the rd1 Cone ERG and Survival. To further test the hypothesis that down-regulating the visual cycle benefits RP cone function and survival, a second strain, Rpe65L450M, was tested. RPE65 is expressed in the RPE and is a critical enzyme for the visual cycle through its activity in regenerating 11-cis retinoids (37). The L450M allele of RPE65 is naturally present in C57BL/6 strains and results in a slower RPE visual cycle and photoreceptor dark adaptation (28, 38), although to a lesser extent than Rlbp1 deficiency. Accordingly, it does not cause cone degeneration in normal conditions. Mice carrying the Rpe65L450M allele were crossed to rd1 and the ERG responses and cone numbers were characterized. Double homozygous Rpe65L450M;rd1 mice showed an improved cone ERG at 10 and 100 cd s/m2, two of the four flash intensities tested, compared to the homozygous Rpe65L450;rd1 controls that were derived from the same founders and carried the wild-type allele of Rpe65 (Fig. 5 A and B). We also observed significantly more ARR3+ cones in the central retina of double homozygous Rpe65L450M;rd1 mice than those of Rpe65L450;rd1 control mice (Fig. 5 C and D). Ectopic RPE65 and LRAT Expression in Cones Reduces rd1 Cone Survival. As the loss of RPE65 activity or RLBP1 benefitted cone survival and function, it was of interest to determine whether the overexpression of related visual cycle genes might produce the opposite phenotype. All-trans retinol needs to travel from the photoreceptors to the RPE to be eventually converted to 11-cis retinal. RPE65 and LRAT are two critical enzymes that limit the conversion of all-trans retinol to 11-cis retinol in the RPE (39, 40), and 11-cis retinol can be further oxidized to 11-cis retinal by 11-cis RDH enzymes in the RPE (41). Subsequently, 11-cis retinal travels back to photoreceptors from the RPE to regenerate the visual pigments. Cones, but not rods, can directly use 11-cis retinol from Müller glial cells to regenerate the visual pigments (42, 43). If RPE65 and LRAT are ectopically expressed in cones, retinoids would not need to travel between cones and RPE/Müller glia, PNAS 2023 Vol. 120 No. 23 e2217885120 https://doi.org/10.1073/pnas.2217885120 3 of 10 Fig. 2. Effects of Rlbp1 deficiency on RP mouse ERG. (A) Images from immunohistochemistry carried out on P21 rd1;Rlbp1+/+ and rd1;Rlbp1−/− retinal cross-sections stained with DAPI (gray) and CRALBP (green). RPE: retinal pigmented epithelium; IPL: inner plexiform layer; GCL: ganglion cell layer. N = 3. (B) Representative dark-adapted ERG traces from P20 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes. (C) Ensemble-averaged dark-adapted ERG b-wave amplitude from P20 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes (same as in B). (D) Representative dark-adapted ERG traces from P30 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes. (E) Ensemble-averaged dark-adapted ERG b-wave amplitude from P30 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes (same as in E). (F) Representative dark-adapted ERG traces from P40 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes. (G) Ensemble-averaged dark-adapted ERG b-wave amplitude from P40 rd1;Rlbp1+/+ and rd1;Rlbp1−/− eyes (same as in F). (H) Representative dark-adapted ERG traces from P90 Rho−/−;Rlbp1+/+ and Rho−/−;Rlbp1−/− eyes. (I) Ensemble-averaged dark-adapted ERG b-wave amplitude from P90 Rho−/−;Rlbp1+/+ and Rho−/−;Rlbp1−/− eyes (same as in H). Error bar: SEM. NS: not significant; P > 0.05, P < 0.05, P < 0.01, P < 0.001, *P < or << 0.0001. The number in the round brackets “()” indicates the number of eyes within each group. which might increase the chromophore turnover for cone opsin regeneration and their accumulation in cones. AAV vectors using a cone-specific promoter, RO1.7, which was made from a human red opsin promoter (44–46), were created to express RPE65 and LRAT in cones. They were injected along with a trace amount of AAV-RedO-H2BGFP for cone labeling. Injections were made into rd1 retinas at P0, and ERGs were measured at P20. No ERG changes were noted at P20 between Rpe65 + Lrat-transduced rd1 retina and controls (i.e., H2BGFP only) (Fig. 6A). When retinas were examined for cone survival at P50, the number of cones was significantly lower in the AAV-Rpe65 + Lrat-transduced retinas compared to the control at P50 (Fig. 6 B and C). In P30 wild- type retinas, AAV-Rpe65 + Lrat infection resulted in no noticeable alteration in cone morphology or survival compared to the control, as probed by peanut agglutinin (PNA) or cone opsin staining (Fig. 6D and SI Appendix, Fig. S3). ALMS1 Disruption Reduces Cone Function and Accelerates Cone Dark Adaptation. The above results suggested that an augmented visual cycle is harmful to cones, and we wondered whether any additional evidence from other models could support this notion. Searching the literature for other rod/cone disease genes, we took note of Alms1, the gene disrupted in the Alström syndrome. Alström syndrome is a multisystem disease that affects vision, hearing, heart function, as well as other systems. Loss of function of ALMS1 causes early-onset cone–rod dystrophy in humans (47). A previous characterization of Alms1Gt(XH152)Byg gene trap mice (herein referred as Alms1−/−) showed decreased scotopic– photopic ERG amplitude and mislocalized rhodopsin in rods (48). As the mechanism for visual deterioration in ALMS1 is not clear, we set out to further investigate the physiological responses in Alms1−/− mice by thoroughly examining the dynamics of cone dark adaptation. To determine the consequence of Alms1 disruption on cone func- tion and dark adaptation, Alms1−/− mice were crossed to mice with a null mutation in Gnat1, Gnat1irdr (49), herein referred to as Gnat1−/−. Gnat1−/− mice have no rod function and thus ERGs or transretinal recordings would measure only cone function. The ERG responses from the double-mutant (Alms1−/−;Gnat1−/−) mice showed a significant reduction of cone b-wave amplitude at most of the 4 of 10 https://doi.org/10.1073/pnas.2217885120 pnas.org Fig. 3. Effects of Rlbp1 deficiency on RP cone opsin expression. (A) Images from immunohistochemistry carried out on P14 rd1;Rlbp1+/+ and rd1;Rlbp1−/− flat- mounted retinas stained for cone opsins (a mixture of antibodies to OPN1SW and OPN1MW; white). Higher-magnification images of boxed regions are shown below. N = 8. (B) Images from immunohistochemistry carried out on P21 rd1;Rlbp1+/+ and rd1;Rlbp1−/− flat-mounted retinas stained with anticone opsins (OPN1SW and OPN1MW; white). N = 8. (C) Images from immunohistochemistry carried out on P21 rd1;Rlbp1+/+ and rd1;Rlbp1−/− retinal cross-sections stained with DAPI (gray) and anticone opsins (OPN1SW + OPN1MW; green). N = 4. flash intensities compared to the Alms1+/+;Gnat1−/− mice that were derived from the same founders (Fig. 7 A and B). The normalized intensity–response curves for the Alms1−/−;Gnat1−/− double mutant and the Alms1+/+;Gnat1−/− eyes were comparable, suggesting that cone b-wave half-response intensity (Ib 1/2) was not affected by Alms1 disruption (Fig. 7 B, Inset). Transretinal recordings (i.e., ex vivo Fig. 4. Effects of Rlbp1 deficiency on RP cone survival. (A) Images from immunohistochemistry carried out on P21 rd1;Rlbp1+/+ and rd1;Rlbp1−/− flat-mounted retinas stained with ARR3 (white). (B) Images from immunohistochemistry carried out on P40 rd1;Rlbp1+/+ and rd1;Rlbp1−/− flat-mounted retinas stained for ARR3 (white). High-magnification images of boxed regions are shown below. (C) Quantification of ARR3-positive cones in the central retina for different groups (same as in B), using a previous published method (12). Error bar: SD. P < or <<0.0001. The number in the round brackets “()” indicates the number of retinas within each group. PNAS 2023 Vol. 120 No. 23 e2217885120 https://doi.org/10.1073/pnas.2217885120 5 of 10 Fig. 5. RP cone ERG and survival in the Rpe65-L450M variant. (A) Representative dark-adapted ERG traces from P20 homozygous rd1;Rpe65L450 and homozygous rd1;Rpe65L450M eyes. (B) Ensemble-averaged dark-adapted ERG b-wave amplitude from P20 homozygous rd1;Rpe65L450 and homozygous rd1;Rpe65L450M eyes (same as in A). Error bar: SEM. (C) Images from immunohistochemistry carried out on P40 rd1;Rpe65L450 and homozygous rd1;Rpe65L450M flat-mounted retinas stained for ARR3 (white). Higher-magnification images of boxed regions are shown below. (D) Quantification of ARR3-positive cones in the central retina for different groups (same as in C). Error bar: SD. NS: not significant; P > 0.05, P < 0.01, *P < 0.001, P < or << 0.0001. The number in the round brackets “()” indicates the number of eyes within each group. ERG) were used to directly assess the cone function as it allows the isolation of the a-wave (Fig. 7C). Similar to the ERG b-wave results, the cone photoresponse amplitude in Alms1−/−;Gnat1−/− mice was significantly reduced compared to Alms1+/+;Gnat1−/− at most of the flash intensities (Fig. 7D), suggesting that Alms1 disruption directly impairs cone function. The dim flash kinetics and the a-wave 1/2) of the Alms1−/−;Gnat1−/− cones showed half-response intensity (Ia no difference from the Alms1+/+;Gnat1−/− cones (Fig. 7 C and D, Insets), suggesting that the compromised cone function (i.e., dark current capacity) was not caused by issues in the amplification of the phototransduction cascade. The Alms1 gene trap disruption was investigated for effects on the kinetics of cone dark adaptation following exposure to bright light. Bright light bleaches most of the cone visual pigment, and the ERG can be used to measure the recovery of cone sensitivity following the bleach. Following bright light exposure, cone pig- ment regeneration and dark adaptation are driven by the chromo- phore, which is recycled by both the canonical visual cycle through the RPE and the cone-specific visual cycle through the Müller glial cells (50). The cone b-wave sensitivity (Sb f) recovery was signifi- cantly increased in the Alms1−/−;Gnat1−/− retina in the initial ~20 min following the bleach (Fig. 7E), implying an enhanced effi- ciency of chromophore turnover for cones in Alms1−/− mice. The initial phase of cone recovery is driven by the retina visual cycle, whereas the later stage of cone dark adaptation is driven by the RPE visual cycle (28). To examine the operation of the cone-specific visual cycle directly, cone dark adaptation was measured in isolated retinas using transretinal recordings. Under these conditions, with the RPE removed, pigment regeneration is driven exclusively by the retina visual cycle (51). Consistent with the in vivo ERG results, the recovery of cone sensitivity in the isolated mutant retinas was significantly enhanced compared to control retinas in the 2 to 12 min postbleach (Fig. 7F). These results suggest that the cone-specific retinal visual cycle is enhanced in the Alms1−/− retinas. The correlation between the enhanced chromophore recy- cling and the suppression of cone function in Alms1 mutant mice is consistent with our observations in rd1 and Rho−/− mice and suggests that a faster than normal chromophore recycling in cones could be detrimental to their function and long-term survival. Discussion The goal of many gene therapies and small-molecule therapeutics for IRDs is to prolong cone survival and function. In our own studies using gene therapies aimed at a gene-agnostic approach, we found increased cone survival, but very little improvement in the cone ERG (11, 12, 14). Similarly, improvements in the cone ERG responses in other studies have been modest (3, 4, 13). This motivated an examination of cone function in untreated RP mice. Here, we report that cone ERG responses were nearly undetectable at an early stage of degeneration in mouse RP models (Fig. 1). This was surprising as the cones were all present, and were notice- ably, but not yet drastically, altered in their morphology. By exam- ining several strains of mutant mice, as well as the time course of the cone ERG loss, it is clear that the loss of the cone ERG is correlated with the loss of rods, rather than the absence of rod phototransduction, as expected. However, a recent study reported that the photopic light response from retinal ganglion cells persists in 7-mo-old Cngb1neo/neo mice, a RP strain in which most of the rods would have been lost by this age, as judged by the thickness of ONL (52). Another study, which employed patch-clamping, reported that the light response of cones persists in 9-wk-old rd10 6 of 10 https://doi.org/10.1073/pnas.2217885120 pnas.org also true, in that RP cone loss was exacerbated by misexpression, directly within cones, of the genes that can directly provide them with chromophore (Fig. 6). Interestingly, the cones in Rho−/− mice have an acceleration in their dark adaptation, which reflects the speed of chromophore turnover in cones (54). This acceleration was seen before rods were gone, suggesting an enhanced access of recycled chromophore from the RPE in RP cones, possibly due to less competition from rods in the absence of rhodopsin. All-trans retinal is produced from 11-cis retinal upon photon absorption. It is believed to be toxic to rods when it is present in abnormally high amounts. This notion is supported by a series of studies using Abca4−/−;Rdh8−/− double homozygous mice, which lack the necessary machinery to clear all-trans retinal from rods (55–57). Although it has not been measured or reported, the total amount of retinoids in RP eyes might not be greatly reduced upon rod death, due to a lack of active clearance mechanisms for multiple forms of retinoids. As aldehydes are reactive electrophiles that can directly damage protein thiols and amines (see review ref. 58), an excessive amount of unbound and free retinaldehyde may directly harm cones. Rods might buffer the lipophilic retinoids by providing a local sink via their lipid-rich inner and outer segments. The ability of rods to buffer retinoids might regulate the amount of free 11-cis retinal available to cones, and/or the rods might absorb the all-trans retinal/retinol from cone phototransduction to retard their trans- port to the RPE/Müller glia for further recycling. In the Rho−/− retina, before rods are gone, there may be an overall lower level of 11-cis retinal in dark-adapted retinas due to the lack of rhodopsin (59). However, the level of free unbound 11-cis retinal per cone could be higher in Rho−/− than that in wild-type retinas. In fact, turning down the visual cycle alleviated cone dysfunction in the Rho−/− strain when rods were gone (Figs. 1 and 2). These results suggest that it may not be the total amount, but the free, unbound form of 11-cis retinal and/or all-trans retinal that harms cones. Beyond the mechanism discussed above, phototransduction itself might also contribute to the poor function and survival of RP cones. This possibility is supported by the death of rods in light damage models due, at least partially, to overactive photot- ransduction (60). Although light damage models are not equiva- lent to RP degeneration, a similar mechanism might affect RP cones, i.e., there might be a reduction in cone phototransduction if the supply of 11-cis retinal is reduced, leading to protection of cones. In addition, cones use NADPH to reduce all-trans retinal, which is generated from 11-cis retinal upon light activation, to all-trans retinol (61). A reduction of 11-cis retinal might reduce the consumption of NADPH in RP cones. The NADPH thus saved may provide the RP cones with more reductive power, which can be used to fight oxidative stress (62). Our results may also help to explain a recent finding regarding a protective effect of retinoic acid (RA) on RP cone survival. We found that RA, produced by ALDH1A1 in peripheral Müller glia, protected peripheral cones in a RP mouse model (36). This enzyme might contribute to cone survival by changing the retinaldehyde to the acid form. It is clear that a transcriptional readout of RA is also involved as there is at least a partial protection of cones via an activated form of the RA receptor. However, both mechanisms may contribute to cone survival. Recordings from the Alms1−/− mice, a cone–rod dystrophy dis- ease model, provide an additional set of data to consider. We show here that an Alms1 disruption induces an acceleration of cone dark adaptation. This is not due to the loss of rods, since at the age tested (~6 wk old), the number of rods is similar between Alms1−/− mutants and littermate controls (48). We noticed that Alms1 RNA is enriched in cones relative to rods or Müller glia using a single-cell transcriptomic database (63). ALMS1 is Fig. 6. RP cone ERG and survival with ectopic expression of RPE65 and LRAT in cones. (A) Representative dark-adapted ERG traces from P20 rd1 eyes P0- infected with adeno-associated viruses (AAVs) encoding RPE65, LRAT, and H2BGFP (AAV8-RedO-Rpe65, ≈1 × 109 vg/eye, AAV8-RedO-Lrat, ≈1 × 109 vg/eye, plus AAV8-RedO-H2BGFP, 2.5 × 108 vg/eye) or control (AAV8-RedO-H2BGFP, 2.5 × 108 vg/eye). N = 3. (B) Representative P50 rd1 flat-mounted retinas after P0 infection with Rpe65 + Lrat, or control (same as in A). (C) Quantification of H2BGFP-positive cones within the ½ radius of P50 rd1 retinas transduced with Rpe65 + Lrat, and control (same as in B). Error bar: SD. P < or << 0.0001. The number in the round brackets “()” indicates the number of eyes within each group. (D) Representative P30 WT flat-mounted retinas stained with PNA after P0 infection with Rpe65 + Lrat, or control (same as in A). Right panels are high- magnification images from the regions indicated by boxes in the Middle panels. mice (19). While quantification of rods using a sensitive assay was carried out for rd1 and Rho−/− in a previous study from our lab (17), a sensitive assay, such as RT-PCR, was not used to measure the remaining rods in these studies of Cngb1neo/neo (52) or rd10 mice (19). Moreover, Cngb1neo/neo and rd10 mice seem to be on the C57BL/6 background which carries the L450M allele of Rpe65 (23, 53). Thus, it is not clear whether the two reports (19, 52) are in conflict with the findings presented here. Our results suggest a few possibilities regarding the mechanism of cone dysfunction and degeneration in RP. A partial rescue of the cone ERG, and of cones themselves, was seen in mutants that have impaired chromophore transport or recycling. This led to the hypothesis that the amount of chromophore, and/or its derivatives, contributes to cone toxicity. There have been studies suggesting that a partially suppressed visual cycle, as defined by a slower turnover rate of chromophore, is protective to rods in certain conditions: Rlbp1 deficiency protects rods from light-induced damage (34), as does the Rpe65L450M mutation (38). In both of these cases, there was a delay in the dark adaptation of rods and cones as well as slowed chromophore turnover (28, 34, 35, 38). Here, we found that these two genetic conditions also led to prolonged cone survival and function in RP mouse models (Figs. 2–5). The converse was PNAS 2023 Vol. 120 No. 23 e2217885120 https://doi.org/10.1073/pnas.2217885120 7 of 10 Fig. 7. Cone pathway electrophysiology response and dark adaptation of six-week-old Alms1-mutant mice on Gnat1−/− background. (A) Representative dark- adapted in vivo ERG traces from control (i.e., Alms1+/+;Gnat1−/−, same for all below) and Alms1−/− (i.e., Alms1−/−;Gnat1−/−, same for all below) eyes. Flash intensities eliciting traces are labeled on the right side in log cd s/m2. (B) Averaged dark-adapted in vivo ERG cone b-wave intensity–response curves of control (n = 6) and mutant eyes (n = 6). Inset: the corresponding normalized intensity–response curves of the dark-adapted cone b-wave. (C) Representative ex vivo transretinal recording traces from control and Alms1−/− retinas. The eliciting flash intensity of the red trace was 13,878 photons/μm2. Inset: averaged normalized dim-flash responses (elicited by 1,387 photons/μm2) of control (n = 6) and Alms1−/− retina (n = 6). (D) Averaged intensity–response curve of control (n = 6) and Alms1−/− groups (n = 8). Inset: normalized intensity–response curve of control (n = 6) and Alms1 transformations−/− groups (n = 8). (E) In vivo ERG of cone b-wave sensitivity recovery DA, was 161 ± 10 μV m2/cd s for control (n = 5) following 90% photobleach, driven by both RPE and Müller glia visual cycles. The prebleach b-wave sensitivity, Sf and 110 ± 7 μV m2/cd s for Alms1−/− eyes (n = 6), P < 0.01. (F) Transretinal recordings of cone sensitivity recovery following 90% photobleach, driven only by the DA was 9.2 ± 2.3 nV μm2/ph for control (n = 6) and 5.9 ± 1.9 nV μm2/ph for Alms1−/− retinas (n = 8), P > 0.05. Error bar: SEM. Unlabeled: Müller glia visual cycle. Sf not significant; P > 0.05, P < 0.05, P < 0.01, **P < 0.001. believed to be part of the basal body and centrosome (64, 65), which are critical elements of the connecting cilium of photore- ceptors (66). This structure is important for transport between the inner and outer segments of photoreceptors. The observation of an acceleration in dark adaptation suggests that Alms1 disrup- tion enhanced the rate of the cone-specific visual cycle (Fig. 7). The inner segments of cones get 11-cis retinol from Müller glial cells (42, 43, 51, 67). 11-cis retinol is transported to the outer segments, presumably through the connecting cilium, and becomes oxidized to 11-cis retinal in the outer segments (68). In addition, ex vivo experiments showed that 11-cis retinal could travel from inner segment to outer segment in cones, but not in rods (69), as may occur if a small pool of 11-cis retinal ever exists in the retina in vivo. We speculate that ALMS1 might serve to regulate this transport, ensuring that an excessive amount of 11-cis retinoids do not get transported to the cone’s outer seg- ment. The loss or the dysfunction of ALMS1 might facilitate the transport of the 11-cis retinoids to the outer segment, thus enhancing cone chromophore turnover and contributing to the deterioration of cones in Alström syndrome. In summary, our results suggest a unique mechanism by which the degeneration of rods could lead to the loss of cone function and survival by leading to abnormally high, and likely toxic, levels of visual chromophore in, or surrounding, cones. Materials and Methods Animals. rd1 mice carry the homozygous Pde6brd1 allele from the FVB strain. CD1 mice were purchased from Charles River Laboratories and used as the WT mice for histology. Gnat1tm1Clma and Rhotm1Phm mice were generated and gifted 8 of 10 https://doi.org/10.1073/pnas.2217885120 pnas.org by Janis Lem (Tufts University, MA) (26, 70). Rlbp1tm1Jsa mice were generated and gifted by John Saari (University of Washington, WA) (34). RPE65L450M allele is on the C57BL/6J genetic background (38) and was purchased from The Jackson Laboratory. These mice were crossed to each other and genotyped by RT-PCR using primers (Transnetyx, Cordova, TN). FVB and Rhotm1Phm strains carry the RPE65L450 allele. Alms1Gt(XH152)Byg mice (48) were crossed to Gnat1irdr mice (provided by Bo Chang, The Jackson Laboratory, ME), to generate double homozygous mice, and were genotyped in the Naggert Lab (The Jackson Laboratory, ME). All mice were raised in default 12:12 light–dark cycle of animal facilities. The ambient light at the bottom of the mouse cages varied between 10 and 50 lx. In Vivo ERG. ERGs were recorded from mouse eyes using an Espion E3 System (Diagonsys LLC) or LKC® system as described previously (35, 71). Mice were dark adapted overnight and anesthetized with ketamine–xylazine cocktail (100/10 mg/kg) prior to the recordings. The pupils of the animals were dilated with 1% tropicamide eye-drops. Throughout the experiments, the body temperature was kept warm using a heating pad. ERG signals were picked up from the mouse cornea by electrodes immersed in phosphate-buffered saline (PBS). Before running the tests, the mice were stabilized for 15 min in darkness. For dark-adapted intensity–response exper- iments, the tests started from dim to bright flashes, and the averaged cone b-wave amplitude of multiple traces was measured at one intensity. Prior to the dark adapta- tion test, the dim flash response was recorded using several 0.238 cd s/m2 flashes for normalization purpose. An estimated 90% of the visual pigment was photobleached using a custom-made green LED light source. The sensitivity recovery was measured after the bleach and normalized to the prebleach dim flash response to construct the kinetics of dark adaptation as previously described (28). Ex Vivo Transretinal Recording. Alms1−/−;Gnat1−/− retinas were dissected from the eyes of overnight dark-adapted mice, which were euthanized by CO2, under an infrared light microscope in accordance with the institutional guidelines of Washington University. A custom-made recording chamber was used to mount the retina with the photoreceptor side facing the light source (72, 73). 37 °C Locke’s solution bubbled with 95% O2/5% CO2 was used to perfuse the retina in the chamber. The perfusion solution also contained 30 μM DL-AP4 to block the mGluR-mediated synaptic transmission, inhibiting the response from ON-bipolar cells (i.e., the b-wave of ERG). Prior to the testing flashes, the retina was stabi- lized for 15 min in the recording chamber in complete darkness. Responses from photoreceptors were induced by a 505-nm LED light. The parameters of flashes, such as intensity and duration, were fine-controlled through Plamp9 software (Molecular Devices). With light stimuli spanning ~5 log units, the transretinal recording signals of cones (i.e., the voltage change across the retina) were ampli- fied, digitized, and recorded in a computer to construct the intensity–response curve of cones. For the dark adaptation test, a bright 3-s light was first delivered to the retina to bleach an estimated 90% of the visual pigment. Then, the recovery of photoresponses was recorded immediately using a pre-programmed protocol over a ~12-min period. To reconstruct the dark adaptation of the sensitivity, the recorded recovery was normalized to its prebleach level of dim flash response as previously described (35). Subretinal Injection of AAVs. The AAVs were designed, prepared, and delivered to eyes as previously described (14). We used Gibson assembly to clone AAV-RO1.7- Rpe65 and AAV-RO1.7-Lrat. The cDNAs of mouse Rpe65 (#EX-Mm35203-M02) and Lrat (#EX-Mm12130-M02) were purchased from GeneCopoeia (Rockville, MD). The vectors were packaged into the AAV8 capsid, produced following transfection of 293T cells, and concentrated using iodixanol gradients as previously described (4, 74). AAVs were mixed together and diluted with PBS before injection into the subretinal space of P0 mouse eyes as previously described (4, 75). Histology and Cone Quantification. The frozen sections and flat mounts of retinas were collected as previously described (76). Two methods of cone quan- tification were performed, based on previously established methods: 1) cone arrestin staining, and counting with ImageJ (12). 2) Labeling with coinjected AAV-RedO-H2BGFP, and counted with a MATLAB script (14). Data, Materials, and Software Availability. All study data are included in the article and SI Appendix. ACKNOWLEDGMENTS. We thank Jürgen K. Naggert at Jackson Laboratory for material and advisory support; Paula Montero-Llopis (Microscopy Resources on the North Quad) of Harvard Medical School; Li Tan, Wanying Li, and Kangning Sang (Optical Imaging Core Facility) of Shanghai Research Center for Brain Science and Brain-Inspired Intelligence for technical support. This work was funded by Howard Hughes Medical Institute (to C.L.C.), NIH grants K99EY030951 (to Y.X. before June 30, 2022), Lingang Laboratory startup fund (to Y.X. after July 20, 2022), and R01EY030912 (to V.J.K.). We acknowledge support from NIH grant P30EY034070 and from an unrestricted grant from Research to Prevent Blindness (to the Gavin Herbert Eye Institute at the University of California, Irvine). G.B.C. was supported by NIH grant R01HD036878. Author affiliations: aLingang Laboratory, 200031, Shanghai, China; bDepartment of Genetics, Blavatnik Institute, Harvard Medical School, Boston, MA 02115; cDepartment of Ophthalmology, Harvard Medical School, Boston, MA 02115; dDepartment of Ophthalmology & Visual Sciences, Washington University School of Medicine, St. Louis, MO 63110; eHHMI, Boston, MA 02115; and fThe Jackson Laboratory, Bar Harbor, ME 04609 1. 2. 3. D. T. Hartong, E. L. Berson, T. P. Dryja, Retinitis pigmentosa. Lancet 368, 1795–1809 (2006). D. S. Narayan, J. P. M. Wood, G. Chidlow, R. J. Casson, A review of the mechanisms of cone degeneration in retinitis pigmentosa. Acta Ophthalmol. 94, 748–754 (2016). L. C. Byrne et al., Viral-mediated RdCVF and RdCVFL expression protects cone and rod photoreceptors in retinal degeneration. J. Clin. Invest. 125, 105–116 (2015). 4. W. Xiong, A. E. MacColl Garfinkel, Y. Li, L. I. Benowitz, C. L. Cepko, NRF2 promotes neuronal survival in neurodegeneration and acute nerve damage. J. Clin. Invest. 125, 1433–1445 (2015). 5. N. Aït-Ali et al., Rod-derived cone viability factor promotes cone survival by stimulating aerobic 6. 7. 8. glycolysis. Cell 161, 817–832 (2015). A. Venkatesh et al., Activated mTORC1 promotes long-term cone survival in retinitis pigmentosa mice. J. Clin. Invest. 125, 1446–1458 (2015). Y. Yang et al., Functional cone rescue by RdCVF protein in a dominant model of retinitis pigmentosa. Mol. Ther. 17, 787–795 (2009). K. Komeima, B. S. Rogers, P. A. Campochiaro, Antioxidants slow photoreceptor cell death in mouse models of retinitis pigmentosa. J. Cell. Physiol. 213, 809–815 (2007). 16. K. Komeima, B. S. Rogers, L. Lu, P. A. Campochiaro, Antioxidants reduce cone cell death in a model of retinitis pigmentosa. Proc. Natl. Acad. Sci. U.S.A. 103, 11300–11305 (2006). 17. C. Punzo, K. Kornacker, C. L. Cepko, Stimulation of the insulin/mTOR pathway delays cone death in a mouse model of retinitis pigmentosa. Nat. Neurosci. 12, 44–52 (2009). 18. Y. Chinchore et al., Transduction of gluconeogenic enzymes prolongs cone photoreceptor survival and function in models of retinitis pigmentosa. bioRxiv [Preprint] (2019), https://doi. org/10.1101/569665 (Accessed 14 November 2020). 19. E. M. Ellis et al., Cones and cone pathways remain functional in advanced retinal degeneration. Curr. Biol. 33, 1513–1522.e4 (2023). 20. G. B. Jaissle et al., Evaluation of the rhodopsin knockout mouse as a model of pure cone function. Invest. Ophthalmol. Vis. Sci. 42, 506–13 (2001). 21. K. Toda, R. A. Bush, P. Humphries, P. A. Sieving, The electroretinogram of the rhodopsin knockout mouse. Vis. Neurosci. 16, 391–398 (1999). 22. M. M. Humphries et al., Retinopathy induced in mice by targeted disruption of the rhodopsin gene. Nat. Genet. 15, 216–219 (1997). 9. M. Ohnaka et al., Long-term expression of glial cell line-derived neurotrophic factor slows, but does not stop retinal degeneration in a model of retinitis pigmentosa. J. Neurochem. 122, 1047–1053 (2012). 10. D. M. Wu et al., Nrf2 overexpression rescues the RPE in mouse models of retinitis pigmentosa. JCI 23. B. Chang et al., Two mouse retinal degenerations caused by missense mutations in the β-subunit of rod cGMP phosphodiesterase gene. Vision Res. 47, 624–633 (2007). 24. K. M. Nishiguchi et al., Gene therapy restores vision in rd1 mice after removal of a confounding Insight 6, e145029 (2021). 11. S. K. Wang, Y. Xue, C. L. Cepko, Microglia modulation by TGF-β1 protects cones in mouse models of retinal degeneration. J. Clin. Invest. 140, 4360–4369 (2020). 12. S. K. Wang, Y. Xue, C. L. Cepko, Augmentation of CD47/SIRPα signaling protects cones in genetic models of retinal degeneration. JCI Insight 6, e150796 (2021). 13. S. K. Wang, Y. Xue, P. Rana, C. M. Hong, C. L. Cepko, Soluble CX3CL1 gene therapy improves cone survival and function in mouse models of retinitis pigmentosa. Proc. Natl. Acad. Sci. U.S.A. 116, 10140–10149 (2019). 14. Y. Xue et al., AAV-Txnip prolongs cone survival and vision in mouse models of retinitis pigmentosa. Elife 10, e66240 (2021). mutation in Gpr179. Nat. Commun. 6, 6006 (2015). 25. D. B. Farber, R. N. Lolley, Cyclic guanosine monophosphate: Elevation in degenerating photoreceptor cells of the C3H mouse retina. Science 186, 449–451 (1974). 26. P. D. Calvert et al., Phototransduction in transgenic mice after targeted deletion of the rod transducin alpha -subunit. Proc. Natl. Acad. Sci. U.S.A. 97, 13913–13918 (2000). 27. A. Morshedian et al., Light-driven regeneration of cone visual pigments through a mechanism involving RGR opsin in Müller glial cells. Neuron 102, 1172–1183.e5 (2019). 28. A. V. Kolesnikov, P. H. Tang, R. O. Parker, R. K. Crouch, V. J. Kefalov, The Mammalian cone visual cycle promotes rapid m/l-cone pigment regeneration independently of the interphotoreceptor retinoid- binding protein. 2. J. Neurosci. 31, 7900–7909 (2011). 15. S. Mohand-Said et al., Normal retina releases a diffusible factor stimulating cone survival in the 29. F. Vinberg et al., The Na+/Ca2+, K+ exchanger NCKX4 is required for efficient cone-mediated retinal degeneration mouse. Proc. Natl. Acad. Sci. U. S. A. 95, 8357–8362 (1998). vision. Elife 6, e24550 (2017). PNAS 2023 Vol. 120 No. 23 e2217885120 https://doi.org/10.1073/pnas.2217885120 9 of 10 30. A. E. Allen, M. A. Cameron, T. M. Brown, A. A. Vugler, R. J. Lucas, Visual responses in mice lacking 53. J. Chen et al., Channel modulation and the mechanism of light adaptation in mouse rods. J. critical components of all known retinal phototransduction cascades. PLoS One 5, e15063 (2010). Neurosci. 30, 16232 (2010). 31. A. L. Lyubarsky et al., Functionally rodless mice: Transgenic models for the investigation of cone 54. A. V. Kolesnikov et al., Retinol dehydrogenase 8 and ATP-binding cassette transporter 4 modulate function in retinal disease and therapy. Vision Res. 42, 401–415 (2002). 32. S. S. Nikonov, R. Kholodenko, J. Lem, E. N. Pugh, Physiological features of the S- and M-cone dark adaptation of M-cones in mammalian retina. J. Physiol. 593, 4923–4941 (2015). 55. A. Maeda et al., Primary amines protect against retinal degeneration in mouse models of photoreceptors of wild-type mice from single-cell recordings. J. Gen. Physiol. 127, 359–74 (2006). retinopathies. Nat. Chem. Biol. 8, 170–178 (2011). 33. M. M. Abd-El-Barr et al., Genetic dissection of rod and cone pathways in the dark-adapted mouse 56. Y. Chen et al., Mechanism of all-trans-retinal toxicity with implications for stargardt disease and retina. J. Neurophysiol. 102, 1945–1955 (2009). age-related macular degeneration. J. Biol. Chem. 287, 5059–5069 (2012). 34. J. C. Saari et al., Visual cycle impairment in cellular retinaldehyde binding protein (CRALBP) 57. A. Maeda et al., Involvement of all-trans-retinal in acute light-induced retinopathy of mice. J. Biol. knockout mice results in delayed dark adaptation. Neuron 29, 739–748 (2001). Chem. 284, 15173–15183 (2009). 35. Y. Xue et al., CRALBP supports the mammalian retinal visual cycle and cone vision. J. Clin. Invest. 58. P. O’Brien, A. Siraki, N. Shangari, Aldehyde sources, metabolism, molecular toxicity mechanisms, 125, 727–738 (2015). and possible effects on human health. Crit. Rev. Toxicol. 35, 609–662 (2005). 36. R. Amamoto, G. K. Wallick, C. L. Cepko, Retinoic acid signaling mediates peripheral cone 59. Y. Liang et al., Rhodopsin signaling and organization in heterozygote rhodopsin knockout mice. J. photoreceptor survival in a mouse model of retina degeneration. Elife 11, e76389 (2022). Biol. Chem. 279, 48189–48196 (2004). 37. T. M. Redmond et al., Rpe65 is necessary for production of 11-cis-vitamin A in the retinal visual cycle. 60. W. Hao et al., Evidence for two apoptotic pathways in light-induced retinal degeneration. Nat. Genet. Nat. Genet. 20, 344–351 (1998). 32, 254–260 (2002). 38. A. Wenzel, C. E. Reme, T. P. Williams, F. Hafezi, C. Grimm, The Rpe65 Leu450Met variation increases retinal resistance against light-induced degeneration by slowing rhodopsin regeneration. J. Neurosci. 21, 53–58 (2001). 39. G. Moiseyev, Y. Chen, Y. Takahashi, B. X. Wu, J.-X. Ma, RPE65 is the isomerohydrolase in the retinoid visual cycle. Proc. Natl. Acad. Sci. U.S.A. 102, 12413–12418 (2005). 61. P. Ala-Laurila et al., Visual cycle: Dependence of retinol production and removal on photoproduct decay and cell morphology. J. Gen. Physiol. 128, 153–169 (2006). 62. C. Punzo, W. Xiong, C. L. Cepko, Loss of daylight vision in retinal degeneration: Are oxidative stress and metabolic dysregulation to blame? J. Biol. Chem. 287, 1642–1648 (2012). 63. K. Shekhar et al., Comprehensive classification of retinal bipolar neurons by single-cell 40. M. L. Batten et al., Lecithin-retinol acyltransferase is essential for accumulation of all-trans-retinyl transcriptomics. Cell 166, 1308–1323.e30 (2016). esters in the eye and in the liver. J. Biol. Chem. 279, 10422–10432 (2004). 41. R. O. Parker, R. K. Crouch, Retinol dehydrogenases (RDHs) in the visual cycle. Exp. Eye Res. 91, 788–792 (2010). 42. G. J. Jones, R. K. Crouch, B. Wiggert, M. C. Cornwall, G. J. Chader, Retinoid requirements for recovery of sensitivity after visual-pigment bleaching in isolated photoreceptors. Proc. Natl. Acad. Sci. U.S.A. 86, 9606–9610 (1989). 64. T. Hearn et al., Subcellular localization of ALMS1 supports involvement of centrosome and basal body dysfunction in the pathogenesis of obesity, insulin resistance, and type 2 diabetes. Diabetes 54, 1581–1587 (2005). 65. V. J. Knorz et al., Centriolar association of ALMS1 and likely centrosomal functions of the ALMS motif-containing proteins C10orf90 and KIAA1731. Mol. Biol. Cell 21, 3617–3629 (2010). 66. M. P. Krebs et al., Mouse models of human ocular disease for translational research. PLoS One 12, 43. P. Ala-Laurila, M. C. Cornwall, R. K. Crouch, M. Kono, The action of 11-cis-retinol on cone opsins and e0183837 (2017). intact cone photoreceptors. J. Biol. Chem. 284, 16492–16500 (2009). 67. S. Sato, V. J. Kefalov, cis Retinol oxidation regulates photoreceptor access to the retina visual cycle 44. V. Busskamp et al., Genetic reactivation of cone photoreceptors restores visual responses in retinitis and cone pigment regeneration. J. Physiol. 594, 6753–6765 (2016). pigmentosa. Science 329, 413–417 (2010). 68. S. Sato, R. Frederiksen, M. C. Cornwall, V. J. Kefalov, The retina visual cycle is driven by cis retinol 45. Y. Wang et al., A locus control region adjacent to the human red and green visual pigment genes. oxidation in the outer segments of cones. Vis. Neurosci. 34, E004 (2017). Neuron 9, 429–440 (1992). 69. J. Jin, G. J. Jones, M. C. Cornwall, Movement of retinal along cone and rod photoreceptors. Vis. 46. G. J. Ye et al., Cone-specific promoters for gene therapy of achromatopsia and other retinal diseases. Neurosci. 11, 389–399 (1994). Hum. Gene Ther. 27, 72–82 (2016). 70. J. Lem et al., Morphological, physiological, and biochemical changes in rhodopsin knockout mice. 47. T. Etheridge, E. R. Kellom, R. Sullivan, J. N. Ver Hoeve, M. A. Schmitt, Ocular evaluation and Proc. Natl. Acad. Sci. U.S.A. 96, 736–741 (1999). genetic test for an early Alström Syndrome diagnosis. Am. J. Ophthalmol. Case Rep. 20, 100873 (2020). 71. W. Xiong et al., AAV cis-regulatory sequences are correlated with ocular toxicity. Proc. Natl. Acad. Sci. U.S.A. 116, 5785–5794 (2019). 48. G. B. Collin et al., Alms1-disrupted mice recapitulate human Alström syndrome. Hum. Mol. Genet. 72. T. R. Sundermeier et al., R9AP overexpression alters phototransduction kinetics in iCre75 mice. 14, 2323–2333 (2005). Invest. Ophthalmol. Vis. Sci. 55, 1339–1347 (2014). 49. M. Miyamoto et al., Visual electrophysiological features of two naturally occurring mouse models 73. T. R. Sundermeier et al., DICER1 is essential for survival of postmitotic rod photoreceptor cells in with retinal dysfunction. Curr. Eye Res. 31, 329–335 (2006). mice. FASEB J. 28, 3780–3791 (2014). 50. V. J. Kefalov, Rod and cone visual pigments and phototransduction through pharmacological, 74. J. C. Grieger, V. W. Choi, R. J. Samulski, Production and characterization of adeno-associated viral genetic, and physiological approaches. J. Biol. Chem. 287, 1635–1641 (2012). vectors. Nat. Protoc. 1, 1412–1428 (2006). 51. J.-S. Wang, M. E. Estevez, M. C. Cornwall, V. J. Kefalov, Intra-retinal visual cycle required for rapid and 75. T. Matsuda, C. L. Cepko, Controlled expression of transgenes introduced by in vivo electroporation. complete cone dark adaptation. Nat. Neurosci. 12, 295–302 (2009). Proc. Natl. Acad. Sci. U.S.A. 104, 1027–1032 (2007). 52. M. L. Scalabrino et al., Robust cone-mediated signaling persists late into rod photoreceptor 76. S. Wang, C. Sengel, M. M. Emerson, C. L. Cepko, A gene regulatory network controls the binary fate degeneration. Elife 11, e80271 (2022). decision of rod and bipolar cells in the vertebrate retina. Dev. Cell 30, 513–527 (2014). 10 of 10 https://doi.org/10.1073/pnas.2217885120 pnas.org
10.1073_pnas.2301985120	RESEARCH ARTICLE \| BIOCHEMISTRY OPEN ACCESS The membrane electric field regulates the PIP2-binding site to gate the KCNQ1 channel Venkata Shiva Mandalaa,b and Roderick MacKinnona,b,1 Edited by David Clapham, HHMI, Ashburn, VA; received February 3, 2023; accepted April 13, 2023 Voltage-dependent ion channels underlie the propagation of action potentials and other forms of electrical activity in cells. In these proteins, voltage sensor domains (VSDs) regulate opening and closing of the pore through the displacement of their positive-charged S4 helix in response to the membrane voltage. The movement of S4 at hyperpolarizing membrane voltages in some channels is thought to directly clamp the pore shut through the S4–S5 linker helix. The KCNQ1 channel (also known as Kv7.1), which is important for heart rhythm, is regulated not only by membrane voltage but also by the signaling lipid phosphatidylinositol 4,5-bisphosphate (PIP2). KCNQ1 requires PIP2 to open and to couple the movement of S4 in the VSD to the pore. To understand the mechanism of this voltage regulation, we use cryogenic electron microscopy to visualize the movement of S4 in the human KCNQ1 channel in lipid membrane vesicles with a voltage difference across the membrane, i.e., an applied electric field in the membrane. Hyperpolarizing voltages displace S4 in such a manner as to sterically occlude the PIP2-binding site. Thus, in KCNQ1, the voltage sensor acts primarily as a regulator of PIP2 binding. The voltage sensors’ influence on the channel’s gate is indirect through the reaction sequence: voltage sensor movement → alter PIP2 ligand affinity → alter pore opening. KCNQ1 channel \| Kv7.1 channel \| voltage sensor \| cryo-EM \| membrane potential Voltage sensor domains (VSDs) are integral membrane proteins that undergo conforma- tional changes in response to voltage differences across the cell membrane. These domains regulate pore opening and closing in voltage-dependent ion channels (1) and enzymatic activity in voltage-dependent phosphatases (2). Voltage sensors have a conserved structure consisting of four transmembrane (TM) helices (S1 to S4) that form a helical bundle (3–6). The fourth helix, S4, contains a repeated sequence of positive-charged amino acids (typically arginines), every third residue that confers sensitivity to voltage. Inside the lipid bilayer, a gating charge transfer center, composed of aspartate, glutamate, and phenylala- nine residues, stabilizes the arginines one at a time as they traverse the hydrophobic core of the membrane (7, 8). The movement of S4 in response to the TM voltage difference is ultimately responsible for the regulation of protein activity. This mechanism underlies the action potential in neurons (1, 9) and the initiation of muscle contraction (4, 10), among other cellular processes. While the structure of a VSD is highly conserved across all voltage-dependent ion channels, there are two configurations for VSD attachment to the pore of the channel (formed by the S5 and S6 helices). In the so-called domain-swapped channels (Fig. 1A), which include voltage-dependent K+ (Kv) channels 1 to 9, Na+ (Nav) channels, Ca2+ (Cav) channels, and most transient receptor potential channels, the VSD of one subunit interacts with the pore domain of an adjacent subunit, connected through a long interfacial helix— the S4–S5 linker (7, 11–16). Meanwhile, in nondomain-swapped channels (Fig. 1A) such as Kv10-12, Slo1, and hyperpolarization-activated cyclic nucleotide-gated (HCN) chan- nels, the VSD contacts the pore domain of the same subunit through a short S4–S5 loop (17–19). This naturally raises the question: how do the conserved VSDs mediate voltage-dependent gating in these two sets of channels with different structures? We have shown recently that in a nondomain-swapped channel Eag (Kv10.1) (20), the S4 helix on the cytoplasmic side forms an interfacial helix in the hyperpolarized (i.e., negative voltage inside) conformation, which functions as a constrictive cuff around the pore, prevent- ing it from opening. Domain-swapped channels already have an interfacial S4–S5 linker helix that contacts the S6 helix in the depolarized (i.e., no applied or positive inside voltage) con- formation (Fig. 1A) (7, 16), suggesting a gating mechanism that is distinct from that in nondomain-swapped channels. In domain-swapped channels, it has been proposed that the displacement of S4 in response to a hyperpolarizing potential moves the S4–S5 linker helix into a position that clamps the pore shut by pushing down on the S6 helical bundle (7, 16). Structures of domain-swapped channels in detergent micelles at zero mV with chemical Significance Voltage-gated ion channels underlie electrical signaling in cells. The structures and functions of voltage-dependent K+, Na+, and Ca2+ and transient receptor potential ion channels have been studied extensively since their discovery. Despite these efforts, it is still not well understood how the voltage sensors in these different ion channels change their conformation in response to membrane voltage changes, and how these movements regulate the opening or closing of the channel’s gate. This study presents structures of the human KCNQ1 (Kv7.1) voltage– dependent and phosphatidylinositol 4,5-bisphosphate (PIP2)- dependent K+ channel in electrically polarized lipid vesicles using cryogenic electron microscopy, showing how the voltage sensors influence gating indirectly by regulating the ability of PIP2 to bind to the channel. Author affiliations: aLaboratory of Molecular Neuro- biology and Biophysics, The Rockefeller University, New York, NY 10065; and bHHMI, The Rockefeller University, New York, NY 10065 Author contributions: V.S.M. and R.M. designed research; V.S.M. performed research; V.S.M. and R.M. analyzed data; and V.S.M. and R.M. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2301985120/-/DCSupplemental. Published May 16, 2023. PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 1 of 12 cross-links, toxins, mutations, and metal affinity bridges thought to mimic the hyperpolarized condition are supportive of this mecha- nism (21–27). Here, we present a cryo-EM analysis of the domain-swapped human KCNQ1 (Kv7.1) channel in lipid mem- brane vesicles with a hyperpolarizing voltage generated across the membrane, illustrating—at least in some domain-swapped chan- nels—a different gating mechanism than previously thought. Results The Rationale for Polarizing KCNQ1. The KCNQ1 Kv channel, also known as Kv7.1, is the pore-forming subunit of the slow delayed rectifier potassium channel (IKS) (28, 29) that plays an important role in the repolarization phase of cardiac action potentials (30, 31). Mutations in the kcnq1 gene are associated with several congenital cardiac diseases, including long and short QT syndromes as well as familial atrial fibrillation (32). Importantly, KCNQ1 and other Kv7 members are regulated both by membrane voltage and the signaling lipid phosphatidylinositol 4,5-bisphosphate (PIP2) (33–36). The voltage sensors close the channel at hyperpolarizing membrane voltages, while PIP2 is required for the channel to open. When PIP2 is depleted in the membrane, such as when phospholipase C is activated through stimulation of Gq-coupled receptors (34, 37), the voltage sensors undergo voltage-dependent conformational changes, but the pore does not open at depolarizing voltages (38, 39). PIP2 is thus thought to be required for the coupling of voltage sensor movements to pore opening. In other words, KCNQ1 is thought to act as a ligand-regulated voltage-dependent channel, where the binding of PIP2 allows the channel to be gated by the membrane potential. In the absence of PIP2 at zero mV, we would expect a closed pore and depolarized voltage sensors, which is exactly the KCNQ1 struc- ture observed in detergent micelles (12, 40). If we now apply a hyperpolarizing voltage across the membrane, the pore should remain closed, but the voltage sensors should adopt the polarized conformation. Because in this circumstance the voltage sensors do not have to perform mechanical work to close the pore, it should be easier to move the voltage sensors when the pore is already closed (due to the absence of PIP2). We note that we exclude the KCNE beta subunits (41) in this study because they are known to modify the voltage sensitivity of KCNQ1 and thus could trap the voltage sensor in a specific conformation (for instance, KCNE3 appears to stabilize the depolarized conformation) (40, 42). KCNQ1 Reconstitution and Polarization. The human KCNQ1 channel was purified as a complex with the structurally obligate subunit calmodulin (40, 43) in the presence of Ca2+ and reconstituted into liposomes composed of 90: 5: 5 1-palmitoyl-2- Fig. 1. Structures of voltage-dependent ion channels and the preparation of unpolarized and polarized KCNQ1 (Kv7.1) proteoliposomes. (A) The two domain arrangements in voltage-dependent ion channels. Channels are shown with α-helix cylinders and one of the four subunits colored blue and the S4–S5 linker colored red. In domain-swapped channels (Left), the VSD of one subunit interacts with the pore domain of an adjacent subunit and is connected to the pore domain through a long interfacial helix––the S4–S5 linker (red). The structure of Kv1.2 paddle chimera (PDB ID: 2R9R) (7) is shown as an example. In nondomain-swapped channels (Right), the VSD interacts with the pore domain of the same subunit through a short S4–S5 loop. The Eag channel (PDB ID: 8EOW) (20) is shown here as an example. (B) Schematic of the protocol used to obtain polarized vesicles for cryo-EM analysis. Kv7.1 is reconstituted into liposomes with symmetrical KCl, and valinomycin (val.) is added to mediate K+-flux. The external KCl is exchanged for NaCl using a buffer-exchange column. Potassium efflux through valinomycin generates a potential difference across the membrane such that the inside of the vesicle is negative with respect to the outside. Unpolarized and polarized vesicles containing Kv7.1 were frozen on a holey carbon grid for structure determination. (C) Two-dimensional class-averages of membrane-embedded Kv7.1 in unpolarized (Left) and polarized (Right) vesicles from cryo-EM. (D) Liposome-based flux assay to test polarization of vesicles. Recordings (n = 5, mean ± SD) were made using empty vesicles (black) or vesicles with Kv7.1 (blue). Addition of the H+-ionophore CCCP allows entry of protons, which is detected by quenching of the fluorescent reporter ACMA. Protons enter when the K+ ionophore valinomycin is added. 2 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org oleoyl-sn-glycero-3-phosphocholine (POPC) to 1-palmitoyl-2- oleoyl- sn-glycero-3-phosphoglycerol (POPG) to cholesterol [wt: wt: wt] with 300 mM KCl (SI Appendix, Fig. S1 A and B). Following our work on Eag (20), valinomycin was added to the vesicles and the extravesicular solution was exchanged to 300 mM NaCl using a buffer-exchange column (Fig. 1B). The valinomycin-mediated K+-efflux generates a membrane voltage with an upper limit of about −145 mV, such that the inside is negative with respect to the outside. These polarized vesicles were immediately applied to a holey carbon grid and frozen for cryogenic electron microscopy (cryo-EM) analysis (Fig. 1C and SI Appendix, Fig. S1C). Grids for an unpolarized control lacking the buffer exchange step (i.e., with symmetric 300 mM KCl) were also prepared. The permeability of these KCNQ1-containing liposomes to small ions was tested using a liposome flux assay (Fig. 1D) (44). The vesicles prepared in 300 mM KCl (without valinomycin) were diluted into a buffer with a fluorescent dye, 9-amino-6-chloro-2- methoxyacridine (ACMA), and isotonic NaCl to generate a K+ gradient. The proton ionophore carbonyl cyanide m-chlorophenylhydrazone (CCCP) was added to allow H+ influx, which leads to quenched ACMA fluorescence. Without valinomycin, no flux was detected, consistent with the channel being tightly closed under these con- ditions. Subsequent addition of the K+-selective ionophore valino- mycin gave rise to rapid quenching of ACMA in both KCNQ1 proteoliposomes and in control liposomes without protein (Fig. 1D), indicating that the valinomycin-generated membrane potential is stable for at least a few minutes. Identification of Three Structural Classes in the Polarized Dataset. We collected large cryo-EM datasets on polarized and unpolarized vesicles using the same microscope, and the structures of KCNQ1 in both were determined using single- particle analysis (SI Appendix, Figs. S2–S4 and Table S1). As we found for the Kv channel Eag, channels were reconstituted exclusively in an inside-in orientation and thus, when polarized, experience hyperpolarizing (i.e., negative inside) potentials under the applied electric field (Fig. 1C). After two rounds of three- dimensional (3D) classification to select for the best subset of particles in each dataset, we carried out 3D classification without alignment with a mask on the TM domain while imposing C4 symmetry (SI Appendix, Fig. S2). The unpolarized dataset showed little heterogeneity: 88% of the particles were in a homogeneously “up” (detailed below) conformation that closely resembled the detergent structure of KCNQ1, while the remaining particles were in an indeterminate state. Meanwhile, the polarized dataset was noticeably more heterogeneous, with only 34% of particles in a homogeneously up conformation, 19% consistent with an “intermediate” conformation, 10% consistent with a “down” conformation, and the remaining indeterminate. Classification without symmetry on a symmetry-expanded particle set showed that these 34% of particles had all four voltage sensors in an up conformation—suggesting that these channels are in vesicles that have lost the ion gradient, and likely do not reflect the distribution of voltage sensor states under the applied potential. Similar classification on the remaining 66% of particles showed classes with voltage sensors in different states, indicating that the higher proportion of indeterminate particles in the polarized dataset is due to a mixture of conformations. In summary, the observation of distinct structural classes for the voltage sensor in the polarized but not the unpolarized dataset indicates that these conformational changes are likely caused by the application of an electric field (the alternative being due to Na+ in the external solution). From the unpolarized dataset, the best up structure (C4-symmetric; SI Appendix, Fig. S3) had an overall resolution of 2.9 Å (Fig. 2A and SI Appendix, Fig. S4). We solved three structures from the polarized dataset (SI Appendix, Figs. S3 and S4): C4-symmetric up and inter- mediate structures with overall resolutions of 3.4 Å and 6.2 Å, respec- tively, and a C1-symmetric down structure from a symmetry-expanded particle set with an overall resolution of 6.8 Å. The up structures from the unpolarized and polarized datasets are nearly identical (SI Appendix, Fig. S5 A and B), so we focus on the better-resolved former structure. We note that one interesting difference between the two up structures regards the occupancy of K+ ions in the selectivity filter (SI Appendix, Fig. S5 C and D). In the polarized sample, due to the low extravesic- ular concentration of K+, density is only visible at the first and third positions in the selectivity filter, while density is present at all four positions in the unpolarized sample. Similar differences were observed in our previous study on Eag (20) and are qualitatively consistent with crystal structures of KcsA solved under symmetrical high and low K+ concentrations (45). The Up Conformation of the Voltage Sensor. The up map is best defined in the TM domain, with local resolution estimates of ~2.4 to 2.8 Å for much of S1 through S6 (SI Appendix, Fig. S4D). Density for individual hydrogen-bonded water molecules is visible in the voltage sensor (SI Appendix, Fig. S6A). These water molecules do not represent a bulk water-filled crevice, but nevertheless undoubtedly contribute to the stabilization of positive-charged residues (20). Tightly bound phospholipid and sterol molecules (SI Appendix, Fig. S6B) are also visible at both the outer and inner leaflets of the membrane. These features are not discussed further in this paper but are highlighted to demonstrate the feasibility of obtaining high-quality cryo-EM reconstructions in lipid bilayers. A structural model was built by fitting the detergent structure of KCNQ1 (40) and making adjustments where needed (Fig. 2 B–D). The up structure in lipid bilayers is very similar to the depolarized structure in detergent micelles. The S4–S5 linker is an α-helix from I257 to G245 and a short loop (Q244 to D242) connects the S4–S5 linker to S4 (Fig. 2D). The S4 is a 310 helix from V241 to R237 and an α-helix from L236 to T224 (Fig. 2D). The S3–S4 loop is partially flexible—with the four residues (GQVF) in between K218 (top of S3) and A223 (top of S4) not well defined—a point we shall return to later. The six positive-charged residues in S4 are positioned as such (Fig. 2C): R6 (R243) lies below the gating charge transfer center. H5 (H240) occupies the gating charge transfer center consisting of F167 from S2 and the negative-charged E170 and D202 from S2 and S3, respectively. R4 (R237) is directly above the gating charge transfer center and interacts with E160 in S2. Q3 (Q234), R2 (R231), and R1 (R228) lie further toward the extracellular side of the membrane. Q3 lies within the voltage sensor helical bundle, R2 is at the periphery, and R1 is pointed toward the headgroups of the phospholipid bilayer (Figs. 2C and 3 A–C). The Down and Intermediate Conformations of the Voltage Sensor. The intermediate and down maps are less well defined due to heterogeneity, but clear differences in the main chain compared to the up map (modeled in Fig. 3A) were used to build partial models. We compare the down (Fig. 3 E and F) and intermediate (Fig. 3D) maps to an up map that is filtered to a comparable resolution (Fig. 3 B and C). Compared to the up map, the down map shows a dramatic change in the bottom half of S4, near the intracellular surface (Fig. 3 C and F and Movie S1). At the intracellular surface, the loop connecting PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 3 of 12 Fig. 2. Structure of KCNQ1 in lipid vesicles with the voltage sensor in the up conformation. (A) Cryo-EM density map of the up structure of the KCNQ1 channel from the unpolarized dataset. Each channel subunit is shown in a different color and calmodulin (CaM) is shown in magenta. Bound lipids or sterols are shown as gray density. (B) Structure of the KCNQ1–CaM complex (cartoon representation) showing domains within one monomer (blue) from the N- to C-terminus: voltage sensor, pore domain, and the cytosolic domain. The other monomers are colored gray for clarity and the bound calmodulin is colored magenta. (C) Stereoview of the KCNQ1 voltage sensor (Cα trace) in the up (depolarized) conformation. The six positive charges in S4 (α carbon marked by blue spheres), three negative charges in S2 and S3 (E160, E170, and D202), and the hydrophobic Phe in S2 (F167, green sticks) are shown in stick-and-ball representation. (D) Stereoview of the main chain in S4 and the S4–S5 linker (stick representation) in the up conformation. The α carbons of the six positive charges in S4 are marked by blue spheres. Regions with different secondary structures are indicated: α-helix (green), 310 helix (cyan), and loop (magenta). S4 to the S4–S5 linker helix becomes lengthened by eight or nine amino acids. The lengthening occurs while the S4–S5 linker helix on the C-terminal side of W248, whose side chain density is apparent even at the lower resolution of the down map, remains unchanged in its position. Given that the S4–S5 linker helix does not move, the lengthened loop must result from amino acids originating in the downward displacement of the S4 helix and the four residues in the S3–S4 loop. The density in the newly formed extended loop suggests that as S4 moves downward, it forms a broken helix ~30° relative to the bilayer normal, and an extended loop (Fig. 3F and Movie S1). On the extracellular side, the top of helical densities for both S3 and S4 appears embedded about one turn below the expected plane of the extracellular membrane surface, while a short loop connecting them reaches to the extracellular surface (Fig. 3 B and E). 4 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 3. Characterization of electric field–induced movements in the KCNQ1 voltage sensor. (A) Stereoview (Cα trace) of the voltage sensor in the up model. For reference, the positions of the alpha carbons of the six positive charges in S4 are marked by blue spheres, that of K218 at the top of S3 is marked by a red sphere, and that of W248 at the end of the S4–S5 linker is marked by a green sphere. F167 in S2, which is part of the gating charge transfer center, is shown in magenta stick representation. (B and E) Stereoviews of the top part of S3 and S4 in the lowpass-filtered up model and map (unpolarized dataset, B) and in the down model and map (polarized dataset, E). (C, D and F) Stereoview of the bottom part of S4 and the S4–S5 linker in the lowpass-filtered up model and map (unpolarized dataset, C), intermediate (inter.) model and map (polarized dataset, D), and in the down model and map (polarized dataset, F). The up map was lowpass filtered to 6.5 Å to facilitate comparison to the down and intermediate maps. In the absence of side chain density, given the large conforma- tional change in the position of S4 required to form the large loop on the intracellular side, we could not build a model of this region with certainty in the polypeptide register. We built two tentative polyalanine models into the continuous main chain density, one invoking a three helical turn displacement of S4 and the other invoking a two helical turn displacement (SI Appendix, Fig. S7). One and four helical turn models are incompatible with the observed density. The three helical turn displacement, which would place Q3-R6 below, R2 in, and R1 above the gating charge transfer, more reliably accounts for density, but additional data will be needed to establish this conclusion. We note at this point that the mechanism presented in the current study (to be dis- cussed) does not rely on modeling side chains or the detailed register of the S4 helix, because the main chain movements we do observe clearly interfere with the PIP2-binding site and thus explain the basic mechanism of this channel’s gating. In contrast to the down structure, the intermediate structure largely preserves the secondary structure of the up conformation but displays a ~4 Å downward displacement of the loop connecting S4 to the S4–S5 linker (Fig. 3 C and D). As in the down structure, the S4–S5 linker helix does not move appreciably. Given that the motion is likely to be a rigid body movement of S4, we included S4 sidechains in the structural model of the intermediate conformation. The PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 5 of 12 intermediate structure places R6 and H5 below the gating charge transfer center; R4 in the gating charge transfer center; and Q3, R2, and R1 above the gating charge transfer center. In all three structures, the pore appears tightly closed, as expected in the absence of PIP2 (SI Appendix, Fig. S8C). The pore radius is ~1 Å at S349 in the up structure, which is notably smaller than the radius of a hydrated K+ ion (~4 Å). While the side chain of S349 is not visible in the intermediate and down maps, the position of S6 is the same as in the up structure with a closed pore and different from the PIP2-bound structure with an open pore (SI Appendix, Fig. S8C), thus being consistent with a closed pore. Discussion The Relationship between Voltage Sensor Movements and PIP2 Binding. The three structures presented in this study delineate the movement of the S4 helix in the KCNQ1 channel in response to polarization, while the pore of the channel is closed in all the three cases due to the absence of PIP2 in our preparation. The structure of PIP2-bound KCNQ1 with an open pore and the voltage sensor in the up conformation was already determined (40, 46, 47). By comparing these structures, we can deduce how the voltage sensor movements relate to PIP2 binding. The PIP2-binding site in KCNQ1 comprises positive-charged and polar residues in the S4–S5 linker, S4 helix, the S2–S3 foot, and the S0 helix (Fig. 4A, see also Fig. 2C). This structure of the pocket is maintained when the pore of the channel is open or closed as long as the voltage sensor is in the up conformation (Fig. 4B). In other words, when the voltage sensor is up, PIP2 can bind to this pocket and promote channel opening, as described previously (33–36, 40). We note that all the structures of KCNQ1 solved in the presence of PIP2 show an open pore, but only some also show a large conformational change in the cytoplasmic domain (SI Appendix, Fig. S8 C–E) (40, 46). The relationship between the two is not clear, but it is apparent that PIP2-binding causes the pore to open, which is what we focus on here. Overlays of the intermediate (Fig. 4C) and down (Fig. 4D) volt- age sensor conformations with the PIP2-bound, voltage sensor up conformation show that the PIP2-binding site is reshaped when S4 moves. The position of S4 in the down conformation sterically occludes the PIP2-binding site altogether (Fig. 4D). Thus, while the voltage sensor is in the down conformation, PIP2 cannot bind to the channel and open the pore. In the intermediate conformation, the residues that bind PIP2 are displaced relative to one another due to the movement of S4 (Fig. 4C). This intermediate conformational change would likely alter the affinity of the PIP2-binding site, but it might not definitively preclude the binding of PIP2. Voltage-Dependent Regulation of PIP2 Binding in KCNQ1. A mechanism for voltage-dependent regulation of KCNQ1 channel activity thus follows (Fig. 5E). We have made a movie to visualize the sequence of events (Movie S2). At hyperpolarized membrane voltages (i.e., at the resting potential of a cell, corresponding to our polarized vesicles), the voltage sensor is in the down conformation, which prevents PIP2 from binding because the site is occluded. Depolarization drives the S4 helix up, which is coupled to the formation of the PIP2 binding site. PIP2 can then bind, which causes the pore to open through an allosteric mechanism (40, 46, 47). In other words, KCNQ1 activity is modulated by a ligand (PIP2), the binding of which is regulated by the voltage sensor. This is different in detail from a mechanism in which the binding of PIP2 permits voltage sensor conformational changes to regulate the pore through direct mechanical coupling (Fig. 5E). This voltage-dependent regulation of PIP2-binding mechanism is compatible with electrophysiological studies of KCNQ channels, Fig. 4. The relationship between voltage sensor movements and PIP2 binding. (A) Stereoview (gray Cα trace) of the PIP2-bound structure of KCNQ1 (PDB ID: 6V01) (40) with the voltage sensor in the up conformation and an open pore. Positive-charged and polar residues that interact with PIP2 are labeled and shown as gray sticks (α carbon marked by gray spheres) and PIP2 is shown as yellow sticks. (B) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the up conformation and a closed pore (blue Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). (C) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the intermediate conformation and a closed pore (magenta Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). (D) Stereoview of the PIP2-free structure of KCNQ1 with the voltage sensor in the down conformation and a closed pore (green Cα trace) and the PIP2-bound structure shown in panel A (gray Cα trace). In panels (B–D), the α carbon positions of the PIP2-interacting residues are shown as spheres in the same color as the α carbon trace. which show that voltage sensor movements slightly precede pore opening (48, 49), that PIP2 is required for activity (33–36), and that in the absence of PIP2, the voltage sensors move but the pore does not open (38). Moreover, if the voltage sensors were to perform work directly on the pore to open it, and if PIP2 was required for this coupling, one would expect a shift in the voltage activation midpoint (i.e., as measured by the movement of S4) depending on whether PIP2 is bound or not. But, the movement of S4 happens at the same membrane voltage whether PIP2 is present in the mem- brane or not (38), suggesting that the voltage sensors do not perform 6 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 5. The structure of S4 and the S4–S5 linker of Kv7.1 (KCNQ1) and Kv2.1 determined in lipid vesicles. (A) Sequence alignment of S4 and the S4–S5 linker for all domain-swapped Kv channel families. All members of the Kv7 family are included for comparison to Kv7.1. Residues conserved across all families are highlighted in blue. (B) Cryo-EM density map of the human Kv2.1 channel determined in lipid vesicles. Each channel subunit is shown in a different color and associated lipids or sterols are shown as gray density. (C and D) Stereoviews of the connection between S4 and the S4–S5 linker (stick representation) in the depolarized conformations of Kv7.1 (C) and Kv2.1 (D) overlaid with cryo-EM density (blue mesh). (E) Cartoons depicting the new gating model (Top) and the old gating model (Bottom) for KCNQ1. The pore domain is colored orange, the voltage sensor is gray, the S4 helix is blue, the S4–S5 linker is green, and PIP2 is depicted as a magenta hexagon. In the new model, the voltage sensor regulates the binding of PIP2 by occluding the binding site in the down conformation (Left). When membrane depolarization occurs, the voltage sensor moves to the up conformation (Middle), which then allows PIP2 to bind to the channel and open the pore (Right). In the old model that is inconsistent with our data, a PIP2-binding site is present in the down conformation, allowing PIP2 to bind to the channel. work on the pore at depolarized potentials to open it. Finally, voltage clamp fluorometry using a reporter on the S3–S4 linker shows two components: a larger fluorescence change that has a midpoint of ~−60 mV and a smaller change in fluorescence with a midpoint of ~30 mV (50, 51). The pore begins to open during the first (more negative voltage) fluorescence change, which has led to the proposal that there are two open states of the channel. These observations could be related to the intermediate and up voltage sensor confor- mations that we observe. Unique Features of the KCNQ1 Voltage Sensor. KCNQ1 is unique among domain-swapped Kv channels in its requirement of PIP2 to open. To this point, a comparison of the KCNQ1 structure to that of a different domain-swapped Kv channel is informative. A primary sequence alignment from S4 through the S4–S5 linker for the Shaker channel, one member from each of the domain-swapped Kv channel families (Kv1-9), and other members of the Kv7 (KCNQ) family, is given in Fig. 5A. Stereoviews of the S4 to S4–S5 helix linker connection along with cryo-EM density PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 7 of 12 are shown for the depolarized structures of Kv7.1 (Fig. 5C, see also Fig. 2D) and human Kv2.1 (52), both in lipid vesicles (Fig. 5D). The Kv2.1 structure was determined to an overall resolution of 3.0 Å (Fig. 5B and SI Appendix, Fig. S9). An important difference between Kv7.1 and Kv2.1 becomes apparent at the junction of S4 and the S4–S5 linker. In KCNQ1, these residues form a helix– loop–helix motif, with three flexible amino acids (G245, G246, and T247) in or adjacent to the loop. Meanwhile, in Kv2.1, the junction is a helix–turn–helix motif. In other words, Kv7.1 has a natural propensity to form a loop in this region, which is not shared by the domain-swapped channel Kv2.1. The S4 movement that we observe in the intermediate and down conformations is centered exactly at this flexible “GGT” motif. Moreover, this motif (and in fact, most of the S4–S5 linker) is conserved among Kv7 family members but is absent in other domain-swapped Kv channels (Fig. 5A), indicating that it is a hallmark of the PIP2- gated KCNQ channels. Whether Shaker-related channels under an applied electric field undergo similar or distinct voltage sensor movements compared to KCNQ1 remains to be seen. Implications for Other Voltage-Dependent Channels. In KCNQ1, membrane polarization causes the S4 helix to displace by one helical turn (~5 Å) in the intermediate structure and most likely three helical turns (~15 Å) in the down structure, but the S4–S5 linker helix does not move appreciably (SI Appendix, Fig. S8A). One might argue that this is because the pore is closed due to the lack of PIP2. But, structures of KCNQ1 with an open pore are known (40, 46), and the S4–S5 linker helix occupies a similar position in those as well (SI Appendix, Fig. S8B). This finding suggests that the position of the S4–S5 linker helix is not strictly coupled to pore opening and closing in the KCNQ1 channel. It is still possible that small movements in the S4–S5 linker bias the conformational state of the pore, but S4–S5 helix movements are minimal when the S4 helix is displaced. What does this static S4–S5 helix in KCNQ1, if anything, suggest for other domain-swapped channels like Shaker-related Kv channels, Nav, or Cav channels? When the first molecular structure of a eukar- yotic voltage-dependent ion channel—the Shaker-related Kv1.2— was determined (16), the S4–S5 linker was found to contact S6 directly. A simple mechanical model for voltage-dependent regula- tion of the pore was proposed: When S4 moves in response to an electric field, the amino terminal end of the S4–S5 linker is displaced, applying a force on S6 and causing pore closure through straighten- ing of the S6 helix at a conserved “PxP” motif (53). Many years later, structures of chemically cross-linked or trapped Nav channels (21, 24, 25) and metal bridge–linked Kv4.2 channels (26) showed that it is indeed possible to trap channels in conformations consistent with the simple mechanical model. We observe in the present study, however, that KCNQ1 does not function according to this model. While KCNQ1 is an outlier among domain-swapped voltage-dependent channels for the reasons discussed above, we remain open minded to the possibility that the simple mechanical model assumed for other domain-swapped voltage-dependent ion channels (16, 21, 24, 25), despite support from mutational and chemical crossbridge data, could be incorrect. Mutations and chem- ical crossbridges likely do not replicate the forces applied to a polar- ized voltage sensor in membranes because an electric force field acts on all charged atoms spread throughout the protein. Ultimately, to know whether the simple mechanical model is correct for other domain-swapped channels, we will need to determine their structures in lipid bilayers with an applied electrostatic force field. Comparison of Voltage Sensor Movements in EAG and KCNQ1. We now know how the voltage sensors in two potassium channels, KCNQ1 (domain-swapped) and Eag (nondomain-swapped) (20), undergo conformational changes in response to an applied voltage difference across the membrane. For comparison, side views of the voltage sensors in these channels are shown in Fig. 6. In both channels, as S4 displaces downward (i.e., toward the cytoplasm), an extended interfacial segment is formed through a break in S4, which is accompanied by a remodeling of the connection between S4 and the S4–S5 linker helix (KCNQ1; Fig. 6A) or S4 and S5 (Eag; Fig. 6B) (20). Apparently, because S4 is both charged and hydrophobic, an interfacial location is energetically more favorable than an aqueous location. The extra amino acids that account for the downward displacement of S4 originate from the S3–S4 linker and the top of S3 in both channels. While the S4 displacement and interfacial helix formation are similar in KCNQ1 and Eag, the helices bend in opposite direc- tions with respect to the pore. In Eag, the polarized S4 bends toward the pore axis, causing it to clamp down on the pore-lining S6 helix, which prevents pore opening (Fig. 6B) (20). In KCNQ1, the polarized S4 bends away from the pore axis so that it occludes the PIP2-binding site. These variations show how the same struc- tural element—a voltage sensor—confers conformational sensi- tivity to an electric field in two Kv channels that differ both in their voltage sensor configuration (i.e., domain-swapped versus nondomain-swapped) and in their modulation by other effectors. Future studies of other voltage-dependent ion channels might uncover other interesting mechanisms for coupling the movement of S4 to gating the pore. On the Magnitude of S4 Displacement in KCNQ1. As we state above, our inability to define the register of the S4 helix main chain (SI Appendix, Fig. S7) prevents us from distinguishing with certainty whether the down map, with its occluded PIP2-binding site, corresponds to a two or three helical turn displacement of S4. A two-turn displacement was anticipated because that is what we observed in a polarized Eag channel (20), and what has been seen in cross-linked Nav and HCN voltage sensors (21, 23–25). Moreover, if S4 can displace three helical turns, it must pass through a two-turn-displaced intermediate. Why then would we not observe this intermediate? A possible answer lies in the unique S4 sequence of KCNQ1, which contains a neutral glutamine at “charged” position 3 (Fig. 5A). A two helical turn displacement would place the neutral glutamine into the highly negative- charged gating charge transfer center (Fig. 2C). For this reason, conformations with one or three helical turn displacements (which both place an arginine in the gating charge transfer center) may be energetically more stable in KCNQ1 than a conformation with two. If this is the case, a two helical turn displacement would function as a transient energy barrier in the conformational change of the voltage sensor. The notion that different residues are stable to varying degrees when they occupy the gating charge transfer center is apparent from functional measurements in other voltage-dependent ion channels. For instance, in the Shaker channel, it has been shown that it is more favorable for a lysine than an arginine to occupy the gating charge transfer center (8). Depending on the position of the substitution within the S4 helix, either the open or the closed state of the channel can be stabilized (corresponding to an up or down conformation of the voltage sensor). Given that even two positive-charged residues, arginine or lysine, can differ in their relative stability, it seems quite possible that a neutral glutamine behaves differently than an arginine. It is also useful to look at another example: Consider the Shaker channel and the domain-swapped channel Kv2.1 (Fig. 5A). Both have lysine at the fifth position (K5) in the gating 8 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org Fig. 6. Comparison of voltage sensor movements in KCNQ1 and Eag. (A and B) Stereoviews of KCNQ1 (A) and Eag (B) showing the up conformation (Left, red S4) and the down conformation (Right, blue S4). The view looks through one voltage sensor in each channel toward the pore axis. The approximate position of the lipid membrane bilayer is marked by yellow lines and the protein is shown in Cα trace representation. charge transfer center. In Shaker, all four residues above the gating charge transfer center are arginines, while Kv2.1 has a glutamine at position one (Q1) followed by three arginines. The gating charge estimated by nonlinear membrane capacitance for Shaker is ~12 to 14 elementary charges per channel (~3 to 3.5 per voltage sensor) and that for Kv2.1 is only ~6 to 7 per channel (~1.5 to 2 per voltage sensor) (54, 55). The gating charge esti- mates for Shaker indicate that the first residue (R1) does not traverse the membrane potential. Yet the presence of a neutral glutamine at the first position in Kv2.1 reduces the apparent gating charge in half. We suppose that in the down conformation of Kv2.1, there is a tendency for R2 to neutralize the extracellular negative-charged residue while R3 occupies the gating charge transfer center, consistent with the net movement of ~2 gating charges. These observations are consistent with the idea that it is more favorable for an arginine (compared to a glutamine) to interact with negative-charged residues in the voltage sensor. In summary, this study provides the structural description of a domain-swapped Kv channel in a lipid bilayer under the influence of a polarizing electric field. The structures reveal a mechanism in which the voltage sensor regulates the affinity of PIP2, thus con- trolling its ability to gate the pore. Materials and Methods Cell Lines. Sf9 (Spodoptera frugiperda Sf21) cells were used for production of baculovirus and were cultured in Sf-900 II SFM medium (GIBCO) supplemented with 100 U/mL penicillin and 100 U/mL streptomycin at 27 °C under atmospheric CO2. HEK293S GnTl− cells were used for protein expression and were cultured in Freestyle 293 medium (GIBCO) supplemented with 2% fetal bovine serum, 100 U/mL penicillin, and 100 U/mL streptomycin at 37 °C in 8% CO2. Expression and Purification of the KCNQ1–Calmodulin Complex. The KCNQ1(Kv7.1)–calmodulin complex (hereby referred to as KCNQ1) was expressed and purified as described before (40), with slight modifications. We used a con- struct corresponding to human KCNQ1 with N-terminal and C-terminal trunca- tions, leaving residues 76 to 620. The construct was cloned into the BacMan expression vector with a C-terminal green fluorescent protein (GFP)-His6 tag linked by a preScission protease (PPX) site (56). A separate BacMan expression vector without a tag was used for vertebrate calmodulin (CaM). Bacmids were generated for KCNQ1 and CaM using DH10Bac Escherichia coli cells. Baculoviruses for KCNQ1 and CaM were produced in SF9 cells transfected with bacmid DNA using the Cellfectin II reagent (Invitrogen). Baculovirus was amplified three times in suspension cultures of SF9 cells grown at 27 °C. Four liters of suspension cultures of HEK293S GnTI- at ~3 × 106 cells/mL were infected with 12% (v/v) of 5:1 KCNQ1:CaM baculovirus at 37 °C for ~8 h. Protein expres- sion was induced by adding 10 µM sodium butyrate, and the incubation tem- perature was changed to 30 °C for the duration of expression. Cell pellets were harvested ~48 h after induction and flash frozen in liquid nitrogen for later use. Four liters of cell pellet were resuspended in ~100 mL of lysis buffer (25 mM Tris pH 8.0, 300 mM KCl, 1 mM MgCl2, 5 mM CaCl2, 2 mM dithiothreitol (DTT), 1 µg/mL leupeptin, 1 µg/mL pepstatin, 1 mM benzamidine, 1 µg/mL aprotinin, 1 mM phenylmethylsulfonyl fluoride, 1 mM 4-(2-aminoethyl) benzenesulfo- nyl fluoride, and 0.1 mg/mL DNase), stirred for 10 min at 4 °C, and Dounce homogenized with a loose pestle till homogenous. The resultant suspension was clarified by centrifugation at 39,800 × g for 15 min at 4 °C. The pellet was resuspended in ~100 mL of lysis buffer and Dounce homogenized with a tight PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 9 of 12 pestle. To extract the KCNQ1–CaM complex, we added 15 mL of a 10%:2% n-no- decyl-β-D-maltopyranoside (DDM):cholesteryl hemisuccinate (CHS) mixture and stirred for 1 h at 4 °C. The mixture was clarified by centrifugation at 39,800 × g for 30 min at 4 °C. The supernatant was bound to ~2.5 mL GFP nanobody-coupled Sepharose resin (prepared in-house) (57) in a 250-mL conical centrifuge tube (Corning) by gentle rotation for 1 h at 4 °C. The resin was transferred to a glass gravity flow column (Bio-Rad) and washed with ~30 column volumes of wash buffer (10 mM Tris pH 8.0, 300 mM KCl, 0.05%:0.01% DDM:CHS, 1 mM CaCl2, and 2 mM DTT). The resin was resuspended in five column volumes of wash buffer, PPX (prepared in-house) was added at a concentration of 0.05 mg/mL to remove the GFP tag, and the solution was rotated for 1 h at 4 °C. The cleaved protein was collected in the flow through and a subsequent wash step with five column volumes of wash buffer. The protein was concentrated to ~500 µL at 3,000 × g and 4 °C using a 15-mL Amicon spin concentrator with a 100-kDa molecular weight cutoff mem- brane. The concentrated protein was filtered through a Corning 0.2 µm spin filter and then purified by size-exclusion chromatography (SEC) using a Superose 6 Increase column (10/300 GL) preequilibrated with SEC buffer (10 mM Tris pH 8.0, 300 mM KCl, 0.025%:0.005% DDM:CHS, 1 mM CaCl2, and 5 mM DTT). Fractions containing hKCNQ1 and calmodulin (SI Appendix, Fig. S1 A and B) were pooled and concentrated to an A280 of 3.8 mg/mL at 3,000 × g and 4 °C using a 4-mL Amicon spin concentrator with a 100-kDa molecular weight cutoff membrane. Purified protein was immediately used for reconstitution into liposomes. Reconstitution of the KCNQ1–Calmodulin Complex and Cryo-EM Grid Preparation. The purified KCNQ1 complex was reconstituted into liposomes con- sisting of 90%: 5%: 5% POPC:POPG:cholesterol (wt: wt: wt, Avanti Polar Lipids) (20). The phospholipids and sterol were mixed together in chloroform at a concentration of 10 mg/mL. Ten milligrams of the lipid mixture were dried to a thin film in a screw-top glass tube under a gentle stream of argon. The lipid film was further dried for ~3 h in a room-temperature vacuum desiccator, and then resuspended at a concentra- tion of 10 mg/mL by gentle vortexing in reconstitution buffer (10 mM Tris pH 8.0, 300 mM KCl and 1 mM DTT). Small unilamellar vesicles (SUVs) were formed by bath sonication (Branson Ultrasonics M1800) at room temperature till the solution was mostly transparent (A400 ~ 0.2), which typically took ~40 min. To permeabilize but not solubilize the lipid vesicles, the detergent C12E10 was added to the 10 mg/mL lipid stock solution to a final concentration of 2 mg/mL (5:1 lipid:detergent, wt/wt) and incubated on ice for ~15 min. Two hundred microliters of this permeabilized vesicle solution was mixed with 27 µL of the purified KCNQ1 complex (3.8 mg/ mL) and 173 µL of reconstitution buffer, giving a total reaction volume of 400 µL (chosen to ensure proper mixing in a 1.5 mL Eppendorf tube), a protein:lipid ratio of 1:20 (wt/wt), and a final lipid concentration of 5 mg/mL. The lipid–protein–detergent mixture was incubated on ice for ~1.5 h. Detergent was removed using adsorbent Bio-Beads SM-2 resin (Bio-Rad) by adding 20 mg of a 50% (wt/vol) Bio-Beads slurry in reconstitution buffer and rotating at 4 °C for ~14 h. The biobeads procedure was repeated twice again for 3 h each at 4 °C to ensure complete removal of detergent. The suspension was bath sonicated briefly (twice for 10 s each) after the biobeads step to minimize vesicle clumping. Polarized and unpolarized vesicles were prepared from the same batch of proteoliposomes. From an 8 mM stock in dimethyl sulfoxide, 2 µM valinomycin was added to the proteoliposomes and incubated for 30 min on ice. Polarized vesicles were prepared as follows: 70 µL of the above solution was added to a 0.5-mL Zeba spin desalting column (40 kDa cutoff, Thermo Scientific), preequili- brated with sodium reconstitution buffer (10 mM Tris pH 8.0 and 300 mM NaCl), to exchange the external K+ for Na+. The sample was centrifuged for ~20 to 30 s at room temperature at 1,500 × g and ~20 µL of flow-through containing vesicles was collected. The residual external K+ concentration is about 1 mM (20). Onto a glow-discharged Quantifoil R1.2/1.3 400 mesh Au grid, 3.5 µL of the polarized vesicle solution was immediately applied. The vesicle solution was incubated on the grid for 3 min at 20 °C under a humidity of 100%. The grid was then manually blotted from the edge of the grid using a piece of filter paper. Another 3.5 µL of the polarized vesicle solution was applied to the same grid for 20 s (58), and then the grid was blotted for 3 s with a blotting force of 0 and flash frozen in liquid ethane using a FEI Vitrobot Mark IV (FEI). Each grid with polarized vesicles used a freshly buffer exchanged sample. Grids for the unpolarized vesicles were frozen by skipping the buffer exchange step, i.e., directly applying the proteoliposomes (with valinomycin) on the Quantifoil grids. Expression and Purification of Kv2.1. Full-length human Kv2.1 (NP_004966.1) with a C-terminal GFP-His6 tag linked by a PPX site and full-length 14-3-3 protein epsilon (empirically found to increase the yield of Kv2.1, XP_040497056.1) were both cloned into a pBig1a vector from the biGBac system (59). Bacmids and baculovirus were generated, and protein was expressed in HEK293S GnTI- cells as described above for KCNQ1. The channel (hKv2.1) was purified following essentially the same protocol as KCNQ1 except that 150 mM KCl (instead of 300 mM KCl) was used for the wash buffer and calcium chloride was not included after the lysis buffer step. The final purification step entailed SEC on a Superose 6 Increase column (10/300 GL) preequilibrated with SEC buffer (10 mM Tris pH 8.0, 150 mM KCl, 0.03%:0.006% DDM:CHS, and 5 mM DTT). Fractions containing hKv2.1 (SI Appendix, Fig. S9A) were pooled and concentrated to an A280 of 1.4 mg/mL at 3,000 × g and 4 °C. Reconstitution of Kv2.1 and Cryo-EM Grid Preparation. Purified protein was reconstituted into liposomes of 90%:5%:5% POPC:POPG:cholesterol prepared in 150 mM KCl using a protein:lipid ratio of 1:20 (wt/wt), following the same protocol as for KCNQ1. A fourfold-molar excess of hanatoxin (compared to hKv2.1 monomers) isolated from Chilean rose tarantula (Grammostola rosea) venom (60) was incubated with the proteoliposomes before freezing grids, but the toxin was not visible in the cryo-EM reconstructions. Grids were frozen exactly as described for unpolarized vesicles containing KCNQ1 (but without added valinomycin). Liposome Flux Assay. The flux assay was carried out as described before (44), with minor modifications. The proteoliposome vesicles or control vesicles without protein (subjected to a mock reconstitution) prepared in 300 mM KCl were diluted 10-fold in isotonic sodium buffer (10 mM Tris pH 8.0 and 300 mM NaCl) imme- diately prior to the assay. Six microliters of the diluted vesicle solution was mixed with 6 µL ACMA solution (10 mM Tris pH 8.0, 300 mM NaCl, and 5 mM ACMA) and 12 µL buffer (10 mM Tris pH 8.0 and 300 mM NaCl). ACMA fluorescence was recorded every 5 s (excitation wavelength = 410 nm, emission wavelength = 490 nm) using a 384-well plate (Grainger) on a fluorescence plate reader (Tecan Infinite M1000). After the ACMA fluorescence stabilized, 6 µL of CCCP solution (10 mM Tris pH 8.0, 300 mM NaCl, and 15 mM CCCP) was added. The resultant KCNQ1-dependent flux, or in this case, the lack thereof because of the absence of PIP2, was measured. At the end of the assay, 2 µL of a 1.2-µM valinomycin solution (in 10 mM Tris pH 8.0 and trace dimethyl sulfoxide) was added to initiate K+ efflux from all the vesicles and determine the minimum ACMA fluorescence. The fluorescence data for each run were normalized by the fluorescence value right before the addition of CCCP (i.e., at 90 s). The normalized data were averaged across five independent measurements, and the mean and SDs are reported. Cryo-EM Data Acquisition and Processing. Data for the polarized and unpo- larized KCNQ1 liposomes were collected on the same microscope—a 300-keV FEI Titan Krios2 microscope located at the HHMI Janelia Research Campus. The microscope was equipped with a Gatan Image Filter (GIF) BioQuantum energy filter and a Gatan K3 camera. A total of 33,057 movies (polarized sample) or 19,998 movies (unpolarized sample) were recorded on Quantifoil grids in super- resolution mode using SerialEM (61). The movies were recorded with a physical pixel size of 0.839 Å (superresolution pixel size of 0.4195 Å) and a target defocus range of −1.0 to −2.0 µm. The total exposure time was ~2 s (fractionated into 50 frames) with a cumulative dose of ~60 e−/Å2. The data-processing workflow is detailed in SI Appendix, Figs. S2 and S3, and followed the same strategy we previously reported for Eag (20). Data processing was carried out using cryoSPARC v3.3.1 (62) and RELION 4.0 (63). The super- resolution movies were gain-normalized, binned by a factor of 2 with Fourier cropping, and corrected for full-frame and sample motion using the Patch motion correction tool (grid = 15 × 10). Contrast transfer function parameters were esti- mated from the motion-corrected micrographs using the Patch CTF estimation tool, which uses micrographs without dose weighting. All subsequent processing was performed on motion-corrected micrographs with dose weighting. Particle picking was initially carried out using the Blob picker. 2D classes with clear protein density were used to train a TOPAZ picking model (64), which was used to pick additional particles. Particles with clear protein density after 2D classification were pooled and duplicate picks were removed. An ab initio model was generated from 2D classes with clear secondary structure features, and 3D classification and refinement was carried out either in cryoSPARC or RELION as detailed in SI Appendix, Figs. S2 and S3. 10 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org Data for the hKv2.1 liposomes were collected on a 300-keV FEI Titan Krios microscope located at the HHMI Janelia Research Campus. The microscope was equipped with a spherical aberration corrector (Cs corrector), a GIF BioQuantum energy filter, and a Gatan K3 camera. A total of 17,007 movies were recorded on a single Quantifoil grid in superresolution mode using SerialEM. The movies were recorded with a physical pixel size of 0.844 Å (superresolution pixel size of 0.422 Å) and a target defocus range of −1.0 to −2.0 µm. The total exposure time was ~2 s (fractionated into 50 frames) with a cumulative dose of ~60 e−/Å2. Data processing was carried out as described for KCNQ1 liposomes. Model Building and Refinement. A structural model for the up conformation was built by docking the structure of KCNQ1–CaM in detergent micelles (PDB ID: 6UZZ) (40) into the up map and making adjustments where needed. The model was edited and refined using the ISOLDE (65) plugin in ChimeraX v1.2.0 (66) or WinCoot v0.98.1 (67) followed by real-space refinement in Phenix (68). The down and intermediate models were built starting from the up model. The up model was initially fit in the intermediate or down maps as a rigid body using Phenix. The S4 helix and the surrounding regions were manually adjusted and then a final step of real-space refinement was carried out in Phenix. The quality of the final models was evaluated using the MolProbity plugin in Phenix (SI Appendix, Table S1). Graphical representations of models and cryo-EM density maps were prepared using PyMOL (69) and ChimeraX. Data, Materials, and Software Availability. Cryo-EM density maps of the KCNQ1 channel with the voltage sensor in the up, intermediate, and down conformation have been deposited in the electron microscopy data bank under accession codes EMD-40508 (70), EMD-40509 (71), and EMD-40510 (72), respectively. Atomic coordinates of the KCNQ1 channel with the voltage sen- sor in the up, intermediate, and down conformation have been deposited in the protein data bank under accession codes 8SIK (73), 8SIM (74), and 8SIN (75), respectively. ACKNOWLEDGMENTS. We thank Rui Yan, Zhiheng Yu, and the team at the Howard Hughes Medical Institute Janelia CryoEM Facility for their effort in cryo-EM microscope operation and data collection; Mark Ebrahim, Johanna Sotiris, and Honkit Ng at the Evelyn Gruss Lipper Cryo-EM Resource Center for assistance with cryo-EM grid screening; Yi Chun Hsiung for assistance with insect and mammalian cell cultures; Dr. Chen Zhao and other members of the MacKinnon Lab for helpful discussions; and Dr. Jue Chen and her group for their input. Dr. Chia-Hseuh Lee carried out the cloning and initial biochemical characterization of the Kv2.1 channel. V.S.M. is supported by the Jane Coffin Childs Memorial Fund Fellowship. R.M. is an investigator in the HHMI. 1. 2. B. Hille, Ionic Channels of Excitable Membranes (Sinauer Associates, ed. 3, 2001) (January 27, 2023). Y. Murata, H. Iwasaki, M. Sasaki, K. Inaba, Y. Okamura, Phosphoinositide phosphatase activity coupled to an intrinsic voltage sensor. Nature 435, 1239–1243 (2005). 30. T. Jespersen, M. Grunnet, S.-P. Olesen, The KCNQ1 potassium channel: From gene to physiological function. Physiology 20, 408–416 (2005). 31. J. Robbins, KCNQ potassium channels: Physiology, pathophysiology, and pharmacology. Pharmacol. Ther. 90, 1–19 (2001). 3. M. Noda et al., Primary structure of electrophorus electricus sodium channel deduced from cDNA 32. P. L. Hedley et al., The genetic basis of long QT and short QT syndromes: A mutation update. Hum. 4. 5. 6. 7. 8. 9. sequence. Nature 312, 121–127 (1984). T. Tanabe et al., Primary structure of the receptor for calcium channel blockers from skeletal muscle. Nature 328, 313–318 (1987). B. L. Tempel, D. M. Papazian, T. L. Schwarz, Y. N. Jan, L. Y. Jan, Sequence of a probable potassium channel component encoded at Shaker locus of Drosophila. Science 237, 770–775 (1987). Y. Jiang et al., X-ray structure of a voltage-dependent K+ channel Nature 423, 33–41 (2003). S. B. Long, X. Tao, E. B. Campbell, R. MacKinnon, Atomic structure of a voltage-dependent K+ channel in a lipid membrane-like environment Nature 450, 376–382 (2007). X. Tao, A. Lee, W. Limapichat, D. A. Dougherty, R. MacKinnon, A gating charge transfer center in voltage sensors. Science 328, 67–73 (2010). A. L. Hodgkin, A. F. Huxley, A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117, 500–544 (1952). Mutat. 30, 1486–1511 (2009). 33. H. Zhang et al., PIP2 activates KCNQ channels, and its hydrolysis underlies receptor-mediated inhibition of M currents. Neuron 37, 963–975 (2003). 34. B.-C. Suh, B. Hille, Recovery from muscarinic modulation of M current channels requires phosphatidylinositol 4,5-bisphosphate synthesis. Neuron 35, 507–520 (2002). 35. G. Loussouarn et al., Phosphatidylinositol-4,5-bisphosphate, PIP2, controls KCNQ1/KCNE1 voltage- gated potassium channels: A functional homology between voltage-gated and inward rectifier K+ channels EMBO J. 22, 5412–5421 (2003). 36. C. P. Ford, P. L. Stemkowski, P. A. Smith, Possible role of phosphatidylinositol 4,5, bisphosphate in luteinizing hormone releasing hormone-mediated M-current inhibition in bullfrog sympathetic neurons. Eur. J. Neurosci. 20, 2990–2998 (2004). 37. P. Delmas, D. A. Brown, Pathways modulating neural KCNQ/M (Kv7) potassium channels. Nat. Rev. 10. K. G. Beam, C. M. Knudson, J. A. Powell, A lethal mutation in mice eliminates the slow calcium Neurosci. 6, 850–862 (2005). current in skeletal muscle cells. Nature 320, 168–170 (1986). 38. M. A. Zaydman et al., Kv7.1 ion channels require a lipid to couple voltage sensing to pore opening. 11. J. Payandeh, T. Scheuer, N. Zheng, W. A. Catterall, The crystal structure of a voltage-gated sodium Proc. Natl. Acad. Sci. U.S.A. 110, 13180–13185 (2013). channel. Nature 475, 353–358 (2011). 39. J. Cui, Voltage-dependent gating: Novel insights from KCNQ1 channels. Biophys. J. 110, 14–25 12. J. Sun, R. MacKinnon, Cryo-EM structure of a KCNQ1/CaM complex reveals insights into congenital (2016). long QT syndrome. Cell 169, 1042–1050.e9 (2017). 40. J. Sun, R. MacKinnon, Structural basis of human KCNQ1 modulation and gating. Cell 180, 340–347. 13. J. Wu et al., Structure of the voltage-gated calcium channel Cav1.1 at 3.6 Å resolution. Nature 537, e9 (2020). 191–196 (2016). 14. H. Shen et al., Structure of a eukaryotic voltage-gated sodium channel at near-atomic resolution. Science 355, eaal4326 (2017). 41. Z. A. McCrossan, G. W. Abbott, The MinK-related peptides. Neuropharmacology 47, 787–821 (2004). 42. R. Barro-Soria, M. E. Perez, H. P. Larsson, KCNE3 acts by promoting voltage sensor activation in KCNQ1. Proc. Natl. Acad. Sci. U.S.A. 112, E7286–E292 (2015). 15. M. Liao, E. Cao, D. Julius, Y. Cheng, Structure of the TRPV1 ion channel determined by electron cryo- 43. L. Shamgar et al., Calmodulin is essential for cardiac IKS channel gating and assembly: Impaired microscopy. Nature 504, 107–112 (2013). function in long-QT mutations. Circ. Res. 98, 1055–1063 (2006). 16. S. B. Long, E. B. Campbell, R. MacKinnon, Voltage sensor of Kv1.2: Structural basis of electromechanical coupling. Science 309, 903–908 (2005). 44. Z. Su, E. C. Brown, W. Wang, R. MacKinnon, Novel cell-free high-throughput screening method for pharmacological tools targeting K + channels Proc. Natl. Acad. Sci. U.S.A. 113, 5748–5753 (2016). 17. C.-H. Lee, R. MacKinnon, Structures of the human HCN1 hyperpolarization-activated channel. Cell 45. J. H. Morais-Cabral, Y. Zhou, R. MacKinnon, Energetic optimization of ion conduction rate by the K+ 168, 111–120.e11 (2017). selectivity filter Nature 414, 37–42 (2001). 18. X. Tao, R. K. Hite, R. MacKinnon, Cryo-EM structure of the open high-conductance Ca2+-activated 46. D. Ma et al., Structural mechanisms for the activation of human cardiac KCNQ1 channel by electro- K+ channel Nature 541, 46–51 (2017). mechanical coupling enhancers. Proc. Natl. Acad. Sci. U.S.A. 119, e2207067119 (2022). 19. J. R. Whicher, R. MacKinnon, Structure of the voltage-gated K+ channel Eag1 reveals an alternative 47. Y. Zheng et al., Structural insights into the lipid and ligand regulation of a human neuronal KCNQ voltage sensing mechanism Science 353, 664–669 (2016). channel. Neuron 110, 237–247.e4 (2022). 20. V. S. Mandala, R. MacKinnon, Voltage-sensor movements in the Eag Kv channel under an applied electric field. Proc. Natl. Acad. Sci. U.S.A. 119, e2214151119 (2022). 21. T. Clairfeuille et al., Structural basis of α-scorpion toxin action on Nav channels. Science 363, eaav8573 (2019). 22. J. Guo et al., Structure of the voltage-gated two-pore channel TPC1 from Arabidopsis thaliana. Nature 531, 196–201 (2016). 48. K. J. Ruscic et al., IKs channels open slowly because KCNE1 accessory subunits slow the movement of S4 voltage sensors in KCNQ1 pore-forming subunits. Proc. Natl. Acad. Sci. U.S.A. 110, E559–E566 (2013). 49. F. Miceli, E. Vargas, F. Bezanilla, M. Taglialatela, Gating currents from Kv7 channels carrying neuronal hyperexcitability mutations in the voltage-sensing domain. Biophys. J. 102, 1372–1382 (2012). 50. P. Hou et al., Inactivation of KCNQ1 potassium channels reveals dynamic coupling between voltage 23. C.-H. Lee, R. MacKinnon, Voltage sensor movements during hyperpolarization in the HCN channel. sensing and pore opening. Nat. Commun. 8, 1730 (2017). Cell 179, 1582–1589.e7 (2019). 24. G. Wisedchaisri et al., Resting-state structure and gating mechanism of a voltage-gated sodium channel. Cell 178, 993–1003.e12 (2019). 51. M. A. Zaydman et al., Domain–domain interactions determine the gating, permeation, pharmacology, and subunit modulation of the IKs ion channel. Elife 3, e03606 (2014). 52. H. Misonou, D. P. Mohapatra, J. S. Trimmer, Kv2.1: A voltage-gated K+ channel critical to dynamic 25. H. Xu et al., Structural basis of Nav1.7 inhibition by a gating-modifier spider toxin. Cell 176, control of neuronal excitability NeuroToxicology 26, 743–752 (2005). 702–715.e14 (2019). 26. W. Ye et al., Activation and closed-state inactivation mechanisms of the human voltage-gated KV4 53. Y. Jiang et al., The open pore conformation of potassium channels. Nature 417, 523–526 (2002). 54. S. K. Aggarwal, R. MacKinnon, Contribution of the S4 segment to gating charge in the shaker K+ channel complexes. Mol. Cell 82, 2427–2442.e4 (2022). channel Neuron 16, 1169–1177 (1996). 27. G. Huang et al., Unwinding and spiral sliding of S4 and domain rotation of VSD during the 55. S. K. Aggarwal, Analysis of the Voltage Sensor in a Voltage-Activated Potassium Channel (Harvard electromechanical coupling in Nav1.7. Proc. Natl. Acad. Sci. U.S.A. 119, e2209164119 (2022). 28. J. Barhanin et al., KvLQT1 and IsK (minK) proteins associate to form the IKS cardiac potassium current. Nature 384, 78–80 (1996). University, Cambridge, MA, 1996) (January 27, 2023). 56. A. Goehring et al., Screening and large-scale expression of membrane proteins in mammalian cells for structural studies. Nat. Protoc. 9, 2574–2585 (2014). 29. M. C. Sanguinetti et al., Coassembly of KVLQT1 and minK (IsK) proteins to form cardiac IKS 57. A. Kirchhofer et al., Modulation of protein properties in living cells using nanobodies. Nat. Struct. potassium channel. Nature 384, 80–83 (1996). Mol. Biol. 17, 133–138 (2010). PNAS 2023 Vol. 120 No. 21 e2301985120 https://doi.org/10.1073/pnas.2301985120 11 of 12 58. L. Tonggu, L. Wang, Cryo-EM sample preparation method for extremely low concentration 67. P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features and development of Coot. Acta Crystallogr. D liposomes. Ultramicroscopy 208, 112849 (2020). Biol. Crystallogr. 66, 486–501 (2010). 59. F. Weissmann et al., biGBac enables rapid gene assembly for the expression of large multisubunit 68. D. Liebschner et al., Macromolecular structure determination using X-rays, neutrons and protein complexes. Proc. Natl. Acad. Sci. U.S.A. 113, E2564–E2569 (2016). 60. K. J. Swartz, R. MacKinnon, An inhibitor of the Kv2.1 potassium channel isolated from the venom of a Chilean tarantula. Neuron 15, 941–949 (1995). 61. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of electrons: Recent developments in Phenix. Acta Crystallogr. Sect. Struct. Biol. 75, 861–877 (2019). 69. L. L. C. Schrödinger, The PyMOL Molecular Graphics System (Ver. 2, Schrödinger, 2015). 70. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the up conformation. EMDB. https:// specimen movements. J. Struct. Biol. 152, 36–51 (2005). www.ebi.ac.uk/emdb/EMD-40508. Deposited 14 April 2023. 62. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: Algorithms for rapid unsupervised 71. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the intermediate conformation. EMDB. cryo-EM structure determination. Nat. Methods 14, 290–296 (2017). https://www.ebi.ac.uk/emdb/EMD-40509. Deposited 14 April 2023. 63. D. Kimanius, L. Dong, G. Sharov, T. Nakane, S. H. W. Scheres, New tools for automated cryo-EM 72. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the down conformation. EMDB. https:// single-particle analysis in RELION-4.0. Biochem. J. 478, 4169–4185 (2021). www.ebi.ac.uk/emdb/EMD-40510. Deposited 14 April 2023. 64. T. Bepler et al., Positive-unlabeled convolutional neural networks for particle picking in cryo-electron 73. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the up conformation. PDB. https://www. micrographs. Nat. Methods 16, 1153–1160 (2019). rcsb.org/structure/8SIK. Deposited 14 April 2023. 65. T. I. Croll, ISOLDE: A physically realistic environment for model building into low-resolution electron- 74. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the intermediate conformation. PDB. density maps. Acta Crystallogr. Sect. Struct. Biol. 74, 519–530 (2018). https://www.rcsb.org/structure/8SIM. Deposited 14 April 2023. 66. E. F. Pettersen et al., UCSF ChimeraX: Structure visualization for researchers, educators, and 75. V. S. Mandala, R. MacKinnon, KCNQ1 with voltage sensor in the down conformation. PDB. https:// developers. Protein Sci. 30, 70–82 (2021). www.rcsb.org/structure/8SIN. Deposited 14 April 2023. 12 of 12 https://doi.org/10.1073/pnas.2301985120 pnas.org
10.1103_physrevd.107.023530	Editors' Suggestion Featured in Physics PHYSICAL REVIEW D 107, 023530 (2023) Joint analysis of Dark Energy Survey Year 3 data and CMB lensing from SPT and Planck. II. Cross-correlation measurements and cosmological constraints C. Chang ,1,2 Y. Omori,1,2,3,4 E. J. Baxter,5 C. Doux,6 A. Choi,7 S. Pandey,6 A. Alarcon,8 O. Alves,9,10 A. Amon,4 F. Andrade-Oliveira,9 K. Bechtol,11 M. R. Becker,8 G. M. Bernstein,6 F. Bianchini,3,4,12 J. Blazek,13,14 L. E. Bleem,8,2 H. Camacho,15,10 A. Campos,16 A. Carnero Rosell,10,17,18 M. Carrasco Kind,19,20 R. Cawthon,21 R. Chen,22 J. Cordero,23 T. M. Crawford,1,2 M. Crocce,24,25 C. Davis,4 J. DeRose,26 S. Dodelson,16,27 A. Drlica-Wagner,1,2,28 K. Eckert,6 T. F. Eifler,29,30 F. Elsner,31 J. Elvin-Poole,32,33 S. Everett,34 X. Fang,29,35 A. Fert´e,30 P. Fosalba,24,25 O. Friedrich,36 M. Gatti,6 G. Giannini,37 D. Gruen,38 R. A. Gruendl,19,20 I. Harrison,23,39,40 K. Herner,28 H. Huang,29,41 E. M. Huff,30 D. Huterer,9 M. Jarvis,6 A. Kovacs,17,18 E. Krause,29 N. Kuropatkin,28 P.-F. Leget,4 P. Lemos,31,42 A. R. Liddle,43 N. MacCrann,44 J. McCullough,4 J. Muir,45 J. Myles,3,4,46 A. Navarro-Alsina,47 Y. Park,48 A. Porredon,32,33 J. Prat,1,2 M. Raveri,6 R. P. Rollins,23 A. Roodman,4,46 R. Rosenfeld,10,49 A. J. Ross,32 E. S. Rykoff,4,46 C. Sánchez,6 J. Sanchez,28 L. F. Secco,2 I. Sevilla-Noarbe,50 E. Sheldon,51 T. Shin,6 M. A. Troxel,22 I. Tutusaus,24,25,52 T. N. Varga,53,54 N. Weaverdyck,9,26 R. H. Wechsler,3,4,46 W. L. K. Wu,4,46 B. Yanny,28 B. Yin,16 Y. Zhang,28 J. Zuntz,55 T. M. C. Abbott,56 M. Aguena,10 S. Allam,28 J. Annis,28 D. Bacon,57 B. A. Benson,1,2,28 E. Bertin,58,59 S. Bocquet,60 D. Brooks,31 D. L. Burke,4,46 J. E. Carlstrom,1,2,8,61,62 J. Carretero,37 C. L. Chang,1,2,8 R. Chown,63,64 M. Costanzi,65,66,67 L. N. da Costa,10,68 A. T. Crites,1,2,69 M. E. S. Pereira,70 T. de Haan,71,72 J. De Vicente,50 S. Desai,73 H. T. Diehl,28 M. A. Dobbs,74,75 P. Doel,31 W. Everett,76 I. Ferrero,77 B. Flaugher,28 D. Friedel,19 J. Frieman,2,28 J. García-Bellido,78 E. Gaztanaga,24,25 E. M. George,79,72 T. Giannantonio,36,80 N. W. Halverson,76,81 S. R. Hinton,82 G. P. Holder,20,75,83 D. L. Hollowood,34 W. L. Holzapfel,72 K. Honscheid,32,33 J. D. Hrubes,84 D. J. James,85 L. Knox,86 K. Kuehn,87,88 O. Lahav,31 A. T. Lee,26,89 M. Lima,10,90 D. Luong-Van,84 M. March,6 J. J. McMahon,1,2,61,62 P. Melchior,91 F. Menanteau,19,20 S. S. Meyer,1,2,61,92 R. Miquel,37,93 L. Mocanu,1,2 J. J. Mohr,94,95,96 R. Morgan,11 T. Natoli,1,2 S. Padin,1,2,97 A. Palmese,35 F. Paz-Chinchón,19,80 A. Pieres,10,68 A. A. Plazas Malagón,91 C. Pryke,98 C. L. Reichardt,12 M. Rodríguez-Monroy,50 A. K. Romer,42 J. E. Ruhl,99 E. Sanchez,50 K. K. Schaffer,2,61,100 M. Schubnell,9 S. Serrano,24,25 E. Shirokoff,1,2 M. Smith,101 Z. Staniszewski,99,30 A. A. Stark,102 E. Suchyta,103 G. Tarle,9 D. Thomas,57 C. To,32 J. D. Vieira,20,83 J. Weller,53,54 and R. Williamson104,1,2 (DES & SPT Collaborations) 1Department of Astronomy and Astrophysics, University of Chicago, Chicago, Illinois 60637, USA 2Kavli Institute for Cosmological Physics, University of Chicago, Chicago, Illinois 60637, USA 3Department of Physics, Stanford University, 382 Via Pueblo Mall, Stanford, California 94305, USA 4Kavli Institute for Particle Astrophysics & Cosmology, P. O. Box 2450, Stanford University, Stanford, California 94305, USA 5Institute for Astronomy, University of Hawai‘i, 2680 Woodlawn Drive, Honolulu, Hawaii 96822, USA 6Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, Pennsylvania 19104, USA 7California Institute of Technology, 1200 East California Blvd, MC 249-17, Pasadena, California 91125, USA 8Argonne National Laboratory, 9700 South Cass Avenue, Lemont, Illinois 60439, USA 9Department of Physics, University of Michigan, Ann Arbor, Michigan 48109, USA 10Laboratório Interinstitucional de e-Astronomia - LIneA, Rua Gal. Jos´e Cristino 77, Rio de Janeiro, RJ - 20921-400, Brazil 11Physics Department, 2320 Chamberlin Hall, University of Wisconsin-Madison, 1150 University Avenue Madison, Wisconsin 53706-1390 12School of Physics, University of Melbourne, Parkville, Victoria 3010, Australia 13Department of Physics, Northeastern University, Boston, Massachusetts 02115, USA 14Laboratory of Astrophysics, École Polytechnique F´ed´erale de Lausanne (EPFL), Observatoire de Sauverny, 1290 Versoix, Switzerland 15Instituto de Física Teórica, Universidade Estadual Paulista, São Paulo, Brazil 16Department of Physics, Carnegie Mellon University, Pittsburgh, Pennsylvania 15312, USA 17Instituto de Astrofisica de Canarias, E-38205 La Laguna, Tenerife, Spain 18Universidad de La Laguna, Dpto. Astrofísica, E-38206 La Laguna, Tenerife, Spain 2470-0010=2023=107(2)=023530(25) 023530-1 © 2023 American Physical Society C. CHANG et al. PHYS. REV. D 107, 023530 (2023) 19Center for Astrophysical Surveys, National Center for Supercomputing Applications, 1205 West Clark St., Urbana, Illinois 61801, USA 20Department of Astronomy, University of Illinois at Urbana-Champaign, 1002 W. Green Street, Urbana, Illinois 61801, USA 21Physics Department, William Jewell College, Liberty, Missouri, 64068 22Department of Physics, Duke University Durham, North Carolina 27708, USA 23Jodrell Bank Center for Astrophysics, School of Physics and Astronomy, University of Manchester, Oxford Road, Manchester, M13 9PL, United Kingdom 24Institut d’Estudis Espacials de Catalunya (IEEC), 08034 Barcelona, Spain 25Institute of Space Sciences (ICE, CSIC), Campus UAB, Carrer de Can Magrans, s/n, 08193 Barcelona, Spain 26Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, California 94720, USA 27NSF AI Planning Institute for Physics of the Future, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213, USA 28Fermi National Accelerator Laboratory, P. O. Box 500, Batavia, Illinois 60510, USA 29Department of Astronomy/Steward Observatory, University of Arizona, 933 North Cherry Avenue, Tucson, Arizona 85721-0065, USA 30Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena, California 91109, USA 31Department of Physics & Astronomy, University College London, Gower Street, London, WC1E 6BT, United Kingdom 32Center for Cosmology and Astro-Particle Physics, The Ohio State University, Columbus, Ohio 43210, USA 33Department of Physics, The Ohio State University, Columbus, Ohio 43210, USA 34Santa Cruz Institute for Particle Physics, Santa Cruz, California 95064, USA 35Department of Astronomy, University of California, Berkeley, 501 Campbell Hall, Berkeley, California 94720, USA 36Kavli Institute for Cosmology, University of Cambridge, Madingley Road, Cambridge CB3 0HA, United Kingdom 37Institut de Física d’Altes Energies (IFAE), The Barcelona Institute of Science and Technology, Campus UAB, 08193 Bellaterra (Barcelona) Spain 38University Observatory, Faculty of Physics, Ludwig-Maximilians-Universitat, Scheinerstrasse 1, 81679 Munich, Germany 39Department of Physics, University of Oxford, Denys Wilkinson Building, Keble Road, Oxford OX1 3RH, United Kingdom 40School of Physics and Astronomy, Cardiff University, CF24 3AA, United Kingdom 41Department of Physics, University of Arizona, Tucson, Arizona 85721, USA 42Department of Physics and Astronomy, Pevensey Building, University of Sussex, Brighton, BN1 9QH, United Kingdom 43Instituto de Astrofísica e Ciências do Espaço, Faculdade de Ciências, Universidade de Lisboa, 1769-016 Lisboa, Portugal 44Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge CB3 0WA, United Kingdom 45Perimeter Institute for Theoretical Physics, 31 Caroline St. North, Waterloo, Ontario N2L 2Y5, Canada 46SLAC National Accelerator Laboratory, Menlo Park, California 94025, USA 47Instituto de Física Gleb Wataghin, Universidade Estadual de Campinas, 13083-859, Campinas, SP, Brazil 48Kavli Institute for the Physics and Mathematics of the Universe (WPI), UTIAS, The University of Tokyo, Kashiwa, Chiba 277-8583, Japan 49ICTP South American Institute for Fundamental Research Instituto de Física Teórica, Universidade Estadual Paulista, São Paulo, Brazil 50Centro de Investigaciones Energ´eticas, Medioambientales y Tecnológicas (CIEMAT), Madrid, Spain 51Brookhaven National Laboratory, Bldg 510, Upton, New York 11973, USA 52D´epartement de Physique Th´eorique and Center for Astroparticle Physics, Universit´e de Gen`eve, 24 quai Ernest Ansermet, CH-1211 Geneva, Switzerland 53Max Planck Institute for Extraterrestrial Physics, Giessenbachstrasse, 85748 Garching, Germany 54Universitäts-Sternwarte, Fakultät für Physik, Ludwig-Maximilians Universität München, Scheinerstrasse 1, 81679 München, Germany 55Institute for Astronomy, University of Edinburgh, Edinburgh EH9 3HJ, United Kingdom 023530-2 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) 56Cerro Tololo Inter-American Observatory, NSF’s National Optical-Infrared Astronomy Research Laboratory, Casilla 603, La Serena, Chile 57Institute of Cosmology and Gravitation, University of Portsmouth, Portsmouth, PO1 3FX, United Kingdom 58CNRS, UMR 7095, Institut d’Astrophysique de Paris, F-75014, Paris, France 59Sorbonne Universit´es, UPMC Univ Paris 06, UMR 7095, Institut d’Astrophysique de Paris, F-75014, Paris, France 60Ludwig-Maximilians-Universität, Scheiner- strasse 1, 81679 Munich, Germany 61Enrico Fermi Institute, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 62Department of Physics, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 63Department of Physics & Astronomy, The University of Western Ontario, London Ontario N6A 3K7, Canada 64Institute for Earth and Space Exploration, The University of Western Ontario, London Ontario N6A 3K7, Canada 65Astronomy Unit, Department of Physics, University of Trieste, via Tiepolo 11, I-34131 Trieste, Italy 66INAF-Osservatorio Astronomico di Trieste, via G. B. Tiepolo 11, I-34143 Trieste, Italy 67Institute for Fundamental Physics of the Universe, Via Beirut 2, 34014 Trieste, Italy 68Observatório Nacional, Rua Gal. Jos´e Cristino 77, Rio de Janeiro, RJ - 20921-400, Brazil 69Department of Astronomy & Astrophysics, University of Toronto, 50 St George St, Toronto, Ontario, M5S 3H4, Canada 70Hamburger Sternwarte, Universität Hamburg, Gojenbergsweg 112, 21029 Hamburg, Germany 71High Energy Accelerator Research Organization (KEK), Tsukuba, Ibaraki 305-0801, Japan 72Department of Physics, University of California, Berkeley, California 94720, USA 73Department of Physics, IIT Hyderabad, Kandi, Telangana 502285, India 74Department of Physics and McGill Space Institute, McGill University, 3600 Rue University, Montreal, Quebec H3A 2T8, Canada 75Canadian Institute for Advanced Research, CIFAR Program in Gravity and the Extreme Universe, Toronto, Ontario M5G 1Z8, Canada 76Department of Astrophysical and Planetary Sciences, University of Colorado, Boulder, Colorado 80309, USA 77Institute of Theoretical Astrophysics, University of Oslo, P.O. Box 1029 Blindern, NO-0315 Oslo, Norway 78Instituto de Fisica Teorica UAM/CSIC, Universidad Autonoma de Madrid, 28049 Madrid, Spain 79European Southern Observatory, Karl-Schwarzschild-Straße 2, 85748 Garching, Germany 80Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, United Kingdom 81Department of Physics, University of Colorado, Boulder, Colorado, 80309, USA 82School of Mathematics and Physics, University of Queensland, Brisbane, QLD 4072, Australia 83Department of Physics, University of Illinois Urbana-Champaign, 1110 West Green Street, Urbana, Illinois 61801, USA 84University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois 60637, USA 85Center for Astrophysics \| Harvard & Smithsonian, 60 Garden Street, Cambridge, Massachusetts 02138, USA 86Department of Physics, University of California, One Shields Avenue, Davis, California 95616, USA 87Australian Astronomical Optics, Macquarie University, North Ryde, New South Wales 2113, Australia 88Lowell Observatory, 1400 Mars Hill Rd, Flagstaff, Arizona 86001, USA 89Department of Physics, University of California, Berkeley, California, 94720, USA 90Departamento de Física Matemática, Instituto de Física, Universidade de São Paulo, CP 66318, São Paulo, SP, 05314-970, Brazil 91Department of Astrophysical Sciences, Princeton University, Peyton Hall, Princeton, New Jersey 08544, USA 92Department of Physics, University of Chicago, 5640 South Ellis Avenue, Chicago, Illinois, 60637, USA 93Institució Catalana de Recerca i Estudis Avançats, E-08010 Barcelona, Spain 94Ludwig-Maximilians-Universität, Scheiner- str. 1, 81679 Munich, Germany 95Excellence Cluster Universe, Boltzmannstr. 2, 85748 Garching, Germany 96Max-Planck-Institut fur extraterrestrische Physik, Giessenbachstrasse 85748 Garching, Germany 97California Institute of Technology, 1200 East California Boulevard, Pasadena, California 91125, USA 98School of Physics and Astronomy, University of Minnesota, 116 Church Street SE Minneapolis, Minnesota 55455, USA 99Department of Physics, Case Western Reserve University, Cleveland, Ohio 44106, USA 023530-3 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) 100Liberal Arts Department, School of the Art Institute of Chicago, Chicago, Illinois USA 60603 101School of Physics and Astronomy, University of Southampton, Southampton, SO17 1BJ, United Kingdom 102Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, Massachusetts 02138, USA 103Computer Science and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831 104Jet Propulsion Laboratory, California Institute of Technology, Pasadena, California 91109, USA (Received 4 April 2022; accepted 30 September 2022; published 31 January 2023) Cross-correlations of galaxy positions and galaxy shears with maps of gravitational lensing of the cosmic microwave background (CMB) are sensitive to the distribution of large-scale structure in the Universe. Such cross-correlations are also expected to be immune to some of the systematic effects that complicate correlation measurements internal to galaxy surveys. We present measurements and modeling of the cross-correlations between galaxy positions and galaxy lensing measured in the first three years of data from the Dark Energy Survey with CMB lensing maps derived from a combination of data from the 2500 deg2 SPT-SZ survey conducted with the South Pole Telescope and full-sky data from the Planck satellite. The CMB lensing maps used in this analysis have been constructed in a way that minimizes biases from the thermal Sunyaev Zel’dovich effect, making them well suited for cross-correlation studies. The total signal-to-noise of the cross-correlation measurements is 23.9 (25.7) when using a choice of angular scales optimized for a linear (nonlinear) galaxy bias model. We use the cross-correlation measurements to obtain constraints on cosmological parameters. For our fiducial galaxy sample, m ¼ 0.272þ0.032 which consist of four bins of magnitude-selected galaxies, we find constraints of Ω −0.052 and S8 ≡ σ8 −0.028 ) when assuming linear (nonlinear) galaxy bias in our modeling. Considering only the cross-correlation of galaxy shear with −0.061 and S8 ¼ 0.740þ0.034 CMB lensing, we find Ω −0.029 . Our constraints on S8 are consistent with recent cosmic shear measurements, but lower than the values preferred by primary CMB measurements from Planck. −0.044 and S8 ¼ 0.734þ0.035 m ¼ 0.245þ0.026 m ¼ 0.270þ0.043 ¼ 0.736þ0.032 ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ m=0.3 Ω −0.028 (Ω p DOI: 10.1103/PhysRevD.107.023530 I. INTRODUCTION Significant progress has been made recently in using cross-correlations between galaxy imaging and cosmic microwave background (CMB) surveys to constrain cos- mological parameters. These developments have come naturally as ongoing galaxy and CMB surveys collect increasingly sensitive data across larger and larger over- lapping areas of the sky. The Dark Energy Survey [DES, 1] is the largest galaxy weak lensing survey today, covering ∼5000 deg2 of sky that is mostly in the southern hemi- sphere. By design, the DES footprint overlaps with high- resolution CMB observations from the South Pole Telescope [SPT, 2], enabling a large number of cross- correlation analyses [3–12]. Although CMB photons originate from the high-redshift Universe, their trajectories are deflected by low-redshift structures as a result of gravitational lensing—these are the same structures traced by the distributions of galaxies and the galaxy weak lensing signal measured in optical galaxy surveys. Cross-correlating CMB lensing with galaxy sur- veys therefore allows us to extract information stored in the large-scale structure. t κ CMBi, CMB, and hγ In this work, we analyze both hδgκ the cross correlation of the galaxy density field δg and the CMB CMBi,1 the weak lensing convergence field κ cross correlation of the galaxy weak lensing shear field γ and κ CMB. Notably, these two two-point functions correlate measurements from very different types of surveys (galaxy surveys in the optical and CMB surveys in the millimeter), and are therefore expected to be very robust to systematic biases impacting only one type of survey. Furthermore, CMB lensing is sensitive to a broad range of redshift, with peak sensitivity at redshift z ∼ 2; galaxy lensing, on the other hand, is sensitive to structure at z ≲ 1 for current surveys. As a result, the CMB lensing cross-correlation functions, hδgκ CMBi, are expected to increase in signal-to-noise relative to galaxy lensing correlations as one considers galaxy samples that extend to higher redshift. Our analysis relies on the first three years (Y3) of galaxy observations from DES and a CMB lensing map con- structed using data from the 2500 deg2 SPT-SZ survey [13] CMBi þ hγ κ t 1The “t” subscript denotes the tangential component of shear, which will be discussed in Sec. IV. 023530-4 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) t κ CMBi þ hγ and Planck [14]. The combined signal-to-noise of the hδgκ CMBi measurements used in the present cosmological analysis is roughly a factor two larger than in the earlier DES þ SPT results presented in [11], which used first year (Y1) DES data. This large improvement in signal-to-noise derives from two main advancements: (1) We have adopted a different methodology in con- structing the CMB lensing map, which results in much lower contamination from the thermal Su- nyaev Zel’dovich (tSZ) effect, allowing small-scale information to be used in the cosmological analysis. This methodology is described in [15]. (2) Data from DES Y3 covers an area approximately three times larger than DES Y1 and is slightly deeper. Along with the significant increase in signal-to-noise, we have also updated our models for the correlation functions to include a number of improvements following [16]. These include an improved treatment of galaxy intrinsic align- ments, inclusion of magnification effects on the lens galaxy density, and application of the so-called lensing ratio likelihood described in [17]. The analysis presented here is the second of a series of three papers: In [15] (Paper I) we describe the construction of the combined, tSZ-cleaned SPT þ Planck CMB lensing map and the methodology for the cosmological analysis. In this paper (Paper II), we present the data measurements of the cross-correlation probes hδgκ CMBi, a series of diagnostic tests, and cosmological constraints from this cross-correlation combination. In [18] (Paper III), we will present the joint cosmological constraints from hδgκ CMBi and the DES-only 3 × 2 pt probes,2 and tests of consistency between the two, as well as constraints from a joint analysis with the CMB lensing auto-spectrum. CMBi þ hγ CMBi þ hγ κ κ t t Suprime-Cam Subaru Similar analyses have recently been carried out using different galaxy imaging surveys and CMB data. [19] studied the cross-correlation of the galaxy weak lensing from the Hyper Strategic Program Survey [HSC-SSP, 20] and the Planck lensing map [21]; [22] used the same HSC galaxy weak lensing measurement to cross-correlate with CMB lensing from the POLARBEAR experiment [23]; [24] cross-correlated gal- axy weak lensing from the Kilo-Degree Survey [KiDS, 25] and the CMB lensing map from the Atacama Cosmology Telescope [ACT, 26]; and [27] cross-correlated the galaxy density measured in unWISE data [28] with Planck CMB lensing. Compared to these previous studies, in addition to the new datasets, this paper is unique in that we combine hδgκ CMBi. Moreover, our analysis uses the CMBi and hγ κ t 2The 3 × 2 pt probes refer to a combination of three two-point functions of the galaxy density field δg and the weak lensing shear field γ: galaxy clustering hδgδgi, galaxy-galaxy lensing hδgγti and cosmic shear hγγi. same modeling choices and analysis framework as in [16], making it easy to compare and combine our results later (i.e. Paper III). The structure of the paper is as follows. In Sec. II we briefly review the formalism of our model for the two cross- correlation functions and the parameter inference pipeline (more details can be found in Paper I). In Sec. III we review the data products used in this analysis. In Sec. IV we introduce the estimators we use for the correlation func- tions. In Sec. V we describe out blinding procedure and unblinding criteria. In Sec. VI we present constraints on cosmological parameters as well as relevant nuisance parameters when fitting to the cross-correlation functions. Finally we conclude in Sec. VII. II. MODELING AND INFERENCE We follow the theoretical formalism laid out in Paper I and [29] for this work. Here, we summarize only the main equations relevant to this paper. Following standard con- vention, we refer to the galaxies used to measure δg as lens galaxies, and the galaxies used to measure γ as source galaxies. power spectra: Using Angular Limber approximation3 the cross-spectra between CMB lensing convergence and galaxy density/shear can be related to the matter power spectrum via: [31], the Z CκCMBXiðlÞ ¼ dχ qκCMBðχÞqi XðχÞ χ2 (cid:3) l þ 1=2 χ (cid:4) ; zðχÞ ; PNL ð1Þ where X ∈ fδg; γg, i labels the redshift bin, PNLðk; zÞ is the nonlinear matter power spectrum, which we compute using CAMB and HALOFIT [32,33], and χ is the comoving distance to redshift z. The weighting functions, qXðχÞ, describe how the different probes respond to large-scale structure at different distances, and are given by qκCMBðχÞ ¼ 3Ω mH2 0 2c2 χ aðχÞ χ(cid:2) − χ χ(cid:2) ; qi δg ðχÞ ¼ biðzðχÞÞni δg ðzðχÞÞ dz dχ γðχÞ ¼ qi 3H2 0Ω m 2c2 χ aðχÞ Z χh χ dχ0ni γðzðχ0ÞÞ dz dχ0 χ0 − χ χ0 ; ð2Þ ð3Þ ð4Þ 3In [30], the authors showed that at DES Y3 accuracy, the Limber approximation is sufficient for galaxy-galaxy lensing and cosmic shear but insufficient for galaxy clustering. Given the primary probe in this work, hδgκ CMBi, are at much lower signal-to-noise than galaxy-galaxy lensing and cosmic shear, we expect that Limber approximation is still a valid choice. CMBi þ hγ κ t 023530-5 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) where H0 and Ω m are the Hubble constant and matter density parameters, respectively, aðχÞ is the scale factor corresponding to comoving distance χ, χ(cid:2) denotes the comoving distance to the surface of last scattering, bðzÞ is the galaxy bias as a function of redshift, and ni δg=γðzÞ are the normalized redshift distributions of the lens/source galaxies in bin i. We note that the above equations assumes linear galaxy bias, which is our fiducial model. Modeling the nonlinear galaxy bias involves changes to both Eq. (1) and Eq. (3) (see below). Correlation functions: The angular-space correlation functions are then computed via X wδi gκCMBðθÞ ¼ wγi t κCMBðθÞ ¼ l X l 2l þ 1 4π FðlÞPlðcosðθÞÞCδi gκCMBðlÞ; 2l þ 1 4πlðl þ 1Þ FðlÞP2 lðcosθÞCκi γκCMBðlÞ; ð5Þ ð6Þ where Pl and P2 l are the lth order Legendre polynomial and associated Legendre polynomial, respectively, and FðlÞ describes filtering applied to the κ CMB maps. For correlations with the κ CMB maps, we set FðlÞ ¼ BðlÞHðl − l − lÞ, where HðlÞ is a step function and BðlÞ ¼ expð−0.5lðl þ 1Þσ2Þ with σ ≡ θ FWHM= FWHM describes the beam applied to the CMB lensing maps (see discussion of l max, and θ FWHM choices in Sec. III, and further discussion in Paper I). minÞHðl ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 8 ln 2 , and θ min, l max p Galaxy bias: We consider two models for the galaxy bias. Our fiducial choice is a linear bias model where bðzÞ ¼ bi is not a function of scale and is assumed to be a free parameter for each tomographic bin i. The second bias model is described in [34] and is an effective 1-loop model with renormalized nonlinear galaxy bias parameters: bi 1 (linear bias), bi 2 (local quadratic bias), bi s2 (tidal quadratic bias) and bi 3nl (third-order non-local bias). The latter two parameters can be derived from bi 1, making the total number of free parameters for this bias model two per tomographic bin i. To use this model, we replace the combination of biPNL in Equation (1) with Pgm described in [34]. Intrinsic alignment (IA): Galaxy shapes can be intrinsi- cally aligned as a result of nearby galaxies evolving in a common tidal field. IA modifies the observed lensing signal. We adopt the five-parameter (a1,η1,a2,η2,bta) tidal alignment tidal torquing model (TATT) of [35] to describe galaxy IA. a1 and η1 characterize the amplitude and redshift dependence of the tidal alignment; a2 and η2 characterize the amplitude and redshift dependence of the tidal torquing effect; bta accounts for the fact that our measurement is weighted by the observed galaxy counts. In Sec. VI B, we will also compare our results using the nonlinear alignment model a simpler IA model, [NLA, 36]. The TATT model is equivalent to the NLA model in the limit that a2 ¼ η2 ¼ bta ¼ 0. Impact of lensing magnification on lens galaxy density: Foreground structure modulates the observed galaxy den- sity as a result of gravitational magnification. The effect of magnification can be modeled by modifying Eq. (3) to include the change in selection and geometric dilution quantified by the lensing bias coefficients Ci g: qi δg ðχÞ → qi δg ðχÞð1 þ Ci gκi gÞ; ð7Þ where κi in [29] and the values of Ci to the values listed in Table I. g is the tomographic convergence field, as described g are estimated in [37] and fixed Uncertainty in redshift distributions: We model uncer- tainty in the redshift distributions of the source galaxies with shift parameters, Δzi, defined such that for each redshift bin i, niðzÞ → niðz − Δi zÞ: ð8Þ For the lens sample, we additionally introduce a stretch parameter (σz) when modeling the redshift distribution, as motivated by [38]: niðzÞ → σi zniðσi z½z − hzi(cid:3) þ hzi − Δi zÞ; ð9Þ where hzi is the mean redshift. Uncertainty in shear calibration: We model uncertainty in the shear calibration with multiplicative factors defined such that the observed CκCMBγ is modified by CκCMBγiðlÞ → ð1 þ miÞCκCMBγiðlÞ; ð10Þ where mi is the shear calibration bias for source bin i. they can, however, Lensing ratio (or shear ratio, SR): The DES Y3 3 × 2 pt analysis used a ratio of small-scale galaxy lensing measure- ments to provide additional information, particularly on source galaxy redshift biases and on IA parameters. These ratios are not expected to directly inform the cosmological constraints; improve constraints via degeneracy breaking with nuisance parameters. The lensing ratios can therefore be considered as another form of systematic calibration, in a similar vein to, e.g., spectro- scopic data used to calibrate redshifts, and image simulations used to calibrate shear biases. In [17], it was demonstrated that the lensing ratio measurements are approximately independent of the 3 × 2 pt measurements, making it trivial to combine constraints from 3 × 2 pt and lensing ratios at the likelihood level. Unless otherwise mentioned, all our analy- ses will include the information from these lensing ratios. We investigate their impact in Sec. VI B. Angular scale cuts: The theoretical model described above is uncertain on small scales due to uncertainty in our understanding of baryonic feedback and the galaxy-halo 023530-6 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) TABLE I. Prior values for cosmological and nuisance param- eters included in our model. For the priors, U½a; b(cid:3) indicates a uniform prior between a and b, while N ½a; b(cid:3) indicates a Gaussian prior with mean a and standard deviation b. δðaÞ is a Dirac Delta function at value a, which effectively means that the parameter is fixed at a. Note that the fiducial lens sample is the first 4 bins of the MAGLIM sample. The two high-redshift MAGLIM bins and the REDMAGIC sample are shown in gray to indicate they are not part of the fiducial analysis. Parameter inference: We assume a Gaussian likelihood4 for the data vector of measured correlation functions, ⃗d, given a model, ⃗m, generated using the set of parameters ⃗p: ln Lð⃗dj ⃗mð ⃗pÞÞ ¼ − 1 2 XN ij ðdi − mið ⃗pÞÞTC−1 ij ðdj − mjð ⃗pÞÞ; ð11Þ Parameter Ωm As × 109 ns Ωb h Ωνh2 × 104 a1 a2 η1 η2 bta MAGLIM b1(cid:4)(cid:4)(cid:4)6 b1(cid:4)(cid:4)(cid:4)6 1 b1(cid:4)(cid:4)(cid:4)6 2 C1(cid:4)(cid:4)(cid:4)6 l Δ1…6 z × 102 σ1…6 z REDMAGIC b1(cid:4)(cid:4)(cid:4)5 b1(cid:4)(cid:4)(cid:4)5 1 b1(cid:4)(cid:4)(cid:4)5 2 C1(cid:4)(cid:4)(cid:4)5 l z × 102 Δ1…5 σ1…4 z Prior U½0.1; 0.9(cid:3) U½0.5; 5.0(cid:3) U½0.87; 1.07(cid:3) U½0.03; 0.07(cid:3) U½0.55; 0.91(cid:3) U½6.0; 64.4(cid:3) U½−5.0; 5.0(cid:3) U½−5.0; 5.0(cid:3) U½−5.0; 5.0(cid:3) U½−5.0; 5.0(cid:3) U½0.0; 2.0(cid:3) U½0.8; 3.0(cid:3) U½0.67; 3.0(cid:3) U½−4.2; 4.2(cid:3) δð1.21Þ, δð1.15Þ, δð1.88Þ, δð1.97Þ, δð1.78Þ, δð2.48Þ N ½−0.9; 0.7(cid:3), N ½−3.5; 1.1(cid:3), N ½−0.5; 0.6(cid:3), N ½−0.7; 0.6(cid:3), N ½0.2; 0.7(cid:3), N ½0.2; 0.8(cid:3) N ½0.98; 0.062(cid:3), N ½1.31; 0.093(cid:3), N ½0.87; 0.054(cid:3), N ½0.92; 0.05(cid:3), N ½1.08; 0.067(cid:3), N ½0.845; 0.073(cid:3) U½0.8; 3.0(cid:3) U½0.67; 2.52(cid:3) U½−3.5; 3.5(cid:3) δð1.31Þ, δð−0.52Þ, δð0.34Þ, δð2.25Þ, δð1.97Þ N ½0.6; 0.4(cid:3), N ½0.1; 0.3(cid:3), N ½0.4; 0.3(cid:3), N ½−0.2; 0.5(cid:3), N ½−0.7; 1.0(cid:3) δð1Þ, δð1Þ, δð1Þ, δð1Þ, N ½1.23; 0.054(cid:3) METACALIBRATION m1…4 × 103 Δ1…4 z × 10−2 N ½−6.0; 9.1(cid:3), N ½−20.0; 7.8(cid:3), N ½−24.0; 7.6(cid:3), N ½−37.0; 7.6(cid:3) N ½0.0; 1.8(cid:3), N ½0.0; 1.5(cid:3), N ½0.0; 1.1(cid:3), N ½0.0; 1.7(cid:3) connection (or, nonlinear galaxy bias). We take the approach of only fitting the correlation functions on angular scales we can reliably model. In Paper I we determined the corresponding angular scale cuts by requiring the cosmo- logical constraints to not be significantly biased when prescriptions for unmodeled effects are introduced. In Figs. 2 and 19 the scale cuts are marked by the gray bands. where the sums run over all of the N elements in the data and model vectors. The posterior on the model parameters is then given by: Pð ⃗mð ⃗pÞj⃗dÞ ∝ Lð⃗dj ⃗mð ⃗pÞÞPpriorð ⃗pÞ; ð12Þ where Ppriorð ⃗pÞ is a prior on the model parameters. Our choice of priors is summarized in Table I. The covariance matrix used here consists of an analytical lognormal covariance combined with empirical noise estimation from simulations. The covariance has been extensively validated in Paper I. In Appendix A Fig. 11 we show that the diagonal elements of our final analytic covariance are in excellent agreement with a covariance estimated from data using jackknife resampling. Our modeling and inference framework is built within the COSMOSIS package [40] and is designed to be consistent with those developed as part of [16]. We generate parameter samples using the nested sampler POLYCHORD [41]. III. DATA A. CMB lensing maps There are two major advances in the galaxy and CMB data used here relative to the DES Y1 and SPT analysis presented in [9,10]. First, for the CMB map in the SPT footprint, we used the method developed in [42] and described in Paper I to remove contamination from the tSZ effect by combining data from SPT and Planck. Such contamination was one of the limiting factors in our Y1 analysis. Second, the DES Y3 data cover a significantly larger area on the sky than the DES Y1 data. Consequently, the DES Y3 footprint extends beyond the SPT footprint, necessitating the use of the Planck-only lensing map [14] over part of the DES Y3 patch. As discussed in Paper I, the different noise properties and filtering of the two lensing maps necessitates separate treatment throughout. The “SPT þ Planck” lensing map, which overlaps with the DES footprint at < −40 degrees in declination, is filtered by l max ¼ 5000 and a Gaussian smooth- ing of θ FWHM ¼ 6 arcmin. This map is produced using the combination of 150 GHz data from the 2500 deg2 SPT-SZ survey [e.g., 13], Planck 143 GHz data, and the min ¼ 8, l 4See e.g., [39] for tests of the validity of this assumption in the context of cosmic shear, which would also apply here. 023530-7 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) tSZ-cleaned CMB Planck temperature map generated using the Spectral Matching Independent Component Analysis the SMICA-noSZ map). The (SMICA) algorithm (i.e. “Planck” lensing map, which overlaps with the DES footprint at > −39.5 degrees in declination, is filtered by max ¼ 3800 and a Gaussian smoothing of min ¼ 8, l l θ FWHM ¼ 8 arcmin is applied. This map is reconstructed using the Planck SMICA-noSZ temperature map alone. We leave a small 0.5 deg gap between the two lensing maps to reduced the correlation between structures on the boundaries. The resulting effective overlapping areas with DES are 1764 deg2 and 2156 deg2 respectively for the SPT þ Planck and Planck patches respectively. B. The DES Y3 data products is mostly in the southern hemisphere, DES [43] is a photometric survey in five broadband filters (grizY), with a footprint of nearly 5000 deg2 of sky that imaging hundreds of millions of galaxies. It employs the 570- megapixel Dark Energy Camera [DECam, 1] on the Cerro Tololo Inter-American Observatory (CTIO) 4 m Blanco telescope in Chile. We use data from the first three years (Y3) of DES observations. The foundation of the various DES Y3 data products is the Y3 Gold catalog described in [44], which achieves S=N ∼ 10 for extended objects up to i ∼ 23.0 over an unmasked area of 4143 deg2. In this work we use three galaxy samples: two lens samples for the galaxy density-CMB lensing correlation, hδgκ CMBi, and one source sample for the galaxy shear-CMB lensing correlation, hγ CMBi. We briefly describe each sample below. These samples are the same as those used in [16] and we direct the readers to a more detailed description of the samples therein. κ t 1. Lens samples: MAGLIM and REDMAGIC We will show results from two lens galaxy samples named MAGLIM and REDMAGIC. Following [16], the first four bins of the MAGLIM sample will constitute our fiducial sample, though we show results from the other bins and samples to help understand potential systematic effects in the DES galaxy selection. The MAGLIM sample consists of 10.7 million galaxies selected with a magnitude cut that evolves linearly with the photometric redshift estimate: i < 4zphot þ 18. zphot is deter- mined using the Directional Neighborhood Fitting algorithm [DNF, 45]. [46] optimized the magnitude cut to balance the statistical power of the sample size and the accuracy of the photometric redshifts for cosmological constraints from galaxy clustering and galaxy-galaxy lensing. MAGLIM is divided into six tomographic bins. The top panel of Fig. 1 shows the per-bin redshift distributions, which have been validated using cross-correlations with spectroscopic gal- axies in [38]. Weights are derived to account for survey systematics, as described in [47]. FIG. 1. Redshift distribution for the tomographic bins for the galaxy samples used in this work: the MAGLIM lens sample (top), the REDMAGIC lens sample (middle) and the METACAL source sample (bottom). The fiducial lens sample only uses the first four bins of the MAGLIM sample, or the solid lines. We perform tests with the nonfiducial samples (dashed lines) for diagnostic purposes. The REDMAGIC sample consists of 2.6 million luminous red galaxies (LRGs) with small photometric redshift errors [48]. REDMAGIC is constructed using a red sequence template calibrated via the REDMAPPER algorithm [49,50]. The lens galaxies are divided into five tomographic bins. The redshift distributions are shown in the middle panel of Fig. 1. These distributions are estimated using draws from the redshift probability distribution functions of the individual REDMAGIC galaxies. As with MAGLIM, [38] validates the redshift distributions, and [47] derives sys- tematics weights. We note that in [16] the two high-redshift bins were excluded in MAGLIM due to poor fits in the 3 × 2 pt analysis, while the REDMAGIC sample was excluded due to an internal tension between galaxy-galaxy lensing and galaxy clustering. With the addition of CMB lensing cross- correlations, one of the aims of this work will be to shed light on potential systematic effects in the lens samples. We briefly discuss this issue in Sec. VI D but there will be a more in-depth discussion in Paper III when we combine with the 3 × 2 pt probes. 2. Source sample: METACALIBRATION For the source sample, we use the DES Y3 shear catalog presented in [51], which contains over 100 million galaxies. 023530-8 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) The galaxy shapes are estimated using the METACALIBRATION algorithm [52,53]. The shear catalog has been thoroughly tested in [51,54]. In [54], the authors used realistic image simulations to constrain the multiplicative bias of the shear estimate to be at most 2%–3%, primarily attributed to a shear- dependent detection bias coupled with object blending effects. The residual shear calibration biases are folded into the modeling pipeline and are listed in Table I. The source galaxies are divided into four tomographic bins based on the SOMPZ algorithm described in [55], utilizing deep field data described in [56] and image simulations described in [57]. The bottom panel of Fig. 1 shows the redshift distributions, which have been validated in [58,17]. IV. CORRELATION FUNCTION ESTIMATORS Our estimator for the galaxy-CMB lensing correlation [Eq. (5)] is hδgκ CMBðθαÞi ¼ hδgκ CMBðθαÞi0 − hδRκ CMBðθαÞi; ð13Þ where hδgκ CMBðθαÞi0 ¼ 1 δgκCMB θα N XNg XNpix i¼1 j¼1 ηδg i ηκCMB j κ CMB;jΘαðj ˆθi − ˆθjjÞ ð14Þ and Our estimator for the galaxy shear-CMB lensing corre- lation [Eq. (6)] is PNgal i¼1 hγ κ CMBðθαÞi ¼ t PNpix j¼1 ηe κ P ηκCMB i j sðθαÞ CMB;jeij t ηκCMB ηe i j Θαðj ˆθi − ˆθjjÞ ; ð16Þ where eij is the component of the corrected ellipticity t oriented orthogonally to the line connecting pixel j and the CMB value in the pixel is κj source galaxy. The κ CMB and ηe i and ηκCMB are the weights associated with the source galaxy j and the κ CMB pixel, respectively. The weights for the source galaxies are derived in Gatti et al. [51] and combines the signal-to-noise and size of each galaxy. sðθαÞ is the METACALIBRATION response. We find that sðθÞ is approx- imately constant over the angular scales of interest, but different for each redshift bin. We carry out these mea- surements using the TREECORR package5 [59] in the angular range 2.50 < θ < 250.00. Note that Eq. (16) does not require subtracting a random component as in Eq. (13) since unlike a density field, the mask geometry cannot generate an artificial signal in a shear field. κ CMBi and hγ CMBi correla- tion functions are shown in Fig. 2. The hδgκ CMBi mea- surements using the REDMAGIC sample are shown in Appendix C. The signal-to-noise (S/N) of the different measurements are listed in Table II. Here, signal-to-noise is calculated via The measured MAGLIM hδgκ t v u u t ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ XN C−1 ij dj dT i ij ; ð17Þ hδRκ CMBðθαÞi ¼ 1 NRκCMB θα XNrand XNpix i¼1 j¼1 ηδR i ηκCMB j κ CMB;jΘαðj ˆθi − ˆθjjÞ; S=N ≡ ð15Þ where the sum in i is over all galaxies and the sum in j is δgκCMB over all pixels in the CMB convergence map; N θα ) is the number of galaxy-κ δRκCMB (N CMB pixel (random- θα CMB pixel) pairs that fall within the angular bin θα; ηδg, κ ηδR and ηκCMB are the weights associated with the galaxies, the randoms and the κ CMB pixels. The weights for the galaxies/randoms are derived in Rodríguez-Monroy [47] using a combination of maps of survey properties (e.g., seeing, depth, airmass) to correct for any spurious signals in the large-scale structure, while the κ CMB weights account for differences in the noise levels of pixels in the κ CMB map. The random catalog is used to sample the selection function of the lens galaxies, and has a number density much higher than the galaxies. ˆθi (ˆθj) is the angular position of galaxy i (pixel j), and Θα is an ˆθi − ˆθjj falls in the angular indicator function that is 1 if j bin θα and 0 otherwise. t κ where d is the data vector of interest and C is the covariance matrix. The final signal-to-noise of the fiducial hδgκ CMBi þ hγ CMBi data vector after the linear bias scale cuts is 23.9, about two times larger than in the Y1 study [11]—the main improvement, in addition to the increased sky area, comes from extending our analysis to smaller scales, enabled by the tSZ-cleaned CMB lensing map. The tSZ signal is correlated with large-scale structure, and can propagate into a bias in the estimated κ CMB if not mitigated. In the DES Y1 analysis presented in [11], tSZ cleaning was not imple- mented at the κ CMB map level, necessitating removal of small-scale CMB lensing correlation measurements from the model fits. This problem was particularly severe for hγ CMBi. Comparing results for the SPT þ Planck and Planck patches in Table II, the SPT þ Planck area domi- nates the signal-to-noise before scale cuts in all the probes, even with a smaller sky area. This is due to the lower noise level of the SPT maps. However, since the higher signal-to- κ t 5https://github.com/rmjarvis/TreeCorr. 023530-9 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) FIG. 2. Measurement of the MAGLIM galaxy density-CMB lensing correlation (top) and galaxy shear-CMB lensing correlation (bottom). For each set of measurements, the upper row shows measurement with the SPT þ Planck CMB lensing map and the lower row shows measurement with the Planck CMB lensing map. The shapes and amplitudes are different due to the difference in the L cut and smoothing of the CMB lensing map. The light (dark) shaded regions in the hδgκ CMBi panels indicate the data points removed when assuming linear (nonlinear) galaxy bias, while the shaded regions in the hγ CMBi panels show the data points removed in all cases (only two bins require scale cuts). The dashed dark gray line shows the best-fit fiducial model for the fiducial lens sample, while the χ2 per degree of freedom (ν) evaluated at the best-fit model with scale cuts for linear galaxy bias model is shown in the upper left corner of each panel. κ t t κ CMBi, even though hδgκ noise necessitates a more stringent scale cut, the resulting signal-to-noise after scale cuts is only slightly higher for the SPT þ Planck patch. Finally, comparing hδgκ CMBi and hγ CMBi starts with ∼75% more signal-to-noise before scale cuts compared to hγ CMBi, the scale cuts remove significantly more signal in hδgκ CMBi compared to hγ CMBi. This is due to limits in our ability to model nonlinear galaxy bias on small scales—indeed we see that the signal-to-noise in hδgκ CMBi increases by 13% when switching from linear to nonlinear galaxy bias model. Overall, these signal-to-noise levels are consistent with the forecasts in Paper I. κ κ t t V. BLINDING AND UNBLINDING Following [16], we adopt a strict, multilevel blinding procedure in our analysis designed to minimize the impact of experimenter bias. The first level of blinding occurs at the shear catalog level, where all shears are multiplied by a 023530-10 secret factor [51]. The second level of blinding occurs at the two-point function level, where we follow the procedure outlined in [60] and shift the data vectors by an unknown amount while maintaining the degeneracy between the different parts of the data vector under the same cosmology. The main analyses in this paper were conducted after the unblinding of the shear catalog, so the most relevant blinding step is the data vector blinding. Below we outline the list of tests that were used to determine whether our measurement is sufficiently robust to unblind: (i) Pass all tests described in Appendix B, which indicate no outstanding systematic contamination in the data vectors. These tests include: (1) check for spurious correlation of our signal with survey property maps, (2) check the cross-shear component of hγ CMBi, (3) check the impact of weights used for the lens galaxies, (4) check the effect of the point-source mask in the CMB lensing map on our measurements, and (5) check that cross-correlating an external large-scale κ t JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) TABLE II. Signal-to-noise for the hδgκCMBi þ hγtκCMBi data vector when different scale cuts are applied. Rows involving the two high-redshift MAGLIM bins and the REDMAGIC sample are shown in gray to indicate that they are not part of the fiducial analysis. the different parts of Scale cuts None Linear bias Nonlinear bias SPT þ Planck hδgκCMBi MAGLIM hδgκCMBi MAGLIM 6 bin hδgκCMBi REDMAGIC hγ CMBi MAGLIM κ t 26.8 30.2 23.7 15.0 17.9 20.5 17.0 10.4 CMBi MAGLIM CMBi MAGLIM 6 bin CMBi REDMAGIC Planck hδgκ hδgκ hδgκ hγtκCMBi MAGLIM Combined hδgκCMBi MAGLIM 32.2 hγtκCMBi MAGLIM 18.2 hδgκCMBi þ hγtκCMBi MAGLIM 34.8 14.5 17.4 14.2 13.4 13.1 15.9 12.5 10.4 19.6 16.9 23.9 17.3 20.0 15.7 13.4 13.8 16.8 12.8 10.4 22.2 16.9 25.7 structure tracer (the cosmic infrared background in this case) with different versions of our CMB lensing maps yields consistent results. (ii) With unblinded chains, use the posterior predictive distribution (PPD) method developed in [61] to evaluate the consistency between the two subsets of the data vectors that use different CMB lensing maps (i.e. the SPT þ Planck patch and the Planck patch). The p-value should be larger than 0.01. (iii) With unblinded chains, verify that the goodness-of- fit of the data with respect to the fiducial model has a p-value larger than 0.01 according to the same PPD framework. CMBi þ hγ Except for the first step, all the above are applied to the hδgκ CMBi data vectors with the fiducial analysis choices (ΛCDM cosmology and linear galaxy bias scale cuts), for the first four bins of the MAGLIM lens sample. κ t VI. PARAMETER CONSTRAINTS FROM CROSS-CORRELATIONS OF DES WITH CMB LENSING CMBi þ hγ Following the steps outlined in the previous section, we found (1) no evidence for significant systematic biases in our measurements, as shown in Appendix B, (2) we obtain a p-value greater than 0.01 when comparing the hδgκ CMBi constraints from the Planck region to constraints from the SPT þ Planck region, and (3) the goodness-of-fit test of the fiducial hδgκ CMBi unblinded chain has a p-value greater than 0.01. In the following, we will quote the precise p-values obtained from these tests using the updated covariance matrix. CMBi þ hγ κ κ t t With all the unblinding tests passed, we froze all analysis choices and unblinded our cosmological constraints. We then updated the covariance matrix to match the best-fit parameters from the cosmological analysis.6 The results we present below use the updated covariance matrix. The main constraints on cosmological parameters are summarized in Table III. A. Cosmological constraints from cross-correlations In Fig. 3 we show constraints from hδgκ CMBi using the first 4 bins of the MAGLIM sample. For compari- son, we also show constraints from hγ CMBi-only, cosmic shear (from [62,63]), and 3 × 2 pt (from [16]). κ t We find that our analysis of hδgκ CMBi þ hγ CMBi þ hγ CMBi gives κ κ t t the following constraints: m ¼ 0.272þ0.032 Ω −0.052 ; σ8 ¼ 0.781þ0.073 −0.073 ; S8 ¼ 0.736þ0.032 −0.028 : κ t m constraints. While hδgκ As can be seen from Fig. 3 and expected from Paper I, the constraints are dominated by hγ CMBi slightly improving the Ω CMBi by itself does not tightly constrain cosmology because of the degeneracy with galaxy bias, the shape information in hδgκ CMBi provides additional information on Ωm when combined with hγ CMBi, with hδgκ κ CMBi. t t t κ κ CMBi þ hγ Figure 3 also shows constraints from DES-only probes, including cosmic shear and 3 × 2 pt. We find that the constraints on S8 from hδgκ CMBi are compa- rable to those from cosmic shear and 3 × 2 pt, and in reasonable agreement. The uncertainties of the hδgκ CMBi þ hγ CMBi constraints on S8 are roughly 30% (70%) larger than that of cosmic shear (3 × 2 pt). We will perform a complete assessment of consistency between these probes in Paper III. We can also see that the degeneracy direction of the hδgκ CMBi constraints are slightly differ- ent from 3 × 2 pt, which will help in breaking degeneracies when combined. CMBi þ hγ κ t We consider constraints from the SPT þ Planck and Planck patches separately in Fig. 4. As discussed earlier in Sec. V, the consistency of these two patches was part of the unblinding criteria, thus these two constraints are consistent 6This procedure is the same as in [16]. Since we cannot know the cosmological and nuisance parameters exactly before running the full inference, a set of fiducial parameters were used to generate the first-pass of the covariance that was used for all blinded chains. After unblinding, we update the parameters to values closer to the best-fit parameters from the data. After confirming that the 5 × 2 pt best-fit constraints Paper III are consistent with the 3 × 2 pt best-fit constraints, we chose to use the 3 × 2 pt best-fit parameters for evaluating the covariance matrix, as this makes our modeling choices more consistent with that of [16]. 023530-11 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) TABLE III. ΛCDM constraints on Ω CMBi þ hγ CMBi and different lens samples. We show the constraints using both linear and nonlinear galaxy bias. The last column shows the p-value corresponding to the goodness of fit for the chain. The parts shown in gray indicate that they are not part of the fiducial samples. m, σ8 and S8 using hδgκ κ t Dataset hγtκCMBi MAGLIM hδgκCMBi þ hγtκCMBi MAGLIM 4 bin linear galaxy bias hδgκCMBi þ hγtκCMBi MAGLIM 4 bin nonlinear galaxy bias hδgκCMBi þ hγtκCMBi MAGLIM 6 bin linear galaxy bias hδgκCMBi þ hγtκCMBi MAGLIM 6 bin nonlinear galaxy bias hδgκCMBi þ hγtκCMBi REDMAGIC linear galaxy bias hδgκ CMBi REDMAGIC nonlinear galaxy bias CMBi þ hγ κ t under the PPD metric. We find a p-value of 0.37 (0.33) when comparing the Planck (SPT þ Planck) results to constraints from SPT þ Planck (Planck). We also observe that the constraints are somewhat tighter in the SPT þ Planck patch in S8, consistent with the slightly larger signal-to-noise (see Table II). We note however, that the signal-to-noise before scale cuts of the SPT þ Planck patch is significantly larger than the Planck patch due to the lower noise and smaller beam size of the SPT lensing map (for hδgκ CMBi: 26.8 vs. 17.9; for hγ CMBi: 15.0 vs. 10.4), though most of the signal-to-noise is on the small scales which we had to remove due to uncertainties in the theoretical modeling. This highlights the importance of improving the small-scale modeling in future work. κ t FIG. 3. Constraints on cosmological parameters Ω from hδgκ κ show the corresponding constraints from hγ shear and 3 × 2 pt for comparison. m, σ8, and S8 CMBi using the MAGLIM sample. We also CMBi-only, cosmic t CMBi þ hγ κ t σ8 0.790þ0.080 −0.092 0.781þ0.073 −0.073 0.820þ0.079 −0.067 0.755þ0.071 −0.071 0.769þ0.071 −0.071 0.793þ0.072 −0.083 0.794þ0.069 −0.069 Ωm 0.270þ0.043 −0.061 0.272þ0.032 −0.052 0.245þ0.026 −0.044 0.288þ0.037 −0.053 0.273þ0.034 −0.047 0.266þ0.036 −0.050 0.253þ0.030 −0.046 S8 0.740þ0.034 −0.029 0.736þ0.032 −0.028 0.734þ0.035 −0.028 0.732þ0.032 −0.029 0.727þ0.035 −0.028 0.738þ0.034 −0.030 0.723þ0.033 −0.030 PPD p-value 0.72 0.50 0.51 0.45 0.45 0.39 0.41 B. Lensing ratio and IA modeling As discussed in Sec. II, we have included the lensing ratio likelihood in all our constraints. As was investigated in detail in [17], the inclusion of the lensing ratio informa- tion mainly constrains the IA parameters and source galaxy redshift biases. The TATT IA model adopted here is a general and flexible model that allows for a large range of possible IA contributions. As such, it is expected that including the lensing ratio could have a fairly large impact for data vectors that are not already constraining the IA parameters well. We now examine the effect of the lensing ratio on our fiducial hδgκ CMBi constraints by first removing the lensing ratio prior in our fiducial result, and then doing the same comparison with a different, CMBi þ hγ κ t m, σ8, and S8 FIG. 4. Constraints on cosmological parameters Ω using the hδgκ κ CMBi probes. We also show the con- t straints only using the SPT þ Planck area and only using the Planck area. CMBi þ hγ 023530-12 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) t κ CMBi þ hγ other probes in the plot. We note that this degeneracy is likely sourced by the lensing ratio likelihood, which on its own is degenerate in the η1 − η2 plane. This is consistent with what we have seen in the simulations in Paper I. The fact that it appears more prominent in hδgκ CMBi than in the other probes is partly related to the fact that a1 and a2 are constrained to be further away from zero in the case of hδgκ CMBi, allowing η1 and η2 (the redshift evolution of the terms associated with a1 and a2) to be constrained better. Another relevant factor is that hδgκ CMBi probes slightly larger redshift ranges than cosmic shear and 3 × 2 pt due to the CMB lensing kernal, which allows for a longer redshift lever arm to constrain η1 and η2, resulting in qualitatively different behaviors in the η1 − η2 parameter space. CMBi þ hγ CMBi þ hγ κ κ t t C. Nonlinear galaxy bias FIG. 5. Constraints on cosmological parameters Ωm, σ8, and S8 using the hδgκCMBi þ hγtκCMBi probes with and without includ- ing the lensing ratio (SR) likelihood, and when assuming the NLA IA model instead of our fiducial IA model TATT. more restrictive IA model, the NLA model (see Sec. II). These results are shown in Fig. 5. We make several observations from Fig. 5. First, the lensing ratio significantly tightens the constraints in the S8 direction (roughly a factor of 2), as expected from Paper I. Second, without the lensing ratio, different IA models result in different S8 constraints, with TATT resulting in ∼40% larger uncertainties than NLA. This is expected given that TATT is a more general model with three more free parameters to marginalize over compared to NLA. That being said, the constraints are still fully consistent when using the different IA models. Third, when lensing ratio is included, there is very little difference in the constraints between the two different IA models. This suggests that the IA constraints coming from the lensing ratio are sufficient to make the final constraints insensitive to the particular IA model of choice. Finally, it is interesting to look at the constraints on the IA parameters for our fiducial hδgκ CMBi analy- sis with and without the lensing ratio. We show this in Fig. 6, and compare them with constraints from cosmic shear [62,63] and 3 × 2 pt [16]. We find two noticeable degeneracies in these parameters: CMBi þ hγ κ t (i) The lensing ratio restricts the a1 − a2 parameter space to a narrow band. This is seen in the cosmic shear and 3 × 2 pt results, as well as the hδgκ CMBi þ hγ CMBi results, although hδgκ CMBi pre- fers somewhat higher a2 values. CMBi þ hγ κ κ (ii) There is a noticeable η1 − η2 degeneracy that shows CMBi þ hγ CMBi and not in the up uniquely in hδgκ κ t t t As discussed in Sec. II, we test a nonlinear galaxy bias model in addition to our baseline linear galaxy bias analysis. With a nonlinear galaxy bias model we are able to use somewhat smaller scales and utilize more signal in the data (see Table II). In Fig. 7 we show the cosmological constraints of our fiducial hδgκ CMBi data vector with the nonlinear galaxy bias model. We find that the constraints between the two different galaxy bias models are consistent. There is a small improvement in the Ω m direction, which is not surprising given that nonlinear bias impacts hδgκ CMBi improves the Ω CMBi, and hδgκ m con- straints relative to hγ κ CMBi alone. The overall improvement t is nevertheless not very significant, as hγ CMBi is domi- nating the constraints. CMBi þ hγ κ κ t t D. Comparison with alternative lens choices t κ CMBi þ hγ We have defined our fiducial lens sample to be the first four bins of the MAGLIM sample. This choice is informed by the 3 × 2 pt analysis in [16], where alternative lens samples were also tested but were deemed to be potentially contaminated by systematic effects and therefore not used in the final cosmology analysis. Here, we examine the hδgκ CMBi constraints using the two alternative choices for lenses: (1) including the two high-redshift bins in MAGLIM to form a 6-bin MAGLIM sample, and (2) the REDMAGIC lens sample. As we have emphasized through- out the paper, since the galaxy-CMB lensing cross- correlation is in principle less sensitive to some of the systematic effects, these tests could potentially shed light on the issues seen in [16]. We only examine the hδgκ CMBi constraints here, but will carry out a more extensive investigation in combination with the 3 × 2 pt probes in Paper III. CMBi þ hγ κ t In Fig. 8 we show constraints from hδgκ CMBi lens samples: 4-bin MAGLIM CMBi þ hγ κ t using the three different 023530-13 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) FIG. 6. Constraints on S8 and the IA parameters from our fiducial hδgκ include the hδgκ CMBi þ hγ CMBi constraints without the lensing ratio (SR) likelihood for comparison. CMBi þ hγ t κ CMBi results, cosmic shear and 3 × 2 pt. We also κ t t κ (fiducial), 6-bin MAGLIM and REDMAGIC. The best-fit parameters as well as the goodness-of-fit are listed in Table III. Broadly, all three constraints appear to be very consistent with each other. This is not surprising given that the constraining power is dominated by hγ CMBi as we discussed earlier. In [16] it was shown that for the 3 × 2 pt analysis, both the 6-bin MAGLIM and the REDMAGIC samples give goodness-of-fits that fail our criteria, while for hδgκ CMBi all three samples give acceptable goodness-of-fits values as seen in Table III. This could imply that the systematic effects that contaminated the other correlation functions in 3 × 2 pt are not affecting the hδgκ CMBi results strongly. Compared to the fiducial constraints, the constraining power of the 6-bin MAGLIM sample is slightly higher in the Ω m direction due to the added signal-to-noise from the high-redshift bins, while CMBi þ hγ CMBi þ hγ κ κ t t m and S8. the constraining power of the REDMAGIC sample is slightly lower in both Ω The DES Y3 3 × 2 pt analyses found that the poor fits for the alternative lens samples can be explained by inconsistent galaxy bias between galaxy-galaxy lensing hδgγ ti and galaxy clustering hδgδgi. That is, when allowing the galaxy bias to be different in galaxy-galaxy lensing and galaxy clustering, improves signifi- the goodness-of-fit cantly. Operationally, this is achieved in [16] by adding a free parameter, Xlens, defined such that Xi lens ¼ bi hδgγti=bi hδgδgi; ð18Þ hδgγti (bi where bi hδgγ hδgδgi) is the linear galaxy bias parameter for ti (hδgδgi) in lens galaxy redshift bin i. Xlens is expected 023530-14 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) FIG. 7. Fiducial hδgκ logical parameters Ω galaxy bias models. CMBi þ hγ CMBi constraints on cosmo- m, σ8, and S8 using linear and nonlinear κ t in general the constraints from hδgκ Our CMB lensing cross-correlation analysis provides an interesting way to explore this systematic effect. In essence, with fixed cosmology, we can fit for galaxy bias using hδgκ CMBi and compare with the galaxy bias derived from hδgγti and hδgδgi. Our results are shown in Fig. 9. We find that CMBi on galaxy bias are weaker than both galaxy-galaxy lensing and galaxy clustering, this is expected due to the lower signal-to-noise. As such, the hδgκ CMBi-inferred galaxy bias values are largely consistent with both galaxy-galaxy lensing and galaxy clustering. There are a few bins, though, where hδgκ CMBi does show a preference for the galaxy bias values to agree more with one of the two probes. Noticeably, for the last two MAGLIM bins, hδgκ CMBi prefers a galaxy bias value that is closer to that inferred by galaxy clustering. On the other hand, for the highest two REDMAGIC bins, hδgκ CMBi prefers galaxy bias values that are closer to galaxy-galaxy lensing. These findings are investigations on Xlens consistent with the various described in [34,64] and suggest potential issues in the measurements or modeling of galaxy-galaxy lensing in the two high-redshift MAGLIM bins and galaxy clustering in the REDMAGIC sample.7 However, we caution that these results can be cosmology-dependent, and change slightly if a different cosmology is assumed. t t κ E. Implications for S8 tension In Fig. 10, we compare our constraints on S8 from hγ κ CMBi to those from recent measurements of cosmic shear from galaxy surveys (light blue circles) as well as other recent hγ CMBi constraints (dark blue squares). We κ show only the constraint CMBi (rather than t hδgκ κ to compare only CMBi) since we want measurements of gravitational lensing. These lensing measurements are not sensitive to the details of galaxy bias, unlike hδgκ CMBi. We see that the constraints on S8 obtained from hγ κ CMBi in this work (gray band) are for the t first time comparable to the state-of-the-art cosmic shear measurements. CMBi þ hγ from hγ t FIG. 8. Fiducial constraints on cosmological parameters Ω m, σ8, and S8 using the hδgκ CMBi þ hγ CMBi probes compared with using the REDMAGIC lens sample instead of the MAGLIM lens sample. κ t Figure 10 also shows the inferred value of S8 from the primary CMB (black triangles), as measured by Planck [21], ACT DR4 [65], combining ACT DR4 and the Wilkinson Microwave Anisotropy Probe [WMAP, 65], and SPT-3G [66]. As discussed in several previous works [e.g., 62,63,67] and can be seen in the figure, there is a ∼2.7σ tension8 between the S8 value inferred from cosmic to equal 1 in the case of no significant systematic effects. ≠ 1 for the two high-redshift In [16] it was found that Xlens bins in the MAGLIM sample and for all bins in the REDMAGIC though there was not enough information to sample, determine whether the systematic effect was in hδgγ ti or hδgδgi. 7In particular, [34] tested an alternative REDMAGIC sample and suggested potential remedies to the systematic effect in REDMAGIC that will be explored in future work. 8Here we are quoting the 1D parameter difference in S8, or 8 − S2 , where the superscript 1 and 2 refer ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 8Þ þ σ2ðS2 σ2ðS1 8Þ ðS1 to the two datasets we are comparing. 8Þ= p 023530-15 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) FIG. 9. With fixed cosmological parameters, the inferred galaxy bias from hδgκ the MAGLIM sample (top) and the REDMAGIC sample (bottom). CMBi, galaxy-galaxy lensing and galaxy clustering, for FIG. 10. Comparison of late-time measurements of S8 from lensing-only data (cosmic shear hγγi and galaxy shear-CMB lensing cross- correlation hγ CMBi) to the inferred value of S8 from the primary CMB. κ t shear and the Planck primary CMB constraint—cosmic shear results prefer a lower S8 value. This is intriguing given that it could indicate an inconsistency in the ΛCDM model. We also see that the other CMB datasets are currently much less constraining, but show some variation, with the lowest S8 value from SPT-3G fairly consistent with all the cosmic shear results. With this work, we can now meaningfully add hγ CMBi into this comparison, and as we see in Fig. 10, the hγ CMBi constraints on S8 are also largely below that coming from the primary CMB. This is potentially exciting, since the hγ CMBi measurements come from a cross-correlation between two very different surveys, and are therefore κ κ κ t t t expected to be highly robust to systematic errors. Our results therefore lend support to the existence of the S8 tension. In Paper III we will perform a more rigorous and complete analysis of the consistency of our constraints here with other datasets. VII. SUMMARY We have presented measurements of two cross- correlations between galaxy surveys and CMB lensing: the galaxy position-CMB lensing correlation (hδgκ CMBi), and the galaxy shear-CMB lensing correlation (hγ κ CMBi). t These measurements are sensitive to the statistics of 023530-16 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) large-scale structure, and are additionally expected to be very robust to many observational systematics. Our measurements make use of the latest data from the first three years of observations of DES, and a new CMB lensing map constructed explicitly for cross-correlations using SPT and Planck data. In particular, our fiducial results are from four tomographic bins of the MAGLIM lens galaxy sample. The signal-to-noise of the full data vector without angular scale cuts is ∼30; the part of the data vector used for cosmological inference has a signal-to- noise of ∼20. The main reduction of the signal-to-noise comes from uncertainty in the modeling of nonlinear galaxy bias, which necessitates removal of the small-angle hδgκ CMBi correlation measurements. Compared to the DES Y1 analysis, the signal-to-noise increased by a factor of ∼2 and we are no longer limited by contamination of tSZ in the CMB lensing map. −0.044 ; S8 ¼ 0.734þ0.035 The joint analysis of these two cross-correlations results −0.052 ; S8 ¼ 0.736þ0.032 m ¼ 0.272þ0.032 in the constraints Ω −0.028 m ¼ 0.245þ0.026 (Ω −0.028 ) when assuming lin- ear (nonlinear) galaxy bias in our modeling. For S8, these constraints are more than a factor of 2 tighter than our DES Y1 results, ∼30% looser than constraints from DES Y3 cosmic shear and ∼70% looser than constraints from DES Y3 3 × 2 pt. We highlight here several interesting findings from this work: t t t κ κ κ (i) We find that hγ κ CMBi þ hγ CMBi þ hγ CMBi dominates the constraints in the hδgκ CMBi combination, confirming t our findings from the simulated analysis in Paper I. (ii) We find that the lensing ratio has a large impact on the hδgκ CMBi constraints, improving the S8 constraints by ∼40%. In addition, the hδgκ CMBi þ hγ CMBi data vector constrains the η1 − η2 degen- eracy direction, something not seen in the DES Y3 3 × 2 pt data vectors. (iii) We investigate the use of two alternative lens samples for the analysis: the 6-bin MAGLIM sample and the REDMAGIC sample. In contrast to the fiducial DES Y3 3 × 2 pt analysis, we find that the hδgκ CMBi analysis using the two alter- native lens samples pass our unblinding criteria and show no signs of systematic contamination. (iv) With fixed cosmology, we use the hδgκ CMBi þ hγ κ t t κ CMBi þ hγ CMBi data vector to constrain the galaxy bias values using the 6-bin MAGLIM sample and the REDMAGIC sample. For the two high-redshift MAGLIM bins, we find bias values that agree more with galaxy clustering. On the other hand, for the REDMAGIC sample, we find bias values more con- sistent with galaxy-galaxy lensing. These provide additional information for understanding the sys- tematic effect seen in [16] from these two alternative lens samples. κ (v) Comparing with previous cosmic shear and hγ CMBi constraints, we find that in line with previous findings, our hγ CMBi constraint on S8 is lower than the primary CMB constraint from Planck. In addition, for the first time, hγ CMBi has achieved comparable precision to state-of-the-art cosmic shear constraints. κ κ t t t The constraints derived in this paper from hδgκ CMBi þ hγ κ CMBi can now be compared and combined with the t DES Y3 3 × 2 pt probes [16], which we will do in Paper III. We will present therein our final combined results along with tests for consistency with external datasets. It is however intriguing that with the galaxy-CMB lensing cross-correlation probes alone, our datasets provide very competitive constraints on the late-time large-scale struc- ture compared to galaxy-only probes. Due to the relative insensitivity to certain systematic effects, this additional constraint is especially important for cross-checking and significantly improving the robustness of the galaxy-only results. Another unique aspect of this work compared to other cross-correlation analyses is that we have carried out our work in an analysis framework that is fully coherent with the galaxy-only probes, making it easy to compare and combine. Looking forward to the final datasets from DES, SPT, and ACT, as well as datasets from the Vera C. Rubin Observatory’s Legacy Survey of Space and Time9 (LSST), the ESA’s Euclid mission,10 the Roman Space Telescope,11 (SO), and CMB Stage-413 the Simons Observatory12 (CMB-S4), our results show that there are significant opportunities for combining the galaxy and CMB lensing datasets to both improve the constraints on cosmological parameters and to make the constraints themselves more robust to systematic effects. ACKNOWLEDGMENTS C. C. and Y. O. are supported by DOE grant No. DE- SC0021949. The South Pole Telescope program is sup- ported by the National Science Foundation (NSF) through the grant No. OPP-1852617. Partial support is also pro- vided by the Kavli Institute of Cosmological Physics at the University of Chicago. Argonne National Laboratory’s work was supported by the U.S. Department of Energy, Office of Science, Office of High Energy Physics, under contract No. DE-AC02- 06CH11357. Work at Fermi National Accelerator Laboratory, a DOE-OS, HEP User 9https://www.lsst.org. 10https://www.euclid-ec.org. 11https://roman.gsfc.nasa.gov. 12https://simonsobservatory.org/. 13https://cmb-s4.org/. 023530-17 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) support the National Center the University of Facility managed by the Fermi Research Alliance, LLC, was supported under Contract No. DE-AC02- 07CH11359. The Melbourne authors acknowledge support from the Australian Research Council’s Discovery Projects scheme (No. DP200101068). The McGill authors acknowledge funding from the Natural Sciences and Engineering Research Council of Canada, Canadian Institute for Advanced research, and the Fonds de recherche du Qu´ebec Nature et technologies. The CU Boulder group acknowledges support from NSF Grant No. AST-0956135. The Munich group acknowledges the support by the ORIGINS Cluster (funded by the Deutsche Forschungsge- meinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy—EXC-2094–390783311), the MaxPlanck-Gesellschaft Faculty Fellowship Program, and the Ludwig-Maximilians-Universität München. J. V. acknowledges from the Sloan Foundation. Funding for the DES Projects has been provided by the U.S. Department of Energy, the U.S. National Science the Ministry of Science and Education of Foundation, Spain, the Science and Technology Facilities Council of the United Kingdom, the Higher Education Funding Council for England, for Supercomputing Illinois at Urbana- Applications at Champaign, the Kavli Institute of Cosmological Physics at the University of Chicago, the Center for Cosmology and Astro-Particle Physics at the Ohio State University, the Mitchell Institute for Fundamental Physics and Astronomy at Texas A&M University, Financiadora de Estudos e Projetos, Fundação Carlos Chagas Filho de Amparo `a Pesquisa do Estado do Rio de Janeiro, Conselho Nacional de Desenvolvimento Científico e Tecnológico and the Minist´erio da Ciência, Tecnologia the Deutsche Forschungsgemeinschaft and the Collaborating Institutions in the Dark Energy Survey. The Collaborating Institutions are Argonne National Laboratory, the University of California at Santa Cruz, the University of Cambridge, Centro de Investigaciones Energ´eticas, Medioambientales y Tecnológicas-Madrid, the University of Chicago, University College London, the DES-Brazil Consortium, the University of Edinburgh, the Eidgenössische Technische Hochschule (ETH) Zürich, Fermi National Accelerator Laboratory, the University of Illinois at Urbana-Champaign, the Institut de Ci`encies de l’Espai (IEEC/CSIC), the Institut de Física d’Altes Energies, Lawrence Berkeley National Laboratory, the Ludwig-Maximilians Universität München and the associated Excellence Cluster Universe, the University of Michigan, NFS’s NOIRLab, the University of Nottingham, The Ohio State University, the University of Pennsylvania, the University of Portsmouth, SLAC National Accelerator Laboratory, Stanford University, the University of Sussex, Inovação, e system is Texas A&M University, and the OzDES Membership Consortium. Based in part on observations at Cerro Tololo Inter-American Observatory at NSF’s NOIRLab (NOIRLab Prop. ID 2012B-0001; PI: J. Frieman), which is managed by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation. The DES data management supported by the National Science Foundation under Grants No. AST-1138766 and No. AST-1536171. The DES participants from Spanish institutions are partially supported by MICINN under grants No. ESP2017-89838, No. PGC2018-094773, No. PGC2018-102021, No. SEV-2016-0588, No. SEV- 2016-0597, and No. MDM-2015-0509, some of which include ERDF funds from the European Union. I. F. A. E. is partially funded by the CERCA program of the Generalitat de Catalunya. Research leading to these results has received funding from the European Research Council under the European Union’s Seventh Framework Program (No. FP7/2007-2013) including ERC grant agreements No. 240672, No. 291329, and No. 306478. We acknowl- edge support from the Brazilian Instituto Nacional de Ciência e Tecnologia (INCT) do e-Universo (CNPq grant No. 465376/2014-2). This manuscript has been authored by Fermi Research Alliance, LLC under Contract No. DE- AC02-07CH11359 with the U.S. Department of Energy, Office of Science, Office of High Energy Physics. APPENDIX A: JACKKNIFE COVARIANCE MATRIX We have performed extensive validation tests on our methodology of modeling in the covariance matrix in Paper I. The ultimate check, however, is to compare the covariance matrix with a data-driven jackknife covariance matrix. The jackknife covariance incorporates naturally the noise in the data as well as any non-cosmological spatial variation in the data that might be important. This comparison was done after unblinding and the update of the covariance described in footnote 6, and is only used as a confirmation—that is, we cannot change any analysis choices based on this check. into 80 patches) In Fig. 11 we show the diagonal elements of the covariance matrix (calculated using the jackknife delete-one block jackknife method by dividing the foot- print lens sample, compared with our fiducial covariance matrix. We find excellent agreement between them on all scales, both hδgκ κ CMBi and hγ CMBi, and on both the SPT þ Planck t and Planck patch. the fiducial for 023530-18 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) FIG. 11. Comparison between the diagonal elements of the jackknife covariance and our fiducial covariance matrix (analytical covariance with noise-noise correction applied). APPENDIX B: DIAGNOSTIC TESTS We perform a number of diagnostic tests to make sure that our measurements are not significantly contaminated by potential systematic effects. As we have discussed in Sec. I, cross-survey correlations like those presented here to possible are expected to be inherently more robust systematic effects. In addition, extensive tests have been done on both the galaxy and CMB data products in [13,47,51,62,63,68]. We perform a series of diagnostic tests specific to the cross-correlation probes. 1. Cross-correlation with survey property maps If a given contaminant associated with some survey property simultaneously affects the galaxy and the CMB fields that we are cross-correlating, the cross-correlation signal will contain a spurious component is not cosmological. A possible example is dust, which could simultaneously contaminate the CMB lensing map (through thermal emission in CMB bands) and the galaxy density field (through extinction). In addition to dust, we consider several other possible survey properties. This test is designed to detect such effects. We calculate the correlation statistic, Xf S, between the observables of interest and various survey property maps: that Xf SðθÞ ¼ hκ CMB SðθÞihfSðθÞi hSSðθÞi ; ðB1Þ CMBi and Xf where S is the survey property map of interest, and f is either δg or γ t. This expression captures correlation of the systematic with both κ CMB and f, and is normalized to have the same units as hfκ the CMBi. Henceforth, we omit θ-dependence in the notation for simplicity, but note that all the factors in Eq. (B1) are functions of θ. Unless the systematic map is correlated with both f and κ CMB, it will not bias hfκ S will be consistent with zero. Note that Xf S should also be compared with the statistical uncertainty of hfκ CMBi, as a certain systematic could be significantly detected but have little impact on the final results if it is much smaller than the statistical uncertainty. CMBi, we consider two S fields: stellar density and extinction. For hγ CMBi, we look in addition at two fields associated with PSF modeling errors. The quantities q and w measure the point-spread function (PSF) modeling residuals as introduced in [51], q ¼ e(cid:2) is the difference of the true ellipticity of the PSF as measured by stars and that inferred by the PSF model, and − TmodelÞ=T(cid:2), where T is a measure of size w ¼ e(cid:2)ðT(cid:2) of the PSF, is the impact on the PSF model ellipticity when For hδgκ − emodel κ t 023530-19 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) FIG. 12. The measured systematic contamination of hδgκCMBi for the MAGLIM lens sample, as assessed by Eq. (B1), for the SPT þ Planck field (top) and the Planck field (bottom) and for different redshift bins. For reference, the gray band shows 10% of the statistical uncertainties for the corresponding data vectors. In all cases, the measured bias is significantly below the statistical uncertainties on the hδgκCMBi measurements. FIG. 13. Same as Fig. 12 but for the REDMAGIC lens sample. − Tmodel. As both q and w are the PSF size is wrong by T(cid:2) spin-2 quantities like the ellipticity, we first decompose them into E and B modes using the same method used for generating weak lensing convergence maps in [69]. We then use the E-mode maps as the S maps to perform the cross-correlation test. The rationale here is that if there is a nontrivial E-mode component, it could signify con- tamination in the shear signal and will correlate with the shear field. Figures 12–14 show the result of our measured Xf S for the different parts of the data vector. For comparison, we also plot the statistical uncertainty on the data vector; given that the statistical uncertainties are much larger than the measured biases in all cases, we scale the statistical uncertainties by 0.1 (hδgκ CMBi) and 0.01 CMBi). The χ2 values per degree of freedom for (hγ κ t the Xf S measurements with respect to the null model are shown in Tables IV–VI together with the probability-to- exceed (PTE) values. The χ2 as well as the error bars on the plots are derived from jackknife resampling where we use 65 equal-area jackknife patches for the SPT þ Planck footprint and 85 patches for the Planck area. To obtain a more reliable jackknife covariance, we measure Xf S using 10 angular bins instead of the 20 bins used for the data vectors. In general, most of the systematic effects are very consistent with zero. For hδgκ CMBi, we find that the absolute level of the potential systematic effects as quantified by Xf S is 1-2 orders of magnitudes smaller than the statistical errors. There appears to be more cross-correlation for the SPT þ Planck area, especially with extinction. All of the PTE values of these cross-correlations are above our threshold 023530-20 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) FIG. 14. The measured systematic contamination of hγtκCMBi, as assessed by Eq. (B1), for the SPT þ Planck field (top) and the Planck field (bottom) and for different redshift bins. The gray band shows 1% of the statistical uncertainties for the corresponding data vectors. t κ for concern of 0.01, so we deem these results acceptable. CMBi, we find that the absolute levels of the Xf For hγ S measurements is much lower (> 2 orders of magnitude)— this is expected as it is much less obvious how the survey property maps will leave an imprint on the shear field. Interestingly, we also find that overall the error bars are larger in the SPT þ Planck patch compared to the Planck patch. This can be due to the survey property maps containing higher fluctuation in the SPT þ Planck area as part of the footprint is close to the galactic plane or the Large Magellanic Cloud (LMC). spatial 2. Cross-shear component t × κ κ κ During the measurement of hγ CMBi, we additionally measure its cross-shear counterpart hγ CMBi. We replace et in Eq. (16) with e×, the corrected ellipticity oriented 45° to the line connecting map pixel and the source galaxy. The correlation hγ CMBi should be consistent with zero. Any × significant detection of hγ CMBi could signal systematic effects in the hγ κ × CMBi measurements. Our results are shown in Fig. 15 with the χ2 per degree of freedom and PTE values listed in Table VI. We find no significant detection of hγ κ CMBi in all parts of the data vector. κ × t TABLE IV. The χ2 per degree of freedom for the systematics diagnostics quantity [Eq. (B1)] for the MAGLIM hδgκ CMBi mea- surements. The different columns represent the different survey properties S, whereas the different rows are for the tomographic bins in both the SPT þ Planck patch and the Planck patch. The corresponding PTE values are listed in the parentheses. 3. hδgκCMBi measurements with and without weights As discussed in [47], weights are applied to the lens galaxies in order to remove correlations with various survey properties. When performing the hδgκ CMBi measurement in SPT þ Planck Planck S Bin 1 2 3 4 5 6 1 2 3 4 5 6 Stellar density Extinction TABLE V. Same as Table IV but for the REDMAGIC lens sample. χ2=d:o:f: (PTE) 0.42 (0.85) 0.10 (0.99) 0.21 (0.98) 0.13 (0.99) 0.22 (0.98) 0.36 (0.93) 0.02 (0.99) 0.12 (0.99) 0.15 (0.99) 0.08 (0.99) 0.06 (0.99) 0.05 (0.99) 0.90 (0.49) 0.65 (0.71) 0.64 (0.72) 1.12 (0.34) 1.34 (0.21) 1.66 (0.10) 0.40 (0.87) 0.26 (0.96) 0.28 (0.96) 0.33 (0.93) 0.21 (0.98) 0.18 (0.98) SPT þ Planck Planck S Bin 1 2 3 4 5 1 2 3 4 5 Stellar density Extinction χ2=d:o:f: (PTE) 0.09 (0.99) 0.50 (0.83) 0.42 (0.88) 0.28 (0.96) 0.73 (0.64) 0.09 (0.99) 0.09 (0.99) 0.05 (0.99) 0.04 (0.99) 0.04 (0.99) 0.20 (0.97) 0.56 (0.78) 0.38 (0.91) 0.76 (0.62) 1.13 (0.33) 0.38 (0.89) 0.16 (0.99) 0.19 (0.98) 0.16 (0.99) 0.16 (0.99) 023530-21 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) TABLE VI. The χ2 per degree of freedom for the systematics diagnostics quantity [Eq. (B1)] for the hγ CMBi measurements. The different columns represent the different survey properties S, whereas the different rows are for the tomographic bins in both the SPT þ Planck patch and the Planck patch. The corresponding PTE values are listed in the parentheses. The last column lists the corresponding numbers for the cross-shear measurement described in Sec. B 2. κ t SPT þ Planck Planck S Bin 1 2 3 4 1 2 3 4 Stellar density Extinction PSF model error q χ2=d:o:f: (PTE) PSF model error w γ× 0.12 (0.99) 0.17 (0.99) 0.32 (0.94) 0.20 (0.97) 0.09 (0.99) 0.09 (0.99) 0.12 (0.99) 0.16 (0.99) 0.12 (0.99) 0.38 (0.95) 0.39 (0.90) 0.41 (0.86) 0.06 (0.99) 0.04 (0.99) 0.07 (0.99) 0.18 (0.99) 0.34 (0.96) 0.20 (0.99) 0.40 (0.89) 0.19 (0.97) 0.11 (0.99) 0.25 (0.98) 0.19 (0.99) 0.27 (0.98) 0.15 (0.99) 0.18 (0.99) 0.30 (0.95) 0.15 (0.98) 0.08 (0.99) 0.17 (0.99) 0.14 (0.99) 0.18 (0.99) 1.11 (0.34) 1.18 (0.29) 0.60 (0.75) 1.91 (0.07) 1.15 (0.31) 1.28 (0.23) 1.16 (0.31) 1.12 (0.33) the effect of Eq. (14), these weights are applied (i.e. the ηδg). In a cross- these weights will be non- correlation, negligible if the systematic effect that is being corrected by the weights also correlates with the CMB lensing map. We note that this test is not always a null-test, as we consider it more correct to use the weights. Rather, it shows qualitatively the level of the correction from these weights—naively, the smaller the correction to start with, the less likely the residual contamination will be. In Fig. 16 we show the difference between the hδgκ CMBi measurements with and without using the lens weights, for the two lens samples. To understand the significance of these results, we calculate the Δχ2 between the data vectors with and without weights for the fiducial MAGLIM sample, using the analytic covariance for the data vector and find a Δχ2 of 1.23 after scale cuts. Propagating this into cosmo- logical constraints by running two chains using hδgκ CMBi with and without weights (fixing galaxy bias) gives a negligible 0.02σ shift in the Ω − S8 plane. It is also worth m pointing out that we see that the weights most significantly affect the two high-redshift bins in the MAGLIM sample, this is likely due to the fact that the high-redshift bins are fainter and more affected by the spatially varying observing conditions. 4. Biases from source masking In constructing the CMB lensing maps for this analysis, we apply a special procedure at the locations of bright point sources to reduce their impact on the output lensing maps. As described in more detail in Paper I, the CMB lensing estimator that we use involves two CMB maps, or “legs.” One of these is high-resolution map (i.e. the SPT þ Planck temperature map), and the other is a low-resolution tSZ-cleaned map (i.e. the Planck SMICAnosz temperature map). To reduce the impact of point sources, we inpaint the point sources with fluxes 6.4 < F < 200 mJy using the method described in [70]. The total inpainted area is roughly 3.6% of the map. The corresponding location in the tSZ-cleaned map are left untouched. We expect this procedure to result in a reasonable estimate of κ CMB at the FIG. 15. Cross-correlation between than cross-component of shear with CMB lensing for the SPT þ Planck field (top) and the Planck field (bottom) and for different redshift bins. The gray band shows the statistical uncertainties for hγ κ CMBi. t 023530-22 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) FIG. 18. Cross-correlation between CIB and the Planck lensing map in the North patch (solid gray), the Planck lensing map in the South patch (open red), and the SPT þ Planck lensing map (black). FIG. 16. The difference in the hδgκCMBi cross-correlation between the two lens galaxy samples MAGLIM and REDMAGIC with CMB lensing when using weights and without weights, over the statistical uncertainty of the measurement σ. locations of the point sources, given that only one leg is inpainted, and the area being inpainted is small (such that Gaussian constrained inpainting predicts the pixels values of the inpainted region well) although it is possible that the noise properties of these regions differ somewhat from the map as a whole. To test whether the inpainting procedure results in any bias, we also measure the cross-correlation with the lensing map after masking (i.e. completely removing) all the point sources down to 6.4 mJy. We show in Fig. 17 the difference in the data vectors using the alternative mask and the fiducial one. We find that there is no coherent difference in the correlation measurements across the range of angular scales considered. There is, however, some scatter about our nominal measurements. The level of this scatter is small, roughly 0.25 and 0.50σ across the full range of angular scales for hδgκ CMBi and hγ CMBi respectively.14 Given that such scatter is expected to have negligible impact on our results, and since some scatter between the data points is expected simply due to the different selection of pixels in the masked and unmasked CMB lensing maps, we do not find this to be a cause for worry. Our baseline results will use the unmasked version of the CMB lensing map. κ t 5. Variations in the CMB lensing map Our fiducial analysis uses the SPT þ Planck map in the Dec < −40° region and the Planck lensing map in the region Dec > −39.5°. We left a 0.5° gap between the two maps to avoid correlation between the large-scale structure on the boundary. Here we like to verify that the cross- correlation of our CMB lensing maps with another large- scale structure tracer is consistent between the two patches and the two CMB lensing data sets. We choose to use the cosmic infrared background (CIB) map from [71]15 as this large-scale structure tracer. We carry out the following two 14This scatter results from the slightly higher-noise region caused by the half-leg lensing reconstruction, with the point sources left in the non-inpainted map effectively behaving as noise. 15Here we use the nH ¼ 2.5e20 cm−1 maps as defined in [71]. FIG. 17. Difference in the data vectors using the alternative mask and the fiducial one. This test is only done for the SPT þ Planck patch, as it is specific to the SPT lensing reconstruction. 023530-23 C. CHANG et al. PHYS. REV. D 107, 023530 (2023) FIG. 19. Same as Fig. 2 but for the REDMAGIC sample. tests: (1) we compare the cross-correlations between the CIB map and the Planck lensing map split into two sub- regions (the “North” region with DEC > −39.5° and the “South” region with DEC < −40°), and verify that they are consistent; (2) we compare in the South patch the cross- correlations between the CIB map with either the Planck CMB lensing map or our SPT þ Planck lensing map, and verify that they are consistent. The resulting correlation measurements are shown in Fig. 18—the high signal-to-noise is expected due to the significant overlap in the kernels of the two tracers. We make two comparisons: (1) CIB × Planck North vs. CIB × Planck South: We find a two-sample χ2=ν of 24.28=20 with a PTE of 0.23. This demonstrates that the two patches are consistent with each other. (2) SPT þ Planck vs. Planck South: We compute the two-sample χ2, and find χ2=ν ¼ 23.9=20., with a PTE of 0.25. This demonstrates that the two mea- surements are consistent with each other. We note that there are two caveats associated with these that, at cross-correlation measurements. The first is there may be residuals. Second, 545 GHz, galactic emission is non-negligible, and while the CIB maps from [71] are intended to be free of galactic dust, CMB correlation is most sensitive to redshifts higher than those probed by DES galaxies, thus we are extrapolating the results above to lower redshift. the CIB-κ APPENDIX C: REDMAGIC RESULTS In this appendix we show the results for the second lens sample—the REDMAGIC sample. The data vector is shown in Fig. 19 with signal-to-noise values listed in Table II. We find that (1) no significant systematic effects were found as described in Appendix B, (2) we get a p-value greater than 0.01 when comparing the hδgκ CMBi con- straints from Planck to constraints from SPT þ Planck, and (3) the goodness-of-fit of the fiducial hδgκ CMBi þ hγ CMBi unblinded chain corresponds to a p-value greater than 0.01. These results allowed us to unblind our results, and the final constraints are listed in Table III and the fiducial constraints are shown in Fig. 8. CMBi þ hγ κ κ t t [1] B. Flaugher, H. T. Diehl, K. Honscheid et al. (DES [6] E. Baxter, J. Clampitt, T. Giannantonio et al., Mon. Not. R. Collaboration), Astron. J. 150, 150 (2015). Astron. Soc. 461, 4099 (2016). [2] J. E. Carlstrom, P. A. R. Ade, K. A. Aird et al., Publ. Astron. [7] E. J. Baxter, S. Raghunathan, T. M. Crawford et al., Mon. Soc. Pac. 123, 568 (2011). Not. R. Astron. Soc. 476, 2674 (2018). [3] B. Soergel, S. Flender, K. T. Story et al., Mon. Not. R. [8] J. Prat, E. Baxter, T. Shin et al., Mon. Not. R. Astron. Soc. Astron. Soc. 461, 3172 (2016). 487, 1363 (2019). [4] D. Kirk, Y. Omori, A. Benoit-L´evy et al., Mon. Not. R. [9] Y. Omori, T. Giannantonio, A. Porredon et al., Phys. Rev. D Astron. Soc. 459, 21 (2016). 100, 043501 (2019). [5] T. Giannantonio, P. Fosalba, R. Cawthon et al., Mon. Not. [10] Y. Omori, E. J. Baxter, C. Chang et al., Phys. Rev. D 100, R. Astron. Soc. 456, 3213 (2016). 043517 (2019). 023530-24 JOINT ANALYSIS OF …. II. CROSS-CORRELATION … PHYS. REV. D 107, 023530 (2023) [11] T. M. C. Abbott et al. (DES and SPT Collaborations), Phys. [42] M. S. Madhavacheril and J. C. Hill, Phys. Rev. D 98, Rev. D 100, 023541 (2019). 023534 (2018). [12] M. Costanzi, A. Saro, S. Bocquet et al., Phys. Rev. D 103, 043522 (2021). [43] B. Flaugher, Int. J. Mod. Phys. A 20, 3121 (2005). [44] I. Sevilla-Noarbe, K. Bechtol, M. Carrasco Kind et al., [13] Y. Omori, R. Chown, G. Simard et al., Astrophys. J. 849, Astrophys. J. Suppl. Ser. 254, 24 (2021). 124 (2017). [45] J. De Vicente, E. Sánchez, and I. Sevilla-Noarbe, Mon. Not. [14] N. Aghanim, Y. Akrami et al. (Planck Collaboration), R. Astron. Soc. 459, 3078 (2016). Astron. Astrophys. 641, A8 (2020). [46] A. Porredon et al. (DES Collaboration), Phys. Rev. D 103, [15] Y. Omori et al., preceding paper, Phys. Rev. D 107, 023529 043503 (2021). (2023). [16] T. M. C. Abbott et al. (DES Collaboration), Phys. Rev. D 105, 023520 (2022). [47] M. Rodríguez-Monroy, N. Weaverdyck, J. Elvin-Poole et al., Mon. Not. R. Astron. Soc. 511, 2665 (2022). [48] E. Rozo, E. S. Rykoff, A. Abate et al., Mon. Not. R. Astron. [17] C. Sánchez, J. Prat, G. Zacharegkas et al., Phys. Rev. D 105, Soc. 461, 1431 (2016). 083529 (2022). [49] E. S. Rykoff, E. Rozo, M. T. Busha et al., Astrophys. J. 785, [18] T. M. C. Abbott et al. (DES and SPT Collaborations), following paper, Phys. Rev. D 107, 023531 (2022). [19] G. A. Marques, J. Liu, K. M. Huffenberger, and J. C. Hill, 104 (2014). [50] E. S. Rykoff, E. Rozo, D. Hollowood et al., Astrophys. J. Suppl. Ser. 224, 1 (2016). Astrophys. J. 904, 182 (2020). [51] M. Gatti, E. Sheldon, A. Amon et al., Mon. Not. R. Astron. [20] H. Aihara, N. Arimoto, R. Armstrong et al., Publ. Astron. Soc. 504, 4312 (2021). Soc. Jpn. 70, S4 (2018). [21] P. A. R. Ade, N. Aghanim et al. (Planck Collaboration), [52] E. Huff and R. Mandelbaum, arXiv:1702.02600. [53] E. S. Sheldon and E. M. Huff, Astrophys. J. 841, 24 Astron. Astrophys. 594, A15 (2016). (2017). [22] T. Namikawa, Y. Chinone, H. Miyatake et al., Astrophys. J. [54] N. MacCrann, M. R. Becker, J. McCullough et al., Mon. 882, 62 (2019). Not. R. Astron. Soc. 509, 3371 (2022). [23] Z. D. Kermish, P. Ade, A. Anthony et al., Proc. SPIE Int. [55] J. Myles, A. Alarcon, A. Amon et al., Mon. Not. R. Astron. Soc. Opt. Eng. 8452, 1C (2012). Soc. 505, 4249 (2021). [24] N. C. Robertson, D. Alonso, J. Harnois-D´eraps et al., [56] W. G. Hartley, A. Choi, A. Amon et al., Mon. Not. R. Astron. Astrophys. 649, A146 (2021). Astron. Soc. 509, 3547 (2022). [25] J. T. A. de Jong, G. A. Verdoes Kleijn, K. H. Kuijken et al., [57] S. Everett, B. Yanny, N. Kuropatkin et al., Astrophys. J. Exp. Astron. 35, 25 (2013). Suppl. Ser. 258, 15 (2022). [26] D. S. Swetz, P. A. R. Ade, M. Amiri et al., Astrophys. J. [58] M. Gatti, G. Giannini, G. M. Bernstein et al., Mon. Not. R. Suppl. Ser. 194, 41 (2011). Astron. Soc. 510, 1223 (2022). [27] A. Krolewski, S. Ferraro, and M. White, J. Cosmol. [59] M. Jarvis, G. Bernstein, and B. Jain, Mon. Not. R. Astron. Astropart. Phys. 12 (2021) 028. Soc. 352, 338 (2004). [28] E. F. Schlafly, A. M. Meisner, and G. M. Green, Astrophys. [60] J. Muir, G. M. Bernstein, D. Huterer et al., Mon. Not. R. J. Suppl. Ser. 240, 30 (2019). Astron. Soc. 494, 4454 (2020). [29] E. Krause, X. Fang, S. Pandey et al., arXiv:2105.13548. [30] X. Fang, E. Krause, T. Eifler, and N. MacCrann, J. Cosmol. [61] C. Doux, E. Baxter, P. Lemos et al., Mon. Not. R. Astron. Soc. 503, 2688 (2021). Astropart. Phys. 05 (2020) 010. [62] A. Amon, D. Gruen, M. A. Troxel et al., Phys. Rev. D 105, [31] D. N. Limber, Astrophys. J. 117, 134 (1953). [32] A. Lewis, A. Challinor, and A. Lasenby, Astrophys. J. 538, 023514 (2022). [63] L. F. Secco, S. Samuroff, E. Krause et al., Phys. Rev. D 105, 473 (2000). 023515 (2022). [33] R. Takahashi, M. Sato, T. Nishimichi, A. Taruya, and M. [64] A. Porredon, M. Crocce, J. Elvin-Poole et al., arXiv: Oguri, Astrophys. J. 761, 152 (2012). 2105.13546. [34] S. Pandey, E. Krause, J. DeRose et al., Phys. Rev. D 106, [65] S. Aiola, E. Calabrese, L. Maurin et al., J. Cosmol. 043520 (2022). Astropart. Phys. 12 (2020) 047. [35] J. A. Blazek, N. MacCrann, M. A. Troxel, and X. Fang, [66] D. Dutcher, L. Balkenhol, P. A. R. Ade et al., Phys. Rev. D Phys. Rev. D 100, 103506 (2019). 104, 022003 (2021). [36] S. Bridle and L. King, New J. Phys. 9, 444 (2007). [37] J. Elvin-Poole, N. MacCrann et al., arXiv:2209.09782. [38] R. Cawthon et al. (DES Collaboration), Mon. Not. R. [67] M. Asgari, C. Heymans, H. Hildebrandt et al., Astron. Astrophys. 624, A134 (2019). [68] J. Prat, J. Blazek, C. Sánchez et al., Phys. Rev. D 105, Astron. Soc. 513, 5517 (2022), arXiv:2012.12826. 083528 (2022). [39] C.-H. Lin, J. Harnois-D´eraps, T. Eifler, T. Pospisil, R. Mandelbaum, A. B. Lee, and S. Singh, Mon. Not. R. Astron. Soc. 499, 2977 (2020). [40] J. Zuntz, M. Paterno, E. Jennings et al., Astron. Comput. 12, 45 (2015). [69] N. Jeffrey, M. Gatti, C. Chang et al., Mon. Not. R. Astron. Soc. 505, 4626 (2021). [70] A. Benoit-L´evy, T. D´echelette, K. Benabed, J.-F. Cardoso, D. Hanson, and S. Prunet, Astron. Astrophys. 555, A37 (2013). [41] W. J. Handley, M. P. Hobson, and A. N. Lasenby, Mon. Not. [71] D. Lenz, O. Dor´e, and G. Lagache, Astrophys. J. 883, 75 R. Astron. Soc. 453, 4384 (2015). (2019). 023530-25
10.1073_pnas.2302325120	RESEARCH ARTICLE \| BIOCHEMISTRY OPEN ACCESS Membrane protein isolation and structure determination in cell-derived membrane vesicles Xiao Taoa,b,1 , and Roderick MacKinnona,b,2 , Chen Zhaoa,b,1 Edited by Donald Engelman, Yale University, New Haven, CT; received February 9, 2023; accepted March 27, 2023 Integral membrane protein structure determination traditionally requires extraction from cell membranes using detergents or polymers. Here, we describe the isolation and structure determination of proteins in membrane vesicles derived directly from cells. Structures of the ion channel Slo1 from total cell membranes and from cell plasma membranes were determined at 3.8 Å and 2.7 Å resolution, respectively. The plasma membrane environment stabilizes Slo1, revealing an alteration of global helical packing, polar lipid, and cholesterol interactions that stabilize previously unre- solved regions of the channel and an additional ion binding site in the Ca2+ regulatory domain. The two methods presented enable structural analysis of both internal and plasma membrane proteins without disrupting weakly interacting proteins, lipids, and cofactors that are essential to biological function. cell membrane vesicle \| proteoliposome \| cryo-EM \| membrane protein \| ion channel Historically, the study of integral membrane protein structures has followed the develop- ment of methods to stabilize membrane proteins outside the membrane. For more than 40 years, detergents have been the mainstay approach (1–3). More recently, methods aimed to better simulate the membrane environment have been developed, including lipid cubic phase crystallization (4), bicelles (5), lipid nanodiscs (6–9), and styrene maleic acid lipid particles (10). Very recently, several structures were determined after isolating membrane proteins in detergent and then reconstituting them back into lipid vesicles (11–19). While reconsti- tution of membrane proteins into lipid vesicles is a valuable approach for structural studies, the procedure is very difficult in some cases, especially if the membrane protein requires stabilization in a low critical micelle concentration detergent. Moreover, lipids used in the reconstitution will not likely replicate the composition of a cell membrane before the membrane composition is known. Of equal importance, the detergent extraction step prior to reconstitution will most assuredly result in the loss of cofactors and associated proteins that may be weakly bound yet necessary for normal biological function. Ideally, we wish to see the structures of biological molecules in the cellular environment. To this aim, techniques such as cryoelectron tomography (20) combined with specialized sample preparation methods, e.g., focused ion beam milling (21), are being developed. To complement this strategy, we wondered whether it would be possible to isolate frag- ments of native cell membranes in the form of small vesicles for structural analysis at atomic resolution. To this end, we have developed two separate procedures, total membrane and plasma membrane, which, for exemplar purposes, we apply to the structure determi- nation of a potassium channel in cell membranes, with an immediate return in the depth of our understanding of this potassium channel. Results Significance Structural studies of membrane proteins historically have relied almost exclusively on first removing membrane proteins from the membrane with dispersive agents like detergents. But dispersive agents destabilize many membrane proteins, and in all membrane proteins, they remove weakly associated lipids, cofactors, and proteins essential to normal biological function. The procedures presented in this paper allow structure determination at atomic resolution without ever removing the proteins from their membrane environment. Thus, the structures of proteins that heretofore could not be extracted stably from the membrane can now be analyzed, and weakly associated lipid molecules, cofactors, and proteins normally lost in isolation are retained. These methods have immediately become the approach of choice for membrane protein structure determination in our laboratory. Enrichment of Slo1-Containing Vesicles Derived from the Total Membrane. For this study, we used a mammalian Slo1 (high-conductance Ca2+-activated K+ channel) channel derived from Homo sapiens (22), engineered at the DNA level to contain an extracellular ALFA tag (23) at its N terminus and an intracellular green fluorescent protein (GFP) tag at its C terminus (Fig. 1A). This channel, which is functional when expressed in cells (SI Appendix, Fig. S1), permitted the enrichment of vesicles containing Slo1 in both intracellular-side-out (inside-out) and extracellular-side-out (outside-out) orientations (Fig. 1B). To produce cell membrane–derived vesicles containing Slo1, the ALFA-Slo1- GFP construct was heterologously expressed in HEK293 GnTl- cells from suspension cultures using the BacMam method (24). To isolate Slo1-containing vesicles from the total cell membrane fraction, cells were first disrupted by sonication into small unilamellar vesicles (SUVs), likely exhibiting both inside-out and outside-out orientations (Fig. 1B). These vesicles originate mainly from Author contributions: X.T., C.Z., and R.M. designed research; X.T. and C.Z. performed research; X.T., C.Z., and R.M. analyzed data; and X.T., C.Z., and R.M. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1X.T. and C.Z. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2302325120/-/DCSupplemental. Published April 25, 2023. PNAS 2023 Vol. 120 No. 18 e2302325120 https://doi.org/10.1073/pnas.2302325120 1 of 9 Fig. 1. Preparation of Slo1-containing vesicles from the total cell membrane. (A) Construct design of the human Slo1 channel. The extracellular ALFA tag is colored in magenta, and the intracellular GFP tag is colored in green. (B) Purification procedure of Slo1-containing vesicles from the total membrane. The endoplasmic reticulum (ER) membrane is colored in light gray, and the plasma membrane (PM) is colored in dark gray. (C) SDS-PAGE of the total membrane vesicles enriched with Slo1. (D) Representative micrograph of Slo1-containing vesicles from the total membrane. (E) 2D class averages of Slo1 channel from the total cell membrane. intracellular membranes such as the endoplasmic reticulum (ER) and the Golgi apparatus because the surface plasma membrane contributes less than 10% of the total membrane in cells (Table 12-2 in ref. 25). We then removed nucleic acids using anion exchange chromatography and enriched Slo1-containing vesicles using a resin conjugated to a GFP nanobody (Fig. 1B). A Coomassie-stained, denaturing gel of the final sample showed that Slo1 was highly enriched, with only two other major contaminating proteins; 70 kDa heat shock protein (HSP70) and the protease used to elute bound vesicles from the resin (Fig. 1C). Note that this enrichment captures only vesicles containing Slo1 with an inside-out orienta- tion (Fig. 1 D and E). The size range of these vesicles was typically 20 to 100 nm diameter. After selecting particles using Topaz (26) with a model trained on manually picked particles, we sorted these particles by two-dimensional (2D) classification, which unambig- uously revealed Slo1 channels (Fig. 1E). Additionally, the mem- brane curvatures from the 2D class averages (Fig. 1E) were consistent with our purification protocol that enriched Slo1 chan- nels with an inside-out orientation (Fig. 1B). Structural Analysis of Slo1 from Total Membrane Vesicles. We determined the cryoelectron microscopy (cryo-EM) structure of the Slo1 channel in vesicles from the total cell membrane preparation at a resolution of 3.8 Å (Fig. 2A and SI Appendix, Figs. S2 and S3A and Table S1). The structure exhibited two-fold symmetry, contrasting the four-fold symmetric structures that we previously determined with detergent-solubilized Slo1 in both Ca2+-free and Ca2+-bound conformations (22, 27, 28). The two-fold symmetric organization is most evident within the cytoplasmic gating ring and is most easily appreciated when viewing the Slo1 channel along its central axis from the intracellular side. At its most intracellular extent, the intersubunit distance for two pairs of diagonally associated protomers differs by more than 10 Å (Fig. 2B). However, at the level of the inner membrane leaflet, the channel becomes more four- fold symmetric, and near the center of the membrane and selectivity filter, the channel follows four-fold symmetry (Fig. 2B). Because of the breakdown of four-fold symmetry on the intracellular side of the channel, if the two-fold symmetric channel is superposed onto its four-fold symmetric counterpart by aligning their selectivity filters, one pair of diagonally associated cytoplasmic domains is rotated and resides closer to the membrane than the other (Fig. 2C). A similar two-fold symmetric organization was also reported when detergent- purified Slo1 was reconstituted into lipid vesicles (18). We wondered whether the two-fold symmetry is a consequence of the presence of a membrane or the concentrations of Ca2+ used when isolating vesicles, which was intermediate between the low Ca2+ (containing ethylenediaminetetraacetic acid (EDTA) or eth- ylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid (EGTA)) or high Ca2+ (10 mM added Ca2+) concentrations used when purifying Slo1 in detergent (22, 27, 28). To address this issue, we purified Slo1 in detergent (22) using the same intermediate Ca2+ concentration used in the total membrane vesicle preparation, i.e., no EDTA or EGTA and no additional Ca2+. Solving the structure, we observed two-fold symmetry, indicating that the two-fold sym- metric structure is a function of the Ca2+ concentration, not the presence of a membrane (Fig. 2D). As a point separate from the main aim of the present paper, the symmetry result indicates that while Slo1 adopts predominantly four-fold symmetric structures in the absence or in the presence of high levels of Ca2+ (closed and open, respectively), intermediate levels of Ca2+ favor breakdown of this symmetry. This finding implies that gating transitions between closed and open conformations can occur through conformational changes within individual subunits that are, at least to some degree, uncoupled from each other. Such breaking of four-fold symmetry has also been observed in other members of the tetrameric cation channel family (29–35). 2 of 9 https://doi.org/10.1073/pnas.2302325120 pnas.org Fig. 2. Structural analysis of Slo1 from the total membrane vesicles. (A) Overall cryo-EM map of Slo1 from the total membrane vesicles. The protomers in Slo1 are colored in dark and light blue. The lipid bilayer (yellow) is contoured at a low threshold to show the curvature of the membrane. (B) Two-fold symmetry of Slo1 from the total membrane vesicles. Slabs of cryo-EM density parallel to the membrane plane are taken at the intracellular-most end of the Slo1 gating ring (slab 1) and the pore region (slab 2). (C) Structural comparison of the two protomers from neighboring subunits. The structures are superposed according to the selectivity filter and the pore helix. (D) Two-fold symmetric structure of Slo1 determined in detergent using the same EDTA-free buffer. The slab of cryo-EM density at the same region in B is shown. The density at the center of the pore entryway in slab 2 is likely due to detergent molecules. Enrichment of Slo1-Containing Vesicles Derived from the Plasma Membrane. Many integral membrane proteins require full maturation to reach the plasma membrane in a fully functional state. To isolate Slo1-containing plasma membrane vesicles from the total cell membrane, we took advantage of the discovery of chemically induced giant plasma membrane vesicles (GPMVs) (36). GPMVs were induced from live HEK293 GnTl− cells expressing ALFA-Slo1-GFP by N-ethylmaleimide (NEM) treatment in the presence of Ca2+ (Fig. 3A). These GPMVs were then sonicated to SUVs that were further enriched by ALFA-nanobody affinity resin (Fig. 3A). In contrast to the total membrane vesicle isolation method, this ALFA-nanobody enrichment method will isolate the Slo1-containing vesicles with an outside-out orientation. In fact, this is the orientation for the majority of the GPMV-derived SUVs, as GFP nanobody resin that interacts with the intracellular GFP tag (Fig. 1A) fails to enrich Slo1-containing vesicles. This outcome is consistent with the literature indicating that GPMVs tend to maintain the outside-out orientation of membrane proteins (37, 38). The Coomassie-stained denaturing gel of the final sample con- tained only one major band corresponding to Slo1 (Fig. 3B). Western blot analysis showed that the plasma membrane marker, Na+/K+ ATPase, was enriched in the GPMVs compared to the ER marker ERp57, in contrast to the high level of the ER marker in the whole cell (Fig. 3C). Cryo-EM micrographs of the plasma membrane vesicles have a clean background with the top or bot- tom views of Slo1 readily identifiable (Fig. 3D). As in the total membrane method, we selected particles using Topaz and per- formed a 2D classification, which again unambiguously revealed the presence of the Slo1 channel (Fig. 3E). The membrane curva- tures of the 2D classes indicate that the Slo1 channels almost exclusively adopt an outside-out orientation (Fig. 3E), consistent with our purification protocol. Structural Analysis of Slo1 from Plasma Membrane Vesicles. We determined the cryo-EM structure of the Slo1 channel in plasma membrane-derived vesicles at a resolution of 2.7 Å (Fig. 4A and SI Appendix, Figs. S3B and S4 and Table S1). The high quality of the cryo-EM map resolved the transmembrane domain at a higher resolution than previously published structures of Slo1 in detergent or in reconstituted lipid vesicles (18, 22, 27, 28). Regions of the channel not resolved in our past work (22, 27, 28) appear to have been stabilized by the plasma membrane environment. Some of these will be discussed in a separate section below. The two best-defined Slo1 structural classes from the plasma membrane vesicles are four-fold symmetric (C4), consistent with the higher Ca2+ concentration (Fig. 4 and SI Appendix, Fig. S4). These are essentially identical to each other and called the C4 structure. Two additional classes exhibit two-fold symmetry and define a C2 structure, which is distinct from the C2 structure observed in the intermediate Ca2+ concentrations (SI Appendix, Figs. S4 and S5). In the following, we focus our description on the higher resolution C4 structure in plasma membrane vesicles (Fig. 4). Both high-affinity Ca2+-binding sites per subunit in the gating ring are occupied due to the presence of 2 mM Ca2+ in the solution used to induce GPMV formation (Fig. 4C). The Mg2+ site is also occupied, likely because the normal intracellular Mg2+ concentration is in the millimolar range (Fig. 4C). Consistent with full occupation of the two Ca2+ and one Mg2+ ion-binding site per subunit, the gating ring engages the trans- membrane domain as in previously determined Ca2+- and Mg2+-bound structures of Slo1 in detergent (Fig. 4C) (22, 27, 28), and the pore’s gate is wide open, establishing an active state of the Slo1 channel in a cell membrane. This structure, together with the demonstration that ion channels are functional in GPMVs (39), supports the suitability of GPMVs for the struc- tural biology of membrane proteins. PNAS 2023 Vol. 120 No. 18 e2302325120 https://doi.org/10.1073/pnas.2302325120 3 of 9 Fig. 3. Preparation of Slo1-containing vesicles from the plasma membrane. (A) Purification procedure of Slo1-containing vesicles from the plasma membrane (PM). The endoplasmic reticulum (ER) membrane is colored in light gray, and the plasma membrane is colored in dark gray. (B) SDS-PAGE of the PM vesicles enriched with Slo1. (C) Western blot showing the relative abundance of the PM marker (Na+/K+ ATPase) and the ER marker (ERp57). Normalized PM marker levels for the two different samples are shown as bar graph, beneath (n = 3). (D) Representative micrograph of Slo1-containing vesicles from the plasma membrane. (E) 2D class averages of the Slo1 channel from the plasma membrane vesicles. Properties of the Slo1 Channel Revealed in Plasma Membrane Vesicles. The plasma membrane environment has provided a structure of Slo1 that exhibits a much higher degree of order throughout the transmembrane region, enabling clear definition of regions that were poorly resolved in the previous cryo-EM structures in detergent (22, 27, 28) (SI Appendix, Fig. S6). Notably, helices on the perimeter such as the S0 helix, which makes little contact with the rest of Slo1 and was barely resolved in structures with detergent (22, 27, 28), are highly ordered (Fig. 5 A, Right and SI Appendix, Fig. S6A). This higher degree of order may be a consequence of the chemical composition and physical properties of the cell plasma membrane. Quite unexpectedly, we also observe a global repositioning of the transmembrane helices in the plasma membrane compared to detergent micelles. Fig. 5A shows a superposition of the transmembrane helices: While a small rotation is evident, an unmistakable expansion of the helices away from the central axis of the channel caught our attention. This does not involve the selectivity filter, nor the pore helices, which are essentially identically positioned in the two structures, but other helices are expanded away from the center. We think that it is unlikely that the curvature of the vesicle membrane caused this because the expansion is similar in both the outer and inner membrane leaflets. We suspect that the expansion, together with the higher degree of order, results from interactions of the channel with its plasma membrane environment. A myriad of lipid molecules can be observed surrounding and within the transmembrane domain that were absent in the deter- gent structures (Fig. 5B and Movie S1). These lipids include both phospholipids and cholesterol. Compositional asymmetry with a preponderance of cholesterol in the outer membrane leaflet is evident. Indeed, a considerable fraction of the membrane-facing surface of Slo1 in contact with the outer leaflet is plastered with cholesterol molecules (Fig. 5C). This observation is consistent with a study showing up to a 12-fold higher concentration of choles- terol in the outer membrane leaflet of some mammalian cells (40). It thus appears that the membranes of GPMVs reflect cell plasma membranes in many respects. Numerous phospholipids make specific interactions with Slo1. We point out one phospholipid in particular because of its proximity to the S4-S5 linker, which couples the voltage sensor to the cytoplasmic Ca2+ sensor (Fig. 5D). Its headgroup interacts directly with the linker, and its alkyl chain runs into the membrane parallel to and contacting the S5 helix. Outside the membrane in the cytoplasmic Ca2+ sensor, we have discovered a previously undetected cation binding site (Fig. 5E). This binding site is adjacent to the known Ca2+ site 2 (Fig. 4C), with a center-to-center distance between ions about 5.4 Å. The Ca2+ is coordinated by oxygen atoms from two carboxylate side chains and three main chain carbonyl oxygen atoms. The new cation is coordi- nated by four main chain carbonyl oxygen atoms (Fig. 5E). The nearest carboxylate oxygen (coordinating the neighboring Ca2+) is about 3.5 Å away. The ion-oxygen distances for the carbonyl oxygen atoms as well as the coordination number are compatible with the new cation being either Ca2+ or Na+; however, given the net charge of the coordinating cage, the newly discovered ion is likely to be Na+ (41). Na+ is known to regulate the Slo2 channel (42–44), a close relative of Slo1. It will be interesting to examine the role of this new ion-binding site in the function of Slo1. Discussion In this study, we showed using two different methods how to isolate a membrane protein from cells without ever removing the protein from its lipid environment and thus maintaining a more native environment. The first approach should be suitable for isolating membrane proteins that normally reside in intracellular membranes, especially—but not limited to—the ER and Golgi. The second approach permits isolation of plasma membrane pro- teins derived from the plasma membrane and is thus important for the study of fully mature plasma membrane proteins and their interactions with surrounding lipids and other proteins. Native lipid vesicles from cells were used decades ago for electrophysiol- ogy studies (45–48) and more recently for native mass spectrom- etry of protein complexes (49). Here, we enriched these cell-derived vesicles containing a specific membrane protein and determined its high-resolution structures using cryo-EM. 4 of 9 https://doi.org/10.1073/pnas.2302325120 pnas.org A B C Mg2+ site 36.4 Å 36.4 Å Ca2+ Ca2+ site 1 Ca2+ Mg2+ Ca2+ site 1 Ca2+ site 2 Ca2+ site 2 Mg2+ site Fig. 4. Structural analysis of Slo1 from the plasma membrane vesicles. (A) Overall cryo-EM map of Slo1 from the total membrane vesicles. The protomers in Slo1 are colored in dark and light blue. The lipid bilayer (yellow) is contoured at a low threshold to show the curvature of the membrane. (B) Four-fold symmetry of Slo1 from the plasma membrane vesicles. Slabs of cryo-EM density parallel to the membrane plane are taken at the intracellular-most end of the Slo1 gating ring indicated in A. (C) The two Ca2+-binding sites and the Mg2+-binding sites are occupied. The Ca2+ ion is colored in orange, and the Mg2+ ion is colored in purple. The resolutions of the protein structures from cell-derived membranes are comparable to those we have solved using deter- gent micelles. In fact, the structure of Slo1 from GPMVs is better defined in regions on the perimeter of the protein such as the S0 transmembrane helix. We assume that these improved regions were somewhat disordered in earlier studies because of the dis- persive nature of the detergent environment. Perhaps owing to the stabilization provided by a plasma membrane-like environ- ment, we also discovered new features of Slo1, including some structural components not previously resolved, specific interac- tions with membrane phospholipids and cholesterol, and a new cation-binding site in the Ca2+ sensor. The advantages to cell-derived membrane vesicles are multiple. 1) They eliminate the need to remove a protein from the mem- brane, which is essential for proteins that are unstable outside the membrane. 2) They preserve a local environment for the protein that is closer to its natural environment in the cell. 3) Combined, the two isolation methods should permit the analysis of proteins resident in internal membranes, in plasma membranes, and plasma membrane proteins along their pathway of assembly and matura- tion. 4) It seems likely that associated molecular partners such as small molecules, lipids, and other proteins that normally dissociate under traditional purification methods will be discovered by ana- lyzing samples prepared by the described methods. In other words, the methods should enable the structural study of transient mem- brane protein complexes. We enumerate several potential limitations to structural analysis with cell-derived membrane vesicles. 1) Native proteins that may exist in membranes at low abundance will be more challenging. The Slo1 channel, even with heterologous expression, did not express to very high levels, and thus enrichment using an affinity tag was important. With gene modification, native proteins can be similarly tagged and enriched. Further experience will be needed before we know how well these methods apply to proteins present at low density. 2) Slo1, with its large cytoplasmic structure, was easily iden- tifiable and amenable to single-particle cryo-EM analysis. Smaller, more membrane-embedded proteins like certain G-protein-coupled receptors, for example, will be more challenging. In these more challenging cases, fiducials such as antibody fragments directed against the target protein ought to help (50, 51). 3) Because thin ice is a requirement for cryo-EM, the lipid vesicles must be made sufficiently small to fit into the thin ice layer. Of course, this same constraint also applies when studying membrane proteins reconsti- tuted into lipid vesicles. Small vesicles have high membrane curva- ture, and therefore, a sizeable bending force is applied to proteins embedded in them (15). In our experience with five different K+ channels reconstituted into lipid vesicles, the structures are very similar to those observed in detergent micelles (except when mem- brane voltage is applied across the membrane). In Piezo1, a mech- anosensitive channel that naturally changes its shape in response to bending forces, vesicle size modulates its shape (14). From these observations, we think that most membrane proteins are relatively shape-insensitive to the bending forces applied to them in small vesicles; however, it is likely that in some cases, the bending forces will alter a protein’s structure. This possibility should be kept in mind when using these methods. 4) The fourth "limitation" is more of an uncertainty, but of great importance. To what extent are GPMVs like plasma membranes? Although previous studies PNAS 2023 Vol. 120 No. 18 e2302325120 https://doi.org/10.1073/pnas.2302325120 5 of 9 Fig. 5. New structural features of Slo1 observed in plasma membrane vesicles. (A) Global conformational change of the Slo1 transmembrane domain. The models are superposed according to the transmembrane domain. The Slo1 model determined in plasma membrane (PM) vesicles is colored in blue, and the Slo1 model determined in detergent is colored in gray. A magnified view of the S0 helix and its cryo-EM density are shown on the Right. (B) Lipid (phospholipid and cholesterol) molecules observed in the cryo-EM map. The Slo1 model is colored in two shades of blue, and the cryo-EM densities for the lipid molecules are colored in light yellow. (C) Cholesterol molecules identified in the cryo-EM map. Cryo-EM densities for putative cholesterol molecules are colored in light yellow. (D) Phospholipid-binding site at the S4-S5 linker. The cryo-EM density for the phospholipid is colored in light yellow. (E) New cation (likely Na+) binding site identified in the Slo1 structure from plasma membrane vesicles. The Na+ ion is colored in light blue, and the Ca2+ ion is colored in orange. Right side is a close-up view of the magenta box on the left.The atomic model and cryo-EM density map are shown as wall-eyed stereoscopic images. reported some preliminary analyses (52, 53), there is much to be learned about the composition and organization of GPMVs pro- duced by the method we used. This is a subject of intense study in our laboratory. In summary, the methods presented here offer a less disruptive approach to the structure determination of membrane proteins. We anticipate that they will accelerate the study of membrane and membrane-associated proteins, especially by permitting the isola- tion of transient complexes. Materials and Methods Cell Culture. Spodoptera frugiperda Sf9 cells (ATCC CRL-1711) were cultured in Sf-900 II SFM medium supplemented with 100 U/mL penicillin and 100 U/mL streptomycin at 27 °C. HEK293S GnTl− cells (ATCC CRL-3022) were cultured in Freestyle 293 medium supplemented with 2% fetal bovine serum (FBS), 100 U/ mL penicillin, and 100 U/mL streptomycin at 37 °C. PtK2 cells (ATCC CCL-56) were cultured in ATCC-formulated Eagle's Minimum Essential Medium (ATCC 30-2003) supplemented with 10% FBS. Construct Design. The DNA sequence encoding ALFA peptide (23) was inserted into the N terminus of the human Slo1 construct used in the previous cryo-EM study (22). The resulting protein has the green fluorescent protein (GFP) followed by a 1D4 antibody recognition sequence (TETSQVAPA) on the C terminus, sepa- rated by an human rhinovirus (HRV)-3C protease cleavage site (SNSLEVLFQ/GP), as well as the ALFA tag at the N terminus. Protein Expression. Heterologous expression of Slo1 was carried out using the BacMam method as previously described (22, 24). In brief, bacmids were gener- ated by transforming the ALFA-Slo1-HRV3C-GFP-1D4 construct into Escherichia coli DH10Bac cells. Baculoviruses were then produced by transfecting Sf9 cells with the bacmid using Cellfectin II (Invitrogen). Baculoviruses, after two rounds of amplification, were used for cell trans- duction. Suspension cultures of HEK293S GnTI- cells were grown at 37 °C to a density of ~3 × 106 cells/mL and infected with 10% (v:v) of the baculovirus. After 20 h, 10 mM sodium butyrate was supplemented, and the temperature was shifted to 30 °C. Cells were harvested ~40 h after the temperature switch. For total membrane isolation, cell pellets were flash-frozen in liquid N2 and used later. For plasma membrane isolation, cell pellets were used immediately without freezing. 6 of 9 https://doi.org/10.1073/pnas.2302325120 pnas.org Total Membrane Vesicle Preparation. All steps were performed at 4 °C in the cold room. Frozen pellet from 4 L cells was thawed and resuspended in 160 mL lysis buffer [20 mM K-HEPES pH 7.4, 300 mM KCl, 0.5 mM MgCl2, and 5 mM dithiothreitol (DTT)] supplemented with 10 µg/mL leupeptin, 10 µg/mL pepstatin A, 1 mM benzamidine, 2 μg/mL aprotinin, 0.3 mg/mL AEBSF, ~100 µg/mL DNase, and ~100 µg/mL RNase A. The cell resuspension was then homogenized in a Dounce homogenizer for ~30 to 60 strokes. The homogenate was transferred to a metal beaker and sonicated in an ice bath with a probe sonicator (1/2” tip with the Branson102-C converter) at 60% power. The homogenate was disrupted by 30 s pulses four times, with a 30 s delay in between pulses. The sonicated homogenate was then spun at 12,000 g for 10 min. EDTA was added to the supernatant containing small vesicles, and then, the sample was spun again at 12,000 g for 10 min. We added EDTA to potentially reduce divalent ion–mediated vesicle and protein aggregation, but we have not tested whether incorporating EDTA is necessary. EDTA was not included in the previous steps because divalent ions are important for the activities of DNases and RNases. The resulting supernatant was filtered through 0.8 μm syringe filters and loaded onto a 20 mL Q Sepharose Fast Flow resin (Cytiva 17-0510-01) column preequilibrated with supplemented lysis buffer containing EDTA in a gravity column. The flow- through was collected and resin further washed with 3 column volumes (CVs) of EDTA-containing supplemented lysis buffer. The flow-through and 3 CVs of wash buffer were combined, and 1 mM fluorinated fos-choline-8 (FFC8, Anatrace, F300F) was added to minimize non- specific hydrophobic interactions with the GFP nanobody–conjugated affinity resin (54). The resulting sample was incubated with 4 mL GFP-nanobody resin (CNBr-activated Sepharose 4B resin from GE Healthcare) in batch at 4 °C for 1 h. The resin was loaded onto a gravity column and subsequently washed with 80 mL wash buffer A containing 20 mM K-HEPES, pH 7.4, 500 mM KCl, 5 mM DTT, 5 mM EDTA, and 1 mM FFC8 and 80 mL wash buffer B containing 20 mM K-HEPES, pH 7.4, 300 mM KCl, and 5 mM DTT. The bound vesicles were eluted by incubating with HRV-3C protease at a final concentration of ~0.014 mg/mL at 4 °C for 2 h. The elution and ~8 mL of additional wash with wash buffer B were combined and concentrated to OD280 ~2 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa). The sample purity was evaluated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) (Bio-Rad Mini-PROTEAN® TGX™ Precast Gels 4 to 15%). In Fig. 1C, Slo1 was identified by residual GFP fluorescence (after HRV 3C protease cleavage) measured in gel. HSP70 was identified by mass spectrometry of a similar preparation. HRV 3C protease was identified by its size on the SDS- PAGE of the purified protease by itself. Plasma Membrane Vesicle Preparation. One liter of live cells (~3 to 4 × 106/ mL) was harvested by centrifugation at 3,000 g for 10 min. The cell pellet was resus- pended in 200 mL GPMV buffer containing 10 mM K-HEPES, pH 7.4, 140 mM NaCl, 10 mM KCl, and 2 mM CaCl2. The cells were centrifuged again at 3,000 g for 10 min and then resuspended in 400 mL GPMV buffer supplemented with NEM (Thermo Scientific™ Pierce™, 23030) at 7.5 mM. The cell resuspension was transferred to four 250 mL baffled flasks to contain 100 mL in each flask. The flasks were then incubated in a 37 °C incubator shaking at 130 rpm for 1.5 to 2 h. The fast shaking helps to dis- lodge the GPMVs from the cells, resulting in smaller unilamellar vesicles (SUVs). At the end of the incubation, the flasks were shaken by hand for ~30 s. The suspension was then spun down at 3,000 g at 4 °C for 10 min to remove cells and large GPMVs. Next, the supernatant was supplemented with 10% glycerol to prevent aggre- gation and sonicated with a probe sonicator (Branson, 1/2” tip with Branson102-C converter) at 40% power for three 30 s pulses with ~1 min chilling on ice in between pulses. The vesicles were then pelleted by ultracentrifugation at 100,000 g at 4 °C for 40 min in a Ti70 rotor. The membrane vesicle pellet from every ~25 mL sample was triturated in ~1 mL GPMV buffer supplemented with 10% glycerol by gently squirting the buffer toward the pellet (with a 200 μL tip). The resuspended vesicles in the ultracentrifugation tubes were then sonicated in a water bath sonicator (Branson M1800) with ~10 s pulses at room temperature until the solution was opalescent (generally within 30 to 60 s). The vesicles were then centrifuged at 3,500 g at 4 °C for 10 min to remove the remaining aggregates. The supernatant (~20 mL) was incubated with 1 mL ALFA Selector CE resin (NanoTag) preequilibrated with GPMV buffer supplemented with 10% glycerol at 4 °C overnight. The following day, the ALFA Selector CE resin with bound vesicles was first batch-washed twice with ~20 mL glycerol-supplemented GPMV buffer and spun at 1,000 g at 4 °C for 1 min to collect resin, followed by a second wash with 15 mL GPMV buffer supplemented with 10% glycerol. The resin was then loaded onto a gravity column and washed further with another 5 mL GPMV buffer containing glycerol, followed by 15 mL GPMV buffer without glycerol. The vesicles were then eluted 3 times with 5 CVs, 5 CVs, and 3 CVs of GPMV buffer supplemented with 0.2 mM ALFA peptide (NanoTag) by incubating at RT for 30 min. The vesicles were kept on ice after being eluted. All three elutions were then combined and concentrated to an OD280 ~1.5 using an Amicon 2 mL concentrator (molecular weight cutoff of 100 kDa). 2+ . Purification of Purification of Slo1 in Detergent with Intermediate Ca the ALFA-tagged Slo1 protein was performed as previously described with modi- fications of the buffer compositions used (22). In brief, cells were gently disrupted by stirring in a hypotonic solution containing 10 mM Tris-HCl pH 8.0, 3 mM DTT, 1 mM EDTA supplemented with protease inhibitors including 0.1 μg/mL pepstatin A, 1 μg/mL leupeptin, 1 μg/mL aprotinin, 0.1 mg/mL soy trypsin inhibitor, 1 mM benzamidine, 0.1 mg/mL 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride (AEBSF), and 1 mM phenylmethylsulfonyl fluoride (PMSF). Cell lysate was then centrifuged for 30 min at 30,000 g, and then, the pellet was homogenized by a Dounce homogenizer in a buffer containing 20 mM Tris-HCl, pH 8.0, 320 mM KCl supplemented with protease inhibitors including 0.1 μg/mL pepstatin A, 1 μg/mL leupeptin, 1 μg/mL aprotinin, 0.1 mg/mL soy trypsin inhibitor, 1 mM benzamidine, 0.1 mg/mL AEBSF, and 0.2 mM PMSF. The lysate was extracted with 10 mM lauryl maltose neopentyl glycol and 2 mM cholesteryl hemisuccinate for an hour with stirring and then centrifuged for 40 min at 30,000 g. Supernatant was added to GFP nanobody–conjugated affinity resin (CNBr-activated Sepharose 4B resin from GE Healthcare) preequilibrated with wash buffer (20 mM K-HEPES, pH 7.4, 450 mM KCl, 0.06% digitonin (Sigma) and 0.1 mg/ml 1-palmitoyl-2-oleoyl-sn-glycero-3-phos- phoethanolamine (POPE):1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC):1- palmitoyl-2-oleoyl-sn-glycero-3-phosphate (POPA), 5:5:1 (w:w:w). The suspension was mixed by nutating for ~2 h. Beads were first washed with 10 CVs of wash buffer in a batch mode and then collected on a column by gravity, washed with another 20 CVs of wash buffer. The protein was then digested on resin with HRV 3C protease (~20:1 w:w ratio) overnight with gentle rocking. Flow-through was then collected, concentrated, and further purified on a Superose-6 size exclusion column (10/300 GL) in 20 mM HEPES-KOH, pH 7.4, 450 mM KCl, 5mM DTT, 0.06% digitonin, and 0.05 mg/mL POPE:POPC:POPA, 5:5:1 (w:w:w). All purification pro- cedures were carried out either on ice or at 4 °C. The peak fractions corresponding to the tetrameric Slo1 channel were concentrated to about 7.5 mg/mL using a 4 mL Amicon concentrator (molecular weight cutoff of 100 kDa) and used for preparation of cryo-EM sample grids. Excised Inside-Out Patch Recordings. The function of the ALFA-Slo1-HRV3C- GFP-1D4 construct used for all the cryo-EM sample preparation was verified by measuring the Ca2+ sensitivity of the channel in PtK2 cells in voltage-clamp inside-out patch configuration. Specifically, 1.2 μg of the plasmid was transfected into PtK2 cells at about 50 to 60% confluency using FuGENE HD transfection reagent following the manu- facturer’s instructions (Promega). Cells were transferred to 30 °C after transfection, and recordings were carried out 18 to 24 h posttransfection. Pipettes of borosilicate glass (Sutter Instruments; BF150-86-10) were pulled to ~2 to 3 MΩ resistance with a micropipette puller (Sutter Instruments; P-97) and polished with a microforge (Narishige; MF-83). All recordings were per- formed at room temperature in voltage-clamp excised inside-out patch config- uration with an Axopatch 200B amplifier (Molecular Devices), Digidata 1440A analog-to-digital converter interfaced with a computer, and pClamp10.5 software (Axon Instruments, Inc) for controlling membrane voltage and data acquisition. The recorded signal was filtered at 1 kHz and sampled at 10 kHz. The bath solution contained 20 mM Na-HEPES, 136 mM K-gluconate, 4 mM KCl, and 10 mM glucose, pH 7.4 (adjusted with NaOH), with an osmolarity of ~300 Osm/L. The bath solution supplemented with 2 mM MgCl2 was used as the pipette solution. Ionic current under a voltage-family protocol was measured with local perfusion of bath solution or bath solution with an additional 10 μM CaCl2 using a fast-pressurized microperfusion system (ALA Scientific; ALAVC3 × 8 PP). Note here that the Ca2+ concentration refers to the amount of Ca2+ added from a stock of CaCl2, not the free [Ca2+]. PNAS 2023 Vol. 120 No. 18 e2302325120 https://doi.org/10.1073/pnas.2302325120 7 of 9 Grid Preparation and Data Collection. Quantifoil R1.2/1.3 400 mesh holey carbon gold grids were glow-discharged for 22 s. Then, 3.5 μL concentrated mem- brane vesicles were applied to freshly glow-discharged grids and left for 3 min at 22 °C with a humidity of 100%. The grids were then blotted manually from the edge with a piece of filter paper. Another 3.5 μL sample was applied, and the grids were blotted using a Vitrobot Mark IV with a blot force of 0 and blot time of 3 s after 20 s of incubation. The grids were then flash-frozen in liquid ethane and stored in liquid nitrogen until data collection. The concentrated detergent sample was supplemented with 2.9 mM FFC8 immediately prior to grid preparation. Then, 3.5 μL of the mixture was applied to freshly glow-discharged Quantifoil R0.6/1 300 mesh holey carbon gold grids at 22 °C with a humidity of 100%. The grids were blotted using a Vitrobot Mark IV with a blot force of 22 and blot time of 4 s after 15 s of incubation, then flash-frozen in liquid ethane, and stored in liquid nitrogen until data collection. The dataset for Slo1-containing total membrane vesicles was collected on a 300 keV Titan Krios transmission electron microscope equipped with a Gatan K3 Summit camera, an energy filter (slit width 20 eV), and a CS corrector. A total of 21,348 movies were collected in a superresolution mode with a physical pixel size of 1.08 Å and a target defocus value of −1.0 to −2.0 μm. Every movie has a total exposure time of 2 s separated into 40 frames with a dose rate of 30 e−/ pixel/s, giving a total dose of 51.4 e−/Å2 (1.286 e−/Å2/frame). The dataset for Slo1-containing plasma membrane vesicles was collected on a 300 keV Titan Krios transmission electron microscope equipped with a cold-field emission gun and an energy filter (slit width 6 eV). A total of 24,876 movies were recorded by a Falcon 4i camera with a physical pixel size of 0.743 Å and a target defocus value of −1.0 to −2.0 μm. The movies have 1,505 internal frames and a total dose of 60 e−/Å2. The dataset for the Slo1 detergent sample was collected at a 300 keV Titan Krios transmission electron microscope equipped with a Gatan K3 Summit camera, an energy filter (slit width 20 eV), and a CS corrector. A total of 18,495 movies were collected in the superresolution mode with a physical pixel size of 1.08 Å and a target defocus value of −0.7 to −2.2 μm. Every movie has a total exposure time of 2 s separated into 40 frames with a dose rate of 30 e−/pixel/s, giving a total dose of 51.4 e−/Å2 (1.286 e−/Å2/frame). Cryo-EM Data Processing for Slo1 Total Membrane Vesicles. The raw movies were motion-corrected by MotionCor2 (55) in Relion V3.1 (56). The CTF param- eters were estimated by patch CTF estimation in cryoSPARC (V4.2.0) (57), and the initial 1,000 particles were manually picked from 164 Topaz (26) denoised micrographs. These particles were then used to train a Topaz model that was used for particle picking from all micrographs. Duplicate particles were removed, and multiple rounds of 2D classification and deep 2D classification (12) were carried out to select particles with clear Slo1 density. The initial model of Slo1 in total membrane vesicles was generated by ab initio reconstruction with C1 symmetry, and then, the particles were further sorted by heterogenous refinement. The sorted particles were subjected to another round of 2D classification in cryoSPARC (V4.2.0). The particles belonging to classes of side views (54%), top or bottom views (14%), and tilted views (32%) were used separately to train three different Topaz models, which were then used to pick particles from all micro- graphs. Combined particles were then subjected to 2D classification to remove "junk" particles. The selected particles were then used for ab initio reconstruction, heterogenous refinement, and homogeneous refinement in cryoSPARC (V4.2.0). It should be noted that refinement with C2 symmetry consistently yielded bet- ter-defined maps with higher resolution than refinement with C4 symmetry. After a best model was achieved, the particles were further sorted by a 3D clas- sification without alignment using Relion (V4.0-beta-1), with a user-defined mask covering the transmembrane domain. The resulting 85,135 particles were further refined using local refinement with C2 symmetry to generate the final cryo-EM map. Cryo-EM Data Processing for Slo1-Containing Plasma Membrane Vesicles. The raw movies in EER format were divided into 50 fractions and upsampled by a factor of 2 (8k rendering) and were motion-corrected by patch motion correction with further Fourier cropping by a factor of 2, after which the CTF parameters were estimated by patch CTF estimation in cryoSPARC V4.0.1. The initial 1,363 particles manually picked from 42 micrographs were sorted by a 2D classification. The particles belonging to classes with clear Slo1 features were used to train a Topaz model for particle picking from all micrographs. Duplicated particles were removed, and the resulting particles were subjected to multiple rounds of 2D classification to remove false-positive and low-abundance classes. A total of 597,254 particles were selected, and an initial model was gener- ated through ab initio reconstruction in cryoSPARC V4.0.1. Heterogenous refine- ment with the Slo1 initial model and 2 bad models from the ab initio reconstruction yielded a good class of Slo1 containing 433,531 particles. The selected particles were further sorted by several rounds of 2D classification to remove false-positive and suboptimal classes. Orientation and translational parameters for the resulting 330,063 particles were then refined using the homogenous refinement algorithm followed by local refinement in cryoSPARC V4.0.1 with C4 symmetry. The refined particle images were subjected to Relion’s (V4.0.0) 3D classification algorithm without a mask, skipping image alignment while imposing C2 symmetry. Four out of the 6 requested classes yielded 3D reconstructions representing Slo1 with high resolutions. Further refinement of each of these 4 good classes with the non- uniform refinement algorithm in cryoSPARC V4.0.1 while applying C1, C2, or C4 demonstrated that 2 classes obey C4 symmetry, while the other 2 appear to obey C2 symmetry. Note that the conformational differences of the 2 pairs of protomers in these plasma membrane vesicle C2 classes are much smaller than the differences observed in the C2 classes of the total membrane vesicle or detergent sample prepa- rations with an intermediate [Ca2+]. Local refinement of the 2 C4 classes and 2 C2 classes from plasma membrane vesicles yielded maps with resolutions of 2.7 Å, 2.9 Å, 3.0 Å, and 3.1 Å, respectively (SI Appendix, Fig. S4), as assessed by Fourier shell correlation using the 0.143 cutoff. 2+ ]. Cryo-EM Data Processing of Slo1 in Detergent under Intermediate [Ca Dose-fractionated superresolution images were 2 × 2 down-sampled by Fourier cropping for motion correction with MotionCorr2 (5 × 5 patches). The parameters of the contrast transfer function were estimated by Ctffind4 (58). Following motion correction, ~1,500 particles from a subset of the images were interactively selected using Relion to generate templates representing different views for automated particle selection with Gautomatch (59). The autopicked particles were then sub- jected to 2 rounds of 2D classification in cryoSPARC V4.0.1 to remove junk particles or particles belonging to low-abundance classes. An initial model was generated from the selected particles using ab initio reconstruction in cryoSPARC V4.0.1. Orientation and translational parameters of ~1,032 k selected particle images were refined with the homogeneous refinement algorithm followed by local refinement in cryoSPARC V4.0.1 imposing C1 symmetry. The refined particle images were subjected to Relion’s (V4.0.0) 3D classification algorithm without image alignment or a mask, requesting six classes. Orientation and translational parameters for the ~123k particles in the best class were refined using homo- geneous refinement, followed by nonuniform refinement and local refinement (imposing C2 symmetry, which consistently yielded better-defined maps with higher resolution than refinement with C4 symmetry, similar to the total mem- brane vesicle data) in cryoSPARC V4.0.1, resulting in a map at a resolution of 3.3 Å before postprocessing. Data, Materials, and Software Availability. Cryo-EM density maps and atomic coordinates of the hSlo1 channel in total membrane vesicles (Ca2+-free and EDTA-free), in plasma membrane vesicles, and in digitonin (Ca2+-free and EDTA-free) have been deposited in the Electron Microscopy Data Bank under accession codes EMD-40038 (60), and EMD-40044 (61), EMD-40045 (62) and in the Protein Data Bank under accession codes 8GH9 (63), 8GHF (64), and 8GHG (65), respectively. ACKNOWLEDGMENTS. We thank Yi Chun Hsiung for assistance with cell culture, George Vaisey for advice, and all MacKinnon Lab and Chen Lab members for helpful discussions. We thank Mark Ebrahim, Johanna Sotiris, and Honkit Ng at the Evelyn Gruss Lipper Cryo-EM Resource Center of Rockefeller University for assistance with cryo-EM data collection of the total membrane vesicle sam- ple and the detergent sample. We thank Rui Yan, Zhiheng Yu, and the team at the HHMI Janelia Cryo-EM Facility for assistance with cryo-EM data collection of the plasma membrane vesicles. R.M. is an Investigator in the Howard Hughes Medical Institute. Author affiliations: aLaboratory of Molecular Neurobiology and Biophysics, The Rockefeller University, New York, NY 10065; and bHHMI, The Rockefeller University, New York, NY 10065 8 of 9 https://doi.org/10.1073/pnas.2302325120 pnas.org 1. A. Helenius, K. Simons, Solubilization of membranes by detergents. Biochim. Biophys. Acta 415, 29–79 (1975). 33. L. Zubcevic, S. Le, H. Yang, S. Y. Lee, Conformational plasticity in the selectivity filter of the TRPV2 ion channel. Nat. Struct. Mol. Biol. 25, 405–415 (2018). 2. H. Michel, D. Oesterhelt, Three-dimensional crystals of membrane proteins: Bacteriorhodopsin. Proc. 34. L. Zubcevic, A. L. Hsu, M. J. Borgnia, S. Y. Lee, Symmetry transitions during gating of the TRPV2 ion 3. 4. 5. 6. 7. Natl. Acad. Sci. U.S.A. 77, 1283–1285 (1980). J. Deisenhofer, O. Epp, K. Miki, R. Huber, H. Michel, X-ray structure analysis of a membrane protein complex. Electron density map at 3 A resolution and a model of the chromophores of the photosynthetic reaction center from Rhodopseudomonas viridis. J. Mol. Biol. 180, 385–398 (1984). E. M. Landau, J. P. Rosenbusch, Lipidic cubic phases: A novel concept for the crystallization of membrane proteins. Proc. Natl. Acad. Sci. U.S.A. 93, 14532–14535 (1996). C. R. Sanders II, J. P. Schwonek, Characterization of magnetically orientable bilayers in mixtures of dihexanoylphosphatidylcholine and dimyristoylphosphatidylcholine by solid-state NMR. Biochemistry 31, 8898–8905 (1992). T. H. Bayburt, Y. V. Grinkova, S. G. Sligar, Self-assembly of discoidal phospholipid bilayer nanoparticles with membrane scaffold proteins. Nano Lett. 2, 853–856 (2002). I. G. Denisov, Y. V. Grinkova, A. A. Lazarides, S. G. Sligar, Directed self-assembly of monodisperse phospholipid bilayer Nanodiscs with controlled size. J. Am. Chem. Soc. 126, 3477–3487 (2004). 8. M. L. Nasr et al., Covalently circularized nanodiscs for studying membrane proteins and viral entry. 9. Nat. Methods 14, 49–52 (2017). S. Zhang et al., One-step construction of circularized nanodiscs using SpyCatcher-SpyTag. Nat. Commun. 12, 5451 (2021). channel in lipid membranes. Elife 8, e45779 (2019). 35. W. Ye et al., Activation and closed-state inactivation mechanisms of the human voltage-gated K(V)4 channel complexes. Mol. Cell 82, 2427–2442.e4 (2022). 36. R. E. Scott, Plasma membrane vesiculation: A new technique for isolation of plasma membranes. Science 194, 743–745 (1976). 37. T. Baumgart et al., Large-scale fluid/fluid phase separation of proteins and lipids in giant plasma membrane vesicles. Proc. Natl. Acad. Sci. U.S.A. 104, 3165–3170 (2007). 38. S. W. Lyu, J. F. Wang, L. Chao, Constructing supported cell membranes with controllable orientation. Sci. Rep. 9, 2747 (2019). 39. M. L. Jacobs et al., Probing the force-from-lipid mechanism with synthetic polymers. bioRxiv [Preprint] (2022), https://doi.org/10.1101/2022.05.20.492859 (Accessed 21 May 2022). 40. S. L. Liu et al., Orthogonal lipid sensors identify transbilayer asymmetry of plasma membrane cholesterol. Nat. Chem. Biol. 13, 268–274 (2017). 41. O. C. Gagne, F. C. Hawthorne, Bond-length distributions for ions bonded to oxygen: Alkali and alkaline-earth metals. Acta Crystallogr. B Struct. Sci. Cryst. Eng. Mater. 72, 602–625 (2016). 42. C. R. Bader, L. Bernheim, D. Bertrand, Sodium-activated potassium current in cultured avian neurones. Nature 317, 540–542 (1985). 43. R. K. Hite et al., Cryo-electron microscopy structure of the Slo2.2 Na(+)-activated K(+) channel. 10. T. J. Knowles et al., Membrane proteins solubilized intact in lipid containing nanoparticles bounded Nature 527, 198–203 (2015). by styrene maleic acid copolymer. J. Am. Chem. Soc. 131, 7484–7485 (2009). 44. R. K. Hite, R. MacKinnon, Structural titration of Slo2.2, a Na(+)-dependent K(+) channel. Cell 168, 11. L. Wang, F. J. Sigworth, Structure of the BK potassium channel in a lipid membrane from electron 390–399.e11 (2017). cryomicroscopy. Nature 461, 292–295 (2009). 12. X. Yao, X. Fan, N. Yan, Cryo-EM analysis of a membrane protein embedded in the liposome. Proc. Natl. Acad. Sci. U.S.A. 117, 18497–18503 (2020). 13. M. E. Falzone, R. MacKinnon, Gβγ activates PIP 2 hydrolysis by recruiting and orienting PLCβ on the membrane surface. bioRxiv [Preprint] (2022). https://doi.org/10.1101/2022.12.20.521270 (Accessed 20 December 2022). 14. C. A. Haselwandter, Y. R. Guo, Z. Fu, R. MacKinnon, Quantitative prediction and measurement of Piezo’s membrane footprint. Proc. Natl. Acad. Sci. U.S.A. 119, e2208027119 (2022). 45. E. Moczydlowski, R. Latorre, Gating kinetics of Ca2+-activated K+ channels from rat muscle incorporated into planar lipid bilayers. Evidence for two voltage-dependent Ca2+ binding reactions. J. Gen. Physiol. 82, 511–542 (1983). 46. C. Miller, J. E. Bell, A. M. Garcia, "The potassium channel of sarcoplasmic reticulum" in Current Topics in Membranes and Transport, F. Bronner, Ed. (Academic Press, 1984), vol. 21, pp. 99–132. 47. W. N. Green, L. B. Weiss, O. S. Andersen, Batrachotoxin-modified sodium channels in planar lipid bilayers. Characterization of saxitoxin- and tetrodotoxin-induced channel closures. J. Gen. Physiol. 89, 873–903 (1987). 15. C. A. Haselwandter, Y. R. Guo, Z. Fu, R. MacKinnon, Elastic properties and shape of the Piezo 48. R. MacKinnon, C. Miller, Functional modification of a Ca2+-activated K+ channel by dome underlying its mechanosensory function. Proc. Natl. Acad. Sci. U.S.A. 119, e2208034119 (2022). trimethyloxonium. Biochemistry 28, 8087–8092 (1989). 49. D. S. Chorev et al., The use of sonicated lipid vesicles for mass spectrometry of membrane protein 16. V. S. Mandala, R. MacKinnon, Voltage-sensor movements in the Eag Kv channel under an applied complexes. Nat. Protoc. 15, 1690–1706 (2020). electric field. Proc. Natl. Acad. Sci. U.S.A. 119, e2214151119 (2022). 50. J. S. Bloch et al., Development of a universal nanobody-binding Fab module for fiducial-assisted 17. Z. Melville, K. Kim, O. B. Clarke, A. R. Marks, High-resolution structure of the membrane-embedded cryo-EM studies of membrane proteins. Proc. Natl. Acad. Sci. U.S.A. 118 (2021). skeletal muscle ryanodine receptor. Structure 30, 172–180.e3 (2022). 51. X. Wu, T. A. Rapoport, Cryo-EM structure determination of small proteins by nanobody-binding 18. L. Tonggu, L. Wang, Structure of the Human BK Ion Channel in Lipid Environment. Membranes scaffolds (Legobodies). Proc. Natl. Acad. Sci. U.S.A. 118 (2021). (Basel) 12, 758 (2022). 52. H. Keller, M. Lorizate, P. Schwille, PI(4,5)P2 degradation promotes the formation of cytoskeleton-free 19. X. Yang et al., Structure deformation and curvature sensing of PIEZO1 in lipid membranes. Nature model membrane systems. Chemphyschem 10, 2805–2812 (2009). 604, 377–383 (2022). 53. K. R. Levental et al., Polyunsaturated lipids regulate membrane domain stability by tuning 20. J. Mahamid et al., Visualizing the molecular sociology at the HeLa cell nuclear periphery. Science membrane order. Biophys. J. 110, 1800–1810 (2016). 351, 969–972 (2016). 54. P. C. Fridy et al., A robust pipeline for rapid production of versatile nanobody repertoires. Nat. 21. M. Marko, C. Hsieh, R. Schalek, J. Frank, C. Mannella, Focused-ion-beam thinning of frozen-hydrated Methods 11, 1253–1260 (2014). biological specimens for cryo-electron microscopy. Nat. Methods 4, 215–217 (2007). 55. S. Q. Zheng et al., MotionCor2: Anisotropic correction of beam-induced motion for improved cryo- 22. X. Tao, R. MacKinnon, Molecular structures of the human Slo1 K(+) channel in complex with beta4. electron microscopy. Nat. Methods 14, 331–332 (2017). Elife 8, e51409 (2019). 56. J. Zivanov et al., New tools for automated high-resolution cryo-EM structure determination in 23. H. Gotzke et al., The ALFA-tag is a highly versatile tool for nanobody-based bioscience applications. RELION-3. Elife 7, e42166 (2018). Nat. Commun. 10, 4403 (2019). 57. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: Algorithms for rapid unsupervised 24. A. Goehring et al., Screening and large-scale expression of membrane proteins in mammalian cells cryo-EM structure determination. Nat. Methods 14, 290–296 (2017). for structural studies. Nat. Protoc. 9, 2574–2585 (2014). 58. A. Rohou, N. Grigorieff, CTFFIND4: Fast and accurate defocus estimation from electron micrographs. 25. B. J. Albert et al., Molecular Biology of the Cell (Garland Science, New York, ed. 4, 2002). 26. T. Bepler et al., Positive-unlabeled convolutional neural networks for particle picking in cryo-electron J. Struct. Biol. 192, 216–221 (2015). 59. K. Zhang, Gautomatch v0.56 [internet]. 2020. Available from: https://www.mrc-lmb.cam.ac.uk/ micrographs. Nat. Methods 16, 1153–1160 (2019). kzhang/ (cited 3 April 2020). 27. R. K. Hite, X. Tao, R. MacKinnon, Structural basis for gating the high-conductance Ca(2+)-activated 60. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in total membrane vesicle. EMDB. https:// K(+) channel. Nature 541, 52–57 (2017). www.ebi.ac.uk/emdb/EMD-40038. Deposited 9 March 2023. 28. X. Tao, R. K. Hite, R. MacKinnon, Cryo-EM structure of the open high-conductance Ca(2+)-activated 61. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in plasma membrane vesicles. EMDB. K(+) channel. Nature 541, 46–51 (2017). https://www.ebi.ac.uk/emdb/EMD-40044. Deposited 9 March 2023. 29. H. Shen et al., Structure of a eukaryotic voltage-gated sodium channel at near-atomic resolution. 62. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in digitonin, Ca2+-free and EDTA-free. Science 355, eaal4326 (2017). EMDB. https://www.ebi.ac.uk/emdb/EMD-40045. Deposited 9 March 2023. 30. X. Pan et al., Structure of the human voltage-gated sodium channel Na(v)1.4 in complex with beta1. 63. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in total membrane vesicles. PDB. https:// Science 362, eaau2486 (2018). www.rcsb.org/structure/8GH9. Deposited 9 March 2023. 31. J. She et al., Structural insights into the voltage and phospholipid activation of the mammalian 64. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in plasma membrane vesicles. PDB. TPC1 channel. Nature 556, 130–134 (2018). https://www.rcsb.org/structure/8GHF. Deposited 9 March 2023. 32. H. Shen et al., Structural basis for the modulation of voltage-gated sodium channels by animal 65. X. Tao, C. Zhao, R. MacKinnon, cryo-EM structure of hSlo1 in digitonin, Ca2+-free and EDTA-free. toxins. Science 362, eaau2596 (2018). PDB. https://www.rcsb.org/structure/8GHG. Deposited 9 March 2023. PNAS 2023 Vol. 120 No. 18 e2302325120 https://doi.org/10.1073/pnas.2302325120 9 of 9
10.1089_crispr.2023.0015	The CRISPR Journal Volume 6, Number 3, 2023 Mary Ann Liebert, Inc. DOI: 10.1089/crispr.2023.0015 RESEARCH ARTICLE The Transposon-Encoded Protein TnpB Processes Its Own mRNA into xRNA for Guided Nuclease Activity Suchita P. Nety,1–5 Han Altae-Tran,1–5 Soumya Kannan,1–5 F. Esra Demircioglu,1–5 Guilhem Faure,1–5 Seiichi Hirano,1–5 Kepler Mears,1–5 Yugang Zhang,1–5 Rhiannon K. Macrae,1–5 and Feng Zhang1–5,* Abstract TnpB is a member of the Obligate Mobile Element Guided Activity (OMEGA) RNA-guided nuclease family, is harbored in transposons, and likely functions to maintain the transposon in genomes. Previously, it was shown that TnpB cleaves double- and single-stranded DNA substrates in an RNA-guided manner, but the bio- genesis of the TnpB ribonucleoprotein (RNP) complex is unknown. Using in vitro puriﬁed apo TnpB, we demon- strate the ability of TnpB to generate guide omegaRNA (xRNA) from its own mRNA through 5¢ processing. We also uncover a potential cis-regulatory mechanism whereby a region of the TnpB mRNA inhibits DNA cleavage by the RNP complex. We further expand the characterization of TnpB by examining xRNA processing and RNA- guided nuclease activity in 59 orthologs spanning the natural diversity of the TnpB family. This work reveals a new functionality, xRNA biogenesis, of TnpB, and characterizes additional members of this biotechnologically useful family of programmable enzymes. Introduction Proteins associated with the IS200/605 family of trans- posable elements have been found to be RNA-guided DNA-targeting enzymes.1,2 These proteins, namely TnpB, IscB, and IsrB, associate with a noncoding RNA, termed omegaRNA or xRNA. The xRNA is often encoded in the same locus as the protein and enables TnpB to per- form guided cleavage of DNA substrates. These IS200/ 605 transposon loci are ﬂanked by transposon ends.3 The 3¢ transposon end (right end or RE) demarcates the 3¢ end of the xRNA scaffold, which is followed by a guide sequence that is encoded outside of the transpo- son.1,2 IS200/605 family transposons are most often mo- bilized by TnpA, a transposase with a single catalytic tyrosine (Y1) that preferentially excises from and inserts into single-stranded DNA (ssDNA).4 TnpB is not known to be a transposase, but rather is thought to function akin to a homing endonuclease5 that prevents loss of the trans- poson upon excision.1,2,6 TnpB comprises a diverse family of proteins which have become associated with CRISPR arrays multiple times throughout evolution to give rise to various CRISPR-Cas12 subtypes.7–10 Biochemical experiments have shown that TnpB cleavage of double-stranded DNA (dsDNA) is dependent on both the complementarity of the guide region of the xRNA to the target substrate and the presence of a target-adjacent motif (TAM) 5¢ to the target sequence.1,2 This preference for a 5¢ TAM is analogous to the 5¢ protospacer-adjacent motif (PAM) preference of Cas12.11 In addition to cleaving dsDNA, TnpB exhibits target-speciﬁc TAM-independent cleav- age of ssDNA, as well as collateral (off-target) cleavage of ssDNA substrates in the presence of target-containing dsDNA or ssDNA.1,2 TnpB is a promising candidate for development as a human genome editing tool due to its relatively com- pact size (350–550 amino acids) compared to most Cas nucleases (e.g., SpCas9 1368 amino acids (aa),12 SaCas9 1Howard Hughes Medical Institute, Cambridge, Massachusetts, USA; 2Broad Institute of MIT and Harvard, Cambridge, Massachusetts, USA; 3McGovern Institute for Brain Research at MIT, Cambridge, Massachusetts, USA; 4Department of Brain and Cognitive Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA; and 5Department of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. Address correspondence to: Feng Zhang, Broad Institute and MIT, Cambridge, MA 02139, USA, E-mail: [email protected] ª Suchita P. Nety, et al. 2023; Published by Mary Ann Liebert, Inc. This Open Access article is distributed under the terms of the Creative Commons License [CC-BY] (http:// creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 232 TNPB PROCESSES ITS OWN GUIDE RNA 233 1053 aa,13 AsCas12a 1307 aa).11 The compact domain structure of TnpB is similar to Cas12f, consisting of N-terminal recognition and wedge (WED) domains con- nected to the RuvC catalytic domain by a linker.14–16 This compact size enables packaging into adeno- associated viruses (AAVs), a clinically validated delivery modality.17 Although AAVs have a number of advan- tages for gene therapy, they have a limited packaging capacity of 4.7 kb, which is too low for most Cas enzymes and the necessary regulatory elements and guide RNA. However, small Cas enzymes such as UnCas12f1 (529 aa)18,19 and AsCas12f1 (422 aa)20 can be engineered as genome or base editors and packaged into a single AAV, highlighting the promise of compact programma- ble nucleases for systemically-administered gene editing therapies. To further explore the biotechnological potential of TnpB, we sought to further characterize the biochemical activity of diverse TnpB enzymes, focusing on the forma- tion and identity of the xRNA. Given the broad diversity of the TnpB family, we sought to catalog and sample this diversity to explore the conserved features of TnpB func- tion. Collectively, these results provide insight into the mechanism of TnpB and serve as a starting point for future endeavors to optimize TnpB for genome engineering. Materials and Methods Protein puriﬁcation Wild-type AmaTnpB (Addgene no. 176587) and RuvC- mutant E271A AmaTnpB were puriﬁed as described previously.1 TnpB with an N-terminal His14-MBP-TEV protease cleavage site was expressed in a pET45b(+) plasmid backbone in Rosetta 2(DE3) cells (Novagen). Cells were grown at 37(cid:2)C in terriﬁc broth (TB) medium supplemented with 100 lg/mL ampicillin and 30 lg/mL chloramphenicol overnight. One liter TB with 100 lg/mL ampicillin was inoculated with a 3 mL overnight culture, grown to an OD600 of 0.6–0.8, and subsequently induced with 0.2 mM IPTG and grown at 18(cid:2)C for 24 h. Cells were harvested by centrifugation and resuspended in Buffer A (50 mM Tris pH 8, 1 M NaCl, 5% glycerol, 40 mM imidazole, and 5 mM b-mercaptoethanol) supple- mented with benzonase (Sigma) and protease inhibitors (phenylmethylsulfonyl ﬂuoride and Roche cOmplete, ethylenediaminetetraacetic acid-free) and then lysed by two passes with a high-pressure homogenizer (LM20 Microﬂuidizer, Microﬂuidics). After clearing the lysate by centrifugation, the soluble fraction was bound to Ni-NTA agarose (Qiagen). The beads were ﬁrst washed with Buffer B (50 mM Tris pH 8, 2 M NaCl, 5% glycerol, 40 mM imidazole, and 5 mM b-mercaptoethanol) and subsequently with Buffer A and Buffer C (50 mM Tris pH 8, 500 mM NaCl, 5% glyc- erol, 40 mM imidazole, and 5 mM b-mercaptoethanol). TnpB protein was then eluted in Buffer D (50 mM Tris pH 8, 500 mM NaCl, 5% glycerol, 300 mM imidazole, and 5 mM b-mercaptoethanol), incubated with 480 lg TEV protease, and dialyzed overnight against Buffer E (20 mM Tris pH 7.5, 500 mM NaCl, 5% glycerol, 0.5 mM TCEP). The protein was puriﬁed using a Resource S column (Cytiva) against a 0.2–2 M NaCl gra- dient. Peak fractions containing TnpB protein were pooled and dialyzed overnight against Buffer E. Protein was concentrated to 5 lM, aliquoted, snap-frozen, and stored at (cid:2)80(cid:2)C. In vitro RNA and DNA cleavage reactions RNA substrates were prepared from PCR-generated DNA templates through in vitro transcription reactions with the NEB T7 HiScribe Kit and puriﬁed with the Zymo Clean & Concentrator-25 Kit. The 1221-nt target DNA substrate was generated by PCR and contains the cognate target sequence with a 5¢ TAM of TCAC, which when cleaved generates 531-nt and 690-nt frag- ments. Sequences for RNA and DNA substrates are pro- vided in Supplementary File S1. In vitro reactions were prepared with ﬁnal concentrations of 1 lM TnpB, 1 lM RNA substrate(s), 10 nM DNA substrate, and 1 U/lL mu- rine RNase inhibitor (NEB) in 20 mM HEPES and 5 mM MgCl2. After incubation at 55(cid:2)C for 30 min, the reactions were either (1) treated with RNase A (Qiagen), Protei- nase K (NEB) and puriﬁed with PCR puriﬁcation col- umns (Qiagen) or (2) treated with DNase I (NEB) and Proteinase K (NEB). RNase-treated samples were visual- ized on 2% E-Gel EX agarose gels (Invitrogen), and DNase-treated samples were denatured and visualized on 6% TBE-Urea polyacrylamide gels (Thermo Fisher Scientiﬁc). The DNase-treated samples were sequenced using the NEBNext Multiplex Small RNA Library Prep Set for Illumina (NEB). Ortholog curation and selection TnpBs for experimental characterization (Supplementary File S2) were sampled from diverse TnpBs by selecting different leaves from the phylogenetic tree described in a comprehensive survey of TnpB/Cas12 diversity.8 Selected sequences were prioritized on the basis of their available contig lengths to ensure that short range (£3 kb) genomic associations of the transposon locus were captured appropriately, and whenever possible, se- quences from complete genomes in the NCBI collection were prioritized due to their generally higher accuracy of assembled contigs relative to those from Joint Genome and Whole Genome Shotgun assembled Institute 234 NETY ET AL. metagenomes. Locus RE boundaries (i.e., the boundary between the xRNA scaffold and guide sequence) were determined by aligning the 3¢ end of the locus with related locus sequences with MAFFT.21 From the align- ments, the guide-scaffold boundary was identiﬁed as the downstream-most position in which a sharp drop in sequence conservation occurred. For the phylogenetic analysis presented in Figure 3, representative TnpBs were selected from the compre- hensive TnpB/Cas12 study8 that cover the main clades (Typical TnpBs, Derived TnpBs, and clades containing catalytic rearrangements of the RuvC-II (RII-r3 and 5) or RuvC-III (RIIIr-4) domain), major branches of TnpB, and some selected Cas12s (Supplementary File S3). These selected TnpB/Cas12 sequences, along with the experimentally studied TnpB sequences, were aligned using MAFFT-einsi,21 then trimmed using TrimAl22 with a gap threshold of 0.5. The LG+G4 substitution model for phylogenetic inference was selected using Model- Finder by optimizing the corrected Akaike Information Score.23 A phylogenetic tree was then inferred using IQTree224 using the following parameters: -nstop 500 -bnni—ninit 5000—ntop 100—nbest 20, and 2000 ultra- fast bootstraps25 and ﬁnally visualized with iTOL.26 Following this, all genes in each locus were searched using HMMER327 against hidden Markov model proﬁles for Cas1, Cas2, Cas4,1 as well as Y1 TnpA, and IS607- like Serine Recombinases using PF01797 and PF00239, respectively, from PFAM.28 All CRISPRs were predicted using CRT.29 Association to CRISPR arrays, Y1 TnpA, and IS607-like Serine Recombinases was determined by the presence of the respective feature (as identiﬁed above) within 1 kb of the protein coding sequence in the genomic locus. The multiple sequence alignment was used to determine for each sequence if each of the three residues in the RuvC catalytic triad was intact by comparing if the residue aligned at the catalytic triad po- sition matched the expected residue (D for RuvC-I, E for RuvC-II, and D for RuvC-III). Noncanonical RuvC resi- dues are indicated by the abbreviations RII-r (RuvC-II rearrangement) and RIII-r (RuvC-III rearrangement). IVTT 5¢ RACE DNA templates were synthesized by Twist Biosciences or ampliﬁed by PCR from bacterial genomic DNA. All templates included a 5¢ UTR containing a T7 promo- ter and ribosomal binding site for expression in the in vitro transcription and translation PURExpress (IVTT) kit (NEB). Reactions were prepared with 110 ng of DNA template and 1 U/lL murine RNase in- hibitor (NEB) and incubated at 37(cid:2)C for 2 h. RNA was extracted with TRIzol reagent (Thermo Fisher Scientiﬁc) and puriﬁed with Zymo Clean & Concentrator-96. 5¢ rapid ampliﬁcation of cDNA ends (RACE) was carried out by annealing 1 lM of primer annealing to the 3¢ end of the xRNA (Supplementary File S2) to 1000 ng of puriﬁed RNA at 65(cid:2)C for 15 min and 25(cid:2)C for 15 min. Next, 375 nM 5¢ SR adaptor (NEB) was denatured and ligated to the RNA using T4 RNA Ligase 1 (NEB) for 1 h at 25(cid:2)C. Reverse transcription was carried out using Super- Script IV RT (Invitrogen). cDNA was extracted by PCR puriﬁcation (Qiagen) and ampliﬁed with 12 cycles of PCR using NEBNext High Fidelity 2X PCR Master Mix (NEB) with one primer annealing to the 5¢ adaptor and one primer annealing to the 3¢ end of the xRNA, fol- lowed by a second round of PCR with 18 cycles with primers adding i7 and i5 Illumina adaptors and barcodes. Ampliﬁed libraries were gel extracted, quantiﬁed using Qubit Fluorometric Quantiﬁcation (Thermo Fisher Scien- tiﬁc), and subjected to single-end sequencing on an Illu- mina MiSeq with the following parameters: read 1-300 cycles, index 1-8 cycles, index 2-8 cycles. Reads were trimmed of adaptors and aligned to template sequences using Geneious Prime. xRNA scaffolds were annotated by selecting se- quences with a clear 5¢ start site (i.e., where a substantial portion of reads corresponded to a single start site). These annotated scaffolds were used for secondary structure prediction and the TAM screen. Raw RNA-seq sequenc- ing data can be accessed at the National Center for Bio- technology Sequence Read Archive (BioProject PRJNA954882). RNA secondary structure was visualized using Vienna RNAfold.30,31 RNA se- quences were aligned using mafft-xinsi.21 RNAalifold32 and R2R33 were used to generate covariance models. For orthologs where no clear could be ascertained, the longest RNA species observed was used for the TAM screen. Information start site IVTT TAM screen DNA templates encoding T7 promoter-driven TnpB pro- teins were generated by PCR from custom synthesis products or bacterial genomic DNA. DNA templates encoding T7 promoter-driven xRNA scaffolds with a 20-nt guide sequence were also generated by PCR and used to prepare RNA with the T7 HiScribe Kit. IVTT reactions were prepared with 150 ng protein template, 5000 ng RNA, and 1 U/lL murine RNase inhibitor in the PURExpress kit (NEB). After 4 h at 37(cid:2)C, 50 ng of an 8N TAM library plasmid was added to each reaction, and the reaction proceeded for an additional 20 min at 37(cid:2)C. Reactions were treated with 10 lg RNase A (Qia- gen) and 8 U Proteinase K (NEB) each followed by a 5 min incubation at 37(cid:2)C. DNA was extracted by PCR TNPB PROCESSES ITS OWN GUIDE RNA 235 puriﬁcation, and adaptors were ligated using the NEB- Next UltraII DNA Library Prep Kit for Illumina (NEB) with the NEBNext Adaptor for Illumina (NEB) as per the manufacturer’s protocol. Following adaptor ligation, cleaved products were ampliﬁed using one primer spe- ciﬁc to the TAM library backbone and one primer spe- ciﬁc to the NEBNext adaptor with 12 cycles of PCR. After a second round of 18-cycle PCR with primers add- ing the i7 and i5 Illumina adaptors and barcodes, ampli- ﬁed libraries were gel-extracted, quantiﬁed using Qubit Fluorometric Quantiﬁcation, and subjected to single- end sequencing on an Illumina MiSeq with the following parameters: read 1-80 cycles, index 1-8 cycles, index 2-8 cycles. TAM enrichment was analyzed and visualized using a custom Python script.1,34 Raw TAM screen sequencing data can be accessed at the National Center for Biotech- nology Information Sequence Read Archive (BioProject PRJNA954882). Results TnpB processes its own mRNA into xRNA Given the evolutionary relationship between TnpB and Cas12s, we hypothesized that the ability of Cas12s to pro- cess pre-crRNA into crRNA35–40 may have originated from analogous functions in TnpB. We therefore investi- gated whether an exemplar TnpB ortholog from Alicyclo- bacillus macrosporangiidus (AmaTnpB) possesses RNA processing activity to generate xRNA. ‰ (A) Schematic illustrating transcription and FIG. 1. translation of the transposon locus, whereby processing of the mRNA yields the active TnpB RNP complex. (B) Substrates utilized to test RNA processing activity, containing AmaTnpB coding sequence (blue), putative xRNA scaffold (orange), guide (pink), and padding sequence (gray). (C) (Top) Denaturing TBE-urea gel of apo purified wild-type AmaTnpB or DRuvC mutant (containing a point mutant in the RuvC-II catalytic site) incubated with each of the four RNA substrates illustrated in (B). Components are present in a 1:1 molar ratio of TnpB protein:xRNA. (Bottom) Agarose E-gel demonstrating cleavage of a 1221-nt dsDNA substrate under the same conditions. Both gels are stained with SYBR gold. (D) RNA-seq of substrate 2 with and without active AmaTnpB in vitro, illustrating processing of the 5¢ section from the xRNA. Bars represent the start site of RNA species from RNA-seq. xRNA, omegaRNA; dsDNA, double-stranded DNA; RNP, ribonucleoprotein. 236 NETY ET AL. To directly interrogate whether TnpB can process RNA (Fig. 1A), we incubated puriﬁed apo form wild- type AmaTnpB or a catalytically dead RuvC-II point mu- tant (DRuvC) with 1:1 molar ratios of four different in vitro transcribed RNA substrates, as well as a target DNA substrate (Fig. 1B). These RNA substrates include a random negative control sequence (substrate 1), the hy- pothesized xRNA with a 20-nt guide sequence, which was previously determined based on RNP pulldown of a closely related ortholog1 (substrate 2), the full TnpB ORF extended to include the nonoverlapping xRNA and guide sequence (substrate 3), and the hypothesized xRNA with an additional 59 nt of 3¢ padding sequence (substrate 4). After incubation, a fraction of the sample was treated with DNase to visualize the RNA species on a denaturing gel, and the remaining sample was treated with RNase to visualize cleavage of the DNA substrate. On the denaturing gel, no processed RNA substrates were visualized from incubation of TnpB with substrate 1 (Fig. 1C). Upon TnpB incubation with RNA substrate 2, the hypothesized xRNA sequence was processed to a 126-nt sequence, which we conﬁrmed by RNA sequ- encing and hereafter refer to as the processed xRNA (Fig. 1D). Substrate 3, which resembles the native mRNA sequence, was processed to this same 126-nt spe- cies (Fig. 1C). Substrate 4 was processed to a 185-nt spe- cies. The difference in length between the processed species resulting from substrates 2/3 and 4 suggests that the 59-nt 3¢ ﬂanking sequence is not processed by TnpB. The DRuvC mutant did not exhibit RNase activity on any substrate, suggesting that TnpB, and not an RNase contaminant from puriﬁcation, is responsible for RNA processing and further that the RuvC domain of TnpB is responsible for xRNA biogenesis. We next investigated whether xRNAs processed by TnpB support target cleavage by examining the RNase- treated fractions of the same reactions above, which include a 1221-nt dsDNA substrate containing the cog- nate target sequence and TAM for AmaTnpB (5¢ TCAC).1 Wild-type TnpB utilized the processed xRNA from substrates 2 and 3 to cleave the DNA target, and the inclusion of 3¢ padding sequence in substrate 4 did not hinder substrate cleavage (Fig. 1C). These results imply that although TnpB is equipped to process 5¢, but not 3¢, sequence, the enzyme can perform cleavage of its DNA substrate regardless of extra sequence padding the xRNA. This ﬁnding is consistent with the fact that TnpB from Deinococcus radiodurans (Dra2TnpB) utilizes only the proximal 12 nt of the guide sequence for DNA target- ing, regardless of the length of guide sequence provided, due to the lack of interaction between the protein and distal end of the xRNA: target heteroduplex.15,16 Cis-regulation of DNA cleavage by AmaTnpB mRNA When assaying for activity, we noticed that DNA cleav- age in the presence of substrate 3 is weak compared to substrates 2 and 4 (Fig. 1C). Therefore, we explored the hypothesis that the TnpB mRNA (i.e., the extra 5¢ sequ- ence in substrate 3 compared to substrate 2) exerts an inhibitory effect on its own DNA cleavage activity. We prepared reactions with active TnpB protein, the 126-nt processed xRNA, different 3¢ truncations of the mRNA, and DNA substrate (Fig. 2A). Compared to a reac- tion with a scrambled 1190-nt RNA (Fig. 2A lane 12) or without an additional RNA species (lane 1), the 5¢ 825 nt of the mRNA does not interfere with efﬁcient DNA cleav- age. However, a sequence or structural feature present in the mRNA between 825 and 875 nt (Fig. 2B) results in a sub- stantial reduction in DNA cleavage activity (Fig. 2A lanes 5–7). We conﬁrmed that a 125-nt RNA fragment encom- passing this region substantially reduces the DNA cleavage (Supplementary Fig. S1). We hypothesize that xRNA sec- ondary structure (Fig. 2C) may be disrupted by its comple- mentarity to this region of the mRNA (Fig. 2D), pointing to a potential mechanism for the inhibitory effect. Extensive xRNA diversity in survey of TnpB orthologs We then turned our attention to the natural diversity of TnpB systems to assess the prevalence of xRNA biogen- esis and to further understand xRNA structural and sequence diversity. TnpB is highly abundant in bacteria and archaea, and there is substantial diversity found within this family of proteins.1,41 To begin experimentally studying this diversity, we sub- sampled the full set of TnpB systems.8 We focused on members of the IS200/IS605/IS607 transposon superfam- ily, that is, those lacking association with CRISPR arrays. We further excluded TnpBs associated with transposases besides the canonical Y1 or serine recombinases or non- mobile orthologs, as those TnpBs may perform alternate functions. We also imposed a requirement that the 3¢ end of the xRNA be well-conserved within the clade, to facil- itate accurate prediction of the guide sequence. We focused on TnpBs from the ﬁve major clades, which are deﬁned by particular conﬁgurations of the RuvC catalytic aa motif: Typical TnpBs (RuvC-III DRDXN), Derived TnpBs (RuvC-III NADXN), and clades containing catalytic rear- rangements of the RuvC-II (RII-r3 and 5) or RuvC-III (RIII-r4) domain (Fig. 3).8 We note that there is no a priori expectation that TnpB proteins from these latter three clades will be catalytically inactivated; although the pre- dicted RuvC motif is atypical, compensatory mutations may permit catalytic function, akin to the natural variation in RNase H-like domain catalytic motifs.42 TNPB PROCESSES ITS OWN GUIDE RNA 237 (A) Agarose gel of DNA cleavage by AmaTnpB with 126-nt processed xRNA (lane 1) and additional RNA FIG. 2. species, including various 3¢ truncations of the mRNA (lanes 2–11) and a scrambled negative control (lane 12). Components are present in a 1:1:1 molar ratio of TnpB protein:mRNA:xRNA. (B) Schematic of AmaTnpB mRNA, highlighting RuvC catalytic residues and hypothesized inhibitory region, which is adjacent to the RuvC-II catalytic site. (C) Predicted MFE secondary structure of the 106-nt processed AmaTnpB xRNA scaffold, illustrating four stem- loop regions. (D) Predicted co-folding of 106-nt processed xRNA scaffold and 50-nt inhibitory region of the AmaTnpB mRNA. MFE, minimum free energy. Based on this information, we selected 59 TnpB ortho- logs that span the phylogenetic diversity within the con- straints outlined above. These orthologs range in length from 353 to 550 aa, with some proteins having as little as 7% aa sequence identity to each other (Supplementary File S2). To investigate these 59 TnpB orthologs at higher throughput, we expressed the TnpB-xRNA-encoding loci from a single DNA template in IVTT reactions. As TnpB processes only the 5¢ end, we utilized 5¢ RACE to determine the 5¢ processing site of each xRNA. By priming cDNA synthesis from the 3¢ end of the xRNA this technique captures the xRNA scaffold scaffold, only, excluding the guide region. We note that not all orthologs may exhibit proper expression and folding under IVTT conditions and that the absence of processing in our assay does not necessarily rule out the possibility that the ortholog has activity under different conditions. The xRNA sequences generated by the TnpB proteins we tested generally fall into one of two categories: those with at least one clear 5¢ processing site (30/59 orthologs) and those without a clear site based on the coverage plots of the RNA reads (Supplementary Fig. S2). The IVTT 5¢ RACE assay recapitulated the AmaTnpB in vitro pro- cessing experiments, generating a 106-nt scaffold (ortho- log 6 in Supplementary Fig. S2). Some orthologs, such as Dra2TnpB (ortholog 22), showed multiple apparent processing sites, consistent with previous observations suggesting either incomplete or promiscuous RNase activity.2,15 The 30 orthologs with evidence of process- ing ability are found throughout the tree and are not con- ﬁned to speciﬁc clades (Fig. 3). FIG. 3. Phylogenetic tree of TnpB protein sequences from publicly available genomes and metagenomes, illustrating representative sequences from the ﬁve major clades of TnpB and CRISPR-associated TnpBs (i.e., Cas12s). A schematic of the RuvC subdomains (I, II, III, ZF) for each clade illustrates the predicted catalytic residues (pink) at each site. Fifty-nine orthologs were sampled from the tree (dots), labeled with numbers. Labels with an asterisk indicate orthologs demonstrating active xRNA processing in IVTT reactions and whose predicted structures are illustrated in Supplementary Figure S3. Colored dots indicate 5¢ TAM sequence identiﬁed from the IVTT TAM screen and are shown in more detail in Supplementary Figure S5. IVTT, in vitro transcription and translation; TAM, target- adjacent motif; ZF, zinc ﬁnger. 238 TNPB PROCESSES ITS OWN GUIDE RNA 239 Processed xRNA species ranged from 79 to 466 nt and have predicted structures rich in hairpins, of which the number and relative orientation vary widely. The RE of IS200/605 transposons is known to contain a subterminal hairpin, which we ﬁnd in almost all orthologs within 10 nt of the 3¢ end of the xRNA scaffold (Fig. 4, Supplemen- tary Fig. S3). This appears to be the only consistently conserved feature among these divergent TnpB xRNA scaffolds. The transposase TnpA is known to interact with subterminal hairpins at the 5¢ and 3¢ ends of the transposon,43 but the 3¢ hairpin may be functionally important for TnpB as well. Recently solved cryo-EM structures of Dra2TnpB illustrate how a linker in the WED domain interacts with the stem of this 3¢ hair- pin.15,16 Furthermore, this interaction is present in exper- imentally solved structures of all Cas12 subtypes to date, pointing to an evolutionarily conserved nuclease-guide RNA interaction (Supplementary Fig. S4).14,15,37,39,44–48 Limited diversity of TAM sequences We utilized the newly-determined xRNA sequences to conduct a screen for DNA nuclease activity of the 59 TnpB orthologs in IVTT reactions to ascertain 5¢ TAM preferences. For orthologs without a clear 5¢ start site of the xRNA, we utilized the longest RNA species observed in the sequencing. Using this screen, we recov- ered the known TAM sequences of TnpB orthologs char- acterized previously, including AmaTnpB (TCAC)1 and Dra2TnpB (TTGAT).2 In total, 27 out of 59 orthologs demonstrated TAM activity in the screen; these active orthologs are broadly distributed among the different TnpB clades and branches (Fig. 3). One ortholog in the RII-r5 clade is active, conﬁrming that catalytic rearrangements of the RuvC catalytic site can still support guided dsDNA cleavage. We noted several orthologs that process xRNA but do not show activity in the IVTT TAM screen and speculate that these TnpBs may target alternate substrates besides dsDNA, consistent with the diversity of substrates targeted by Cas12s.49,50 TAMs were found to be a maximum of 5 nt long, with relatively little degeneracy in the positions (Supplemen- tary Fig. S5). Nearly all characterized TAMs are AT- rich, although a small subset is G-rich. This relative lack of diversity in the TAM sequence is striking com- pared to, for example, Cas9, which is a relatively less abundant and diverse protein family compared to TnpB1,8 but still exhibits a wide variation in PAMs.51 The limited TAM diversity of TnpBs may be explained by the co-evolution of TnpA and TnpB, as TnpA also utilizes the TAM to recognize the transposon ssDNA for excision and insertion.1,2,6,43,52 Although it is not yet clear how, if at all, TnpB interacts with TnpA and the transposon DNA, the fact that both proteins possess the same TAM sequence is a clear constraint on the TAM diversity. A B (A) Covariance model of xRNA scaffold structures from 11 orthologs in the clade of Typical TnpBs. In this FIG. 4. clade, the only conserved structure is the 3¢ stem-loop within 10 nt of the start of the guide sequence. (B) Covariance model of xRNA scaffold structures of 16 orthologs in the clade of Derived TnpBs. In this clade, two 5¢ stem loops, as well as the 3¢ stem loop, are conserved. 240 NETY ET AL. Discussion The biochemical activity of OMEGA nucleases was previ- ously found to include RNA-guided ssDNA and dsDNA cleavage, but no RNAse activity was evident.1 Although heterologous expression of TnpB loci in Escherichia coli generates processed xRNA species that physically associ- ate with TnpB protein, it was unclear whether these RNA species were the result of TnpB RNA processing activity or endogenous host RNases.1,2,15 In this study, we show that TnpB can process its own mRNA into xRNA to generate active RNP complexes. In the cell, it is possible that endogenous host RNases as- sist in processing, especially at the 3¢ end of the guide. We observed that an intact RuvC catalytic domain is es- sential for TnpB xRNA processing activity. We therefore infer that the TnpB RuvC domain performs the nucleo- philic attack on the RNA, which is consistent with the ac- tivities of Cas12c237 and Cas12j40 but in contrast to Cas12a35,36 and Cas12i,38,39 which instead use the WED domain for crRNA processing. These ﬁndings attri- bute the additional function of RNA processing to this class of RNA-guided nucleases. The observation that part of the AmaTnpB mRNA exerts an inhibitory effect on cleavage of DNA raises the question of whether this phenomenon is conserved among different TnpB orthologs. IS200/605 transposons are known to contain a variety of post-transcriptional cis- regulatory mechanisms to regulate TnpA activity, includ- ing mRNA secondary structure and an antisense small RNA,53 suggesting that TnpB activity may also be regu- lated by similar mechanisms. For TnpB, the inhibitory region of the mRNA may base pair with the xRNA, thereby disrupting its structure and ability to bind to TnpB. The inhibitory effect may also be acting on the RNP complex through a different mechanism. Whether this inhibitory effect can be recapitulated in cells and its functional consequence remains to be explored. According to recent evidence, TnpB improves reten- tion of its transposon whereby an xRNA expressed from one transposon locus targets transposon-lacking versions of that locus in the same cell, that is, loci where the transposon has not yet inserted or those which have undergone excision.6 Therefore, it appears that TnpB DNA cleavage serves to ﬁx the transposon- containing locus in the population by eliminating loci that have undergone excision1 and/or homing to un- inserted loci.2,54 Under the TnpB mRNA could serve as a temporally-sensitive signal of ac- tive transcription of the transposon and its continued presence in the genome and, therefore, exert negative feedback on TnpB DNA cleavage to prevent unnecessary genome instability. this hypothesis, We also note that the negative feedback exerted by TnpB mRNA on DNA cleavage should be taken into ac- count when utilizing TnpB for applications in heterolo- gous systems. Codon optimization of the TnpB ORF sequence up until the 5¢ end of the xRNA will likely ab- rogate the inhibitory effect, ultimately maximizing enzy- matic activity. Our exploration of 59 different TnpB loci reconstituted in IVTT revealed a substantial amount of diversity in the xRNAs of these orthologs. Among these orthologs, the xRNAs contain a minimum of two predicted stem-loop structures but otherwise vary widely in length and overall topology. Furthermore, parts of the xRNA with predicted secondary structure may in fact be disordered or ﬂexible, as was found with Dra2TnpB, for which a fully functional xRNA can be reconstituted by maintaining the structured domains of the xRNA.15 Overall, our demonstration of TnpB RNA processing and observation of a cis-regulatory mechanism comple- ment studies of the biological function of OMEGA effec- tors. Furthermore, our survey of the natural diversity of TnpB-xRNA complexes adds to the toolbox of this bio- technologically promising family of enzymes. Acknowledgments The authors are grateful to Osamu Nureki for sharing structural data for Dra2TnpB in advance of publication (PDB: 8h1j). The authors thank all the members of the Zhang Lab for helpful advice and support. Authors’ Contributions S.P.N. and F.Z. conceived and designed the study; S.P.N., F.E.D., S.H., K.M., and Y.Z. performed experiments; S.P.N., H.A.-T., S.K., and G.F. analyzed data; F.Z. super- vised the research and experimental design with support from R.K.M.; S.P.N., R.K.M., and F.Z. wrote the article with input from all authors. Disclaimer The content is solely the responsibility of the authors and does not necessarily represent the ofﬁcial views of the National Institute of General Medical Sciences or the National Institutes of Health. Author Disclosure Statement F.Z. is a cofounder of Editas Medicine, Beam Therapeu- tics, Pairwise Plants, Arbor Biotechnologies, Proof Diag- nostics, and Aera Therapeutics. F.Z. is a scientiﬁc advisor for Octant. Funding Information S.P.N. is supported by award nos. T32GM007753 and T32GM144273 from the National Institute of General TNPB PROCESSES ITS OWN GUIDE RNA 241 Medical Sciences. F.Z. is supported by an NIH grant (2R01HG009761-05); Howard Hughes Medical Institute; Poitras Center for Psychiatric Disorders Research at MIT; Hock E. Tan and K. Lisa Yang Center for Autism Research at MIT; K. Lisa Yang and Hock E. Tan Molec- ular Therapeutics Center at MIT; K. Lisa Yang Brain– Body Center at MIT; Broad Institute Programmable Therapeutics Gift Donors; The Pershing Square Founda- tion, William Ackman, and Neri Oxman; James and Pat- ricia Poitras; BT Charitable Foundation; Asness Family Foundation; the Phillips family; David Cheng; and Rob- ert Metcalfe. Supplementary Material Supplementary File S1 Supplementary File S2 Supplementary File S3 Supplementary Figure S1 Supplementary Figure S2 Supplementary Figure S3 Supplementary Figure S4 Supplementary Figure S5 References 1. Altae-Tran H, Kannan S, Demircioglu FE, et al. The widespread IS200/IS605 transposon family encodes diverse programmable RNA-guided endo- nucleases. Science 2021;374(6563):57–65; doi: 10.1126/science.abj6856 2. Karvelis T, Druteika G, Bigelyte G, et al. Transposon-associated TnpB is a programmable RNA-guided DNA endonuclease. Nature 2021; 599(7886):692–696; doi: 10.1038/s41586-021-04058-1 16. Sasnauskas G, Tamulaitiene G, Druteika G, et al. TnpB structure reveals minimal functional core of Cas12 nuclease family. Nature 2023; 616(7956):384–389; doi: 10.1038/s41586-023-05826-x 17. Au HKE, Isalan M, Mielcarek M. Gene therapy advances: A meta-analysis of AAV usage in clinical settings. Front Med 2021;8:809118; doi: 10.3389/ fmed.2021.809118 18. Kim DY, Lee JM, Moon SB, et al. Efﬁcient CRISPR editing with a hyper- compact Cas12f1 and engineered guide RNAs delivered by adeno- associated virus. Nat Biotechnol 2021; doi: 10.1038/s41587-021-01009-z 19. Kim DY, Chung Y, Lee Y, et al. Hypercompact adenine base editors based on transposase B guided by engineered RNA. Nat Chem Biol 2022;18: 1005–1013; doi: 10.1038/s41589-022-01077-5 20. Wu Z, Zhang Y, Yu H, et al. Programmed genome editing by a miniature CRISPR-Cas12f nuclease. Nat Chem Biol 2021; doi: 10.1038/s41589-021- 00868-6 21. Katoh K, Standley DM. MAFFT multiple sequence alignment software version 7: Improvements in performance and usability. Mol Biol Evol 2013;30(4):772–780; doi: 10.1093/molbev/mst010 22. Capella-Gutie´ rrez S, Silla-Martı´nez JM, Gabaldo´ n T. trimAl: A tool for automated alignment trimming in large-scale phylogenetic analyses. Bioinformatics 2009;25(15):1972–1973; doi: 10.1093/bioinformatics/ btp348 23. Kalyaanamoorthy S, Minh BQ, Wong TKF, et al. ModelFinder: Fast model selection for accurate phylogenetic estimates. Nat Methods 2017;14(6): 587–589; doi: 10.1038/nmeth.4285 24. Nguyen L-T, Schmidt HA, von Haeseler A, et al. IQ-TREE: A fast and effective stochastic algorithm for estimating maximum-likelihood phylogenies. Mol Biol Evol 2015;32(1):268–274; doi: 10.1093/molbev/ msu300 25. Hoang DT, Chernomor O, von Haeseler A, et al. UFBoot2: Improving the ultrafast bootstrap approximation. Mol Biol Evol 2018;35(2):518–522; doi: 10.1093/molbev/msx281 26. Letunic I, Bork P. Interactive tree of life (iTOL) v3: An online tool for the display and annotation of phylogenetic and other trees. Nucleic Acids Res 2016;44(W1):W242–5; doi: 10.1093/nar/gkw290 3. Siguier P, Gourbeyre E, Chandler M. Bacterial insertion sequences: Their genomic impact and diversity. FEMS Microbiol Rev 2014;38(5):865–891; doi: 10.1111/1574-6976.12067 27. Mistry J, Finn RD, Eddy SR, et al. Challenges in homology search: HMMER3 and convergent evolution of coiled-coil regions. Nucleic Acids Res 2013;41(12):e121; doi: 10.1093/nar/gkt263 4. Ton-Hoang B, Guynet C, Ronning DR, et al. Transposition of ISHp608, member of an unusual family of bacterial insertion sequences. EMBO J 2005;24(18):3325–3338; doi: 10.1038/sj.emboj.7600787 28. Mistry J, Chuguransky S, Williams L, et al. Pfam: The protein families database in 2021. Nucleic Acids Res 2021;49(D1):D412–D419; doi: 10.1093/nar/gkaa913 5. Stoddard BL. Homing endonuclease structure and function. Q Rev 29. Bland C, Ramsey TL, Sabree F, et al. CRISPR recognition tool (CRT): Biophys 2005;38(1):49–95; doi: 10.1017/S0033583505004063 6. Meers C, Le H, Pesari SR, et al. Transposon-encoded nucleases use guide RNAs to selﬁshly bias their inheritance. bioRxiv 2023;2023.03.14.532601; doi: 10.1101/2023.03.14.532601 7. Shmakov S, Smargon A, Scott D, et al. Diversity and evolution of class 2 CRISPR-Cas systems. Nat Rev Microbiol 2017;15(3):169–182; doi: 10.1038/nrmicro.2016.184 8. Altae-Tran H, Shmakov S, Makarova KS, et al. Diversity, evolution, and classiﬁcation of the RNA-guided nucleases TnpB and Cas12. n.d. A tool for automatic detection of clustered regularly interspaced palindromic repeats. BMC Bioinform 2007;8:209; doi: 10.1186/1471- 2105-8-209 30. Gruber AR, Lorenz R, Bernhart SH, et al. The Vienna RNA websuite. Nucleic Acids Res 2008;36(Web Server issue):W70-4; doi: 10.1093/nar/ gkn188 31. Wiegreffe D, Alexander D, Stadler PF, et al. RNApuzzler: Efﬁcient outer- planar drawing of RNA-secondary structures. Bioinformatics 2019; 35(8):1342–1349; doi: 10.1093/bioinformatics/bty817 9. Shmakov S, Abudayyeh OO, Makarova KS, et al. Discovery and functional characterization of diverse class 2 CRISPR-Cas systems. Mol Cell 2015; 60(3):385–397; doi: 10.1016/j.molcel.2015.10.008 32. Bernhart SH, Hofacker IL, Will S, et al. RNAalifold: Improved consensus structure prediction for RNA alignments. BMC Bioinform 2008;9:474; doi: 10.1186/1471-2105-9-474 10. Makarova KS, Wolf YI, Iranzo J, et al. Evolutionary classiﬁcation of CRISPR– Cas systems: A burst of class 2 and derived variants. Nat Rev Microbiol 2019;18(2):67–83; doi: 10.1038/s41579-019-0299-x 33. Weinberg Z, Breaker RR. R2R—software to speed the depiction of aesthetic consensus RNA secondary structures. BMC Bioinform 2011;12:3; doi: 10.1186/1471-2105-12-3 11. Zetsche B, Gootenberg JS, Abudayyeh OO, et al. Cpf1 is a single RNA- guided endonuclease of a class 2 CRISPR-Cas system. Cell 2015;163(3): 759–771; doi: 10.1016/j.cell.2015.09.038 34. Tareen A, Kinney JB. Logomaker: Beautiful sequence logos in Python. Bioinformatics 2020;36(7):2272–2274; doi: 10.1093/bioinformatics/ btz921 12. Cong L, Ran FA, Cox D, et al. Multiplex genome engineering using CRISPR/ Cas systems. Science 2013;339(6121):819–823; doi: 10.1126/science .1231143 35. Fonfara I, Richter H, Bratovicˇ M, et al. The CRISPR-associated DNA-cleaving enzyme Cpf1 also processes precursor CRISPR RNA. Nature 2016; 532(7600):517–521; doi: 10.1038/nature17945 13. Ran FA, Cong L, Yan WX, et al. In vivo genome editing using Staphylo- coccus aureus Cas9. Nature 2015;520(7546):186–191; doi: 10.1038/ nature14299 36. Swarts DC, van der Oost J, Jinek M. Structural basis for guide RNA pro- cessing and seed-dependent DNA targeting by CRISPR-Cas12a. Mol Cell 2017;66(2):221.e4–233.e4; doi: 10.1016/j.molcel.2017.03.016 14. Takeda SN, Nakagawa R, Okazaki S, et al. Structure of the miniature type V-F CRISPR-Cas effector enzyme. Mol Cell 2021;81(3):558.e3–570.e3; doi: 10.1016/j.molcel.2020.11.035 37. Kurihara N, Nakagawa R, Hirano H, et al. Structure of the type V-C CRISPR- Cas effector enzyme. Mol Cell 2022;82(10):1865.e4–1877.e4; doi: 10.1016/j.molcel.2022.03.006 15. Nakagawa R, Hirano H, Omura SN, et al. Cryo-EM structure of the 38. Zhang H, Li Z, Xiao R, et al. Mechanisms for target recognition and transposon-associated TnpB enzyme. Nature 2023;616(7956):390–397; doi: 10.1038/s41586-023-05933-9 cleavage by the Cas12i RNA-guided endonuclease. Nat Struct Mol Biol 2020;27(11):1069–1076; doi: 10.1038/s41594-020-0499-0 242 NETY ET AL. 39. Huang X, Sun W, Cheng Z, et al. Structural basis for two metal-ion catalysis of DNA cleavage by Cas12i2. Nat Commun 2020;11(1):5241; doi: 10.1038/s41467-020-19072-6 40. Pausch P, Al-Shayeb B, Bisom-Rapp E, et al. CRISPR-CasF from huge phages is a hypercompact genome editor. Science 2020;369(6501): 333–337; doi: 10.1126/science.abb1400 41. Kapitonov VV, Makarova KS, Koonin EV. ISC, a novel group of bacterial and archaeal DNA transposons that encode Cas9 homologs. J Bacteriol 2015;198(5):797–807; doi: 10.1128/JB.00783-15 42. Majorek KA, Dunin-Horkawicz S, Steczkiewicz K, et al. The RNase H-like superfamily: New members, comparative structural analysis and evo- lutionary classiﬁcation. Nucleic Acids Res 2014;42(7):4160–4179; doi: 10.1093/nar/gkt1414 43. Guynet C, Hickman AB, Barabas O, et al. In vitro reconstitution of a single- stranded transposition mechanism of IS608. Mol Cell 2008;29(3):302– 312; doi: 10.1016/j.molcel.2007.12.008 44. Yamano T, Nishimasu H, Zetsche B, et al. Crystal structure of Cpf1 in complex with guide RNA and target DNA. Cell 2016;165(4):949–962; doi: 10.1016/j.cell.2016.04.003 45. Yang H, Gao P, Rajashankar KR, et al. PAM-dependent target DNA rec- ognition and cleavage by C2c1 CRISPR-Cas endonuclease. Cell 2016;167(7):1814.e12–1828.e12; doi: 10.1016/j.cell.2016.11.053 46. Liu J-J, Orlova N, Oakes BL, et al. CasX enzymes comprise a distinct family of RNA-guided genome editors. Nature 2019;566(7743):218–223; doi: 10.1038/s41586-019-0908-x 47. Li Z, Zhang H, Xiao R, et al. Cryo-EM structure of the RNA-guided ribo- nuclease Cas12g. Nat Chem Biol 2021;17(4):387–393; doi: 10.1038/ s41589-020-00721-2 48. Park J-U, Tsai AW-L, Rizo AN, et al. Structures of the holo CRISPR RNA- guided transposon integration complex. Nature 2023;613(7945):775– 782; doi: 10.1038/s41586-022-05573-5 49. Yan WX, Hunnewell P, Alfonse LE, et al. Functionally diverse type V CRISPR-Cas systems. Science 2019;363(6422):88–91; doi: 10.1126/ science.aav7271 50. Wang JY, Pausch P, Doudna JA. Structural biology of CRISPR-Cas immu- nity and genome editing enzymes. Nat Rev Microbiol 2022;20(11):641– 656; doi: 10.1038/s41579-022-00739-4 51. Gasiunas G, Young JK, Karvelis T, et al. A catalogue of biochemically diverse CRISPR-Cas9 orthologs. Nat Commun 2020;11(1):5512; doi: 10.1038/s41467-020-19344-1 52. Lavatine L, He S, Caumont-Sarcos A, et al. Single strand transposition at the host replication fork. Nucleic Acids Res 2016;44(16):7866–7883; doi: 10.1093/nar/gkw661 53. Ellis MJ, Trussler RS, Haniford DB. A cis-encoded sRNA, Hfq and mRNA secondary structure act independently to suppress IS200 transposition. Nucleic Acids Res 2015;43(13):6511–6527; doi: 10.1093/nar/gkv584 54. Kaur D, Kuhlman TE. IS200/IS605 family-associated TnpB increases transposon activity and retention. bioRxiv 2022;2022.10.12.511977; doi: 10.1101/2022.10.12.511977 Received: March 22, 2023 Accepted: April 21, 2023 Online Publication Date: May 24, 2023
10.1073_pnas.2221613120	RESEARCH ARTICLE \| GENETICS OPEN ACCESS The retrotransposon R2 maintains Drosophila ribosomal DNA repeats Jonathan O. Nelsona,b,1, Alyssa Slickoa,b , and Yukiko M. Yamashitaa,b,c,1 Edited by R. Scott Hawley, Stowers Institute for Medical Research, Kansas City, MO; received December 20, 2022; accepted May 3, 2023 Ribosomal DNA (rDNA) loci contain hundreds of tandemly repeated copies of ribo- somal RNA genes needed to support cellular viability. This repetitiveness makes it highly susceptible to copy number (CN) loss due to intrachromatid recombination between rDNA copies, threatening multigenerational maintenance of rDNA. How this threat is counteracted to avoid extinction of the lineage has remained unclear. Here, we show that the rDNA-specific retrotransposon R2 is essential for restorative rDNA CN expansion to maintain rDNA loci in the Drosophila male germline. The depletion of R2 led to defective rDNA CN maintenance, causing a decline in fecun- dity over generations and eventual extinction. We find that double-stranded DNA breaks created by the R2 endonuclease, a feature of R2’s rDNA-specific retrotransposi- tion, initiate the process of rDNA CN recovery, which relies on homology-dependent repair of the DNA break at rDNA copies. This study reveals that an active retrotrans- poson provides an essential function for its host, contrary to transposable elements’ reputation as entirely selfish. These findings suggest that benefiting host fitness can be an effective selective advantage for transposable elements to offset their threat to the host, which may contribute to retrotransposons’ widespread success throughout taxa. ribosomal DNA \| retrotransposons \| Drosophila \| germline Ribosomal RNAs (rRNAs) account for 80 to 90% of all transcripts in eukaryotic cells (1). To meet this demand, the ribosomal DNA (rDNA) gene that codes for rRNA is tandemly repeated hundreds of times, making up rDNA loci on eukaryotic chromosomes. This repet- itive structure is susceptible to intrachromatid recombination that causes rDNA copy num- ber (CN) loss (Fig. 1A), which is a major cause of replicative senescence in budding yeast (2). Evidence of similar rDNA CN instability has been noted in some tissues from aged dogs and humans (3, 4). While rDNA instability in somatic tissues may lead to insufficient ribosomal activity and disrupt cellular function, perhaps leading to disease and threatening the health of the individual (5), rDNA instability in the germline threatens survival of the entire species due to the potential degeneration of rDNA loci over successive generations. rDNA CN is variable between individuals of most species but maintained within a consistent range throughout the population (6), implying that rDNA CN is dynamically maintained through transgenerational series of CN losses and re-expansions. Indeed, age-associated rDNA CN loss occurs in the Drosophila male germline and is inherited by the next gener- ation, but the flies that inherited reduced rDNA CN re-expand rDNA in their germline to ensure sufficient rDNA is transmitted to their progeny (7). Similarly, intensive studies have revealed that rDNA CN expansion in yeast maintains rDNA repeat abundance over gen- erations through sister chromatid recombination at rDNA loci (2). Germline rDNA CN restoration has been best studied in Drosophila, particularly in the phenomenon of “rDNA magnification,” first described over 50 years ago as the process wherein aberrant rDNA loci bearing minimal rDNA repeats recover to a normal rDNA CN (8, 9). rDNA magnification is proposed to be accomplished through unequal sister chromatid exchange (USCE) during the homologous recombination (HR)-mediated repair of double-stranded breaks (DSBs) at the rDNA locus (10). This proposed USCE mechanisms is similar to the model of tandem rDNA repeat expansion in yeast (11), except that it results in one sister chromatid gaining rDNA copies at the cost of “stealing” them from the other sister. Indeed, rDNA magnification requires genes involved in homologous recombination (HR)-mediated repair (12, 13), supporting the model of USCE-mediated rDNA CN expan- sion. We have demonstrated that the expansion of rDNA CN in the germline of the progeny from old fathers requires the same set of genes as rDNA magnification (7), suggesting that the mechanisms of rDNA magnification normally serve to maintain rDNA CN across gen- erations. This rDNA CN expansion likely operates in male germline stem cells (GSCs) (7), which support sperm production throughout adulthood through persistent asymmetric divisions that produce a self-rendered GSC and a differentiating daughter destined for sperm differentiation (14). We recently found that GSC divisions have biased inheritance of sister Significance Retrotransposons are mobile genetic elements that occupy a large fraction of eukaryotic genomes, but are generally regarded as genomic parasites that do not contribute to host biology. This study reveals that the Drosophila retrotransposon R2 has a function essential to maintain its hosts genome. R2 specifically mobilizes within the large number of tandem ribosomal DNA (rDNA) repeats needed for proper ribosomal function. rDNA repeats are frequently lost from the genome, and restoration of these lost copies within the germline is required to maintain rDNA over successive generations. We find that R2 activity stimulates the expansion of rDNA copies needed to maintain rDNA throughout a population, indicating that R2 mutually benefits the survival of its host genome and its own propagation. Author affiliations: aWhitehead Institute for Biomedical Research, Cambridge, MA 02142; bHHMI, Cambridge, MA 02142; and cDepartment of Biology, School of Science, Massachusetts Institute of Technology, Cambridge, MA 02142 Author contributions: J.O.N. and Y.M.Y. designed research; J.O.N. and A.S. performed research; J.O.N. contributed new reagents/analytic tools; J.O.N. and A.S. analyzed data; and J.O.N. and Y.M.Y. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221613120/-/DCSupplemental. Published May 30, 2023. PNAS 2023 Vol. 120 No. 23 e2221613120 https://doi.org/10.1073/pnas.2221613120 1 of 9 chromatids that preferentially retains the sister chromatid with more rDNA copies in the GSC during rDNA CN expansion, leading us to propose that USCE followed by retention of the expanded rDNA locus in the GSC achieves rDNA magnification (15). While these observations may explain how rDNA CN expansion occurs in the germline, the underlying factors that control this process, particularly the source of DNA breaks at rDNA loci that may initiate USCE, remain elusive. Metazoan rDNA genes frequently contain insertions of rDNA-specific transposable elements (TEs), such as the retrotrans- poson R2 in Drosophila. R2 is found throughout arthropods and R2-like elements are widely present across taxa, including Cnidaria, Planaria, nematodes, fish, birds, and reptiles (16, 17). These TEs use their sequence-specific nuclease to mobilize specifically within rDNA loci (18), inserting into rDNA genes and likely disrupting 28S rRNA function (19) (SI Appendix, Fig. S1A). TEs are generally regarded as genomic parasites, serving only their replication throughout the host genome via their mobilization, which typically has detrimental, or at best neutral, mutagenetic effects. Many TEs mitigate their potential detriment to the host through biased mobi- lization at “harmless” insertion sites, such as repetitive, noncoding, or heterochromatic regions (20). Here we show that the Drosophila R2 retrotransposon actively contributes to host functions needed for rDNA CN expansion: inhibition of R2 expression in the germline disrupts rDNA maintenance, resulting in GSC loss and reduced fecundity over successive generations. We find that the R2 rDNA-specific endonuclease is required for R2 to induce rDNA CN expansion, indicating that R2 activity is the source of DSBs at rDNA that can stimulate USCE. We propose that R2 is a “mutu- alistic” TE whose mobilization activity benefits host fitness, which in turn benefits their own evolutionary success. Results R2 Is Required for Normal Germline rDNA CN and GSC Maintenance during Aging. To test the potential impact of R2 in the Drosophila male germline, we conducted RNAi-mediated knockdown of R2 in the Drosophila male germline (nos-gal4- driven expression of RNAi lines, nos>R2i-1 or R2i-2, hereafter) (SI Appendix, Fig. S1A). RNAi constructs were specifically designed to target the open reading frame (ORF) part of R2, which is included in the mature R2 RNA that is translated to produce the sequence-specific endonuclease/reverse transcriptase protein that executes R2 retrotransposition (19, 21). Because translation of R2 ORF occurs in the cytoplasm, the RNAi machinery (which mainly operates in the cytoplasm) is expected to efficiently target R2 ORF expression. Indeed, we observed R2 transcripts are efficiently knocked down by expression of these RNAi constructs, without disrupting expression of the rDNA arrays where they are inserted (SI Appendix, Fig. S1 B–H). Since R2 is a multicopy TE, we could not achieve complete elimination of R2 transcripts, but our knockdown was similarly efficient to previously reported use of RNAi-mediated knockdown of TE transcripts that functionally disrupted TE activity (22). Surprisingly, we found that RNAi-mediated knockdown of R2 resulted in premature loss of GSCs during aging (Fig. 1 B–F). GSCs continuously produce differentiating germ cells to sustain sperm production throughout adulthood and thus are the source of the genome passed to the next generation (14). Whereas newly eclosed R2 RNAi males contained similar numbers of GSCs to controls, GSC number more rapidly declined during aging in R2 knockdown males compared to controls (Fig. 1F). Given that R2 is specifically inserted into rDNA, we also examined the effect of R2 knockdown on rDNA stability. Using highly quantitative droplet digital PCR (ddPCR), we found that while RNAi-mediated knockdown of R2 had no effect in young testes, older RNAi expressing animals had reduced rDNA copy number compared to controls (Fig. 1 G and H), suggesting rDNA CN loss is more severe during aging when R2 is inhibited. rDNA CN loss in the germline was further confirmed by DNA FISH on the meiotic chromosomes (SI Appendix, Fig. S2 A–F). Interestingly, one of the RNAi constructs (R2i-2) suffered rapid rDNA CN loss Fig. 1. R2 is required for GSC maintenance via rDNA CN maintenance during Drosophila male germline aging. (A) Model of rDNA repeat instability. (B–E) GSCs in 0- and 40-d old control (B and C) and R2 RNAi testes (D and E). Yellow dotted circle = GSC. GSC signaling niche indicated by . Green = Vasa, White = Fascillin III, Blue = DAPI. (Scale bar, 7.5 µM). (F) Average GSCs per testis in control and two R2 RNAi lines during aging. indicates P < 10−3 determined by Student’s t test. Error = 95% CI. (G and H) Testis rDNA CN determined by ddPCR. P value by Student’s t test. Error = 95% CI. (I) Average GSCs per testis during aging in Control and two R2 RNAi lines containing mini-X chromosome. Error = 95% CI. 2 of 9 https://doi.org/10.1073/pnas.2221613120 pnas.org within the first 10 d of adulthood, but recovered by 20 d of age (Fig. 1G). This effect may be due to the incomplete efficiency of RNAi knockdown, combined with the variable multicopy nature of R2 within rDNA, which appears to leave a small population of germ cells that retain the ability to express R2 in the presence of the RNAi (SI Appendix, Fig. S1B). We speculate that severe rDNA loss caused by the R2i-2 RNAi may rapidly select for such germ cells, which become enriched within the testis, accounting for the observed recovery in rDNA CN at later ages. rDNA CN insufficiency is likely the primary cause of GSC loss in R2 RNAi animals, because increasing total rDNA CN via introduction of a mini-chromosome harboring an rDNA locus (23) suppressed the premature GSC loss caused by R2 knockdown (Fig. 1I). These results revealed that R2 contributes to sustaining GSC population during aging through rDNA CN maintenance, uncovering an unanticipated benefit of the R2 retrotransposon to the host, despite the widely held view of mobile TEs being genetic parasites. R2 Is Necessary and Sufficient for rDNA Magnification. Historically, rDNA magnification has been assessed through the observation of the emergence of offspring with normal cuticle from fathers with abnormal (“bobbed”) cuticle caused by insufficient rDNA CN (8) (Fig. 2A). Drosophila melanogaster rDNA loci reside on the sex chromosomes (X and Y) (24), and rDNA magnification is typically assayed as the recovery of X chromosome rDNA CN: X chromosome rDNA loci harboring the minimal viable amount of rDNA (bbZ9, SI Appendix, Fig. S3A) undergoes magnification when combined with a Y chromosome lacking rDNA (bbZ9/Ybb0, “magnifying males” hereafter) (8) (SI Appendix, Fig. S3B). Importantly, the use of Ybb0 is required to induce rDNA magnification, presumably because a normal Y chromosome provides sufficient rDNA copy number (and thus does not activate the CN sensing mechanism to trigger magnification). Accordingly, rDNA magnification is typically not observed in males with a normal Y chromosome containing intact rDNA (bbZ9/Y +, “nonmagnifying males” hereafter) (9). We found that R2 knockdown reduces rDNA magnification from 13.73% in control conditions (bbZ9/Ybb0; nos-gal4, n = 233) to 0% in R2i-1 (bbZ9/Ybb0; nos-gal4/UAS-R2i-1, n = 181, P = 5.6 × 10−7) and 2.36% in R2i-2 animals (bbZ9/Ybb0; nos-gal4/UAS-R2i-2, n = 127, P = 9.9 × 10−4) (Fig. 2B). Moreover, the quantification of rDNA CN by ddPCR revealed that 87.5% of bbZ9 chromosomes increased rDNA CN in magnifying males (n = 96, P = 1.8 × 10−4), with an average increase of 18.29 rDNA copies across all samples (n = 96, P = 3.1 × 10−12) (Fig. 2C). Importantly, this observed increase in rDNA CN under magnifying conditions was detected from animals randomly selected across all offspring, including those that still had bobbed cuticle phenotype: this demonstrates that rDNA magnification broadly increases rDNA CN throughout the germline, despite only 13.73% of bbZ9 chromosomes recovering enough CN to support normal cuticle development. This rDNA CN increase in magnifying males is also eliminated upon R2 knockdown (Fig. 2C). These results reveal that R2 is required for rDNA CN expansion during rDNA magnification. Interestingly, we found that rDNA magnification was blocked only when the R2 RNAi constructs were expressed by the nos-gal4 driver in early germ cells (including GSCs), but not when expressed in later germ cells by the bam-gal4 driver (Materials and Methods) (Fig. 2B). These results indicate that R2 primarily functions in the earliest stages of germ cells (including GSCs) to support rDNA magnification. We further found that R2 is sufficient for rDNA CN expansion. Ectopic expression of transgenic R2 in the germline (SI Appendix, Fig. S4 A–F) induced rDNA magnification of the bbZ9 locus in nonmagnifying males (bbZ9/Y +) (Fig. 2 D and E). We found 3.3% Fig. 2. R2 is necessary and sufficient for rDNA magnification. (A) Diagram of rDNA magnification at the bbZ9 rDNA locus, during which dorsal cuticle defect (red arrow) revert to normal cuticle. (B) Percent magnified offspring determined by cuticular phenotype in offspring from magnifying males. P value determined by chi-squared test. Error = 95% CI. (C) Mean bbZ9 locus rDNA CN determined by ddPCR in daughters from males. P value determined by Student’s t test. Error = 95% CI. (D) Percent magnified offspring from nonmagnifying males. P value determined by chi-squared test. Error = 95% CI. (E) Mean bbZ9 locus rDNA CN determined by ddPCR in daughters from nonmagnifying males. Nonmagnifying condition is the same data as panel C. P value determined by Student’s t test. Error = 95% CI. For all experiments, control condition contains nos-gal4 alone with no RNAi or transgene. of female offspring from males expressing transgenic R2 (bbZ9/Y +; UAS-R2/+; nos-gal4/+) exhibited magnification (normal cuticle) (Fig. 2D, n = 877, P = 3.2 × 10–5), compared to control (bbZ9/Y +; nos-gal4 without R2 expression), which never showed any magni- fication, when dorsal cuticle phenotype was used as an assay. Thus, transgenic R2 expression is sufficient to induce rDNA magnifica- tion, albeit at a low frequency. Importantly, reversion of the cuticle phenotype was heritable to the subsequent F2 generation through- out our experiments, confirming that CN restoration occurred in the germline (SI Appendix, Fig. S5 A–C). Quantification of rDNA CN by ddPCR revealed that ectopic overexpression of R2 in non- magnifying males (bbZ9/Y +) also increases the average rDNA CN at bbZ9 rDNA loci among all offspring, again regardless of cuticular phenotype (Fig. 2E, n = 94, P = 0.0256), revealing R2 expression induces rDNA CN expansion broadly among inherited rDNA loci. Critically, expression of a nuclease dead R2 transgene (NucDeadR2) in nonmagnifying males (SI Appendix, Fig. S4 A–E) failed to induce rDNA magnification (Fig. 2D), suggesting that the nuclease activity of R2 is essential for its ability to induce rDNA CN expansion. The R2 Endonuclease Is Required to Induce rDNA CN Expansion. In yeast, rDNA CN expansion is initiated by DSBs at the rDNA intergenic sequence, which induces sister chromatid PNAS 2023 Vol. 120 No. 23 e2221613120 https://doi.org/10.1073/pnas.2221613120 3 of 9 recombination that results in rDNA gene duplication (25). All proposed models of Drosophila rDNA CN expansion [the most prominent model being USCE (10)] require an initiating DSB at the rDNA locus (Fig. 3A and SI Appendix, Fig. S6 A and B). Indeed, artificial introduction of DSBs at rDNA loci by I-CreI endonuclease expression has been reported to induce rDNA magnification (26), but the endogenous factor that induces rDNA magnification remained unclear. R2 is capable of creating DSBs through sequential nicking of opposite DNA strands during retrotransposition (16). It has been speculated that DSBs created during R2 retrotransposition may be an initiating event of rDNA magnification (27), although this possibility has yet to be empirically tested. We found that rDNA magnification is associated with an elevation in DSBs in GSCs: the frequency of γH2Av-positive GSCs is increased in magnifying males (bbZ9/ Ybb0) compared to nonmagnifying males (bbZ9/Y +) (Fig. 3 B, C, and E; n = 519, P = 8.8 × 10−4). Strikingly, we observed that knockdown of R2 in magnifying males reduced the frequency of γH2Av-positive GSCs to levels comparable to nonmagnifying males (Fig. 3 D and E; n = 537, P = 7.1 × 10−4 for R2i-1; n = 521, P = 7.9 × 10−4 for R2i-2), indicating that R2 is responsible for the DSBs formed in GSCs during rDNA magnification. Furthermore, we confirmed that expression of transgenic R2, but not NucDeadR2, indeed induces chromosomal breaks at rDNA loci identified by chromosome spreads (SI Appendix, Fig. S4 B–D). R2 overexpression (but not NucDeadR2) in the germline also increased the frequency of GSCs with DSBs, identified by γH2Av expression (SI Appendix, Fig. S4E). Taken together, these results suggest that rDNA-specific endonuclease activity of R2 creates DSBs at the rDNA loci that may in turn induce rDNA CN expansion. Importantly, R2 transgenes (UAS-R2 and UAS-R2NucDead) contain synonymous mutations that confer resistance to the R2i-1 RNAi, and the expression of functional R2 but not NucDeadR2 was able to rescue the disruption of GSC homeostasis caused by R2 RNAi (SI Appendix, Fig. S4F). These findings confirm that the defects in GSC homeostasis upon R2 RNAi expression are indeed due to loss of R2 function, and suggest that R2 endonuclease activity is required for R2 contribution to rDNA maintenance in GSCs. R2 Is Dynamically Regulated within GSCs in Response to rDNA CN. Given the threat R2 mobilization poses to the host genome, both by disruption of rRNA function and causing excessive DSB formation (16), how is the potential benefit of R2 to rDNA CN maintenance balanced with the detriment of R2 retrotransposition? We found R2 expression in the germline is specifically derepressed under conditions of reduced rDNA CN, potentially explaining how the conflicting consequences of R2 expression are resolved. Using RNA fluorescence in situ hybridization (RNA FISH) to examine R2 expression at a single-cell resolution, we found that the frequency of GSCs expressing R2 was significantly increased in magnifying males (bbZ9/Ybb0), whereas nonmagnifying males (bbZ9/Y +) rarely expressed R2 (Fig. 4 A–C; n = 231, P = 1.7 × 10−10). Moreover, we found that GSCs from aged animals and the sons of old fathers, which inherit reduced rDNA CN (7), also exhibited a higher frequency of R2 expression compared to GSCs from young flies (Fig. 4D and SI Appendix, Fig. S7 A and B; n = 1,247, P = 8.3 × 10−4 for old animals; n = 1,107, P = 1.5 × 10−4 for offspring). Importantly, the frequency of R2 expression among GSCs in the sons of old fathers returned to the basal level after 20 d of age, when rDNA CN was shown to have recovered (7) (Fig. 4D and SI Appendix, Fig. S7C; n = 617, P = 0.036). These results indicate that R2 expression is dynamically regulated in response to changing rDNA CN. Taken together, we propose that R2 expression is finely tuned to function when most beneficial to the host while minimizing unnecessary exposure to the harmful effects of transposition. R2-Mediated rDNA CN Expansion Is Required for Multigenerational Maintenance of Germline Function. Based on the finding that R2 plays a critical role in maintaining germline rDNA CN, we postulated that R2 is essential to prevent continuous multigenerational rDNA Fig. 3. Derepressed R2 creates DSBs in GSCs during rDNA magnification. (A) Diagram of rDNA CN expansion by unequal sister chromatid exchange during DSB repair at rDNA loci. Recombination between misaligned rDNA copies during DSB repair result in crossovers that create unequal sister chromatid exchange that increases rDNA CN on one chromatid. (B–D) Detection of DSBs in the early adult male germline by anti-γH2Av staining. R2 RNAi expressed under the nos-gal4 driver. Non-RNAi conditions contain the nos-gal4 driver alone. GSCs indicated by yellow dotted circle. Blue = DAPI, Green = vasa, Magenta = γH2Av, white = FasIII. The hub is indicated by . (Scale bar, 10 µM). (E) Percentage of γH2Av positive GSCs. P value determined by chi-squared test. Error = 95% CI. (F) Number of GSCs per testis in R2i-1 condition coexpressing R2 transgenes. P value determined by Student’s t test. Error = 95% CI. 4 of 9 https://doi.org/10.1073/pnas.2221613120 pnas.org Fig. 4. R2 expression is regulated in response to changes in rDNA copy number. (A and B) R2 expression in GSCs (yellow dotted circle). Blue = DAPI, Green = R2 mRNA. Isolated R2 channel in A′ and B′. The hub is indicated by . R2 positive cells GSCs are marked by yellow arrowhead. (C) Percentage of R2 positive GSCs in nonmagnifying (Y +/bbZ9) and magnifying (Ybb0/bbZ9) animals. (D) Percentage of R2 positive GSCs in newly eclosed and aged adults from young or old fathers. P values determined by chi-squared test. Error = 95% CI. loss capable of causing the extinction of the lineage. In C. elegans, the loss of genome integrity is known to cause gradual loss of fertility, a phenotype known as mortal germline (morg) (28). To test whether R2-mediated rDNA maintenance is required to maintain fertility through generations, we established multiple independent lines expressing R2 RNAi in their germline and tracked their fecundity at each generation through the ability of each line to produce sufficient offspring to establish a new generation (SI Appendix, Fig. S8). While nearly all control lines survived throughout the duration of the experiments, we found that lines expressing the R2i-1 RNAi construct failed to consistently produce sufficient progeny, with over half failing by the fourth generation (Fig. 5A) (n = 43, P = 3 × 10−6), indicating that R2 is essential for continuity of the germline lineage. Surviving males of extinguishing R2i-1 lineages had ~20% reduction in rDNA CN compared to control lines (n = 22, P = 0.031) (Fig. 5B). With the R2i-2 RNAi, the lineage was maintained relatively well, after initial sharp drop (Fig. 5A): Considering that R2 knockdown by the R2i-2 construct exhibits only transient rDNA CN decrease at day 10 (SI Appendix, Fig. S2A), which we speculate is produced through selection of germ cells that have sporadically retained R2 expression in the RNAi condition, this effect may also quickly select for lineages insensitive to R2 knockdown that have retained rDNA CN expansion activity. Taken together, these results suggest that R2-mediated maintenance of rDNA contributes to germline immortality. Discussion Our findings reveal an unanticipated “function” of retrotransposon activity to benefit the host genome through a role in rDNA CN maintenance. Tandem repetitive DNA sequences are among the elements of the eukaryotic genome most vulnerable to genomic excision (29), and essential tandemly repeated coding and non- coding elements likely require active mechanisms that serve their Fig. 5. R2 is required to maintain rDNA CN and fertility over successive generations. (A) Kaplan–Meier curve of lineage survival in control (nos-gal4 driver alone and two R2 RNAi expressing via the nos-gal4 driver) lineages. Each lineage constitutes an individual data point. P values determined by log rank test. (B) rDNA CN determined by ddPCR in males of control animals at the 6th generation or R2 RNAi lineages at their terminating generation. P value determined by Student’s t test. Error = 95% CI. (C) Model of the role of R2 in germline rDNA CN maintenance. PNAS 2023 Vol. 120 No. 23 e2221613120 https://doi.org/10.1073/pnas.2221613120 5 of 9 maintenance. We propose that rDNA loci are maintained by DSBs generated by R2 in GSCs with reduced rDNA CN, which stim- ulate sister chromatid exchange that results in rDNA CN expan- sion to restore CN (Fig. 5C). We recently reported that USCE in GSCs followed by nonrandom segregation of the “expanded” sister chromatid to the self-renewed GSC during mitosis likely mediates rDNA magnification (15). The present study suggests that R2 functions upstream of USCE, creating the DSBs that stimulate this repeat expansion mechanism, revealing the function that TEs plays for the host. This work provides experimental evidence that supports the model for R2 to initiate rDNA magnification first suggested over 30 years ago by Hawley and Marcus (27), which was not possible to test at the time in the absence of methods to inhibit the expression of multicopy TEs (e.g., RNAi). TEs can be a major source of genomic instability, generating DNA breaks, disrupting genes as they mobilize, or creating oppor- tunities for recombination between TE insertions (30). There is often selective advantage for TEs to minimize the threat of their mobilization through restricting their expression or limiting their range of mobilization (20). Our proposed role for R2 in rDNA maintenance suggests that contributing to host functions may also be a robust adaptive feature for TEs through creating a mutualistic host–TE relationship. Since DSB formation by the R2 endonu- clease relies on reverse transcriptase activity (18), it remains unclear whether full integration of a new R2 insertion or DSB generation alone stimulates rDNA CN expansion. Further analysis that can separate R2 endonuclease activity from reverse transcription will reveal if R2 contributes to rDNA CN expansion beyond DSB formation alone. There are several descriptions of TEs providing benefit to their host through their coding or noncoding features being repurposed for host function, but these “co-opted” TEs lack their own ability to mobilize and replicate (31). There are very few well-described examples of eukaryotic “mutualistic” TEs, whose active mobilization benefits host fitness (32). The only other func- tionally demonstrated mutualistic TEs are the telomere-bearing element (TBE) DNA transposases in the ciliate Oxytricha trifallax (33). The TBEs execute the large-scale genome rearrangement needed for O. trifallax macronuclear development, and knockdown of TBE transposons by RNAi disrupts macronuclear assembly (33). The first proposed “functional” TEs are the retrotransposons that constitute the telomeres of most Drosophila species (Het-A, TART, TAHRE) (34). The absence of telomerase from these species, com- bined with evidence that suggests their transposition may be licensed by the host, lead to the model that retrotransposition of these telomeric TEs maintains telomeres in the fly (35). Although retrotransposition of telomeric TEs is indeed required to establish new telomeres at broken chromosome ends (36), it has yet to be functionally demonstrated whether these TEs are required to main- tain existing telomeres. On the contrary, recombination-based, TE-independent telomere extension has been observed to be a major source of telomere extension in Drosophila (37) and some Drosophila species completely lack functional telomeric TEs, appearing to rely solely on this recombination-based telomere extension (38). Therefore, while telomeric TEs may potentially be mutualistic elements, their requirement for telomere maintenance and host fitness remains unclear, and these elements may have instead simply found a safe-haven for insertion at telomeres. There appears to be a fine line between mutualistic element and oppor- tunistic parasite, and further discovery of functional instances of mutualistic TEs will be critical to understand how host–TE rela- tionships may shift between parasitism and mutualism. For the mutualistic host–TE relationship to exist, R2 expression is likely dynamically regulated through the interaction with the host, such that its expression is limited to only when it can be beneficial (i.e., decreased rDNA CN). This control may be achieved through modulation of a number of mechanisms that can regulate R2 expression. The piRNA pathway is the major repressor of TEs in the germline, including R2 (39). Indeed, R2 is normally repressed in the male germline, but becomes dere- pressed only when rDNA CN becomes insufficient (Fig. 4 A–C). Thus, it is possible that the activity of the piRNA pathway is modulated such that R2 becomes derepressed only when germ cells experience insufficient rDNA CN (e.g., changes in the activ- ity of piRNA core machinery or a reduction in R2-specific piR- NAs). It is also possible that the transcription of R2 may be regulated. Lacking its own Pol II promoter, R2 transcription is dependent on transcription of the rDNA copy where it is inserted (40), suggesting changes in rDNA transcription upon rDNA CN loss may alter R2 expression. Indeed, we observed large-scale tran- scriptional changes occur at rDNA loci in GSCs upon rDNA CN loss (7), suggesting regulation over rDNA transcription may con- trol R2 expression in response to rDNA CN loss. rDNA tran- scription and stability are broadly impacted by siRNA-mediated histone methylation (41, 42), and modification of this activity may underlie the increased R2 expression upon rDNA reduction. Furthermore, disruption of ribosome processing has also been shown to selectively induce R2 expression (43). This indicates that ribosomal abundance or function may be the molecular sensor that triggers R2 expression upon rDNA CN reduction, perhaps through a compensatory activation of hitherto repressed rDNA copies, including those containing R2. Future investigation to uncover how these mechanisms may control R2 expression in response to rDNA CN are critical to understanding how R2 can be utilized for the host’s benefit. The elucidation of the nature of the rDNA CN sensing mech- anism is critical to understand how R2 expression is regulated in response to rDNA CN. Curiously, in our R2 RNAi conditions we observed robust defects in germ line viability and function while detecting modest deficiencies in rDNA CN itself, suggesting R2 expression and rDNA CN expansion in the germline may be triggered by relatively small reductions in rDNA CN. This modest effect may be partly due to survivorship bias in our sampling, since rDNA CN cannot be measured in nonviable animals that have fewer rDNA copies than required for viability. Additionally, the phenotypic effects of these small CN changes may be due to a physiological need to maintain large rDNA CN beyond the small subset needed for transcription at any given time (44). rDNA loci are a common site of transcription–replication collisions that can create DNA breaks when replication forks progress through highly transcriptionally active regions of the rDNA (45). Accordingly, cells must compartmentalize rDNA copies into those that are actively transcribed and others that are being replicated to avoid transcription–replication collisions. Because of the need of com- partmentalization, cells must carry more rDNA CN than mini- mally required to transcribe sufficient rRNA for cells’ survival. The transcription–replication collisions become more likely to occur as rDNA loci shrink, and their increased frequency com- promises the efficiency of DNA damage repair activity (2). Indeed, the reduction in untranscribed rDNA copies have been observed to increase the sensitivity to DNA damage in yeast (44). Given that preserving genomic integrity is a top priority for germ cells, and these cells have a low tolerance for DNA damage (46), it is likely that the germline would be particularly sensitive to reduc- tions in rDNA CN. Furthermore, it is unclear what genetic or environmental factors influence the necessary number of untran- scribed rDNA copies, and variation in the demand for surplus rDNA may contribute to the large variation in rDNA CN between Drosophila strains. Such deviations in rDNA CN requirements 6 of 9 https://doi.org/10.1073/pnas.2221613120 pnas.org may underlie the inconsistency in the correlation between rDNA CN and phenotypic effect that we observed between experimental conditions. Future investigation into the direct causes of rDNA CN reduction to disrupt GSC physiology, in particular the roles of rRNA synthesis and transcription–replication collisions, is crit- ical to fully understand the selective forces that impact the inter- action between R2 and its host genome. The widespread presence of R2 and other rDNA-specific TEs in both vertebrates and invertebrates (17) suggests that similar host–TE mutualism may support rDNA CN maintenance throughout Metazoa. Interestingly, many of the rDNA-specific TEs have little sequence similarity to R2, instead appearing to be derived from other nonspecific TEs (17), suggesting this host–TE mutualism may have evolved multiple times over the course of evolution. In summary, our study provides an example of mutu- alistic retrotransposons in the maintenance of eukaryotic genomes, and we propose that such host–TE relationships may be wide- spread throughout eukaryotes. Materials and Methods Immunofluorescence. Immunofluorescence staining of testes was performed as previously described (47). Briefly, testes were dissected in PBS, fixed in 4% formaldehyde in PBS for 30 min, then briefly washed two times in PBS containing 0.1% Triton-X (PBS-T), followed by washing in PBS-T for 30 min. After washes, samples were incubated at 4 °C overnight with primary antibody in 3% bovine serum albumin (BSA) in PBS-T. Samples were washed three consecutive times for 20 min in PBS-T, then incubated at 4 °C overnight with secondary antibody in 3% BSA in PBS-T, washed three times again in PBS-T for 20 min, and mounted in VECTASHIELD with DAPI (Vector Labs). The following primary antibodies were used: rat antivasa (1:20; DSHB; developed by A. Spradling), mouse anti-Fascillin III (1:200; DSHB; developed by C. Goodman), and rabbit anti-γ-H2AvD pS137 (1:200; Rockland). Images were taken with a Leica Stellaris 8 confocal microscope with 63× oil-immersion objectives and processed using Fiji (ImageJ) software. RNA FISH and Image Quantification. RNA FISH samples were prepared as previously described (7). In short, dissected testes were fixed in 4% formaldehyde in PBS for 30 min, briefly washed in PBS, and permeabilized in 70% ethanol overnight at 4°. Samples were then briefly rinsed in 2× SSC with 10% formamide prior to hybridization with 50 nM probes overnight at 37°. Samples were washed twice in 2× SSC with 10% formamide for 30 min and mounted in VECTASHIELD with DAPI (Vector Labs). Samples were imaged using a Leica Stellaris 8 confocal microscope with 63× oil-immersion objectives and processed using Fiji (ImageJ) software. R2 Stellaris FISH probe set was designed and synthesized by Biosearch Technologies. ITS probe sequence is listed in SI Appendix, Table S1. For ITS RNA FISH signal quantification, nonsaturating images were taken of optimized z-slices throughout each imaged cell. ITS signal intensity was quantified using Fiji (ImageJ) software and signal intensity was summed across all z-slices for each cell. The summed intensity of each GSC was normalized to the summed intensity of a somatic Cyst Stem Cell within similar z planes. Only GSCs with a suitable normalizing Cyst Stem Cell within similar z planes were scored. DNA Isolation. Testis DNA was isolated from 50 pooled dissected testes frozen in liquid N2. DNA isolation was performed according to previously described methods for isolation from Drosophila tissues (48). DNA was isolated from indi- vidual Drosophila animals using a modified protocol of the DNeasy Blood and Tissue DNA extraction kit (Qiagen). In short, individual animals were homogenized in 200 µL Buffer ATL containing proteinase K using a pipette tip in Eppendorf tubes, vortexed for 15 s, and incubated for 1.5 h at 56°. Samples were then prepared following the manufacturer’s protocol after incubation. All DNA samples were quantified and checked for purity by NanoDrop One spectrophotometer (ThermoFisher). rDNA Copy Number Measurement by Droplet Digital PCR (ddPCR). Thirty nanograms of DNA sample were used per 20 µL ddPCR for control gene reactions (RpL and Upf1), and 0.3 ng of DNA per 20 µL ddPCR for 28S rDNA reactions. Primers and probes for reactions are listed at SI Appendix, Table S1. ddPCR were carried out according to the manufacturer’s (Bio-Rad) protocol. In short, master mixes containing ddPCR Supermix for Probes (No dUTP) (Bio-Rad), DNA samples, primer/probe mixes, and HindIII-HF restriction enzyme (New England Biolabs) for 28S rDNA reactions (no restriction enzyme for control gene reactions) were prepared in 0.2-mL Eppendorf tubes, and incubated at room temperature for 15 min to allow for restriction enzyme digestion. ddPCR droplets were gener- ated from samples using QX200 Droplet Generator (Bio-Rad) and underwent complete PCR cycling on a C100 deep-well thermocycler (Bio-Rad). Droplet fluorescence was read using the QX200 Droplet Reader (Bio-Rad). Sample copy number was determined using Quantasoft software (Bio-Rad). rDNA copy number per genome was determined by 28S sample copy number multiplied by 100 (due to the 100× dilution of sample in the 28S reaction compared to control reaction) divided by control gene copy number multiplied by the expected number of control gene copies per genome (2 for RpL in all samples; 2 for Upf1 in female samples; 1 for Upf1 in male samples). The 28S copy number values determined by each control gene was averaged to determine 28S copy number for each sample. RNA Isolation. Fifty dissected testes were pooled and frozen in liquid N2 for each RNA isolation sample. Samples were homogenized in 400 µL TRIzol™ (ThermoFisher Scientific) and RNA was isolated using Direct-zol™ RNA Miniprep kit (Zymo Research) according to manufacturer directions, including on-column DNase I treatment. All RNA samples were quantified and checked for purity by Nanodrop One spectrophotometer (ThermoFisher). Quantification of R2 Expression by Reverse Transcriptase (RT)-ddPCR. Approximately 20 ng of total RNA was used per 20 µL RT-ddPCR for R2 reactions, and 0.2 ng of total RNA were used per 20 µL control gene (Tubulin) reaction. Tubulin primers and probe are listed at SI Appendix, Table S1, and R2 primers and probe mix were designed by Bio-Rad (Assay ID: dCNS858096478). ddPCR droplets were generated from samples using QX200 Droplet Generator (Bio-Rad) and underwent RT-PCR and endpoint PCR on a C100 deep-well thermocycler (Bio- Rad). Droplet fluorescence was read using the QX200 Droplet Reader (Bio-Rad). RNA quantitation was determined using QuantaSoft software (Bio-Rad), and R2 counts were normalized to Tubulin concentration for all samples. Normalized R2 expression values were then set relative to the average R2 expression value in control conditions. Quantification of ETS Expression by qRT-PCR. Approximately 1 μg of total RNA was used for cDNA synthesis via SuperScript III First-Strand Synthesis (Invitrogen) using random hexamer primers, according to manufacturer direc- tions. Real-time PCR was done using SYBR® Green PCR Master Mix (Applied Biosystems) and assessed with a QuantStudio 6 Flex system (Applied Biosystems). ETS expression values were normalized to GAPDH. Primers used are listed in SI Appendix, Table S1. Mitotic and Meiotic Chromosome Spread, DNA FISH, and Quantification. Mitotic chromosome spreads in neuroblast cells, DNA FISH, and imaging were all done as previously described (49). In short, brains were dissected from male third instar larvae in PBS and fixed in 25 µL of acetic acid and 4% formaldehyde in PBS. Samples were applied to Superfrost plus slides and manually squashed under a coverslip, then immediately frozen in liquid N2. After freezing slides were removed from N2, coverslip removed, and slides were dehydrated in 100% ethanol and dried at room temperature. DNA FISH hybridization was performed in 20 µL of 50% formamide, 10% dextran sulfate, 2× SSC buffer and 0.5 µM each probe applied directly to the sample on the slide and covered with a cover slip. Samples were incubated at 95° for 5 min, cooled and wrapped in parafilm, then incubated overnight at room temperature in a dark humid chamber. Coverslips were removed and slides were washed three times for 15 min in 0.1× SSC, dried, and then mounted in VECTASHIELD with DAPI (Vector Labs). Samples were imaged using a Leica Stellaris 8 confocal microscope with 63× oil-immersion objectives and processed using Fiji (ImageJ) software. Meiotic chromosome spreads were prepared from dissected testes and imaged in the same manner. Relative Y:X rDNA fluorescence quantification was determined as previously described (7). Probes used for this study are as follows: 359, 5′-AGGATTTAGGGAAATTAATTTTT GGATCAATTTTCGCATTTTTTGTAAG-3′-Cy5; (TAGA)6-Cy5; IGS, 5′-AGTGAAAAATGTTGA AATATTCCCATATTCTCTAAGTATTATAGAGAAAAGCCATTTTAGTGAATGGA-3′-Alexa488; (AATAC)6-Cy3; and (AATAAAC)6-Cy3. PNAS 2023 Vol. 120 No. 23 e2221613120 https://doi.org/10.1073/pnas.2221613120 7 of 9 Generation of rDNA Deletion Animals. rDNA copy number loss was induced during larval development in yw/Y; HS-I-CreI, Sb/TM6B males with a y, w X chro- mosome by I-CreI expression as previously described (26). In brief, parental ani- mals mated and laid eggs for 3 d, then removed from food. After one additional day of larval development, animals were exposed to 37 °C heat shock for 45 min on two consecutive days. To identify X chromosomes with significant rDNA copy number reduction (bb), adult males that experienced I-CreI expression were mated to bb158/FM6 females, and virgin non-FM6 daughters (bb/bb158) were screened for the bobbed phenotype. 28 out of 946 non-FM6 daughters screened were bobbed. To isolate potentially reduced rDNA loci and remove HS-I-CreI from the background, TM6B containing bobbed females were individually mated to wildtype males. Male offspring from each individual female candidates were subsequently individually mated to bb158/FM6 females. Any mating that failed to produce non-FM6 daughters were eliminated (due to having the bb158 and not the candidate bb chromosome). All viable non-FM6 daughters were dou- ble-checked for the bobbed phenotype, and stocks with all bobbed non-FM6 daughters had FM6 containing siblings collected and used to establish a bb/FM6 stock. This method isolated the novel rDNA deletion allele, bbZ9, used in this study. Drosophila Genetics. All Drosophila lines used in this study are found in SI Appendix, Table S2. All animals we reared on standard Bloomington medium at 25°. All aging was done in roughly 1:1 mixed presence of males and females, provided fresh food every 4 to 6 d. UAS-R2 RNAi strains were designed using SNAP-DRAGON shRNA target software, and oligos containing target hairpin sequence were cloned into the WALLIUM20 vector for phiC31 site-directed integration into the Drosophila genome for expression of a short hairpin to create endogenous miRNAs (50). The target sequence for the R2i-1 construct is 1481-CCGGTTGAACTCATCAATCAA-1502. The target sequence for the R2i-2 construct is 432-CCAGACGAACTTGATGAAGAA-453. All UAS-R2 transgenes were synthesized into pUAST:attB by VectorBuilder (Chicago, IL) for site-directed integration. Importantly, these target sequences were specifically designed to target the R2 ORF encoding the R2 retrotransposase. ORF-containing R2 mRNA are expected to be translated in the cytoplasm, where it is subjected to silencing by the canonical RNAi mechanism. The UAS-R2 transgene contains the R2 ORF tagged with 3xFLAG tag at N terminus, cloned into the pUAST:attB vector. The UAS- R2 transgene also contains sense mutations at the R2i-1 target sequence to render the transgene insensitive to this RNAi (1481-CCGGTTGAACTCATCAATCAA-1502 to 1481-ACGTCTTAATAGCAGTATTAA-1502. The Nuclease-Dead UAS-R2 transgene is identical to the UAS-R2 transgene except for 3001-AAACCAGAC-3009 to GCC and 3097-AAAATCAATAGA-3108 to 3097-GCCATCAATGCC-3108, which are analogous mutations to those demonstrated to disrupt B. mori R2 endonucelase activity (51). All injections and selection of animals containing integrated transgenes were performed by BestGene, Inc (Chino Hills, CA). rDNA Magnification and Heritability Assays. Males containing the bbZ9 allele were mated in bulk to bb158/FM6, Bar females. bbZ9/bb158 female offspring were selected based on the absence of the Bar dominant marker, and scored for cuticular phenotype. To determine the frequency of heritability of magnified offspring, unmated bbZ9/bb158 female F1 animals with wild-type cuticles were collected and individually mated with 3 bb158/Y males. Since homozygous bb158 animals are lethal, all viable female F2 animals are bbZ9/bb158 and were scored for cuticular phenotype. Each individual F1 animal was scored for their ability to produce any offspring with wildtype cuticles, and for the percentage of their F2 female offspring to have wildtype cuticles. Lineage Survival Assay. Independent lineages of nos-gal4/CyO; UAS-R2 RNAi/Tm6B or nos-gal4/CyO; TM2/TM6B animals were established by collecting siblings of the indicated genotypes from nos-Gal4/CyO;TM2/TM6B males mated to Sp/CyO;UAS-R2 RNAi/TM6B females. At each generation in each lineage, three males were mated with five females for 5 d, and offspring were collected 10 d after mated ended. Any lineages that did not have at least three males and five females at collection time were terminated due to insufficient animals. Statistics. For all comparisons of percentage of samples with categorical values (percent γH2Av or R2 positive cells), significance was determined by chi-squared test, and error bars were generated using the Confidence Interval for a Population Proportion formula. For all comparisons of samples with independent values (number of GSCs; rDNA copy number), significance was determined by Student’s t test between experimental and control conditions, unless otherwise indicated. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. We thank the Bloomington Drosophila Stock Center, Kyoto Drosophila Stock Center and Developmental Studies Hybridoma Bank for reagents. We thank the Yamashita lab members and Dr. Andy Clark for discussion and comments on the manuscript. We thank American Type Culture Collection (ATCC) for design of the 28S and RpL ddPCR assays. This research was supported by the Howard Hughes Medical Institute and the John Templeton Foundation. J.O.N. was supported by an American Cancer Society Postdoctoral Fellowship (133949-PF-19-133-01-DMC). 1. 2. 3. 4. 5. 6. 7. 8. 9. A. F. Palazzo, E. S. Lee, Non-coding RNA: What is functional and what is junk? Front. Genet. 6, 2 (2015). T. Kobayashi, Ribosomal RNA gene repeats, their stability and cellular senescence. Proc. Jpn. Acad. Ser B 90, 119–129 (2014). B. L. Strehler, M.-P. Chang, L. K. Johnson, Loss of hybridizable ribosomal DNA from human post- mitotic tissues during aging: I. Age-dependent loss in human myocardium. Mech. Ageing Dev. 11, 371–378 (1979). R. Johnson, B. L. Strehler, Loss of genes coding for ribosomal RNA in ageing brain cells. Nature 240, 412–414 (1972). D. O. Warmerdam, R. M. F. Wolthuis, Keeping ribosomal DNA intact: A repeating challenge. Chromosome Res. 27, 57–72 (2019). J. O. Nelson, G. J. Watase, N. Warsinger-Pepe, Y. M. Yamashita, Mechanisms of rDNA copy number maintenance. Trends Genet. 35, 734–742 (2019). K. L. Lu, J. O. Nelson, G. J. Watase, N. Warsinger-Pepe, Y. M. Yamashita, Transgenerational dynamics of rDNA copy number in Drosophila male germline stem cells. Elife 7, e32421 (2018). F. M. Ritossa, Unstable redundancy of genes for ribosomal RNA. Proc. Natl. Acad. Sci. U.S.A. 60, 509–516 (1968). K. D. Tartof, Regulation of ribosomal RNA gene multiplicity in Drosophila melanogaster. Genetics 73, 57–71 (1973). 16. T. H. Eickbush, D. G. Eickbush, Integration, regulation, and long-term stability of R2 retrotransposons. Microbiol. Spectr. 3, 1127–1146 (2015). 17. K. K. Kojima, H. Fujiwara, Long-term inheritance of the 28S rDNA-specific retrotransposon R2. Mol. Biol. Evol. 22, 2157–2165 (2005). 18. J. Yang, H. S. Malik, T. H. Eickbush, Identification of the endonuclease domain encoded by R2 and other site-specific, non-long terminal repeat retrotransposable elements. Proc. Natl. Acad. Sci. U.S.A. 96, 7847–7852 (1999). 19. D. G. Eickbush, T. H. Eickbush, R2 retrotransposons encode a self-cleaving ribozyme for processing from an rRNA cotranscript. Mol. Cell Biol. 30, 3142–3150 (2010). 20. T. Sultana, A. Zamborlini, G. Cristofari, P. Lesage, Integration site selection by retroviruses and transposable elements in eukaryotes. Nat. Rev. Genet. 18, 292–308 (2017). 21. J. L. Jakubczak, Y. Xiong, T. H. Eickbush, Type I (R1) and type II (R2) ribosomal DNA insertions of Drosophila melanogaster are retrotransposable elements closely related to those of Bombyx mori. J. Mol. Biol. 212, 37–52 (1990). 22. L. Krug et al., Retrotransposon activation contributes to neurodegeneration in a Drosophila TDP-43 model of ALS. PLoS Genet. 13, e1006635 (2017). 23. D. L. Lindsley, L. Sandler, The meiotic behavior of grossly deleted X chromosomes in Drosophila melanogaster. Genetics 43, 547–563 (1958). 24. F. M. Ritossa, K. C. Atwood, S. Spiegelman, A molecular explanation of the bobbed mutants of 10. K. D. Tartof, Unequal mitotic sister chromatid exchange as the mechanism of ribosomal RNA gene Drosophila as partial deficiencies of “ribosomal” DNA. Genetics 54, 819–34 (1966). magnification. Proc. Natl. Acad. Sci. U.S.A. 71, 1272–1276 (1974). 25. S. Gangloff, H. Zou, R. Rothstein, Gene conversion plays the major role in controlling the stability of 11. T. Kobayashi, Strategies to maintain the stability of the ribosomal RNA gene repeats. Genes Genet. large tandem repeats in yeast. EMBO J. 15, 1715–1725 (1996). Syst. 81, 155–161 (2006). 26. S. Paredes, K. A. Maggert, Expression of I-CreI endonuclease generates deletions within the rDNA of 12. R. S. Hawley, K. D. Tartof, The effect of mei-41 on rDNA redundancy in Drosophila melanogaster. Drosophila. Genetics 181, 1661–1671 (2009). Genetics 104, 63–80 (1983). 27. R. S. Hawley, C. H. Marcus, Recombinational controls of rDNA redundancy in Drosophila. Annu. Rev. 13. R. S. Hawley, C. H. Marcus, M. L. Cameron, R. L. Schwartz, A. E. Zitron, Repair-defect mutations inhibit rDNA magnification in Drosophila and discriminate between meiotic and premeiotic magnification. Proc. Natl. Acad. Sci. U.S.A. 82, 8095–8099 (1985). Genet. 23, 87–120 (1989). 28. C. Smelick, S. Ahmed, Achieving immortality in the C. elegans germline. Ageing Res. Rev. 4, 67–82 (2005). 14. M. T. Fuller, A. C. Spradling, Male and female Drosophila germline stem cells: Two versions of 29. R. E. Brown, C. H. Freudenreich, Structure-forming repeats and their impact on genome stability. immortality. Science 316, 402–404 (2007). Curr. Opin. Genet. Dev. 67, 41–51 (2021). 15. G. J. Watase, J. O. Nelson, Y. M. Yamashita, Nonrandom sister chromatid segregation mediates rDNA 30. H. H. KazazianJr., J. V. Moran, Mobile DNA in health and disease. N. Engl. J. Med. 377, 361–370 copy number maintenance in Drosophila. Sci. Adv. 8, eabo4443 (2022). (2017). 8 of 9 https://doi.org/10.1073/pnas.2221613120 pnas.org 31. L. Sinzelle, Z. Izsvák, Z. Ivics, Molecular domestication of transposable elements: From detrimental 41. N. Warsinger-Pepe, D. Li, Y. M. Yamashita, Regulation of nucleolar dominance in Drosophila parasites to useful host genes. Cell Mol. Life Sci. 66, 1073–1093 (2009). melanogaster. Genetics 214, 991–1004 (2020). 32. R. L. Cosby, N.-C. Chang, C. Feschotte, Host–transposon interactions: Conflict, cooperation, and 42. J. C. Peng, G. H. Karpen, H3K9 methylation and RNA interference regulate nucleolar organization cooption. Gene Dev. 33, 1098–1116 (2019). and repeated DNA stability. Nat. Cell Biol. 9, 25–35 (2007). 33. M. Nowacki et al., A functional role for transposases in a large eukaryotic genome. Science 324, 43. F. He, A. James, H. Raje, H. Ghaffari, P. DiMario, Deletion of Drosophila Nopp140 induces subcellular 935–938 (2009). ribosomopathies. Chromosoma 124, 191–208 (2015). 34. B. S. Young, A. Pession, K. L. Traverse, C. French, M. L. Pardue, Telomere regions in drosophila share 44. S. Ide, T. Miyazaki, H. Maki, T. Kobayashi, Abundance of ribosomal RNA gene copies maintains complex DNA sequences with pericentric heterochromatin. Cell 34, 85–94 (1983). genome integrity. Science 327, 693–696 (2010). 35. M.-L. Pardue, P. G. DeBaryshe, Drosophila telomeres: A variation on the telomerase theme. Fly 2, 45. Y. Takeuchi, T. Horiuchi, T. Kobayashi, Transcription-dependent recombination and the role of fork 101–110 (2008). collision in yeast rDNA. Gene Dev. 17, 1497–1506 (2003). 36. K. L. Traverse, M. L. Pardue, A spontaneously opened ring chromosome of Drosophila melanogaster 46. B. S. Heyer, A. MacAuley, O. Behrendtsen, Z. Werb, Hypersensitivity to DNA damage leads to has acquired He-T DNA sequences at both new telomeres. Proc. Natl. Acad. Sci. U.S.A. 85, 8116–8120 (1988). increased apoptosis during early mouse development. Gene Dev. 14, 2072–2084 (2000). 47. J. Cheng et al., Centrosome misorientation reduces stem cell division during ageing. Nature 456, 37. T. Kahn, M. Savitsky, P. Georgiev, Attachment of HeT-A Sequences to chromosomal termini in Drosophila melanogaster may occur by different mechanisms. Mol. Cell Biol. 20, 7634–7642 (2000). 599–604 (2008). 48. A. M. Huang, E. J. Rehm, G. M. Rubin, Quick Preparation of Genomic DNA from Drosophila. Cold Spring Harb. Protoc. 2009, pdb.prot5198 (2009). 38. B. Saint-Leandre, S. C. Nguyen, M. T. Levine, Diversification and collapse of a telomere elongation 49. M. Jagannathan, N. Warsinger-Pepe, G. J. Watase, Y. M. Yamashita, Comparative analysis of satellite mechanism. Genome Res. 29, 920–931 (2019). 39. B. Czech et al., piRNA-guided genome defense: From biogenesis to silencing. Annu. Rev. Genet. 52, DNA in the Drosophila melanogaster species complex. G3 (Bethesda) 7, 693–704 (2017). 50. J.-Q. Ni et al., A genome-scale shRNA resource for transgenic RNAi in Drosophila. Nat. Methods 8, 131–157 (2018). 405–407 (2011). 40. J. A. George, T. H. Eickbush, Conserved features at the 5′ end of Drosophila R2 retrotransposable elements: Implications for transcription and translation. Insect Mol. Biol. 8, 3–10 (1999). 51. A. Govindaraju, J. D. Cortez, B. Reveal, S. M. Christensen, Endonuclease domain of non-LTR retrotransposons: Loss-of-function mutants and modeling of the R2Bm endonuclease. Nucleic Acids Res. 44, 3276–3287 (2016). PNAS 2023 Vol. 120 No. 23 e2221613120 https://doi.org/10.1073/pnas.2221613120 9 of 9
10.1073_pnas.2306965120	RESEARCH ARTICLE \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS IL- 6 trans- signaling in a humanized mouse model of scleroderma , Kriti Agrawalc,d, Esen Sefikb, Anahi V. Odellb, Elizabeth Cavese, Nancy C. Kirkiles- Smithb, Valerie Horsleya,e, Monique Hinchclifff, Ian D. Odella,b Jordan S. Pobera,b,g , Yuval Klugerc,d,g , and Richard A. Flavellb,h,1 Contributed by Richard A. Flavell; received April 27, 2023; accepted July 31, 2023; reviewed by Shervin Assassi and Thomas Krieg Fibrosis is regulated by interactions between immune and mesenchymal cells. However, the capacity of cell types to modulate human fibrosis pathology is poorly understood due to lack of a fully humanized model system. MISTRG6 mice were engineered by homologous mouse/human gene replacement to develop an immune system like humans when engrafted with human hematopoietic stem cells (HSCs). We utilized MISTRG6 mice to model scleroderma by transplantation of healthy or scleroderma skin from a patient with pansclerotic morphea to humanized mice engrafted with unmatched allogeneic HSC. We identified that scleroderma skin grafts contained both skin and bone marrow–derived human CD4 and CD8 T cells along with human endothelial cells and pericytes. Unlike healthy skin, fibroblasts in scle- roderma skin were depleted and replaced by mouse fibroblasts. Furthermore, HSC engraftment alleviated multiple signatures of fibrosis, including expression of collagen and interferon genes, and proliferation and activation of human T cells. Fibrosis improvement correlated with reduced markers of T cell activation and expression of human IL- 6 by mesenchymal cells. Mechanistic studies supported a model whereby IL- 6 trans- signaling driven by CD4 T cell–derived soluble IL- 6 receptor complexed with fibroblast- derived IL- 6 promoted excess extracellular matrix gene expression. Thus, MISTRG6 mice transplanted with scleroderma skin demonstrated multiple fibrotic responses centered around human IL- 6 signaling, which was improved by the presence of healthy bone marrow–derived immune cells. Our results highlight the importance of IL- 6 trans- signaling in pathogenesis of scleroderma and the ability of healthy bone marrow–derived immune cells to mitigate disease. scleroderma \| systemic sclerosis \| fibrosis \| humanized mice \| interleukin 6 Scleroderma is an umbrella term that incorporates patients with systemic sclerosis (SSc), an immune- mediated disease that causes fibrosis and vasculopathy of the skin and internal organs (1, 2), and morphea (also called localized scleroderma), that causes skin fibrosis without internal organ involvement. Skin gene expression studies of scleroderma patients showed similar expression profiles between morphea and an inflammatory subset of SSc, suggesting shared pathogenic processes (3). The most severe form of localized scleroderma is pansclerotic morphea, which causes fibrosis of all layers of the skin, and the childhood form was recently found to have elevated IL- 6 expression reg- ulated by STAT4 gain of function mutations (4). As with most human autoimmune diseases, scleroderma has complex genetic contribution (5–8), and its pathogenesis involves several immune and mesenchymal cell types. Immune cells such as monocytes (9, 10), plasmacytoid dendritic cells (11, 12), and type 2 innate lymphoid cells (13) communicate by incompletely understood mechanisms with fibroblasts and pericytes to regulate expression of collagen and other extracellular matrix (ECM) genes. A better mechanistic understanding of scleroderma requires in vivo disease models that can emulate the cellular composition and regulatory signals that occur in human patients. Investigation of scleroderma in mice is hindered by a lack of mouse models that fully recapitulate the complexity of human disease. To model the human immune system in mice, human- specific growth factors and cytokines are required for proper immune cell develop- ment from the bone marrow. We generated the MISTRG6 humanized mouse strain (14) to provide physiologic expression of human CSF1 (monocytes and tissue macrophage devel- opment) (15), CSF2/IL3 (lung alveolar macrophages) (16), SIRPA (macrophage tolerance to human cells) (17), THPO (hematopoiesis and platelets) (18), and IL6 (improved engraft- ment and antibody responses) (19) knocked into their respective mouse loci on a Rag2- /- Il2rg- /- background, thereby permitting development of both innate and adaptive lineages from engrafted human hematopoietic stem cells (HSCs). As both adaptive and innate immune cells are important regulators of fibrosis, the development of a comprehensive Significance Scleroderma is an autoimmune disease that causes skin and internal organ fibrosis. No mouse model has been identified that fully recapitulates human disease. Therefore, we tested whether the MISTRG6 strain of humanized mice could be transplanted with healthy or scleroderma human skin grafts. We found that healthy and scleroderma skin grafts retained skin and bone marrow–derived immune cells. Moreover, fibrosis in scleroderma skin grafts was alleviated by engraftment with unmatched allogeneic hematopoietic stem cells. Mechanistic studies supported a trans- signaling model whereby CD4 T cell–derived soluble interleukin- 6 (IL- 6) receptor complexed with interferon- driven IL- 6 cytokine drives skin fibrosis. Thus, scleroderma skin transplants to MISTRG6 humanized mice recapitulate a key signaling pathway of scleroderma and may be a useful model to study human disease. Author contributions: I.D.O., N.C.K.- S., and R.A.F. designed research; I.D.O., K.A., A.V.O., and N.C.K.- S. performed research; E.C., V.H., and M.H. contributed new reagents/analytic tools; I.D.O., K.A., E.S., and J.S.P. analyzed data; and I.D.O., K.A., E.S., M.H., J.S.P., Y.K., and R.A.F. wrote the paper. Reviewers: S.A., University of Texas Health Science Center at Houston; and T.K., Universitat zu Koln. Competing interest statement: R.A.F. is a consultant for GSK and Zai Lab Ltd. I.D.O. and R.A.F. are founders of Plythera, Inc. I.D.O. has research support from Ventus Pharmaceuticals. The other authors declare no competing interests. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. Published September 5, 2023. PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 1 of 12 human immune repertoire by MISTRG6 mice may provide a highly suited system to model those interactions in scleroderma. To fully model immune–mesenchymal interactions in mice, HSC- engrafted mice require the presence of human tissue, such as by transplantation of human skin. However, skin grafts alone do not contain a full panoply of bone marrow–derived human immune cells, which may be important regulators of disease. We hypothe- sized that engrafting both human HSC and skin to MISTRG6 mice would allow examination of key immune–mesenchymal interac- tions in scleroderma. We identified that MISTRG6 mice could be stably engrafted with human skin and contained a broad range of human immune and mesenchymal cell types. Skin grafts from a patient with pansclerotic morphea to HSC- engrafted MISTRG6 mice demonstrated that healthy bone marrow–derived T cells were able to improve pathologic signatures of fibrosis including ECM gene expression and T cell activation. These findings underscore the importance of bone marrow–derived immune cells in regulating fibrotic skin disease. Results Healthy Human Skin Grafting onto MISTRG Mice. Acceptance of human HSC and vascularized skin grafts varies by immunodeficient mouse strains, in particular, whether innate immune responses reject certain cell types and/or tissues (20). To determine what human cell types are retained in skin grafts to MISTRG mice, we transplanted split- thickness skin grafts (STSG) of healthy human skin onto the backs of 8- wk- old MISTRG mice. Staples were removed after 14 d, and the grafts were allowed to heal and mature for a total of 7 wk. As shown in Fig. 1A, two skin grafts were transplanted to each mouse and were readily apparent by their brown pigmentation because the skin donor was of African American ancestry. The skin grafts appeared viable and intact on histology (Fig. 1B). Typical findings of STSG were observed including epidermal hyperplasia, absence of hair follicles, and degeneration of sebaceous and eccrine glands (21). Also typical of STSG, modest contracture (22) of the graft area occurred. A C hCD31 100 µm D hCD146 B 500 µm mCD31 merge hCD31 merge 100 µm E mCD34 hCD34 100 µm merge hCD45 merge hCD11c merge 100 µm mCD45 F 100 µm G mCD11c 100 µm Fig. 1. Human skin grafting onto MISTRG mice. (A) MISTRG mice were transplanted with STSG to the dorsal flanks, visible by the brown pigmentation. (B) Histology of a representative skin graft after transplantation for 7 wk. (C–G) Immunofluorescence images of skin grafts with species- specific antibodies against endothelial cells (C), pericytes (D), fibroblasts (E), and immune cells (F and G). Slides were imaged with a Keyence BZ- X800 microscope (B and E) or Leica DMI6000 B Microscope (C, D, F, and G). Images are all 10× magnification, and lower power images in B and E were digitally stitched together. 2 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org Overall, skin grafts from healthy human skin to MISTRG mice showed good viability and stability in the first 7 wk. To ascertain what human and mouse cell types were present in the skin grafts, we obtained immunofluorescence images using human and mouse- specific antibodies. Both human and mouse vasculature were present in skin graft dermis by the presence of CD31+ endothelial cells (Fig. 1C). Mouse vessels coursed through the entire thickness of the dermis intermixed between the human blood vessels. Human vessels retained their associated pericytes detected by CD146+ staining (Fig. 1D). The graft dermis also showed the presence of human fibroblasts by CD34+ positivity, which was sharply demarcated from underlying murine fascia (Fig. 1E). Mouse myeloid immune cells infiltrated both the dermal and epidermal compartments of the skin graft (Fig. 1F), which included CD11c+ dendritic cells and macrophages (Fig. 1G). However, we did not observe human immune cells within skin grafts to MISTRG mice by immunofluorescence imaging. In sum- mary, healthy human skin grafts to MISTRG mice are viable and retain multiple human mesenchymal cell populations including fibroblasts, endothelial cells, and pericytes alongside multiple murine structural and immune cell types. Healthy Human Skin Retains T Cells Long Term in MISTRG6 Mice. The presence of the human IL- 6 gene in the MISTRG strain of humanized mice (MISTRG6) dramatically improves HSC engraftment and enhances B cell development (19). We considered whether human IL- 6 would modify skin graft cellular constituents and graft longevity. We transplanted STSG of healthy human skin from a second healthy donor this time to MISTRG6 mouse recipients, and mice were allowed to age up to an additional 9 mo after transplantation. Even after 9 mo, skin grafts were readily visible by the naked eye and contained some of the original skin pigmentation (Fig. 2A). To assess immune and mesenchymal cells present in the skin grafts, we digested the grafts to generate single- cell suspensions and analyzed the cellular components by fluorescence- activated cell sorting (FACS) compared to MISTRG6 mouse control skin. In contrast to MISTRG mice, skin grafts on MISTRG6 mice retained human immune cells in the skin, which consisted almost entirely of T cells (Fig. 2B). Given their long- term retention in human skin and absence in control MISTRG6 mouse skin, the T cells were likely tissue resident. Accordingly, at least ¼ of the T cells showed positive CD103 staining on FACS, as CD103 mediates adhesion to the skin epidermis. In addition to human T cells, skin grafts retained human fibroblasts, detected by CD26 and CD34 staining (Fig. 2C), human endothelial cells marked by CD31 (Fig. 2D), and human pericytes positive for CD146 (Fig. 2E). Therefore, MISTRG6 mice support healthy human skin grafts with long- term retention of human tissue- resident T cells as well as fibroblasts, endothelial cells, and pericytes for at least up to 9 mo after engraftment. A B 5 4 D C m C 6 2 D C m D 1 3 D C m Mouse Skin Human Skin Human Skin hCD45 Human Skin 3 0 1 D C h 4 3 D C m hCD3 Human Skin hCD26 Human Skin hCD34 hCD45 Mouse Skin hCD26 Human Skin 5 4 D C m 6 2 D C m E 6 4 1 D C m hCD31 hCD146 Fig. 2. Healthy human skin retains T cells long term in MISTRG6 mice. (A) MISTRG6 mice were transplanted with STSG to the dorsal flanks and followed for 9 mo. Residual graft is visible by brown pigmentation and highlighted with the dotted line. (B–E) Skin grafts transplanted for 9 mo were digested to create single- cell suspension and analyzed by FACS for human and mouse immune and mesenchymal cell populations. PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 3 of 12 Transplantation of Scleroderma Skin to MISTRG6 Mice. Scleroderma skin has multiple characteristics that could make skin transplantation difficult, such as increased stiffness and dysregulated metabolism (23). We report a patient with adult- onset pansclerotic morphea, a severe subtype of localized scleroderma that affects all layers of the skin but spares internal organs, who was highly motivated to enroll in our study and donate STSG for transplantation to humanized mice. Our patient had long- standing progressive disease despite therapy with prednisone, cyclosporine, cyclophosphamide, thalidomide, tocilizumab, and extracorporeal photopheresis (ECP). Ongoing therapy at the time of skin grafting was monthly sessions of ECP, most recently done 1 wk prior to skin graft. A baseline punch biopsy of his skin is shown in Fig. 3A, demonstrating full- thickness fibrosis of the entire dermal layer. We prepared a cohort of MISTRG6 recipient mice by injecting 19 neonatal MISTRG6 mice intrahepatically each with 10,000 CD34+ HSC (Fig. 3B). Peripheral blood was monitored at 6 and 8 wk postengraftment for immune reconstitution. The six MISTRG6 mice with the most favorable immune engraftment (#8, 11, 13, 15, 16, and 19) were selected as recipients for patient skin grafts (Fig. 3C) and contained a high percentage of the myeloid lineage (Fig. 3D). At 8 wk after HSC engraftment, a 0.75- mm- thick STSG (Fig. 3E) was obtained from the patient’s back, cut into ~1 cm2 squares, and transplanted two per mouse onto the backs of 6 HSC- engrafted MISTRG6 mice and 4 unengrafted control MISTRG6 mice. Because the skin dermis contains spatially distributed populations of fibroblasts (24), we confirmed that profibrotic fibroblasts marked by CD26 were present in the donor skin grafts (Fig. 3F) (25). The skin grafts were allowed to heal and mature for 4 wk (Fig. 3G), at which point the mice were euthanized and skin grafts analyzed for histology and scRNA- Seq. The effects of HSC engraftment in combination with skin trans- plantation could each contribute to the presence of circulating and tissue immune cells. To interrogate circulating immune cells, we performed FACS analysis of peripheral blood of HSC- engrafted (+HSC) MISTRG6 mice compared to unengrafted (−HSC) con- trol MISTRG6 mice. We observed circulating myeloid and adap- tive immune subsets only in the HSC- engrafted mice which were absent in all the unengrafted controls (Fig. 3H). This signifies that human skin graft–derived immune cells did not detectably pop- ulate the blood. To investigate the presence of immune cells in the skin tissue, we assessed skin histology and immunofluorescence images of skin grafts. By histology, scleroderma skin grafts on HSC- unengrafted MISTRG6 mice showed a prominent cellular infiltrate in the dermis similar to what we observed with healthy skin on MISTRG mice (Fig. 3 I, Upper). In contrast, scleroderma skin grafts of HSC- engrafted mice showed less cellularity, suggest- ing fewer numbers of one or more cell types (Fig. 3 I, Lower). Nonetheless, human scleroderma skin of both HSC- engrafted and - unengrafted mice contained numerous human and mouse immune cells labeled by CD45 (Fig. 3J). This observation suggests that like healthy skin grafts, scleroderma skin grafts retained res- ident human immune cells in MISTRG6 mice. Therefore, scle- roderma skin grafts contained a chimeric assortment of mouse and human immune cells of skin and bone marrow origin. In contrast, skin- derived human immune cells did not circulate in the blood to an appreciable level. scRNA- Seq Analysis of Transplanted Scleroderma Skin. To understand how skin and bone marrow–derived human immune cells alter the cellular signaling in scleroderma skin grafts, we performed scRNA- Seq on 3 of the HSC- engrafted (+HSC) compared to 3 HSC- unengrafted (−HSC) scleroderma skin grafts 4 wk after transplantation onto MISTRG6 humanized mice. Immediately after euthanasia of the mice, the skin grafts were excised and digested to create single- cell suspensions. Live cells were sorted and processed with the 10× Chromium Single Cell Controller to create barcoded single- cell cDNA libraries, and sequencing reads were aligned to mouse and human genomes. Because MISTRG6 mice express human transgenes by mouse cells, we also needed to align reads to a combined human and mouse genome to assess expression of human genes by mouse cells. Uniform Manifold Approximation and Projection (UMAP) embedding of the single- cell cDNA identified 5 cell clusters of human cells, including CD4 and CD8 T cells, monocytes, pericytes, and a population of proliferating cells (Fig. 4A). Bone marrow–derived T cells could be identified by expression of the long noncoding RNA XIST due to the sex mismatch of HSC and skin donors (female HSC, male scleroderma skin) (Fig. 4B). This showed that 20% of the human CD4 and CD8 T cells in the HSC- engrafted mice were bone marrow derived, whereas the other 80% were retained from the original human skin. Unexpectedly, no human fibroblasts were detected in our scRNA- Seq data of scleroderma skin grafts independent of HSC engraftment. The absence of human fibroblasts in scleroderma skin grafts was corroborated by immunofluorescence imaging of scleroderma skin grafts before and after transplantation using human and mouse PDGFRA antibodies (Fig. 4C). Prior to skin transplantation, human fibroblasts were apparent in the scleroderma donor skin, whereas 4 wk posttransplantation, scleroderma skin grafts were largely devoid of human fibroblasts with only rare human PDGFRA positive cells (arrowheads) and were replaced by mouse fibroblasts. Thus, scleroderma skin grafts retained human CD4 and CD8 T cells and pericytes but lost human fibroblasts. In the scRNA- Seq data that aligned to the mouse genome, we observed 15 clusters of mouse cells (Fig. 4D). Murine cell types included two clusters of fibroblasts as well as pericytes, endothelial cells (both vascular and lymphatic), monocytes, and macrophages, and smaller groups of neurons, Schwann cells, muscle, and mast cells. In the scleroderma skin grafts, mouse cells predominated in number with 13,805 mouse cells compared to 1,057 human cells. Engraftment with human HSC associated with different propor- tions of mouse cell populations in scleroderma skin grafts. Scleroderma skin grafts of HSC- engrafted mice showed 70% fewer mouse monocytes and macrophages and a new population of mouse erythroid cells. Although the erythroid cells were merely due to anemia secondary to HSC engraftment (14), the alterations in myeloid populations may reflect differences in immune–fibro- blast signaling. Altogether, our findings demonstrated that human scleroderma skin grafts contained human resident and bone mar- row–derived T cells and were also infiltrated with mouse mesen- chymal and myeloid immune cells. To identify gene expression modules altered in scleroderma skin grafts by bone marrow–derived immune cells in an unbiased man- ner, we analyzed our scRNA- Seq data for differentially abundant (DA) cell subpopulations using DA- Seq (26). This approach detects cell subpopulations whose abundance differs between two states. We observed two DA regions in the human cells and seven DA regions in the mouse cells. Among human T cells, DA region 1 highlighted an increased abundance of skin CD4 T cells in HSC- engrafted mice in blue, which contrasts with relatively more abundant CD8 T cells in HSC- unengrafted mice in red (Fig. 4E). Human DA region 2 highlighted proliferating T cells, which may reflect local expansion of T cells in HSC- unengrafted mice. To understand what genes might be driving these observations, we calculated differential gene expression of HSC- engrafted vs. - unengrafted human CD4 and CD8 T cells. We identified decreased 4 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org A E F G I 500 μm 500 μm hCD26 250 μm B C C H -HSC D D 5 4 D C m +HSC 9 1 D C h 8 D C h -HSC 250 μm +HSC hCD45 hCD33 hCD4 J -HSC mCD45 hCD45 merge 100 μm +HSC mCD45 hCD45 merge 250 μm 100 μm Fig. 3. Transplantation of scleroderma skin to MISTRG6 mice. (A) Histology of punch biopsy from scleroderma skin donor. (B) Diagram of the experimental setup. (C) Human peripheral blood immune cell engraftment levels of 8- wk- old prospective MISTRG6 skin transplant recipients. (D) Peripheral blood immune cell proportions of highest engrafted MISTRG6 mice. (E) Histology of STSG from the scleroderma donor at the same magnification as A. (F) Immunofluorescence imaging of STSG with the fibroblast marker CD26. (G) Scleroderma skin grafts on euthanized MISTRG6 mouse at completion of the experiment. (H) Peripheral blood human immune cells of MISTRG6 mice with scleroderma skin comparing HSC engrafted (+HSC) and unengrafted (−HSC) mice. (I) Histology of scleroderma skin grafts to HSC- unengrafted and - engrafted MISTRG6 mice. (J) Scleroderma skin graft immune cell staining by CD45 comparing HSC- engrafted and - unengrafted mice. Histology and immunofluorescent images were obtained with a Keyence BZ- X800 microscope. Low- power images are 10× magnification and stitched together. PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 5 of 12 A D +HSC -HSC B XIST C donor skin post-transplant hPDGFRa T cells XIST pos: 102 XIST neg: 409 40 (cid:31)m mPdgfra 40 (cid:31)m +HSC -HSC merge E DAseq F 2 +HSC -HSC 1 H HLA-DRB1 ** * HLA-DQA1 ns CD4 CD8 CD4 CD8 DAseq G +HSC -HSC 2 m D i +HSC -HSC 3 7 4 6 5 1 2 Dim1 Log2(fold change) -1 0 1.5 Extracellular Matrix Genes I ** Human IL-6 ns * **** Interferon Signature Genes IGF-related Genes J Fibroblast 1 Fibroblast 2 Endothelial Pericyte +HSC -HSC Fig. 4. scRNA- Seq analysis of transplanted scleroderma skin. (A) UMAP embedding of human cell populations in scleroderma skin grafts comparing HSC engrafted (+HSC) to unengrafted mice. (B) Human XIST expression of HSC- engrafted mice to distinguish bone marrow–derived immune cells. (C) Immunofluorescence images of scleroderma skin graft before and after transplantation of human and mouse PDGFRA. Arrowheads indicate human fibroblasts in right panels. (D) UMAP embedding of mouse cell populations in scleroderma skin grafts. (E) DAseq regions of human cells. (F) Expression of MHC class II genes by human CD8 T cells. (G) DAseq regions of mouse cells. (H) Heatmap of differentially expressed genes with P value <0.05 by mouse fibroblasts, pericytes, endothelial cells, monocytes, and macrophages. Fold change refers to ratio unengrafted/HSC engrafted. (I) Violin plot of human IL- 6 expression by mouse mesenchymal cell types. (J) Model of human T cell interactions with mouse myeloid and mesenchymal cells. IL- 6 expression by multiple cell types induces expression of MHC class II molecules on human T cells along with elevated ECM expression, which is down- regulated by healthy bone marrow–derived immune cells. Data in F, H, and I are analyzed with a nonparametric Wilcoxon rank- sum test adjusted with Bonferroni correction using all features in the dataset, with P < 0.05, P < 0.01, P < 0.001, and *P < 0.0001. Immunofluorescent images were obtained with a Keyence BZ- X800 microscope at 40× magnification. 6 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org expression of MHC class II genes by both CD4 and CD8 T cells in HSC- engrafted compared to - unengrafted mice (Fig. 4F). The differential expression of these genes was not affected by the presence of bone marrow–derived (XIST positive) T cells. Notably, the dif- ferentially expressed MHC class II alleles we observed are significant predictors of mortality in scleroderma patients (27). Although the function of MHC class II expression by human T cells is not entirely clear, it is a well- established marker of T cell activation (28). Thus, loss of MHC II expression by scleroderma T cells suggests decreased T cell activation in HSC- engrafted samples. Altogether, our results suggest that activation of scleroderma skin–derived CD4 and CD8 T cells is decreased by HSC engraftment. Among mouse cells, DA regions of HSC- engrafted mice high- lighted a decrease in monocytes, macrophages, and endothelial cells (regions 1 to 3) and a region of increased fibroblasts (region 7) (Fig. 4G). Proliferating cells were again highlighted in HSC- unengrafted mice (region 4), as well as erythroid cells in HSC- engrafted mice (region 6). NK cells in region 5 were highlighted, but these were disproportionally present in only one of the three scleroderma skin grafts, so were excluded from further analysis. To understand the gene expression changes associated with the DA regions, we calculated differential gene expression of cells from skin grafts of HSC- engrafted vs. - unengrafted mice followed by pathway analysis. In HSC- unengrafted mice, we observed significantly up- regulated expression of the myofibroblast marker aSMA (Acta2) by fibroblasts along with increased expression of many ECM genes by fibroblasts, pericytes, and endothelial cells (Fig. 4H). These same cell types showed upregulation of insulin growth factor related genes, another profibrotic signaling pathway (29). Multiple cell types in skin grafts of HSC- unengrafted mice, including fibroblasts, peri- cytes, endothelial cells, monocytes, and macrophages, also showed elevated expression of interferon signature genes, which are associ- ated with scleroderma severity (30, 31). Therefore, scleroderma skin grafts of HSC- unengrafted mice showed a fibrotic expression signa- ture by murine cells including higher ECM expression, growth factor signaling, and interferon response. We considered what signal might be driving the array of cellular alterations in scleroderma skin, including activation and prolifer- ation of human T cells along with elevated ECM and ISG expres- sion. The profibrotic cytokine IL- 6 has been linked to each of these observations including induction of T cell proliferation and activation (32), induction of collagen expression by dermal fibro- blasts (33), and regulation by type I interferon (34). IL- 6 is ele- vated in the skin and blood of patients with localized scleroderma and SSc (35, 36) and was recently identified to be overexpressed by fibroblasts of patients with childhood- onset pansclerotic mor- phea (4). We therefore assessed the effects of HSC engraftment on human IL- 6 expression in the scleroderma xenografts. We found that HSC- unengrafted mice had significantly higher expres- sion of human IL- 6 by mouse fibroblasts, endothelial cells, and pericytes (Fig. 4I), which corresponds with their activated and proliferating T cells and elevated ECM expression compared to HSC- engrafted mice. Elevated expression of human IL- 6 thus links the T cell and mesenchymal changes characteristic of fibrotic scleroderma skin. Altogether, the up- regulated transcriptional programs in scleroderma skin of HSC- unengrafted compared to HSC- engrafted mice demonstrated a more fibrotic program through elevated expression of multiple fibrosis signature genes, including aSMA, ECM, and interferon response, which all center around elevated expression of human IL- 6 (Fig. 4J). IL- 6 Trans- Signaling between CD4 T Cells and Fibroblasts. To evaluate how IL- 6 signaling may drive fibrosis in scleroderma skin grafts, we investigated the effects of T cell activation on expression of MHC class II and IL6 receptor alpha subunit (IL6Ra). We focused on CD4+ T cells because they were previously shown to shed soluble IL6Ra (sIL6Ra) (37). We activated peripheral blood CD4+ T cells with CD3/CD28 microbeads for three days and measured surface expression of MHC class II and membrane- bound IL6Ra (mIL6Ra) by FACS. We found that activated CD4+ T cells up- regulated both HLA- DR and IL6Ra on their cell surface compared to unstimulated CD4+ T cells (Fig. 5 A and B). sIL6Ra was detectable in the T cell supernatant of activated CD4+ T cells, but not unstimulated controls, demonstrating that IL6Ra is cleaved from CD4+ T cell surface (Fig. 5C). Shedding of IL6Ra from the cell surface is a characteristic feature of IL- 6 trans- signaling, in which sIL6Ra binds the IL- 6 cytokine and then complexes with gp130 on target cells to induce an inflammatory response (38). Our results confirm the association of T cell activation with upregulation of MHC class II and suggest CD4+ T cells as a source of sIL6Ra to activate IL- 6 trans- signaling. We considered whether IL- 6 trans- signaling promotes fibrosis more than IL- 6 cytokine alone using cultured fibroblasts. We first investigated which IL- 6 signaling components are expressed in different primary fibroblast sources. We measured IL6, IL6R (sol- uble isoform sIL6R and transmembrane isoforms tmIL6R), and IL6ST (gp130) expression in human foreskin fibroblasts (HFF), adult normal dermal fibroblasts (NDF), and SSc fibroblasts. SSc fibroblasts were isolated from three patients with limited cutaneous SSc of 13 (SSc1), 4 (SSc2), and 10 (SSc3) years disease duration. SSc1 and SSc2 were not treated with immunomodulating therapy, whereas SSc3 was taking mycophenolic acid for SSc interstitial lung disease at the time of skin biopsy. Among fibroblasts, we found that HFF expressed the highest amount of IL6 and IL6ST com- pared to NDF and SSc fibroblasts (Fig. 5D). SSc1 fibroblasts expressed higher levels of sIL6R and tmIL6R than those from SSc2 and SSc3, although not as much sIL6R as HFF. Because HFF recapitulated important features of EGFR signaling in SSc in our previous studies (40) and overall showed the highest expression of IL- 6- related genes, we used these and SSc fibroblasts in subsequent studies to measure their responses to recombinant type I interferon (IFNa2). We found that type I interferon induced IL6 expression by HFF, SSc1, and SSc3 fibroblasts (Fig. 5E). SSc3 fibroblasts also showed higher expression of fibrotic ECM genes TNC and COL1A1 in response to IFNa2, which may reflect their higher fold change in IL6 expression compared to HFF and other SSc fibro- blasts. These results align with our scRNA- Seq data showing ele- vated ISG signature and IL6 expression by fibroblasts in scleroderma skin grafts. We then checked whether recombinant IL- 6 or IL- 6 in combination with sIL6R could induce fibrotic gene expression. We found that in both HFF and SSc1 fibroblasts, the combination of IL- 6 and sIL6Ra was superior at inducing expression of IL6 itself and the profibrotic ECM gene TNC compared to IL- 6 alone (Fig. 5 F and G). To further assess IL- 6 trans- signaling on HFF and SSc fibroblasts, we used an IL- 6- IL6Ra fusion protein termed hyper- IL6 (41). We found that hyper- IL6 significantly induced expression of IL6 and TNC in HFF and SSc fibroblasts (Fig. 5H). Hyper- IL6 also induced type I collagen expression in HFF, but not SSc fibro- blasts. Altogether, our results indicate that IL- 6 trans- signaling can drive expression of multiple ECM genes, which may be regulated by sIL6Ra derived from activated CD4+ T cells (Fig. 5I). In sum- mary, scleroderma skin grafts to humanized mice were character- ized by markers of T cell activation through increased expression of MHC class II genes. Activated CD4+ T cells produce sIL6Ra, and in turn, sIL6Ra can bind IL- 6 to drive excess ECM gene expression by fibroblasts. Thus, IL- 6 trans- signaling appears to be a fundamental component driving fibrosis in our humanized mouse model of scleroderma. PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 7 of 12 Unstimulated Activated B C IL6Ra A - R D A L H D E F SSc1 Fibroblasts G HFF I H Fig. 5. IL- 6 trans- signaling drives ECM expression in fibroblasts. (A–C) CD4+ T cells from peripheral blood were activated with CD3/CD28 Dynabeads for 3 d, n = 3 per group; data are representative of 2 independent experiments. (A) Representative FACS plot of HLA- DR and IL6Ra staining. (B) Quantification of HLA- DR and IL6Ra expression by CD4+ T cells. (C) sIL6Ra levels in CD4 T cell supernatant, n.d. not detected. (D) Relative expression of IL- 6 signaling genes in fibroblasts, n = 3 to 5 per group. (E) IL6 expression by HFF and SSc fibroblasts incubated with IFNa2, n = 7 (HFF) and 4 (SSc1- 3) per group; data were combined from 2 independent experiments. (F and G) IL6 and TNC expression when SSc1 and HFF fibroblasts were incubated with IL6 and/or sIL6Ra for 4 h, n = 4 per group; data are representative from 2 independent experiments. (H) IL6 and ECM gene expression by HFF and SSc fibroblasts in response to hyper- IL6 for 4 h, n = 4 per group. (I) Model of IL- 6 trans- signaling in scleroderma skin grafts. tmIL6Ra is cleaved by ADAM17 (39), thereby generating the soluble receptor. IL6 expression is induced by type I interferon, which can then bind sIL6Ra and gp130 on fibroblasts to drive ECM expression along with IL6 positive feedback loop. Data are mean ± SD (P < 0.05, P < 0.01, P < 0.001, and **P < 0.0001) analyzed with one- way ANOVA with Tukey multiple- comparisons test (D, F, and G) and unpaired two- tailed Student t test (B, E, and H). 8 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org Discussion Humanized mouse models provide an in vivo system to study mech- anisms of complex human diseases. They have been used to model human tropic infections such as hepatitis C (42) and SARS- CoV- 2 (43) as well as understand effects of human immune cells on tumors with patient- derived xenografts (e.g., refs. (44, 45)). Humanized mice also have high potential to better model human autoimmune diseases. Autoimmune diseases such as scleroderma and lupus occur in patients with predisposing genetic risk and a triggering event such as infection or cancer that induces immune dysregulation and disease development (1). To fully encompass these complex traits in mice, human immune cells and tissue are required. We therefore tested whether the MISTRG6 strain of humanized mice could be engrafted with healthy human skin and characterized the immune– mesenchymal interactions when transplanted with scleroderma skin. We found that in MISTRG6, but not MISTRG mice, human skin retained resident human T cells. This difference is likely due to the function of IL- 6 to promote T cell differentiation and improved engraftment in MISTRG6 compared to MISTRG mice (19). Unexpectedly, however, although healthy human skin retained fibroblasts, fibroblasts were depleted from scleroderma skin and were replaced by mouse fibroblasts. Thus, MISTRG6 mice showed differential acceptance of healthy vs. fibrotic human skin xenografts. Healthy human skin grafts retain both human immune and mes- enchymal cells long term, but fibrotic scleroderma skin is depleted of human mesenchymal cells. Generation of humanized mice with skin grafts involves a coor- dinated sequence of events. After perinatal engraftment of mice with human HSC, fresh skin is required 8 to 10 wk later. For healthy skin grafts, we depended on good- quality human skin from discarded mastectomy and abdominoplasty surgical speci- mens. Unpredictable availability of fresh healthy human skin could limit the feasibility of large numbers of experiments. After obtaining human skin, an STSG was cut to reduce the thickness of human skin so that it could approximate that of the mouse recipient and remain viable. Human skin has four fibroblast pop- ulations that have heterogeneous spatial distribution (46, 47). Consequently, STSG of human skin was likely enriched in pap- illary dermal fibroblasts which express CD26 and WNT pathway genes (47), which are key signaling molecules in fibrosis (25, 48). The papillary dermis also contains more T cells than the reticular dermis (47), which may have facilitated their retention in the skin grafts. In sum, transplanting skin to humanized mice requires specific timing of procedures, and STSG may be enriched in T cells and fibroblast populations important in fibrosis. A concern in humanized mice containing HSC and skin from different donors is development of graft- vs.- host disease (GvHD). GvHD occurs when there is HLA or minor histocompatibility anti- gen mismatch between immune cells and tissue. GvHD is charac- terized by elevated proinflammatory cytokines including IL- 4, IL- 6, IL- 17, IL2Ra, and TNF (49), among which IL- 6 has been proposed as key driver of disease (50, 51). To avoid bone marrow biopsy or peripheral mobilization of HSC, which can promote scleroderma (52), we engrafted MISTRG6 mice with unmatched HSC prior to transplantation of scleroderma skin grafts. Although healthy human skin was transplanted with long- term viability in MISTRG6 mice, we observed rejection of fibroblasts from scleroderma skin grafts. Scleroderma fibroblast rejection occurred independently of HSC engraftment, suggesting that it was mediated by mouse rather than human immune cells. Furthermore, engraftment with unmatched HSC reduced human IL- 6 expression, which we would expect to have increased in human immune cell–mediated GvHD. Thus, although mismatched HSCs were used, our findings indicate that rejection of scleroderma fibroblasts occurred by mouse immune cells rather than human ones and suggest a loss of tolerance to a specific cell type independent of HLA. Innate immune checkpoints may influence the capacity of humanized mice to tolerate fibroblasts in scleroderma skin trans- plants. The interaction between SIRPA and CD47 acts as a negative regulatory signal to inhibit macrophage- mediated phagocytosis (53). Survival of fibroblasts in fibrotic tissue requires upregulation of immunoregulatory molecules to avoid killing by immune cells. As such, inhibition of the “don’t eat me” molecule CD47 has improved skin fibrosis in mice by inducing macrophage- mediated killing of myofibroblasts (54). By extension, our results showing depletion of scleroderma fibroblasts from skin xenografts may be due to insufficient activation of mouse macrophage Sirpa by human CD47 on scleroderma fibroblasts. Although the underlying reason for loss of tolerance remains unclear, contributing factors may include somatic mutations due to phototherapy, previous immunomodulating therapies, higher oxygen demand by sclero- derma skin, or higher levels of fibroblast apoptosis or senescence. We hypothesize that while human fibroblasts were tolerated in healthy skin grafts, scleroderma fibroblasts were seen as foreign and killed by mouse macrophages. Loss of scleroderma fibroblasts thus opened a cellular niche in the skin for immigration of mouse fibroblasts into scleroderma skin grafts. IL- 6 is a well- established profibrotic cytokine whose elevated expression is associated with severity of scleroderma skin fibrosis (55, 56). In MISTRG6 humanized mice, improvement in fibrosis of scleroderma skin grafts correlated with reduced expression of human IL- 6. Regulation of IL- 6 expression is highly complex (38), so multiple mechanisms may explain how healthy bone marrow– derived immune cells could down- regulate IL- 6 expression in scleroderma skin grafts. For example, transcription of IL- 6 is neg- atively regulated by multiple receptors including PPARa as well as estrogen and glucocorticoid receptors (57–59). Degradation of IL- 6 mRNA is enhanced by RNA- binding proteins and microR- NAs (60–62). Moreover, the presence of healthy human immune cells in scleroderma skin grafts may act as a sink for IL- 6 protein and disrupt a positive feedback loop, such as been described for IL- 6 and IL- 17A (63). More studies are required to fully under- stand regulation of IL- 6 expression in scleroderma skin grafts. The capacity of IL- 6 trans- signaling to drive fibrosis has been analyzed in mouse models of skin and lung fibrosis. In the skin, IL6R neutralization with a monoclonal antibody prevented but did not improve existing skin fibrosis in sclerodermoid GvHD model (64). In bleomycin- induced lung fibrosis via intraperitoneal injections, lung macrophages were shown to shed elevated amounts of IL6Ra, and its inhibition with recombinant gp130Fc reduced markers of myofibroblasts and improved lung function in mice (65). In vitro studies further suggested that collagen expression induced by IL- 6 trans- signaling may occur indirectly through TGFβ pathway activation (66). Our scleroderma skin grafts to humanized mice support IL- 6 trans- signaling as a key driver of skin fibrosis in pansclerotic morphea and align our results with recent identification of STAT4 mutations driving IL- 6 expression in the childhood form of pansclerotic morphea (4). The impor- tance of IL- 6 trans- signaling was further supported by our in vitro studies of SSc fibroblasts derived from patients with limited cuta- neous SSc, suggesting IL- 6 as a commonly dysregulated signaling pathway in multiple scleroderma subsets. The ability of T cells to produce sIL6Ra suggests that both macrophages and CD4 T cells may be relevant cell types in promoting fibrosis in patients with scleroderma. IL- 6 trans- signaling induced expression of the profi- brotic gene TNC in addition to its previously recognized ability to induce type I collagen, thereby expanding the number of PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 9 of 12 important ECM genes it regulates. In sum, through the retention of human T cells in scleroderma skin, scleroderma skin grafts to humanized mice permitted identification of IL- 6 trans- signaling as a T cell–regulated process in scleroderma pathogenesis. A limitation of our study was the use of a single donor of scle- roderma skin. However, from this single donor, we were able to generate 18 skin grafts distributed onto 9 mice because of the patient’s willingness to provide an STSG that could be divided into smaller pieces. A second limitation was the unexpected depletion of human fibroblasts from scleroderma skin xenografts. As a result, we were not able to analyze immune–mesenchymal signaling between human cell types. We were interested in the effects of human T cell–derived interferon- γ on fibrosis but could not assess its role because activation of interferon- γ receptor by interferon- γ is species- specific (67). Instead, infiltrating mouse cells into scle- roderma skin grafts showed a clear fibrotic signature that was mod- ulated by HSC engraftment. This allowed us to assess whether scleroderma skin and bone marrow–derived human T cells altered fibrosis signatures via mouse cell types. Future studies and genetic alterations will be needed to achieve maintenance of human fibro- blasts in scleroderma skin in MISTRG6 humanized mice. Myeloablation followed by autologous stem cell transplantation improves event- free and overall survival of patients with severe scleroderma (68) and is associated with normalized T cell receptor diversity (69) and restoration of regulatory T cells (70). Our obser- vations with humanized mice align with these clinical findings in that healthy bone marrow–derived hematopoietic cells improved fibrosis signatures in scleroderma skin. The main effect of healthy human bone marrow was by down- regulating IL- 6 expression and its multiple downstream effects on fibrosis. We also considered whether bone marrow–derived monocytes might be antifibrotic, but their gene expression in scleroderma skin grafts did not show regulatory or antifibrotic signals. In summary, we show that healthy and scleroderma skin can be transplanted to MISTRG6 humanized mice, which maintain human T cells long term. In scleroderma skin, expression of human IL- 6 corresponds with multiple pathologic features including T cell activation and ECM expression. Future work will be needed to clarify the mechanisms of human fibroblast depletion in this model. Materials and Methods Study Patients. Skin from deidentified surgical specimens was used in experi- ments of healthy skin transplantation onto humanized mice. Prior to transplantation, low- quality specimens, such as having extensive stretch markings, were excluded. For STSG of scleroderma skin, a patient with pansclerotic morphea was enrolled in the approved study (Yale Human Investigation Committee #1511016816). Patient skin biopsies for SSc fibroblast isolation were part of a second approved study (Yale Human Investigation Committee #2000026608). Animals. MISTRG (14) and MISTRG6 (19) mice were generated on Rag2−/− IL2rg−/− 129xBalb/c background with genes for human CSF1 (M- CSF), IL3, SIRPA, THPO, CSF2 (GM- CSF), and IL6 knocked into their respective mouse loci. Mice were maintained under specific pathogen- free conditions in the Yale animal facilities. Mouse experi- ments were conducted under a protocol approved by the Yale University Institutional Animal Care and Use Committee and in accordance with AAALAC guidelines. Transplantation of Human CD34+ Hematopoietic Progenitor Cells into Mice. Recipient mice were engrafted with human hematopoietic progenitor cells as previously described (14, 71). Briefly, human fetal liver samples were cut into small fragments and treated with collagenase D (Roche) 100 ng/mL for 45 min at 37 °C. Human CD34+ cells from the resulting cell suspension were purified from the fetal liver by density gradient centrifugation (Lymphocyte Separation Medium, MP Biomedicals) followed by positive immunomagnetic selection with the EasySepTM Human CD34 Positive Selection Kit (StemCell). Cells were frozen in FBS (fetal bovine serum) containing 10% dimethyl sulfoxide (DMSO) and stored in liquid nitrogen. For intrahepatic engraftment, newborn 1 to 3- d- old pups were irradiated with 80 rad using a cabinet irradiator (X- RAD 320) and then injected with 10,000 fetal liver CD34+ cells in PBS into the liver with a 22- gauge needle (Hamilton Company). Skin Transplantation. For transplantation of scleroderma skin, a 6 × 10 cm area of skin was anesthetized on the left back with 1% lidocaine with epinephrine. The site was cleansed with 3 cycles of chlorhexidine application. Using a sterilized Zimmer electric dermatome, a 5 × 7 cm by 0.75 mm thick partial- thickness skin graft was obtained and placed in cold RPMI (Roswell Park Memorial Institute) media. The wound was bandaged with Promogran Prisma (44% oxidized regener- ated cellulose (ORC), 55% Collagen, and 1% silver- ORC) under Tielle nonadhesive hydropolymer and 2 large Tegaderm bandages. The patient was discharged home in good condition and tolerated the procedure without complication. The donor site was assessed weekly for 4 wk for proper healing. Cultured Human CD4+ T Cells and Fibroblasts. Peripheral blood mononuclear cells (PBMCs) from deidentified healthy donors were isolated by density gradient centrifugation with Ficoll- Paque PLUS. CD4+ T cells were isolated from PBMCs by negative selection using the MojoSort Human CD4 T Cell Isolation Kit (BioLegend). CD4+ T cells were activated by incubation in Gibco RPMI 1640 Medium containing 10% FBS and 100 U/mL penicillin–streptomycin at 37 °C in 5% CO2 humidified incubator with Human T- Activator CD3/CD28 Dynabeads (Gibco) for 3 d compared to unstimulated controls in media alone. The supernatant of activated and unstim- ulated CD4 T cells was stored at −80 °C, and cells were immediately processed for staining. After blocking Fc receptors with TruStain FcX (BioLegend), cells were stained for HLA- DR and IL6Ra, fixed in 1% paraformaldehyde, and analyzed on the BD LSR II flow cytometer. sIL6Ra from CD4+ T cell supernatant was quantified using the Human IL- 6R alpha Quantikine ELISA Kit (R&D Systems). HFFs and normal adult dermal fibroblasts were purchased from ATCC (SCRC- 1041 and PCS- 201- 010). Isolation of SSc skin fibroblasts was modified from ref. (72). Briefly, 4- mm punch biopsies were obtained from clinically involved skin on the left forearm. From each biopsy, subcutaneous fat was trimmed, and the remain- ing skin was minced into approximately 10 evenly sized pieces and plated on a 6- well tissue culture dish coated with 0.1% gelatin at 2 to 3 skin pieces per well. The SSc skin pieces were cultured in Dulbecco’s Modified Eagle Medium (DMEM) F12 containing 10% FBS and 100 U/mL penicillin–streptomycin 0.8 mL per well with 0.2 mL media added every 2 d. After 7 d of culture, media were increased to 2 mL total and replaced every 2 d until SSc fibroblasts were confluent for passaging. For gene expression studies, fibroblasts were seeded in DMEM containing 1% FBS overnight at 37 °C in 5% CO2 humidified incubator. The following day, fresh media containing the indicated cytokines were added, and RNA was isolated after 4 h. Human cytokines and their concentrations included IFNa2 1,000 U/mL (BioLegend 592704), IL- 6 100 ng/mL (R&D Systems 206- IL), IL6Ra 100 ng/mL (R&D Systems 227- SR), and Human IL- 6/IL- 6R alpha Protein Chimera 100 ng/mL (R&D Systems 8954- SR). Relative gene expression was normalized to the housekeeping gene UBC (73) and calculated using the 2−ΔΔCt method (74). qPCR primers are listed below. Gene IL6 IL6R IL6ST IL6R (75) sIL6R (75) UBC COL1A1 TNC Forward primer Reverse primer GCAGAAAAAGGCAAAGAATC CTACATTTGCCGAAGAGC CTGGAAACTATTCATGCTACC GACTGTTCTGAAACTTCCTC AAATTGAAGCCATAGTCGTG TTAAAATTGTGCCTTGGAGG CATTGCCATTGTTCTGAGGTTC GTGCCACCCAGCCAGCTATC GCGACAAGCCTCCCAGGTTC GTGCCACCCAGCCAGCTATC CGTCACTTGACAATGCAG TGTTTTCCAGCAAAGATCAG GCTATGATGAGAAATCAACCG TCATCTCCATTCTTTCCAGG GTGGGATCCTCTAGACATTG GTGATCTCTCCCTCATCTTC Histology and Immunofluorescence of Skin Sections. For histology, a strip of skin graft was placed in 10% neutral buffered formalin for 24 h prior to embedding in paraffin. Samples were processed at the Yale Pathology Tissue Services. Photos were taken using the Keyence BZ- X800 microscope at 10× mag- nification and stitched together using their software. For immunohistochemistry analysis, a strip of skin was placed in OCT, frozen, and stored at −80 °C. Using Leica CM1850 cryostat, 8 to 10- μm sections were cut, and slides were stained 10 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org for the indicated human and mouse antibodies. All antibodies were purchased from BioLegend other than human CD11c (Cell Marque), human PDGFRA (Cell Signaling Technology #5241, 1/400 dilution), and mouse Pdgfra (R&D Systems #AF1062, 13 mg/mL) and used at 1/100 dilution. Human antibodies CD31 CD146 CD34 CD45 CD11c CD26 HLA- DR IL6Ra Clone WM59 SHM- 57 561 HI30 5D11 BA5b L243 UV4 Mouse antibodies Clone CD31 CD34 CD45 CD11c CD26 MEC13.3 HM34 30- F11 N418 H194- 112 Single- Cell Analysis. Each cDNA library generated from a skin graft sample was sequenced paired- end on 1 lane with 75 base- pair read length using the Illumina HiSeq 2500 System generating at least 75,000 reads per cell. The 10× genomics Cell Ranger pipeline was used to align the reads, perform clustering and gene expression analysis, and aggregate the samples with normalized read counts. All 10× experiments were completed with the same 3′ chemistry and high- throughput sequencer to avoid batch effects. CellRanger was run 3 times using 3 different genomes: mm10, hg19, and a combined mm10 and GRCh38 genome. The analysis for the mm10, hg19, and combined mm10- GRCh38 genome was conducted separately but followed the same procedure. The raw matrices from Cell Ranger were processed and analyzed using the Seurat version 3 R toolkit for single- cell genomics (76, 77). For the matrices aligned to the mm10 genome and the combined mm10- GRCh38 genome, low- quality cells with a high mitochondrial percentage of over 7.5% were filtered out. For the matrices aligned to the hg19 genome, low- quality cells with a mitochondrial percentage of over 10% were fil- tered out. The data were then normalized (log normalized using the default scaling factor of 10,000) and scaled. Principal components were calculated and the data were visualized by UMAP embedding. After determining the clusters, canonical marker genes were used to identify the cell types present. Once the cell types were identified, we used DAseq (26) to identify the regions that were DA between HSC engraftment and unengrafted samples. From DAseq, identified cell types with DA regions and then analyzed differential gene expression in those cell types using Seurat. Pathway analysis of differentially expressed genes with adjusted P value of less than 0.05 was completed with Metascape (78). Statistical Analysis. Differential gene expression analysis of scRNA- Seq data was performed using Cell Ranger and Loupe Cell Browser software (10× Genomics), which uses a variant of the negative binomial exact test from sSeq and the asymptotic beta test in edgeR depending on sample size (79, 80). Statistical analysis and graphs were generated using GraphPad Prism v9. Pairwise comparisons were analyzed using the two- tailed Student t test and multiple comparison with one- way ANOVA. For all graphs, P < 0.05, P < 0.01, P < 0.001, and ***P < 0.0001. Data, Materials, and Software Availability. Raw scRNA sequencing data gen- erated in this study were deposited to the NCBI Gene Expression Omnibus under the accession number GSE240009 (81). ACKNOWLEDGMENTS. We thank our patient with scleroderma who enabled this study; Guilin Wang, Christopher Castaldi, and the staff of the Yale Center for Genome Analysis; and Amos Brooks and the staff at the Yale Pathology Tissue Services. This work was supported by a grant from the Benaroya Research Institute at Virginia Mason. Dr. I.D.O. is supported by NIH NIAMS K08 AR077689, and previous career development awards from the Dermatology Foundation and CTSA Grant Number UL1 TR001863 from the National Center for Advancing Translational Science, a component of the NIH. J.S.P. and R.A.F. are supported by NIH NIAID R21 AI159580. M.H. is supported by NIH NIAMS R01 AR073270. This work was also supported by the Howard Hughes Medical Institute (R.A.F.). This article is subject to HHMI’s Open Access to Publications policy. HHMI lab heads have previously granted a nonexclusive CC BY 4.0 license to the public and a sublicensable license to HHMI in their research articles. Author affiliations: aDepartment of Dermatology, Yale University School of Medicine, New Haven, CT 06520; bDepartment of Immunobiology, Yale University School of Medicine, New Haven, CT 06520; cProgram in Computational Biology and Bioinformatics, Yale University, New Haven, CT 06511; dProgram in Applied Mathematics, Yale University, New Haven, CT 06511; eDepartment of Molecular, Cellular, and Developmental Biology, Yale University, New Haven, CT 06520; fDepartment of Internal Medicine, Section of Rheumatology, Allergy and Immunology, Yale School of Medicine, New Haven, CT 06520; gDepartment of Pathology, Yale University, New Haven, CT 06511; and hHHMI, Chevy Chase, MD 20815 1. C. P. Denton, D. Khanna, Systemic sclerosis. Lancet 390, 1685–1699 (2017), 10.1016/S0140- 6736(17)30933- 9. 2. A. Gabrielli, E. V. Avvedimento, T. Krieg, Scleroderma. N. Engl. J. Med. 360, 1989–2003 (2009). 3. H. W. Chen et al., Gene expression signatures in inflammatory and sclerotic morphea skin and sera 17. T. Strowig et al., Transgenic expression of human signal regulatory protein alpha in Rag2- /- gamma(c)- /- mice improves engraftment of human hematopoietic cells in humanized mice. Proc. Natl. Acad. Sci. U.S.A. 108, 13218–13223 (2011). 18. A. Rongvaux et al., Human thrombopoietin knockin mice efficiently support human hematopoiesis distinguish morphea from systemic sclerosis. J. Invest. Dermatol., 10.1016/j.jid.2023.02.036 (2023). in vivo. Proc. Natl. Acad. Sci. U.S.A. 108, 2378–2383 (2011). 4. H. Baghdassarian et al., Variant STAT4 and response to ruxolitinib in an autoinflammatory 5. 6. 7. 8. 9. syndrome. N. Engl. J. Med. 388, 2241–2252 (2023), 10.1056/NEJMoa2202318. P. Dieude et al., BANK1 is a genetic risk factor for diffuse cutaneous systemic sclerosis and has additive effects with IRF5 and STAT4. Arthritis Rheum. 60, 3447–3454 (2009). T. R. Radstake et al., Genome- wide association study of systemic sclerosis identifies CD247 as a new susceptibility locus. Nat. Genet. 42, 426–429 (2010). A. C. Mak et al., Brief report: Whole- exome sequencing for identification of potential causal variants for diffuse cutaneous systemic sclerosis. Arthritis Rheumatol. 68, 2257–2262 (2016). E. Lopez- Isac et al., GWAS for systemic sclerosis identifies multiple risk loci and highlights fibrotic and vasculopathy pathways. Nat. Commun. 10, 4955 (2019). B. M. Kraling, G. G. Maul, S. A. Jimenez, Mononuclear cellular infiltrates in clinically involved skin from patients with systemic sclerosis of recent onset predominantly consist of monocytes/ macrophages. Pathobiology 63, 48–56 (1995). 10. M. K. D. Scott et al., Increased monocyte count as a cellular biomarker for poor outcomes in fibrotic diseases: A retrospective, multicentre cohort study. Lancet Respir. Med. 7, 497–508 (2019). 11. M. D. Ah Kioon et al., Plasmacytoid dendritic cells promote systemic sclerosis with a key role for TLR8. Sci. Transl. Med. 10, eaam8458 (2018). 12. L. van Bon et al., Proteome- wide analysis and CXCL4 as a biomarker in systemic sclerosis. N. Engl. J. Med. 370, 433–443 (2014). 13. T. Wohlfahrt et al., Type 2 innate lymphoid cell counts are increased in patients with systemic sclerosis and correlate with the extent of fibrosis. Ann. Rheum. Dis. 75, 623–626 (2015), 10.1136/ annrheumdis- 2015- 207388. 14. A. Rongvaux et al., Development and function of human innate immune cells in a humanized mouse model. Nat. Biotechnol. 32, 364–372 (2014). 15. C. Rathinam et al., Efficient differentiation and function of human macrophages in humanized CSF- 1 mice. Blood 118, 3119–3128 (2011). 16. T. Willinger et al., Human IL- 3/GM- CSF knock- in mice support human alveolar macrophage development and human immune responses in the lung. Proc. Natl. Acad. Sci. U.S.A. 108, 2390–2395 (2011). 19. H. Yu et al., A novel humanized mouse model with significant improvement of class- switched, antigen- specific antibody production. Blood 129, 959–969 (2017). 20. N. C. Kirkiles- Smith et al., Development of a humanized mouse model to study the role of macrophages in allograft injury. Transplantation 87, 189–197 (2009). 21. T. M. Johnson, D. Ratner, B. R. Nelson, Soft tissue reconstruction with skin grafting. J. Am. Acad. Dermatol. 27, 151–165 (1992). 22. L. F. Rose et al., Recipient wound bed characteristics affect scarring and skin graft contraction. Wound Repair Regen. 23, 287–296 (2015). 23. X. Zhao et al., Metabolic regulation of dermal fibroblasts contributes to skin extracellular matrix homeostasis and fibrosis. Nat. Metab. 1, 147–157 (2019). 24. R. R. Driskell et al., Distinct fibroblast lineages determine dermal architecture in skin development and repair. Nature 504, 277–281 (2013). 25. Y. Rinkevich et al., Skin fibrosis. Identification and isolation of a dermal lineage with intrinsic fibrogenic potential. Science 348, aaa2151 (2015). 26. J. Zhao et al., Detection of differentially abundant cell subpopulations in scRNA- seq data. Proc. Natl. Acad. Sci. U.S.A. 118, e2100293118 (2021). 27. S. Assassi et al., Clinical and genetic factors predictive of mortality in early systemic sclerosis. Arthritis Rheum. 61, 1403–1411 (2009). 28. T. M. Holling, E. Schooten, P. J. van Den Elsen, Function and regulation of MHC class II molecules in T- lymphocytes: Of mice and men. Hum. Immunol. 65, 282–290 (2004). 29. D. M. Hernandez et al., IPF pathogenesis is dependent upon TGFbeta induction of IGF- 1. FASEB J. 34, 5363–5388 (2020). 30. H. J. Kim et al., Classification of parenchymal abnormality in scleroderma lung using a novel approach to denoise images collected via a multicenter study. Acad. Radiol. 15, 1004–1016 (2008). 31. M. L. Eloranta et al., Type I interferon system activation and association with disease manifestations in systemic sclerosis. Ann. Rheum. Dis. 69, 1396–1402 (2010). 32. M. Lotz et al., B cell stimulating factor 2/interleukin 6 is a costimulant for human thymocytes and T lymphocytes. J. Exp. Med. 167, 1253–1258 (1988). PNAS 2023 Vol. 120 No. 37 e2306965120 https://doi.org/10.1073/pnas.2306965120 11 of 12 33. T. Hugle et al., Tumor necrosis factor- costimulated T lymphocytes from patients with systemic sclerosis trigger collagen production in fibroblasts. Arthritis Rheum. 65, 481–491 (2013). 34. N. Ito et al., Induction of interleukin- 6 by interferon alfa and its abrogation by a serine protease inhibitor in patients with chronic hepatitis C. Hepatology 23, 669–675 (1996). 35. H. Ihn, S. Sato, M. Fujimoto, K. Kikuchi, K. Takehara, Demonstration of interleukin- 2, interleukin- 4 and interleukin- 6 in sera from patients with localized scleroderma. Arch Dermatol. Res. 287, 193–197 (1995). 36. J. C. O’Brien et al., Transcriptional and cytokine profiles identify CXCL9 as a biomarker of disease activity in morphea. J. Invest. Dermatol. 137, 1663–1670 (2017). 37. E. M. Briso, O. Dienz, M. Rincon, Cutting edge: Soluble IL- 6R is produced by IL- 6R ectodomain shedding in activated CD4 T cells. J. Immunol. 180, 7102–7106 (2008). 38. T. Tanaka, M. Narazaki, T. Kishimoto, IL- 6 in inflammation, immunity, and disease. Cold Spring Harb. Perspect. Biol. 6, a016295 (2014). 39. V. Matthews et al., Cellular cholesterol depletion triggers shedding of the human interleukin- 6 receptor by ADAM10 and ADAM17 (TACE). J. Biol. Chem. 278, 38829–38839 (2003). I. D. Odell et al., Epiregulin is a dendritic cell- derived EGFR ligand that maintains skin and lung fibrosis. Sci. Immunol. 7, eabq6691 (2022). 40. 41. M. Fischer et al., A bioactive designer cytokine for human hematopoietic progenitor cell expansion. Nat. Biotechnol. 15, 142–145 (1997). 42. M. L. Washburn et al., A humanized mouse model to study hepatitis C virus infection, immune response, and liver disease. Gastroenterology 140, 1334–1344 (2011). 43. E. Sefik et al., A humanized mouse model of chronic COVID- 19. Nat. Biotechnol. 40, 906–920 (2021), 10.1038/s41587- 021- 01155- 4. 44. Y. Song et al., A highly efficient and faithful MDS patient- derived xenotransplantation model for pre- clinical studies. Nat. Commun. 10, 366 (2019). 45. R. Das et al., Microenvironment- dependent growth of preneoplastic and malignant plasma cells in humanized mice. Nat. Med. 22, 1351–1357 (2016), 10.1038/nm.4202. 46. L. Sole- Boldo et al., Single- cell transcriptomes of the human skin reveal age- related loss of fibroblast priming. Commun. Biol. 3, 188 (2020). 47. C. Philippeos et al., Spatial and single- cell transcriptional profiling identifies functionally distinct human dermal fibroblast subpopulations. J. Invest. Dermatol. 138, 811–825 (2018), 10.1016/j.jid.2018.01.016. 48. J. Wei et al., Wnt/beta- catenin signaling is hyperactivated in systemic sclerosis and induces Smad- dependent fibrotic responses in mesenchymal cells. Arthritis Rheum. 64, 2734–2745 (2012). 49. R. Zeiser, B. R. Blazar, Pathophysiology of chronic graft- versus- host disease and therapeutic targets. 50. N. Engl. J. Med. 377, 2565–2579 (2017). I. Tawara et al., Interleukin- 6 modulates graft- versus- host responses after experimental allogeneic bone marrow transplantation. Clin. Cancer Res. 17, 77–88 (2011). 51. X. Chen et al., Blockade of interleukin- 6 signaling augments regulatory T- cell reconstitution and attenuates the severity of graft- versus- host disease. Blood 114, 891–900 (2009). 52. G. R. Hill et al., Stem cell mobilization with G- CSF induces type 17 differentiation and promotes scleroderma. Blood 116, 819–828 (2010). 59. A. Ray, K. E. Prefontaine, Physical association and functional antagonism between the p65 subunit of transcription factor NF- kappa B and the glucocorticoid receptor. Proc. Natl. Acad. Sci. U.S.A. 91, 752–756 (1994). 60. Z. Xu et al., miR- 365, a novel negative regulator of interleukin- 6 gene expression, is cooperatively regulated by Sp1 and NF- kappaB. J. Biol. Chem. 286, 21401–21412 (2011). 61. V. Palanisamy, A. Jakymiw, E. A. Van Tubergen, N. J. D’Silva, K. L. Kirkwood, Control of cytokine mRNA expression by RNA- binding proteins and microRNAs. J. Dent. Res. 91, 651–658 (2012). 62. J. G. Kang et al., Kaposi’s sarcoma- associated herpesvirus ORF57 promotes escape of viral and human interleukin- 6 from microRNA- mediated suppression. J. Virol. 85, 2620–2630 (2011). 63. H. Ogura et al., Interleukin- 17 promotes autoimmunity by triggering a positive- feedback loop via interleukin- 6 induction. Immunity 29, 628–636 (2008). 64. D. Le Huu et al., IL- 6 blockade attenuates the development of murine sclerodermatous chronic graft- versus- host disease. J. Invest. Dermatol. 132, 2752–2761 (2012). 65. T. T. Le et al., Blockade of IL- 6 Trans signaling attenuates pulmonary fibrosis. J. Immunol. 193, 3755–3768 (2014). 66. S. O’Reilly, M. Ciechomska, R. Cant, J. M. van Laar, Interleukin- 6 (IL- 6) trans signaling drives a STAT3- dependent pathway that leads to hyperactive transforming growth factor- beta (TGF- beta) signaling promoting SMAD3 activation and fibrosis via Gremlin protein. J. Biol. Chem. 289, 9952–9960 (2014). 67. P. Sultan et al., Pig but not human interferon- gamma initiates human cell- mediated rejection of pig tissue in vivo. Proc. Natl. Acad. Sci. U.S.A. 94, 8767–8772 (1997). 68. K. M. Sullivan et al., Myeloablative autologous stem- cell transplantation for severe scleroderma. N. Engl. J. Med. 378, 35–47 (2018). 69. P. A. Muraro et al., Thymic output generates a new and diverse TCR repertoire after autologous stem 70. cell transplantation in multiple sclerosis patients. J. Exp. Med. 201, 805–816 (2005). I. de Kleer et al., Autologous stem cell transplantation for autoimmunity induces immunologic self- tolerance by reprogramming autoreactive T cells and restoring the CD4+CD25+ immune regulatory network. Blood 107, 1696–1702 (2006). 71. E. Sefik et al., Inflammasome activation in infected macrophages drives COVID- 19 pathology. Nature 606, 585–593 (2022), 10.1038/s41586- 022- 04802- 1. 72. M. Vangipuram, D. Ting, S. Kim, R. Diaz, B. Schule, Skin punch biopsy explant culture for derivation of primary human fibroblasts. J. Vis. Exp. 77, e3779 (2013). 73. S. L. Chua, W. C. See Too, B. Y. Khoo, L. L. Few, UBC and YWHAZ as suitable reference genes for accurate normalisation of gene expression using MCF7, HCT116 and HepG2 cell lines. Cytotechnology 63, 645–654 (2011). 74. K. J. Livak, T. D. Schmittgen, Analysis of relative gene expression data using real- time quantitative PCR and the 2(- Delta Delta C(T)) Method. Methods 25, 402–408 (2001). 75. H. Nakanishi et al., Interleukin- 6/soluble interleukin- 6 receptor signaling attenuates proliferation and invasion, and induces morphological changes of a newly established pleomorphic malignant fibrous histiocytoma cell line. Am. J. Pathol. 165, 471–480 (2004). 53. S. B. Willingham et al., The CD47- signal regulatory protein alpha (SIRPa) interaction is a therapeutic target for human solid tumors. Proc. Natl. Acad. Sci. U.S.A. 109, 6662–6667 (2012). 54. T. Lerbs et al., CD47 prevents the elimination of diseased fibroblasts in scleroderma. JCI Insight 5, 76. A. Butler, P. Hoffman, P. Smibert, E. Papalexi, R. Satija, Integrating single- cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol. 36, 411–420 (2018), 10.1038/nbt.4096. e140458 (2020). 55. M. Hasegawa et al., Serum levels of interleukin 6 (IL- 6), oncostatin M, soluble IL- 6 receptor, and soluble gp130 in patients with systemic sclerosis. J. Rheumatol. 25, 308–313 (1998). 56. S. Sato, M. Hasegawa, K. Takehara, Serum levels of interleukin- 6 and interleukin- 10 correlate with total skin thickness score in patients with systemic sclerosis. J. Dermatol. Sci 27, 140–146 (2001). 57. P. Delerive et al., Peroxisome proliferator- activated receptor alpha negatively regulates the vascular inflammatory gene response by negative cross- talk with transcription factors NF- kappaB and AP- 1. J. Biol. Chem. 274, 32048–32054 (1999). 58. R. L. Jilka et al., Increased osteoclast development after estrogen loss: Mediation by interleukin- 6. Science 257, 88–91 (1992). 77. T. Stuart et al., Comprehensive integration of single- cell data. Cell 177, 1888–1902.e21 (2019). 78. Y. Zhou et al., Metascape provides a biologist- oriented resource for the analysis of systems- level datasets. Nat. Commun. 10, 1523 (2019). 79. D. Yu, W. Huber, O. Vitek, Shrinkage estimation of dispersion in Negative Binomial models for RNA- seq experiments with small sample size. Bioinformatics 29, 1275–1282 (2013). 80. M. D. Robinson, G. K. Smyth, Moderated statistical tests for assessing differences in tag abundance. 81. Bioinformatics 23, 2881–2887 (2007). I. D. Odell et al., IL- 6 trans- signaling in a humanized mouse model of scleroderma. NCBI’s Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE240009. Deposited 03 August 2023. 12 of 12 https://doi.org/10.1073/pnas.2306965120 pnas.org
10.1073_pnas.2218085120	RESEARCH ARTICLE \| BIOCHEMISTRY OPEN ACCESS In diverse conditions, intrinsic chromatin condensates have liquid-like material properties Bryan A. Gibsona,b,1, Claudia Blaukopfc, Tracy Loud, Lifeng Chena,b, Lynda K. Doolittlea,b, Ilya Finkelsteine,f,g and Michael K. Rosena,b,2 , Geeta J. Narlikard, Daniel W. Gerlichc, Contributed by Michael K. Rosen; received October 23, 2022; accepted March 23, 2023; reviewed by Anthony A. Hyman and Richard A. Young Nuclear DNA in eukaryotes is wrapped around histone proteins to form nucle- osomes on a chromatin fiber. Dynamic folding of the chromatin fiber into loops and variations in the degree of chromatin compaction regulate essential processes such as transcription, recombination, and mitotic chromosome segregation. Our understanding of the physical properties that allow chromatin to be dynamically remodeled even in highly compacted states is limited. Previously, we reported that chromatin has an intrinsic capacity to phase separate and form dynamic liquid-like condensates, which can be regulated by cellular factors [B. A. Gibson et al., Cell 179, 470–484.e421 (2019)]. Recent contradictory reports claim that a specific set of solution conditions is required for fluidity in condensates that would otherwise be solid [J. C. Hansen, K. Maeshima, M. J. Hendzel, Epigenetics Chromatin 14, 50 (2021); H. Strickfaden et al., Cell 183, 1772–1784.e1713 (2020)]. We sought to resolve these discrepancies, as our ability to translate with confidence these biophysical observations to cells requires their precise characterization. Moreover, whether chromatin assemblies are dynamic or static affects how processes such as transcription, loop extrusion, and remodeling will engage them inside cells. Here, we show in diverse conditions and without specific buffering components that chromatin fragments form phase separated fluids in vitro. We also explore how sample preparation and imaging affect the experimental observation of chromatin condensate dynamics. Last, we describe how liquid-like in vitro behaviors can translate to the locally dynamic but globally constrained chromatin movement observed in cells. chromatin \| phase separation \| biomolecular condensate To maintain integrity during mitosis and fit into the nucleus, the eukaryotic genome must undergo substantial compaction (1). Chromatin is compacted by affinity-based interac- tions within the fiber and motor-driven extrusion of dynamic loops by protein complexes of the structural maintenance of chromosome (SMC) family. Together, these activities regulate many essential functions, including transcription, recombination, DNA repair, and chromosome segregation (2–5). Individual genomic loci are constrained to move only within a locally defined region inside the nucleus, controlled by interchromatin interactions, physical crosslinks induced by macromolecular complexes, and attachment of chromatin to static nuclear structures (6–8). A detailed account of the physical mechanisms that package the genome is critical, given the importance of spatial organization in regulating DNA-templated processes such as transcription, DNA replication, and DNA repair (9, 10). In a previous report, we described how chromatin has an intrinsic capacity to phase separate, producing liquid-like condensates with cell-like DNA density (11). Among other advances, this work shed light on the physical mechanism underlying a well-described assay for chromatin self-assembly, historically performed by adding superphysiological concentrations of divalent cation alone (12, 13). These intrinsic chromatin condensates, which refers here to factor-independent nucleosome-driven phase separation, can be reg- ulated by cellular factors in kind with their functions in genome regulation (11, 14). We suggested that interchromatin interaction through intrinsic condensation could represent a “ground state” for chromatin organization, molded or disrupted in cells by different regulatory factors (11, 15–25). Recent reports have called this work into question, sug- gesting that without specific buffering components, intrinsic chromatin condensates are solid, reflecting the globally constrained organization of chromatin in cells (26, 27). The distinction between liquid-like and solid-like behavior of chromatin condensates is impor- tant because many nuclear processes rely on dynamic rearrangements of chromatin. Whether such dynamics, especially those on short length scales, can occur through simple thermal fluctuations (as in a liquid-like state) or require input of energy (as in a solid-like Significance The organization of eukaryotic chromatin is important in many nuclear processes. Recent studies have shown that chromatin fragments can self-assemble by phase separation into micron-scale structures in the presence of salt in vitro. There are discrepancies regarding whether these structures generally have liquid-like or solid-like behaviors, an important distinction in considering how processes such as transcription and chromosome remodeling by loop extrusion can occur in cells. Here, we resolve conflicting reports by demonstrating that chromatin condensates have liquid-like behaviors in diverse solution conditions and describing aspects of sample handling that can lead to artifactual solid-like behaviors. Our data suggest how chromatin can be dynamic on short length scales but restrained on long length scales, as observed in cells. Author contributions: B.A.G., D.W.G., G.J.N., and M.K.R. designed research; B.A.G., C.B., T.L., L.C., and L.K.D. performed research; B.G. contributed new reagents/ analytic tools; B.A.G., C.B., T.L., L.C., G.J.N., I.F., D.W.G., and M.K.R. analyzed data; and B.A.G., I.F., G.J.N., D.W.G., and M.K.R. wrote the paper. Reviewers: A.A.H., Max-Planck-Institut fur molekulare Zellbiologie und Genetik; and R.A.Y., Massachusetts Institute of Technology. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1Present address: Department of Cell and Molecular Biology, St. Jude Children’s Research Hospital, Memphis, TN 38105. 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2218085120/-/DCSupplemental. Published April 24, 2023. PNAS 2023 Vol. 120 No. 18 e2218085120 https://doi.org/10.1073/pnas.2218085120 1 of 10 state) impacts mechanistic considerations of many processes, and the ability of in vitro chromatin condensates to model them. Here, we examine in detail the effect of solution conditions on the properties of intrinsic chromatin condensates. We find that condensates composed of small chromatin fragments are fluid, similar to a recent report (28); no unique solution composition is needed for their liquid-like properties. We also examine how sam- ple preparation and imaging strategies can lead to mischaracteri- zation of chromatin condensates. Last, we make efforts to clarify how the liquid-like organization of condensates might translate to chromatin dynamics in cells. Results Bovine Serum Albumin (BSA) and Dithiothreitol (DTT) Are Dispensable for the Liquid-Like Properties of Condensates Formed through Intrinsic Phase Separation of Chromatin. In prior work (11), somewhat complex solutions were used to explore the nature of condensates formed from chromatin, most typically containing tris(hydroxymethyl)aminomethane (Tris) buffer, acetate, potassium, magnesium, BSA, DTT, ethylenediaminetetraacetic acid (EDTA), glycerol, and oxygen-scavenging components (glucose oxidase, catalase, and glucose). The composition of this solution was an effort to mimic the cellular milieu (acetate, potassium, BSA, glycerol, and DTT) and reduce photodamage of condensates during fluorescence microscopy (oxygen-scavenging components and DTT). Recent reports have suggested that BSA and DTT in these buffers lead chromatin condensates to exhibit artifactual liquid-like behavior and that their omission reveals the mesoscale material properties of condensates to be solid-like and constrained (26, 27). We set out to rigorously explore the effect of buffer conditions on chromatin condensate behavior. We assembled dodecameric nucleosomal arrays by salt-mediated dialysis of reconstituted and unlabeled histone octamers and a DNA template with 12 repeats of Widom’s 601 nucleosome posi- tioning element (Fig. 1A). Using differential interference contrast microscopy, we observed in a minimal phase separation buffer composed of 25 mM Tris-acetate, 150 mM potassium acetate, and 1 mM magnesium acetate the formation of micron-sized spherical condensates that rounded upon fusion (Fig. 1B) and maintained a consistent total volume following coalescence (Fig. 1C). Droplet fusion followed by rounding to a spherical shape is a hallmark of fluids. The rate at which rounding occurs is a conse- quence of the relationship between the surface tension ( 𝛾 ) and viscosity ( 𝜂 ) of condensates (29). Simple fluids coalesce according to the equation 𝜏 ≈ 𝜂 ⋅ l , where l is the diameter of condensates 𝛾 prior to fusion and 𝜏 is the characteristic relaxation time during coalescence. To determine 𝜏 for each instance of condensate fusion, we measured the change in aspect ratio ( AR ) over time ( t ) during condensate fusion and found these values fit well to an exponential decay, AR = 1 + (ARinit − 1) ⋅ e−t ∕𝜏 , where ARinit is the initial aspect ratio following the onset of fusion (SI Appendix, Fig. S1). Plotting 𝜏 versus l from many fusion events (N = 177) showed clear linearity, with relaxation times on the order of sec- onds, indicating that intrinsic chromatin condensates in this min- imal buffer are fluid (Fig. 1D). The slope of this plot gives the inverse capillary velocity for these condensates in this solution, which is a quantitative measure of the distinctive ratio of surface tension ( 𝛾 ) and viscosity ( 𝜂 ) of the material. We note that although this analysis reports on the viscosity of the solution, most biomo- lecular condensates are not simple Newtonian fluids but rather complex network fluids with viscoelastic behaviors. Viscoelasticity arises from the hierarchy of interaction strengths and timescales between the molecules (30–32). Full characterization of chromatin condensates will thus require rheological analyses across a range of length- and timescales, and is likely to reveal elasticity at small scales while viscosity dominates at the larger scales relevant to A D B E C F Fig. 1. Intrinsic chromatin condensates are fluid without BSA and DTT. (A) Graphical depiction of the dodecameric nucleosomal arrays used for experimentation. (B) Differential interference contrast microscopy images of a fusion event between intrinsic chromatin condensates in the indicated buffer. (C) Dot plot representation of the inferred total volume of condensates before and after fusion. (D) Relaxation time versus length scale (sum of prefusion diameters) for 177 individual instances of condensate fusion in the buffer composition indicated in Fig. 1B. Inverse capillary velocity, the characteristic ratio of surface tension, 𝜂 , and viscoscity, 𝛾 , is derived from the linear fit (red line) of the plots’ slope. (E) Differential interference contrast microscopy images of intrinsic chromatin condensate fusion in the buffer indicated in Fig. 1B supplemented with BSA (0.1 mg/mL, Left), DTT (5 mM, Middle), or BSA and DTT (0.1 mg/mL and 5 mM, respectively, Right). (F) Bar chart of inverse capillary velocities ( ± SD of 2 biological replicates) of intrinsic chromatin condensates in the buffer indicated in Fig. 1B, buffer with BSA, or buffer with BSA and DTT. For each condition, the fusion events per replicate are: buffer (177 and 68), +BSA (147 and 81), +DTT (183 and 68), +BSA+DTT (184 and 93). Scale bars, in white, are 4 μm. 2 of 10 https://doi.org/10.1073/pnas.2218085120 pnas.org droplet fusion (and fluorescence recovery after photobleaching, FRAP, experiments below). Condensates formed through intrinsic phase separation of chromatin in a solution containing BSA, DTT, or BSA and DTT also coalesced and became round (Fig. 1E). The inverse capillary velocity was identical within error for condensates formed in minimal phase separation buffer alone, or buffer with BSA, DTT, or BSA and DTT (Fig. 1F). These data show that BSA and DTT are not responsible for liquid-like material properties of intrinsic chromatin condensates. Condensates Formed by Intrinsic Phase Separation of Chromatin Are Liquid-Like in a Variety of Solutions. We next explored how different anions and buffering systems affected the material properties of intrinsic chromatin condensates to ascertain whether their fluidity results from a particular component. We assayed the material properties of intrinsic chromatin condensates formed in solutions containing Tris buffer and sodium or potassium salts with chloride, acetate, or glutamate anions. Chloride is a typical anion used for biochemistry in buffered salt solutions. Previously, we used acetate to mimic small-molecule anions in cells; glutamate is the predominant anion found in cells (33). We also used piperazine-N,N′-bis(2-ethanesulfonic acid), pH adjusted with KOH (PIPES-KOH), a buffer/salt often used in fluorescence-based assays that reconstitute cellular processes, including microtubule dynamics (34, 35). First, we determined the phase diagram for dodecameric nucle- osomal arrays at 500 nM nucleosome concentration for each buffer (Fig. 2 A–D). Condensates formed at similar concentrations of mono- and divalent salt in each buffering system, though glu- tamate anions required slightly higher concentrations of salt. In buffers containing chloride anion, condensate formation required at least 2 mM magnesium or the inclusion of glycerol (SI Appendix, Fig. S2 A–F). While the source of this effect is not clear, it could arise from the well-described propensity of glycerol to shield charged peptide side chains from salt (36). Altogether, these data show that intrinsic chromatin condensation occurs robustly across many buffer compositions. For each buffering system, we chose a combination of mono- and divalent ions that resemble physiological concentrations in cells. In these solution conditions, both unlabeled (Fig. 2 E–H) and AlexaFluor 488-labeled nucleosomal arrays (SI Appendix, Fig. S2 G–J) rounded in seconds following fusion. Moreover, con- densate size increased over the course of at least 2 h (Fig. 2 I–L), most likely through condensate fusion (11). These data suggest that in different buffers, intrinsic chromatin condensates are fluid. To probe the dynamics of molecules within these condensates, we photobleached a portion of condensates and measured the recovery of fluorescence using condensates composed of AlexaFluor 488-labeled dodecameric nucleosomal arrays in each of the buff- ered salt solutions (Fig. 2 M–T). These partial-droplet fluorescence recovery after photobleaching (FRAP) experiments was carried out using glass treatments that reduce condensate motion (see below) to aid in the quantitation of photobleach recovery. This preparation affects condensates in chloride buffers more strongly than others, resulting in adherence to the surface and nonspherical shapes. Still, in each buffer condition, we observed rapid and full fluorescence recovery from photobleaching in minutes (Fig. 2 Q–T). Notably, condensates in buffers with glutamate, the pre- dominate anion in cells, recovered approximately three-fold more rapidly from photobleaching as compared to chloride, acetate, and PIPES-KOH buffered salt solutions (based on t1/2 of fluores- cence recovery). These data demonstrate that in a variety of simple buffers, intrinsic chromatin condensates are fluid. Condensate Fluidity Is Retarded by a Nonphysiologic Solution, but not by Several Other Factors. The material properties of biomolecular condensates are an emergent phenomenon, where small differences between molecules and their interactions can impart substantial effects. We next sought to explore whether small differences in nucleosome arrays vs nucleosomal arrays or the conditions used to assay chromatin condensates might have significant effects on their dynamics and liquid-like behavior. Reconstituting nucleosome arrays from bacterially purified com- ponents is a complex biochemical procedure (37), and small errors can result in underassembly, partial assembly, or overassembly of nucleosome arrays, which result, respectively, in free nucleosome positioning sequences, subnucleosomal structures (e.g., tetra- or hexasomes), or aggregates of nonnucleosomal histones on chro- matinized DNA. Intrinsic chromatin condensates composed of improperly formed nucleosome arrays would likely affect their material properties, so we have accounted here for potential differ- ences in the quality of nucleosome arrays by performing key exper- iments with independent materials from multiple laboratories with experience in chromatin reconstitution (Fig. 2 and SI Appendix, Fig. S2; also, see Experimental Methods), with each demonstrating clear liquid-like material properties. We first explored how long linker DNA lengths might affect chromatin droplet fluidity. In cells, linker DNA length is highly regulated. While each eukaryotic organism, cell type, and genomic region can harbor short (~20 bp in Saccharomyces cerevisiae) to long (~90 bp in Thyone briareus) average linker lengths (38), across eukaryotes, there is a predisposition for nucleosomes to be placed every 10n+5 base pairs from one another (e.g., 5, 15, and 25) (39–41). 10n-spaced (e.g., 10, 20, and 30) polynucleosome arrays can adopt hierarchically folded two-start zig-zag fibers in vitro (42, 43), while 10n+5-spaced arrays prefer to interact with other chromatin frag- ments and form intrinsic chromatin condensates (11), demon- strating how the specific DNA template used in these assays can impact chromatin droplet formation and perhaps the material properties of the condensates that are formed. We assembled nucleosome arrays using a DNA template that purported to produce condensates with more solid-like material properties (27). This template has 60 bp internucleosome linker DNA lengths, longer than those we had previously employed (15 to 45 bp), and 4 bp palindromic single-stranded DNA overhangs, which might act as a source of nonnucleosomal valency for this template (Fig. 3A). We prepared chromatin using this DNA tem- plate and found that intrinsic chromatin condensates fused and rounded in seconds in a buffer lacking BSA or DTT, composed of 25 mM Tris-acetate, 150 mM potassium acetate, and 1 mM magnesium acetate (Fig. 3B). In partial-droplet FRAP assays in the presence of either BSA or BSA and DTT, these condensates each recovered in minutes within error of one another (Fig. 3 C–E). These experiments demonstrate that altered material prop- erties do not arise from differences in DNA template or an effect from BSA in the presence of DTT. We next tested whether the concentration of nucleosome arrays or size of chromatin condensates might alter their properties. We assembled chromatin condensates at 10 nM nucleosome concen- tration (0.83 nM nucleosome arrays) in a buffer composed of 25 mM Tris-acetate, 100 mM potassium acetate, and 2 mM mag- nesium acetate. In partial-droplet FRAP on large droplets and half-droplet FRAP on small droplets, recovery from photobleach occurred in minutes (Fig. 3 F and G), similar to condensates formed with 1 μM nucleosome concentrations (Fig. 2 N and R). These data demonstrate that chromatin concentration and con- densate size do not appreciably change intrinsic chromatin con- densate fluidity. PNAS 2023 Vol. 120 No. 18 e2218085120 https://doi.org/10.1073/pnas.2218085120 3 of 10 A E I M Q B F J N R C G K O S D H L P T Fig. 2. Intrinsic chromatin condensates are fluid in diverse buffers. Phase diagrams for intrinsic chromatin condensate formation in (A) Tris-chloride, (B) Tris- acetate, (C) Tris-glutamate, and (D) PIPES-KOH buffers. Dark circles indicate the presence of condensates, and representative images are in SI Appendix, Fig. S2. With materials produced and experiments performed in the Narlikar lab, bright-field light microscopy images of intrinsic chromatin condensate fusion in (E) Tris-chloride, (F) Tris-acetate, (G) Tris-glutamate, and (H) PIPES-KOH buffers. Boxplots of intrinsic chromatin condensate diameters following induction of phase separation in (I) Tris-chloride, (J) Tris-acetate, (K) Tris-glutamate, or (L) PIPES-KOH-based buffers. Bars marked with different letters are significantly different from one another (Student’s t test, P < 1 × 10−7). Fluorescence microscopy images of partial-droplet FRAP of intrinsic chromatin condensates, in green, composed of nucleosomal arrays labeled with AlexaFluor 488 in (M) Tris-chloride, (N) Tris-acetate, (O) Tris-glutamate, or (P) PIPES-KOH-based buffers. Quantification of partial- droplet FRAP of intrinsic chromatin condensates in (Q) Tris-chloride, (R) Tris-acetate, (S) Tris-glutamate, or (T) PIPES-KOH-based buffers. Fluorescence signal is normalized to pre-bleach droplet intensity and error bars are SD of six technical replicates. Scale bars, in white, are 4 μm. 4 of 10 https://doi.org/10.1073/pnas.2218085120 pnas.org A B C D E OAc F H G I Fig. 3. Intrinsic chromatin condensates are fluid in most conditions, but not in superphysiologic magnesium alone. (A) Graphical depiction of a long linker-length 12 × 601 DNA template (27). (B) Confocal fluorescence microscopy of intrinsic chromatin condensates composed of AlexaFluor 488-labeled long linker-length nucleosomal arrays, in green, undergoing fusion. Confocal fluorescence microscopy of partial-droplet FRAP of intrinsic chromatin condensates composed of AlexaFluor 488-labeled long linker-length nucleosomal arrays, in green, formed in the presence of (C) 0.1 mg/mL BSA or (D) 0.1 mg/mL BSA and 5 mM DTT. (E) Quantification of partial-droplet FRAP of intrinsic chromatin condensates with BSA or BSA and DTT, in blue and red, respectively. Fluorescence signal is normalized to pre-bleach droplet intensity and error bars are SD of six technical replicates. (F) Partial-droplet FRAP or (G) half-droplet FRAP of large or small intrinsic chromatin condensates, respectively, formed at 10 nM nucleosome concentration in minimal phase separation buffer. (H) Partial-droplet FRAP and (I) quantitation of fluorescence recovery for intrinsic chromatin condensates induced to form at 10 nM nucleosome concentration with 4 mM magnesium acetate. Scale bars, in orange and white, are 4 and 1 μm, respectively. Last, we explored the dynamics of intrinsic chromatin con- densates formed in superphysiologic concentrations of magne- sium without monovalent salt. These, or similar, nonphysiologic conditions have sometimes been used to study chromatin self-assembly in the past (44). We formed chromatin condensates at 10 nM or 1 μM nucleosome concentration in a buffer com- posed of 25 mM Tris-acetate and 4 mM magnesium acetate and observed in each condition minimal recovery from photobleach in partial-droplet FRAP assays (Fig. 3 H and I and SI Appendix, Fig. S2K). These data demonstrate that without monovalent salt, chromatin condensates formed in 4 mM magnesium exhibit solid-like material properties. While low ionic strength in this buffer could lead to long lifetime charge–charge interactions, it is not clear why condensates in magnesium alone should have solid-like properties. Regardless of mechanism, networks of interaction between chromatin fragments likely differ within chromatin condensates formed with these nonphysiologic buff- ers, complicating interpretations regarding the behaviors of chro- matin in cells. Sample Preparation Affects Condensate Movement and Internal Dynamics. Similar to single-molecule biochemical imaging studies (45), it is common when studying biomolecular condensates to prepare the cover glass surface to prevent artifactual wetting of biomolecules. In a study where intrinsic chromatin condensates were found to be solid-like (27), chromatin condensates were deposited onto raw glass by centrifugation prior to fluorescence microscopy (Fig. 4A). In our previous studies (11), we passivated the glass surface with methoxy polyethylene glycol (mPEG) and BSA to prevent the adherence of macromolecules and allowed condensates to settle onto the surface by gravity to minimize force- mediated perturbation (Fig. 4B and figure S1E of ref. 11). We investigated whether these differences affected the motion and physical properties of chromatin condensates. Chromatin condensates deposited by centrifugation onto raw glass did not appreciably move during 2 min of observation by fluorescence microscopy and exhibited nonspherical morphology consistent with adhesion to the surface (Fig. 4 C and D). In contrast, intrinsic chromatin condensates settled by gravity onto mPEGylated PNAS 2023 Vol. 120 No. 18 e2218085120 https://doi.org/10.1073/pnas.2218085120 5 of 10 B E BSA F D H A C G t n e m e c a l p s i d d e r a u q s I J K Fig. 4. Condensate movement and dynamics is affected by microscopy glass preparation. Graphical depiction of techniques used to prepare intrinsic chromatin condensates for fluorescence microscopy imaging: (A) Intrinsic chromatin condensates can be spun onto raw glass using a centrifuge (27). (B) Alternatively, intrinsic chromatin condensates can be added to a 384-well microscopy plate and brought by gravity to rest on mPEGylated and BSA-passivated glass (11). Movement of a single or many intrinsic chromatin condensates, following their preparation for fluorescence microscopy imaging on untreated glass (C and D) and prepared glass (E and F). (C and E) The movement of an individual condensate across 2 min in 10 s intervals is overlaid in orange on fluorescence microscopy images of AlexaFluor 488-labeled intrinsic chromatin condensates, in green. (D and F) The relative movement of many condensates determined across 2 min in 500 ms intervals. (G) Plot of mean squared displacement ( ± SE) over lag time, 𝜏 , for intrinsic chromatin condensates between 4 and 8 μm in diameter following centrifugation onto untreated glass (gray dots) or settling by gravity onto prepared glass (black dots). The diffusion coefficient, indicated in orange ± SE, of intrinsic chromatin condensates can be calculated from the slope of the linear fit (dashed line) of the plotted data. For droplets centrifuged onto untreated glass, three replicates with 11,222, 8,114, and 14,092 trajectories extracted from 171, 147, and 238 droplets were used for analysis, respectively. For droplets settled onto passivated glass, three replicates with 6,563, 7,900, and 8,179 trajectories extracted from 106, 100, and 101 droplets were used for analysis, respectively. (H) Bar chart of the diffusion coefficients of intrinsic chromatin condensates following their preparation for microscopy with and without centrifugation, mPEGylation of the microscopy glass, and BSA passivation of the microscopy well. Error bars are SD of four technical replicates. Confocal fluorescence microscopy of whole-droplet FRAP of intrinsic chromatin condensates composed of AlexaFluor 488-labeled long linker-length nucleosomal arrays, in green, settled onto (I) untreated or (J) mPEGylated glass. (K) Quantification of whole-droplet FRAP recovery of intrinsic chromatin condensates on raw or mPEGylated glass, in gray and black, respectively. Fluorescence signal is normalized to pre-bleach droplet intensity and error bars are SD of six technical replicates. Panels C–H used nucleosome arrays with a 25 base pair internucleosome linker length. Panels I–K used nucleosome arrays with a 60 base pair internucleosome linker length. Scale bars, in white, are 4 μm. 6 of 10 https://doi.org/10.1073/pnas.2218085120 pnas.org and BSA-passivated glass moved many microns in distance, remained spherical, and underwent fusion (Fig. 4 E and F). We quantified the movement in these two conditions by measuring the mean squared displacement by lag time and found that condensates settled onto prepared glass were mobile, with a diffusion coefficient of 0.035 ± 0.005 μm2/s for condensates between 4 and 8 μm in diameter, while those deposited onto raw glass were not (Fig. 4G and SI Appendix, Fig. S3 A–D and K). To understand what experimental parameter led to these differ- ences, we quantified condensate movement with and without centrifugation, mPEGylation, and BSA passivation. Time-lapse imaging showed that diffusive condensate movement requires mPEGylation and BSA passivation, though some subdiffusive mobility is retained without passivation so long as glass is mPE- Gylated and condensates are not centrifuged onto the surface (Fig. 4H and SI Appendix, Fig. S3 D–K). The microscopy sample preparation can thus impact condensate movement and fusion. We considered whether BSA leaching from the passivated glass surface might lead to liquid-like condensate properties. Three pieces of data argue against this possibility. First, our photobleach- ing experiments, which show rapid recovery, are carried out in the absence of BSA passivation (Fig. 2 M–T). Second, condensates move, albeit with restriction, in the absence of BSA passivation (Fig. 3H). Third, condensates fuse with comparable kinetics in the presence or absence of BSA passivation (SI Appendix, Fig. S3L). Thus, the liquid-like behavior of intrinsic chromatin condensates is not a consequence of BSA passivation. Given the strong effects of slide surfaces on condensate move- ment, we next asked how glass treatment might affect the physical properties of the condensates themselves. Using long linker-length chromatin (Fig. 3A), we photobleached entire condensates to probe the extent of fluorescence recovery resulting from chromatin exchange between the condensed and dilute phases. This is distinct from partial-droplet FRAP in Fig. 3, which principally measures the movement of chromatin within a condensate. On raw glass, we observed appreciable recovery of fluorescence in the presence of BSA and DTT as described in other work (Fig. 4I) (27). In contrast, condensates settled onto mPEGylated glass did not sub- stantially recover (Fig. 4 J and K), which we hypothesized previ- ously (11) is due to the very low concentration of chromatin in solution (note that differences in partial versus whole-droplet FRAP recoveries were addressed in our previous study). These data show that microscopy preparations affect not just the movement of intrinsic chromatin condensates, but also their exchange with molecules in solution. While we do not understand the basis of this difference, condensates centrifuged or settled to a strongly adherent glass surface will be flattened, perhaps appreciably so. In contrast, condensates settled onto a well-passivated surface will remain spherical and shielded from the glass. The additional inter- actions between a flat condensate and glass may influence pho- tobleaching recovery and might exhibit sensitivity to specific buffering components. In a previous report where biochemical experiments were performed without these additions (27), intrinsic chromatin condensates demonstrated solid-like behavior, raising the possibility that photocrosslinking might have limited chromatin mobility in their condensate imaging experiments. We therefore explored the effect of ROS mitigation on photocrosslinking of intrinsic chromatin condensates. We developed an assay to measure light-induced photo- crosslinking of intrinsic chromatin condensates. In this assay, condensates were formed in a buffer where free magnesium was required for their formation (Fig. 2B, 2 mM Mg (OAc)2 and 50 mM KOAc). The concentration of monovalent salt in this buffer is insufficient to induce nucleosomal arrays to phase sepa- rate. Under these conditions, condensates can be dissolved by chelation of magnesium with EDTA (Fig. 5A). We hypothesized that photocrosslinking condensates would prevent their dissolu- tion by EDTA. We formed intrinsic chromatin condensates with 1 in 80 his- tone proteins conjugated to a fluorophore in a magnesium-dependent phase separation buffer. Exposure of these condensates to 20 W/cm2 of fluorescent light for 500 ms ( 𝜆 = 488 nm), comparable to that used on our microscope in a typical imaging experiment, prevented their dissolution by EDTA (Fig. 5B). Condensates in adjacent fields, which had not been exposed to light, were dissolved 1 min after the addition of EDTA. Light-induced solidification of condensates did not occur with fivefold less fluorophore or 10-fold less light (Fig. 5 C and D). Shorter exposure to light of higher intensity also led to condensate solidification, demonstrating that the totality and not duration of light exposure drives condensate solidification (Fig. 5E). Addition of an oxygen scavenging system to the buffer prevents light-induced condensate solidification (Fig. 5F), although its inclusion can alter condensate properties (SI Appendix, Fig. S4). Together, these data demonstrate that imaging intrinsic chromatin condensates can cause their solidification and suggest that this results from light-induced ROS production and photocrosslinking. Furthermore, these data highlight how minimizing light exposure, fluorophore density, and including oxygen scavengers can prevent artifactual hardening of condensates. We next sought to understand how the inclusion of BSA and/ or DTT can influence photocrosslinking of intrinsic chromatin condensates. Adding 100 ng/μL BSA, as used in our own and other studies (11, 26, 27), did not prevent condensate solidifica- tion (Fig. 5G). In 5 mM DTT, light exposure and EDTA addition resulted in loss of spherical condensates but left aggregates in solution, suggesting partial but incomplete mitigation of photo- crosslinking (Fig. 5H). Adding BSA and DTT together prevented condensate solidification, enabling their dissolution upon EDTA addition. While the mechanism by which BSA, or some compo- nent in commercially available BSA, can inhibit photocrosslinking is unclear, these observations suggest that BSA and DTT can act in concert to reduce light-induced hardening of intrinsic chroma- tin condensates (Fig. 5I). BSA and DTT Mitigate Photocrosslinking of Intrinsic Chromatin Condensates. Having analyzed how differences in sample preparation can alter condensate movement and FRAP recovery, we next examined the effects of imaging parameters. Laser excitation can produce radical oxygen species (ROS) that react with and crosslink neighboring molecules. Such light-induced crosslinking can cause artifactual hardening of biomolecular condensates (46). ROS production and photocrosslinking of molecules are typically mitigated in biochemical imaging studies by including reducing agents in buffers, limiting fluorophore concentration, minimizing laser excitation, and scavenging soluble oxygen in solution (47–49). Intrinsic Chromatin Condensates Show Length-Dependent Dynamics. The cellular chromatin polymer is vastly longer than the nucleosome arrays investigated here. According to classical polymer theory, this additional length would add constraints on polymer movement due to increased adhesion to neighboring molecules (50). As a step toward addressing this issue, we reconstituted chromatin in vitro with 7, 12, or 17 nucleosomes by altering the number of repeats of Widom’s 601 nucleosome positioning sequence, while keeping the internucleosome linker lengths constant. Chromatin condensates composed of these arrays were formed at 1 μM nucleosome concentration in a physiologic PNAS 2023 Vol. 120 No. 18 e2218085120 https://doi.org/10.1073/pnas.2218085120 7 of 10 A C G B D E F H I Fig. 5. DTT and BSA mitigate photocrosslinking during fluorescence microscopy. (A) Diagram depicting an assay to detect photocrosslinking of intrinsic chromatin condensates. (Left) Magnesium-dependent intrinsic chromatin condensates are exposed to fluorescent light prior to the addition of super- stoichiometric quantities of EDTA. (Right) Photo- crosslinked condensates fail to dissipate following chelation of magnesium. (B) Confocal fluorescence microscopy images of intrinsic chromatin condensates composed of nucleosomal arrays where 1 in 80 histone molecules are labeled with AlexaFluor 488. Images are following exposure to fluorescent light and both before (Left) and after (Right) the addition of EDTA. Confocal fluorescence microscopy images of intrinsic chromatin condensates imaged, as in Fig. 5B, with (C) less fluorophore, (D) less exposure, (E) more laser power with less exposure, or (F) the inclusion of oxygen scavenging components. Confocal fluorescence microscopy images of intrinsic chromatin condensates formed in the presence of (G) BSA, (H) DTT, or (I) BSA and DTT and imaged as described in Fig. 5B. Fluorescent microscopy images before and after the addition of EDTA were processed separately. All experiments were performed using nucleosome arrays with 25 base pair internucleosome repeat length. Scale bars, in white, are 10 μm. salt solution and assayed for changes in their dynamics using FRAP (Fig. 6 A–D). We found that increased chromatin length results in more limited recovery from photobleach. Condensates composed of even longer nucleosome arrays would be expected to exhibit more solid-like properties, as demonstrated recently with other biomolecular condensates (51, 52). Still, for very long polymers, short sections will retain dynamics at short length scales while moving little at longer lengths (50). Thus, an intrinsic chromatin condensate composed of chromosome-length fragments would be locally dynamic but exhibit little recovery from photobleach, like the dynamics of the genome observed in cells (53). fusion assays) under a wide range of physiologically relevant solution conditions. Quantification of rounding after fusion and partial-droplet FRAP recovery show that BSA and DTT impart no effect on condensate fluidity, even when using a DNA template that had exhibited solid-like behaviors (27). Others have recently come to similar conclusions (28). From a series of experiments, we show that fluid condensates can appear solid-like without passivation of glass or when ROS-limiting components are omitted. Our results have important implications on the behavior of chromatin and the use of phase-separated chromatin condensates to study nuclear processes. Discussion The Liquid-Like Properties of Intrinsic Chromatin Condensates. Here, we present data demonstrating that intrinsic chromatin condensates composed of short nucleosome arrays are fluid (likely viscoelastic) over the course of minutes (in FRAP and droplet Regulated Solidification of Chromatin Assemblies in Cells. We have shown intrinsic chromatin condensates are fluid, but it remains possible that chromatin assemblies may solidify in cells even on short length scales as part of a regulated biological process. ROS can crosslink and solidify chromatin (Fig. 5) and are produced in cells as a by-product of cellular processes. ROS are produced at A B C D Fig. 6. Length-dependent effects on chromatin condensate dynamics. Confocal fluorescence microscopy images of partial-droplet FRAP of intrinsic chromatin condensates, in green, composed of AlexaFluor 488-labeled arrays that are (A) 7, (B) 12, or (C) 17 nucleosomes in length. (D) Quantification of partial-droplet FRAP of intrinsic chromatin condensates composed of 7, 12, or 17 nucleosome-long arrays in blue, green, and purple, respectively. Fluorescence signal is normalized to pre-bleach droplet intensity and error bars are SD of six technical replicates. Scale bars, in white, are 4 μm. 8 of 10 https://doi.org/10.1073/pnas.2218085120 pnas.org large by mitochondrial metabolism or inflammatory cell signaling (54), and at specific genomic loci by enzymes like lysine-specific demethylase 1, whose removal of histone lysine methylation produces not just hydrogen peroxide, but also formaldehyde, which can crosslink and arrest chromatin movement (6, 55). It will be interesting to examine in future studies whether, and how, chromatin dynamics might be slowed to a solid-like state as part of normal cellular signaling and functions. Bridging Fluid Condensates to Chromatin Dynamics in the Cell. A large body of data on the spatial organization and movement of loci in different cell types has demonstrated that on short length scales chromatin is highly dynamic. Superresolution and single- molecule fluorescence imaging have shown nucleosomes compact into 30 to 50 nm chromatin assemblies called “clutches” (56), which further assemble into chromatin domains with a radius of ~100 to 300 nm (57–61). Analyses of their motion have shown that individual nucleosomes move within these domains on tens of milliseconds timescales (62, 63) and the domains themselves move on hundreds of milliseconds to seconds timescales (6, 57, 58, 61, 64, 65). In both regimes, movement is subdiffusive and/ or confined (6, 58, 61, 63–66), in part due to constraints on a given chromatin segment imparted by adhesions to surrounding structures, which increase with length of the segment (i.e., number of adhesions) (64, 67). While poorly understood ATP-dependent processes can affect longer-length chromatin motion (65, 68), movement at small scales (e.g., short chromatin assemblies, limited radius) is thought to primarily occur via passive thermal fluctuations rather than actively driven processes (6, 57, 58, 64, 65, 67). Thus, short range/timescale movement reflects the dynamics of local inter- nucleosome contacts that are subject to changes induced by his- tone acetylation and binding of linker histone H1 (59, 63). These local dynamics are likely necessary for many genome functions, such as enhancer–promoter interactions (69), loop extrusion by SMC complexes (70, 71), and homologous pairing of sequences during meiosis and DNA repair (71, 72). Lack of movement at greater scales (~400 nm or larger) arises from multiple con- straints, including the large size of chromosomes, crosslinking macromolecules (e.g., SMC complexes, adaptor proteins), and attachment of chromatin to nuclear structures (e.g., nuclear bod- ies, nuclear lamina) (6, 66, 73–75). These constraints lead to the well-described reticence of chromatin in cells to recover from photobleaching (27, 73, 76–80). In condensates that form through interactions between small chromatin fragments alone, these larger-scale constraints are not present, allowing micrometer-scale movement and photobleach recovery. These long-range behaviors of intrinsic chromatin condensates in vitro very likely reflect the interactions that govern short length/times- cale chromatin dynamics in cells (81). As numerous cellular processes depend on short-range chromatin dynamics, the reported absence of dynamics in chromatin condensates in vitro (26, 27) is thus unlikely to be physiologic, except perhaps in very specific biological situations (see above). The length-dependent FRAP recovery behaviors shown in Fig. 6 underscore an important issue when studying condensates in vitro. Decades of study have demonstrated that the structure and func- tion of discrete macromolecular complexes in vitro inform in a straightforward fashion on the structure and function of those factors in vivo. In contrast, the properties of condensates generated in vitro (e.g., size, structure, and behavior) require care in their translation to cellular correlates. In this regard, we propose that factors that influence “mesoscale” genome dynamics in cells will not be readily observable when studying intrinsic chromatin con- densates generated from kilobase-scale DNA stretches. Mesoscale genome dynamics, defined as the larger-scale motion that deter- mines photobleach recovery of chromatin in cells, are likely gov- erned by short-range chromatin interactions translated to genome-relevant scales in the context of complicating factors that crosslink and adhere chromatin to physical structures of the nucleus. The utility of the reconstituted system of phase-separated nucleosomal arrays is the ability to study how factors influence short-range chromatin dynamics using a macroscopic technique like FRAP. Experimental Methods Detailed methods for expression and purification of recombinant proteins and DNA, assembly of nucleosome arrays, preparation of slide surfaces, imaging, condensate crosslinking and fusion assays, and image analysis are provided in SI Appendix. Lead Contact and Materials Availability Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Michael K. Rosen ([email protected]). Data, Materials, and Software Availability. Datasets and software are availa- ble by requests to the corresponding author. Microscopy images data have been deposited in Dryad (doi:10.5061/dryad.83bk3j9ws) (82). ACKNOWLEDGMENTS. The research was supported by the HHMI, a Paul G. Allen Frontiers Distinguished Investigator Award (to M.K.R.), grants from the NIH (R35 GM141736 to M.K.R., R35 GM127020 to G.J.N., and F32GM129925 to B.A.G.), the Welch Foundation (I-1544 to M.K.R.), the European Research Council under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 101019039, to D.W.G.), the Austrian Science Fund (SFB F34-06 and DK W1238 to D.W.G.), the Wiener Wissenschafts-, Forschungs- und Technologiefonds (LS17-003 and LS19-001 to D.W.G.), and the NSF (NSF-1921794 to G.J.N.). We thank Tieqiao (Tim) Zhang for measuring light intensity in our confocal microscope. Author affiliations: aDepartment of Biophysics, University of Texas Southwestern Medical Center, Dallas, TX 75390; bHHMI, University of Texas Southwestern Medical Center, Dallas, TX 75390; cInstitute of Molecular Biotechnology of the Austrian Academy of Sciences, Vienna BioCenter, 1030 Vienna, Austria; dDepartment of Biochemistry and Biophysics, University of California, San Francisco, CA 94158; eDepartment of Molecular Biosciences, University of Texas at Austin, Austin, TX 78712; fInstitute for Cellular and Molecular Biology, University of Texas at Austin, Austin, TX 78712; and gCenter for Systems and Synthetic Biology, University of Texas at Austin, Austin, TX 78712 1. 2. 3. 4. D. E. Olins, A. L. Olins, Chromatin history: Our view from the bridge. Nat. Rev. Mol. Cell Biol. 4, 809–814 (2003). T. Misteli, The self-organizing genome: Principles of genome architecture and function. Cell 183, 28–45 (2020). L. Mirny, J. Dekker, Mechanisms of chromosome folding and nuclear organization: Their interplay and open questions. Cold Spring Harb. Perspect. Biol. 14, a040147 (2022). P. Batty, D. W. Gerlich, Mitotic chromosome mechanics: How cells segregate their genome. Trends Cell Biol. 29, 717–726 (2019). 6. W. F. Marshall et al., Interphase chromosomes undergo constrained diffusional motion in living cells. Curr. Biol. 7, 930–939 (1997). 7. H. A. Shaban, A. Seeber, Monitoring the spatio-temporal organization and dynamics of the genome. 8. 9. Nucleic Acids Res. 48, 3423–3434 (2020). L. A. Mirny, M. Imakaev, N. Abdennur, Two major mechanisms of chromosome organization. Curr. Opin. Cell Biol. 58, 142–152 (2019). E. H. Finn, T. Misteli, Molecular basis and biological function of variability in spatial genome organization. Science 365 (2019). 5. M. J. Rowley, V. G. Corces, Organizational principles of 3D genome architecture. Nat. Rev. Genet. 19, 10. J. Dekker, L. Mirny, The 3D genome as moderator of chromosomal communication. Cell 164, 789–800 (2018). 1110–1121 (2016). PNAS 2023 Vol. 120 No. 18 e2218085120 https://doi.org/10.1073/pnas.2218085120 9 of 10 11. B. A. Gibson et al., Organization of chromatin by intrinsic and regulated phase separation. Cell 179, 48. R. Dixit, R. Cyr, Cell damage and reactive oxygen species production induced by fluorescence 470–484.e421 (2019). 12. J. C. Hansen, Conformational dynamics of the chromatin fiber in solution: Determinants, mechanisms, and functions. Annu. Rev. Biophys. Biomol. Struct. 31, 361–392 (2002). microscopy: Effect on mitosis and guidelines for non-invasive fluorescence microscopy. Plant J. 36, 280–290 (2003). 49. C. Joo et al., Real-time observation of RecA filament dynamics with single monomer resolution. Cell 13. K. Maeshima et al., Nucleosomal arrays self-assemble into supramolecular globular structures 126, 515–527 (2006). lacking 30-nm fibers. Embo J. 35, 1115–1132 (2016). 14. S. Sanulli et al., HP1 reshapes nucleosome core to promote phase separation of heterochromatin. 50. M. Rubinstein, R. H. Colby, Polymer Physics (OUP Oxford, 2003), p. 456. 51. M. M. Keenen et al., HP1 proteins compact DNA into mechanically and positionally stable phase Nature 575, 390–394 (2019). separated domains. Elife 10 (2021). 15. A. G. Larson et al., Liquid droplet formation by HP1alpha suggests a role for phase separation in 52. F. Muzzopappa, M. Hertzog, F. Erdel, DNA length tunes the fluidity of DNA-based condensates. heterochromatin. Nature 547, 236–240 (2017). Biophys. J. 120, 1288–1300 (2021). 16. A. R. Strom et al., Phase separation drives heterochromatin domain formation. Nature 547, 53. M. M. Tortora, H. Salari, D. Jost, Chromosome dynamics during interphase: A biophysical 241–245 (2017). perspective. Curr. Opin. Genet. Dev. 61, 37–43 (2020). 17. A. Boija et al., Transcription factors activate genes through the phase-separation capacity of their 54. H. Sies, D. P. Jones, Reactive oxygen species (ROS) as pleiotropic physiological signalling agents. activation domains. Cell 175, 1842–1855.e1816 (2018). Nat. Rev. Mol. Cell Biol. 21, 363–383 (2020). 18. M. H. Kagey et al., Mediator and cohesin connect gene expression and chromatin architecture. 55. Y. Shi et al., Histone demethylation mediated by the nuclear amine oxidase homolog LSD1. Cell Nature 467, 430–435 (2010). 119, 941–953 (2004). 19. C. H. Li et al., MeCP2 links heterochromatin condensates and neurodevelopmental disease. Nature 56. M. A. Ricci, C. Manzo, M. F. García-Parajo, M. Lakadamyali, M. P. Cosma, Chromatin fibers are formed 586, 440–444 (2020), 10.1038/s41586-020-2574-4. by heterogeneous groups of nucleosomes in vivo. Cell 160, 1145–1158 (2015). 20. B. R. Sabari et al., Coactivator condensation at super-enhancers links phase separation and gene 57. T. Nozaki et al., Dynamic organization of chromatin domains revealed by super-resolution live-cell control. Science 361 (2018). imaging. Mol. Cell 67, 282–293 e287 (2017). 21. A. J. Plys et al., Phase separation of Polycomb-repressive complex 1 is governed by a charged 58. S. S. Ashwin, T. Nozaki, K. Maeshima, M. Sasai, Organization of fast and slow chromatin revealed by disordered region of CBX2. Genes Dev 33, 799–813 (2019). single-nucleosome dynamics. Proc. Natl. Acad. Sci. U.S.A. 116, 19939–19944 (2019). 22. L. Wang et al., Histone modifications regulate chromatin compartmentalization by contributing to a 59. J. Otterstrom et al., Super-resolution microscopy reveals how histone tail acetylation affects DNA phase separation mechanism. Mol. Cell 76, 646–659.e646 (2019). 23. J. M. Eeftens, M. Kapoor, D. Michieletto, C. P. Brangwynne, Polycomb condensates can promote epigenetic marks but are not required for sustained chromatin compaction. Nat. Commun. 12, 5888 (2021). 24. D. L. J. Lafontaine, J. A. Riback, R. Bascetin, C. P. Brangwynne, The nucleolus as a multiphase liquid condensate. Nat. Rev. Mol. Cell Biol. 22, 165–182 (2021). 25. D. S. W. Lee, N. S. Wingreen, C. P. Brangwynne, Chromatin mechanics dictates subdiffusion and compaction within nucleosomes in vivo. Nucleic Acids Res. 47, 8470–8484 (2019). 60. M. Lakadamyali, M. P. Cosma, Visualizing the genome in high resolution challenges our textbook understanding. Nat. Methods 17, 371–379 (2020). 61. Y. Itoh, E. J. Woods, K. Minami, K. Maeshima, R. Collepardo-Guevara, Liquid-like chromatin in the cell: What can we learn from imaging and computational modeling? Curr. Opin. Struct. Biol. 71, 123–135 (2021). coarsening dynamics of embedded condensates. Nat. Phys. 17, 531–538 (2021). 62. J. Lerner et al., Two-parameter mobility assessments discriminate diverse regulatory factor behaviors 26. J. C. Hansen, K. Maeshima, M. J. Hendzel, The solid and liquid states of chromatin. Epigenetics in chromatin. Mol Cell 79, 677–688.e676 (2020). Chromatin 14, 50 (2021). 63. P. A. Gómez-García et al., Mesoscale modeling and single-nucleosome tracking reveal remodeling of 27. H. Strickfaden et al., Condensed chromatin behaves like a solid on the mesoscale in vitro and in clutch folding and dynamics in stem cell differentiation. Cell Reports 34, 108614 (2021). living cells. Cell 183, 1772–1784.e1713 (2020). 64. H. Hajjoul et al., High-throughput chromatin motion tracking in living yeast reveals the flexibility of 28. Q. Chen et al., Chromatin liquid-liquid phase separation (LLPS) is regulated by ionic conditions and the fiber throughout the genome. Genome Res 23, 1829–1838 (2013). fiber length. Cells 11 (2022). 65. V. Levi, Q. Ruan, M. Plutz, A. S. Belmont, E. Gratton, Chromatin dynamics in interphase cells revealed 29. M. Feric et al., Coexisting liquid phases underlie nucleolar subcompartments. Cell 165, 1686–1697 by tracking in a two-photon excitation microscope. Biophys. J. 89, 4275–4285 (2005). (2016). 66. T. Nozaki et al., Dynamic organization of chromatin domains revealed by super-resolution live-cell 30. T. Mittag, R. V. Pappu, A conceptual framework for understanding phase separation and addressing imaging. Mol. Cell 67, 282–293.e287 (2017). open questions and challenges. Mol. Cell 82, 2201–2214 (2022). 67. J. R. Chubb, S. Boyle, P. Perry, W. A. Bickmore, Chromatin motion is constrained by association with 31. D. W. Sanders et al., Competing protein-RNA interaction networks control multiphase intracellular nuclear compartments in human cells. Curr. Biol. 12, 439–445 (2002). organization. Cell 181, 306–324.e328 (2020). 68. A. Zidovska, D. A. Weitz, T. J. Mitchison, Micron-scale coherence in interphase chromatin dynamics. 32. P. Yang et al., G3BP1 is a tunable switch that triggers phase separation to assemble stress granules. Proc. Natl. Acad. Sci. U.S.A. 110, 15555–15560 (2013). Cell 181, 325–345.e328 (2020). 33. J. O. Park et al., Metabolite concentrations, fluxes and free energies imply efficient enzyme usage. Nat. Chem. Biol. 12, 482–489 (2016). 34. T. Mitchison, M. Kirschner, Dynamic instability of microtubule growth. Nature 312, 237–242 (1984). 35. A. Elie et al., Tau co-organizes dynamic microtubule and actin networks. Sci. Rep. 5, 9964 (2015). 36. M. A. Mendes, J. M. Chies, A. C. de Oliveira Dias, S. A. Filho, M. S. Palma, The shielding effect of glycerol against protein ionization in electrospray mass spectrometry. Rapid. Commun. Mass Spectrom. 17, 672–677 (2003). 69. H. Chen et al., Dynamic interplay between enhancer-promoter topology and gene activity. Nat. 70. Genet. 50, 1296–1303 (2018). I. F. Davidson, J. M. Peters, Genome folding through loop extrusion by SMC complexes. Nat. Rev. Mol. Cell Biol. 22, 445–464 (2021). 71. Y. Kim, Z. Shi, H. Zhang, I. J. Finkelstein, H. Yu, Human cohesin compacts DNA by loop extrusion. Science 366, 1345–1349 (2019). 72. M. H. Hauer, S. M. Gasser, Chromatin and nucleosome dynamics in DNA damage and repair. Genes. Dev. 31, 2204–2221 (2017). 37. K. Luger, T. J. Rechsteiner, T. J. Richmond, Expression and purification of recombinant histones and 73. J. R. Abney, B. Cutler, M. L. Fillbach, D. Axelrod, B. A. Scalettar, Chromatin dynamics in interphase nucleosome reconstitution. Methods Mol. Biol. 119, 1–16 (1999). 38. O. Perisic, R. Collepardo-Guevara, T. Schlick, Modeling studies of chromatin fiber structure as a function of DNA linker length. J. Mol. Biol. 403, 777–802 (2010). 39. D. Lohr, K. E. Van Holde, Organization of spacer DNA in chromatin. Proc. Natl. Acad. Sci. U.S.A. 76, 6326–6330 (1979). nuclei and its implications for nuclear structure. J. Cell Biol. 137, 1459–1468 (1997). 74. T. Cremer et al., Rabl’s model of the interphase chromosome arrangement tested in Chinese hamster cells by premature chromosome condensation and laser-UV-microbeam experiments. Hum. Genet. 60, 46–56 (1982). 75. D. Gerlich et al., Global chromosome positions are transmitted through mitosis in mammalian cells. 40. K. Brogaard, L. Xi, J. P. Wang, J. Widom, A map of nucleosome positions in yeast at base-pair Cell 112, 751–764 (2003). resolution. Nature 486, 496–501 (2012). 76. M. A. Lever, J. P. H. Th’ng, X. Sun, M. J. Hendzel, Rapid exchange of histone H1.1 on chromatin in 41. L. N. Voong et al., Insights into nucleosome organization in mouse embryonic stem cells through living human cells. Nature 408, 873–876 (2000). chemical mapping. Cell 167, 1555–1570.e1515 (2016). 77. H. Kimura, P. R. Cook, Kinetics of core histones in living human cells: Little exchange of H3 and H4 42. T. Schalch, S. Duda, D. F. Sargent, T. J. Richmond, X-ray structure of a tetranucleosome and its and some rapid exchange of H2B. J. Cell Biol. 153, 1341–1353 (2001). implications for the chromatin fibre. Nature 436, 138–141 (2005). 78. E. Meshorer et al., Hyperdynamic plasticity of chromatin proteins in pluripotent embryonic stem 43. F. Song et al., Cryo-EM study of the chromatin fiber reveals a double helix twisted by cells. Dev. Cell 10, 105–116 (2006). tetranucleosomal units. Science 344, 376–380 (2014). 79. T. Higashi et al., Histone H2A mobility is regulated by its tails and acetylation of core histone tails. 44. X. Lu, J. M. Klonoski, M. G. Resch, J. C. Hansen, In vitro chromatin self-association and its relevance Biochem. Biophys. Res. Commun. 357, 627–632 (2007). to genome architecture. Biochem. Cell Biol. Biochimie et Biol. Cell. 84, 411–417 (2006). 80. A. S. Hansen, I. Pustova, C. Cattoglio, R. Tjian, X. Darzacq, CTCF and cohesin regulate chromatin loop 45. C. Joo, T. Ha, Preparing sample chambers for single-molecule FRET. Cold Spring Harb. Protoc. 2012, stability with distinct dynamics. Elife 6 (2017). 1104–1108 (2012). 81. K. Rippe, Liquid-liquid phase separation in chromatin. Cold Spring Harb. Perspect. Biol. 14, 46. J. A. Ditlev et al., A composition-dependent molecular clutch between T cell signaling condensates a040683 (2021), 10.1101/cshperspect.a040683. and actin. eLife 8 (2019). 47. Q. Zheng, S. Jockusch, Z. Zhou, S. C. Blanchard, The contribution of reactive oxygen species to the photobleaching of organic fluorophores. Photochem. Photobiol. 90, 448–454 (2014). 82. B. A. Gibson, et al., In diverse conditions intrinsic chromatin condensate have liquid-like material properties. Dryad. https://datadryad.org/stash/share/MOxYnmIkc6BaV2r44QUuIvM-uHFZrRA- ggA2VClmL0c. Deposited 13 April 2023. 10 of 10 https://doi.org/10.1073/pnas.2218085120 pnas.org
10.1073_pnas.2108421118	BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model Ye Hea,1, Joshua Riverab,1, Miklos Diossyc,d, Haohui Duanb, Christian Bowman-Colina, Rachel Reeda, Rebecca Jenningse, Jesse Novake, Stevenson V. Tranb, Elizabeth F. Cohena, David Szutsf, Anita Giobbie-Hurderg, Roderick T. Bronsone, Adam J. Bassh, Sabina Signorettie, Zoltan Szallasic,d, David M. Livingstona,2, and Shailja Pathaniab,2 aDepartment of Cancer Biology, Dana-Farber Cancer Institute, Boston, MA 02215; bCenter for Personalized Cancer Therapy, University of Massachusetts Boston, Boston, MA 02125; cDepartment of Statistics, Danish Cancer Society Research Center, Copenhagen, 2100, Denmark; dComputational Health Informatics Program, Boston Children’s Hospital, Boston, MA 02215; eDepartment of Pathology, Brigham and Women’s Hospital, Boston, MA, 02215; fInstitute of Enzymology, Research Centre for Natural Sciences, Hungarian Academy of Sciences, Budapest, 1117, Hungary; gDivision of Biostatistics, Department of Data Science, Dana-Farber Cancer Institute, Boston, MA 02215; and hDivision of Hematology and Oncology, Columbia University/Herbert Irving Comprehensive Cancer Center, New York, NY 10032 Contributed by David M. Livingston, August 20, 2021 (sent for review May 4, 2021); reviewed by Ronny Drapkin and Shyam K. Sharan BRCA1 germline mutations are associated with an increased risk of breast and ovarian cancer. Recent findings of others suggest that BRCA1 mutation carriers also bear an increased risk of esophageal and gastric cancer. Here, we employ a Brca1/Trp53 mouse model to show that unresolved replication stress (RS) in BRCA1 heterozygous cells drives esophageal tumorigenesis in a model of the human equivalent. This model employs 4-nitroquinoline-1-oxide (4NQO) as an RS-inducing agent. Upon drinking 4NQO-containing water, Brca1 heterozygous mice formed squamous cell carcinomas of the distal esophagus and forestomach at a much higher frequency and speed (∼90 to 120 d) than did wild-type (WT) mice, which remained largely tumor free. Their esophageal tissue, but not that of WT con- trol mice, revealed evidence of overt RS as reflected by intracellular CHK1 phosphorylation and 53BP1 staining. These Brca1 mutant tu- mors also revealed higher genome mutation rates than those of control animals; the mutational signature SBS4, which is associated with tobacco-induced tumorigenesis; and a loss of Brca1 heterozy- gosity (LOH). This uniquely accelerated Brca1 tumor model is also relevant to human esophageal squamous cell carcinoma, an often lethal tumor. BRCA1 \| replication stress \| haploinsufficiency \| mouse model Germline BRCA1 mutations predispose humans to an ele- vated cancer risk and especially that of the breast and ovary (1, 2). In recent times the suggestion of an increased risk of esophageal cancer in BRCA1 mutation carriers has also been reported, e.g., in an individual with a germline BRCA1 mutation (3). Also, a complete clinical response to platinum treatment was observed in a patient with BRCA1 mutant esophageal cancer (4). Furthermore, the overall esophageal squamous cell carcinoma (ESCC) risk in BRCA1 carriers is significant (relative risk [RR] of 2.9 [95% CI 1.1 to 6.0]) (5, 6). This is in keeping with the obser- vation that a relatively frequent loss of heterozygosity (LOH) is detected in the BRCA1-containing region of chromosome 17 in squamous cell carcinoma of the esophagus (7–9). BRCA1 maintains genome integrity by engaging in multiple cellular processes, including the repair of DNA damage (10, 11), including double strand breaks (DSBs), stalled replication forks, and other abnormalities. Stalled forks, when not resolved, can lead to mutations or can collapse into DSBs (12–15). Both outcomes are components of what is commonly referred to as replication stress (RS), which, when chronic, can serve as a cancer driving force (16–18). Loss of certain DNA damage repair functions in BRCA1 mutant tumor cells also renders these cells sensitive to platinum- based derivatives and PARP inhibitors (19, 20). Success of these agents in suppressing BRCA1 mutant tumor growth has made them therapeutic agents of choice for treating BRCA1 mutant cancer (21, 22). Loss of BRCA1 function either by germline deletion and/or promoter hypermethylation is now a predictive classifier of response to these agents (23). Currently, multiple Brca1 mouse models facilitate the study of BRCA1 loss-associated tumorigenesis. Complete loss of BRCA1 is embryonically lethal (24); thus, successful, tumorigenic models either conditionally delete both alleles of Brca1 in a tissue of interest or express a hypomorphic mutant version of Brca1 (25–27). For example, conditional Brca1 loss can be driven by Cre-mediated deletion of two Brca1 floxed alleles in a tissue of choice. And mice bearing hypomorphic Brca1 mutant alleles, like S C I T E N E G Significance Although germline heterozygous BRCA1 mutations predispose human carriers to cancer, heterozygous mouse BRCA1 muta- tions do not. We find that exposure to a source of upper gas- trointestinal replication stress (RS) elicited a marked cancer incidence in Brca1+/−;Trp53+/− heterozygous mice but not in wild-type mice (Brca1+/+;Trp53+/− ). Oral delivery of 4 nitro- quinoline-1-oxide induced esophageal epithelial RS, an in- creased esophageal mutation rate, loss of esophageal Brca1 heterozygosity (LOH), and accelerated esophageal tumorigen- esis. These data underscore the necessity of combining other- wise nontumorigenic BRCA1 heterozygosity with simultaneous induction of RS for the generation of not only BRCA1 mutant esophageal cancer, but also a dramatically accelerated form thereof. These results strongly imply that RS is both a major contributor to BRCA1 cancer development and a marked accelerant thereof. Author contributions: D.M.L. and S.P. designed research; Y.H., J.R., H.D., R.R., R.J., J.N., S.V.T., and S.P. performed research; J.R., M.D., C.B-C., E.F.C., D.S., A.G-H., R.T.B., A.J.B., S.S., Z.S., D.M.L., and S.P. analyzed data; and D.M.L. and S.P. wrote the paper. Reviewers: R.D., University of Pennsylvania; S.K.S., National Cancer Institute Competing interest statement: D.M.L. is a scientific advisor to Constellation Pharma, a Science Partner of Nextech Ventures (Zurich, CH), a Science Advisory Board member of the Sidney Kimmel Cancer Center (Johns Hopkins School of Medicine), a Science Advisor to the Pezcoller Foundation (Trento, Italy), and a Special Advisor to the Director of Break Through Cancer, none of which poses a conflict of interest with the substance of this paper. This open access article is distributed under Creative Commons Attribution-NonCommercial- NoDerivatives License 4.0 (CC BY-NC-ND). 1Y.H. and J.R. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/ doi:10.1073/pnas.2107208118/-/DCSupplemental. Published October 4, 2021. PNAS 2021 Vol. 118 No. 41 e2108421118 https://doi.org/10.1073/pnas.2108421118 \| 1 of 11 delta 11, and those that express an incompletely functioning or a truncated version of the Brca1 protein instead of the full-length polypeptide also develop Brca1 tumors (26, 28, 29). Often, BRCA1 knockout (KO) mice also incur a loss of p53 function, which results in accelerated tumor formation, and, in the case of hypomorphic Brca1 mutant mice, their survival as well (25, 27, 30). Mice in both models develop tumors, including mammary and ovarian tumors, which, on average, take ∼1 y to develop. Brca1 delta 11 (Brca1Δ11/Δ11 ) mice, which synthesize a markedly deleted but still partially functional Brca1 allele, also form esophageal tu- mors that can be accelerated by addition of the oxidative stress– inducing agent methyl-N-amylnitrosamine (MNAN) to their drinking water (31). However, the role of BRCA1 haploinsufficiency has not been extensively evaluated in mouse models. One reason is that Brca1 heterozygous mice did not generate tumors more often or faster than their wild-type (WT) counterparts (27, 32). This is unlike the increased predisposition to cancer observed in human BRCA1 mutation carriers (who carry a germline loss-of-function mu- tation in a single BRCA1 allele). They manifest a significantly greater than normal breast and/or ovarian cancer incidence by age 70 (1, 2). BRCA1 haploinsufficiency can be linked to increased genomic instability, in part, because of its defective participation in stalled fork repair and replication stress suppression (33) and, possibly, because of its role in regulating SIRT1 levels and affecting pRb pathway activation (34). Given the importance of RS develop- ment in tumorigenesis (16, 18, 35), this effect would be a logical contributor to BRCA1 mutant cancer development. Of note, BRCA1 heterozygous human cells are haploinsufficient for RS suppression (33), raising the possibility that this defect operates as a general contributor to the increased tumorigenicity observed in many germline BRCA1 heterozygous families. To test this hypothesis, we have established a Brca1 mutant esophageal mouse cancer model that is capable of addressing the role of replication stress accumulation in BRCA1 mutant cancer. Here one allele of Brca1 and one of Trp53 were deleted through the action of Meox2Cre, which acts very early during embryogenesis (embryonic day 5 [E5]) (36) and results in the development of Brca1 and p53 heterozygosity in all tissues. Using this mouse model, we have found that BRCA1 deficiency in replication stress suppression is enhanced by exposure to 4-nitroquinoline-1-oxide (4NQO) in BRCA1 heterozygous tissue where it serves as an efficient and ab- normally rapid driver of tumor formation. Results 4NQO Induces Replication Stress in Mice Bearing Germline Conditional Brca1 and Trp53 Alleles. Unlike the increased predisposition to cancer observed in human BRCA1 mutation carriers, heterozygous Brca1 mouse models do not manifest an increased cancer incidence (27). Given that 1) BRCA1 heterozygous cells (including Brca1 heterozygous mouse cells) are defective in stalled replication fork repair and RS suppression (33), and that 2) chronic RS is a general contributor to tumorigenesis in certain mammalian species (16, 18, 35), we reasoned that, if RS were a necessary driver of murine Brca1 mutant cancer, then increasing RS above the levels observed in Brca1 heterozygous mouse tissue might induce tu- morigenesis in this tissue and not in the same tissue of wild-type control mice. Conceivably, in a typically short mouse life span (∼18 to 24 mo), insufficient, indigenous RS accumulates in Brca1 heterozygous animals to elicit tumorigenesis. To investigate the role of RS in BRCA1 mutant tumorigenesis, we established a Brca1 heterozygous mouse model where RS was actively heightened, pharmacologically. Specifically, 4NQO, a known RS-inducing agent (37), was administered to mice by adding it to their drinking water. 4NQO is a known carcinogen, which primarily forms DNA adducts at guanine residues (38, 39). More specifically, conditional loss-of-function Brca1 and Trp53 mice were studied, wherein loxP sites flanked Brca1 exons 5 to 13 and Trp53 exons 2 to 10 (Fig. 1A) (27). We generated four different mouse cohorts to address the role of BRCA1 heterozygosity in tu- mor formation: Brca1flox/wt;Trp53flox/wt;Meox2Cre (hereafter referred to as BPM), Trp53flox/wt;Meox2Cre (PM), Brca1flox/wt;Meox2Cre (BM), and WT controls that also bore a Meox2Cre allele (Fig. 1B). In these mice, Cre recombinase was expressed under the control of the endogenous Meox2 promoter, which has been shown to express as early as E5 (36). The early expression of Meox2Cre insured early deletion of the floxed Brca1 and/or Trp53 alleles from all mouse tissues. We confirmed the deletion of the floxed Brca1 and Trp53 alleles by genotyping tail cuts from these mice (Fig. 1C). All mice generated in this study were healthy and fertile. To determine whether cells derived from mouse tissue were susceptible to 4NQO-induced replication stress, we studied mouse embryonic fibroblasts (MEFs). Upon replication stress develop- ment, stalled replication forks become coated with phosphorylated replication protein A (pRPA32), which is a prerequisite for efficient stalled replication fork repair (40). We showed earlier that BRCA1 heterozygous cells are defective in loading pRPA32 on chromatin following hydroxyurea (HU)-induced replication stress (33). We first tested whether 4NQO is similarly capable of inducing replication stress. To address this question, we analyzed esophageal squamous cell carcinoma cell line KYSE410 and studied the loading of pRPA32 on chromatin as a manifestation of replication stress induction following 4NQO treatment. As shown in Fig. 1D, 4NQO induced pRPA32 (S33) loading in control cells (siLuc-treated cells), but not in siBRCA1- treated KYSE410 cells (Fig. 1D) confirming that, as shown before with other RS-inducing agents like hydroxy- urea and ultraviolet (UV) (33, 41), pRPA32 loading following 4NQO treatment is BRCA1 dependent. We next asked whether this is true for mouse cells as well. We found that MEFs derived from Brca1 heterozygous mice (Brca1+/− ), but not those derived from Brca1 wild-type (Brca1+/+ ) MEFs, are defective in loading pRPA32 on chromatin following the induction of 4NQO-induced replication stress (Fig. 1E). These results confirmed that 4NQO is indeed a RS-inducing agent and that mouse Brca1 heterozygous mouse cells manifest a defective replication stress response, just like BRCA1 heterozygous human cells (33). Increased Replication Stress in Forestomach Tissue of BPM Mice Exposed to Oral 4NQO. Previous studies have employed two dif- ferent modes of introducing 4NQO into mice to study carcinogen- induced tumor formation. 4NQO was either “painted” on to the tongues of mice or was added to their drinking water (42). We employed the latter approach. Having established that 4NQO induces replication stress in MEFs, we next asked whether 4NQO exposure elicits replication stress in vivo. Specifically, we analyzed mouse tissue and whether Brca1 heterozygous mouse tissue is especially susceptible to accumulating such DNA damage. Given that we were introducing 4NQO into mice via their drinking water, we focused on the state of their esophageal/ forestomach tissue. Two mice from each of the four, afore- mentioned mouse cohorts drank water containing 200 μg/mL of 4NQO for 2 d. The other animals served as controls. Mice were killed 24 h postoral 4NQO treatment, and their esophageal/ forestomach tissue was harvested and analyzed for signs of rep- lication stress. These sections were stained for the protein 53BP1 (SI Appendix, Fig. S1A), a marker of replication stress (43, 44), and a clear increase in 53BP1-positive nuclei was observed in tissue sections collected from the 4NQO-treated mice. Impor- tantly, BPM mice revealed a higher incidence of strong 53BP1- positive nuclei compared with mice in the other cohorts (BM, PM, and WT) (Fig. 1 F and G). Stained sections were analyzed using ImageJ and Matlab, and 53BP1-positive pixel values for each image were determined by measuring levels that were above 2 of 11 \| PNAS https://doi.org/10.1073/pnas.2108421118 He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model A Brca1 Trp53 1 2 3 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 1 2 3 4 5 6 7 8 9 10 11 LoxP sites B Genotype Brca1flox/wt;p53flox/wt;Meox2cre (BPM) Brca1wt/wt;p53flox/wt;Meox2cre (PM) Brca1flox/wt;p53wt/wt ;Meox2cre (BM) Brca1wt/wt;p53wt/wt ;Meox2cre (WT) siBRCA1 siLuc + + siLuc - - siBRCA1 E (PM) (BPM) EP1-2 EP1-1 4NQO - + - + C WT PM BM BPM D 4NQO Brca1 del Brca1 flox Brca1 wildtype p53 del p53 flox p53 wildtype Meox2+cre Meox2 wildtype +4NQO WT PM BM BPM F 1 P B 3 5 BRCA1 pRPA32 (S33) Lamin B1 G y t i s n e t n I 1 P B 3 5 r a e l c u N e g a r e v A 180 160 140 120 100 pRPA32 (S4/S8) Lamin B1 p= < 0.0145 * p= < 0.0001 ** p= 0.0017 S C I T E N E G T W M P M B M P B ) M P B ( O Q N 4 o N WT PM BM BPM Forestomach/Esophagus Sections H E & H Fig. 1. Use of 4NQO as a RS-inducing agent in mice carrying conditional Brca1 and Trp53 alleles. (A) Schematic representation of the location of loxP sites in the Brca1 and Trp53 loci in the mouse genome. (B) List of genotypes (BPM, PM, BM, and WT) used in this study. (C) PCR-based analysis to confirm germline deletion of the floxed segments of Brca1 and Trp53 by Meox2-driven Cre recombinase. Genomic DNA was extracted from mouse tail cuts and amplified using primers for the floxed and/or the deleted region of the gene. Details of the PCR primers and their locations are provided in Materials and Methods. A representative gel depicting the presence of WT, deleted (del), floxed (flox), and Meox2-Cre alleles is shown for all four genotypes (WT, PM, BM, and BPM). (D) Western blot analysis of the human esophageal squamous cell cancer line, KYSE410 nuclear extracts. The extracts were blotted for phosphorylated RPA (pRPA32, phosphorylated at S33), BRCA1, and Lamin B1 (loading control). pRPA32 accumulation was studied before and after 4NQO treatment in control (siLuc) and BRCA1-depleted (siBRCA1) cells. (E) Western blot analysis of phosphorylated, chromatin-associated RPA32 (pRPA32, phosphorylated at S4/S8) in extracts isolated from 4NQO-treated and -untreated MEFs. MEFs from Brca1flox/wt;Meox2Cre (Brca1+/− ) and Brca1wt/wt;Meox2Cre (Brca1+/+) mice were cultured in the presence and absence of 4NQO for 3 h. In both D and E, cells were harvested, and equivalent amounts of lysate were electrophoresed, blotted, and the blot probed with an anti-pRPA32 antibody. A nonspecific band (D) and Lamin B1 (E) served as loading controls. (F and G) Effect of 4NQO treatment (4NQO delivered via the drinking water) on replication stress in BPM mice. Accumulation of 53BP1 in nuclei was used to assess induction of replication stress. Stomach sections were collected from mice who drank water with and without 4NQO for 2 d and analyzed for accumulation of 53BP1 by IHC. Forestomach/esophageal tissue sections from WT, PM, BM, and BPM mice (n = 2 mice for each genotype) were collected upon 4NQO exposure and stained for 53BP1 (IHC). Images were captured at 40X magnification and analyzed with ImageJ and MatLab_R2019b software. Mice heterozygous for Brca1 and Trp53 (BPM) showed enriched mean pixel intensity for 53BP1 compared with other genotypes. A two-tailed paired t test was used to determine statistical significance. (H) H&E staining of esophageal tissue. All four genotypes (WT, BM, PM, and BPM) show normal esophageal histology. (Scale bars in black, 200 mm.) He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model PNAS \| 3 of 11 https://doi.org/10.1073/pnas.2108421118 background (>150 pixels/μm2). These values were averaged for each sample (Fig. 1G). Higher nuclear 53BP1 pixel intensities in BPM tissue com- pared with all the other genotypes, suggests that dual Brca1/p53 heterozygosity resulted in increased replication stress in the nuclei of BPM tissue compared with BM, PM, and/or WT tissue. Although there was an increased accumulation of 53BP1-positive nuclei in BPM mice, the hematoxylin and eosin (H&E) stains confirmed that such a short exposure to 4NQO (2 d) did not affect the normal histology of the esophageal tissue in mice from any of the four genotypes (Fig. 1H). Brca1;Trp53 Double Heterozygous BPM Mice Are Prone to Esophageal and Forestomach Cancer when Faced with the Development of Replication Stress. Having confirmed that there is increased rep- lication stress in Brca1/p53 double heterozygous mouse esophagus/ forestomach tissue upon 4NQO exposure, we then asked whether increased replication stress in the esophagus/stomach tissue of these mice is tumorigenic. Prior results of others have suggested that RS is potentially tumorigenic (45, 46). To study whether the codevelopment of Brca1 heterozygosity and upper gastrointestinal (GI)-focused replication stress is tumorigenic in the esophagus, we exposed mice to 4NQO-containing drinking water. The animals were administered 200 μg/mL 4NQO in drinking water for a pe- riod of 75 d and then followed for 45 d during which the animals received only normal, drug-free drinking water (Fig. 2A). A common water stock was prepared and was equally dis- tributed in drinking bottles. Bottles were replaced every 2 d. Mice from the various genotypes (BPM, PM, BM, and WT) (Fig. 2B) were housed in the same cage (four mice per cage) and all drank from the same water source. This ensured that all mice in our study were equivalently exposed to 4NQO, which was dis- solved in a polyethylene glycol (PEG)/H2O solution. Control mice received PEG-containing water as a form of vehicle-only exposure. Equivalent exposure to 4NQO was confirmed by the observation that the incidence of benign tongue papillomatosis, a common occurrence in mice that drink 4NQO-containing water, was similar in mice of different genotypes (SI Appendix, Fig. S1B). All mice were killed at the end of a 45-d observation period, and their stomach, esophagus, and tongue tissue were harvested for analysis. We worked with n = 107 mice divided among four genotypes, as detailed in Fig. 2B. They were scored for tumor formation at the time of death. We observed a significantly higher incidence of esophageal/forestomach tumorigenesis in BPM mice after 4NQO treatment compared with control mice (WT) (Fig. 2 C and D). Forestomach sections were further analyzed to determine the A B C D Fig. 2. Brca1flox/wt;p53flox/wt;MeoX2cre double heterozygous mice (BPM) are prone to esophageal and/or forestomach cancer upon exposure to replication stress induced by 4NQO. (A) Schematic representation of the animal drug treatment protocol that was employed in this study. (B) The table lists the four genotypes (BPM, PM, BM, and WT) and the number of mice per cohort that were used in this study. (C) Representative macroscopic images that show formation of esophageal/forestomach tumors in BPM mice and lack of such tumors in WT mice after undergoing the same 4NQO treatment as detailed in A. (D) Macroscopic images of the forestomach in WT vs. BPM mice that underwent 4NQO treatment as detailed in A. WT mice show mostly tumor-free for- estomach and BPM mice show large fused tumors in the forestomach. Detailed microscopic analysis of WT and BPM forestomach is provided in Fig. 3. 4 of 11 \| PNAS https://doi.org/10.1073/pnas.2108421118 He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model S C I T E N E G incidence of papilloma and/or tumor development (Fig. 3 A and B). After accounting for death arising from unknown causes during the course of the experiment, we analyzed n = 97 mice belonging to four different genotype cohorts: BPM (n = 31), BM (n = 15), PM (n = 23), and WT (n = 28) (Fig. 3C) in search of tumorigenesis in each cohort. Given that both male and female mice were present in this study, we first tested whether males or females were affected differently. No tumor development gender bias was observed. Moreover, no mice in the PEG control arm (n = 16) formed tumors. By contrast, when different genotypes were compared for 4NQO-associated tumorigenesis, only Brca1;Trp53 double heterozygous (BPM) mice revealed a significantly increased in- cidence of tumorigenesis compared with WT mice (odds ratio [OR]: 8.7, 95% CI: 1.8 to 42.4, P = 0.008) (Fig. 3 D and E). Although BPM mice revealed the highest frequency of tu- morigenesis compared with other genotypes, they revealed low levels of oral papilloma development. In contrast to this obser- vation, BM mice exhibited an increased incidence of oral pap- illomatosis relative to the other genotypes (Fig. 3 C and E). One possible explanation for the relatively low-level papilloma for- mation in BPM mice compared with the relatively high incidence in BM mice is that the loss of p53 in BPM mice allowed their precancerous cells to progress further along the path to tumor- igenesis than did BM cells (which have lost Brca1 but expressed WT p53). Esophageal Squamous Cell Carcinoma Reveals Krt14 and Phosphorylated Chk1 Expression. The increased expression of keratins such as cyto- keratin 14 (K14) is a known manifestation of esophageal carcino- genesis, especially in human squamous cell esophageal carcinoma (47, 48). Thus, we analyzed K14 expression in BPM esophageal they were strongly K14 positive tumors and confirmed that (Fig. 4A). PM tumors were similarly positive for K14 staining. We also tested whether phosphorylated CHK1, a reporter of DNA damage and of replication stress, was expressed in these tumors. Previous studies had shown that, in response to repli- cation stress, ATR phosphorylates CHK1 kinase at S317 and S345 (49). To determine whether tumors of any one genotype experi- enced higher levels of a given replication stress marker than others, we analyzed pChk1(S345) in esophageal tumors in mice of different genotypes (Fig. 4B). The overall percentage of cells with nuclear pChk1(S345) staining was significantly higher in BPM tumors compared with WT or BM or PM tumors (Fig. 4 B and C). This implies that higher levels of replication stress and, likely, DNA damage were present in Brca1/Trp53 double heterozygous (BPM) tumor cells than in tumor cells that appeared in any of the other cohorts. BPM tumors also stained positively for 53BP1, signaling signs of DNA damage in these tumors (Fig. 4D). We also checked p63 status in BPM tumors, given that one of the traditional markers of esophageal squamous cell carcinoma is p63 (50). BPM tumor sections manifested strong p63 staining (Fig. 4D), while normal esophageal sections from BPM mice did not (SI Appendix, Fig. S1C). Whole-Exome Sequencing Reveals an Increased C > A Mutation Frequency in 4NQO-Induced Tumors in BPM Mice. To determine the mutational load and the abundance of relevant genomic al- terations in tumors from BPM compared with PM mice, we carried out whole-exome sequencing (WES). Given that mutational load closely correlates with tumor ag- gressiveness (51), we asked whether there were any differences in the nature and/or the extent of mutational signatures in tumors from one genotype compared with another. A sample list is shown in Fig. 5A. Genomic DNA used for WES was extracted from microdissected tumor sections of BPM mice (samples S1 and S2), PM mice (samples S4 and S5), of papilloma tissue from a BPM mouse (sample S7), and tissue from a normal forest- omach/esophageal region of BPM mice (samples S3 and S6) (Fig. 5A). We analyzed the WES data in an effort to assess the roles of certain mutational signatures that exist in the relevant tumor material. Homologous recombination deficiency (HRD) signatures are enriched in cancers harboring biallelic inactivation of BRCA1 or BRCA2 (52). An HRD-related mutational signature, referred to as COSMIC signature 3, is characterized by a specific single nu- cleotide variation pattern and a distinct pattern of insertions and deletions. There was no accumulation of signature 3 in any of the tumor samples, including Brca1 mutant tumors (BPM). Previous studies with Brca1 mutant mouse tumors also have failed to detect the strong accumulation of an HRD signature unlike what was detected in human BRCA1 mutant tumors (53–55). However, the WES analysis did reveal that C > A mutations were four- to fivefold more abundant in the 4NQO-induced tu- mors in BPM mice compared with PM mice (Fig. 5B). These mutations were observed only in tumor tissue as opposed to tissue of the same genotype that had been exposed to 4NQO but remained tumor free (e.g., in samples S3 and S7). These findings are consistent with the clonal nature of the tumor cells and suggest that cells harboring C > A mutations were particularly prone to becoming tumorigenic of all of the cell types that were analyzed. Moreover, a four- to fivefold increase in C > A mutations in BPM tumors, compared with PM tumors inefficient repair of 4NQO-induced strongly suggested that damage in Brca1 heterozygous cells in BPM mice contributed to tumor formation. 4NQO and its metabolites have a strong preference to form adducts with guanine (38). Either 4NQO- guanine adducts, or its metabolite-driven 8-oxo-guanine (8-oxoG) and 8-hydroxydeoxyguanosine (8OHdG) DNA adducts (56, 57) could contribute to increased C > A mutations after 4NQO treatment. 8-oxoG and 8OHdG DNA adducts have been shown previously to accumulate in cells treated with 4NQO, and both are commonly associated with C > A transversions because of mispairing with adenine during DNA replication (58). Furthermore, although there was a significant increase in C > A mutations in tumor cells, the contribution from other base substitutions was relatively low. This pattern is consistent with the presence of a single nucleotide variation-based mutational signature, SBS4 (59), that dominated the other mutational out- comes in the tumor. Of note, this signature accumulates in cancers that are strongly associated with tobacco exposure or appears in cell lines that have been exposed to benzo-o-pyrene, a tobacco smoke carcinogen (59). Both of these agents, like 4NQO, primarily form guanine adducts. We also determined the frequency of indels (deletions and insertions) in these samples. Though we detected a strong cor- relation between 4NQO exposure and an increase in indel fre- quency, we did not detect a tissue genotype-based effect on indel frequency. This suggests that such events (insertions and dele- tions) do not necessarily depend on tissue genotype and are more dependent on exposure of the tissue to a relevant carcinogen. Finally, in keeping with an increased C > A mutation signature in BPM- compared with PM-based tumor cells, we also found that the percentage of genome alterations (i.e., amplifications and deletions) was, on average, greater than twofold higher in BPM tumors than PM tumors (Fig. 5C). Brca1 and Trp53 LOH in Esophageal Tumors Formed following 4NQO Exposure. BRCA1 mutant cancers frequently reveal a loss of BRCA1 heterozygosity (LOH), largely arising from loss of the wild- type allele of BRCA1 and retention of the mutant allele (60–64). Recently, however, some reports have indicated that BRCA1 LOH might not be as common an event as previously suggested in Brca1 tumors (65). Studies where microdissected tumor cells were He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model PNAS \| 5 of 11 https://doi.org/10.1073/pnas.2108421118 A B C a m o l l i p a P l l e C s u o m a u q S a m o n i c r a C Normal Papilloma Tumor N % N % N % 9 4 6 29.0 26.7 26.1 13 46.4 5 8 11 10 16.1 17 54.8 53.3 3 20.0 47.8 6 26.1 35.7 5 17.9 Fisher’s exact = 0.002 Genotype BPM BM PM WT D Association between Genotype and Tumor Development Comparison Odds Ratio 95% CI of Odds Ratio P-value Tumor Tumor Tumor B1wt/wt;p53flox/wt (PM) B1wt/wt;p53wt/wt (WT) B1flox/wt;p53wt/wt (BM) B1flox/wt;p53flox/wt (BPM) B1wt/wt;p53wt/wt (WT) B1wt/wt;p53wt/wt (WT) 2.66 3.28 8.65 0.41 17.4 0.31 0.42 1.77 25.6 0.26 42.4 0.008 E Association between Genotype and Papilloma Development Comparison Odds Ratio 95% CI of Odds Ratio P-value Papilloma B1wt/wt;p53flox/wt (PM) B1wt/wt;p53wt/wt (WT) Papilloma Papilloma B1flox/wt;p53wt/wt (BM) B1flox/wt;p53flox/wt (BPM) B1wt/wt;p53wt/wt (WT) B1wt/wt;p53wt/wt (WT) 2.76 4.45 1.31 0.53 0.81 0.27 14.2 0.23 24.4 0.09 6.3 0.74 Fig. 3. Brca1flox/wt;p53flox/wt;MeoX2cre double heterozygous mice (BPM) are prone to esophageal and/or forestomach carcinoma upon replication stress. (A and B) Microscopic analysis of lesions in the forestomach of mice following 4NQO treatment. Representative histological images (H&E stains) to document the existence of esophageal papilloma (A) and squamous cell carcinoma (B) observed in mice following 4NQO exposure. (A) Papilloma tissue growing upward into the lumen of the esophagus and (B) depicts small nests of squamous tumor cells invading the stroma. It also contains a tumor cyst filled with keratin. (C) Summary of genotype by outcome (normal, papilloma, and/or tumor). The lesions were classified as papilloma or squamous cell carcinoma based on similarity to representative images shown in A and B. (D) Association between genotype and tumor development. The table summarizes the results of a multinomial logistic regression model of tumor development. Compared with WT animals, mice with a Brca1flox/wt;Trp53flox/wt;Meox2Cre genotype (BPM) experienced a statistically significant likelihood of tumor development (OR: 8.7, 95% CI: 1.8 to 42.4, P = 0.008). (E) Association between genotype and papilloma devel- opment. Papilloma incidence in WT mice was compared with that in BM, PM, and BPM mice. 6 of 11 \| PNAS https://doi.org/10.1073/pnas.2108421118 He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model A 4 1 t r K B E & H 1 k h C p C s l l e c e v i t i s o p 1 K H C p f o % BPM non-tumor BPM tumor WT BM PM BPM pChk1 pChk1 pChk1 pChk1 100 80 60 40 20 10 p= 0.0076 p= <0.0001 p= <0.0001 T W M M P B Genotype M P B D WT PM BM BPM E & H 1 P B 3 5 BPM E & H 3 6 p S C I T E N E G Fig. 4. Cytokeratin 14 and phosphorylated CHK1 expression in esophageal squamous cell carcinoma (SCC) of mice. (A) Representative IHC sections to assess keratin 14 (Krt14) expression in forestomach tissue from nontumor and tumor tissue harvested from BPM mice. (B) Representative pictures of PM, BM, and BPM tumors stained with pCHK1 antibody. Upper panels show the H&E of the sections that were stained for pCHK1. Approximate region on the H&E slide that corresponds to the pCHK1-stained section is shown in the boxed region. (Scale bars in black, 200 mm and scale bars in white, 100 mm.) pCHK1 im- munofluorescence analysis reveals increased pCHK1 expression in BPM esophageal/forestomach tumor tissue compared with PM or BM mice. (C) Percentage of pCHK1-positive cells was determined for tumor sections collected from mice of different genotypes. BPM tumor sections contained significantly more pChk1-positive cells per section than BM, PM, and WT cells (D). analyzed for BRCA1 LOH showed that BRCA1 LOH is a het- erogeneous tumor-based outcome (66–68). To address the extent of LOH of Brca1 and Trp53, we extracted genomic DNA from 4NQO-induced tumors and car- ried out qPCR. To enrich for tumor-associated DNA, tumor sections were microdissected in order to isolate tumor-enriched segments. Genomic DNA from tail cuts (TCs) of the corre- sponding mice was also extracted. DNA from the tumor (TU) and the corresponding TC was collected from n = 6 BPM, n = 6 PM, and n = 2 BM and WT mice. Actin primers were used to amplify the genomic actin locus DNA as an internal control. Primers were designed such that they amplified regions within loxP site encompassed segments of Brca1 and Trp53. These pri- mers were used to amplify a specific ∼100-bp region within the respective genes in both TCs and TU genomic DNA extracts from all samples. We chose this region of Brca1 and Trp53 be- cause it allowed one to compare the gene dosage for Brca1 and Trp53 upon Cre-mediated deletion in the germline (i.e., tail cuts) with that in the tumors. The TCs served as positive controls to confirm loss of one allele of Brca1 and/or Trp53 upon action of Cre. The fold change in gene dosage (Brca1 and p53) was cal- culated by normalizing results to those obtained with WT tail cut genomic DNA. As shown in Fig. 5D, qPCR-based analysis showed that BPM mice experienced a reduction in Brca1 gene dosage in the tumors compared with tail cuts, implying the loss of one or both Brca1 alleles in the tumors. No such additional loss of Brca1 or Trp53 was observed in the BM or the WT tumors compared with their tail cut genomic DNA. This could imply that the genomic insta- bility in Brca1 heterozygous (BM) tumors is less than that ob- served in Brca1/Trp53 double het (BPM) mice, where there was additional loss (LOH) for both Brca1 and Trp53. This could also account for reduced tumor rate in BM mice compared with BPM. Furthermore, we did detect a reduced Brca1 gene dosage in PM tumors compared with the corresponding PM tail cuts which, we believe, could be an indication of increased genomic instability in these tumors (Fig. 5D). Similarly, there was also p53 gene dosage loss in BPM compared with BM or WT tumors (Fig. 5E). We also analyzed the WES sequencing data using GATK HaplotypeCaller (69) in genomic vcf mode from the Broad He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model PNAS \| 7 of 11 https://doi.org/10.1073/pnas.2108421118 C E A B D F Fig. 5. DNA mutational signature and Brca1 and Trp53 LOH analysis in the esophageal tumors formed upon exposure to 4NQO. (A) List of mouse tissue samples that were used for WES. Genomic DNA was extracted from microdissected tumor sections (samples S1, S2, S4, and S5), papilloma sections (sample S7), and from the normal forestomach/esophageal region (samples S3 and S6). (B) Number of somatic mutations in the WES samples listed in A, after filtering on the quality, depth, and S-score parameters returned by isomut, and the removal of the clustered germline contaminants. Clusters were identified as a group of variants with a mean distance between the neighboring mutations less than 10% of the average distance between neighboring mutations across the entire exome. (C) Percentage genome altered (amplifications and deletions) was analyzed for samples S1 and S2 (BPM tumors), S4 and S5 (PM tumors), and S3 (BPM mouse, 4NQO exposure, no tumor) and S6 (BPM mouse no 4NQO exposure/no tumor). Each sample was compared with the same normal sample (sample S6). Copy number analysis was performed as described in Materials and Methods. (D and E) Forestomach/esophageal tumor tissue sections were microdissected and genomic DNA was extracted from them. Genomic DNA from TCs of the corresponding mice was also extracted. Tumor and the corresponding tail cut DNA were collected from n = 6 (BPM), n = 6 (PM), and n = 2 BM and WT mice. Actin primers were used to amplify an actin locus as an internal control. Primers for Brca1 and Trp53 were used to amplify the respective genes in both TCs and TU genomic DNA extracted from all relevant samples. The fold change in gene dosage (Brca1 and Trp53) was calculated by normalizing to WT tail cut genomic DNA. The Brca1 gene dosage is plotted in D and the Trp53 gene dosage is plotted in E. P value <0.05 was considered statistically significant (P < 0.05; P < 0.005; P < 0.0005; **P < 0.0001). (F) Variant allele frequency (VAF) analysis for single nucleotide polymorphisms (SNPs) in exon 10 (analogous to human exon 11) of the mouse Brca1 gene. S1 and S2 samples (BPM tumors) reveal a complete loss of the variant at position chr11:101524509 and a near complete loss of the variant at position chr11:101525014. 8 of 11 \| PNAS https://doi.org/10.1073/pnas.2108421118 He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model S C I T E N E G Institute and confirmed results generated through our qPCR analysis. There were two heterozygous single nucleotide poly- morphisms (SNPs) (rs28288696 C/G and rs28288694 A/G) in these specimens that were located in Brca1, both of them in exon 10 (which is contained within the floxed region of Brca1) (Fig. 5F). Compared with their PM counterparts (S4 and S5), the BPM tu- mor samples (S1 and S2) show a strong reduction in coverage at both of these SNPs and the variant allele frequencies (VAFs) were near zero (Fig. 5F), indicating the presence of LOH in the region. In fact, in samples S1 and S2 the SNP rs28288694 completely disappeared, which strongly suggests the existence of LOH in this region. No such disappearance of SNPs or loss of allele frequency was observed in a neighboring gene, ETV4, where allele frequency for five SNPs was analyzed (Fig. 5F). This implies that the dis- appearance of the Brca1 region-specific SNP (rs28288694) in S1 and S2 cannot simply be attributed to a lack of sequencing depth. This is because the probability that the SNPs were present, with a contribution equal to that of the reference allele, but the se- −7, quencing did not capture them, was 7.6 × 10 respectively, assuming the diploidy of the two samples (Fig. 5F). These findings confirm that 1) in BPM tumors (S1 and S2 sam- ples), only Brca1 heterozygous cells contributed to tumor forma- tion and 2) that the loss of SNPs was specific to the Brca1 region. −6 and 9.5 × 10 Discussion Germline heterozygous loss of BRCA1 is associated with an in- creased lifetime cancer risk and, especially, breast and ovarian cancer (1, 70). Increasing evidence suggests that BRCA1 loss is also associated with additional cancer types, including but not limited to pancreatic, prostate, colon, gastric, and esophageal cancer. Despite the well-established association between BRCA1 heterozygosity and cancer predisposition in humans, there are currently no such Brca1 heterozygous mouse models that faithfully recapitulate this high risk of tumor formation upon BRCA1 heterozygosity. Brca1 heterozygous mice are not tumor prone, and complete loss of Brca1 in mouse models by conditionally deleting it in the tissue of interest rarely triggers tumorigenesis, unless Trp53 is also deleted. This makes it difficult to use Brca1 mouse models to study the role of BRCA1 heterozygosity in tumor formation. Here we report a heterozygous Brca1 mouse model that manifests a dramatic acceleration of disease and provides a potentially valu- able tool to study the mechanism of BRCA1 mutant tumorigenesis. We find that BRCA1 heterozygosity in combination with the acute onset of chemically derived replication stress is a major BRCA1 tumor–driving process. We employed 4NQO, a well-established replication stress inducer, and found that exposure to this agent can elicit esophageal tumor formation in Brca1 heterozygous mice. More specifically, oral delivery of 4NQO to doubly heterozy- gous Brca1/Trp53 (BPM) mice led to squamous cell carcinoma of the distal esophagus and forestomach, while no such 4NQO- promoted tumorigenesis developed in WT and PM mice. Thus, dual Brca1 and Trp53 heterozygosity combined with 4NQO rep- lication stress represented an adequate tumorigenic force in the aforementioned animals (33). One interpretation of these results is that 4NQO provides an added source of replication stress that allows a subtumorigenic effect of this DNA damage-associated, BRCA1/p53 heterozygosity-driven process to 1) reach a tumor- inducing level and 2) to do so with remarkable speed. The proposal of a link between achieving sufficient replication stress and Brca1 mutation–driven ESCC is supported by the high level expression of 53BP1 that was detected after persistent and direct exposure of esophageal/forestomach tissue to 4NQO. We also found that BPM-, but not PM-based tumors contained a high percentage of phosphorylated Chk1 (S345)-positive, RS- affected tumor cells, which is also in keeping with a role for chronic replication stress in BRCA1 mutant tumorigenesis and, possibly, in other cancers, as well (16, 18). We also investigated the nature of mutational signatures in tumors formed in BPM and PM mice. Whole-exome sequencing data obtained from tumor and nontumor tissue genomic DNA showed that an SBS4 mutational signature was enriched in 4NQO- induced BPM tumors but not in PM tumors. This mu- tational signature is known to correlate with tobacco smoking and has also been associated with esophageal cancer (10). It is presumed to be an output of unresolved and unrepaired DNA adducts formed by tobacco exposure as well as in 4NQO- associated BPM tumors. Conceivably, if 4NQO leads to similar or related DNA adducts, when unrepaired, they could also give rise to a C > A mutational signature. The fact that we observed a C > A mutational signature almost exclusively in BPM tumors and not in PM tumors strongly sug- gests that defective repair of 4NQO-induced lesions in Brca1/ Trp53 double heterozygous cells contributed to tumor formation. However, why we did not detect a HRD-associated signature 3 in the BPM tumors is as yet unclear. This observation is similar to what has been reported previously for tumor lines derived from mice where both Brca1 alleles were conditionally deleted in the mammary gland. There too, only minor enrichment of signature 3 in the Brca1-deleted tumors was found (53–55). A possible explanation for a lack of signature 3 in Brca1 mouse tumors is that the tumors grew fast enough through largely un- hindered cell divisions, thereby preventing certain subclones from arising and surviving, and thereby contributing to the overall mutational signature of the tumor. LOH analysis of the BPM tumor DNA did reveal a significant reduction in the Brca1 and Trp53 gene dosage compared with DNA from the tail cuts of the corresponding mice. Though significant, Brca1 LOH was not complete, implying the existence of intra- tumoral heterogeneity. This is in keeping with previous reports that have shown evidence that up to 25% of tumor nuclei retain positive BRCA1 staining in BRCA1 mutant human tumors (65, 68). The BPM mouse model also serves as an excellent addition to existing esophageal squamous cell carcinoma mouse models. Distal esophageal and forestomach squamous cell cancers in mice share strong similarity with human ESCC. Previous attempts to model ESCC in mice have employed many different approaches, including genetically engineered mice [e.g., to inactivate p16 or p53 and/or to overexpress cyclin D1 (71)] or the use of carcino- gens. It has been difficult to study ESCC in genetically engineered mice, partly because of its traditionally long latency, which ranges from 12 to 18 mo and normally appears to be necessary for tumor development. However, BPM tumors were detected between ∼90 and 120 d, representing a dramatic acceleration in tumorigenesis. Of note, this mouse model was generated in mice bearing a mixed genetic background (both C57BL/6 and FVB-N) (27) and the effect of genetic backgrounds on tumorigenesis could affect tumor latency in certain backgrounds. The use of carcinogens, including 4NQO, has been helpful in replicating aspects of human ESCC in mice. For example, in the genetically engineered mice (Brca1/Trp53 double heterozygous mice) used in this study, ESCC was induced in a very short time (on average ∼90 to 120 d). By contrast, in 4NQO-free BPM mice, no esophageal tumors were detected within 18 mo, at which time observation was stopped. Moreover, the ∼90- to 120-d onset of tumor formation was directly dependent on both Brca1 hetero- zygosity and the induction of replication stress. Such a telescoped tumor model may offer an opportunity to gain a more rapid un- derstanding of molecular events that lead to ESCC. In conclusion, we have established a markedly accelerated mouse tumor model to study the effect of replication stress in BRCA1 heterozygous tissue and how it contributes to BRCA1 mutant tumorigenesis. Brca1 heterozygous cells are innately sus- ceptible to accumulating exogenously delivered tumor-promoting replication stress (33), and we hypothesize that unrepaired repli- in an accumulation of genomic cation stress effects result He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model PNAS \| 9 of 11 https://doi.org/10.1073/pnas.2108421118 alterations that trigger the rapid onset of tumorigenesis in 4NQO- treated Brca1+/−;Trp53+/− mice. Replication stress in cases of esophageal cancer in BRCA1 mutation carriers can be generated by exogenous agents like tobacco and alcohol, both of which have been linked to esophageal cancer (72, 73). Although the role of endogenous replication stress in tumorigenesis is well established (74, 75), the nature of the endogenous replication stress-inducing agents that lead to BRCA1 mutant cancer is unclear, as is whether such agents contribute to the tissue specificity observed in human BRCA1 cancer (i.e., to the increased risk for breast and ovarian cancer). There is some indication that estrogen metabolites could be one such endogenous agent that causes increased replication stress in BRCA1 heterozygous cells (76). Clear identification of endogenous agents that contribute to increased replication stress in BRCA1 mutant cancer will be important in understanding the early steps that define BRCA1 mutant tumorigenesis. Finally, although PM and BM mice also developed tumors, they did so at a much lower frequency than BPM mice. More- over, only the BPM tumors revealed 1) an accumulation of 4NQO-induced C > A alterations and 2) evidence of increased Brca1 LOH. Both of these observations strongly suggest that the accelerated tumorigenesis observed in BPM mice is a direct consequence of 4NQO-induced replication stress coupled with Brca1 heterozygosity that, in turn, led to Brca1 LOH, an estab- lished Brca1 mutant cancer driver. Importantly, neither 4NQO- induced DNA damage nor Brca1 heterozygosity, on its own, was enough to drive tumorigenesis in this experimental setting. That said, one might also consider the possibility that intraductal mammary 4NQO delivery could lead to Brca1 heterozygosity– driven mammary tumors. Such an accelerated tumor model system could prove to be invaluable in the detection of the earliest events in BRCA1 heterozygous mutation-driven breast cancer. Materials and Methods Cell Line and Culturing Condition. The human esophageal squamous cell carcinoma KYSE410 cell line was provided by Adam J. Bass’ group at Co- lumbia University, New York, NY. Cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (GIBCO) and 1% penicillin/ streptomycin (Thermo Fisher). Mouse Models. Brca1flox/wt (loxP sites flanking exons 5 to 13 of Brca1), and Trp53flox/wt (loxP sites flanking exons 1 to 10 of Trp53) mice were kindly provided by Jos Jonker’s group, Netherlands Cancer Institute, Amsterdam, The Netherlands. These mice were crossed with Meox2Cre deletor mice purchased from Jackson Labs (Stock No. 003755) to generate the four dif- ferent genotypes (BPM, BM, PM, and WT). All experimental protocols were approved by the Dana-Farber Institutional Animal Care and Use Committee. Experimental Conditions for 4NQO-Induced Tumorigenesis. 4NQO (Sigma-Aldrich, Cat. No. N8141) was dissolved in propylene glycol (Sigma-Aldrich, Cat. No. 398039- 500ML) at 6 mg/mL and diluted in drinking water to a final concentration of 200 μg/mL. Mice were given access to the drinking water at all times during the treatment. Six- to eight-wk-old mice from different genotypes were cohoused such that same sex mice from different genotypes were in the same cage. The water was refilled with fresh 4NQO solution biweekly. Control groups received the same volume of propylene glycol (vehicle control) in their drinking water. After 75 d, 4NQO-containing water was replaced with regular drinking water, and all mice were monitored for 45 additional days and then killed for tissue/tumor collection. Tissue Collection and Sectioning. Whole stomach and distal esophagus from experimental mice, both tumor and normal, were harvested and fixed in 10% formalin. Macroscopic images of stomach and esophagus were recorded and photographed for each experimental mouse. The harvested tissue was em- bedded in paraffin wax, cut into 5-μm sections, and stained with H&E. Pathology processing and analyses were carried out at the Rodent Histo- pathology Core at Harvard Medical School. Statistical Analyses. The relationships between phenotype and outcome were assessed using Fisher’s exact test. More details are provided in SI Appendix. Genomic DNA Extraction for LOH Analysis and WES. Formalin-fixed paraffin- embedded tissue sections were deparaffinized with xylene, washed with ethanol, and rehydrated in deionized water. The tumor areas to be micro- dissected were identified on the unstained section by comparing the tissue with an adjacent section stained with H&E. Microdissection was performed under the microscope, using a sterile syringe needle and/or a scalpel blade. The isolated tissue was placed in a 0.5-mL PCR tube containing DNA ex- traction buffer. Genomic DNA from the microdissection sections was extracted using the QIAmp DNA FFPE tissue kit (Qiagen, 56404) according to the manufacturer’s instructions. Tail cut DNA was repurified with a DNeasy kit (Qiagen) and concentrated with a Microcon DNA Fast Flow Centrifugal Filter Unit with Ultracel membrane (Millipore, Cat. No. MRCF0R100). Immunofluorescence and Immunohistochemistry. Paraffin-embedded sections were deparaffinized in xylene and rehydrated by serial immersion in ethanol and ddH2O. Antigen retrieval was achieved by heating sections in a pressure cooker (Cuisinart CPC-600) for 12 min in pH 6 citrate buffer (Millipore-Sigma, Cat. No. C9999). Tissue sections were washed with phosphate buffered saline (PBS) twice for 10 min, and then blocked for 1 h in blocking buffer (PBS containing 1% bovine serum albumin [BSA], 0.3% Triton X-100, and 10% normal goat serum). Sections were incubated with primary antibody in antibody solution buffer (PBS containing 1% BSA, 0.3% Triton X-100, and 1% normal goat serum) overnight at 4 °C. For immunohistochemistry (IHC), a Novolink Polymer detection system (Leica) was used and counterstained with Gill’s Hematoxylin No. 3. For immu- nofluorescence (IF), tissue sections were subsequently incubated with Alexa Fluor 647 secondary antibody for 30 min at room temperature. Sections were mounted with VECTASHIELD antifade mounting medium containing DAPI (Vectorlabs H-1000). Primary antibodies used were: phospho-CHK1 Ser345 (Cell Signaling Technology, Cat. No. 2341T), 53BP1 (Bethyl Laboratories, Cat. No. A300-272A), and P63 (Santa Cruz, Cat. No. sc-25268). Microscopy and Image Analysis. Images were acquired with an Axio Imager.M2 (Carl Zeiss) equipped with an Axiocam 506 color camera, controlled by Zen software. Images were exported at 16-bit images for subsequent analysis. Image analysis was performed with FIJI and Matlab_9.7 R2019b. Pixel intensity analysis was determined by assessing the average pixel abundance per mi- crometer squared and averaged across images. PCR Primers for Genotyping and LOH Analysis. Details of PCR primers and the related methods are described in detail in SI Appendix. Whole Exome Sequencing Data Analysis. Alignment and postprocessing of the whole exomes, copy number analysis, somatic point mutation calling, somatic signature calling, indel analysis, coverage analysis, and variant allele fre- quencies in the direct vicinity of BRCA1 are described in detail in SI Appendix. Data Availability. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. We express our sincerest thanks to Dr. Jos Jonkers of the Netherlands Cancer Institute for generously providing us with mice carrying floxed alleles of Brca1 and Trp53. This work was funded, in part, by grants from the NIH; The Gray Foundation; the Breast Cancer Research Foun- dation; the Susan G. Komen Foundation; the BRCA Foundation; Mr. Michael Robbins; Jeremy Maltby; Gia Lee; Bruce Rabb; and the Murray Winsten Foun- dation to D.M.L. It was also funded by University of Massachusetts Startup Funds and the Dana-Farber/Harvard Cancer Center (DF/HCC) Breast SPORE CEP to S.P.; the Breast Cancer Research Foundation (Grant ID: BCRF-20-159); Kræftens Bekæmpelse (Grant ID: R281-A16566); CDMRP Prostate Cancer Re- search Program (Grant W81XWH-18-2-00560); Det Frie Forskningsråd Sundhed og Sygdom (Grant ID: 7016-00345B); the Basser Foundation grant to Z.S.; and, in part, by the Dana-Farber/Harvard Cancer Center Kidney Can- cer SPORE Grant P50-CA101942-12 (to S.S.). 1. A. Antoniou et al., Average risks of breast and ovarian cancer associated with BRCA1 or BRCA2 mutations detected in case series unselected for family history: A combined analysis of 22 studies. Am. J. Hum. Genet. 72, 1117–1130 (2003). 2. K. B. Kuchenbaecker et al.; BRCA1 and BRCA2 Cohort Consortium, Risks of breast, ovarian, and contralateral breast cancer for BRCA1 and BRCA2 mutation carriers. JAMA 317, 2402–2416 (2017). 3. J. Starr, B. Ramnaraign, Germline BRCA1 mutated esophageal squamous cell carci- noma. Rare Tumors 12, 2036361320972218 (2020). 4. B. Ramnaraign, E. Altshuler, Complete clinical response of a patient with BRCA1- mutant cervical esophageal squamous cell carcinoma treated with oxaliplatin-based chemotherapy highlights the importance of performing genomic profiling in cancer treatment Curr. Prob. Cancer Case Rep. 3, 100069 (2021). 10 of 11 \| PNAS https://doi.org/10.1073/pnas.2108421118 He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model S C I T E N E G 5. J. Mersch et al., Cancers associated with BRCA1 and BRCA2 mutations other than 40. A. Maréchal, L. Zou, RPA-coated single-stranded DNA as a platform for post- breast and ovarian. Cancer 121, 269–275 (2015). translational modifications in the DNA damage response. Cell Res. 25, 9–23 (2015). 6. A. Moran et al., Risk of cancer other than breast or ovarian in individuals with BRCA1 41. S. Pathania et al., BRCA1 is required for postreplication repair after UV-induced DNA and BRCA2 mutations. Fam. Cancer 11, 235–242 (2012). damage. Mol. Cell 44, 235–251 (2011). 7. E. M. Petty, L. M. Kalikin, M. B. Orringer, D. G. Beer, Distal chromosome 17q loss in Barrett’s esophageal and gastric cardia adenocarcinomas: Implications for tumori- genesis. Mol. Carcinog. 22, 222–228 (1998). 8. J. Dunn et al., Multiple target sites of allelic imbalance on chromosome 17 in Barrett’s oesophageal cancer. Oncogene 18, 987–993 (1999). 9. T. Mori et al., Frequent loss of heterozygosity in the region including BRCA1 on chromosome 17q in squamous cell carcinomas of the esophagus. Cancer Res. 54, 1638–1640 (1994). 10. C.-X. Deng, R.-H. Wang, Roles of BRCA1 in DNA damage repair: A link between de- velopment and cancer. Hum. Mol. Genet. 12, R113–R123 (2003). 11. R. Roy, J. Chun, S. N. Powell, BRCA1 and BRCA2: Different roles in a common pathway of genome protection. Nat. Rev. Cancer 12, 68–78 (2011). 12. D. Branzei, M. Foiani, The DNA damage response during DNA replication. Curr. Opin. Cell Biol. 17, 568–575 (2005). 13. D. Branzei, M. Foiani, The checkpoint response to replication stress. DNA Repair (Amst.) 8, 1038–1046 (2009). 14. R. M. Jones, E. Petermann, Replication fork dynamics and the DNA damage response. Biochem. J. 443, 13–26 (2012). 15. A. J. Osborn, S. J. Elledge, L. Zou, Checking on the fork: The DNA-replication stress- response pathway. Trends Cell Biol. 12, 509–516 (2002). 16. V. G. Gorgoulis et al., Activation of the DNA damage checkpoint and genomic in- stability in human precancerous lesions. Nature 434, 907–913 (2005). 17. T. D. Halazonetis, V. G. Gorgoulis, J. Bartek, An oncogene-induced DNA damage model for cancer development. Science 319, 1352–1355 (2008). 18. S. Negrini, V. G. Gorgoulis, T. D. Halazonetis, Genomic instability—An evolving hall- mark of cancer. Nat. Rev. Mol. Cell Biol. 11, 220–228 (2010). 19. D. S. P. Tan, S. B. Kaye, Chemotherapy for patients with BRCA1 and BRCA2-mutated ovarian cancer: Same or different? American Society of Clinical Oncology educational book. American Society of Clinical Oncology. Annual Meeting 10.14694/EdBoo- k_AM.2015.35.114, 114–121 (2015). 20. T. A. Yap, R. Plummer, N. S. Azad, T. Helleday, The DNA damaging revolution: PARP inhibitors and beyond. American Society of Clinical Oncology educational book. American Society of Clinical Oncology. Annual Meeting 39, 185–195 (2019). 21. H. E. Bryant et al., Specific killing of BRCA2-deficient tumours with inhibitors of poly(ADP-ribose) polymerase. Nature 434, 913–917 (2005). 22. H. Farmer et al., Targeting the DNA repair defect in BRCA mutant cells as a thera- peutic strategy. Nature 434, 917–921 (2005). 23. O. Kondrashova et al.; Australian Ovarian Cancer Study (AOCS), Methylation of all BRCA1 copies predicts response to the PARP inhibitor rucaparib in ovarian carcinoma. Nat. Commun. 9, 3970 (2018). 24. L. C. Gowen, B. L. Johnson, A. M. Latour, K. K. Sulik, B. H. Koller, Brca1 deficiency results in early embryonic lethality characterized by neuroepithelial abnormalities. Nat. Genet. 12, 191–194 (1996). 25. J. Dine, C.-X. Deng, Mouse models of BRCA1 and their application to breast cancer research. Cancer Metastasis Rev. 32, 25–37 (2013). 26. S. S. Kim et al., Hyperplasia and spontaneous tumor development in the gynecologic system in mice lacking the BRCA1-Delta11 isoform. Mol. Cell. Biol. 26, 6983–6992 (2006). 27. X. Liu et al., Somatic loss of BRCA1 and p53 in mice induces mammary tumors with features of human BRCA1-mutated basal-like breast cancer. Proc. Natl. Acad. Sci. U.S.A. 104, 12111–12116 (2007). 28. T. Ludwig, P. Fisher, S. Ganesan, A. Efstratiadis, Tumorigenesis in mice carrying a truncating Brca1 mutation. Genes Dev. 15, 1188–1193 (2001). 29. R. Bachelier et al., Normal lymphocyte development and thymic lymphoma formation in Brca1 exon-11-deficient mice. Oncogene 22, 528–537 (2003). 30. R. Perets et al., Transformation of the fallopian tube secretory epithelium leads to high-grade serous ovarian cancer in Brca;Tp53;Pten models. Cancer Cell 24, 751–765 (2013). 31. L. Cao et al., Absence of full-length Brca1 sensitizes mice to oxidative stress and carcinogen-induced tumorigenesis in the esophagus and forestomach. Carcinogenesis 28, 1401–1407 (2007). 32. E. M. Michalak, J. Jonkers, Studying therapy response and resistance in mouse models for BRCA1-deficient breast cancer. J. Mammary Gland Biol. Neoplasia 16, 41–50 (2011). 33. S. Pathania et al., BRCA1 haploinsufficiency for replication stress suppression in pri- mary cells. Nat. Commun. 5, 5496 (2014). 34. M. Sedic et al., Haploinsufficiency for BRCA1 leads to cell-type-specific genomic in- stability and premature senescence. Nat. Commun. 6, 7505 (2015). 35. J. Bartkova et al., DNA damage response as a candidate anti-cancer barrier in early human tumorigenesis. Nature 434, 864–870 (2005). 36. M. D. Tallquist, P. Soriano, Epiblast-restricted Cre expression in MORE mice: A tool to distinguish embryonic vs. extra-embryonic gene function. Genesis 26, 113–115 (2000). 37. H. Duan et al., E3 ligase RFWD3 is a novel modulator of stalled fork stability in BRCA2- deficient cells. J. Cell Biol. 219, e201908192 (2020). 38. D. J. Downes et al., Characterization of the mutagenic spectrum of 4-nitroquinoline 1-oxide (4-NQO) in Aspergillus nidulans by whole genome sequencing. G3 (Bethesda) 4, 2483–2492 (2014). 39. M. Ikenaga, H. Ichikawa-Ryo, S. Kondo, The major cause of inactivation and mutation by 4-nitroquinoline 1-oxide in Escherichia coli: Excisable 4NQO-purine adducts. J. Mol. Biol. 92, 341–356 (1975). 42. X. H. Tang, B. Knudsen, D. Bemis, S. Tickoo, L. J. Gudas, Oral cavity and esophageal carcinogenesis modeled in carcinogen-treated mice. Clin. Cancer Res. 10, 301–313 (2004). 43. S. K. Sotiriou et al., Mammalian RAD52 functions in break-induced replication repair of collapsed DNA replication forks. Mol. Cell 64, 1127–1134 (2016). 44. J. Her, C. Ray, J. Altshuler, H. Zheng, S. F. Bunting, 53BP1 mediates ATR-Chk1 signaling and protects replication forks under conditions of replication stress. Mol. Cell. Biol. 38, e00472-17 (2018). 45. H. Gaillard, T. García-Muse, A. Aguilera, Replication stress and cancer. Nat. Rev. Cancer 15, 276–289 (2015). 46. M. Macheret, T. D. Halazonetis, DNA replication stress as a hallmark of cancer. Annu. Rev. Pathol. 10, 425–448 (2015). 47. R. A. L. Schoop, M. H. M. Noteborn, R. J. Baatenburg de Jong, A mouse model for oral squamous cell carcinoma. J. Mol. Histol. 40, 177–181 (2009). 48. X. L. Gao et al., Cytokeratin-14 contributes to collective invasion of salivary adenoid cystic carcinoma. PLoS One 12, e0171341 (2017). 49. H. C. Reinhardt, M. B. Yaffe, Kinases that control the cell cycle in response to DNA damage: Chk1, Chk2, and MK2. Curr. Opin. Cell Biol. 21, 245–255 (2009). 50. M. A. DiMaio, S. Kwok, K. D. Montgomery, A. W. Lowe, R. K. Pai, Immunohisto- chemical panel for distinguishing esophageal adenocarcinoma from squamous cell carcinoma: A combination of p63, cytokeratin 5/6, MUC5AC, and anterior gradient homolog 2 allows optimal subtyping. Hum. Pathol. 43, 1799–1807 (2012). 51. R. Barroso-Sousa et al., Prevalence and mutational determinants of high tumor mu- tation burden in breast cancer. Ann. Oncol. 31, 387–394 (2020). 52. H. Davies et al., HRDetect is a predictor of BRCA1 and BRCA2 deficiency based on mutational signatures. Nat. Med. 23, 517–525 (2017). 53. Y. Huo et al., Genetic interactions among Brca1, Brca2, Palb2, and Trp53 in mammary tumor development. NPJ Breast Cancer 7, 45 (2021). 54. L. Hámori et al., Establishment and characterization of a Brca1-/-, p53-/- mouse mammary tumor cell line. Int. J. Mol. Sci. 21, 1185 (2020). 55. Á. Póti et al., Correlation of homologous recombination deficiency induced muta- tional signatures with sensitivity to PARP inhibitors and cytotoxic agents. Genome Biol. 20, 240 (2019). 56. Y. Arima et al., 4-Nitroquinoline 1-oxide forms 8-hydroxydeoxyguanosine in human fibroblasts through reactive oxygen species. Toxicol. Sci. 91, 382–392 (2006). 57. T. Nunoshiba, B. Demple, Potent intracellular oxidative stress exerted by the carcin- ogen 4-nitroquinoline-N-oxide. Cancer Res. 53, 3250–3252 (1993). 58. A. Viel et al., A specific mutational signature associated with DNA 8-oxoguanine persistence in MUTYH-defective colorectal cancer. EBioMedicine 20, 39–49 (2017). 59. L. B. Alexandrov et al.; PCAWG Mutational Signatures Working Group; PCAWG Consortium, The repertoire of mutational signatures in human cancer. Nature 578, 94–101 (2020). 60. J. M. Schildkraut et al., Loss of heterozygosity on chromosome 17q11-21 in cancers of women who have both breast and ovarian cancer. Am. J. Obstet. Gynecol. 172, 908–913 (1995). 61. S. D. Merajver et al., Germline BRCA1 mutations and loss of the wild-type allele in tumors from families with early onset breast and ovarian cancer. Clin. Cancer Res. 1, 539–544 (1995). 62. R. S. Cornelis et al.; The Breast Cancer Linkage Consortium, High allele loss rates at 17q12-q21 in breast and ovarian tumors from BRCAl-linked families. Genes Chro- mosomes Cancer 13, 203–210 (1995). 63. S. D. Merajver et al., Somatic mutations in the BRCA1 gene in sporadic ovarian tu- mours. Nat. Genet. 9, 439–443 (1995). 64. S. L. Neuhausen, C. J. Marshall, Loss of heterozygosity in familial tumors from three BRCA1-linked kindreds. Cancer Res. 54, 6069–6072 (1994). 65. K. N. Maxwell et al., BRCA locus-specific loss of heterozygosity in germline BRCA1 and BRCA2 carriers. Nat. Commun. 8, 319 (2017). 66. T. A. King et al., Heterogenic loss of the wild-type BRCA allele in human breast tu- morigenesis. Ann. Surg. Oncol. 14, 2510–2518 (2007). 67. C. L. Clarke et al., Mapping loss of heterozygosity in normal human breast cells from BRCA1/2 carriers. Br. J. Cancer 95, 515–519 (2006). 68. F. C. Martins et al., Evolutionary pathways in BRCA1-associated breast tumors. Cancer Discov. 2, 503–511 (2012). 69. A. McKenna et al., The Genome Analysis Toolkit: A MapReduce framework for ana- lyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 (2010). 70. G. J. Mann et al.; Kathleen Cuningham Consortium for Research in Familial Breast Cancer, Analysis of cancer risk and BRCA1 and BRCA2 mutation prevalence in the kConFab familial breast cancer resource. Breast Cancer Res. 8, R12 (2006). 71. K. Ishida et al., Current mouse models of oral squamous cell carcinoma: Genetic and chemically induced models. Oral Oncol. 73, 16–20 (2017). 72. N. D. Freedman et al., A prospective study of tobacco, alcohol, and the risk of esophageal and gastric cancer subtypes. Am. J. Epidemiol. 165, 1424–1433 (2007). 73. D. J. Uhlenhopp, E. O. Then, T. Sunkara, V. Gaduputi, Epidemiology of esophageal cancer: Update in global trends, etiology and risk factors. Clin. J. Gastroenterol. 13, 1010–1021 (2020). 74. A. Tubbs, A. Nussenzweig, Endogenous DNA damage as a source of genomic insta- bility in cancer. Cell 168, 644–656 (2017). 75. M. K. Zeman, K. A. Cimprich, Causes and consequences of replication stress. Nat. Cell Biol. 16, 2–9 (2014). 76. K. I. Savage et al., BRCA1 deficiency exacerbates estrogen-induced DNA damage and genomic instability. Cancer Res. 74, 2773–2784 (2014). He et al. BRCA1/Trp53 heterozygosity and replication stress drive esophageal cancer development in a mouse model PNAS \| 11 of 11 https://doi.org/10.1073/pnas.2108421118
10.1073_pnas.2300099120	RESEARCH ARTICLE \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS B cell peripheral tolerance is promoted by cathepsin B protease Marissa Y. Choua,b, Dan Liua,b, Jinping Ana,b, Ying Xua,b, and Jason G. Cystera,b,1 Contributed by Jason Cyster; received January 4, 2023; accepted March 14, 2023; reviewed by Robert Brink and Richard J. Cornall B cells that bind soluble autoantigens receive chronic signaling via the B cell receptor (signal-1) in the absence of strong costimulatory signals (signal-2), and this leads to their elimination in peripheral tissues. The factors determining the extent of soluble autoantigen-binding B cell elimination are not fully understood. Here we demon- strate that the elimination of B cells chronically exposed to signal-1 is promoted by cathepsin B (Ctsb). Using hen egg lysozyme-specific (HEL-specific) immunoglobulin transgenic (MD4) B cells and mice harboring circulating HEL, we found improved survival and increased proliferation of HEL-binding B cells in Ctsb-deficient mice. Bone marrow chimera experiments established that both hematopoietic and non- hematopoietic sources of Ctsb were sufficient to promote peripheral B cell deletion. The depletion of CD4+ T cells overcame the survival and growth advantage provided by Ctsb deficiency, as did blocking CD40L or removing CD40 from the chronically antigen-engaged B cells. Thus, we suggest that Ctsb acts extracellularly to reduce solu- ble autoantigen-binding B cell survival and that its actions restrain CD40L-dependent pro-survival effects. These findings identify a role for cell-extrinsic protease activity in establishing a peripheral self-tolerance checkpoint. B lymphocyte \| signal-1 \| self-tolerance \| cathepsin \| CD40 Self-t.olerance in the B cell compartment is established through multiple checkpoints (1). Antigens that strongly cross-link the B cell receptor (BCR) on developing B cells in the bone marrow (BM) cause receptor editing or deletion. B cells that recognize low-valency soluble autoantigens are not held up at the immature B cell checkpoint but are instead regulated by peripheral checkpoints. One model that has been widely used to study tol- erance induction in response to soluble autoantigen involves immunoglobulin (Ig) trans- genic mice (called MD4) that are specific for hen egg lysozyme (HEL) and transgenic mice that express soluble HEL as a neoself-antigen (called ML5) (2). In double-transgenic mice generated by intercrossing these two lines, the B cells are all autoantigen-engaged and thus chronically receiving B cell receptor (BCR) signals (signal-1). These HEL-binding B cells are not receiving cognate T cell help or being exposed to pathogen-associated molecular patterns (PAMPS) and thus are not receiving signal-2. HEL-engaged B cells have reduced surface IgM levels and a reduced ability to signal in vitro in response to exogenous HEL. When these chronically HEL-engaged B cells are placed in the polyclonal repertoire of wild-type (WT) mice, they undergo follicular exclusion and are deleted from the periphery within a few days (3, 4). This competitive elimination of B cells receiving chronic signal-1 without signal-2 occurs in part due to an increased dependence of the cells on the pro-survival factor BAFF and the limited availability of this factor in mice with a polyclonal B cell repertoire (5, 6). As well as being strongly dependent on BAFF, survival of B cells receiving chronic signal-1 in the absence of signal-2 is augmented by naïve CD4+ T cells (7). Although naïve CD4+ T cells are not conventionally considered to be a source of CD40L (and thus signal-2), our previous work showed that naïve CD4+ T cells constitutively express CD40L, though it is not measurable on the cell surface due to continual modulation by engagement with CD40-expressing cells; when these cells are removed, surface CD40L can be detected (8). Whether additional extrinsic factors beyond BAFF and naïve CD4+ T cell CD40L influence the survival of peripheral B cells chronically exposed to signal-1 is unclear. Cathepsin B (Ctsb) is a widely expressed member of the cysteine cathepsin family (9). It is expressed as a preproenzyme in the endoplasmic reticulum, and it becomes a mature protease in the lysosome. It carries out a variety of functions in the lysosome, such as processing other lysosomal enzymes and mediating the degradation of hormones (9, 10). Ctsb is also found in multiple extracellular locations. Extracellular functions attributed to Ctsb include proteolysis of the extracellular matrix (11), protection of cytotoxic T cells from self-killing (12), generation of soluble TRAIL (TNFSF10) (13), activation of TGFb (14), and cleavage of several chemokines (15). In some of these studies, the extracellular function was thought to occur while Ctsb was associated with the surface of the secreting Significance Self-tolerance in the B cell compartment of soluble autoantigens depends on mechanisms that promote peripheral B cell elimination. In this study, we report that deficiency in the cysteine protease cathepsin B (Ctsb) in mice results in less efficient removal of chronically antigen- engaged B cells from the repertoire. The tolerogenic action of Ctsb is lost in mice lacking the CD40 pathway, and the findings suggest that Ctsb restrains, directly or indirectly, basal CD40L-CD40 pathway activity and thereby helps prevent the activation of autoreactive B cells. Since protease activity is increased at sites of inflammation, our findings point to the possibility that self- tolerance checkpoints are reset in these sites due to proteolytic modulation of cell surface interaction molecules. Author contributions: M.Y.C. and J.G.C. designed research; M.Y.C., D.L., J.A., and Y.X. performed research; M.Y.C., D.L., and J.G.C. analyzed data; M.Y.C. and J.G.C. wrote the paper. Reviewers: R.B., Garvan Institute of Medical Research; and R.J.C., University of Oxford. Competing interest statement: J.G.C. is on the scientific advisory board of BeBio Pharma and MiroBio Ltd. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2300099120/-/DCSupplemental. Published April 11, 2023. PNAS 2023 Vol. 120 No. 16 e2300099120 https://doi.org/10.1073/pnas.2300099120 1 of 9 cell (10). Of particular relevance to B cell biology, a recent study suggested that extracellular Ctsb was required for cleavage of CXCL13 to help establish a gradient of this chemokine for the organization of B cell follicles in lymphoid tissues (16). In this project, we set out to further characterize the influence of Ctsb on B cell follicle organization and function. In the Ctsb-deficient mice studied here, we did not observe a defect in follicular organization. However, when MD4 B cells were trans- ferred into Ctsb-deficient mice containing soluble HEL such that they experienced chronic signal-1, the extent of B cell deletion was reduced compared to transfers into matched WT mice. Our exper- iments establish that hematopoietic and nonhematopoietic cells both contribute to the production of Ctsb that promotes HEL-binding B cell elimination. The effect of Ctsb on HEL-binding B cells was overcome in mice depleted of CD4+ T cells or deficient in CD40L function or in CD40. These findings suggest a pathway by which extracellular protease activity promotes B cell self-tolerance. Results Ctsb-Deficient Mice Have Normal Follicles but Reduced Elimination of HEL-Binding B Cells. Follicular organization in Ctsb-deficient mice was examined by staining spleen and lymph node sections for B cell and follicular dendritic cell (FDC) markers. Since the prior study had reported a requirement for Ctsb to maintain primary follicles and FDCs (16), we examined unimmunized mice. In the Ctsb−/− mouse colony studied here, lymphoid follicle organization in spleen and lymph nodes was intact and FDCs were readily detected (SI Appendix, Fig. S1 A and B). Analysis of Ctsb activity in spleen interstitial fluid using an enzyme activity assay confirmed the Ctsb deficiency (Fig. 1A). These data suggest Ctsb is not essential for establishing the CXCL13 activity needed for follicular organization. We next considered whether Ctsb might influence the fate of different types of follicular B cells. Ctsb-deficient mice had normal frequencies of splenic B cells and CD4+ T cells and slightly reduced CD8+ T cells (SI Appendix, Fig. S1C) as previously reported (16). The mice mounted intact splenic germinal center (GC) responses to sheep red blood cells (SRBCs), a model T-dependent antigen (SI Appendix, Fig. S1D). The chronic GC responses present in mes- enteric lymph nodes and Peyer’s patches were also comparable between Ctsb-deficient and control mice (SI Appendix, Fig. S1E). Since the naïve and GC B cell compartments appeared intact, we next asked whether the fate of B cells experiencing chronic signal-1 was affected. This was tested in Ctsb-deficient mice by taking advan- tage of the finding that when mice harboring low frequencies of MD4 B cells are injected with large amounts of soluble HEL, the HEL-binding B cells experience chronic BCR engagement and are deleted within 3 d (7). This outcome is analogous to the loss that occurs when MD4 B cells are transferred into ML5 mice that express HEL endogenously (4). Using this MD4 cell transfer and HEL injection approach, we observed the expected marked elimi- nation of splenic HEL-binding B cells in control recipients (Fig. 1 B and C). However, the extent of deletion was significantly reduced in Ctsb-deficient mice (Fig. 1C). Similar findings were made in peripheral lymph nodes (SI Appendix, Fig. S2A). Analysis of the surface phenotype of the transferred cells showed higher CD23 and ICAM1 levels on HEL-binding B cells in the Ctsb-deficient recip- ients (Fig. 1 D and E). The reduced expression of CD23 on HEL-engaged B cells is consistent with the ability of BCR signaling to downmodulate this marker (17). Surface IgM was also present at higher levels on a fraction of the HEL-binding B cells in Ctsb-deficient recipients (Fig. 1F). Since IgM downmodulation is highly sensitive to the extent of HEL exposure, we examined serum HEL abundance in the treated control and Ctsb-deficient mice. This was done by incubating MD4 B cells with serum taken from control and Ctsb-deficient mice at day 3 after HEL injection and then staining with HyHEL9 to detect the amount of bound HEL. Control and Ctsb-deficient mouse serum contained comparable amounts of HEL (Fig. 1G), thereby excluding differences in HEL availability as an explanation for the improved survival of HEL-binding B cells in mice lacking Ctsb. Using cell trace violet (CTV) labeling to track cell division in the transferred cells, few cells had divided in control recipients as expected, but there was an increase in cell division in Ctsb-deficient recipients (Fig. 1H). The assessment of the frequencies of divided and undivided B cells showed that both were increased (Fig. 1H and SI Appendix, Fig. S2B), indicating that Ctsb-deficiency augmented both prolif- eration and survival of B cells experiencing chronic BCR signaling. Analysis of the surface marker phenotype in divided and undivided transferred B cells showed a greater difference among the divided B cells, suggesting the mechanism promoting proliferation also contributed to the increased CD23, ICAM1, and IgM (SI Appendix, Fig. S2 C–E). To confirm that the HEL injection approach was an accurate model for events occurring during autoantigen exposure, we crossed Ctsb−/− mice with ML5 mice. Transferred MD4 cells were mostly deleted within 3 days of transfer into control ML5 mice, but the deletion was reduced in Ctsb−/− ML5 mice (Fig. 1I). Serum HEL concentrations in ML5 mice were unchanged by Ctsb deficiency (SI Appendix, Fig. S2F). The HEL-binding B cells in Ctsb-deficient ML5 mice showed elevated CD23, ICAM1, and IgM (SI Appendix, Fig. S2 G–I), and the fraction of cells that divided was also increased (SI Appendix, Fig. S2J). Thus, Ctsb contributes to restraining HEL-autoantigen binding B cell acti- vation and to promoting their removal from the peripheral B cell repertoire. To test for possible effects of Ctsb on BAFF abundance, we took advantage of the finding that the BAFF availability is reflected in the amount that is bound to receptors on the B cell surface (5). Using a polyclonal anti-BAFF serum, the amount of surface BAFF on B cells from control and Cstb-deficient mice was equivalent (SI Appendix, Fig. S3A). As the size of the B cell compartment is highly responsive to BAFF availability (18), the unchanged size of the B cell compartment in Ctsb-deficient mice (SI Appendix, Fig. S1C) provided further evidence that BAFF abundance was not altered by Ctsb deficiency. Ctsb from Hematopoietic and Nonhematopoietic Cells Promotes Peripheral B Cell Tolerance. Our transfer studies utilized WT MD4 B cells and thus established that the action of Ctsb in promoting elimination of B cells receiving chronic signal-1 was cell- extrinsic. To determine whether Ctsb was needed in hematopoietic or nonhematopoietic cells, we generated WT -> Ctsb KO, Ctsb KO -> WT, and Ctsb KO -> Ctsb KO BM chimeras. After 7 to 8 wk of reconstitution, MD4 B cells were transferred, and soluble HEL injected. The deletion efficiency was intact in all the chimeras except the Ctsb KO -> Ctsb KO group (Fig. 1J). These data indicate that Ctsb from either hematopoietic cells or radiation-resistant (most likely nonhematopoietic stromal) cells was sufficient for promoting elimination of HEL-binding B cells. Follicular Exclusion Is Intact in the Absence of Ctsb. To test whether follicular exclusion of HEL-binding B cells was impaired in the absence of Ctsb, spleens were taken from control and Ctsb- deficient mice that harbored MD4 B cells and had been treated with soluble HEL 1 d earlier. Imaging analysis revealed the MD4 B cells were predominantly located at the follicle T zone interface in 2 of 9 https://doi.org/10.1073/pnas.2300099120 pnas.org 1250 A B ) U A ( e c n e c s e r o u l f 1000 750 500 250 0 assay buffer spleen interstitial D 12500 ns E 30000 10000 assay buffer ff spleen interstitial 1 M A C I I F M 20000 10000 ns saline HEL 0 x a m f o % CD23 ICAM1 ns ns I e v i l f o s l l e c B 4 D M x t 3 2 D C I F M 7500 5000 2500 0 x a m f o % G 9 L E H y H I F M 5000 4000 3000 2000 1000 0 ns 0.8 0.6 0.4 C e v i l f o s l l e c B 4 D M x t % 0.2 0.0 saline HEL ns H 100 4 D M x t f o d e d v d % i i 80 60 40 20 0 ns F a M g I I F M 10000 7500 5000 2000 1500 1000 500 0 saline HEL saline HEL saline HEL x a m f o % x a m f o % IgMa J 0.5 e v i l 0.4 f o s l l e c B 4 D M x t 0.3 0.2 CTV ns ns 1.0 0.5 0.15 0.10 % 0.05 0.00 % 0.1 0.0 saline HEL saline HEL ctrl ML5 1:10 1:50 Donor BM : Recipient : WT WT KO KO WT WT KO WT WT KO KO KO saline HEL Fig. 1. Ctsb Promotes Elimination of HEL-Binding B Cells. (A) Ctsb activity in assay buffer or spleen interstitial fluid of control (Ctsb+/+ or Ctsb+/−) or Ctsb−/− (CtsB KO) mice. Data show results from two separate experiments. (B) Schematic of CTV-labeled MD4 B cell adoptive transfer into control or Ctsb-deficient mice, followed by HEL treatment. (C) Frequencies of transferred MD4 B cells in spleens of control or Ctsb-deficient recipients 3 d after saline (n = 16 control, n = 9 KO mice) or HEL treatment (n = 26 control, n = 30 KO mice). (D–F) MFI (Top) and representative histogram plot (Bottom) of CD23 (D), ICAM1 (E), and IgMa (F) on transferred MD4 B cells 3 d after saline (n = 11 control, n = 5 KO) or HEL treatment (n = 22 control, n = 24 KO). (G) MFI of HyHEL9 on MD4 B cells incubated with 1:10 or 1:50 dilutions of sera from control or Ctsb-deficient mice 3 d after saline (n = 7 control, n = 5 KO) or HEL treatment (n = 16 control, n = 17 KO). (H) Percentage of divided transferred MD4 B cells (Top) or representative histogram plot of CTV (Bottom) 3 d after saline (n = 8 control, n = 4 KO) or HEL treatment (n = 16 control, n = 17 KO). (I) Frequencies of transferred MD4 B cells in spleens of control ML5+ (n = 8) or Ctsb-deficient ML5+ (n = 7) recipients 3 d after MD4 B cell adoptive transfer. Control ML5− (n = 3) mice used as deletion control. (J) Frequencies of transferred MD4 B cells in spleens of bone marrow (BM) chimeras 3 days after saline or HEL treatment. The Ctsb genotype of donor BM and recipient mice used to generate the BM chimeras is indicated. In graphs, each data point indicates an individual mouse and lines indicate means. Error bars represent SDs. I is representative of three experiments. Statistical significance for A–I was determined by unpaired t test. NS, not significant; P < 0.05; P < 0.01; P < 0.001; P < 0.0001. Statistical significance for J was determined by ordinary one-way ANOVA, , P = 0.0007. PNAS 2023 Vol. 120 No. 16 e2300099120 https://doi.org/10.1073/pnas.2300099120 3 of 9 both control and Ctsb-deficient recipients (Fig. 2 A–C). Thus, Ctsb is not required for follicular exclusion of antigen-engaged B cells. + T Cell Depletion Overcomes the Effect of Ctsb Deficiency CD4 on HEL-Binding B Cells. Since prior work had established that naïve CD4+ T cells restrain the extent of HEL-binding B cell elimination in WT mice (7), we tested whether the action of Ctsb was influenced by CD4+ T cells. Control and Ctsb-deficient mice were treated with a CD4+ T cell depleting antibody and then used as recipients of MD4 cells and soluble HEL (Fig. 3A). In contrast to the findings in non-treated recipients, there was a similar extent of deletion in control and Ctsb-deficient mice when CD4+ T cells were lacking (Fig. 3B). There was also a loss in the increased levels of CD23, ICAM1 and IgM (Fig. 3 C–E). The effect of Ctsb- deficiency on HEL-binding B cell proliferation was also lost in CD4+ T cell-depleted mice (Fig. 3F). These data suggest that the B cell deletion-promoting effect of Ctsb may involve restraining some function of CD4+ T cells. Previous work had found that naïve CD4+ T cells can limit the extent of HEL autoantigen- binding B cell elimination through provision of CD40L (8). Therefore, we next examined the effect of CD40L blockade. Enhanced HEL-Binding B Cell Persistence in Ctsb-Deficient Hosts Depends on CD40L and CD40. Control and Ctsb-deficient mice were treated with CD40L-blocking antibody, and the fate of MD4 B cells after HEL treatment was examined. Blocking CD40L had a similar effect to CD4+ T cell ablation, overcoming the effect of Ctsb deficiency on HEL-binding B cell accumulation (Fig. 4A). This included a loss in CD23 and ICAM1 induction (Fig. 4 B and C) and a complete loss of increased cell division (Fig. 4D). We then examined CD40L transcript levels in sorted naïve CD4+ T cells and found they were abundant as expected (8) and were unaltered by Ctsb deficiency (Fig. 4E). Moreover, surface CD40L on PMA plus ionomycin-activated control and Ctsb-deficient CD4+ T cells was equivalent (Fig. 4F). CD40 engagement of CD40L causes loss of CD40L from the T cell surface (19–21); however, when purified naïve CD4+ T cells are incubated in a low- density culture and thus in the absence of CD40 exposure, CD40L becomes detectable on the cell surface (8). When control and Ctsb-deficient CD4+ T cells were incubated in this way, the Ctsb- deficient cells showed a trend for more surface display of CD40L, but this difference was not statistically significant (Fig. 4G and SI Appendix, Fig. S3B). Thus, under the conditions tested, Ctsb did not significantly alter CD40L expression. Finally, as a further test of the contribution of CD40L to the augmented survival of HEL-binding B cells in Ctsb-deficient hosts, we cotransferred WT and CD40−/− MD4 B cells into control and Ctsb-deficient hosts and treated them with soluble HEL. The analysis 3 d later showed comparable deletion in both types of recipients (Fig. 4H). CD40- deficiency also prevented elevation of ICAM1 and IgM, though there was still some induction of CD23 (SI Appendix, Fig. S3 C and D). The expression of IgM was widely dispersed in this experiment (SI Appendix, Fig. S3E). CD40 abundance on WT MD4 B cells in control and Ctsb-deficient mice was similar (SI Appendix, Fig. S3F). Taken together, these data suggest that the effect of Ctsb on augmenting deletion of HEL-binding B cells involves a process that limits the CD40L-CD40 engagement that occurs when HEL-binding B cells encounter naïve CD4+ T cells in lymphoid tissues. Although we did not observe an alteration in the polyclonal GC response to complex foreign or gut-associated antigens (SI Appendix, Fig. S1 D and E), our findings suggesting that C e c l i l l o f n i 4 D M x t # / B - T t a 4 D M x t # IgD CD21/35 CD4 MD4 15 10 5 0 control CtsB KO ns HEL A l o r t n o c B O K B s t C Fig. 2. Follicular Exclusion of HEL-Engaged B Cells Is Unaffected by Ctsb Deficiency. (A and B) Immunofluorescence for MD4 GFP B cells (GFP, green) transferred into HEL-treated control (A) or Ctsb-deficient (B) mice stained to detect endogenous B cell follicles (IgD, blue; CD21/35, white) and the T cell zone (CD4, red). Sections were prepared one day after HEL treatment. Two example images are shown and are representative of multiple cross-sections from at least three mice of each type. (Scale bar, 200 µm.) (C) Quantification of proportion of MD4 GFP B cells at the T zone–follicle (T-B) interface one day after HEL treatment. Each data point represents an individual follicle (n = 24 control, n = 25 KO) from sections prepared from at least three mice of each genotype. Lines indicate means, and error bars represent SDs. Statistical significance was determined by unpaired t test. NS, not significant. IgD CD21/35 CD4 MD4 4 of 9 https://doi.org/10.1073/pnas.2300099120 pnas.org A C 8000 3 2 D C I F M 6000 4000 2000 D 2000 1 M A C I I F M 1500 1000 500 ns 0 saline HEL HEL + GK1.5 0 saline HEL HEL + GK1.5 x a m f o % x a m f o % B 1.0 e v i l 0.5 ns 0.2 0.1 0.0 saline HEL HEL + GK1.5 f o s l l e c B 4 D M x t % E 10000 ns ns saline HEL HEL + GK1.5 a M g I I F M 7500 2500 2000 1500 1000 500 0 x a m f o % F 4 D M x t f o d e d v d % i i ns saline HEL HEL + GK1.5 80 60 40 20 0 x a m f o % CD23 ICAM1 IgMa CTV Fig. 3. Depletion of CD4+ T Cells Overcomes the Effect of Ctsb Deficiency on HEL-Binding B Cells. (A) Schematic of CTV-labeled MD4 B cell adoptive transfer into control or Ctsb-deficient mice, followed by GK1.5 CD4+ T cell depletion and HEL treatment. (B) Frequencies of transferred MD4 B cells in spleens of control or Ctsb-deficient recipients 3 d after saline (n = 6 control), HEL (n = 7 control, n = 9 KO), or HEL with GK1.5 treatment (n = 6 control, n = 9 KO). (C–E) MFI (Top) and representative histogram plot (Bottom) of CD23 (C), ICAM1 (D), and IgMa (E) on transferred MD4 B cells 3 d after saline (n = 5 control), HEL (n = 6 control, n = 5 KO), or HEL with GK1.5 treatment (n = 6 control, n = 5 KO). (F) Percentage of divided transferred MD4 B cells (Top) or representative histogram plot of CTV (bottom) 3 d after saline (n = 3 control, n = 2 KO), HEL (n = 6 control, n = 6 KO), or HEL with GK1.5 treatment (n = 6 control, n = 5 KO). Each data point indicates an individual mouse and lines indicate means. Error bars represent SDs. C–F are representative of three experiments. Statistical significance for B–F was determined by unpaired t test. NS, not significant; P < 0.05; P < 0.01, P < 0.001. noncognate T-dependent signals could be increased in the absence of Ctsb led us to perform a further experiment exam- ining cognate T cell help. We transferred MD4 cells into con- trol or Ctsb-deficient recipients and immunized the mice with HEL2x, a form of HEL with reduced affinity for the MD4 BCR (22), coupled to strongly immunogenic SRBCs. Upon analysis at 5 d, although total numbers of MD4 B cells were similar between recipients, we observed an increase in the fraction of MD4 cells that had a GC phenotype in the immunized Ctsb-deficient mice compared to control recipients (Fig. 4 I and J). Moreover, CTV dilution analysis showed that the MD4 cells had undergone more extensive cell division in the Ctsb-deficient recipients (Fig. 4 K and L). These data indicate that under some conditions, Ctsb can restrain the extent of T cell help in response to foreign antigen. Discussion Here we find that the elimination of peripheral B cells experiencing chronic signal-1 is partially dependent on extrinsic Ctsb activity. Since more than one source of Ctsb (hematopoietic or nonhe- matopoietic) was sufficient, we speculate that Ctsb was acting in an extracellular manner to reduce the availability of a factor (or factors) that augments the survival and proliferation of chronically BCR-engaged B cells. Ctsb activity in promoting B cell elimina- tion was lost when CD4+ T cells, CD40L, or CD40 was lacking. We suggest Ctsb acts in part by reducing signaling via the CD40L– CD40 axis during the encounter between autoantigen-binding B cells and naïve CD4+ T cells. The finding of increased GC B cells in Ctsb-deficient mice immunized with a model foreign antigen is in accord with a possible increase in CD40L–CD40 signaling. PNAS 2023 Vol. 120 No. 16 e2300099120 https://doi.org/10.1073/pnas.2300099120 5 of 9 A 1.0 e v i l 0.8 f o s l l e c B 4 D M x t 0.6 0.4 % 0.2 0.0 saline B 10000 ns 3 2 D C I F M 5000 C ns 20000 15000 10000 5000 1 M A C I I F M ns D 4 D M x t f o d e d v d % i i HEL HEL + MR1 0 saline HEL HEL + MR1 0 saline HEL HEL + MR1 E F G l T R P H o t e v i t a e r n o s s e r p x e L 0 4 D C i ns ns 102 101.5 101 100.5 100 L 0 4 D C I F M 2500 2000 1500 1000 500 0 ns 1500 0.0868 1000 L 0 4 D C I F M 500 0 spleen LN no stim PMA/iono ice 2h culture Control CtsB KO I 1.0 e v i l 0.8 J ns f o s l l e c B 4 D M x t 0.6 0.4 % 0.2 0.0 4 D M x t f o C G % 40 30 20 10 0 K 4 D M x t f o d e d v d % i i 80 60 40 20 0 ns unimmunized SRBC-HEL2x unimmunized SRBC-HEL2x undivided 5+ divisions H e v i l f o s l l e c B 4 D M x t % L x a m f o % Control CtsB KO ns 80 60 40 20 0 saline HEL HEL + MR1 WT MD4 CD40KO MD4 ns 0.3 0.2 0.1 0.0 CtsB KO + HEL control + HEL saline CtsB KO + HEL control + HEL saline Control CtsB KO CTV Fig. 4. Enhanced HEL-Binding B Cell Persistence in Ctsb-deficient Hosts Depends on CD40L and CD40. (A) Frequencies of transferred MD4 B cells in spleens of control or Ctsb-deficient recipients 3 d after saline, HEL, or HEL with MR1 CD40L-blocking treatment. (B and C) MFI of CD23 (B) and ICAM1 (C) on transferred MD4 B cells 1.5 d after saline (n = 4 control, n = 4 KO), HEL (n = 8 control, n = 8 KO), or HEL with MR1 treatment (n = 9 control, n = 9 KO). (D) Percentage of divided transferred MD4 B cells 3 d after saline (n = 3 control, n = 4 KO), HEL (n = 6 control, n = 5 KO), or HEL with MR1 treatment (n = 6 control, n = 6 KO). (E) CD40L transcript abundance in naïve CD4+ T cells isolated from spleen and lymph node tissues harvested from control (n = 6) or Ctsb−/− (n = 3) mice. CD40L and hypoxanthine phosphoribosyltransferase (HPRT) transcripts were quantitated by real-time PCR. Data show results from two separate experiments. (F) MFI of CD40L on purified CD4+ T cells from control (n = 7) or Ctsb-deficient (n = 8) mice incubated without (no stim) or with (PMA/iono) PMA and ionomycin for 2 h at 37 °C. (G) MFI of CD40L on purified CD4+ T cells from control (n = 27) or Ctsb-deficient (n = 26) mice kept on ice or incubated in a dilute culture for 2 h at 37 °C. (H) Frequencies of WT or CD40-deficient MD4 B cells in transferred splenocytes in spleens of control or Ctsb-deficient recipients 3 d after saline (n = 2 control, n = 2 KO) or HEL treatment (n = 6 control, n = 6 KO). (I and J) Frequencies of transferred MD4 B cells amongst total cells (I) and of MD4 B cells having a germinal center (GC) phenotype (J) in spleens of unimmunized mice (n = 3) or mice immunized with SRBC-HEL2x (n = 4 control, n = 4 KO). (K) Frequencies of undivided MD4 B cells or MD4 B cells having undergone five or more divisions in spleens of control (n = 4) or Ctsb-deficient mice (n = 4) 5 d after SRBC-HEL2x immunization. (L) Representative histogram plot of CTV labeling of transferred MD4 B cells in indicated recipients 5 d after SRBC-HEL2x immunization. Each data point indicates an individual mouse and lines indicate means. Error bars represent SDs. A–D and H are representative of three experiments. I–L are representative of two experiments. Statistical significance for A–K was determined by unpaired t test. NS, not significant; P < 0.05; P < 0.01; P < 0.001; ***P < 0.0001. However, we cannot exclude the possibility that Ctsb is acting to restrain an independent pathway that cooperates with CD40L– CD40 or other T cell-derived signals. Overall, these findings iden- tify a role for extracellular protease activity in establishing the extent of B cell tolerance to peripheral self-antigen and potentially the magnitude of the response to some foreign antigens. A previous study showed that Ctsb could cleave CXCL13 in vitro and the truncated form was reported to be more potent in promoting 6 of 9 https://doi.org/10.1073/pnas.2300099120 pnas.org B cell migration than the full-length form (16). Images were pro- vided (without further quantification) suggesting that Ctsb-deficient mice have a loss of follicular structure and FDCs in lymph nodes. FDCs are maintained in lymphoid follicles via LTα1β2 signals pro- vided by B cells in a partially CXCL13-dependent manner (23), making the observations in Ctsb-deficient mice internally consistent. In contrast to those observations, we failed to observe a defect in B cell follicular clustering in Ctsb-deficient mice, and the follicles con- tained intact FDC networks. The basis for this discrepancy between studies is unclear but might reflect different breeding histories of the mouse lines or differences in the microbiome between facilities. Although Ctsb was shown to be able to cleave CXCL13 in vitro, it was not shown that the enzyme has this activity in vivo (16). Since our anatomical findings suggest CXCL13 function does not depend on Ctsb, we have not examined whether CXCL13 is cleaved in vivo by Ctsb. Importantly, the influence of Ctsb on autoantigen-engaged B cell survival is unlikely to be via effects on CXCL13 as earlier studies have shown that CXCR5-deficient B cells that cannot respond to CXCL13 undergo an equivalent extent of deletion as WT B cells in the HEL-MD4 model (24). Ctsb acts on a wide range of intracellular and extracellular sub- strates (9, 10). While our study does not identify the molecule(s) that Ctsb acts on to restrain autoreactive B cell survival, our findings in CD4+ T cell-deficient, CD40L-blocked, and CD40-deficient systems point to actions that influence the CD40L–CD40 pathway. This possibility is also supported by the finding of increased ICAM1 and CD23 expression in chronically antigen-engaged B cells in Ctsb-deficient hosts, as both these molecules are CD40-inducible (25, 26). The augmented proliferation in the case of B cells receiving chronic signal-1 and for MD4 B cells provided with cognate T cell help is also consistent with increased CD40 pathway activity. Our inability to detect significantly altered surface abundance of CD40L on CD4+ T cells or CD40 on B cells from Ctsb-deficient mice suggests Ctsb may affect this pathway in an indirect manner. However, it must be kept in mind that the activity of CD40L in vivo that augments autoreactive B cell survival likely reflects transient surface expression on naïve CD4+ T cells (8). Since we are not able to measure this low, transient surface CD40L, we cannot exclude that Ctsb, whether directly or indirectly, reduces the amount of CD40L available from naïve CD4+ T cells during interactions with chronically antigen-engaged B cells. In this regard, it is notable that a disintegrin and metalloprotease 10 (ADAM10) can promote reductions in surface CD40L (21). Work in cancer cell lines has indicated that Ctsb can promote the release of soluble ADAM10 (27). These observations raise the possibility that Ctsb, by increasing soluble ADAM10 in tissues, reduces the amount of CD40L available at interfaces between T cells and B cells, a possibility that merits exploration. Alternatively, the effect of Ctsb could be via a distinct pathway and this pathway can only show an influence on B cell survival and proliferation when the CD40L-CD40 pathway is intact; when this pathway is missing, the Ctsb-regulated pathway may be insufficient to have an influence. Future studies examining global gene expression changes in autoantigen-engaged B cells in control versus Ctsb-deficient recipients at different time points may help determine the signaling pathway(s) most influenced by Ctsb. Our study provides an example of a protease acting to influence the extent of tolerance induction in lymphoid tissues. Although less extensively tested, our work suggests this activity may also restrain the amount of help provided during responses to some foreign anti- gens. Interestingly, a recent report found that deficiency in prolidase, a cytosolic metallopeptidase, causes spontaneous T cell activation and lupus-like autoimmunity (28). It will be of interest in the future to determine whether Ctsb deficiency predisposes to systemic auto- immune disease. Moreover, increases in protease activity are common at sites of inflammation and in tumors. It will be impor- tant in future studies to determine the extent to which immune checkpoints are reset in these sites due to proteolytic modulation of immune cell communication molecules. Materials and Methods Mice. All mice were bred internally, and 6- to 20-wk-old mice of both sexes were used. Cathepsin-deficient [B6;129-Ctsbtm1Jde/J] mice were obtained from JAX and were backcrossed six times to C57BL/6J. CD40-deficient [B6.129P2-Cd40tm1Kik/J] mice were obtained from JAX. MD4 mice [C57BL/6-Tg(IghelMD4)4Ccg/J] and UBC-GFP [Tg(UBC-GFP)30Scha/J] mice were from an internal colony. ML5 mice [Tg(ML5sHEL)5Ccg] were obtained from Julie Zikherman, University of California, San Francisco, CA. CD45.1 B6 [B6.SJL-PtprcaPepcb/BoyCrCrl] mice used for chi- mera recipients and cell transfer experiments were bred internally from founders ordered from JAX. In most experiments, littermates were used as controls and experimental animals were co-caged in groups of two to six whenever possible. All mice were analyzed between 8 and 20 wk of age. Animals were housed in a specific pathogen-free environment in the Laboratory Animal Research Center at UCSF, and all experiments conformed to ethical principles and guidelines approved by the UCSF Institutional Animal Care and Use Committee. Cathepsin B Activity Assay. Spleens from control and Ctsb−/− mice were mashed through a 70-µm cell strainer in 700μL of PBS. Resulting cell suspensions were centrifuged at 1,500 rpm at 4 °C to pellet cells. The interstitial fluid-contain- ing supernatant (or “mashate”) was collected, diluted 1:10, and Ctsb activity was measured using a fluorometric Ctsb activity assay (Abcam). Adoptive Transfer of MD4 B Cells. Spleens and lymph nodes from CD45.1 MD4 mice were macerated and resulting cell suspensions were filtered through a 70-μm mesh into PBS supplemented with 2% FCS and 1mM EDTA. Counting beads were used for the enumeration of cells, and frequency of MD4 B cells was determined by staining of IgMa-positive B cells on the flow cytometer. Cells were labeled with Cell Trace Violet (CTV) (Invitrogen) and 5 to 10 × 106 MD4 B cells were injected intravenously into recipient mice. For positioning of HEL-binding B cells, spleens and lymph nodes from CD45.1 MD4 GFP mice were prepared as above. MD4 B cells were enriched by negative selection using biotinylated antibodies against CD43, CD4, CD8, TCRβ, CD11c, and Ter119, followed by streptavidin-con- jugated beads (EasySep Streptavidin RapidSpheres) to greater than 95% purity. Counting beads were used for the enumeration of cells, and 5 to 10 × 106 MD4 B cells were injected intravenously into recipient mice. For CD40-deficient MD4 experiments, CD45.1 WT MD4 and CD45.2 CD40−/− MD4 spleens and lymph nodes were prepared as above. WT and CD40−/− MD4 splenocytes were mixed such that the MD4 B cells reached a 1:1 ratio. Cells were labeled with CTV, and 10 to 15 × 106 MD4 B cells were injected intravenously into recipient mice. HEL and Antibody Treatments. Recipient mice were injected intravenously with 1 mg hen egg lysozyme (HEL) (Sigma) 1 d after adoptive transfer of MD4 B cells. For CD4+ T cell depletion experiments, 250 μg anti-mouse CD4 GK1.5 monoclo- nal antibody (BioXCell) was injected intravenously 2 to 3 h before HEL treatment on d0 and on d1.5. For CD40L blocking experiments, 250 μg anti-mouse CD40L MR1 monoclonal antibody (BioXCell) was injected intravenously on d0 and d1.5. HEL Serum Measurements. Serum was collected from control or Ctsb−/− mice 3 d after 1 mg HEL treatment or from control ML5+ or Ctsb−/− ML5+ mice. MD4 B cells were plated in a 96-well plate at 5 × 106 cells per mL in PBS supplemented with 2% FCS and incubated with sera at 1:10 and 1:50 dilutions for 20 min on ice. Cells were washed three times and HEL occupancy was measured by staining with HyHEL9–PE-Cy5.5 (in house). Bone Marrow Chimeras. CD45.1 B6 or Ctsb−/− mice were lethally irradiated with 1,100 rads gamma irradiation (split dose separated by 3 h) and then i.v. injected with relevant BM cells under isoflurane anesthesia. Chimeras were used as recipients for adoptive transfer experiments after 8 to 10 wks of reconstitution. Immunizations. Recipient mice were immunized intravenously with 2 × 108 SRBC (Colorado Serum Company) in a volume of 200 μL. Spleens were harvested on day 5. For SRBC-HEL2× experiments, 1 × 105 MD4 B cells were CTV-labeled and transferred into recipient mice. The following day, mice were immunized PNAS 2023 Vol. 120 No. 16 e2300099120 https://doi.org/10.1073/pnas.2300099120 7 of 9 intravenously with 2 × 108 SRBC-HEL2×, and spleens were harvested on day 5 post-immunization. Conjugation of SRBC-HEL2× was done as previously described (29). Briefly, SRBCs were first washed with PBS three times, mixed with 10 μg/mL HEL2× (Gift of R. Brink), crosslinked with EDCI (1-ethyl-3-(3-dimethylaminopropyl) carbodiimide) (Sigma-Aldrich) for 30 min, and washed three times to remove the free HEL. Immunofluorescence. Lymph nodes and spleens were embedded in optimal cutting temperature compound. Cryosections of 7 μm were dried for 1 h, fixed in acetone for 10 min, and dried for 1 h at room temperature. Slides were rehydrated in PBS containing 0.1% fatty acid-free bovine serum albumin (BSA) for 10 min. A solution consisting of 1% normal mouse serum (NMS), 1% normal donkey serum (NDS), 1:100 AF647-conjugated anti-mouse CD21/35 (BD Bioscience, catalog no. 123424) and 1:300 goat anti-mouse IgD (Cedarlane Laboratories, GAM/ IGD(FC)/7S) was used to label FDCs and endogenous naïve B cells, respectively. This solution was incubated with the slides overnight at 4 °C. The slides were then washed in PBS and stained with AMCA-conjugated donkey anti-goat IgG (Jackson Immunoresearch, 705-156-147) at room temperature, and images were captured with a Zeiss AxioObserver Z1 inverted microscope. To track the positioning of GFP MD4 B cells, spleens were fixed in 4% par- aformaldehyde (PFA) for 2 h at 4 °C, washed with phosphate-buffered saline (PBS), submerged in 30% sucrose overnight, and embedded in optimal cutting temperature compound. Cryosections of 7 μm were dried for 1 h at room tem- perature and rehydrated in PBS containing 0.1% fatty acid-free bovine serum albumin (BSA) for 10 min. A solution consisting of 1% NMS and NDS, 1:100 AF488-conjugated rabbit anti-GFP (Invitrogen, catalog no. A21311), 1:100 AF647-conjugated anti-mouse CD21/35, 1:100 biotin-conjugated anti-mouse CD4 (BioLegend, catalog no. 100404), and 1:300 goat anti-mouse IgD was used to label MD4 B cells, FDCs, endogenous CD4 T cells, and endogenous naïve B cells, respectively. These solutions were incubated with the slides overnight at 4 °C. The slides were then washed in PBS and stained with Cy3-conjugated streptavidin (Jackson Immunoresearch, catalog no. 016-160-084) and AMCA-conjugated don- key anti-goat IgG for 1 h at room temperature, and images were captured with a Zeiss AxioObserver Z1 inverted microscope. Quantification of IF Images. Images of immunofluorescent stains of GFP MD4 B cell positioning were quantified using ImageJ (version 1.53). All images were captured at the same magnification using a Zeiss AxioObserver Z1 inverted micro- scope. Images were loaded into ImageJ and the freehand line tool was used to outline the T-B interface of a follicle. The total number of GFP MD4 B cells 20 μm on either side of the T-B interface was quantified. The number of GFP MD4 B cells inside the follicle was also quantified, and the proportion of GFP MD4 B cells at the T-B interface to GFP MD4 B cells in the follicle was calculated. Cell Culture. CD4+ T cells from control and Ctsb−/− lymph nodes were enriched by negative selection using biotinylated antibodies against B220, CD8, CD11c, and Ter119, followed by streptavidin-conjugated beads to greater than 95% purity. Cells were plated in a 96-well plate at 5 × 106 cells per mL in RPMI medium 1640 plus 10% FCS at 37 °C and rested or stimulated for 2 h with 10 ng/mL PMA and 1 μg/mL ionomycin. To facilitate the detection of surface-exposed CD40L, the cells were incubated in the presence of 1 mg/mL anti-CD40L antibody as previ- ously described (30, 31). To mimic CD40 deficiency, we plated CD4+ T cells from control and Ctsb−/− lymph nodes in a dilute culture of 0.5 mL at 5 × 105 cells per mL in 24-well plates and kept on ice or incubated at 37 °C for 2 h. Real-Time PCR Analysis. B6 and Ctsb−/− spleens and lymph nodes were har- vested and CD4+ T cells were enriched by negative selection using biotinylated antibodies against CD19, B220, CD8, CD11c, and Ter119, followed by streptavi- din-conjugated beads to greater than 95% purity. Cell pellets were snap-frozen in liquid nitrogen, and RNA was prepared by using an RNeasy kit (Qiagen). Equivalent amounts of cDNA were used in quantitative PCR on an ABI 7300 sequence detection instrument (Applied Biosystems) by using primer sets with SYBR Green (Bio-Rad). Primer pairs were as follows (forward, F; reverse, R): hypox- anthine phosphoribosyltransferase (HPRT) F, AGGTTGCAAGCTTGCTGGT, and HPRT R, TGAAGTACTCATTATAGTCAAGGGCA; CD40 ligand F, GTGAGGAGATGAGAAGGCAA, and CD40 ligand R, CACTGTAGAACGGATGCTGC. Flow Cytometry. Cells were stained for 20 min on ice in MACS buffer (2% FCS in PBS with 1 mM EDTA) at 0.5 to 1 × 106 cells per well in 96-well round-bottom plates unless otherwise specified. The following monoclonal antibodies were used: B220–BV785 (BioLegend), ICAM1–biotin (BD), followed by streptavidin–BV711 (Fisher), CD45.2–PerCP-Cy5.5 (Tonbo), IgMa–FITC (Fisher), CD40–PE (Fisher), CD23–PE-Cy7 (BioLegend), CD45.1–BV605 (BioLegend), CD40L–PE (BioLegend), and CD40L–biotin (eBioscience), followed by streptavidin–A647 (Fisher). Dead cells were excluded using Fixable Viability Dye eFluor780 (eBioscience no. 65-0865-18). All samples were run on a BD LSRII or BD FACSymphony A1 at 5,000 to 10,000 events per second. Flow cytometry data were analyzed using FlowJo (v10.8.1). BAFF Staining. Staining for BAFF occupancy of BAFFR on B cells was done as previously described (5). In brief, after FcR blocking, spleen cells were incubated with rabbit anti-mouse BAFF-ectodomain serum at a 1:1,000 dilution, followed by 1:100 goat anti-rabbit–biotin with 2% NMS, normal rat serum (NRS), and normal goat serum (NGS), followed by streptavidin–A647 and antibodies to other markers. As a positive control, cells were incubated with recombinant BAFF prior to the antibody staining. Statistical Analyses. Data were analyzed using unpaired Student’s t test, and ordinary one-way ANOVA using Tukey’s multiple comparisons test was performed when comparing one variable across multiple groups. Prism version 9 (GraphPad Software) was used for all statistical analyses and to generate plots. Each experi- ment was repeated at least three times, unless otherwise indicated in the figure legends. In summary graphs, points indicate individual samples and horizontal lines are means. All error bars represent SDs. Data, Materials, and Software Availability. All study data are included in this article and/or SI Appendix. ACKNOWLEDGMENTS. We thank Julie Zikherman for ML5 mice, Julie and Anthony DeFranco for feedback on the project. M.Y.C. was supported by the Biomedical Sciences (BMS) graduate program training grant; J.G.C. is an inves- tigator of the HHMI. This work was supported in part by NIH grant R01 AI45073. Author affiliations: aDepartment of Microbiology and Immunology, University of California, San Francisco, CA 94143; and bHHMI, University of California, San Francisco, CA 94143 1. 2. 3. 4. 5. C. C. Goodnow, J. Sprent, B. Fazekas de St Groth, C. G. Vinuesa, Cellular and genetic mechanisms of self tolerance and autoimmunity. Nature 435, 590–597 (2005). C. C. Goodnow, J. Crosbie, H. Jorgensen, R. A. Brink, A. Basten, Induction of self-tolerance in mature peripheral B lymphocytes. Nature 342, 385–391 (1989). J. G. Cyster, S. B. Hartley, C. C. Goodnow, Competition for follicular niches excludes self-reactive cells from the recirculating B-cell repertoire. Nature 371, 389–395 (1994). J. G. Cyster, C. C. Goodnow, Antigen-induced exclusion from follicles and anergy are separate and complementary processes that influence peripheral B cell fate. Immunity 3, 691–701 (1995). R. Lesley et al., Reduced competitiveness of autoantigen-engaged B cells due to increased dependence on BAFF. Immunity 20, 441–453 (2004). 9. T. Yadati, T. Houben, A. Bitorina, R. Shiri-Sverdlov, The ins and outs of cathepsins: Physiological function and role in disease management. Cells 9, 1679 (2020). 10. S. Roshy, B. F. Sloane, K. Moin, Pericellular cathepsin B and malignant progression. Cancer Metastasis Rev. 22, 271–286 (2003). 11. K. Porter, Y. Lin, P. B. Liton, Cathepsin B is up-regulated and mediates extracellular matrix degradation in trabecular meshwork cells following phagocytic challenge. PLoS One 8, e68668 (2013). 12. K. N. Balaji, N. Schaschke, W. Machleidt, M. Catalfamo, P. A. Henkart, Surface cathepsin B protects cytotoxic lymphocytes from self-destruction after degranulation. J. Exp. Med. 196, 493–503 (2002). 13. S. M. Mariani, P. H. Krammer, Differential regulation of TRAIL and CD95 ligand in transformed cells of the T and B lymphocyte lineage. Eur. J. Immunol. 28, 973–982 (1998). 6. M. Thien et al., Excess BAFF rescues self-reactive B cells from peripheral deletion and allows them to 14. M. Guo, P. A. Mathieu, B. Linebaugh, B. F. Sloane, J. J. Reiners Jr., Phorbol ester activation of a 7. 8. enter forbidden follicular and marginal zone niches. Immunity 20, 785–798 (2004). K. N. Schmidt, J. G. Cyster, Follicular exclusion and rapid elimination of hen egg lysozyme autoantigen-binding B cells are dependent on competitor B cells, but not on T cells. J. Immunol. 162, 284–291 (1999). R. Lesley, L. M. Kelly, Y. Xu, J. G. Cyster, Naive CD4 T cells constitutively express CD40L and augment autoreactive B cell survival. Proc. Natl. Acad. Sci. U. S. A. 103, 10717–10722 (2006). proteolytic cascade capable of activating latent transforming growth factor-betaL a process initiated by the exocytosis of cathepsin B. J. Biol. Chem. 277, 14829–14837 (2002). 15. L. Hasan et al., Function of liver activation-regulated chemokine/CC chemokine ligand 20 is differently affected by cathepsin B and cathepsin D processing. J. Immunol. 176, 6512–6522 (2006). 16. J. Cosgrove et al., B cell zone reticular cell microenvironments shape CXCL13 gradient formation. Nat. Commun. 11, 3677 (2020). 8 of 9 https://doi.org/10.1073/pnas.2300099120 pnas.org 17. L. Jackson, C. T. Cady, J. C. Cambier, TLR4-mediated signaling induces MMP9-dependent cleavage of 24. E. H. Ekland, R. FOrster, M. Lipp, J. G. Cyster, Requirements for follicular exclusion and competitive B cell surface CD23. J. Immunol. 183, 2585–2592 (2009). 18. F. Mackay, P. Schneider, P. Rennert, J. Browning, BAFF AND APRIL: A tutorial on B cell survival. Annu. Rev. Immunol. 21, 231–264 (2003). 19. M. J. Yellin et al., CD40 molecules induce down-modulation and endocytosis of T cell surface T cell-B cell activating molecule/CD40-L. Potential role in regulating helper effector function. J. Immunol. 152, 598–608 (1994). 20. B. Ludewig et al., Spontaneous apoptosis of dendritic cells is efficiently inhibited by TRAP (CD40-ligand) and TNF-alpha, but strongly enhanced by interleukin-10. Eur. J. Immunol. 25, 1943–1950 (1995). 21. D. Yacoub et al., CD154 is released from T-cells by a disintegrin and metalloproteinase domain- containing protein 10 (ADAM10) and ADAM17 in a CD40 protein-dependent manner. J. Biol. Chem. 288, 36083–36093 (2013). 22. D. Paus et al., Antigen recognition strength regulates the choice between extrafollicular plasma cell and germinal center B cell differentiation. J. Exp. Med. 203, 1081–1091 (2006). elimination of autoantigen binding B cells. J. Immunol. 172, 4700–4708 (2004). 25. J. Banchereau et al., The CD40 antigen and its ligand. Annu. Rev. Immunol. 12, 881–922 (1994). 26. L. K. Busch, G. A. Bishop, The EBV transforming protein, latent membrane protein 1, mimics and cooperates with CD40 signaling in B lymphocytes. J. Immunol. 162, 2555–2561 (1999). 27. F. C. Sigloch et al., Proteomic analysis of silenced cathepsin B expression suggests non-proteolytic cathepsin B functionality. Biochim. Biophys. Acta 1863, 2700–2709 (2016). 28. R. Hodgson et al., Prolidase deficiency causes spontaneous T cell activation and lupus-like autoimmunity. J. Immunol. 210, 547–557 (2023). 29. T. Yi, J. G. Cyster, EBI2-mediated bridging channel positioning supports splenic dendritic cell homeostasis and particulate antigen capture. Elife 2, e00757 (2013). 30. Y. Koguchi, T. J. Thauland, M. K. Slifka, D. C. Parker, Preformed CD40 ligand exists in secretory lysosomes in effector and memory CD4+ T cells and is quickly expressed on the cell surface in an antigen-specific manner. Blood 110, 2520–2527 (2007). 23. K. M. Ansel et al., A chemokine driven positive feedback loop organizes lymphoid follicles. Nature 31. J. L. Gardell, D. C. Parker, CD40L is transferred to antigen-presenting B cells during delivery of T-cell 406, 309–314 (2000). help. Eur. J. Immunol. 47, 41–50 (2017). PNAS 2023 Vol. 120 No. 16 e2300099120 https://doi.org/10.1073/pnas.2300099120 9 of 9
10.1073_pnas.2221652120	RESEARCH ARTICLE \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS Enhanced inhibition of MHC-I expression by SARS-CoV-2 Omicron subvariants Miyu Moriyamaa, Carolina Lucasa, Valter Silva Monteiroa , Yale SARS-CoV-2 Genomic Surveillance Initiative1, and Akiko Iwasakia,b,c,2 Contributed by Akiko Iwasaki; received December 21, 2022; accepted March 9, 2023; reviewed by Jonathan C. Kagan and Yoshihiro Kawaoka Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants of concern (VOCs) possess mutations that confer resistance to neutralizing antibodies within the Spike protein and are associated with breakthrough infection and reinfection. By contrast, less is known about the escape from CD8+ T cell-mediated immunity by VOC. Here, we demonstrated that all SARS-CoV-2 VOCs possess the ability to suppress major histocompatibility complex class I (MHC-I) expression. We identified several viral genes that contribute to the suppression of MHC I expression. Notably, MHC-I upregulation was strongly inhibited after SARS-CoV-2 but not influenza virus infection in vivo. While earlier VOCs possess similar capacity as the ancestral strain to suppress MHC-I, the Omicron subvariants exhibited a greater ability to suppress surface MHC-I expression. We identified a common mutation in the E protein of Omicron that further suppressed MHC-I expression. Collectively, our data suggest that in addition to escaping from neutralizing antibodies, the success of Omicron subvariants to cause breakthrough infection and reinfection may in part be due to its optimized evasion from T cell recognition. CD8 T cell \| cytotoxic T lymphocytes \| viral evasion \| MHC \| COVID-19 SARS-CoV-2 has continued to evolve since it was first detected in Wuhan, China in December 2019. Beginning in late 2020, waves of SARS-CoV-2 variants of concern (VOCs) with increased transmissibility and immune evasion capacity have emerged. Increasing breakthrough infection and reinfection events are associated with the emergence of VOCs (1, 2). Breakthrough infections and reinfections are likely driven by significant increases in transmissibility (3), evasion from innate immunity (4, 5), and escape from neutralization by vaccine/infection-induced antibodies (6–9). By contrast, minimal evasion of T cell epitopes has been reported for VOCs (10). In November 2021, the Omicron variant, the newest VOC declared by WHO to date, had emerged. Omicron variant then quickly outcompeted the previously dominant Delta variant and led to the largest surge in COVID-19 cases worldwide. The outstanding features of the Omicron variant are the considerably enhanced escape from the antibody neutralization (9, 11) and increased infectivity (12) than the earlier VOCs, due to its heavily mutated Spike protein. Although the Omicron variant and its subvariants harbor a far greater number of mutations in its genome compared to those in previous VOCs, T cell epitopes remain generally intact (13). CD8+ cytotoxic T lymphocyte (CTL) recognizes and kills infected cells and eliminates the source of replicating viruses. Antigen presentation by major histocompatibility complex class I (MHC-I) is a critical step for the activation of antigen-specific CD8+ T cells and the subsequent killing of infected cells. Viral peptides processed by the cellular proteasome complex are loaded on MHC-I molecule in the endoplasmic reticulum and translocate to the cell surface to be recognized by antigen-specific CD8+ T cells. To successfully establish infection and replicate in the host, many viruses have acquired the ability to inhibit MHC-I processing and presentation of viral antigens (14). Likewise, SARS-CoV-2 utilizes its viral proteins to interfere with the MHC-I pathway (15–19). SARS-CoV-2 ORF8 protein induces autophagic degradation of MHC-I and confers resistance to CTL surveillance (15). Studies from the first 3 month of the pandemic showed a rapid evolution of the SARS-CoV-2 ORF8 gene including isolates with 382 nt deletion spanning the ORF7b-ORF8 gene region (20, 21), which is associated with robust T cell response and a milder clinical outcome (22, 23). These findings collectively raised a question of whether VOC and its ORF8 protein have evolved to further enhance the ability to shut down MHC-I, thereby evading from antigen-specific memory CD8+ T cells established by previous infection or vaccination. Here, we performed a systematic analysis of the capacity of SARS-CoV-2 variants to down-regulate MHC-I presentation. Our data demonstrated vigorous suppression of MHC-I surface expression by the ancestral SARS-CoV-2 and minimal evolution in mod- ulating MHC-I pathway by earlier VOCs. Remarkably, the latest Omicron subvariants have acquired an enhanced ability in modulating MHC-I pathway. Significance Numerous pathogenic viruses have developed strategies to evade host CD8+ T cell-mediated clearance. Here, we demonstrated that SARS-CoV-2 encodes multiple viral factors that can modulate major histocompatibility complex class I (MHC-I) expression in the host cells. We found that MHC-I upregulation was strongly suppressed during SARS-CoV-2, but not influenza virus infection, in vivo. Notably, the Omicron subvariants showed an enhanced ability to suppress MHC-I compared to the original strain and the earlier SARS-CoV-2 variants of concern (VOCs). We identified a mutation in the E protein shared by the Omicron subvariants that further suppressed MHC-I expression. Our results point to the inherently strong ability of SARS-CoV-2 to hinder MHC-I expression and demonstrated that Omicron subvariants have evolved an even more optimized capacity to evade CD8 T cell recognition. Reviewers: J.C.K., Boston Children’s Hospital; and Y.K., University of Wisconsin-Madison. Competing interest statement: A.I. served as a consultant for RIGImmune, Xanadu Bio, Paratus Bio, and Invisishield. A.I. owns equity in RIGImmune and Xanadu Bio. A.I. has patent filings for RIGImmune and Xanadu Bio. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1A complete Surveillance Initiative can be found in the SI Appendix. list of the Yale SARS-CoV-2 Genomic 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221652120/-/DCSupplemental. Published April 10, 2023. PNAS 2023 Vol. 120 No. 16 e2221652120 https://doi.org/10.1073/pnas.2221652120 1 of 10 Results Pre-Omicron SARS-CoV-2 Variants Retain Similar MHC-I Evasion Capacity. To investigate the impact of SARS-CoV-2 infection on MHC-I expression, we infected Calu-3 cells, a commonly used human lung epithelial cell line, with SARS-CoV-2 variants and the ancestral strain (USA-WA1). We tested four VOC (B.1.1.7/ Alpha, B.1.351/Beta, P.1/Gamma, and B.1.617.2/Delta) and three variants of interest (B.1.427/Epsilon, B.1.429/Epsilon, and B.1.526/Iota). To assess MHC-I expression levels, the cells were pregated for single cells and live cells (SI Appendix, Fig. S1). Infection with SARS-CoV-2 variants reduced the viability of the cells by ~30% compared to the mock-infected condition (SI Appendix, Fig. S1 A and B). Within the live cell population, SARS-CoV-2 variants and the ancestral strain similarly down-regulated MHC-I levels after infection (Fig. 1A). We next examined transcriptional levels of MHC-I genes after infection with SARS-CoV-2 variants. Transcriptional levels of MHC-I genes differed depending on the variants (Fig. 1B). The ancestral strain significantly down-regulated human leukocyte antigen (HLA)-A, B, and C genes as previously reported (16). B.1.1.7 and B.1.351 showed a similar reduction in HLA-A, B, and C mRNA expression as the ancestral strain. Other variants showed a weaker downregulation (B.1.526), no significant change (B.1.429), or upregulation (P.1) of HLA class I genes within the infected cells. To gain further mechanistic insights into differential transcriptional levels of MHC-I in SARS-CoV-2 variants infection, we determined if the master transcription factors of MHC-I gene, NLRC5 and IRF1 (24), are differentially expressed in these cells. We observed the upregulation of both NLRC5 and IRF1 by SARS-CoV-2 infection, where some VOCs had lower expression levels compared to the WA1 strain (Fig. 1C). These results indicated that despite induction of NLRC5 and IRF1, SARS-CoV-2 variants maintain a similar capacity to reduce HLA-I mRNA levels as the ancestral virus, except for the P.1 and B.1.429 variants. Given that P.1-infected and B.1.429- infected cells still expressed low levels of MHC-I on the surface, other posttranscriptional mechanisms involved in the MHC-I processing and presentation pathway must account for the reduced surface MHC-I expression. Variant-Specific Mutations Are Found in the ORF8 Gene of SARS- CoV-2. Because previous studies revealed the association between 382-nt deletion spanning the ORF7b-ORF8 gene region and robust interferon (IFN)-γ and T cell responses (22, 23), we next addressed the role of ORF8 in the differential MHC-I regulation by VOCs. We performed multiple sequence alignments of ORF8 amino acid sequences from SARS-CoV-2 variants to see if there are any nonsynonymous mutations. In total, eight nonsynonymous mutations and two deletions were detected from 16 variants examined (SI Appendix, Fig. S2). Notably, a premature stop codon A B C Fig. 1. MHC-I evasion by SARS-CoV-2 variants. (A) Calu-3 cells were infected with SARS-CoV-2 variants at MOI 0.3 for 40 h. The cell surface expression of HLA-ABC was analyzed by flow cytometry. (B and C) Calu-3 cells were infected with SARS-CoV-2 variants at MOI 0.01 for 48 h. The mRNA levels of HLA-A, -B, -C (B), NLRC5, and IRF1 (C) were measured by qRT-PCR. Data are mean ± SD. Data are representative of two to three independent experiments. P < 0.05; P < 0.01; P < 0.001. 2 of 10 https://doi.org/10.1073/pnas.2221652120 pnas.org was introduced at Q27 of B.1.1.7, which truncates the ORF8 polypeptide length and likely alters the protein functionality. The downstream mutations (R52I and Y73C) probably have no further impact on B.1.1.7 ORF8 protein because of the early translation termination by Q27stop mutation. Although pre-Omicron VOC (B.1.1.7/Alpha, B.1.351/Beta, P.1/Gamma, and B.1.617.2/Delta) harbored mutations or deletions in ORF8 protein, ORF8 sequence from BA.1/Omicron variant and its descendants remained intact. Two of the former variants of interest, B.1.429/Epsilon and B.1.526/Iota harbored V100L and T11I mutation, respectively. None of the mutations and deletions were conserved among different lineages. To investigate the prevalence of mutations found in variants, we downloaded 3,059 SARS-CoV-2 genome sequence data from GISAID database (https://www.gisaid. org/). We found that the mutation in a particular amino acid is only exclusively seen in a single lineage (Fig. 2A). ORF8 L84S mutation, which was detected within the first 2 month of the pandemic (25) and corresponding to clade S, was not observed in any of the variants. We also observed the mutations found by multiple sequence alignment are generally highly prevalent, and the proportions ranged from 12.5 to 100% of the lineage (Fig. 2B). These results indicate that the variant-specific mutations were acquired independently during SARS-CoV-2 evolution. Impaired MHC-I Downregulation by B.1.1.7 ORF8 Protein. We next tested whether variant-specific mutations alter MHC-I downregulating capability of the ORF8 protein. To this end, we generated expression plasmids encoding seven ORF8 mutants from SARS-CoV-2 variants (Fig. 2C), and subsequently transfected HEK293T cells with these plasmids for the detection of its effect on the surface MHC-I expression levels. We included SARS-CoV ORF8a/b proteins as negative controls, as they have been shown not to affect MHC-I expression levels (15). Since ORF8 induces degradation of MHC-I via autophagy by interacting with MHC-I and localizing in LC3-positive puncta (15), ORF8 presumably acts on MHC-I downregulation in a cell-intrinsic manner. Indeed, surface MHC-I levels of the cells expressing WT ORF8 protein were much lower than those of the cells without ORF8 expression (Fig. 2D). In addition to the surface MHC-I, intracellular MHC-I molecules were also decreased specifically in ORF8-expressing cells (SI Appendix, Fig. S3A), further supporting the direct role of ORF8 in the control of cellular MHC-I levels. The cysteine 20 residue of SARS-CoV-2 ORF8 is known to form intermolecular disulfide bonds between two ORF8 molecules and stabilize the homodimer (26, 27). Disruption of the Cys20 residue, however, did not affect the regulation of cell surface MHC-I levels (SI Appendix, Fig. S3B). Among seven ORF8 mutants tested, six mutants including L84S, I121L, E92K, del119-120, V100L, and T11I down-regulated surface MHC-I levels of the cells expressing those proteins to a similar extent to WT ORF8 protein (Fig. 2E), while maintaining the expression of an irrelevant cell surface molecule, CD324 (Fig. 2F). On the other hand, Q27Stop ORF8 mutant had a completely abrogated MHC-I downregulation capability compared to the WT ORF8 protein (Fig. 2E). These results indicated that none of the variant-specific mutations enhanced the ability of ORF8 protein to down-regulate MHC-I, and the ORF8 encoded by the B.1.1.7 lineage lost its ability to reduce surface MHC-I expression. Multiple SARS-CoV-2 Viral Proteins Play Redundant Roles in the Downregulation of MHC-I. Given that B.1.1.7 and P.1 variants were able to reduce MHC-I expression levels even though these lineages retain functionally defective ORF8 mutant or are less effective in reducing HLA-I mRNA levels, we investigated the possibility that SARS-CoV-2 encodes multiple viral genes that redundantly act to suppress MHC-I expression. Since nascent MHC-I molecule assembly and peptide loading take place in the endoplasmic reticulum, and peptide-loaded MHC-I molecules are subsequently transported through the secretory pathway, we chose to test whether SARSCoV-2-encoded proteins which localize to these subcellular compartments modulate MHC-I expression (28, 29). We generated expression plasmids encoding the original Wuhan strain SARS-CoV-2 E, M, ORF7a, and ORF7b, and assessed the effect on the surface MHC-I and MHC-II expression levels of HEK293T cells following transfection with these plasmids. We also included HIV Nef as a positive control for downregulating both MHC-I and MHC-II (30, 31), and SARS-CoV ORF8a/b proteins as a negative control. As expected, HIV Nef protein down-regulated both MHC-I and MHC-II levels, whereas SARS- CoV-2 ORF8 specifically targeted MHC-I (Fig. 3 A and B). We found that in addition to ORF8, SARS-CoV-2 E, M, and ORF7a substantially down-regulated MHC-I within the cells expressing these viral proteins (Fig. 3C). Significant reduction of surface MHC-II levels was also observed by expression of these viral proteins (Fig. 3C), albeit to a lesser extent (~20%). These results suggested that SARS-CoV-2 encodes multiple viral genes that are redundantly down-regulating MHC-I likely to ensure viral evasion from MHC-I-mediated CTL recognition. Superior MHC-I Evasion by SARS-CoV-2 Compared to Influenza A Virus In Vivo. In the experiments above, we have shown that SARS- CoV-2 encodes multiple viral proteins that are targeting MHC-I expression, which can synergistically strengthen the capability of the virus to avoid MHC-I presentation. Moreover, we confirmed the previous finding that the MHC-I downregulation is a newly acquired function of SARS-CoV-2 ORF8 protein, which was not seen in SARS-CoV ORF8a/b proteins. Considering these results, we hypothesized that even the ancestral strain of SARS-CoV-2 possesses a superior MHC-I evasion strategy than other respiratory viruses. To assess this hypothesis, we infected C57BL/6J mice intranasally with a mouse-adapted strain of SARS-CoV-2 (SARS- CoV-2 MA10) or influenza A/PR8 virus and analyzed the MHC-I expression levels of lung epithelial cells at 2 d after infection. SARS- CoV-2 MA10 virus harbors two mutations in the Spike protein, three mutations in the ORF1ab, and an F7S mutation in ORF6 compared to the ancestral virus (32). Strikingly, influenza A virus induced robust upregulation of MHC-I in both infected (NP+) and uninfected (NP−) lung epithelial cells, whereas SARS-CoV-2 MA10 up-regulated MHC-I only in uninfected cells (S−), and to a lesser extent than the influenza virus (Fig. 4 A–C). Importantly, MHC-I upregulation was completely abrogated in SARS-CoV-2 MA10-infected (S+) lung epithelial cells, suggesting that SARS- CoV-2 viral proteins are strongly inhibiting MHC-I upregulation in a cell-intrinsic manner. To test if the superior MHC-I inhibition by SARS-CoV-2 is also seen in cultured human cells, we infected Calu- 3 cells with influenza A/PR8 virus or USA-WA1 strain of SARS- CoV-2 and compared the cell surface MHC-I expression at 48 h after infection. Consistent with the in vivo results, we found that influenza A/PR8 virus infection significantly up-regulated MHC-I expression in vitro (Fig. 4D). We observed that SARS-CoV-2 infection down- regulated MHC-I in both infected (S+) and uninfected cells (S−), but more profound effect on infected cells as seen in the infected human cell line (Fig. 4E). These results indicated that SARS-CoV-2 possesses a near complete ability to shut down MHC-I induction within infected cells in vivo, a feature not found in influenza A virus. Omicron Subvariants Down-Regulate MHC-I More Efficiently Than Earlier Isolates. Finally, we tested the ability of the more recent Omicron subvariants to counteract the MHC-I expression. We PNAS 2023 Vol. 120 No. 16 e2221652120 https://doi.org/10.1073/pnas.2221652120 3 of 10 A B C E D F Fig. 2. Unique mutations are found in ORF8 gene of SARS-CoV-2 variants. (A) Mutant proportion in the ORF8 genes of the indicated SARS-CoV-2 variants. The amino acid positions shown are selected based on the results of multiple sequence alignment performed in SI Appendix, Fig.S2. The number of sequences analyzed for each lineage is shown above each graph. (B) Frequency of amino acids at the positions enriched for mutants in each variant. The amino acids shown in gray color correspond to WT. (C) Schematic diagram of ORF8 proteins from SARS-CoV-2 variants. (D–F) HEK293T cells were transfected with plasmids encoding C-terminally Flag-tagged SARS-CoV ORF8a/b, HIV Nef, SARS-CoV-2 ORF8 WT, or SARS-CoV-2 ORF8 variants. Forty-eight hours after transfection, cells were collected and analyzed for the cell surface HLA-ABC expression (D and E) and CD324 (F). Data are shown in raw median fluorescence intensity (MFI) (D) or as the ratio of MFI in Flag+ cells to Flag- cells (E) (n = 3). Data are mean ± SD. Data are representative of two to three independent experiments. P < 0.001. generated an A549 cell line, a human adenocarcinoma cell line that stably expresses ACE2 (A549-hACE2 cells). A549-hACE2 cells were infected with five Omicron subvariants (BA.1, BA.2.12.1, XAF, BA.4, and BA.5) along with earlier isolates (WA1 and B.1.429) for comparison. We further distinguished virally infected (S+) cells from bystander cells (S−) to assess the direct impact of infection on the cell surface MHC-I expression (Fig. 5A), by looking at cell surface HLA-ABC raw MFI (Fig. 5 B and C) or normalized MFI in S+- infected cells (Fig. 5 D and E). Consistent with the observation in Calu-3 cells (Fig. 1A), SARS-CoV-2 infection suppressed MHC-I expression in A549-hACE2 cells (Fig. 5 B–D). Furthermore, MHC-I reduction was specifically seen in S+-infected cells and varied 4 of 10 https://doi.org/10.1073/pnas.2221652120 pnas.org A B C Fig. 3. SARS-CoV-2 viral genes redundantly down-regulate MHC-I. (A–C) HEK293T cells were transfected with expression plasmids encoding C-terminal Flag- tagged viral proteins as indicated. After 48 h, surface expressions of HLA-ABC and HLA-DR were analyzed by flow cytometry. The representative histogram of HLA-ABC (A) and HLA-DR (B) of SARS-CoV ORF8a-Flag, HIV-Nef-Flag, or SARS-CoV-2 ORF8-Flag +/− cells are shown. (C) The normalized ratio of HLA surface expression in Flag+ cells to Flag- cells are shown (n = 3). All SARS-CoV-2 proteins tested here are derived from the ancestral Wuhan strain. Data are mean ± SD. Data are representative of three independent experiments. Statistical significance is calculated versus SARS-CoV ORF8a P < 0.05; **P < 0.001. between different viral strains (Fig. 5B). In contrast, an irrelevant cell surface marker CD324 (E-cadherin) remained unchanged upon infection with SARS-CoV-2 (Fig. 5 C–E). Remarkably, many of the omicron subvariants, such as BA.1, BA.2.12.1, XAF, and BA.4, had a superior capacity to reduce surface MHC-I levels compared to the older isolates (Fig. 5D). To understand the differential ability of the Omicron subvariants to suppress MHC-I expression, we plotted SARS-CoV-2 S protein expression levels against the surface MHC-I levels (Fig. 5 F and G). We observed a negative correlation between S protein expression and MHC-I expression (Fig. 5F). This may explain the differential efficiency of MHC-I evasion among Omicron subvariants. Notably, although both WA1 and BA.1-infected cells expressed similar levels of S protein, the surface MHC-I expression was ~36% lower in BA.1 infection than WA1 (Fig. 5G). Similarly, WA1- and BA.5-infected cells expressed similar levels of MHC-I, even though BA.5 infection expressed ~42% lower viral S protein than WA1. Collectively, these data suggest an exaggerated rate of MHC-I evasion by Omicron subvariants per viral translational unit measured by the S protein expression. To explore the molecular mechanism of the enhanced MHC-I inhibition by Omicron subvariants, we searched for common mutations shared among Omicron subvariants that were used in this study (SI Appendix, Fig. S4). We identified common mutations in the E protein (T9I) and the M protein (Q19E/A63T), which are shared among all Omicron subvari- ants used in this study. To examine whether these common mutations can account for the superior ability of Omicron subvariants to suppress MHC-I, we generated expression plas- mids encoding mutant E and M and assessed the impact of these mutations on the surface MHC-I expression levels in HEK293T cells. We found that T9I mutation within the E protein significantly enhanced the degree of MHC-I downregulation, whereas the M mutations had no impact (Fig. 5H). Collectively, these results underscore the universal capacity of all SARS- CoV-2 strains to mediate the cell-intrinsic reduction of MHC-I expression within the infected cells and highlight the superior ability of the Omicron subvariants in acquiring MHC-I evasion capacity. PNAS 2023 Vol. 120 No. 16 e2221652120 https://doi.org/10.1073/pnas.2221652120 5 of 10 A B C D E Fig. 4. Robust suppression of MHC-I upregulation by SARS-CoV-2 in vivo. (A–C) C57BL/6J mice were infected intranasally with 105 PFU of SARS-CoV-2 MA10 or Influenza virus A/PR8. Two days later, lungs were collected and analyzed for surface MHC-I expressions on epithelial cells of viral protein [SARS-CoV-2 Spike (S) or influenza A virus nucleoprotein (NP)]-positive and -negative populations. (n = 3 to 4). The representative histograms (A) and MFI (B and C) are shown. (D and E) Calu-3 cells were infected with influenza virus A/PR8 (D) or SARS- CoV-2 WA1 (E) at MOI 0.3 for 48 h. The cell surface expression of HLA-ABC was analyzed by flow cytometry. Data are mean ± SD. Data are representative of two independent experiments. P < 0.05; P < 0.01; P < 0.001; N.S., not significant. Discussion CD8+ T cell-mediated elimination of infected cells plays an impor- tant role in the antiviral adaptive immune response. Thus, many viruses have developed ways to avoid the efficient MHC-I mediated antigen presentation to CD8+ T cells. In the current study, we uncovered the intrinsically potent ability of SARS-CoV-2 to shut down the host MHC-I system by using live, authentic SARS- CoV-2 variants as well as the functional analysis of variant-specific mutations in ORF8 gene, a key viral protein for both MHC-I evasion and adaptation to the host. We further identified multiple other viral genes that confer redundant function in MHC-I sup- pression. We show that respiratory epithelial cells infected in vivo with SARS-CoV-2 failed to up-regulate MHC-I, whereas those infected with influenza virus robustly elevated MHC-I expression. Finally, our data revealed that the most recent Omicron subvariants have superior capacity to suppress MHC-I compared to the earlier isolates, and identified a mutation within the E protein common to these Omicron subvariants that possess superior MHC-I sup- pression capacity. We demonstrated that most variants of concern/interest possess unique mutations within ORF8 gene. However, none of these ORF8 mutations led to further reduction of MHC-I levels in cells expressing these molecules. Notably, the Omicron variant and its descendants lacked nonsynonymous mutations in their ORF8 gene, indicating that the mechanisms for superior MHC-I sup- pression by the Omicron sublineages must be located outside the ORF8 protein. Indeed, we identified a T9I mutation within the E protein common to the Omicron sublineages that confers supe- rior inhibition of MHC-I expression. Our data showed that the ORF8 mutation in the B.1.1.7/Alpha lineage abrogated its function in MHC-I modulation. Given the truncation mutation likely rendering ORF8 in the Alpha variant nonfunctional, this raises a question as to how such mutations might be tolerated. Multiple functions beyond MHC-I downreg- ulation are documented for SARS-CoV-2 ORF8, which include inhibition of type I IFN, interferon-stimulated genes (ISGs), or nuclear factor-kappa B (NF-kB) signaling (33–35), epigenetic modulation through histone mimicry (36) and induction of proin- flammatory cytokines from macrophages and monocytes via IL-17RA (37–40). Interestingly, several studies showed that SARS-CoV-2 ORF8 is actively secreted into the cell culture media in a signal peptide-dependent manner when it is overexpressed in vitro (27, 37, 41). Furthermore, ORF8 peptides and anti-ORF8 antibodies can be detected abundantly in serum of patients, sug- gesting the relevance of the active secretion of ORF8 to actual infection in humans (41). The Alpha variant likely acquired com- pensatory mechanisms that enabled its successful transmission until the next variant came along. ORF8 is implicated in adaptation to the human host during the SARS-CoV outbreak (42, 43), and it is known that ORF8 is the hypervariable genomic region among the SARS-CoV and bat SARS-related CoVs (44, 45). Likewise, studies from early in the COVID-19 pandemic observed the variability and rapid evolution of SARS-CoV-2 ORF8 gene (20, 46). Notably, SARS-CoV-2 iso- lates with 382-nt deletion spanning the ORF7b-ORF8 gene region were observed in Singapore (21), which correlated with robust T cell response and a mild clinical outcome (22, 23). Mutations in ORF8 gene thus may play a key role in modulating viral patho- genesis and adaptation to the host by regulating MHC-I levels and ISGs. The enhanced immune evasion by VOCs has been well docu- mented for escape from neutralizing antibodies (6–8) and from innate immune responses (4, 5). Here, we demonstrated that the ability to reduce MHC-I expression remained unchanged through- out the pre-Omicron VOC evolution. These findings suggested three important perspectives on the MHC-I evasion strategy of SARS-CoV-2. First, SARS-CoV-2 utilizes multiple redundant strategies to suppress MHC-I expression. For example, considering B.1.1.7 retained an intact ability to shut down MHC-I, the impaired MHC-I evasion by B.1.1.7 ORF8 is likely compensated 6 of 10 https://doi.org/10.1073/pnas.2221652120 pnas.org A D F C B E G H Fig. 5. Increased suppression of MHC-I by Omicron sublineages. (A–G) A549-hACE2 cells were infected with SARS-CoV-2 variants at MOI 0.3 for 44 h. Cells were collected and analyzed for surface MHC-I expressions on single live cells of SARS-CoV-2 Spike (S)-positive and -negative populations by flow cytometry. (A) Representative flow cytometry plot. Data are shown as the raw MFI (B and C) or the normalized ratio of MFI in SARS-CoV-2 S+ cells to MFI in mock cells (D and E). SARS-CoV-2 S MFI were plotted against HLA-ABC MFI (F and G). (H) HEK293T cells were transfected with plasmids encoding C-terminally Flag-tagged SARS-CoV ORF8a, HIV Nef, SARS-CoV-2 E WT or Omicron mutant (T9I), or SARS-CoV-2 M WT or Omicron mutant (D3G/Q19E/A63T). Forty-eight hours after transfection, cells were collected and analyzed for the cell surface HLA-ABC expression. Data are shown as the ratio of MFI in Flag+ cells to Flag- cells (n = 3). Data are mean ± SD. Data are representative of two independent experiments. P < 0.05; P < 0.01; P < 0.001; N.S., not significant. by the redundant and/or compensatory functions of other viral proteins including E, M, and ORF7a. In addition, B.1.1.7 lineage has been shown to express an increased subgenomic RNA and protein abundance of ORF6 (4), which suppresses MHC-I at the transcriptional level by interfering with STAT1-IRF1-NLRC5 axis (16). The multitiered MHC-I evasion mechanisms thus work redundantly to ensure escape from CTL killing. Second, MHC-I downregulation may not only impair CTL recognition of infected cells for killing but may also impair priming of CD8 T cells. Indeed, the frequency of circulating SARS-CoV-2 specific memory CD8+ T cells in SARS-CoV-2-infected individ- uals are ~10-fold lower than for influenza or Epstein-Barr virus-specific T cell populations (47), which indicates the subop- timal induction of memory CD8+ T cells following SARS-CoV-2 infection in humans. Third, given that the VOC had not further evolved to down-regulate MHC-I more strongly than the original strain except for the Omicron subvariants, SARS-CoV-2 ancestral virus was already fully equipped to escape from CD8+ T cell-mediated immunity with respect to downregulation of MHC-I expression and is under less evolutionary pressure to further optimize the evasion strategy than those from type I IFNs or antibodies. However, mutations and evasion from particular HLA-restricted CTL epitopes have been observed in circulating SARS-CoV-2 and VOCs (48, 49). Genome-wide screening of epitopes suggested the CD8+ T and CD4+ T cell epitopes are broadly distributed throughout SARS-CoV-2 genome (50, 51), and the estimated numbers of epitopes per individual are at least 17 for CD8+ T and 19 for CD4+ T cells, respectively (51), and thus functional T cell evasion by VOCs is very limited (10). This in turn suggests that MHC-I downregulation may be a more efficient way for viruses to avoid CTL surveillance than introducing mutations in epitopes. The importance of MHC-I evasion by SARS-CoV-2 is also high- lighted by the fact that no genetic mutations or variations in the MHC-I pathway have thus far been identified as a risk factor for severe COVID (52), unlike innate immune pathways involving toll-like receptors (TLRs) and type I IFNs (53). SARS-CoV-2 infection in both human and animal models have shown to induce antigen-specific CD8+ T cell responses (54, 55), and the early CTL response correlated with a milder disease outcome in humans (56). Adoptive transfer of serum or IgG from convalescent animals alone, however, is enough to reduce viral load in recipients after SARS-CoV-2 challenge in mice and nonhuman primates (57, 58) and neutralizing antibody is shown to be a strong correlate of protection (57, 59, 60). The protective roles of CD8+ T cell- mediated immunity appear to be more important in the absence of PNAS 2023 Vol. 120 No. 16 e2221652120 https://doi.org/10.1073/pnas.2221652120 7 of 10 the optimal humoral responses/neutralizing antibody (57, 61). Circulating anti-ORF8 antibodies can be used as the highly sensitive clinical marker for SARS-CoV-2 infection early (~14 d) after symp- tom onset (41, 62), which suggests the role of ORF8 in the very early stage of the disease. ORF8-mediated MHC-I downregulation can therefore precede antigen presentation and hinder priming of viral antigen-specific CD8+ T cell immune responses. Robust MHC-I shutdown by SARS-CoV-2 may explain in part the less effective protection by CD8+ T cells and the less impact of CD8+ T cell absence compared with humoral immunity (57). Our study provided evidence of inhibition of MHC-I upregu- lation in SARS-CoV-2-infected cells in both in vitro and in vivo settings. Whether and to what extent the reduction of MHC-I impairs the recognition of infected cells by CTL for killing or impairs the priming of CD8 T cells should be addressed in future studies. We also did not exhaustively examine all viral proteins for their ability to reduce MHC-I expression, nor did we examine the requirement for various mutations in MHC-I suppression. In this study, we demonstrated the increased ability of Omicron variants to suppress MHC-I expression. The cellular mechanisms and con- sequences of enhanced MHC-I inhibition by Omicron variants on infection and disease remain to be determined. Collectively, our data shed light on the intrinsically potent ability of SARS-CoV-2 to avoid the MHC-I mediated antigen presentation to CD8+ T cells. Importantly, we observed a com- plete inhibition of MHC-I upregulation in lung epithelial cells infected with SARS-CoV-2 at the early stage of infection in a mouse model. Since the ability of ORF8 to down-regulate MHC-I is a newly acquired feature in SARS-CoV-2 ORF8 and is absent in SARS-CoV ORF8 (15), it is possible that ORF8 played a role in the efficient replication and transmission of SARS-CoV-2 in humans and contributed to its pandemic potential. Our work provides insights into SARS-CoV-2 pathogenesis and evolution and predicts difficulty for CD8 T cell-based therapeutic approaches to COVID-19. Materials and Methods Mice. Six-to-ten-week-old male C57BL6 mice were purchased from the Jackson Laboratory. All animal experiments in this study complied with federal and insti- tutional policies of the Yale Animal Care and Use Committee. Cell Lines and Viruses. HEK293T cells and A549-hACE2 cells were maintained in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 1% Penicillin- Streptomycin and 10% heat-inactivated fetal bovine serum (FBS). Calu-3 cells were maintained in Minimum Essential Medium (MEM) supplemented with 1% Penicillin-Streptomycin and 10% heat-inactivated FBS. ACE2-TMPRSS2-VeroE6 cells were maintained in DMEM supplemented with 1% sodium pyruvate, 1% Penicillin-Streptomycin and 10% heat-inactivated FBS at 37 °C. Influenza virus A/Puerto Rico/8/34 was kindly provided by Hideki Hasegawa (National Institute of Infectious Diseases in Japan). Virus stocks were propagated in allantoic cav- ities from 10-to-11-d-old fertile chicken eggs for 2 d at 35 °C. Viral titers were determined by standard plaque assay procedure. SARS-CoV-2 MA10 (32) was kindly provided by Ralph S. Baric (University of North Carolina at Chapel Hill). SARS-CoV-2 lineage A (USA-WA1/2020) and B.1.351b (hCoV-19/South Africa/ KRISP-K005325/2020) were obtained from BEI resources. Pre-Omicron line- ages [B.1.1.7(GenBank Accession: MZ202178), B.1.351a (GenBank Accession: MZ202314), P.1 (GenBank Accession: MZ202306), B.1.617.2 (GenBank Accession: MZ468047), B.1.427 (GenBank Accession: MZ467318), B.1.429 (GenBank Accession: MZ467319), and B.1.526 (GenBank Accession: MZ467323)] and Omicron sub-lineages [BA.1 (GenBank Accession: ON425981), BA.2.12.1 (GenBank Accession: ON411581), XAF (GenBank Accession: OP031604), BA.4 (GenBank Accession: ON773234), and BA.5.2.1 (GenBank Accession: OP031606)] were isolated and sequenced as part of the Yale Genomic Surveillance Initiative’s weekly surveillance program in Connecticut, United States, as previously described (63). Virus stocks were propagated and titered as previously described (8, 64). Briefly, TMPRSS2-VeroE6 cells were infected at multiplicity of infection of 0.01 for 3 d and the cell-free supernatant was collected and used as working stocks. All experiments using live SARS-CoV-2 were performed in a biosafety level 3 laboratory with approval from the Yale Environmental Health and Safety office. Viral Genome Sequence Analysis. Pre-Omicron SARS-CoV-2 variant genome sequences (3,067 sequences) were downloaded from GISAID database (https://www.gisaid.org/) as of February 23, 2022. Sequences of Wuhan Hu-1 (GenBank accession: NC_045512.2) and USA-WA1/2020 (GenBank accession: MW811435.1) were obtained from NCBI Virus SARS-CoV-2 Data Hub (https:// www.ncbi.nlm.nih.gov/labs/virus/vssi/#/sars-cov-2). To investigate the prevalence of amino acid mutations, we downloaded up to 965 sequences of each lineage and aligned the ORF8 nucleotide sequences using Jalview software (http:// www.jalview.org/) (Waterhouse et al. Bioinformatics. 2009) by MUSCLE algo- rithm (Edgar RC. Nucleic Acid Res. 2004). Sequences containing undetermined nucleotides within the codon of interest were removed for analysis. ORF8 amino acid sequence alignment was conducted by Jalview software using MUSCLE algorithm. Viral Infection. Mice were fully anesthetized by intraperitoneal injection of keta- mine and xylazine, and intranasally inoculated with 50 µL of phosphate-buffered saline (PBS) containing 1 × 105 PFU of influenza virus A/Puerto Rico/8/34. For SARS-CoV-2 infection in the animal biosafety level 3 facility, mice were anesthe- tized by 30% v/v isoflurane diluted in propylene glycol, and 50 µL of 1 × 105 PFU of SARS-CoV-2 MA10 in PBS was intranasally delivered. For cell culture infection, cells were washed with PBS and infected with SARS-CoV-2 at a multiplicity of infec- tion of 0.01 or 0.3 for 1 h at 37 °C. After 1-h incubation, cells were supplemented with complete media and cultured until sample harvest. Plasmids. pDONR207-SARS-CoV-2 E (#141273), pDONR207-SARS-CoV-2 M (#141274), pDONR207-SARS-CoV-2 ORF7a (#141276), pDONR223-SARS-CoV-2 ORF7b (#141277), and pDONR223-SARS-CoV-2 ORF8 (#141278) were purchased from addgene (65) and used as templates for construction of plasmids express- ing SARS-CoV-2 viral proteins. For HIV Nef expressing plasmid construction, NL4- 3-dE-EGFP (kindly provided by Ya-Chi Ho) was used as a template. The full-length viral genes were amplified by PCR using iProof™ High-Fidelity DNA Polymerase (Bio-Rad), with templates described above and specific primers containing XhoI (XbaI for HIV Nef) and BamHI sites at the 5′ and 3′ ends, respectively. Following restriction enzyme digestion, PCR fragments were cloned into c-Flag pcDNA3 vector (addgene, #20011). For construction of plasmids expressing SARS-CoV viral proteins, oligonucleotides corresponding to both strands of SARS-CoV Tor2 (GenBank accession: NC_004718.3) ORF8a and ORF8b containing XhoI and BamHI sites at the 5′ and 3′ ends were synthesized (IDT) and cloned into XhoI-BamHI site of c-Flag pcDNA3 vector. Mutant SARS-CoV-2 E, M, and ORF8 expressing plasmids were generated by standard PCR-based mutagenesis method. Integrity of inserts was verified by sequencing (Yale Keck DNA sequenc- ing facility). Lung Cell Isolation. Lungs were harvested and processed as previously described (57). In brief, lungs were minced with scissors and digested in RPMI1640 media containing 1 mg/mL collagenase A, 30 µg/mL DNase I at 37 °C for 45 min. Digested lungs were then filtered through a 70-µm cell strainer and treated with ACK buffer for 2 min. After washing with PBS, cells were resuspended in PBS with 1% FBS. Flow Cytometry. Cells were blocked with Human BD Fc Block (Fc1.3216, 1:100, BD Biosciences) in the presence of Live/Dead Fixable Aqua (Thermo Fisher) for 15 min at room temperature. Staining antibodies were added and incubated for 20 min at room temperature. Cells were washed with 2 mM EDTA-PBS and resuspended in 100 µL 2% PFA for 1 h at room temperature. For intracellu- lar staining, PFA-fixed cells were washed and permeabilized with eBioscience FoxP3/Transcription Factor Staining Buffer (Thermo Fisher) for 10 min at 4 °C. Cells were washed once and stained in the same permeabilization buffer con- taining staining antibodies. After 30-min incubation at 4 °C, cells were washed and resuspended in PBS with 1% FBS for analysis on Attune NxT (Thermo Fisher). FlowJo software (Tree Star) was used for the data analysis. Staining antibod- ies are as follows: Hu Fc Block Pure Fc1.3216 (BD, Cat# 564220), APC anti- HLA-ABC (Thermofisher, Cat# 17-9983-42), APC/Cy7 anti-HLA-DR (BioLegend, Cat# 307618), BV421 anti-mouse/human CD324 (Biolegend, Cat# 147319), 8 of 10 https://doi.org/10.1073/pnas.2221652120 pnas.org PE anti-DYKDDDDK Tag (BioLegend, Cat# 637309), AF488 anti-SARS-CoV-2 Spike S1 Subunit (R&D Systems, Cat# FAB105403G), FITC anti-Influenza A NP (Thermofisher, Cat# MA1-7322), PE anti-mouse CD45 (BioLegend, Cat# 109808), BV421 anti-mouse CD31 (BioLegend, Cat# 102423), APC anti-mouse EpCAM (BioLegend, Cat# 118213), and PerCP/Cy5.5 anti-H-2Kb/H-2Db (BioLegend, Cat# 114620). Statistical Analysis. Statistical significance was tested using one-way ANOVA with Tukey’s multiple comparison test. P-values of <0.05 were considered sta- tistically significant. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. Quantitative PCR. SARS-CoV-2-infected cells were washed with PBS and lysed with TRIzol reagent (Invitrogen). Total RNA was extracted using the RNeasy mini kit (QIAGEN) and reverse transcribed into cDNA using the iScript cDNA syn- thesis kit (Bio-Rad). RT-PCR was performed by the CFX96 Touch real-time PCR detection system (Bio-Rad) using iTaq SYBR premix (Bio-Rad) and the follow- ing primers (5′-3′): HLA-A (Forward: AAAAGGAGGGAGTTACACTCAGG, Reverse: GCTGTGAGGGACACATCAGAG), HLA-B (Forward: CTACCCTGCGGAGATCA, Reverse: ACA GCCAGGCCAGCAACA), HLA-C (Forward: CACACCTCTCCTTTGTGACTTCAA, Reverse: CCACC TCCTCACATTATGCTAACA), NLRC5 (Forward: GTCATCCGCCTCTGGAATAAC, Reverse: CT GGTTGTCAAAGAAGGCAAAG), IRF1 (Forward: GAGGAGGTGAAAGACCAGAGCA, Reverse: TAGCATCTCGGCTGGACTTCGA), human GAPDH (Forward: CAACGGATTTGGTCGTATT, Reverse: GATGGCAACAATATCCACTT). The data analysis was performed by a standard comparative Ct method, and the results are shown as a fold change due to infection (2−ΔΔCt) (66). These values were calculated using the following equation: ΔΔCt = [(Ct value of gene of interest − Ct value of GAPDH)infected − (Ct value of gene of interest − Ct value of GAPDH)mock]. ACKNOWLEDGMENTS. We thank Melissa Linehan and Huiping Dong for tech- nical and logistical assistance. We thank Ralph Baric for kindly providing SARS- CoV-2 MA10. We thank Ya-Chi Ho for kindly providing NL4-3-dE-EGFP. We thank Craig Wilen for sharing his technical expertise. We thank Benjamin Israelow and Tianyang Mao for critical reading of the manuscript. We also give special recogni- tion to the services of Ben Fontes and the Yale EH&S Department for their ongoing assistance in safely conducting biosafety level 3 research. This work was in part supported by the Fast Grant from Emergent Ventures at the Mercatus Center and 1R01AI157488. A.I. is an Investigator of the HHMI. M.M. is supported by the Japan Society for Promotion of Science Overseas fellowship. Author affiliations: aDepartment of Immunobiology, Yale University School of Medicine, New Haven, CT 06520; bDepartment of Molecular Cellular and Developmental Biology, Yale University, New Haven CT 06520; and cHHMI, Chevy Chase, MD 20815 Author contributions: M.M., C.L., V.S.M., and A.I. designed research; M.M., C.L., and V.S.M. performed research; Y.S.-C.-2.G.S.I. contributed new reagents/analytic tools; M.M., C.L., V.S.M., and A.I. analyzed data; Y.S.-C.-2.G.S.I. isolated, purified, sequenced, and validated viral RNA genomes, and provided clinical viral isolates for our experiments; and M.M. and A.I. wrote the paper. 1. 2. 3. 4. 5. K. Tao et al., The biological and clinical significance of emerging SARS-CoV-2 variants. Nat. Rev. Genet. 22, 757–773 (2021). T. Kustin et al., Evidence for increased breakthrough rates of SARS-CoV-2 variants of concern in BNT162b2-mRNA-vaccinated individuals. Nat. Med. 27, 1379–1384 (2021). E. C. Sabino et al., Resurgence of COVID-19 in Manaus, Brazil, despite high seroprevalence. Lancet 397, 452–455 (2021). L. G. Thorne et al., Evolution of enhanced innate immune evasion by SARS-CoV-2. Nature 602, 487–495 (2022). K. Guo et al., Interferon resistance of emerging SARS-CoV-2 variants. Proc. Natl. Acad. Sci. U.S.A. 119, e2203760119 (2022). 6. W. F. Garcia-Beltran et al., Multiple SARS-CoV-2 variants escape neutralization by vaccine-induced 7. 8. 9. humoral immunity. Cell 184, 2523 (2021). P. Wang et al., Antibody resistance of SARS-CoV-2 variants B.1.351 and B.1.1.7. Nature 593, 130–135 (2021). C. Lucas et al., Impact of circulating SARS-CoV-2 variants on mRNA vaccine-induced immunity. Nature 600, 523–529 (2021). D. Planas et al., Considerable escape of SARS-CoV-2 Omicron to antibody neutralization. Nature 602, 671–675 (2022). 25. C. Ceraolo, F. M. Giorgi, Genomic variance of the 2019-nCoV coronavirus. J. Med. Virol. 92, 522–528 (2020). 26. T. G. Flower et al., Structure of SARS-CoV-2 ORF8, a rapidly evolving immune evasion protein. Proc. Natl. Acad. Sci. U.S.A. 118, e2021785118 (2021). 27. K. Matsuoka et al., SARS-CoV-2 accessory protein ORF8 is secreted extracellularly as a glycoprotein homodimer. J. Biol. Chem. 298, 101724 (2022). 28. B. Xia et al., Why is the SARS-CoV-2 Omicron variant milder? Innovation (Camb) 3, 100251 (2022). 29. J. Zhang et al., A systemic and molecular study of subcellular localization of SARS-CoV-2 proteins. Signal Transduct Target Ther. 5, 269 (2020). 30. O. Schwartz, V. Marechal, S. Le Gall, F. Lemonnier, J. M. Heard, Endocytosis of major histocompatibility complex class I molecules is induced by the HIV-1 Nef protein. Nat. Med. 2, 338–342 (1996). 31. P. Stumptner-Cuvelette et al., HIV-1 Nef impairs MHC class II antigen presentation and surface expression. Proc. Natl. Acad. Sci. U.S.A. 98, 12144–12149 (2001). 32. S. R. Leist et al., A mouse-adapted SARS-CoV-2 induces acute lung injury and mortality in standard laboratory mice. Cell 183, 1070–1085.e1012 (2020). 33. J. Y. Li et al., The ORF6, ORF8 and nucleocapsid proteins of SARS-CoV-2 inhibit type I interferon signaling pathway. Virus Res. 286, 198074 (2020). 10. A. Tarke et al., Impact of SARS-CoV-2 variants on the total CD4(+) and CD8(+) T cell reactivity in 34. X. Lei et al., Activation and evasion of type I interferon responses by SARS-CoV-2. Nat. Commun. 11, infected or vaccinated individuals. Cell Rep. Med. 2, 100355 (2021). 3810 (2020). 11. S. Cele et al., Omicron extensively but incompletely escapes Pfizer BNT162b2 neutralization. Nature 602, 654–656 (2022). 12. W. F. Garcia-Beltran et al., mRNA-based COVID-19 vaccine boosters induce neutralizing immunity 35. H. Geng et al., SARS-CoV-2 ORF8 forms intracellular aggregates and inhibits IFNgamma-induced antiviral gene expression in human lung epithelial cells. Front. Immunol. 12, 679482 (2021). 36. J. Kee et al., SARS-CoV-2 disrupts host epigenetic regulation via histone mimicry. Nature 610, against SARS-CoV-2 Omicron variant. Cell 185, 457–466.e454 (2022). 381–388 (2022). 13. C. H. GeurtsvanKessel et al., Divergent SARS-CoV-2 Omicron-reactive T and B cell responses in 37. N. Kriplani et al., Secreted SARS-CoV-2 ORF8 modulates the cytokine expression profile of human COVID-19 vaccine recipients. Sci. Immunol. 7, eabo2202 (2022). 14. T. H. Hansen, M. Bouvier, MHC class I antigen presentation: Learning from viral evasion strategies. Nat. Rev. Immunol. 9, 503–513 (2009). macrophages. bioRxiv [Preprint] (2021), https://doi.org/10.1101/2021.08.13.456266 (Accessed 17 March 2022). 38. X. Lin et al., ORF8 contributes to cytokine storm during SARS-CoV-2 infection by activating IL-17 15. Y. Zhang et al., The ORF8 protein of SARS-CoV-2 mediates immune evasion through down- pathway. iScience 24, 102293 (2021). regulating MHC-Iota. Proc. Natl. Acad. Sci. U.S.A. 118, e2024202118 (2021). 16. J. S. Yoo et al., SARS-CoV-2 inhibits induction of the MHC class I pathway by targeting the STAT1- 39. X. Wu et al., Viral mimicry of interleukin-17A by SARS-CoV-2 ORF8. mBio 13, e0040222 (2022). 40. M. Kohyama et al., SARS-CoV-2 ORF8 is a viral cytokine regulating immune responses. Int. Immunol. IRF1-NLRC5 axis. Nat. Commun. 12, 6602 (2021). 17. F. Zhang et al., Inhibition of major histocompatibility complex-I antigen presentation by sarbecovirus ORF7a proteins. Proc. Natl. Acad. Sci. U.S.A. 119, e2209042119 (2022). 18. N. Arshad et al., SARS-CoV-2 accessory proteins ORF7a and ORF3a use distinct mechanisms to downregulate MHC-I surface expression. bioRxiv [Preprint] (2022). https://doi. org/10.1101/2022.05.17.492198 (Accessed 24 May 2022). 19. S. Zheng et al., The SARS-CoV-2 accessory factor ORF7a downregulates MHC class I surface expression. bioRxiv [Preprint] (2022). https://doi.org/10.1101/2022.05.29.493850 (Accessed 5 July 2022). 20. L. Velazquez-Salinas et al., Positive selection of ORF1ab, ORF3a, and ORF8 genes drives the early evolutionary trends of SARS-CoV-2 during the 2020 COVID-19 pandemic. Front. Microbiol. 11, 550674 (2020). 35, 43–52 (2022), 10.1093/intimm/dxac044. 41. X. Wang et al., Accurate diagnosis of COVID-19 by a novel immunogenic secreted SARS-CoV-2 orf8 protein. mBio 11, e02431-20 (2020). 42. Y. Guan et al., Isolation and characterization of viruses related to the SARS coronavirus from animals in southern China. Science 302, 276–278 (2003). 43. C. S. M. E. Consortium, Molecular evolution of the SARS coronavirus during the course of the SARS epidemic in China. Science 303, 1666–1669 (2004). 44. B. Hu et al., Discovery of a rich gene pool of bat SARS-related coronaviruses provides new insights into the origin of SARS coronavirus. PLoS Pathog. 13, e1006698 (2017). 45. J. Cui, F. Li, Z. L. Shi, Origin and evolution of pathogenic coronaviruses. Nat. Rev. Microbiol. 17, 181–192 (2019). 21. Y. C. F. Su et al., Discovery and genomic characterization of a 382-nucleotide deletion in ORF7b and 46. A. Alkhansa, G. Lakkis, L. El Zein, Mutational analysis of SARS-CoV-2 ORF8 during six months of ORF8 during the early evolution of SARS-CoV-2. mBio 11, e01610-20 (2020). COVID-19 pandemic. Gene. Rep. 23, 101024 (2021). 22. B. E. Young et al., Effects of a major deletion in the SARS-CoV-2 genome on the severity of infection and the inflammatory response: an observational cohort study. Lancet 396, 603–611 (2020). 23. S. W. Fong et al., Robust virus-specific adaptive immunity in COVID-19 patients with SARS-CoV-2 47. J. R. Habel et al., Suboptimal SARS-CoV-2-specific CD8(+) T cell response associated with the prominent HLA-A02:01 phenotype. Proc. Natl. Acad. Sci. U.S.A. 117, 24384–24391 (2020). 48. B. Agerer et al., SARS-CoV-2 mutations in MHC-I-restricted epitopes evade CD8(+) T cell responses. Delta382 variant infection. J. Clin. Immunol. 42, 214–229 (2022). Sci. Immunol. 6, eabg6461 (2021). 24. M. L. M. Jongsma, G. Guarda, R. M. Spaapen, The regulatory network behind MHC class I expression. 49. C. Motozono et al., SARS-CoV-2 spike L452R variant evades cellular immunity and increases Mol. Immunol. 113, 16–21 (2019). infectivity. Cell Host Microbe 29, 1124–1136.e1111 (2021). PNAS 2023 Vol. 120 No. 16 e2221652120 https://doi.org/10.1073/pnas.2221652120 9 of 10 50. A. P. Ferretti et al., Unbiased screens show CD8(+) T cells of COVID-19 patients recognize shared epitopes in SARS-CoV-2 that largely reside outside the spike protein. Immunity 53, 1095–1107 e1093 (2020). 58. K. McMahan et al., Correlates of protection against SARS-CoV-2 in rhesus macaques. Nature 590, 630–634 (2021). 59. D. S. Khoury et al., Neutralizing antibody levels are highly predictive of immune protection from 51. A. Tarke et al., Comprehensive analysis of T cell immunodominance and immunoprevalence of symptomatic SARS-CoV-2 infection. Nat. Med. 27, 1205–1211 (2021). SARS-CoV-2 epitopes in COVID-19 cases. Cell Rep. Med. 2, 100204 (2021). 60. K. A. Earle et al., Evidence for antibody as a protective correlate for COVID-19 vaccines. Vaccine 39, 52. C.-H.G. Initiative, Mapping the human genetic architecture of COVID-19. Nature 600, 472–477 4423–4428 (2021). (2021). 61. E. M. Bange et al., CD8(+) T cells contribute to survival in patients with COVID-19 and hematologic 53. Q. Zhang, P. Bastard, C. H. G. Effort, A. Cobat, J. L. Casanova, Human genetic and cancer. Nat. Med. 27, 1280–1289 (2021). immunological determinants of critical COVID-19 pneumonia. Nature 603, 587–598 (2022). 62. A. Hachim et al., ORF8 and ORF3b antibodies are accurate serological markers of early and late SARS-CoV-2 infection. Nat. Immunol. 21, 1293–1301 (2020). 54. V. Joag et al., Cutting edge: Mouse SARS-CoV-2 Epitope reveals infection and vaccine-elicited CD8 63. T. Mao et al., A stem-loop RNA RIG-I agonist protects against acute and chronic SARS-CoV-2 infection T cell responses. J. Immunol. 206, 931–935 (2021). in mice. J. Exp. Med. 219, e20211818 (2022). 55. A. Grifoni et al., SARS-CoV-2 human T cell epitopes: Adaptive immune response against COVID-19. 64. E. Perez-Then et al., Neutralizing antibodies against the SARS-CoV-2 Delta and Omicron variants Cell Host Microbe 29, 1076–1092 (2021). following heterologous CoronaVac plus BNT162b2 booster vaccination. Nat. Med. 28, 481–485 (2022). 56. A. T. Tan et al., Early induction of functional SARS-CoV-2-specific T cells associates with rapid viral 65. D. K. Kim et al., A comprehensive, flexible collection of SARS-CoV-2 coding regions. G3 (Bethesda) clearance and mild disease in COVID-19 patients. Cell Rep. 34, 108728 (2021). 10, 3399–3402 (2020). 57. B. Israelow et al., Adaptive immune determinants of viral clearance and protection in mouse models 66. T. D. Schmittgen, K. J. Livak, Analyzing real-time PCR data by the comparative C(T) method. Nat. of SARS-CoV-2. Sci. Immunol. 6, eabl4509 (2021). Protoc. 3, 1101–1108 (2008). 10 of 10 https://doi.org/10.1073/pnas.2221652120 pnas.org
10.1073_pnas.2221175120	RESEARCH ARTICLE \| MEDICAL SCIENCES OPEN ACCESS Targeting SWI/SNF ATPases in H3.3K27M diffuse intrinsic pontine gliomas Mateus Motaa,b, Stefan R. Swehaa,b, Matt Puna,b,c,d, Siva Kumar Natarajana,b, Yujie Dinga,b, Chan Chunge Alexander R. Judkinsf, Susanta Samajdarg, Xuhong Caoh, Lanbo Xiaoh, Abhijit Paroliah, Arul M. Chinnaiyanh,i,j,k,1 , Debra Hawesf, Fusheng Yangf, , and Sriram Vennetia,b,c,h,i,1 Contributed by Arul M. Chinnaiyan; received December 13, 2022; accepted March 30, 2023; reviewed by Sameer Agnihotri and Stephen Mack Diffuse midline gliomas (DMGs) including diffuse intrinsic pontine gliomas (DIPGs) bearing lysine-to-methionine mutations in histone H3 at lysine 27 (H3K27M) are lethal childhood brain cancers. These tumors harbor a global reduction in the tran- scriptional repressive mark H3K27me3 accompanied by an increase in the tran- scriptional activation mark H3K27ac. We postulated that H3K27M mutations, in addition to altering H3K27 modifications, reprogram the master chromatin remode- ling switch/sucrose nonfermentable (SWI/SNF) complex. The SWI/SNF complex can exist in two main forms termed BAF and PBAF that play central roles in neurodevel- opment and cancer. Moreover, BAF antagonizes PRC2, the main enzyme catalyzing H3K27me3. We demonstrate that H3K27M gliomas show increased protein levels of the SWI/SNF complex ATPase subunits SMARCA4 and SMARCA2, and the PBAF component PBRM1. Additionally, knockdown of mutant H3K27M lowered SMARCA4 protein levels. The proteolysis targeting chimera (PROTAC) AU-15330 that simultaneously targets SMARCA4, SMARCA2, and PBRM1 for degradation exhibits cytotoxicity in H3.3K27M but not H3 wild-type cells. AU-15330 lowered chromatin accessibility measured by ATAC-Seq at nonpromoter regions and reduced global H3K27ac levels. Integrated analysis of gene expression, proteomics, and chro- matin accessibility in AU-15330-treated cells demonstrated reduction in the levels of FOXO1, a key member of the forkhead family of transcription factors. Moreover, genetic or pharmacologic targeting of FOXO1 resulted in cell death in H3K27M cells. Overall, our results suggest that H3K27M up-regulates SMARCA4 levels and combined targeting of SWI/SNF ATPases in H3.3K27M can serve as a potent ther- apeutic strategy for these deadly childhood brain tumors. pediatric brain cancer \| H3K27M mutation \| SWI/SNF complex H3K27M diffuse midline gliomas (DMGs), including diffuse intrinsic pontine gliomas (DIPGs), are lethal childhood brain tumors. The available treatment options, chemo- and radiotherapy, are ineffective, and over 90% of patients die within 1.5 y of diagnosis (1, 2). These tumors harbor missense mutations in histone 3-encoding genes that result in a lysine-to-methionine substitution at position 27 mainly in genes H3-3A (termed H3.3K27M) and to a lesser extent in H3C2 (termed H3.1K27M, collectively referred to as H3K27M) (3–5). H3K27M mutations result in global reduction of the repressive H3K27me3 mark, accompanied by aberrant H3K27ac-enriched enhancers and super- enhancers and deregulation of gene expression. Additionally, abnormal genomic distri- bution of other histone marks, including H3K4me3, H3K36me3, H3K36me2, and H2K119ub, has been reported (6–15). These observations have led to the proposal of inhibitors of factors that mediate histone posttranslational modifications for potential therapies, including inhibitors of H3K27-demethylases, HDACs, EZH2, BET proteins, LSD1, and BMI1 (14–21). However, it is not known whether H3K27M mutations impact chromatin by altering other epigenetic regulators in addition to histone modifications and whether this infor- mation can be leveraged for designing therapeutics. The SWI/SNF complex is a master epigenetic modulator that facilitates nucleosome incorporation and displacement by sliding or evicting histone octamers, resulting in differential chromatin accessibility to transcription factors (22, 23). The complex can exist in two forms called BAF and PBAF with several common, unique, and obligate members. BAF is critical to establish enhancers and superenhancers (24–26). On the contrary, PBAF localizes to active promoters and is defined by the expression of three components—PBRM1, ARID2, and BRD7. The SWI/ SNF complexes are ATP dependent with two main ATPase subunits—SMARCA4 (BRG1) and SMARCA2 (BRM). Several of these key subunits, including the ATPases, are mutated in various tumors, implicating a central role of the SWI/SNF complex in cancers in general (22, 23). Significance H3K27M mutant DMGs including DIPGs are deadly childhood brain cancers. More than 250 clinical trials over the past 50 y have failed to yield effective therapies. This underscores the importance of understanding the biology of these tumors to develop efficacious therapies. Because H3K27M mutations suppress PRC2 function, and PRC2 opposes the BAF form of the SWI/SNF complex, we targeted key members of the SWI/SNF complex as a potential therapy. We demonstrate that a PROTAC tool compound AU-15330 that simultaneously degrades the ATPases SMARCA4 and SMARCA2 and PBRM1 selectively kills H3.3K27M but not H3WT glioma cell lines. Our studies suggest that targeting key components of the SWI/SNF complex can provide a potent therapy for these lethal pediatric brain cancers. Author contributions: M.M., A.M.C., and S.V. designed research; M.M., S.R.S., M.P., S.K.N., Y.D., C.C., D.H., F.Y., A.R.J., X.C., L.X., and A.P. performed research; S.S. contributed new reagents/analytic tools; M.M., S.R.S., M.P., S.K.N., Y.D., C.C., D.H., F.Y., A.R.J., X.C., L.X., and A.P. analyzed data; and M.M., A.M.C., and S.V. wrote the paper. Reviewers: S.A., Children’s Hospital of Pittsburgh; and S.M., St Jude Children’s Research Hospital. Competing interest statement: S.S. is an employee of Aurigene Discovery Technologies. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221175120/-/DCSupplemental. Published April 24, 2023. PNAS 2023 Vol. 120 No. 18 e2221175120 https://doi.org/10.1073/pnas.2221175120 1 of 9 Global H3K27me3 reduction is mediated by H3K27M muta- tions that suppress the functions of the polycomb repressive com- plex 2 (PRC2) which contains the H3K27-specific methyltransferase EZH2 (6–8, 27, 28). PRC2 function can also be antagonized by BAF. This was first established in rhabdoid tumors bearing SMARCB1 loss-of-function mutations (29). Subsequent studies have shown that recruitment of BAF complexes to enhancers evicts PRC2 to lower H3K27me3 levels and enables genomic enrich- ment of H3K27ac (24–26, 29–32). Moreover, suppression of EZH2 is proposed as a therapeutic strategy for both SMARCB1 mutant rhabdoid tumors and H3K27M gliomas (20, 21, 29–31). Because H3K27M similarly antagonizes the function of PRC2, we hypothesized that the SWI/SNF complex plays an important role in H3K27M DMGs. We employed a proteolysis-targeting chimera (PROTAC) degrader compound, AU-15330, that specif- ically degrades SMARCA4, SMARCA2, and PBRM1. Enhancer-addicted prostate cancers are potently suppressed by AU-15330 (33). Since H3K27M tumors are also enhancer addicted (9, 19, 34, 35), we hypothesized that AU-15330 would show selective efficacy in H3K27M as compared to H3WT tumor cells. We addressed this hypothesis using an integrated approach encompassing proteomics and next-generation sequencing. Results SMARCA4 Protein Levels Are Higher in H3.3K27M Compared to H3WT Cells. We began by assessing the levels of various SWI/ SNF-associated proteins including the ATPases SMARCA4 and SMARCA2; PBAF-associated members PBRM1, BRD7, and ARID2; and SMARCB1 in a panel of H3WT (SF188, SJGBM2, and UMPED37), H3.3G34V (KNS42), and H3.3K27M (DIPG007, DIPGXIIIP, SF7761, and BT425) low-passage, patient-derived cell lines. Overall, we noted higher expression of SMARCA4, SMARCA2, and PBRM1 in H3.3K27M versus H3WT and H3.3G34V cell lines (Fig. 1A). We next knocked down (KD) H3-3A to reduce mutant H3.3K27M levels. This was accompanied by an increase in H3K27me3 and lowered H3K27ac levels (Fig. 1B). We determined changes in various SWI/SNF Fig. 1. SMARCA4 protein levels are higher in H3.3K27M compared to H3WT cells. (A) Immunoblots of patient-derived H3 WT (SF188, SJGBM2, and UMPED37), H3.3G34V (KNS42), or H3.3K27M (DIPG007, DIPGXIIIp, and SF7761) cell lines probed for SMARCA4, SMARCA2, PBRM1, ARID2, BRD7, and SMARCB1; western blot for HSP90 and Ponceau staining were used as loading control. (B) Immunoblots of DIPG007 cells with or without stable knockdown (KD) of H3-3A with two independent shRNAs (sh1 or sh2) probed for H3.3K27M, H3K27me3, and H3K27ac; total H3 was probed as loading control. (C) Heatmap of normalized protein abundance (Z-score) of key components of BAF and PBAF (PBRM1, ARID2, and BRD7) SWI/SNF complex in DIPG007 H3.3K27M with or without H3-3A KD acquired by untargeted proteomics (n = 3, each). (D) Immunoblots of DIPG007 cells with or without H3-3A KDs from B probed for SMARCA4, SMARCA2, PBRM1, ARID2, BRD7, and SMARCB1. HSP90 was probed as loading control. (E) Representative images of H3K27M and H3WT (2 cases each) tumors stained for SMARCA4 by immunohistochemistry. (Scale bar, 50 μM.) (F) Matlab-based quantification of SMARCA4 (Y axis, a.u.=arbitrary units determined by number of pixels × pixel intensity; three randomly selected regions/case) in H3WT (n = 5) and H3K27M (n = 6) DMGs. Data were analyzed by two-sided, two-tailed, nonpaired t test. 2 of 9 https://doi.org/10.1073/pnas.2221175120 pnas.org complex members by proteomics on H3.3K27M KD in DIPG007 cells. SMARCA4, ARID2, BRD7, SMARCC2, and SMARCD3 protein levels were significantly reduced (Fig. 1C). On the contrary, PBRM1, SMARCB1, SMARCA5, SMARCC1, SMARCD3, and SMARCD1 levels were significantly increased (Fig. 1C). We validated key proteomic findings by western blotting in DIPG007 cells with both shRNAs. We did not observe similar consistent changes in ARID2, BRD7, and SMARCB1 (Fig. 1D); however, reduction in SMARCA4 and increased PBRM1 were confirmed (Fig. 1D). We next performed immunohistochemistry (IHC) for SMARCA4 in DMG tumor samples (Fig. 1 E and F). IHC confirmed increased SMARCA4 levels in H3K27M compared to H3WT DMGs as observed in patient-derived cell lines (Fig. 1 E and F). Expression (mRNA) levels of Smarca4, Smarca2, and Pbrm1 were not altered in isogenic mouse neuronal stem cells bearing H3.3K27M versus H3.3WT from our previous studies (SI Appendix, Fig. S1A) (36). This suggests that H3K27M may not alter expression but could impact stabilization of the SWI/ SNF complex as suggested by others (37, 38). H3.3K27M Cell Lines Are Sensitive to SMARCA2, SMARCA4, and PBRM1 Protein Degradation Induced by Treatment with the PROTAC Degrader AU-15330. Since SMARCA4, SMARCA2, and PBRM1 levels were elevated in H3.3K27M versus H3WT and H3.3G34V cell lines, we sought to determine the effects of a PROTAC tool compound (AU-15330) designed to simultaneously target all the three components. DIPG007 cells treated with AU-15330 showed reduced levels of SMARCA4 and PBRM1 by proteomics (Fig. 2A) and immunoblotting (Fig. 2B and SI Appendix, Fig. S1B). Additionally, proteomics showed a significant increase in SMARCB1, SMARCA5, SMARCC1, SMARCC2, SMARCD1, and SMARCE1 (Fig. 2A). We tested the effects of AU-15330 in our panel of H3.3K27M, H3WT, and H3G34V cells lines. All cell lines showed lowered SMARCA4, SMARCA2, and PBRM1 levels (Figs. 2 B and C and SI Appendix, Fig. S1B). However, H3.3K27M cells showed far greater sensitivity to AU-15330 as compared to H3WT and H3G34V cell lines (Fig. 2D and SI Appendix, Fig. S1 C and D). Moreover, knockdown of mutant H3.3K27M using our two independent shRNAs rendered H3.3K27M DIPG007 cells insensitive to AU-15330 (Fig. 2E and SI Appendix, Fig. S1E). Importantly, AU-15330 IC50 values were 15-fold lower in H3.3K27M as compared to H3WT, or H3.3G34V (KNS42), and DIPG007 with H3.3K27M KD with both shRNAs (Fig. 2F). These data demonstrate that H3.3K27M cells are far more sensitive to AU- 15330 as compared to H3WT and H3.3G34V mutant cell lines. AU-15330 Reduces Chromatin Accessibility at Nonpromoter Regions in H3.3K27M Mutant Cells. We next elucidated genome-wide changes in chromatin accessibility with AU- 15330 treatment. DIPG007 cells were treated with AU- 15330 (1 µM) or vehicle for 24 h and profiled for genome- wide changes by ATAC-Seq. We noted marked lowering of chromatin accessibility at gene bodies, introns, and distal intergenic regions (SI Appendix, Fig. S2 A–C). These overlapped mainly to nonpromoter genomic areas, while promoter regions were largely unaltered (Fig. 3 A and B and Dataset S1). This corresponded to a reduction in global H3K27ac levels in all the three H3.3K27M cell lines (Fig. 3C). In contrast, overall levels of H3K4me1, H3K4me3, and H3K27me3 remained unaltered even with high concentrations of AU-15330 (10 µM for 48 h, SI Appendix, Fig. S2D). Moreover, H3.3K27M levels were unchanged with AU-15330 treatment (SI Appendix, Fig. S2D). To determine the functional significance of changes in chromatin accessibility, we performed RNA-Seq (Dataset S2) and proteomics (Dataset S3) in parallel in H3.3K27M cells with or without treatment with AU-15330 (1 µM for 24 h) (Fig. 3D). We overlapped down-regulated genes from RNA-Seq and proteomics that also showed lowered chromatin accessibility to identify 161 commonly down-regulated genes (Fig. 3E and Dataset S4). Gene set enrichment analysis (GSEA) of these 161 genes showed downregulation of pathways related to cell adhesion, cell motility, cell morphogenesis, and neurogenesis (Fig. 3F). Chromatin Accessibility, Gene Expression, and Protein Abundance of FOXO1 Are Decreased by AU-15330. We performed an integrated analysis on AU-15330 treatment using motif analyses on genes with lowered chromatin accessibility along with the 161 function- ally down-regulated genes (Fig. 3E) and cross referenced these data with known human transcription factors. These analyses identified FOXO1 (Forkhead Box O1), a key forkhead master transcription- al factor down-regulated by AU-15330 (Fig. 4A and Dataset S5). Reduced chromatin accessibility was noted at two peaks corre- sponding to H3K27ac-marked enhancers at the FOXO1 locus in DIPG007 cells (Fig. 4 B–D). In contrast, chromatin accessibility at the FOXO1 promoter was not altered with AU-15330 treatment (Fig. 4 B–D). Importantly, Hi-C data showed a topologically associ- ated domain formed between the FOXO1 promoter region, the two downstream enhancer peaks, and downstream intergenic regions comprised of LINC00598 and LINC00332 (SI Appendix, Fig. S3A and Fig. 4B). Lowered chromatin accessibility at the FOXO1 locus was accompanied by reduction of both FOXO1 mRNA and protein levels (Figs.3D and 4E) and downregulation of canonical FOXO1 targets by RNA-Seq (SI Appendix, Fig. S3B). Of the FOXO1 tar- gets, we identified downregulation of Ras homolog (RHO) family GTPases (SI Appendix, Fig. S3 C and D). Among the RHO family GTPases, RHOB was one of the 161 genes that were down-regulat- ed by RNA-Seq, proteomics, and ATAC-Seq in AU-15330 versus vehicle-treated DIPG007 cells (Fig. 3 D–F). Therefore, we focused our efforts on assessing the expression of FOXO1 and RHOB in our panel of cells. Both FOXO1 and RHOB protein levels were expressed at higher levels in H3.3K27M versus H3WT or H3G34V cells (Fig. 4F). Importantly, both FOXO1 and RHOB protein levels were reduced upon AU-15330 treatment in all the three H3.3K27M cell lines (Fig. 4G). In contrast, AU-15330 ver- sus vehicle treatment did not show consistent FOXO1 or RHOB alterations in H3WT and H3G34V cells (SI Appendix, Fig. S3E). Finally, both genetic and pharmacologic (FOXO1 inhibitor AS1842856) suppression of FOXO1 resulted in cell death in H3.3K27M cells (Fig. 4H and SI Appendix, Fig. S3F). AS1842856 binds to active FOXO1, but not the Ser256-phosphorylated form (39–41). Accordingly, AS1842856 did not show a consistent pat- tern on p-ser256-FOXO1 levels in H3.3K27M cells (SI Appendix, Fig. S3G). Overall, these results suggest that AU-15330 toxicity is, in part, mediated by downregulation of FOXO1. Discussion The discovery of H3K27M mutations in DIPGs and DMGs has brought chromatin biology to the forefront to define epigenetic treatments for these lethal tumors. Due to inhibition of the function of the PRC2 complex, H3K27M mutations result in global reduc- tion of H3K27me3. This is accompanied by abnormal genomic distribution of other histone posttranslational marks, including H3K27ac, H3K4me3 (including bivalent H3K4me3/H3K27me3), H3K36me2, H3K36me3, and H2AK119ub. To further define aberrant chromatin in H3K27M tumors, we interrogated key PNAS 2023 Vol. 120 No. 18 e2221175120 https://doi.org/10.1073/pnas.2221175120 3 of 9 Fig. 2. H3.3K27M cell lines are sensitive to SMARCA2, SMARCA4, and PBRM1 protein degradation using the PROTAC AU-15330. (A) Heatmap of normalized protein abundance (Z-score) of key SWI/SNF complex components in untreated DIPG007 cells (n = 3) or DIPG007 cells treated with AU-15330 PROTAC (n = 4, 1 µM for 24 h) or vehicle (Veh, DMSO, n = 3), acquired by untargeted proteomics. Data were analyzed by ANOVA; P values indicated are between Veh and AU- 15330-treated cells. (B) Immunoblots of H3.3K27M mutant cell lines (SF7761, BT245, and DIPG007) treated with AU-15330 (1 µM for 24 h) or Veh and probed for SMARCA4, SMARCA2, PBRM1, and SMARCB1. HSP90 was probed as loading control. (C) Immunoblots of H3.3G34V mutant (KNS42) and H3WT (SJGBM2 and SF188) cell lines treated with AU-15330 (1 µM for 24 h) or Veh and probed for SMARCA4, SMARCA2, PBRM1, and SMARCB1. HSP90 was probed as loading control. (D) Cell viability (normalized to Veh, percentage, Y axis) of H3WT (SJGBM2 and SF188), H3.3G34V mutant (KNS42), and H3.3K27M mutant (SF7761, BT245, and DIPG007) cell lines on treatment with different concentrations of AU-15330 (log concentrations, X axis) for 5 d (n = 3 for each concentration/cell line). (E) Cell viability (normalized to Veh, percentage, Y axis) of DIPG007 cells with or without for H3-3A KD with two independent shRNAs (H3-3A sh1 or sh2) treated with different concentrations of AU-15330 (log concentrations, X axis) for 5 d (n = 3 for each concentration/cell line). (F) Half maximal inhibitory concentration (IC50) of AU-15330 (μM=Molar, Y axis) for H3.3K27M mutant (SF7761, BT245, and DIPG007), H3-3A KD sh1 and sh2 DIPG007, H3.3G34V mutant (KNS42), and H3WT (SJGBM2 and SF188) cells based on cell viability curves in (D) and (E). members of the SWI/SNF complex, including the ATPases SMARCA4 and SMARCA2, and key components of PBAF, includ- ing PBRM1, BRD7, and ARID2. All the tested components were higher in H3.3K27M versus H3WT and H3.3G4V cell lines. Upon knockdown of H3.3K27M, we noted a reduction in SMARCA4 accompanied by an increase in PBRM1 levels. In support of these data, previous studies using DNA-barcoded H3.3K27M containing nucleosomes suggest mutant histone inter- actions with both BAF and PBAF (37). In this study, we utilized a PROTAC compound, AU-15330, that simultaneously degrades SMARCA4, SMARCA2, and PBRM1 and is toxic in enhancer-addicted prostate cancers (33). 4 of 9 https://doi.org/10.1073/pnas.2221175120 pnas.org Fig. 3. AU-15330 reduces chromatin accessibility at nonpromoter regions in H3.3K27M mutant cells. (A) Heat maps of peak intensity (plotted as ± 2.0 kb from peak center) and overall peak representation of nonpromoter regions and promoters (plotted as ± 2.0 kb from center, transcriptional start site). (B) Overall peak representation of nonpromoter and promoter regions from 3A. (C) Immunoblots of H3.3K27M mutant cell lines (SF7761, DIPG007, and BT245) treated with AU-15330 (1 µM for 24 h) or Veh and probed for H3K27Ac. Total H3 was probed as loading control. (D) Heatmap of differentially expressed genes (Left, RNA-Seq) and proteins (Right, untargeted proteomics) in DIPG007 cells treated with AU-15330 (1 µM for 24 h) or Veh. Red = up-regulated genes, yellow = up-regulated proteins, and blue = down-regulated genes or proteins (RNA-Seq n = 2/condition; proteomics: n = 3 for Veh and n = 4 for AU-15330). (E) Venn diagram depicting overlap of genes with lowered chromatin accessibility (ATAC-Seq), lowered expression by RNA-Seq, and proteomics in DIPG007 cells treated with AU-15330 versus Veh. (F) Pathway analysis of 161 genes from Venn diagram. Red bars indicate pathways related to development. H3K27M tumors show elevated H3K27ac levels and are also enhancer addicted (9, 19, 34, 35). Accordingly, we noted selective toxicity in H3.3K27M versus H3WT and H3.3G34V cell lines accompanied by a reduction in global H3K27ac levels. AU-15330 lowered chromatin accessibility of genes at nonpromoter regions, and functional RNA-Seq and proteomic analyses showed down- regulation of developmentally related pathways, cell adhesion, and cell locomotion. Similar results upon targeting SMARCA4 in PNAS 2023 Vol. 120 No. 18 e2221175120 https://doi.org/10.1073/pnas.2221175120 5 of 9 Fig. 4. Chromatin accessibility, gene expression, and protein abundance of FOXO1 are lowered by AU-15330 PROTAC. (A) Venn diagram depicting overlap of 161 down-regulated genes from Fig. 3F, with motif analysis of genes with lowered chromatin accessibility by ATAC-Seq, and all known human transcription factor genes. FOXO1 motif is indicated. (B) Hi-C heatmap (Left) depicting a topologically associating domain (TAD) between the FOXO1 promoter and ATAC-Seq peaks 1 and 2 associated with the FOXO1 locus in DIPG007 cells treated with AU-15330 (1 µM for 24 h, tan, n = 2) or Veh (blue, n = 2). H3K27ac (pale blue) track from DIPG007 shows localizations of peaks 1 and 2 with H3K27ac-enriched FOXO1 enhancer sites (pale blue boxes). Magnified FOXO1 locus (rectangle, right) is illustrated. (C and D) Representation (C) and quantification (Y axis, a.u. = arbitrary units normalized to Veh) (D) of ATAC-Seq peak tracks/chromatin accessibility at the FOXO1 promoter and at peak 1 and 2 associated with the FOXO1 locus from (B). (E) FOXO1 mRNA and protein levels (a.u. = arbitrary units normalized to Veh) of DIPG007 cells treated with or without AU-15330 by RNA-Seq (n = 2/condition) and proteomics (n = 3 for Veh; n = 4 for AU-15330). (F) Immunoblots of H3WT (SF188 and SJGBM2), H3.3G34V mutant (KNS42), and H3.3K27M mutant cell lines (SF7761, DIPGXIIIP, BT245, and DIPG007) probed for FOXO1, RHOB, and RHOA/B/C. HSP90 was probed as loading control. (G) Immunoblots of H3.3K27M mutant cell lines (SF7761, BT245, and DIPG007) treated with AU-15330 (1 µM for 24 h) or Veh probed for FOXO1, RHOB, and RHOA/B/C. HSP90 was probed as loading control. Arrows indicate FOXO1 and RHOB. (H) Cell viability assessment (normalized to Veh, percentage, Y axis) of H3.3K27M mutant cell lines DIPG007 with (n=16) or without (n = 8) FOXO1 knock down and DIPG007 (n = 4) and DIPGXIIIp (n = 8) cells treated with the FOXO1 inhibitor AS-1842856 (2, 6, or 10 µM for 48 h) or Veh. H3K27M tumors were also observed by two independent groups while our manuscript was in preparation (42, 43). While our experiments were focused on H3.3K27M tumor cells, both these studies show similar effects on H3.1K27M cells, suggesting that targeting the SWI/SNF complex may be a key therapeutic strategy for H3.3/H3.1K27M DIPGs and DMGs. Our approach of integrating motif analysis with down-regulated transcription factors on treatment with AU-15330 identified the master transcriptional regulator FOXO1. Lowered chromatin accessibility was noted at regions corresponding to FOXO1 enhancers in a topologically associating domain formed at the FOXO1 pro- moter. FOXO1 is important for brain development and deregulated in several cancers, including gliomas (37, 44–48). Alveolar rhabdo- myosarcomas are driven by PAX3–FOXO1 and PAX7–FOXO1 fusion proteins in ~80% of cases (49, 50). Intriguingly, PAX3/7– FOXO1 fusion proteins epigenetically reprogram SMARCA4, result- ing in aberrant enhancer activation that creates a therapeutic dependency with SMARCA4 inhibitors (51, 52). FOXO1 levels were 6 of 9 https://doi.org/10.1073/pnas.2221175120 pnas.org decreased by AU-15330, and corresponding downregulation of FOXO1 canonical pathways was noted, including lowering of RHOB levels in H3K27M but not H3WT cells. Rho GTPases, including RHOB, can regulate several pathways related to signal transduction, cell adhesion, tumor cell invasion, and cell division (53), warranting further future examination of both FOXO1 and RHOB in H3K27M DMGs. However, AU-15330 down-regulated multiple pathways related to cell adhesion, cytoskeletal organi- zation, and morphogenesis, suggesting that toxicity arises from multiple mechanisms that are both FOXO1 dependent and independent. In summary, we note that H3.3K27M gliomas show alterations in key components of both the BAF and PBAF SWI/SNF complex. Therapeutically, a PROTAC compound simultaneously targeting SMARCA4, SMARCA2, and PBRM1 showed selective toxicity in H3.3K27M in our panel of tested cell lines. Our results suggest that targeting central epigenetic regulators including SMARCA4 may be a key therapeutic target for H3K27M DIPGs and DMGs, warranting further biologic and preclinical in vivo studies. Materials and Methods Patient-Derived Cell Lines. Cell lines were cultured in a 5% CO2-humidified incubator at 37 °C. BT245 (obtained from Nada Jabado, University of Montreal), DIPG-007 (HSJD-DIPG007, obtained from Rintaro Hashizume, Northwestern University; RRID: CVCL_VU70), and DIPG-XIII*p (obtained from Michelle Monje, Stanford University; RRID: CVCL_IT41) H3.3K27M cells were cultured in Neurobasal-A (Thermo Fisher Scientific #10888022) and DMEM/F-12 (Thermo Fisher Scientific #11330032) media supplemented with HEPES (1 M) (Thermo Fisher Scientific #15630080), sodium pyruvate (100 mM) (Thermo Fisher Scientific #11360070), MEM nonessential amino acid solution (100X) (Thermo Fisher Scientific #11140050), GlutaMAX™ supplement (Thermo Fisher Scientific #35050061), B-27™ supplement (50X), minus vitamin A (Thermo Fisher Scientific #12587010), 20 ng/μL human EGF (Shenandoah #100-26, Warminster, PA), 20 ng/μL human FGF-BASIC-154aa (Shenandoah #100- 146), 10 ng/μL human PDGF-AA (Shenandoah #100-16), 10 ng/μL human PDGF-BB (Shenandoah #100-18), 2 µg/mL 0.2% heparin solution (StemCell Technologies, Inc. #07980, Cambridge, MA), and antibiotic–antimycotic (100X) (Thermo Fisher Scientific #15240096). SF7761 H3.3K27M cells (obtained from Rintaro Hashizume, Northwestern University; RRID: CVCL_IT45) were cultured in Neurobasal-A (Thermo Fisher Scientific #10888022) media supplemented with serum free N-2 supplement (100X) (Thermo Fisher Scientific #17502048), B-27™ supplement (50X) (Thermo Fisher Scientific #17504044), L-glutamine (200 mM) (Thermo Fisher Scientific #A2916801), 7.5% albumin, bovine frac- tion V (BSA) (dot scientific, inc #DSA30075, Burton, MI), 20 ng/μL human EGF (Shenandoah #100-26), 20 ng/μL human FGF-BASIC-154aa (Shenandoah #100-146), 2 µg/mL heparin solution, 0.2% (StemCell Technologies, Inc. #07980), and penicillin–streptomycin (10,000 U/mL) (Thermo Fisher Scientific #115140122). UMPed37 H3 WT (obtained from Carl Koschmann, University of Michigan) and KNS42 H3.3G34V (obtained from Carl Koschmann, University of Michigan; RRID: CVCL_0378) cells were cultured in DMEM/F-12 (Thermo Fisher Scientific #11320033) supplemented with GlutaMAX™ supplement (Thermo Fisher Scientific #35050-061), 10% fetal bovine serum (VWR #89510- 186, Radnor PA), Normocin™ (InvivoGen #ant-nr-1), and antibiotic–antimycotic (100X) (Thermo Fisher Scientific #15240-096). SF188 (obtained from Craig B. Thompson, Memorial Sloan Kettering Cancer Center; RRID: CVCL_6948) and HEK293T (obtained from Andrew Lieberman, University of Michigan) H3 WT cells were cultured in DMEM, high glucose, no glutamine (Thermo Fisher Scientific #111960-044) supplemented with 10% fetal bovine serum (VWR #89510-186), L-glutamine (200 mM) (Thermo Fisher Scientific #A2916801), and penicillin– streptomycin (10,000 U/mL) (Thermo Fisher Scientific #115140122). SJGBM2 H3 WT cells (obtained from Carl Koschmann, University of Michigan; RRID: CVCL_M141) were cultured in IMDM (Thermo Fisher Scientific #112440-053) supplemented with 20% fetal bovine serum (VWR #89510-186), Normocin™ (InvivoGen #ant-nr-1), and antibiotic–antimycotic (100X) (Thermo Fisher Scientific #15240-096). Chemical Compounds. AU-15330 proteolysis-targeting chimera (PROTAC) compound was provided by Aurigene Biosciences. Cells were treated with 1 µM AU-15330 for 24 h, unless otherwise noted. AS-1842856 (#16761) was purchased from Cayman Chemical, and cells were treated with 2 µM, 5 µM, or 10 µM AS-1842856 for 48 h, unless otherwise noted. Dimethyl sulfoxide (DMSO) (Sigma #D2650) was used as vehicle for both compounds. Gene Silencing. H3-3A. H3-3A gene was knocked down in DIPG007 cells by using shH3-3A-con- taining lentiviral particles generated according to the LentiStarter 3.0 kit (SBI #LV060-1, Palo Alto, CA). Briefly, HEK293T cells were transfected with 700 ng of two independent human shH3-3A plasmids purchased from Sigma (clone IDs: TRCN0000139066 = sh1 or TRCN0000139180 = sh2). HEK293T-transfected media containing lentiviral particles were collected 72 h post transfection and fil- tered with a 0.45 µm-syringe filter. Next, the filtered lentiviral particles were added into media of previously plated DIPG007 cells, and 48 h post transduction, culture media was replaced by fresh growth media. After an additional 48 h, media was sup- plemented with 1 μg/mL puromycin for selection of successfully transduced cells. FOXO1. FOXO1 gene was knocked down in DIPG007 cells by using shFOXO1-con- taining lentiviral particles generated according to the TransIT® Lentivirus System (Mirus #MIR 6655). Briefly, HEK293T cells were transfected with human shFOXO1 plasmids purchased from Sigma (clone IDs: TRCN0000010333 or TRCN0000039580). HEK293T-transfected media containing lentiviral particles were collected 48 h post transfection and filtered with a 0.45 µm-syringe filter. Next, the filtered lentiviral particles were added into media of previously plated DIPG007 cells, and 48 h post transduction, culture media was replaced by fresh growth media. After an additional 48 h, media was supplemented with 1 μg/mL puromycin for selection of successfully transduced cells. Cell Lysate Extraction. Histone extraction. Briefly, cells were washed with phosphate-buffered saline (PBS) (Gibco #10010-023) and centrifuged. Cell pellets were resuspended with a hypotonic lysis buffer (10 mM Tris HCl pH8.0, 1 mM KCl, and 1.5 mM MgCl2) sup- plemented with protease (Sigma #P8340) and phosphatase (APExBIO #K1012) inhibitor cocktails and incubated on a rotator at 4 °C for 30 min for nucleus isolation. The isolated nuclei were pelleted under centrifugation, resuspended with sulfuric acid (0.4 N H2SO4), and incubated on a rotator at 4 °C overnight. Following centrifugation, the supernatant was transferred to a new tube, and trichloroacetic acid (Sigma #T0699) was added followed by a 30 min incubation on ice. Next, the histone mix was centrifuged, supernatant discarded, and purified histones in the tube were washed twice with ice-cold acetone under centrifuga- tion (5 min/wash). After the last acetone wash, histones were air-dried at room temperature for 20 min and resuspended with double-distilled water. Whole-cell protein lysate. Cells were washed with PBS and centrifuged. Cell pel- lets were resuspended with RIPA lysis buffer (Thermo Fisher Scientific #8990) sup- plemented with protease (Sigma #P8340) and phosphatase (APExBIO #K1012) inhibitor cocktails and incubated on a rotator at 4 °C for 1 h. Next, the mix was centrifuged and the supernatant containing whole-cell protein was transferred to a new tube. Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific #23225) was used to quantitate both whole-cell protein lysate and extracted histones. Immunoblotting. Whole-cell protein and histone lysates were resolved in 4 to 15% Mini-PROTEAN® TGX™ Precast Protein Gels (BIO-RAD #4561084) and transferred to a PVDF membrane by using the Trans-Blot Turbo RTA Mini 0.2 µm Transfer Kit (BIO-RAD #1704272). After membrane blocking with 5% milk in tris-buffered saline (TBS) supplemented with 0.1% Tween-20 (TBST) at room temperature for 1 h on a rocker, blots were incubated at 4 °C overnight with the following primary antibodies: (Whole-cell protein lysate) Brg1/SMARCA4 (Cell Signaling #52251), SMARCA2 (Bethyl #A301-015A), PBRM1 (Cell Signaling #91894), ARID2 (Novus Biologicals #NBP1-26615), BRD7 (Thermo Fisher Scientific #PA5-49379), SMARCB1 (Cell Signaling #91735), FOXO1 (Cell Signaling #2880), Phospho-FOXO1 (Ser256) (Cell Signaling #9461), HSP90 (Cell Signaling #4877), and β-actin (Sigma #A5441); (histone) H3K27Ac (Millipore #07-360), H3K4me1 (Cell Signaling #5326), H3K4me3 (Cell Signaling #9751), H3K27me3 (Cell Signaling #9733), H3.3K27M (Sigma #ABE419), and histone H3 (Cell Signaling #3638). Next, immunoblots were washed with TBST and incubated with either goat anti-rabbit IgG (H + L)-HRP conjugate (BIO-RAD #1706515) or goat anti- mouse IgG (H + L)-HRP conjugate (BIO-RAD #1706516) secondary antibodies PNAS 2023 Vol. 120 No. 18 e2221175120 https://doi.org/10.1073/pnas.2221175120 7 of 9 at room temperature for 1 h. Then, after membranes were washed with TBST, immunoblot signals were detected by using either Pierce™ ECL Western Blotting Substrate (Thermo Fisher Scientific #32106) or SuperSignal™ West Pico PLUS Chemiluminescent Substrate (Thermo Fisher Scientific #34578) reagents on an autoradiography film (dotScientific #BDB57-Lite). Cell Viability. Cell viability was assessed using the CellTiter-Glo® 2.0 Cell Viability Assay (Promega #G9242) according to manufacturer’s instructions. Briefly, 5,000 cells were seeded (in triplicate/condition) in a solid bottom white- walled 96-well plate and incubated with different concentrations of the drug for 5 d. Luminescence signal was detected using a BioTek Synergy HTX Multi-Mode microplate reader (Agilent company), and relative light unit (RLU) was analyzed with GraphPad Prism 8.4.3 (GraphPad software). Percentage of cell viability was calculated by dividing the RLU of drug-treated sample by the RLU of the control (drug vehicle) and multiplying by 100. Another method of cell viability analysis was by mixing cells with Trypan Blue Solution, 0.4% (Thermo Fisher Scientific #15250061), in a 1:1 dilution and inserting into a cell-counting chamber slide (Thermo Fisher Scientific #C10312) to read in a Countess™ 3 FL Automated Cell Counter (Thermo Fisher Scientific #AMQAF2000). Immunohistochemistry. Immunohistochemistry was performed on H3K27M and H3WT DMG samples as previously reported (54, 55). The Discovery XT proces- sor (Ventana Medical Systems) or Leica Bond automated staining processor (Leica Biosystems) was used to perform immunohistochemical staining. Each section from H3K27M or H3WT tumor sample was blocked using 10% normal goat serum along with 2% BSA in PBS for a duration of 30 min. After this, each section was treated with a rabbit monoclonal anti-SMARCA4 antibody (1:200, Abcam, ab110641) for a duration of 5 h. This was followed by treating individual tissue sections with biotinylated goat anti-rabbit IgG (PK6101, Vector Labs) at a dilution of 1:200 for a duration of 60 min. Chromogens were detected using the DAB detection kit along with Streptavidin-HRP and Blocker D to reduce background (Ventana Medical Systems). These treatments were performed according to instructions provided by the manufacturer. Subsequently, each section was mounted, dried, and visualized using the scanning system from Aperio Vista (AperioScanscope Scanner). Each slide was visualized through the accompanying AperioImageScope software program. We quantified slides in a blinded manner. An experimenter blinded to the study design visualized each slide at 40× magnification. They then captured images in JPEG format. JPEG images were captured from three randomly selected areas of each H3WT- and H3K27M-stained section. We used our previously published, auto- mated analysis program to quantify each image. This program is MATLAB based and uses the image processing toolbox (54). K-means clustering, color segmentation based on RGB color differentiation, and Otsu’s threshold-based background–fore- ground separation are taken into consideration to arrive at a quantitative score that multiplies extracted pixels with the average intensity for each JPEG image. TMT Mass Spectrometry. Whole-cell protein lysates were collected from DIPG007 cells from the following conditions: (a) untreated (n=3); (b) stably knocked down for H3-3A/H3F3A (H3-3AKD sh1) (n=3); (c) treated with AU-15330 PROTAC (1 μM for 24 h) (n=4); and (d) vehicle (Veh = DMSO for 24 h) (n=3). Samples were then sent to the Proteomics Resource Facility at the University of Michigan for analysis. Briefly, 100 μg of the lysate/sample/replicate was prepared and labeled following the directions of the TMTpro™ 16plex Label Reagent Set (Thermo Fisher Scientific #A44521). Twelve fractions were used for LC-MS/MS analysis. Generated data were aligned to a human protein database (20291 entries; reviewed; downloaded on 12/13/2021) and sorted for high (≤1%) or medium-low (≤5%) FDR confidence. A heatmap of protein abundance was gen- erated using GraphPad Prism 8.4.3 (GraphPad software). RNA Sequencing. DIPG007 cells were treated with 1 µM AU-15330 or DMSO for 24 h, and total RNA was extracted using Trizol (Thermo Fisher #15596026) followed by DNAse (Sigma #DN25) treatment. Preparation of complementary DNA (cDNA) library and sequencing were performed according to the Illumina TruSeq protocol and HiSeq 2000 platform, respectively. The resultant RNA-Seq data were aligned to human reference genome by using Bowtie software and analyzed with the RNA-Seq by Expectation-Maximization (RSEM) software tool. Sorting of differentially expressed genes was performed by using the empirical Bayes hierarchical models (EBSeq), and significantly up-regulated and down-regulated pathways were determined by the Molecular Signatures Database (MSigDB) in GSEA software (https://www.gsea-msigdb.org/gsea/msigdb/). RNA-Seq datasets have been deposited at the Gene Expression Omnibus (GEO) and an accession number is pending. ATAC Sequencing. ATAC sequencing was performed by the epigenetic services of Active Motif company (Active Motif), and samples were prepared according to the company’s protocol. Briefly, following treatment with 1 µM AU-15330 or DMSO for 24 h, DIPG007 cells were centrifuged, resuspended, and incubated in growth media with DNase solution at 37 °C for 30 min. Next, cells were centrifuged, and the pellet was resuspended in ice-cold PBS. Following cell count, a volume accounting for 100,000 cells was transferred to two separate tubes (representing two biological replicates), which were centrifuged and supernatant discarded. The remaining cell pellet was resuspended in ice-cold cryopreservation solution (50% FBS, 40% growth media, and 10% DMSO) and stored in a −80 °C freezer until shipping. ATAC-Seq data were aligned to hg38 human reference genome. Comparative analysis was performed by a standard normalization method, and peaks were determined using the MACS 3.0.0 algorithm at a cutoff of P-value 1×107, without control file, and with the -nomodel option. False peaks were removed according to the ENCODE blacklist. Statistical Analysis. Data were analyzed and statistical tests were applied (unpaired Student’s t test or ANOVA) using GraphPad Prism 8.4.3 (GraphPad software). Venn diagrams were generated by Venny2.1 software (https://bioinfogp. cnb.csic.es/tools/venny/). Statistical significance is determined for P-value < 0.05; error bars represent ± SD. Data, Materials, and Software Availability. All study data are included in the article and/or SI Appendix. RNA-seq and ATAC-seq data have been deposited in the National Center for Biotechnology Information (NCBI)- Gene Expression Omnibus (GEO) under GSE229454 (56). ACKNOWLEDGMENTS. We thank Murali Ramachandra and Akhil Kumar from Aurigene. We also thank Stephanie Ellison for her help in preparing the manu- script. This work was funded by NINDS R01NS110572 (S.V.), NCI R01ACA261926 (S.V.), Chad Tough Defeat DIPG Foundation (S.V. and M.M.), Alex Lemonade Stand Foundation R-accelerator Award (S.V.), the Hyundai Hope on Wheel Foundation (S.V.), and the Rogel Cancer Center PIBS Graduate Student Scholarship (M.P.). A.M.C. is a Howard Hughes Medical Institute Investigator, A. Alfred Taubman Scholar, and American Cancer Society Professor; A.M.C. also acknowledges sup- port from the NCI Outstanding Investigator Award (R35 CA231996). Author affiliations: aLaboratory of Brain Tumor Metabolism and Epigenetics, Department of Pathology, University of Michigan, Ann Arbor, MI 48109; bChad Carr Pediatric Tumor Center, Department of Pediatrics, University of Michigan, Ann Arbor, MI 48109; cCellular and Molecular Biology Program, University of Michigan Medical School, Ann Arbor, MI 48109; dMedical Scientist Training Program, University of Michigan Medical School, Ann Arbor, MI 48109; eDepartment of New Biology, Daegu Gyeongbuk Institute of Science and Technology, Daegu 42988, Korea; fDepartment of Pathology and Laboratory Medicine, Children’s Hospital Los Angeles, Keck School of Medicine University of Southern California, Los Angeles, CA 90027; gAurigene Discovery Technologies, Bengaluru, Karnataka 560100, India; hMichigan Center for Translational Pathology, Department of Pathology, University of Michigan Medical School, Ann Arbor, MI 48109; iRogel Cancer Center, University of Michigan Medical School, Ann Arbor, MI 48109; jDepartment of Urology, University of Michigan Medical School, Ann Arbor, MI 48109; and kHHMI, University of Michigan Medical School, Ann Arbor, MI 48109 1. 2. 3. K. M. Schroeder, C. M. Hoeman, O. J. Becher, Children are not just little adults: Recent advances in understanding of diffuse intrinsic pontine glioma biology. Pediatr. Res. 75, 205–209 (2014). R. Aziz-Bose, M. Monje, Diffuse intrinsic pontine glioma: Molecular landscape and emerging therapeutic targets. Curr. Opin. Oncol. 31, 522–530 (2019). J. Schwartzentruber et al., Driver mutations in histone H3.3 and chromatin remodelling genes in paediatric glioblastoma. Nature 482, 226–231 (2012). 4. 5. 6. G. Wu et al., Somatic histone H3 alterations in pediatric diffuse intrinsic pontine gliomas and non- brainstem glioblastomas. Nat. Genet. 44, 251–253 (2012). D. Sturm et al., Hotspot Mutations in H3F3A and IDH1 Define Distinct Epigenetic and Biological Subgroups of Glioblastoma. Cancer Cell 22, 425–437 (2012). K. M. Chan et al., The histone H3.3K27M mutation in pediatric glioma reprograms H3K27 methylation and gene expression. Genes Dev. 27, 985–990 (2013). 8 of 9 https://doi.org/10.1073/pnas.2221175120 pnas.org 7. 8. 9. S. Bender et al., Reduced H3K27me3 and DNA hypomethylation are major drivers of gene expression in K27M mutant pediatric high-grade gliomas. Cancer Cell 24, 660–672 (2013). A. S. Harutyunyan et al., H3K27M induces defective chromatin spread of PRC2-mediated repressive H3K27me2/me3 and is essential for glioma tumorigenesis. Nat. Commun. 10, 1262 (2019). S. Nagaraja et al., Transcriptional dependencies in diffuse intrinsic pontine glioma. Cancer Cell 31, 635–652.e636 (2017). 10. J. D. Larson et al., Histone H3.3 K27M accelerates spontaneous brainstem glioma and drives restricted changes in bivalent gene expression. Cancer Cell 35, 140–155.e147 (2019). 33. L. Xiao et al., Targeting SWI/SNF ATPases in enhancer-addicted prostate cancer. Nature 601, 434–439 (2021), 10.1038/s41586-021-04246-z. 34. S. Nagaraja et al., Histone variant and cell context determine H3K27M reprogramming of the enhancer landscape and oncogenic state. Mol. Cell 76, 965–980.e912 (2019). 35. B. Krug et al., Pervasive H3K27 acetylation leads to erv expression and a therapeutic vulnerability in H3K27M gliomas. Cancer Cell 35, 782–797.e788 (2019). 36. C. Chung et al., Integrated metabolic and epigenomic reprograming by H3K27M mutations in diffuse intrinsic pontine gliomas. Cancer Cell 38, 334–349.e339 (2020). 11. D. Haag et al., H3.3-K27M drives neural stem cell-specific gliomagenesis in a human iPSC-derived 37. N. Mashtalir et al., Chromatin landscape signals differentially dictate the activities of mSWI/SNF model. Cancer Cell 39, 407–422.e413 (2021). family complexes. Science 373, 306–315 (2021). 12. N. Furth et al., H3–K27M-mutant nucleosomes interact with MLL1 to shape the glioma epigenetic 38. R. Siddaway et al., Oncohistone interactome profiling uncovers contrasting oncogenic mechanisms landscape. Cell Rep. 39, 110836 (2022). 13. N. Harpaz et al., Single-cell epigenetic analysis reveals principles of chromatin states in H3.3-K27M 14. gliomas. Mol. Cell 82, 2696–2713.e2699 (2022). I. Balakrishnan et al., Senescence induced by BMI1 inhibition is a therapeutic vulnerability in H3K27M-Mutant DIPG. Cell Rep. 33, 108286 (2020). and identifies potential therapeutic targets in high grade glioma. Acta Neuropathol. 144, 1027–1048 (2022). 39. T. Nagashima et al., Discovery of novel forkhead box O1 inhibitors for treating type 2 diabetes: Improvement of fasting glycemia in diabetic db/db mice. Mol. Pharmacol. 78, 961–970 (2010). 40. P. Zou et al., Targeting FoxO1 with AS1842856 suppresses adipogenesis. Cell Cycle 13, 3759–3767 15. J. N. Anastas et al., Re-programing chromatin with a bifunctional LSD1/HDAC inhibitor induces (2014). therapeutic differentiation in DIPG. Cancer Cell 36, 528–544.e510 (2019). 41. A. Pandey, G. S. Kumar, A. Kadakol, V. Malek, A. B. Gaikwad, FoxO1 inhibitors: The future medicine 16. R. Hashizume et al., Pharmacologic inhibition of histone demethylation as a therapy for pediatric for metabolic disorders? Curr. Diabetes Rev. 12, 223–230 (2016). brainstem glioma. Nat. Med. 20, 1394–1396 (2014). 17. H. Katagi et al., Radiosensitization by histone H3 demethylase inhibition in diffuse intrinsic pontine glioma. Clin. Cancer Res. 25, 5572–5583 (2019). 42. Y. Mo et al., Epigenome programing by H3.3K27M mutation creates a dependence of pediatric glioma on SMARCA4. Cancer Discov. 12, 2906–2929 (2022), 10.1158/2159-8290.CD-21-1492. 43. E. Panditharatna et al., BAF complex maintains glioma stem cells in pediatric H3K27M-glioma. 18. C. S. Grasso et al., Functionally defined therapeutic targets in diffuse intrinsic pontine glioma. Nat. Cancer Discov. 12, 2880–2905 (2022), 10.1158/2159-8290.CD-21-1491. Med. 21, 555–559 (2015), 10.1038/nm.3855. 19. A. Piunti et al., Therapeutic targeting of polycomb and BET bromodomain proteins in diffuse intrinsic pontine gliomas. Nat. Med. 23, 493–500 (2017), 10.1038/nm.4296. 44. E. Firat, G. Niedermann, FoxO proteins or loss of functional p53 maintain stemness of glioblastoma stem cells and survival after ionizing radiation plus PI3K/mTOR inhibition. Oncotarget 7, 54883–54896 (2016). 20. F. Mohammad et al., EZH2 is a potential therapeutic target for H3K27M-mutant pediatric gliomas. 45. C. J. Lau, Z. Koty, J. Nalbantoglu, Differential response of glioma cells to FOXO1-directed therapy. Nat. Med. 23, 483–492 (2017), 10.1038/nm.4293. Cancer Res. 69, 5433–5440 (2009). 21. G. L. Brien et al., Simultaneous disruption of PRC2 and enhancer function underlies histone H3.3-K27M oncogenic activity in human hindbrain neural stem cells. Nat. Genet. 53, 1221–1232 (2021). 22. J. L. Pulice, C. Kadoch, Composition and Function of Mammalian SWI/SNF Chromatin Remodeling Complexes in Human Disease. Cold Spring Harb. Symp. Quant. Biol. 81, 53–60 (2016). 23. P. Mittal, C. W. M. Roberts, The SWI/SNF complex in cancer - biology, biomarkers and therapy. Nat. Rev. Clin. Oncol. 17, 435–448 (2020). 46. K. Masui et al., mTOR complex 2 controls glycolytic metabolism in glioblastoma through FoxO acetylation and upregulation of c-Myc. Cell Metab. 18, 726–739 (2013). 47. J. Seoane, H. V. Le, L. Shen, S. A. Anderson, J. Massague, Integration of Smad and forkhead pathways in the control of neuroepithelial and glioblastoma cell proliferation. Cell 117, 211–223 (2004). 48. L. Wang, J. Wang, T. Jin, Y. Zhou, Q. Chen, FoxG1 facilitates proliferation and inhibits differentiation by downregulating FoxO/Smad signaling in glioblastoma. Biochem. Biophys. Res. Commun. 504, 46–53 (2018). 24. S. Schick et al., Acute BAF perturbation causes immediate changes in chromatin accessibility. Nat. 49. N. Galili et al., Fusion of a fork head domain gene to PAX3 in the solid tumour alveolar Genet. 53, 269–278 (2021). 25. Y. K. Park et al., Interplay of BAF and MLL4 promotes cell type-specific enhancer activation. Nature communications 12, 1630 (2021). 26. T. Vierbuchen et al., AP-1 Transcription Factors and the BAF Complex Mediate Signal-Dependent Enhancer Selection. Mol. Cell 68, 1067–1082 e1012 (2017). rhabdomyosarcoma. Nat. Genet. 5, 230–235 (1993). 50. S. Mehra et al., Detection of FOXO1 (FKHR) gene break-apart by fluorescence in situ hybridization in formalin-fixed, paraffin-embedded alveolar rhabdomyosarcomas and its clinicopathologic correlation. Diagn. Mol. Pathol. 17, 14–20 (2008). 51. N. Bharathy et al., The HDAC3-SMARCA4-miR-27a axis promotes expression of the PAX3:FOXO1 27. P. W. Lewis et al., Inhibition of PRC2 Activity by a Gain-of-Function H3 Mutation Found in Pediatric fusion oncogene in rhabdomyosarcoma. Sci. Signal 11, eaau7632 (2018). Glioblastoma. Science 340, 857–861 (2013). 52. D. Laubscher et al., BAF complexes drive proliferation and block myogenic differentiation in fusion- 28. S. U. Jain et al., H3 K27M and EZHIP impede H3K27-methylation spreading by inhibiting positive rhabdomyosarcoma. Nat. Commun. 12, 6924 (2021). allosterically stimulated PRC2. Mol. Cell 80, 726–735.e7 (2020), 10.1016/j.molcel.2020.09.028. 53. R. Eckenstaler, M. Hauke, R. A. Benndorf, A current overview of RhoA, RhoB, and RhoC functions in 29. B. G. Wilson et al., Epigenetic antagonism between polycomb and SWI/SNF complexes during vascular biology and pathology. Biochem. Pharmacol. 206, 115321 (2022). oncogenic transformation. Cancer Cell 18, 316–328 (2010). 30. R. T. Nakayama et al., SMARCB1 is required for widespread BAF complex-mediated activation of enhancers and bivalent promoters. Nat. Genet. 49, 1613–1623 (2017). 54. S. Venneti et al., Evaluation of histone 3 lysine 27 trimethylation (H3K27me3) and enhancer of Zest 2 (EZH2) in pediatric glial and glioneuronal tumors shows decreased H3K27me3 in H3F3A K27M mutant glioblastomas. Brain Pathol. 23, 558–564 (2013). 31. B. H. Alver et al., The SWI/SNF chromatin remodelling complex is required for maintenance of 55. C. Chung et al., Integrated metabolic and epigenomic reprograming by H3K27M mutations in lineage specific enhancers. Nat. Commun. 8, 14648 (2017). diffuse intrinsic pontine gliomas. Cancer Cell 38, 334–349.e9 (2020), 10.1016/j.ccell.2020.07.008. 32. C. M. Weber et al., mSWI/SNF promotes Polycomb repression both directly and through genome- 56. M. Mota et al., Targeting SWI/SNF ATPases in H3.3K27M diffuse intrinsic pontine gliomas. NCBI- wide redistribution. Nat. Struct. Mol. Biol. 28, 501–511 (2021). GEO. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE229454. Deposited 12 April 2023. PNAS 2023 Vol. 120 No. 18 e2221175120 https://doi.org/10.1073/pnas.2221175120 9 of 9
10.1021_acssynbio.3c00078	pubs.acs.org/synthbio This article is licensed under CC-BY 4.0 Research Article Tuning Methylation-Dependent Silencing Dynamics by Synthetic Modulation of CpG Density Yitong Ma, Mark W. Budde, Junqin Zhu, and Michael B. Elowitz* Cite This: ACS Synth. Biol. 2023, 12, 2536−2545 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: Methylation of cytosines in CG dinucleotides (CpGs) within promoters has been shown to lead to gene silencing in mammals in natural contexts. Recently, engineered recruitment of methyltransferases (DNMTs) at specific loci was shown to be sufficient to silence synthetic and endogenous gene expression through this mechanism. A critical parameter for DNA methyl- ation-based silencing is the distribution of CpGs within the target promoter. However, how the number or density of CpGs in the target promoter affects the dynamics of silencing by DNMT recruitment has remained unclear. Here, we constructed a library of promoters with systematically varying CpG content, and analyzed the rate of silencing in response to recruitment of DNMT. We observed a tight correlation between silencing rate and CpG content. Further, methylation-specific analysis revealed a constant accumulation rate of methylation at the promoter after DNMT recruitment. We identified a single CpG site between TATA box and transcription start site (TSS) that accounted for a substantial part of the difference in silencing rates between promoters with differing CpG content, indicating that certain residues play disproportionate roles in controlling silencing. Together, these results provide a library of promoters for synthetic epigenetic and gene regulation applications, as well as insights into the regulatory link between CpG content and silencing rate. KEYWORDS: epigenetics, DNA methylation, DNMT3b, synthetic biology ■ INTRODUCTION Methylation of CG dinucleotides (CpGs) plays critical roles in mammalian development, tumor progression, and aging.1−4 These functions result mainly from the ability of CpG methylation to induce and stabilize gene silencing in mammals through multiple mechanisms.5,6 Control of DNA methylation and further gene silencing depends on both trans-acting factors and the DNA sequence itself. Trans-factors include methyl- ation “writers” such as DNA methyl-transferases (DNMTs)7 and “erasers” such as TET18 that establish and alter methylation marks, as well as “readers” such as MeCP2 and histone deacetylases9 that link methylation to regulation of gene transcription.6,10 In mammalian cells, DNA methylation occurs mainly at CpGs. As a result, the distribution of CpG dinucleotides within a given sequence can play a pivotal role in methylation-based gene regulation. At the genome level, regions with different CpG content exhibit distinct methylation patterns,11,12 potentially due to cooperativity between nearby CpGs and other suppressive epigenetic marks, which can generate positive feedback.13,14 Relatively high CpG-density regions (CpG islands) from a human chromosome largely maintain their methylation state when hosted in a transchromosomic mouse model,15 suggesting that DNA sequence composition plays a strong role in establishing stable methylation states. Conversely, insertion of several hundred base pairs of CpG-free DNA can disrupt these patterns, permitting de novo methylation of the surrounding CpG island.16 However, the precise role of CpG sequence context can be difficult to discern at natural loci, where regulation is also affected by many other cis- and trans- acting factors, including cell-type specific methylation writer and reader profiles, neighboring (non-CpG) motifs that recruit epigenetic modifiers, pre-existing chromatin states, and so forth. To clearly delineate the role of DNA sequence in methylation and silencing, one would ideally want to directly compare the methylation and silencing of similar promoters with different CpG distributions, in the same genomic context. Further, because methylation and its effects on gene regulation can both be dynamic,17−19 the ability to control the timing of to a locus and methyltransferases (DNMTs) recruitment Received: February 2, 2023 Published: August 12, 2023 © 2023 The Authors. Published by American Chemical Society 2536 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article Figure 1. rTetR fused DNMT3b catalytic domain methylates and silences a reporter library of different CpG content upon recruitment: (A) schematic of the synthetic methylation-silencing system. rTetR fused DNMT3bCD is expressed constitutively, and upon induction of dox, recruited to the promoter region of a site-specifically integrated Citrine reporter. The recruitment methylates the promoter and further silences the gene expression, with dynamics depending on the promoter’s CpG content. (B) Design of the library of promoters with varying CpG content. 5× tandem TetO binding sites were fused with an insert (or no insert) and a pEF1s synthetic promoter to make the library of promoters. Red lines represent CpG dinucleotides, and the pEF1s promoters are vertically aligned to show sequence homology. follow the resulting changes in gene expression is also desirable. The field of synthetic epigenetics seeks to harness epigenetic regulatory mechanisms to control gene expression on different timescales.20,21 Recent work demonstrated the ability to regulate synthetic or endogenous gene expression by recruiting DNMTs to specific target genes18,22−24 and even create fully synthetic DNA methylation-based systems synthetic epigenetic memory.25 In CHO-K1 cells, transient DNMT recruitment to a locus drives stochastic, irreversible, all-or-none silencing over timescales of about 5 days.18 However, it is unclear how the dynamics of gene silencing depends on the DNA sequence of the regulated target gene. Understanding the effects of sequence composition on silencing rates could provide insights into gene regulation by DNA methylation and also expand the synthetic epigenetic toolbox for fine tuning circuits. for Here, we adapted a previously established DNMT recruit- ment system to analyze the effects of DNA sequence on methylation-dependent silencing.18 We derived a library of promoters with different CpG densities from a synthetic promoter and observed the relationship between CpG density and the silencing dynamics occurring after DNMT recruit- ment. Using a mathematical model of methylation, together with sequencing identifying methylation marks, we showed that the observed gene expression dynamics could be explained by methylation accumulation on the DNA. We also identified several specific CpG elements that appear to play dispropor- tionate roles in silencing dynamics and confirmed that one of them [near the TATA box and transcription start site (TSS)] causes significant changes in silencing dynamics. Our results reveal how CpG density influences silencing dynamics and provide a library of promoters with different silencing rates for synthetic applications. ■ RESULTS AND DISCUSSION system, Construction of a Promoter Library with Varying CpG Content. To investigate the relation between promoter CpG content and its DNMT-dependent silencing rate, we adopted a previously described synthetic methylation-silencing system.18 the catalytic domain of DNMT3b In this (DNMT3bCD) is fused with a reversed tetracycline repressor (rTetR), allowing precise temporal control of the recruitment to a target gene by adding doxycycline (dox) to the culture media.26 Here, to focus on the role of DNA methylation in silencing, we specifically used the mouse gene Dnmt3b′s catalytic domain, spanning amino acids 402−872. This region omits the major heterochromatin-interacting PWWP do- main.27 This construct also incorporates a co-expressed H2B- mCherry fluorescent protein fusion. We stably integrated this construct using the piggyBac transposon system and sorted cell populations with similar mCherry fluorescence level (Figure S1A). This enabled direct comparison of different promoters (see below) with the same DNMT3bCD expression context. We used an H2B-mCitrine28 fluorescent fusion protein as the target gene. This target was driven by one of a set of promoters containing varying densities of CpG (see below). In each promoter, an array of 5 rTetR binding sites (TetO) was the promoter, allowing recruitment of fused upstream of rTetR-DNMT3bCD (Figure 1A). To enable direct compar- 2537 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article Figure 2. Promoters’ silencing rate correlates with their CpG content. (A) Promoter silences with all-or-none kinetics when DNMT3bCD is recruited to the locus, and this silencing is dependent on DNMT3b’s catalytic activity. Cells with the pEF1s(high) promoter were treated with dox for 4 days, and then no dox (release) for 2, 6, and 10 days. Cells are analyzed by flow cytometry at various time points (left) and quantified by comparing to the no dox control (black lines in the left). A log−normal distribution is first fitted onto the no dox control’s positive population, and μ−2σ are used for quantification of silenced fractions. These fractions are then normalized to no dox controls’ silenced fraction (see Materials and Methods, right). The silenced fractions are stable after the release of dox for 2 days, and no silencing is observed in the DNMT3bCI controls. (B) Time course of the silenced fraction of different promoters. Cells were treated with or without dox, and then with 2 days of no dox (release). The fraction of silencing is determined as described in (A): cells with lower fluorescence than μ−2σ of the no dox control group were determined as silenced. The silencing rate is further normalized to the no dox group (see Materials and Methods). For the shorter time scale (left), the same method is used except with a higher time resolution. (C) Summary of the silencing rates in (B). The silencing rate is calculated by subtraction between each pair of neighboring dots and then normalized by time intervals in between, as well as the remaining fraction size. We omitted dot pairs over 80% fraction as the normalization fraction is too small. ison between target promoters at the same genomic context, all reporter cassettes were site-specifically integrated as a single copy into the epigenetically active φC31attP/attB landing site within an artificial chromosome that was previously engineered into CHO cells (MI-HAC CHO cells, also see Materials and Methods).29 We constructed a library of synthetic promoters that differed in their CpG densities. We started with a synthetic version of the human elongation factor 1α promoter [pEF1s(orig), with 18% CpG density], a 544 bp fusion of promoter fragments from the human EF1α promoter and human T-cell virus (HTLV),30 that is commercially available (InvivoGen) and has been used for antibody expression and gene therapies.31,32 To identify conserved CpG elements, we compared both the EF1α fragment and HTLV fragments of this promoter to their natural orthologs, respectively.33 We then removed or added CG pairs into the promoter at non-conserved sites. With this procedure, we generated promoters with varying CpG densities at 9.6 and 24% [pEF1s(low) and pEF1s(high), respectively, Figure 1B, middle]. Next, to broaden the range of CpG densities, we designed an additional DNA segment, inserted upstream of the promoter, containing high (60%) CpG density (Figure 1B, high CpG insert). We altered this CpG insert by swapping out CG with GC dinucleotides, or by replacing C with T, to create a lower CpG density (5.4%) insert, while otherwise preserving its sequence similarity with the high CpG insert (Figure 1B, low CpG insert). Altogether, we combined the three pEF1s promoters with the two inserts, or with no insert, to produce a library of 7 sequences whose overall CpG density ranged 2538 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article Figure 3. Sequencing reveals the constant accumulation of methylation and, potentially, master CpGs. (A) Schematics of the FACS-Sequencing experiment: cells with different promoters are treated with dox, and then FACS-sorted to three bins based on the Citrine brightness (high, med, and low), consisting cells that are “still ON,” “recently silenced,” and “long silenced,” respectively. The first two groups proceed to downstream methylation-specific sequencing (see Materials and Methods). (B−D) Total CpG methylation (B), methylation frequency (C), and total methylation normalized by promoter length (D) accumulates in the promoter with time in the “still ON” cell population. “Still ON” populations are sorted out as indicated in (A) at intended dates, and subsequently analyzed by methylation-specific sequencing (EM-seq), targeting the integrated gene promoter. (E) CpGs around TATA-box and TSS (highlighted in green) show significant difference in methylation between the “still ON” and “recently silenced” group. Methylation percentages of different samples at different days were pooled together for comparison (a total of 10 from “still ON” group compared to 6 from “recently silenced” group). P-values are from Student t-test. (F) Mutation at CpG793 changes promoters’ silencing rate significantly. New cell lines are constructed by introducing point mutations (CG to CC or the inverse) at CpG793 in pEF1s(high), pEF1s(orig), and pEF1s(low) promoters (top). Cell lines are then constructed as described previously in this paper. DNMT3bCD recruitments are induced by dox at day zero and cells were analyzed by flow cytometry at each time point after dox induction. (G) Quantification of the silencing rate (similar method as Figure S3A and eq 1) of time course in (F). We excluded the time point at 0 from the fitting as dox release was not included in this experiment. P-values are calculated based on the estimation and standard error of β from linear regression. 2539 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article from 8.0 to 36% (Figure 1B right). Despite their variation in CpG density, all 7 promoters drove strong expression of the fluorescent protein reporter, producing ∼200-fold greater signal compared to autofluorescence in non-transfected cells, with a 2.8 fold variation of expression level (Figure S1B,C). This difference could be either due to the difference in the promoter sequences or due to altered mRNA secondary structure of 5′UTR (part of the altered promoter sequence), resulting in changes this expression change did not impact quantification of the silenced fraction, as silencing thresholds were set independently for each promoter (see below). As expected, the original pEF1s promoter shows the highest activity, while our alterations to the sequence (both addition or reduction of CpGs) slightly lowered its activity. Therefore, there was no correlation between expression level and CpG content. in RNA half-life.34 However, DNMT-Dependent Silencing Rate Correlates with Promoter CpG Density. Previous analysis of silencing dynamics by DNA methylation in a similar system revealed that transcriptional silencing occurs through stochastic, all-or- none, irreversible events in individual cells.18 To confirm that our system has similar kinetics, we induced DNMT3bCD recruitment to promoters for 4 days and then released the recruitment for 2, 6 and 10 days and measured the Citrine fluorescence via flow cytometry for all three promoter variants [Figures 2A and S2A,B for pEF1s(high), pEF1s(orig), and pEF1s(low), respectively]. As expected, a fraction of the cells were transcriptionally silenced after the induction, and gradually diluted out stable H2B-Citrine fluorescent protein during the “release” phase due to cell divisions (approximately 22 h per division, observed by the shifting position of silenced population peaks). Meanwhile, the active populations (peaks on the right) remained stable in terms of both fluorescence level and cell population fraction, indicating an all-or-none, irreversible kinetics as reported before. Setting a silencing threshold 2 standard deviations (2σ) below the mean fluorescence levels of actively expressing (“no dox”) cells yielded a stable value after 2 days of release (see quantification on the right in Figures 2A and S2A,B), allowing consistent quantitation of all-or-none silencing. Because the control groups were single peaked and exhibited consistent variation, this cutoff occurred at values ranging from 54 to 71% of the mean, depending on cell line (Figure S2C). We also used a catalytically inactive version of the DNMT3bCD protein (P656V and C657D double point mutations,35 noted as DNMT3bCI) as a negative control. Even though the CI versions were expressed at a similar level of the CD version (Figure S1A lower half), no silencing effect was observed when they were recruited to the promoters (Figures 2A and S2A,B, lower half as well as blue lines in the quantifications). This confirmed that the observed silencing effects resulted specifically from DNA methylation activity and interference with not transcription machinery. from other protein interactions or These results established that the promoters silence in a methylation-dependent, all-or-none and irreversible manner and indicate that silencing kinetics can be captured by the dox induce-and-release protocol. To quantify silencing dynamics across the library, we analyzed the dynamics of silent cell accumulation over a time course (up to 19 days) of dox induction with 2-day subsequent release of the recruitment at various time points (Figure 2B, right). As some of the promoters, notably those with insert(high) or pEF1s(high), reached over 50% of silencing only after four days (about two time points in our setup), we added a biological replicate for each of the four fastest promoters with a separate time course with finer time resolution (Figure 2B, left). From the time course, we can conclude that higher overall CpG density on the promoter results in faster silencing dynamics. This result is robust to analysis with a stricter cutoff of 90% expression reduction (Figure S2B). Even with the less sensitive threshold, we observed a similar trend of increased CpG density correlating with faster silencing. The time-course dynamics for each promoter could be summarized by an empirical silencing rate as the silenced fraction per unit time (day), normalized by the remaining active population. Silencing rates varied over nearly an order of magnitude across promoters with various CpG densities. Further, the silencing rate correlated linearly with CpG density over this range (Figure 2C). These results showed that across varying CpG densities, DNMT-dependent silencing dynamics are broadly consistent with a stochastic, all-or-none silencing process, occurring at a rate that depends on CpG density. Silencing Kinetics Follow a Stochastic Switching Model. Previous studies using a similar system with a different version of the EF1α promoter showed that silencing kinetics could be described as a single-step stochastic switching event from the active to the silent state18 (Figure S3A). In the model, each promoter silences stochastically at a time-invariant rate β. Here, we tested whether a similar model could fit our data, if we allowed β(c) to depend on CpG density, c. With this assumption, the size of the active population fraction, A, can be described by a simple differential equation (1) Note that cell proliferation does not need to be explicitly incorporated due to the heritability of the expression state. In this model, the active population, A, decays exponentially over time, t. To test this model, we plotted the time course data (Figure 2B) in terms of the remaining active population, A(t) (Figure S3B), and observed a linear−log relationship, consistent with exponential decay, for every promoter variant. In these plots, the switching rate β(c) ranged from 0.032 to 0.274 d−1, depending on CpG density. These results are consistent with a simple stochastic switching model in which silencing rate is tuned by CpG density. Methylation Accumulates after DNMT Recruitment. Given that promoter silencing depends on the methylation activity of the recruited DNMT3bCD, we next asked whether methylation accumulates at similar or different rates for different promoters. We used fluorescence activated cell sorting (FACS) to isolate the transcriptionally active cell fraction (A in the model) at different times after dox addition (Figure 3A). We then measured promoter CpG methylation profiles using methylation-specific sequencing (EM-seq36) (Materials and Methods, Figure 3A). As expected, methylation accumulated in the transcription- ally active populations, as measured by methylation rate (methyl-CpG over total CpG), total methylation per promoter, as well as total methylation per promoter per bp of DNA (Figure 3B−D respectively). Unexpectedly, however, the rate of methylation accumulation was independent of CpG density, measured as total methylation per bp of DNA (Figure 3D). In fact, the rate of methylation per CpG was greater at promoters 2540 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 AttcAd()d()=·ACS Synthetic Biology pubs.acs.org/synthbio Research Article with lower CpG densities (Figure 3B), while the total number of methylated CpG in the promoter region was similar across is compatible with different promoters. This behavior saturation of methylation capacity of the locally recruited DNMT3bCD. Alternatively, it could also reflect an effective interaction, in which unmethylated CpGs inhibit methylation at nearby CpG sites.13 Single CpG Has a Disproportionate Impact on Silencing Rate. The apparent discrepancy between the CpG density-dependent silencing rate and the density- independent methylation rate provoked the question of whether certain individual CpGs might play disproportionate roles in controlling silencing. Such CpGs would be expected to exhibit significant differences in methylation between cell populations containing active versus recently silenced promoters. To discover such CpGs, we pooled all available sequencing results from different time points. Within the pEF1s region, where all three promoters overlap (∼90% of the pEF1s region) (Figure 3E), we identified three CpGs with significantly different methylation levels between the two expression groups (p < 0.05). Interestingly, two of the most significant CpGs, including the top ranked one (CpG at position 793, or CpG793 for short), are located between the TATA box and the TSS (arrow in Figure 3C), consistent with previous reports suggesting functionally important CpG islands around the TSS.37 CpG793 was among the CpGs that were eliminated in the construction of the low CpG pEF1s(low) promoter, consistent with the lower silencing rate observed for this promoter (Figure 2B). To test for a functional role of CpG793, we mutated it to CC in pEF1s(orig) and pEF1s(high). Conversely, we also reverted this position back to CG in pEF1s(low), where all 22 other CpGs including CpG793 were mutated previously. these constructs provided a set of controlled Together, comparisons in which position 793 was either CC or CG in pEF1s(high), pEF1s(orig), and pEF1s(low) (Figure 3F, top). We analyzed silencing rates (fraction per day normalized by remaining fraction, similar to Figure S3B) for each of these promoters. These rates were significantly reduced in the “CC” variants of pEF1s(orig) (p < 0.001) and pEF1s(low) (p < 0.001), but not the pEF1s(high) promoters, compared to the CG variants (Figures 3F lower, and 3G). Further, silencing rates were similar between the CG variant of pEFs1(low) and the CC variant of pEFs1(orig) (barely significantly different with p = 0.047), even though these two sequences systemati- cally differed at 22 other CpGs. This indicates that the position 793 mutation could almost compensate for the combined effect of 22 other CpG mutations. Finally, we asked if the observed differential silencing dynamics caused by the CC-CG mutation at position 793 could result from disruption or introduction of a known transcription factor binding site. We queried the surrounding sequence (±8 nt, 18 nts in total) against known mouse cis- regulatory elements in CIS-BP38 and filtered for hits expressed in CHO cells39 based on criteria suggested previously40 (Table S1). The only hits observed in both the “low” and “orig” promoters were Gmeb1 and Gmeb2, a pair of proteins that are involved in modulating glucocorticoid receptor-mediated transactivation.41 However, these proteins are not known to be directly involved in epigenetic regulation, to our knowledge. While we cannot rule out the possibility that the observed difference in silencing results from differential binding of sequence-dependent cis-factors, is consistent with the explanation that methylation capability at this position has a disproportionate effect on silencing. it Together, we observed three key results: First, CpG methylation in promoters in the “still ON” population accumulated with time. Second, the silencing rate did not correlate with either methylation rate or total methylation, contrary to expectation. Third, we discovered a specific CpG position that plays a disproportionate, role in controlling silencing rate. functional ■ CONCLUSIONS While effects of sequence on DNMT-dependent gene silencing have long been observed, a controlled system for directly analyzing the effects of sequence on silencing has not been available. Here, we constructed a library of synthetic promoters, featuring varying CpG content and methylation- dependent silencing kinetics (Figure 1). Strikingly, silencing rate correlates directly with CpG content (Figure 2C). However, this correlation could not be explained by a corresponding effect of CpG content on methylation, as methylation accumulated at similar rates in all promoter variants (Figure 3B−D). Finally, we observed evidence that a certain CpG (CpG793), located between the TATA box and the TSS, can play a disproportionate role in control of silencing rate (Figure 3F,G). Together, these results should provide a versatile set of components for engineering synthetic epigenetic circuits with desired silencing behaviors, as well as a foundation for future investigations of the mechanisms of DNMT-dependent silencing. Finally, our observation that the DNA sequence-based substrate of epigenetic modifications could alter the regulation dynamics might also apply into fully synthetic epigenetic circuits.25 A remaining mystery is why the rate of methylation accumulation is correlated neither with the rate of silencing nor with the CpG content of the promoter. Despite the lack of correlation between silencing rate and accumulated methyl- ation, promoter silencing depended on the methylation activity of the recruited DNMT, as a catalytically inactive variant of DNMT3b was not able to initiate silencing in our system (Figure 2A), indicating that de novo DNA methylation is a necessary requirement for promoter silencing in this context. A possible explanation could be that silencing requires at least two distinct steps, mediated by two types of trans- regulatory factors: the first binds to methyl-CpG, and the second binds to CpG in a methylation-independent fashion. If only a small number of methyl-CpG is required for the first, methyl-dependent factor(s), then total CpG density could establish a rate-limiting step for advancing to a silent state. Examples of both types of proteins exist. Methyl-binding domain (MBD) proteins like MeCP2, MBD2, and so forth are known to play key roles in methylation-dependent silenc- ing.6,42 At the same time, CpG islands are known to be able to initiate silencing by recruiting polycomb group proteins independent of methylation in embryonic stem cells differ- entiating into neurons.43 There are also “dual functional” proteins (e.g., TET144 and KDM2B45) that bind to un- methylated CpGs but still promote gene silencing in some cell contexts. Further experiments could help to disentangle the roles of methylation-dependent and independent factors in controlling the rate of silencing. Starting within two days after the release of dox, the active population remained in an actively expressing state (Figure 2541 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article 2A). DNA methylation is actively maintained and, thus, unlikely to dilute out during this period without active recruitment of de-methylation enzymes.46 Therefore, the recruited DNMT3bCD protein may play an additional role in silencing beyond its catalytic activity as a methyltransferase. In fact, full length DNMT3b, even with its catalytic domain deactivated, has significant functions in epigenetic gene regulation through its protein and heterochromatin interacting domain.35 In this study, we specifically recruited the DNMT3b “catalytic domain” (with the PWWP domain deleted). However, this protein still includes the ATRX domain that has been shown to associate with heterochromatin.27 Further investigation will be needed to identify the roles of methyltransferase-dependent and independent activities of DNMT3b. One factor that could complicate our comparison between promoters is the distance from the recruited site (5xtetO) to the promoter’s core. This distance differed in promoters with additional inserts. However, the correlation of silencing rate with CpG density occurred among groups of constructs either lacking or containing the insert, when these groups were considered separately. This suggests that the change in distance to the core promoter (roughly 300 base pairs) in this system is not responsible for silencing rate correlation. Additionally, our discovery of CpG793 playing a dispropor- tionate role in determining silencing dynamics also suggested our model of correlation between CpG density and silencing rate is incomplete. A much larger set of promoter variants containing combinatory mutations on all CpGs may provide a more complete model accounting for the individual effects of each CpG. We believe this issue would be better resolved in the future using a massively parallel reporter assay approach that can access much larger numbers of promoters. Finally, we note that phenotypically, our findings resemble the genetic mechanism of fragile X syndrome (FRX), in which an increased CGG repeat number upstream of the FRM1 gene’s promoter leads to hypermethylation and gene silencing during development.47 The exact molecular mechanism leading to silencing in FRX is not yet fully understood, but various hypotheses, including toxic secondary RNA structure48 and aberrant histone deacetylation,49 have been proposed. It would be interesting to find out to what extent the mechanisms underlying the relationships observed here may be shared with those involved in FRX. ■ MATERIALS AND METHODS Cell Culture Maintaining. CHO cells containing a human artificial chromosome (CHO-HAC)29 were cultured at 37 °C, in a humidified atmosphere with 5% CO2. The growth media consisted of Alpha MEM Earle’s Salts (Irvine Scientific) with 10% Tet Approved FBS (Clontech Laboratories or Avantor) and 1× penicillin/streptomycin (Life Technologies) and 1× GlutaMax (Gibco) added. Cells were passaged according to the standard CHO-K1 cell (CCL-61, ATCC) procedure. Plasmid and Cell Line Construction. All plasmids are constructed using standard cloning techniques, including Gibson Assembly (NEB) and GoldenGate Assembly (NEB). The plasmids and their maps are available for requests at Addgene (addgene.org/browse/article/28233817/). The basal cell line expressing rTetR-DNMT3B (CD and CI version) was constructed by transfection and stable integration following via the PiggyBAC system (System Biosciences), manufacturer’s instructions, followed by blasticidin (Gibco) selection at 10 μg/mL for 5 days. The cells were then sorted for similar mCherry expression (Figure S1A), or single cloned further reporter integration (in the case of finer time course in Figure 2B). For integration of the reporter, methods similar to previous literature18 were used. Briefly, we co-transfected 600 ng reporter plasmid and 200 ng PhiC31 integrase plasmid using Lipofectamine 2000 (Invitrogen). After selection by geneticin (Gibco) at 400 ng/mL for 14 days (Figure S1B), cells were sorted (see below) to isolate the population with expression around the highest peak (Figure S1C, expected expression of single integration, as their system’s single integration rate should be close to 90% after selection29). Flow Cytometry and FACS. Cells were washed by PBS, lifted by 0.25% EDTA−Trypsin (Gibco), and diluted in HBSS (Gibco) with 0.25% of BSA before flow cytometry. Flow cytometry experiments were performed either on MACSQuant VYB Analyzer (Miltenyi Biotec) or CytoFLEX (Beckman Coulter). Analysis of data was done with open source, in-house developed software, EasyFlow (https://github.com/ AntebiLab/easyflow), or EasyFlowQ (https://github.com/ ym3141/EasyFlowQ). FACS was performed with SY3200 Cell Sorter (Sony) at Caltech FLow Cytometry Facility. Enzymatic Methylation-Specific Sequencing and Analysis. Cells were sorted as described above and immediately lysed for DNA extraction (DNeasy Blood & Tissue Kit, Qiagen). Total DNA was then converted with NEBNext Enzymatic Methyl-seq Conversion Module (NEB) according to the manufacturer’s instructions and further amplified (EpiMark Hot Start Taq DNA Polymerase, NEB) with primers targeting a 2.5 kb region containing TetO binding sites, promoter region, and the gene body (nucleotide 1943− 4438 on the none-pEF1s(orig) plasmid). The amplified targets were further prepared into library (Nextera XT Library Prep protocol Illumina) and sequenced on the MiSeq (250 bp pair ended, Illumina) platform. The resulting reads were first trimmed and filtered by Trim Galore! (Babraham Institute) and then aligned and analyzed by Bismark50 and SAMtools51 to generate the methylation calling statistics. Data Processing and Statistical Testing. For calculating the silenced fractions, the background silenced fraction (Sdox−) from the no recruitment control sample (no dox) was subtracted from the observed silenced fraction (Sdox+) from the with recruitment group experiment and further normalized by the “fraction still available for silencing” (“still ON” fraction in the control 1 − Sdox−. Consequently, the silenced rate of a given sample was calculated as follows: All statistical testings in this study were Student’s t-test if not specified. Error bars in Figure 3F,G were generated via bootstrapping. Specifically, for each time point in Figure 3F,G, each of the three “with recruitment” samples were normalized to each of three “no recruitment” control samples, according to the method described above. Therefore, a total of 9 data points were generated, and the error bars represent the standard deviation of these points. 2542 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 SSSS1doxdoxdox=+ACS Synthetic Biology pubs.acs.org/synthbio Research Article ■ ASSOCIATED CONTENT sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.3c00078. Legend for transcription factors that bind differentially only to the CG or CC (at CpG793) version of the promoters; additional characterization of the cell lines; additional characterization of the silencing time course and analysis with alternative criteria; and mathematical model schematics and fittings to the data (PDF) ■ AUTHOR INFORMATION Corresponding Author Michael B. Elowitz − Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, California 91125, United States; Howard Hughes Medical Institute, California Institute of Technology, Pasadena, California 91125, United States; 1221-0967; Email: [email protected] orcid.org/0000-0002- Authors Yitong Ma − Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, California 91125, United States; orcid.org/0000-0003-4446-7326 Mark W. Budde − Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, California 91125, United States; Primordium Labs, Arcadia, California 91006, United States Junqin Zhu − Department of Biology, Stanford University, Stanford, California 94305, United States Complete contact information is available at: https://pubs.acs.org/10.1021/acssynbio.3c00078 formal analysis, Author Contributions investigation, and Y.M.: conceptualization, writing; formal analysis and investigation; M.W.B.: J.Z.: conceptualization, supervision, and funding; M.B.E: conceptu- alization, supervision, writing, and funding. Funding This work is supported by the Defense Advanced Research Projects Agency under contract no. HR0011-17-2-0008, by the National Institutes of Health grant RO1 HD075605A, and by National Science Foundation grant EF-2021552 under subaward UWSC10142. M.B.E. is a Howard Hughes Medical Institute Investigator. Notes The authors declare the following competing financial interest(s): M.W.B. is a founder and employee of Primordium Labs. Plasmids and their maps available for requests at Addgene (addgene.org/browse/article/28233817/). The key cell lines are available upon request. EM-Seq raw and processed data is deposited at Gene Expression Omnibus (GSE224403). Data and codes for analysis and generating figures are available at data.caltech (doi: 10.22002/ct5kt-cv878). ■ ACKNOWLEDGMENTS We thank Jeff Park for technical assistance and advice; Rochelle Diamond and Jamie Tejirina at the Caltech Flow Cytometry and Cell Sorting Facility for technical advice and assistance. Yodai Takei, James Linton, Shiyu Xia, and other the Elowitz lab for critical members of feedback on the manuscript; Lacramioara Bintu and Matt Thomson for scientific input and advice. Part of the content in this article was also included in Y.M.’s doctoral thesis (doi: 10.7907/ w0q1-7s17). This article is subject to HHMI’s Open Access to Publications policy. HHMI lab heads have previously granted a nonexclusive CC BY 4.0 license to the public and a sublicensable license to HHMI in their research articles. Pursuant to those licenses, the author-accepted manuscript of this article can be made freely available under a CC BY 4.0 license immediately upon publication. ■ REFERENCES (1) Smith, Z. D.; Meissner, A. DNA Methylation: Roles in Mammalian Development. Nat. Rev. Genet. 2013, 14, 204−220. (2) Ehrlich, M. DNA Methylation in Cancer: Too Much, but Also Too Little. Oncogene 2002, 21, 5400−5413. (3) Greenberg, M. V. C.; Bourc’his, D. The Diverse Roles of DNA Methylation in Mammalian Development and Disease. Nat. Rev. Mol. Cell Biol. 2019, 20, 590−607. (4) McCabe, M. T.; Brandes, J. C.; Vertino, P. M. Cancer DNA Methylation: Molecular Mechanisms and Clinical Implications. Clin. Cancer Res. 2009, 15, 3927−3937. (5) Attwood, J. T.; Yung, R. L.; Richardson, B. C. DNA Methylation and the Regulation of Gene Transcription. Cell. Mol. Life Sci. 2002, 59, 241−257. (6) Zhu, H.; Wang, G.; Qian, J. Transcription Factors as Readers and Effectors of DNA Methylation. Nat. Rev. Genet. 2016, 17, 551− 565. (7) Lyko, F. The DNA Methyltransferase Family: A Versatile Toolkit for Epigenetic Regulation. Nat. Rev. Genet. 2018, 19, 81−92. (8) Morita, S.; Noguchi, H.; Horii, T.; Nakabayashi, K.; Kimura, M.; Okamura, K.; Sakai, A.; Nakashima, H.; Hata, K.; Nakashima, K.; Hatada, I. Targeted DNA Demethylation in Vivo Using dCas9− peptide Repeat and scFv−TET1 Catalytic Domain Fusions. Nat. Biotechnol. 2016, 34, 1060−1065. (9) Choudhury, S. R.; Cui, Y.; Lubecka, K.; Stefanska, B.; Irudayaraj, J. CRISPR-dCas9 Mediated TET1 Targeting for Selective DNA Demethylation at BRCA1 Promoter. Oncotarget 2016, 7, 46545− 46556. (10) Moore, L. D.; Le, T.; Fan, G. DNA Methylation and Its Basic Function. Neuropsychopharmacology 2013, 38, 23−38. (11) Lövkvist, C.; Dodd, I. B.; Sneppen, K.; Haerter, J. O. DNA Methylation in Human Epigenomes Depends on Local Topology of CpG Sites. Nucleic Acids Res. 2016, 44, 5123−5132. (12) Weber, M.; Hellmann, I.; Stadler, M. B.; Ramos, L.; Pääbo, S.; Rebhan, M.; Schübeler, D. Distribution, Silencing Potential and Evolutionary Impact of Promoter DNA Methylation in the Human Genome. Nat. Genet. 2007, 39, 457−466. (13) Haerter, I. B.; Sneppen, K. Collaboration between CpG Sites Is Needed for Stable Somatic Inheritance of DNA Methylation States. Nucleic Acids Res. 2014, 42, 2235−2244. (14) Bruno, S.; Williams, R. J.; Del Vecchio, D. Epigenetic Cell Memory: The Gene’s Inner Chromatin Modification Circuit. PLoS Comput. Biol. 2022, 18, No. e1009961. (15) Long, H. K.; King, H. W.; Patient, R. K.; Odom, D. T.; Klose, R. J. Protection of CpG Islands from DNA Methylation Is DNA- Encoded and Evolutionarily Conserved. Nucleic Acids Res. 2016, 44, 6693−6706. (16) Takahashi, Y.; Wu, J.; Suzuki, K.; Martinez-Redondo, P.; Li, M.; Liao, H.-K.; Wu, M.-Z.; Hernández-Benítez, R.; Hishida, T.; Shokhirev, M. N.; Esteban, C. R.; Sancho-Martinez, I.; Belmonte, J. C. I. Integration of CpG-Free DNA Induces de Novo Methylation of CpG Islands in Pluripotent Stem Cells. Science 2017, 356, 503−508. (17) Singer, Z. S.; Yong, J.; Tischler, J.; Hackett, J. A.; Altinok, A.; Surani, M. A.; Cai, L.; Elowitz, M. B. Dynamic Heterogeneity and J. O.; Lövkvist, C.; Dodd, 2543 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article J.; Dadon, D.; Young, R. A.; DNA Methylation in Embryonic Stem Cells. Mol. Cell 2014, 55, 319− 331. (18) Bintu, L.; Yong, J.; Antebi, Y. E.; McCue, K.; Kazuki, Y.; Uno, N.; Oshimura, M.; Elowitz, M. B. Dynamics of Epigenetic Regulation at the Single-Cell Level. Science 2016, 351, 720−724. (19) Ziller, M. J.; Gu, H.; Müller, F.; Donaghey, J.; Tsai, L. T.-Y.; Kohlbacher, O.; De Jager, P. L.; Rosen, E. D.; Bennett, D. A.; Bernstein, B. E.; Gnirke, A.; Meissner, A. Charting a Dynamic DNA Methylation Landscape of the Human Genome. Nature 2013, 500, 477−481. (20) Kungulovski, G.; Jeltsch, A. Epigenome Editing: State of the Art, Concepts, and Perspectives. Trends Genet. 2016, 32, 101−113. (21) Nakamura, M.; Gao, Y.; Dominguez, A. A.; Qi, L. S. CRISPR Technologies for Precise Epigenome Editing. Nat. Cell Biol. 2021, 23, 11−22. (22) Van, M. V.; Fujimori, T.; Bintu, L. Nanobody-Mediated Control of Gene Expression and Epigenetic Memory. Nat. Commun. 2021, 12, 537. (23) Liu, X. S.; Wu, H.; Ji, X.; Stelzer, Y.; Wu, X.; Czauderna, S.; Shu, Jaenisch, R. Editing DNA Methylation in the Mammalian Genome. Cell 2016, 167, 233− 247.e17. (24) Nuñez, J. K.; Chen, J.; Pommier, G. C.; Cogan, J. Z.; Replogle, J. M.; Adriaens, C.; Ramadoss, G. N.; Shi, Q.; Hung, K. L.; Samelson, A. J.; Pogson, A. N.; Kim, J. Y. S.; Chung, A.; Leonetti, M. D.; Chang, H. Y.; Kampmann, M.; Bernstein, B. E.; Hovestadt, V.; Gilbert, L. A.; Weissman, J. S. Genome-Wide Programmable Transcriptional Memory by CRISPR-Based Epigenome Editing. Cell 2021, 184, 2503−2519.e17. (25) Park, M.; Patel, N.; Keung, A. J.; Khalil, A. S. Engineering Epigenetic Regulation Using Synthetic Read-Write Modules. Cell 2019, 176, 227−238.e20. (26) Urlinger, S.; Baron, U.; Thellmann, M.; Hasan, M. T.; Bujard, H.; Hillen, W. Exploring the Sequence Space for Tetracycline- Dependent Transcriptional Activators: Novel Mutations Yield Expanded Range and Sensitivity. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 7963−7968. (27) Chen, T.; Tsujimoto, N.; Li, E. The PWWP Domain of Dnmt3a and Dnmt3b Is Required for Directing DNA Methylation to the Major Satellite Repeats at Pericentric Heterochromatin. Mol. Cell. Biol. 2004, 24, 9048−9058. (28) Zacharias, D. A.; Violin, J. D.; Newton, A. C.; Tsien, R. Y. Partitioning of Lipid-Modified Monomeric GFPs into Membrane Microdomains of Live Cells. Science 2002, 296, 913−916. (29) Yamaguchi, S.; Kazuki, Y.; Nakayama, Y.; Nanba, E.; Oshimura, M.; Ohbayashi, T. A Method for Producing Transgenic Cells Using a Multi-Integrase System on a Human Artificial Chromosome Vector. PLoS One 2011, 6, No. e17267. (30) Argentova, V. V.; Aliev, T. K.; Toporova, V. A.; Rybchenko, V. S.; Dolgikh, D. A.; Kirpichnikov, M. P. Studies on the Influence of Different Designs of Eukaryotic Vectors on the Expression of Recombinant IgA. Moscow Univ. Biol. Sci. Bull. 2017, 72, 63−68. (31) Oliveira, D. S. L. D.; Paredes, V.; Caixeta, A. V.; Henriques, N. M.; Wear, M. P.; Albuquerque, P.; Felipe, M. S. S.; Casadevall, A.; Nicola, A. M. Hinge Influences in Murine IgG Binding to Cryptococcus Neoformans Capsule. Immunology 2022, 165, 110− 121. (32) Fu, X.; Zhu, J.; Duan, Y.; Lu, P.; Zhang, K. CRISPR/Cas9 mediated somatic gene therapy for insertional mutations: the vibrator mouse model. Precis. Clin. Med. 2021, 4, 168−175. (33) Sayers, E. W.; Bolton, E. E.; Brister, J. R.; Canese, K.; Chan, J.; Comeau, D. C.; Connor, R.; Funk, K.; Kelly, C.; Kim, S.; Madej, T.; Marchler-Bauer, A.; Lanczycki, C.; Lathrop, S.; Lu, Z.; Thibaud- Nissen, F.; Murphy, T.; Phan, L.; Skripchenko, Y.; Tse, T.; Wang, J.; Williams, R.; Trawick, B. W.; Pruitt, K. D.; Sherry, S. T. Database Resources of the National Center for Biotechnology Information. Nucleic Acids Res. 2022, 50, D20−D26. (34) Jia, L.; Mao, Y.; Ji, Q.; Dersh, D.; Yewdell, J. W.; Qian, S.-B. Decoding mRNA Translatability and Stability from the 5’ UTR. Nat. Struct. Mol. Biol. 2020, 27, 814−821. (35) Nowialis, P.; Lopusna, K.; Opavska, J.; Haney, S. L.; Abraham, A.; Sheng, P.; Riva, A.; Natarajan, A.; Guryanova, O.; Simpson, M.; Hlady, R.; Xie, M.; Opavsky, R. Catalytically Inactive Dnmt3b Rescues Mouse Embryonic Development by Accessory and Repressive Functions. Nat. Commun. 2019, 10, 4374. (36) Vaisvila, R.; Ponnaluri, V. C.; Sun, Z.; Langhorst, B. W.; Saleh, L.; Guan, S.; Dai, N.; Campbell, M. A.; Sexton, B. S.; Marks, K.; Samaranayake, M.; Samuelson, J. C.; Church, H. E.; Tamanaha, E.; Corrêa, I. R., Jr; Pradhan, S.; Dimalanta, E. T.; Evans, T. C., Jr; Williams, L.; Davis, T. B. Enzymatic Methyl Sequencing Detects DNA Methylation at Single-Base Resolution from Picograms of DNA. Genome Res. 2021, 31, 1280−1289. (37) Fenouil, R.; Cauchy, P.; Koch, F.; Descostes, N.; Cabeza, J. Z.; Innocenti, C.; Ferrier, P.; Spicuglia, S.; Gut, M.; Gut, I.; Andrau, J.-C. CpG Islands and GC Content Dictate Nucleosome Depletion in a Transcription-Independent Manner at Mammalian Promoters. Genome Res. 2012, 22, 2399−2408. (38) Weirauch, M. T.; Yang, A.; Albu, M.; Cote, A. G.; Montenegro- Montero, A.; Drewe, P.; Najafabadi, H. S.; Lambert, S. A.; Mann, I.; Cook, K.; Zheng, H.; Goity, A.; van Bakel, H.; Lozano, J.-C.; Galli, M.; Lewsey, M. G.; Huang, E.; Mukherjee, T.; Chen, X.; Reece- Hoyes, J. S.; Govindarajan, S.; Shaulsky, G.; Walhout, A. J. M.; Bouget, F.-Y.; Ratsch, G.; Larrondo, L. F.; Ecker, J. R.; Hughes, T. R. Determination and Inference of Eukaryotic Transcription Factor Sequence Specificity. Cell 2014, 158, 1431−1443. (39) Kondratova, A.; Watanabe, T.; Marotta, M.; Cannon, M.; Segall, A. M.; Serre, D.; Tanaka, H. Replication Fork Integrity and Intra-S Phase Checkpoint Suppress Gene Amplification. Nucleic Acids Res. 2015, 43, 2678−2690. (40) Nagaraj, N.; Wisniewski, J. R.; Geiger, T.; Cox, J.; Kircher, M.; Kelso, J.; Pääbo, S.; Mann, M. Deep Proteome and Transcriptome Mapping of a Human Cancer Cell Line. Mol. Syst. Biol. 2011, 7, 548. (41) Zeng, H.; Kaul, S.; Simons, S. S., Jr. Genomic Organization of Human GMEB-1 and Rat GMEB-2: Structural Conservation of Two Multifunctional Proteins. Nucleic Acids Res. 2000, 28, 1819−1829. (42) Spruijt, C. G.; Vermeulen, M. DNA Methylation: Old Dog, New Tricks? Nat. Struct. Mol. Biol. 2014, 21, 949−954. (43) Deaton, A. M.; Bird, A. CpG Islands and the Regulation of Transcription. Genes Dev. 2011, 25, 1010−1022. (44) Wu, H.; D’Alessio, A. C.; Ito, S.; Xia, K.; Wang, Z.; Cui, K.; Zhao, K.; Eve Sun, Y.; Zhang, Y. Dual Functions of Tet1 in Transcriptional Regulation in Mouse Embryonic Stem Cells. Nature 2011, 473, 389−393. (45) Farcas, A. M.; Blackledge, N. P.; Sudbery, I.; Long, H. K.; McGouran, J. F.; Rose, N. R.; Lee, S.; Sims, D.; Cerase, A.; Sheahan, T. W.; Koseki, H.; Brockdorff, N.; Ponting, C. P.; Kessler, B. M.; Klose, R. J. KDM2B Links the Polycomb Repressive Complex 1 (PRC1) to Recognition of CpG Islands. Elife 2012, 1, No. e00205. (46) Maeder, M. L.; Angstman, J. F.; Richardson, M. E.; Linder, S. J.; Cascio, V. M.; Tsai, S. Q.; Ho, Q. H.; Sander, J. D.; Reyon, D.; Bernstein, B. E.; Costello, J. F.; Wilkinson, M. F.; Joung, J. K. Targeted DNA Demethylation and Activation of Endogenous Genes Using Programmable TALE-TET1 Fusion Proteins. Nat. Biotechnol. 2013, 31, 1137−1142. (47) Garber, K. B.; Visootsak, J.; Warren, S. T. Fragile X Syndrome. Eur. J. Hum. Genet. 2008, 16, 666−672. (48) Ajjugal, Y.; Kolimi, N.; Rathinavelan, T. Secondary Structural Choice of DNA and RNA Associated with CGG/CCG Trinucleotide Repeat Expansion Rationalizes the RNA Misprocessing in FXTAS. Sci. Rep. 2021, 11, 8163. (49) Coffee, B.; Zhang, F.; Warren, S. T.; Reines, D. Acetylated Histones Are Associated with FMR1 in Normal but Not Fragile X- Syndrome Cells. Nat. Genet. 1999, 22, 98−101. (50) Krueger, F.; Andrews, S. R. Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 2011, 27, 1571−1572. 2544 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545 ACS Synthetic Biology pubs.acs.org/synthbio Research Article (51) Danecek, P.; Bonfield, J. K.; Liddle, J.; Marshall, J.; Ohan, V.; Pollard, M. O.; Whitwham, A.; Keane, T.; McCarthy, S. A.; Davies, R. M.; Li, H. Twelve Years of SAMtools and BCFtools. Gigascience 2021, 10(), giab008. DOI: DOI: 10.1093/gigascience/giab008. 2545 https://doi.org/10.1021/acssynbio.3c00078 ACS Synth. Biol. 2023, 12, 2536−2545
10.1073_pnas.2301987120	RESEARCH ARTICLE \| MICROBIOLOGY OPEN ACCESS A role for the Gram- negative outer membrane in bacterial shape determination Elayne M. Fivensona , Andrea Vettigera, Marios F. Sardisa , Patricia D. A. Rohsa , Grasiela Torresa, Alison Forchoha, and Thomas G. Bernhardta,b,1 Edited by Thomas Silhavy, Princeton University, Princeton, NJ; received February 3, 2023; accepted July 21, 2023 The cell envelope of Gram- negative bacteria consists of three distinct layers: the cytoplasmic membrane, a cell wall made of peptidoglycan (PG), and an asymmetric outer membrane (OM) composed of phospholipid in the inner leaflet and lipopoly- saccharide (LPS) glycolipid in the outer leaflet. The PG layer has long been thought to be the major structural component of the envelope protecting cells from osmotic lysis and providing them with their characteristic shape. In recent years, the OM has also been shown to be a load- bearing layer of the cell surface that fortifies cells against internal turgor pressure. However, whether the OM also plays a role in mor- phogenesis has remained unclear. Here, we report that changes in LPS synthesis or modification predicted to strengthen the OM can suppress the growth and shape defects of Escherichia coli mutants with reduced activity in a conserved PG synthesis machine called the Rod complex (elongasome) that is responsible for cell elongation and shape determination. Evidence is presented that OM fortification in the shape mutants restores the ability of MreB cytoskeletal filaments to properly orient the synthesis of new cell wall material by the Rod complex. Our results are therefore consistent with a role for the OM in the propagation of rod shape during growth in addition to its well- known function as a diffusion barrier promoting the intrinsic antibiotic resistance of Gram- negative bacteria. peptidoglycan \| lipopolysaccharide \| membrane \| morphogenesis \| cell envelope Gram- negative bacteria have a characteristic three- layered cell envelope comprised of an inner (cytoplasmic) membrane (IM), a relatively thin cell wall made of peptidoglycan (PG), and an outer membrane (OM). The OM bilayer is asymmetric with phospholipids in the inner leaflet and the lipopolysaccharide (LPS) glycolipid in the outer leaflet. For many years, the PG layer was thought to be the sole load- bearing component of the envelope with the OM primarily serving to protect Gram- negative cells from external insults like antibiotics (1, 2). However, it has recently become clear that in addition to providing a barrier function, the OM can also help cells resist internal turgor pressure (3). What has remained unknown is whether the OM also partners with the PG layer to define cell shape. Here, we report a genetic analysis of PG synthesis and cell shape determination that supports such a role for the OM. The PG heteropolymer is composed of glycan chains with alternating units of N- acetylglucosamine (GlcNAc) and N- acetylmuramic acid (MurNAc) (4). A short peptide is attached to the MurNAc sugar and is used to cross- link adjacent glycans to form the cell wall matrix. Glycosyltransferases catalyze the polymerization of glycan polymers, whereas transpeptidases perform the cross- linking reaction. There are two major classes of PG synthases: class A Penicillin Binding Proteins (aPBPs) and complexes formed between SEDS (Shape, Elongation, Division, Sporulation) proteins and class B PBPs (bPBPs) (1, 2, 5). The aPBPs have both enzymatic functions in a single polypeptide, whereas in the SEDS- bPBP complexes, the SEDS protein promotes glycan polymerization and the bPBP provides the cross- linking activity (6–9). The SEDS- bPBP complexes RodA- PBP2 (6–8, 10) and FtsW- FtsI (9) play essential roles in rod shape determination and cell division, respectively. In both cases, these synthases are part of larger multiprotein assemblies involving cytoskeletal filaments. The rod shape– determining system is called the Rod complex (a.k.a. the elongasome). It promotes the elongation of bacilli and maintains their characteristic rod shape. In addition to RodA- PBP2, the complex includes filaments of the actin- like MreB protein along with three membrane proteins of poorly understood function: MreC, MreD, and RodZ (11–18). The Rod complex has been observed to dynamically rotate around the long axis of the cell as it deposits new PG material to promote cell elongation. PG synthesis is required for the motion and MreB filaments are thought to orient it orthogonally to the long cell axis via a rudder- like mechanism (1, 7, 19–22). Significance The cell wall has traditionally been thought to be the main structural determinant of the bacterial cell envelope that resists internal turgor and determines cell shape. However, the outer membrane (OM) has recently been shown to contribute to the mechanical strength of Gram- negative bacterial envelopes. Here, we demonstrate that changes to OM composition predicted to increase its load- bearing capacity rescue the growth and shape defects of Escherichia coli mutants defective in the major cell wall synthesis machinery that determines rod shape. Our results therefore reveal a previously unappreciated role for the OM in bacterial shape determination in addition to its well- known function as a diffusion barrier that protects Gram- negative bacteria from external insults like antibiotics. Author affiliations: aDepartment of Microbiology, Blavatnik Institute, Harvard Medical School, Boston, MA 02115; and bHHMI, Chevy Chase, MD 20815 Author contributions: E.M.F., P.D.A.R., A.V., and T.G.B. designed research; E.M.F., P.D.A.R., A.V., M.F.S., G.T., and A.F. performed research; E.M.F., P.D.A.R., A.V., M.F.S., G.T., A.F., and T.G.B. analyzed data; and E.M.F. and T.G.B. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2301987120/- /DCSupplemental. Published August 22, 2023. PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 1 of 11 To better understand the Rod complex function, we previously identified nonfunctional variants of MreC in Escherichia coli and selected for suppressor mutations that overcame their shape and viability defects (10, 23). One major class of suppressors encoded hypermorphic variants of PBP2 and RodA that provided impor- tant insight into the mechanism of Rod complex activation and the regulation of SEDS proteins (10). Genetic, structural, and cytological evidence suggests that MreC activates the complex by inducing a conformational change in PBP2, which in turn acti- vates RodA, shifting the complex from an inactive to an active state (10). The role of MreD in the complex is not clear (23, 24). The signals that promote Rod complex activation also remain unknown, but the mechanism may involve the recognition of landmarks in the PG matrix by PBP2 (25). In this report, we study a class of suppressors that restore the growth and shape of mreC hypomorphs. Instead of activating the Rod complex directly, these suppressors function by increasing the production of LPS. Further analysis of the suppression mech- anism revealed that Rod complex mutants are impaired for LPS production. Additionally, we found that modifications to LPS predicted to stiffen the OM restore rod shape in cells defective for MreC by promoting the feedback mechanism via which MreB orients PG synthesis. Thus, our results suggest a potential connec- tion between Rod complex activity and LPS synthesis and argue for a morphogenic role for the OM. Results Increased LPS Synthesis Suppresses a Rod Complex Defect. Cells with mreC(R292H) or mreC(G156D) mutations produce stable MreC protein capable of inducing a dominant- negative growth and shape phenotype (10, 23). Therefore, the altered proteins are likely capable of joining the Rod complex but are defective in stimulating its activity. Mutants with these alleles at the native locus as their sole copy of mreC can be maintained as spheres on minimal medium (M9), but they fail to grow on rich medium (LB). We selected for spontaneous suppressors that restored the growth of these mutants on LB along with their rod shape. In addition to mutants encoding altered PBP2 and RodA described previously (10), the selection also identified suppressors in the ftsH and lapB(yciM) genes encoding regulators of LPS synthesis (Fig. 1 and SI Appendix, Table S1). FtsH is an IM metalloprotease that along with its adapter protein LapB (26, 27) degrades LpxC (UDP- 3- O- acyl- N- acetylglucosamine deacetylase) (28–30), the enzyme that catalyzes the first committed step in LPS synthesis (31, 32). Proteolysis of LpxC is in turn regulated by the essential inner membrane protein YejM (PbgA, LapC), which functions to inhibit LapB activity in a manner that is sensitive to the concentration of LPS in the IM (30, 33–38). When the steady- state concentration is low due to LPS synthesis being balanced with its transport to the OM, YejM blocks LpxC turnover (Fig. 1 A, Top). However, when LPS synthesis outpaces its transport, YejM is inhibited by the buildup of LPS in the inner membrane and LpxC turnover is increased to restore homeostasis (Fig. 1 A, Bottom). Both ftsH suppressors encoded protease variants with substitu- tions in the periplasmic loop of the protein (Fig. 1). One was found as a suppressor of mreC(R292H) and the other as a sup- pressor of mreC(G156D) (Fig. 1B and SI Appendix, Table S1). A mutation in lapB encoding a protein with a deletion of the last eleven C- terminal amino acids was also isolated as a suppressor of mreC(G156D) (Fig. 1B and SI Appendix, Table S1). Although growth rate and morphology were not restored to completely match those of wild- type cells, the suppressors supported full plat- ing efficiency of their respective mreC mutant on LB (Fig. 1B) and switched their morphology from sphere- like to elongated rods (Fig. 1C). Suppression was not allele specific as the ftsH(V41G) mutation originally isolated as a suppressor of mreC(G156D) (SI Appendix, Table S1) also suppressed the growth and shape defects of mreC(R292H) (Fig. 2 A and B). We chose to further characterize the mechanism of suppression by the ftsH(V41G) allele by determining its effect on the cellular concentration of LpxC (Fig. 2C) and LPS (Fig. 2D). In cells with wild- type FtsH, mutants encoding defective MreC variants had decreased levels of both LpxC (Fig. 2C) and LPS (Fig. 2D and SI Appendix, Fig. S1) compared to cells with MreC(WT). We also observed a similar dose- dependent decrease in LpxC levels following treatment with the MreB inhibitor A22, suggesting that a decrease in OM synthesis is not specific to the mreC mutants but instead is a more general response to perturbations to Rod complex activity (SI Appendix, Fig. S2). The ftsH(V41G) allele increased LpxC and LPS levels in cells with either hypo- morphic allele of mreC (Fig. 2 C and D and SI Appendix, Fig. S1). This change resulted in an increase in LPS concentration to near normal in cells with the defective MreC variants (Fig. 2 C and D and SI Appendix, Fig. S1). RNAseq analysis showed that lpxC transcript levels remain unchanged in mreC mutants compared to WT, suggesting that the decrease in LpxC levels results from a change in posttranscriptional control (SI Appendix, Fig. S3). This finding is consistent with previous results showing that lpxC transcript levels remain the same under conditions when LpxC is stabilized, suggesting that LpxC production is not regulated at the transcriptional level (40). Although technical challenges resulting from the extremely low levels of LpxC in the mreC mutants prevented us from measuring its half- life in these strains, we think it is reasonable to conclude that the ftsH(V41G) allele is hypomorphic and likely leads to reduced LpxC turnover and a rise in LPS levels that compensates for the apparent defect in LPS synthesis of the mreC mutants. To determine whether an increase in LPS synthesis is sufficient to suppress the defective mreC alleles, we overexpressed lpxC in the mutants (Fig. 3). Overproduction of LpxC indeed promoted the growth of mreC(R292H) and mreC(G156D) mutants on LB and restored an elongated rod- like shape (Fig. 3). However, sup- pression was not as robust as that promoted by the ftsH(V41G) allele (Figs. 2 and 3), suggesting either that the levels of LPS upon LpxC overproduction were too high and caused mild toxicity or that changes in the turnover of FtsH substrates other than LpxC contribute to the suppressing activity of ftsH(V41G). Suppression was dependent on LpxC activity as the overproduction of a cat- alytically defective LpxC that lacks a degradation signal (desig- nated as ∆C5) (29, 41, 42) failed to promote the elongated growth of cells producing the MreC variants (Fig. 3). Notably, overexpression of lpxC did not suppress an mreC deletion (Fig. 3), arguing that partial Rod complex activity in the mreC(R292H) and mreC(G156D) mutants is required to promote rod shape under suppressing conditions. Overall, our results suggest that the growth and shape defects of the mreC(R292H) and mreC(G156D) mutants is not just due to problems with PG biogenesis. Surprisingly, improper LPS synthesis and OM bio- genesis also appear to be contributing factors. Notably, these findings explain previous reports of OM defects in mutants defec- tive for the Rod system (11). Accordingly, we found that even under the permissive growth condition (minimal media, 30 °C), the mreC mutants are sensitive to a range of antibiotics indicating compromised OM barrier function (SI Appendix, Fig. S4). mreC Mutants Remain Capable of Sensing Perturbations to LPS Synthesis. One explanation for the decrease in LPS production observed in the mreC mutants is that these cells are 2 of 11 https://doi.org/10.1073/pnas.2301987120 pnas.org Fig. 1. Mutations in factors involved in LpxC turnover rescue mreC hypomorphs. (A) Schematic overview of LpxC regulation. Top: When LPS levels are low, YejM interacts with LapB, sequestering it from the FtsH protease, leading to the stabilization of LpxC and increased LPS synthesis. Bottom: When LPS levels are high, LPS accumulates in the outer leaflet of the inner membrane. YejM binds to LPS, allowing LapB to interact with FtsH and target LpxC for degradation, reducing LPS synthesis. (B) WT (HC555), mreC(R292H) (PR5), mreC(R292H) ftsH(V37G) (PR82), mreC(G156D) (PR30), mreC(G156D) lapB(∆379- 389) (PR86), mreC(G156D) ftsH(V41G) (PR88) were cultured for 24 h in minimal medium (M9 + CAA + glu) at 30 °C. Cultures were then normalized to OD600 = 1 and serially diluted and spotted onto LB and M9 + CAA + glu plates. LB plates were incubated for 16 h at 37 °C and M9 plates were incubated for 40 h at 30 °C. Dilution factors are indicated above the spot dilutions. (C) Micrographs of WT (HC555), mreC(R292H) (PR5), mreC(R292H) ftsH(V37G) (PR82), mreC(G156D) (PR30), mreC(G156D) lapB(∆379- 389) (PR86), mreC(G156D) ftsH(V41G) (PR88). Strains were grown overnight in minimal medium (M9 + CAA + glu) at 30 °C. Overnight cultures were then back diluted to OD600 = 0.05 in minimal medium and incubated shaking at 30 °C until OD600 = 0.3- 0.4. Cells were then spun down and resuspended in LB to an OD600 of 0.025 and incubated at 37 °C until OD600 = 0.3- 0.4. Cells were then fixed and imaged. Aspect ratios were analyzed using the FIJI plugin MicrobeJ (39). (Scale bar, 5 µm.) n= 100 cells per group. Statistical significance determined using an unpaired t test with Welch’s correction (not assuming equal SDs). defective in modulating LpxC stability through the YejM/LapB/ FtsH pathway in response to reduced LPS levels (30, 33–38). To test this possibility, we monitored LpxC levels following the overproduction of a hyperactive allele of fabZ (3- hydroxy- acyl- [acyl- carrier- protein] dehydratase) (29), an enzyme that functions early in the phospholipid synthesis pathway (43). Overproduction of this enzyme is expected to increase the flux of common precursors into the phospholipid synthesis pathway at the expense of LPS synthesis. Cells harboring the hyperactive fabZ(L85P) allele were previously reported to have increased levels of LpxC, presumably due to LpxC stabilization in order to restore balance between the two lipid biosynthesis pathways (29, 44). We found that mreC(R292H) cells overexpressing fabZ(L85P) had increased levels of LpxC compared to the uninduced controls and that the magnitude of the increase was comparable to that in WT cells upon induction of the hyperactive fabZ allele (Fig. 3C). We observed a similar result when we treated mreC(R292H) cells with the LpxC inhibitor CHIR- 090 (45, 46), which was also previously shown to promote LpxC stabilization (47) (SI Appendix, Fig. S5). Thus, mreC mutant cells remain capable of sensing an acute reduction in LPS synthesis but fail to respond to and correct their chronic deficit in LpxC and LPS levels. OM Modifications Associated with Increased Stiffness Suppress Cell Shape Defects. We reasoned that increasing LPS synthesis could suppress the shape defect of mreC mutants either by activating the Rod complex similar to previously characterized suppressors in rodA and mrdA encoding RodA- PBP2 (10) or by altering the structural properties of the OM. To test the former possibility, we measured the effect of the ftsH(V41G) allele on Rod complex activity in vivo using a radiolabeling assay. For this assay, a genetic background is used where PG synthesis by the divisome and the aPBPs can be inhibited by SulA production (48–50) and (2- sulfonatoethyl) methanethiosulfonate (MTSES) PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 3 of 11 Fig. 2. FtsH(V41G) increases LpxC and LPS levels in mreC hypomorphs. (A) Cultures of WT(EMF196), mreC(G156D) (EMF197), mreC(R292H) (PR109), ftsH(V41G) (EMF199), mreC(G156D) ftsH(V41G) (PR111), mreC(R292H) ftsH(V41G) (PR110) were incubated in M9 + CAA + glu at 30 °C for 24 h. Cultures were diluted and plated as in Fig. 1. (B) Cultures of the strains listed in (A) were diluted to OD600 = 0.05 in M9 + CAA + glu and incubated at 30 °C until OD600 = 0.2- 0.3. Cultures were gently spun down and resuspended in LB to an OD600 = 0.025 and incubated at 37 °C until OD = 0.2- 0.3. Cells were fixed and imaged (Methods). Aspect ratios were analyzed using the FIJI plugin MicrobeJ (39). (Scale bar, 5 µm.) n = 100 cells per group. Statistical significance was determined as in Fig. 1. (C) Cultures of the strains listed in (A) were grown as described in (B) and an immunoblot for LpxC was performed. (D) Cultures of the strains listed in (A) were grown as described in (B) and analyzed via silver stain for lipid A- core (Top). Samples were normalized to total protein and an immunoblot for RpoA was performed to serve as a loading control. treatment (51), respectively. Rod complex activity can be further isolated by treatment with the PBP2- specific inhibitor mecillinam. This drug blocks the cross- linking activity of PBP2, but the glycosyltransferase activity of RodA remains active, leading to an accumulation of uncross- linked glycan chains. These uncross- linked glycans are known to be rapidly degraded by the lytic transglycosylase Slt (51). Thus, the accumulation of nascent PG turnover products during radiolabeling in the presence of mecillinam, MTSES, and SulA can be used as an indirect measure of Rod complex activity. Unlike the suppressing RodA and PBP2 variants characterized previously (10) that activate nascent PG turnover product accumulation, the ftsH(V41G) allele did not significantly alter Rod complex activity as assessed by the turnover assay (SI Appendix, Fig. S6). Furthermore, the activated PBP2(L61R) variant was found to increase the resistance of cells to the MreB inhibitor A22, another indication of its ability to activate the Rod complex. By contrast, overexpression of lpxC did not increase resistance to A22 (SI Appendix, Fig. S7). Taken together, these results suggest that hyperactivation of LPS synthesis does not suppress the shape and growth defects of mreC mutants by enhancing the PG synthesis activity of the Rod complex. To investigate whether the mechanical stabilization of the OM is the underlying mechanism by which increased LPS synthesis restores shape to the mreC mutants, we sought alternative ways to alter OM stiffness. EDTA strips the OM of magnesium ions, disrupting the lateral interactions between adjacent LPS molecules (52), which has been shown to reduce cell envelope stiffness (3). The addition of EDTA reverses the growth benefit of lpxC over- expression in mreC(R292H) and mreC(G156D) mutants (Fig. 3A). We therefore conclude that LPS packing in the OM is required for the overexpression of lpxC to improve the growth of mreC hypormorphs (Fig. 3). We next investigated the effect of increasing OM stiffness by reintroducing O- antigen in the mreC mutants. LPS is composed of three covalently attached units (53). The base glycolipid is called Lipid A. It is modified by a core oligosaccharide that is conserved among Gram- negative organisms. The core is further modified by longer polysaccharide chains called O- antigens, the composition of which varies between species. Laboratory strains of E. coli K- 12 do not synthesize O- antigen due to an insertion element in wbbL (54), a gene required for producing LPS modified by the O- 16 O- antigen serotype (52). It was previously reported that restoring O- antigen to the OM dramatically increases its stiffness (3). We therefore asked if reintroducing wild- type wbbL to the mreC mutants on an arabinose- inducible plasmid could suppress their growth and shape phenotypes like the overexpression of lpxC (Fig. 4A). Expression of wbbL but not a lacZ control promoted growth of the mreC hypomorphs under the nonpermissive con- dition (LB, 37 °C) and restored their growth as elongated rods (Fig. 4 B and C). Importantly, we did not observe an increase 4 of 11 https://doi.org/10.1073/pnas.2301987120 pnas.org Fig. 3. The overexpression of lpxC restores growth and partially restores shape to mreC hypomorphs. (A) WT (HC555), mreC(G156D) (PR30), mreC(R292H) (PR5), and ∆mreC (EMF150) expressing WT lpxC (pPR111) or lpxC(H265A)∆C5 (pPR115) from an IPTG- inducible plasmid were cultured for 24 h at 30 °C in M9 + CAA + glu. Cultures were diluted and plated on the indicated media as in Fig. 1. All plates contained CM. M9 plates were incubated at 30 °C for 40 h and LB plates were incubated at 30 °C for 24 h. (B) The strains listed in (A) were grown for 24 h at 30 °C in M9 + CAA + glu + CM. Cultures were diluted to OD600 = 0.025 in M9 + CAA + glu + CM + 50 µM IPTG and incubated at 30 °C until OD600 = 0.2- 0.3. Cells were gently pelleted and resuspended in LB + CM + 50 µM IPTG and grown at 37 °C for 1 h 45 min. Cells were then fixed and imaged (Methods). Aspect ratios were analyzed using the FIJI plugin MicrobeJ (39). (Scale bar, 5 µm.) n = 100 cells per group. Statistical significance was determined as in Fig. 1. (C) Immunoblot for LpxC. Cell lysates of WT (HC555) and mreC(R292H) (PR5) cells harboring plasmids expressing fabZ(L85P) from an IPTG- inducible promoter (pEMF137) were cultured in M9 + CAA + glu + CM at 30 °C for 24 h. Cultures were then diluted to OD600 = 0.025 in M9 + CAA + glu + CM and grown at 30 °C until OD600 = 0.2- 0.3. Cells were gently pelleted and resuspended in LB + CM ± IPTG as indicated and grown at 37 °C for 2 h and were subsequently harvested via centrifugation and processed for immunoblotting. PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 5 of 11 LpxC levels in WT or mreC(R292H) cells expressing wbbL com- pared to the lacZ control (SI Appendix, Fig. S8A), indicating that wbbL is not acting by directly increasing LPS synthesis but rather is improving the structure of the envelope by increasing lateral interactions between adjacent LPS molecules in the OM. We also observed that wbbL can partially rescue the shape defects of mreC(R292H) cells when expressed at the native locus, although not to the same extent as expression from a multicopy plasmid (SI Appendix, Fig. S8 B and C). This suppression phenotype is further improved by overexpressing lpxC, suggesting both methods of increasing OM stiffness have an additive effect on cell shape improvement. As we observed with cells overexpressing lpxC, over- expressing wbbL did not improve the shape or growth defects of ∆mreC cells even though they synthesized comparable levels of Fig. 4. Synthesis of O- antigen- modified LPS suppresses the growth and shape defects of mreC hypomorphs. (A) Schematic of strains. The wbbL gene in E. coli K- 12 is disrupted by an insertion element, preventing the synthesis of O- antigen. wbbL is expressed in trans from an arabinose (ara)- inducible promoter, restoring O- antigen synthesis. lacZ is expressed as a control. (B) WT (HC555), mreC(R292H) (PR5), mreC(G156D) (PR30), and ∆mreC (EMF150) expressing wbbL (pEMF130) or lacZ (pEMF134) from an arabinose- inducible promoter were incubated for 24 h in M9 + CAA + glu + tet at 30 °C. Cultures were diluted and plated on the indicated media as in Fig. 1. (C) The strains listed in (A) were grown for 24 h in M9 + CAA + glu + tet at 30 °C and diluted to OD600 = 0.05 in M9 +CAA + ara + tet for 3 h at 30 °C. After 3 h, the cultures were gently pelleted and resuspended in LB + tet + ara. Cells were grown for 2 h at 37 °C. Cells were then fixed and imaged (Methods). Aspect ratios were analyzed using the FIJI plugin MicrobeJ (39). (Scale bar, 5 µm.) n= 100 cells per group. Statistical significance was determined as in Fig. 1. (D) Proemerald Q stain of LPS. The strains listed in (A) were grown as described in (B). Cell lysates were prepared and LPS was analyzed via promerald Q straining. Note: in our experience, the proemeraldQ method of detecting LPS allowed for more consistent visualization of high molecular weight O- antigen modified species compared the silver stain method used in Fig. 2. 6 of 11 https://doi.org/10.1073/pnas.2301987120 pnas.org O- antigen- LPS as the other strains (Fig. 4D). Restoring O- antigen synthesis also did not restore shape to cells deleted for rodZ (SI Appendix, Fig. S9). Therefore, an intact Rod complex is required to mediate the growth and shape changes in mutant cells with a restored O- antigen. From these results, we infer that OM stiffening is the likely mechanism by which changes in LPS syn- thesis or modification restores rod shape to cells with a poorly functioning Rod complex. OM Stiffness and the Directional Motion of MreB Filaments. MreB polymers align along the greatest principal curvature of the cell and are thought to orient the insertion of new PG by the Rod complex perpendicular to the long cell axis via a rudder- like mechanism (55). MreB polymers thus promote growth in a rod shape, but they also require rod shape for their proper alignment. Rod shape is therefore thought to be a self- reinforcing property (21). We reasoned that this rod- shape feedback loop is impaired in the mreC mutants because the reduced activity of the Rod complex fails to build an envelope robust enough to maintain the beginnings of a cylindrical extrusion that can be elongated into a rod via oriented MreB motion. However, strengthening of the OM in the suppressors may overcome this problem by stabilizing the envelope, allowing a partially functional machine to promote the self- enhancing shape determination process. To test this hypothesis, we wanted to track the motion of a functional MreB- mNeon sandwich fusion (SWMreB- mNeon) (7) in mreC hypomorphic cells with and without shape- restoring suppressor mutations. Unfortunately, we were unable to construct strains encoding both the mreC hypomorphic alleles and the SWmreB- mNeon fusion at the native locus because the combination was toxic. Instead, we produced SWMreB- mNeon from the native mreB locus that also contained mreC(WT) and overexpressed the dominant- negative mreC(R292H) allele from a plasmid in cells with or without O- antigen (Fig. 5A). Overexpression of mreC(R292H) caused cells lacking O- antigen to form sphere- like cells, but the shape change was not as dramatic as that observed for cells harboring mreC(R292H) as the sole copy of the gene at the native locus. As expected, rod shape was maintained in O- antigen positive cells overexpressing mreC(R292H). In addition to the differences in shape, the presence of O- antigen also impacted MreB dynamics. Compared to the rod- shaped O- antigen positive cells, cells lacking O- antigen showed a reduction in the number of directionally moving particles and those particles that were moving did not appear to have as consistent of an orientation (Fig. 5 B, C, and E and Movie S1). Particles in the O- antigen positive cells were also less likely to change direction during imaging than those in the cells lacking O- antigen (Fig. 5D). These results argue that the OM contributes to shape determination by providing sufficient envelope stability for MreB- directed PG synthesis to be properly oriented and self- reinforcing. Discussion The OM and PG layers of the Gram- negative envelope share numer- ous connections. Their building blocks are synthesized from com- mon precursors (58–60), and the layers are physically linked by PG- binding proteins anchored in the OM (61–63). Additionally, the insertion of beta- barrel proteins in the OM appears to be spa- tially coordinated with the insertion of new PG material into the mature cell wall matrix (64). Despite these connections, it has only recently been appreciated that the OM plays a role in the mechanical stability of the Gram- negative envelope that rivals that of the cell wall (3, 65–67). Here, we provide evidence that rather than just stiffening the envelope, the OM also plays a critical role in rod shape determination. Additionally, our genetic analysis uncovered an unexpected connection between LPS synthesis and the activity of the Rod complex that elongates the PG matrix, revealing yet another link between the two outermost layers of Gram- negative cells. A morphogenic role for the OM is inferred from the ability of elevated LPS synthesis or O- antigen modification to restore rod- like shape to cells with a partially defective Rod complex. The shape mutants showed a reduced level of LPS and the LPS syn- thesis enzyme LpxC (Fig. 2). The stiffness of the OM is thought to be mediated by the lateral packing of LPS molecules bridged by Mg2+ ions (3). Thus, the OM of the shape defective cells with reduced LPS likely has suboptimal LPS packing and reduced stiff- ness. Increasing LPS synthesis in these cells by stabilizing LpxC or overproducing it is expected to increase the LPS concentration in the OM of these cells, enhancing lateral interactions between LPS molecules to at least partially restore OM mechanical stability (Fig. 3A). Similarly, the addition of O- antigen is likely to stiffen the membrane despite suboptimal LPS levels because the extended glycan chains facilitate long- distance LPS–LPS interactions. How does OM stiffening rescue the Rod complex defect? We propose that it does so by promoting the oriented- synthesis feed- back via which the Rod complex generates rod shape (21) (Fig. 6). A critical feature of this model of shape determination is that rod shape is self- reinforcing due to the curvature preference of MreB filaments that orients them perpendicular to the long cell axis to guide PG synthesis by the Rod complex (55). If the cell wall made by the machinery is not stiff enough to hold the beginnings of a cylindrical shape in the face of turgor pressure, as is likely the case in the mreC mutants, then the feedback loop that elongates the cylinder to generate rod shape cannot be initiated (Fig. 6). This is reminiscent of a similar phenomenon in Gram- positive bacteria with defects in wall teichoic acid (WTA) synthesis (55). Much like LPS, these anionic cell wall polymers have been proposed to stiffen the envelope at least in part through lateral interactions mediated by bridging Mg2+ ions (21). Accordingly, mutants with reduced levels of WTA synthesis can be converted from rods to spheres by removing Mg2+ from the medium (55). Moreover, cell shape can be restored to B. subtills mutants with a partially defec- tive Rod complex by the addition of excess Mg2+(68, 69). Although it remains to be determined whether the mechanism behind shape restoration in this context is based on envelope rigidification or potential effects on the activity of PG cleaving enzymes, the par- allels suggest the attractive possibility that the LPS of Gram- negative bacteria and WTAs of Gram- positive organisms may function similarly to promote cell shape by providing sufficient envelope rigidity to enable the self- reinforcing orientation of PG synthesis by the Rod complex. Given its relevance to antibiotic resistance, the most well- studied role of the OM is as a permeability barrier preventing the entry of bulky and/or hydrophobic drugs. Mutants defective for the Rod complex have been known to have a defective OM permeability barrier for many years (11, 70), but the cause of their increased permeability to antibiotics has been unclear. Our results indicate that the problem is likely caused by a reduction in LPS synthesis in the spherical cells. Whether this reflects a direct or indirect connec- tion between Rod complex activity and the LPS synthesis and/or transport systems is not known. However, the mreC mutants we studied are still capable of responding to reductions in the flux through the LPS synthesis pathway by stabilizing LpxC (Fig. 3C and SI Appendix, Fig. S5). Thus, the defect does not appear to be at the level of the YejM- LapB- FtsH system that monitors the steady- state level of LPS in the outer leaflet of the IM (30, 33–38). One possible model is that LPS synthesis is down- regulated in Rod PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 7 of 11 Fig. 5. MreB dynamics upon Rod complex inactivation by mreC(R292H) in cells with or without O- antigen. (A) Schematic of strains. SWmreB- mNeon cells harbor either wbbL(INS) (AV007) or wbbL+ (EMF210) at the native chromosomal locus, resulting in cells without or with O- antigen- modified LPS, respectively. mreC(R292H) (D) is expressed in trans from an IPTG- inducible promoter (pMS9). (B) wbbL(INS) (AV007) or wbbL+ (EMF210) cells expressing mreC(R292H)D (pMS9). Individual traces of MreB tracks were mapped using the TrackMate feature of FIJI (56, 57). Each track is indicated in a different color. (C) Violin plot of the number of directional MreB tracks per cell area in cells with (EMF210) and without O- antigen (AV007) expressing mreC(R292H)D (pMS9). [n = 30 cells (AV007), n = 31 cells (EMF210)]. Statistical significance determined by an unpaired t test with Welch’s correction. (D) Violin plot of the mean directional change rate of MreB tracks in wbbL (−) and wbbL (+) cells [n = 10,214 tracks (AV007), n = 9,162 tracks (EMF210)]. Statistical significance determined by the Mann–Whitney test. (E) Histogram of the log- log fit (α) values of Individual MreB traces in cells with (EMF210) and without O- antigen (AV007) expressing mreC(R292H)D (pMS9). [n = 18,618 tracks (AV007), n = 15,070 tracks (EMF210)]. Statistical significance determined by the Mann–Whitney test. complex mutants in order to direct common precursors toward PG synthesis in an attempt to restore cell wall integrity. In this case, the cell may be triaging PG synthesis at the expense of the OM. Our results indicate that reduced LPS synthesis in this case is not due to differential transcriptional regulation of lpxC. Further study will be needed to determine whether the regulation works through the YejM- LapB- FtsH pathway or through a different mechanism and to understand if and how the status of the cell wall is sensed as part of the regulatory systems governing OM biogenesis. Notably, several studies have also recently made connections between PG and OM synthesis in Pseudomonas aeruginosa (71), Acinetobacter baumannii (72), and Vibrio cholera (73). Future investigation of these and other PG- OM connections promises to reveal new ways to compromise the permeability barrier of diverse Gram- negative bacteria to sensi- tize them to antibiotics. Methods Bacterial Strains and Growth Conditions. The strains generated and used in this study are derivatives of MG1655 and cultured in LB (1% tryptone, 0.5% yeast extract, 0.5% NaCl) or minimal (M9) medium (74). Minimal medium was sup- plemented with 0.2% Casamino Acids (CAA) and 0.2% glucose (glu) or arabinose (ara) where indicated (see figure legends). Rod complex mutants and controls were maintained on M9 + CAA + glu at 30 °C unless otherwise indicated. Strains harboring plasmids were grown in the presence of antibiotics at the following concentrations (unless indicated differently in the figure legends): 25 µg/mL chlo- ramphenicol (CM), 25 µg/mL kanamycin (Kan), and 10 µ/mL tetracycline (Tet). All strains, plasmids, and primers used in this study are listed in SI Appendix, Tables S2, S3, and S4, respectively. For details, please see SI Appendix, Supporting text. Suppressor Analysis. Suppressors were isolated and analyzed as described previously (10). Western Blots. Cells were pelleted via centrifugation and resuspended in water and 2× Laemmli sample buffer (100 mM Tris- HCl, pH 6.8; 2% SDS; 0.1% bromo- phenol blue; 20% glycerol) at a 1:1 ratio to a final OD600 of 20, boiled for 10 min, and stored at −80 °C. Samples were thawed and sonicated for 1 min twice using a Qsonica tip sonicator with an amplification of 25%. Sample concentration was determined using the Noninterfering (NI) Protein Assay [with bovine serum albu- min (BSA) protein standard] (G Biosciences catalog no. 786- 005). Samples were run on a 15% polyacrylamide gel (LpxC western blots) or 4 to 20% Mini- PROTEAN 8 of 11 https://doi.org/10.1073/pnas.2301987120 pnas.org Fig. 6. Interventions that strengthen the OM restore shape to Rod complex hypomorphs. (A) In wild- type cells, the internal turgor pressure of the cell is countered by the combined mechanical strength of the cell wall and the OM. The Rod complex is fully functional and is orientated by MreB, which aligns along the greatest principle curvature to ensure synthesis perpendicular to the long axis of the cell. (B) In hypomorphic mreC mutants (mreC*), the Rod complex is not able to synthesize sufficient PG or LPS, weakening the envelope and leading to loss of rod shape. The cells no longer form a clearly defined long axis, causing MreB filaments to misalign. The reduced Rod complex activity in these mutants is therefore not properly oriented. (C) When the mechanical strength of the OM is increased, the cell envelope is sufficiently able to resist the internal turgor pressure of the cell to allow for the initiation and propagation of a rod shape by allowing MreB and limited PG synthesis by the Rod complex to properly orient. gels (BioRad cat# 4568095) and transferred to a polyvinylidene difluoride mem- brane. The membrane was rinsed in phosphate- buffered saline containing 0.1% Tween (PBS- T) [10% 10× PBS- T buffer, pH 7.4 (Sigma- Aldrich)] and blocked in 5% milk in PBS- T for 1.5 h. The membrane was incubated in 1% milk- PBS- T containing rabbit anti- LpxC antibody (a generous gift from the Doerrler lab) or mouse anti- RpoA (anti- E. coli RNA polymerase alpha from Biolegend, cat# 663104) diluted 1:10,000. The membranes were incubated at 4 °C O/N rocking and then washed 4× with PBS- T at room temperature (1× quickly followed by 3× for 10 min). For LpxC blots, the membrane was incubated in 0.2% milk dissolved in PBS- T with [HRP]- conjugated anti- rabbit IgG (1:40,000 dilution, Rockland cat# 18–8816- 33). For RpoA western blots, membranes were incubated with anti- mouse IgG HRP at a dilution of 1:3,000 (Thermo Fisher Scientific catalog no. 34577). Membranes were incubated with secondary antibody for 2 h and then washed 5× with PBS- T (1× quickly followed by 4× for 10 min per wash). Membranes were developed using the SuperSignal West Pico Plus chemiluminescent substrate (Thermo Fisher Scientific catalog no. 34577) and imaged using the c600 Azure Biosystems platform. Detecting LPS Using Silver Stain. Cultures were prepared as described in figure legends. For Fig. 2 and SI Appendix, Fig. S1, strains listed in the figure legend were cultured for 24 h at 30 °C in M9 + CAA + glu. Cultures were then diluted to OD600 = 0.05 and grown at 30 ˚C until OD = 0.2- 0.3. Cells were gently pelleted and resuspended in LB (OD600 = 0.025) and grown at 37 °C until OD600 = 0.2- 0.3. Cells were pelleted and resuspended in 1× LDS sample buffer (Invitrogen NP0008) + 4% - mercaptoethanol) to a final OD600 of 20. Pellets were boiled for 10 min and stored at −80 °C. The protein concentration of the samples was measured using the Noninterfering (NI) Protein Assay (with BSA protein stand- ard) (G Biosciences catalog no. 786- 005). RpoA western blots were carried out as described above. For the LPS silver stain, 50 µL of sample was incubated with 1.25 µL of proteinase K (NEB P8107S) for 1 h at 55 °C and then 95 °C for 10 min. Also, 20 µg (volume equivalent) was resolved on a 4 to 12% Criterion XT Bis- Tris gel (Bio- Rad 3450124) at 100V for 2 h. LPS detection via silver stain was performed as described previously (75). First, the gel was fixed overnight in a solution of 200 mL of 40% ethanol and 5% acetic acid. Periodic acid was added to the fixative solution (final concentration of 0.7%). Following a 5 min incubation at room temperature, the gel was washed with 200 mL ultrapure H20 (2× for 30 min, 1× for 1 h). The gel was then incubated with 150 mL of staining solution (0.018 N NaOH, 0.4% NH4OH, and 0.667% Silver Nitrate) for 10 min. The gel was then washed 3× for 15 min in 200 mL ultrapure H20 and developed in developer solution (0.26 mM Citric Acid pH 3.0, 0.014% formaldehyde). The reaction was stopped by removing the developer and replacing it with 100 mL of 0.5% acetic acid. The gel was imaged using the Bio- Rad ChemiDocTM MP Imaging System. Detecting LPS Using the Pro- Q Emerald 300 LPS Gel Stain Kit. WT (HC555), mreC(R292H) (PR5), mreC(G156D) (PR30), and ∆mreC (EMF150) expressing wbbL or lacZ from an arabinose- inducible promoter were incubated for 24 h in M9 + CAA + glu + tet at 30 °C and diluted to OD600 = 0.05 in M9 +CAA + ara + tet for 3 h at 30 °C. After 3 h, the cultures were gently pelleted and resuspended in LB + ara + tet. Cells were grown for an additional 2 h at 37 °C. Cells were pelleted and resuspended in 1× LDS sample buffer (Invitrogen NP0008) + 4% 2- mercaptoethanol) to a final OD600 of 20, boiled for 10 min, and stored at −80 °C. The protein concentration of the samples was measured using the Noninterfering (NI) Protein Assay (with BSA protein standard) (G Biosciences catalog no. 786- 005). RpoA western blots were carried out as described above. For the LPS proemeraldQ stain, 50 µL of sample was incubated with 1.25 µL of proteinase K (NEB P8107S) for 1 h at 55 °C then 95 °C for 10 min. A normalized volume equivalent to 20 µg total protein in the predigested sample was resolved on a 4 to 12% Criterion XT Bis- Tris gel (Bio- Rad 3450124) at 100V for 2 h. The Proemerald Q stain was performed following the manufacturer’s instructions (Pro- Q Emerald 300 LPS gel stain kit- Molecular Probes P20495). The gel was imaged using the Bio- Rad ChemiDocTM MP Imaging System. Phase Contrast Microscopy. Phase contrast micrographs in Figs. 1, 2, 3, and 4 and SI Appendix, Figs. S2, S8, and S9 were all taken using cells fixed in 2.6% in formaldehyde and 0.04% glutaraldehyde. After adding the fixative, cells were incubated at room temperature for 1 h and stored at 4 °C for a maximum of 3 d. To image, cells were immobilized on agarose pads (2%) on 1 mm glass slides (1.5 coverslips). Micrographs in Fig. 1 were taken using a Nikon TE2000 inverted microscope using a 1.4 NA Plan Apo Ph3 objective and Nikon Elements Acquisition Software AR 3.2. Micrographs in Fig. 2 were taken with a Nikon Ti Inverted Microscope using a 1.4 NA Plan Apo 100× Ph3 DM objective and with Nikon Elements 4.30 Acquisition Software. Micrographs in Figs. 3 and 4 and SI Appendix, Figs. S2, S8, and S9 were taken with a Nikon Ti2- E inverted microscope using a 1.45 NA Plan Apo 100× Ph3 DM objective lens and Nikon Elements 5.2 Acquisition Software. Micrographs were processed using rolling ball transformation (radius = 35 pixels) in FIJI (76) prior to length and width quantification using the microbeJ plugin (39). The aspect ratio was calculated by dividing the length measurements by the width measurements. The data were plotted in GraphPad Prism and statistical analysis of aspect ratio done in GraphPad Prism using a parametric unpaired t test assuming gaussian distribution but not equal SD (Welch’s correction). Images were cropped in FIJI (76). 3H- mDAP physiological radiolabeling. PG turnover was determined as described previously (7, 10, 51). Data were plotted on GraphPad Prism. MreB Dynamics. wbbL(INS) (AV007) or wbbL+ (EMF210) cells expressing mreC(R292H)D (pMS9) were back diluted from overnight cultures (1:200) and grown in LB + 1 mM IPTG and incubated at 37 °C until OD600 = ~0.4. Cells were then back diluted a second time to OD600 = 0.05 in LB + 1 mM IPTG and incubated at 37 °C until OD600 = ~0.4. # 1.5 high precision coverslips (Marienfeld) were added to a hydrochloric acid and ethanol and cleaned. Cells were placed onto a 2% (w/v) agarose pad in LB + 1 mM IPTG and imaged at RT on a Nikon Ti inverted microscope equipped with Nikon TIRF Lun- f laser illumination, a Plan Apo 100×, 1.45 NA Ph3 objective lens. Images were recorded using an Andor Zyla 4.2 Plus sCMOS camera and Nikon Elements PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 9 of 11 4.30 acquisition software. Three- minute timelapse series with an acquisition frame rate of 3s were recorded to capture MreB dynamics and overlayed over a single- frame phase contrast reference image using Fiji (76). Particle tracking was performed as described in Navarro et al. (77). Briefly, MreB tracks were detected in TrackMate v6.0.1 (56) using a LoG detector (0.3- µm radius) and the Kalman filter. To analyze the nature of the displacement of each track, the mean square displacement ( MSD ) was calculated using the MATLAB class msdana- lyzer (78). Slopes ( 𝛼 ) of the individual MSD curves were extracted using the Log- log fit of the MSD and the delay time 𝜏 . As the maximum delay time 75% of the track length was used. Only tracks which persisted for longer than 4 timepoints (12 s) and with a R2 for log [MSD] versus log [t] above 0.95 were included in the analysis. MreB filaments engaged in active cell wall synthesis are displaced by the action of the enzymatic activities of RodA and PBP2 (2, 7, 17–20, 22, 79) and thus its MSD curves display slopes of 𝛼 ≈ 2 indicative of a transported particle motion above the rate of Brownian diffusion ( 𝛼 ≈ 1) or confined motion ( 𝛼 > 1). The mean directional change rate was derived from TrackMate and is defined as a measure of the angle between two succeeding links, averaged over all the links of a track, and is reported in radians. RNAseq Analysis. Samples were prepared as described in the figure legend (SI Appendix, Fig. S3) and sent to SeqCenter (https://www.seqcenter.com/). RNA extraction, sequencing, and analysis were performed by SeqCenter following stand- ard protocols. Briefly, samples were treated with DNAse (Invitrogen, RNAse free) and libraries were prepared using Illumina’s Stranded Total RNA Prep Ligation with the Ribo- Zero Plus kit and 10 bp unique dual indices. Sequencing was performed on a NovaSeq 6000, generating paired- end 151- bp reads. Demultiplexing, quality control, and adapter trimming was performed with bcl- convert (v4.1.5). lpxC transcripts were normalized to WT (HC555) and plotted using GraphPad Prism. Data, Materials, and Software Availability. The data that support the findings of this study as well as the associated protocols are all presented in the paper. Bacterial strains and other reagents generated during the course of this study are available from the corresponding author upon reasonable request. ACKNOWLEDGMENTS. We would like to acknowledge all the members of the Bernhardt and Rudner labs for their advice and thoughtful comments throughout the course of this work. We also thank Paula Montero Llopis and the other members of the MicRoN (Microscopy Resources on the North Quad) team at Harvard Medical School for their expertise, support, consultation, and services. We thank Bill Doerrler for the generous gift of the anti- LpxC antibody. We thank Andrew Darwin (NYU) and Teru Ogura (Kumamoto University, Japan) for sharing with us the fabZ(L85P) mutant and Natividad Ruiz for sharing the wbbL+ strain NR2528. This work was supported by the NIH (R01 AI083365 and U19 AI158028 to T.G.B.) and Investigator funds from the Howard Hughes Medical Institute (T.G.B.). E.M.F. is supported by the NSF Graduate Research Fellowship award. P.D.A.R. was supported in part by a predoctoral fellowship from the Canadian Institute for Health Research. A.V. is supported by a EMBO long- term postdoctoral fellowship (ALTF_89- 2019) and a SNF Postdoc Mobility fellowship (P500PB_203143). 1. 2. 3. 4. 5. 6. P. D. Rohs, T. G. Bernhardt, Growth and division of the peptidoglycan matrix. Annu. Rev. Microbiol. 75, 315–336 (2021). A. Typas, M. Banzhaf, C. A. Gross, W. Vollmer, From the regulation of peptidoglycan synthesis to bacterial growth and morphology. Nat. Rev. Microbiol. 10, 123 (2012). E. R. Rojas et al., The outer membrane is an essential load- bearing element in Gram- negative bacteria. Nature 559, 617 (2018). J.- V. Höltje, Growth of the stress- bearing and shape- maintaining murein sacculus of Escherichia coli. Microbiol. Mol. Biol. Rev. 62, 181–203 (1998). E. Sauvage, F. Kerff, M. Terrak, J. A. Ayala, P. Charlier, The penicillin- binding proteins: Structure and role in peptidoglycan biosynthesis. FEMS Microbiol. Rev. 32, 234–258 (2008). A. J. Meeske et al., SEDS proteins are a widespread family of bacterial cell wall polymerases. Nature 537, 634 (2016). 7. H. Cho et al., Bacterial cell wall biogenesis is mediated by SEDS and PBP polymerase families 8. 9. functioning semi- autonomously. Nat. Microbiol. 1, 16172 (2016). K. Emami et al., RodA as the missing glycosyltransferase in Bacillus subtilis and antibiotic discovery for the peptidoglycan polymerase pathway. Nat. Microbiol. 2, 1–9 (2017). A. Taguchi et al., FtsW is a peptidoglycan polymerase that is functional only in complex with its cognate penicillin- binding protein. Nat. Microbiol. 4, 587–594 (2019). 10. P. D. Rohs et al., A central role for PBP2 in the activation of peptidoglycan polymerization by the bacterial cell elongation machinery. PLoS Genet. 14, e1007726 (2018). 27. S. Mahalakshmi, M. Sunayana, L. SaiSree, M. Reddy, yciM is an essential gene required for regulation of lipopolysaccharide synthesis in Escherichia coli. Mol. Microbiol. 91, 145–157 (2014). 28. K. Ito, Y. Akiyama, Cellular functions, mechanism of action, and regulation of FtsH protease. Annu. Rev. Microbiol. 59, 211–231 (2005). 29. T. Ogura et al., Balanced biosynthesis of major membrane components through regulated degradation of the committed enzyme of lipid A biosynthesis by the AAA protease FtsH (HflB) in Escherichia coli. Mol. Microbiol. 31, 833–844 (1999). 30. S. Shu, W. Mi, Regulatory mechanisms of lipopolysaccharide synthesis in Escherichia coli. Nat. Commun. 13, 1–11 (2022). 31. M. S. Anderson, A. D. Robertson, I. Macher, C. R. Raetz, Biosynthesis of lipid A in Escherichia coli: Identification of UDP- 3- O- [(R)- 3- hydroxymyristoyl]- alpha- D- glucosamine as a precursor of UDP- N2, O3- bis [(R)- 3- hydroxymyristoyl]- alpha- D- glucosamine. Biochemistry 27, 1908–1917 (1988). 32. K. Young et al., The envA permeability/cell division gene of Escherichia coli encodes the second enzyme of lipid A biosynthesis. UDP- 3- O- (R- 3- hydroxymyristoyl)- N- acetylglucosamine deacetylase. J. Biol. Chem. 270, 30384–30391 (1995). 33. M. B. Cian, N. P. Giordano, R. Masilamani, K. E. Minor, Z. D. Dalebroux, Salmonella enterica serovar Typhimurium use PbgA/YejM to regulate lipopolysaccharide assembly during bacteremia. Infect. Immun. 88, e00758- 19 (2019). 11. F. O. Bendezú, P. A. De Boer, Conditional lethality, division defects, membrane involution, and 34. T. Clairfeuille et al., Structure of the essential inner membrane lipopolysaccharide–PbgA complex. endocytosis in mre and mrd shape mutants of Escherichia coli. J. Bacteriol. 190, 1792–1811 (2008). Nature 584, 479–483 (2020). 12. F. O. Bendezú, C. A. Hale, T. G. Bernhardt, P. A. De Boer, RodZ (YfgA) is required for proper assembly 35. E. M. Fivenson, T. G. Bernhardt, An essential membrane protein modulates the proteolysis of LpxC to of the MreB actin cytoskeleton and cell shape in E. coli. EMBO J. 28, 193–204 (2009). control lipopolysaccharide synthesis in Escherichia coli. mBio 11, e00939- 20 (2020). 13. S. A. Alyahya et al., RodZ, a component of the bacterial core morphogenic apparatus. Proc. Natl. Acad. Sci. U.S.A. 106, 1239–1244 (2009). 14. T. Kruse, J. Bork- Jensen, K. Gerdes, The morphogenetic MreBCD proteins of Escherichia coli form an essential membrane- bound complex. Mol. Microbiol. 55, 78–89 (2005). 36. R. L. Guest, D. Samé Guerra, M. Wissler, J. Grimm, T. J. Silhavy, YejM modulates activity of the YciM/ FtsH protease complex to prevent lethal accumulation of lipopolysaccharide. mBio 11, e00598- 20 (2020). 37. D. Nguyen, K. Kelly, N. Qiu, R. Misra, YejM controls LpxC levels by regulating protease activity of the 15. M. Leaver, J. Errington, Roles for MreC and MreD proteins in helical growth of the cylindrical cell FtsH/YciM complex of Escherichia coli. J. Bacteriol. 202, e00303- 20 (2020). wall in Bacillus subtilis. Mol. Microbiol. 57, 1196–1209 (2005). 16. R. M. Morgenstein et al., RodZ links MreB to cell wall synthesis to mediate MreB rotation and robust morphogenesis. Proc. Natl. Acad. Sci. U.S.A. 112, 12510–12515 (2015). 38. D. Biernacka, P. Gorzelak, G. Klein, S. Raina, Regulation of the first committed step in lipopolysaccharide biosynthesis catalyzed by LpxC requires the essential protein LapC (YejM) and HslVU protease. Int. J. Mol. Sci. 21, 9088 (2020). 17. D. Shiomi, M. Sakai, H. Niki, Determination of bacterial rod shape by a novel cytoskeletal membrane 39. A. Ducret, E. M. Quardokus, Y. V. Brun, MicrobeJ, a tool for high throughput bacterial cell detection protein. EMBO J. 27, 3081–3091 (2008). 18. M. Wachi, M. Matsuhashi, Negative control of cell division by mreB, a gene that functions in determining the rod shape of Escherichia coli cells. J. Bacteriol. 171, 3123–3127 (1989). 19. R. A. Daniel, J. Errington, Control of cell morphogenesis in bacteria: Two distinct ways to make a rod- shaped cell. Cell 113, 767–776 (2003). and quantitative analysis. Nat. Microbiol. 1, 16077 (2016). 40. P. G. Sorensen et al., Regulation of UDP- 3- O- [R- 3- hydroxymyristoyl]- N- acetylglucosamine deacetylase in Escherichia coli: The second enzymatic step of lipid a biosynthesis. J. Biol. Chem. 271, 25898–25905 (1996). 41. F. Führer, S. Langklotz, F. Narberhaus, The C- terminal end of LpxC is required for degradation by the 20. J. Domínguez- Escobar et al., Processive movement of MreB- associated cell wall biosynthetic FtsH protease. Mol. Microbiol. 59, 1025–1036 (2006). complexes in bacteria. Science 333, 225–228 (2011). 21. E. C. Garner, Toward a mechanistic understanding of bacterial rod shape formation and regulation. Annu. Rev. Cell Dev. Biol. 37, 1–21 (2021). 22. S. Van Teeffelen et al., The bacterial actin MreB rotates, and rotation depends on cell- wall assembly. Proc. Natl. Acad. Sci. U.S.A. 108, 15822–15827 (2011). 23. P. D. Rohs et al., Identification of potential regulatory domains within the MreC and MreD components of the cell elongation machinery. J. Bacteriol. 203, e00493- 20 (2021). 24. X. Liu, J. Biboy, E. Consoli, W. Vollmer, T. den Blaauwen, MreC and MreD balance the interaction between the elongasome proteins PBP2 and RodA. PLoS Genet. 16, e1009276 (2020). 25. G. Özbaykal et al., The transpeptidase PBP2 governs initial localization and activity of the major cell- wall synthesis machinery in E. coli. Elife 9, e50629 (2020). 26. G. Klein, N. Kobylak, B. Lindner, A. Stupak, S. Raina, Assembly of lipopolysaccharide in Escherichia coli requires the essential LapB heat shock protein. J. Biol. Chem. 289, 14829–14853 (2014). 42. J. E. Jackman, C. R. Raetz, C. A. Fierke, Site- directed mutagenesis of the bacterial metalloamidase UDP- (3- O- acyl)- N- acetylglucosamine deacetylase (LpxC). Identification of the zinc binding site. Biochemistry 40, 514–523 (2001). 43. R. J. Heath, C. O. Rock, Roles of the FabA and FabZ β- hydroxyacyl- acyl carrier protein dehydratases in Escherichia coli fatty acid biosynthesis. J. Biol. Chem. 271, 27795–27801 (1996). 44. M. Schäkermann, S. Langklotz, F. Narberhaus, FtsH- mediated coordination of lipopolysaccharide biosynthesis in Escherichia coli correlates with the growth rate and the alarmone (p) ppGpp. J. Bacteriol. 195, 1912–1919 (2013). 45. A. L. McClerren et al., A slow, tight- binding inhibitor of the zinc- dependent deacetylase LpxC of lipid A biosynthesis with antibiotic activity comparable to ciprofloxacin. Biochemistry 44, 16574–16583 (2005). 46. A. W. Barb et al., Inhibition of lipid A biosynthesis as the primary mechanism of CHIR- 090 antibiotic activity in Escherichia coli. Biochemistry 46, 3793–3802 (2007). 10 of 11 https://doi.org/10.1073/pnas.2301987120 pnas.org 47. A. Emiola, S. S. Andrews, C. Heller, J. George, Crosstalk between the lipopolysaccharide and 63. H. Suzuki et al., Murein- lipoprotein of Escherichia coli: A protein involved in the stabilization of phospholipid pathways during outer membrane biogenesis in Escherichia coli. Proc. Natl. Acad. Sci. U.S.A. 113, 3108–3113 (2016). bacterial cell envelope. Mol. Gen. Genet. 167, 1–9 (1978). 64. G. Mamou et al., Peptidoglycan maturation controls outer membrane protein assembly. Nature 606, 48. E. Bi, J. Lutkenhaus, Cell division inhibitors SulA and MinCD prevent formation of the FtsZ ring. J. 953–959 (2022). Bacteriol. 175, 1118–1125 (1993). 65. J. Sun, S. T. Rutherford, T. J. Silhavy, K. C. Huang, Physical properties of the bacterial outer 49. S. C. Cordell, E. J. Robinson, J. Löwe, Crystal structure of the SOS cell division inhibitor SulA and in membrane. Nat. Rev. Microbiol. 20, 236–248 (2022). complex with FtsZ. Proc. Natl. Acad. Sci. U.S.A. 100, 7889–7894 (2003). 66. J. Berry, M. Rajaure, T. Pang, R. Young, The spanin complex is essential for lambda lysis. J. Bacteriol. 50. A. Mukherjee, C. Cao, J. Lutkenhaus, Inhibition of FtsZ polymerization by SulA, an inhibitor of 194, 5667–5674 (2012). septation in Escherichia coli. Proc. Natl. Acad. Sci. U.S.A. 95, 2885–2890 (1998). 67. M. Rajaure, J. Berry, R. Kongari, J. Cahill, R. Young, Membrane fusion during phage lysis. Proc. Natl. 51. H. Cho, T. Uehara, T. G. Bernhardt, Beta- lactam antibiotics induce a lethal malfunctioning of the Acad. Sci. U.S.A. 112, 5497–5502 (2015). bacterial cell wall synthesis machinery. Cell 159, 1300–1311 (2014). 68. A. Formstone, J. Errington, A magnesium- dependent mreB null mutant: Implications for the role of 52. H. Nikaido, Molecular basis of bacterial outer membrane permeability revisited. Microbiol. Mol. Biol. mreB in Bacillus subtilis. Mol. Microbiol. 55, 1646–1657 (2005). Rev. 67, 593–656 (2003). 69. H. Rogers, P. Thurman, R. Buxton, Magnesium and anion requirements of rodB mutants of Bacillus 53. C. Whitfield, M. S. Trent, Biosynthesis and export of bacterial lipopolysaccharides. Annu. Rev. subtilis. J. Bacteriol. 125, 556–564 (1976). Biochem. 83, 99–128 (2014). 70. J. A. Buss et al., Pathway- directed screen for inhibitors of the bacterial cell elongation machinery. 54. X. Rubirés et al., A gene (wbbL) from Serratia marcescens N28b (O4) complements the rfb- 50 Antimicrob. Agents Chemother. 63, e01530- 18 (2019). mutation of Escherichia coli K- 12 derivatives. J. Bacteriol. 179, 7581–7586 (1997). 71. K. R. Hummels et al., Coordination of bacterial cell wall and outer membrane biosynthesis. Nature 55. S. Hussain et al., MreB filaments align along greatest principal membrane curvature to orient cell 615, 300–304 (2023). wall synthesis. Elife 7, e32471 (2018). 56. J.- Y. Tinevez et al., TrackMate: An open and extensible platform for single- particle tracking. Methods 115, 80–90 (2017). 72. B. W. Simpson et al., Acinetobacter baumannii can survive with an outer membrane lacking lipooligosaccharide due to structural support from elongasome peptidoglycan synthesis. mBio 12, e0309921 (2021). 57. D. Ershov et al., TrackMate 7: Integrating state- of- the- art segmentation algorithms into tracking 73. A. I. Weaver et al., Genetic determinants of penicillin tolerance in Vibrio cholerae. Antimicrob. pipelines. Nat. Methods 19, 829–832 (2022). Agents Chemother. 62, e01326- 18 (2018). 58. M. S. Anderson, C. Raetz, Biosynthesis of lipid A precursors in Escherichia coli. A cytoplasmic 74. J. Miller, Experiments in Molecular Genetics (Cold Spring Laboratory Press, Cold Spring Harbor, NY, acyltransferase that converts UDP- N- acetylglucosamine to UDP- 3- O- (R- 3- hydroxymyristoyl)- N- acetylglucosamine. J. Biol. Chem. 262, 5159–5169 (1987). 59. D. N. Crowell, M. S. Anderson, C. Raetz, Molecular cloning of the genes for lipid A disaccharide synthase and UDP- N- acetylglucosamine acyltransferase in Escherichia coli. J. Bacteriol. 168, 152–159 (1986). 60. J. Marquardt, D. Siegele, R. Kolter, C. Walsh, Cloning and sequencing of Escherichia coli murZ and purification of its product, a UDP- N- acetylglucosamine enolpyruvyl transferase. J. Bacteriol. 174, 5748–5752 (1992). 61. V. Braun, K. Rehn, Chemical characterization, spatial distribution and function of a lipoprotein (murein- lipoprotein) of the E. coli cell wall: The specific effect of trypsin on the membrane structure. Eur. J. Biochem. 10, 426–438 (1969). 1972). 75. C.- M. Tsai, C. E. Frasch, A sensitive silver stain for detecting lipopolysaccharides in polyacrylamide gels. Anal. Biochem. 119, 115–119 (1982). 76. J. Schindelin et al., Fiji: An open- source platform for biological- image analysis. Nat. Methods 9, 676–682 (2012). 77. P. P. Navarro et al., Cell wall synthesis and remodelling dynamics determine division site architecture and cell shape in Escherichia coli. Nat. Microbiol. 7, 1621–1634 (2022). 78. N. Tarantino et al., TNF and IL- 1 exhibit distinct ubiquitin requirements for inducing NEMO–IKK supramolecular structures. J. Cell Biol. 204, 231–245 (2014). 62. M. Inouye, J. Shaw, C. Shen, The assembly of a structural lipoprotein in the envelope of Escherichia 79. E. C. Garner et al., Coupled, circumferential motions of the cell wall synthesis machinery and MreB coli. J. Biol. Chem. 247, 8154–8159 (1972). filaments in B. subtilis. Science 333, 222–225 (2011). PNAS 2023 Vol. 120 No. 35 e2301987120 https://doi.org/10.1073/pnas.2301987120 11 of 11
10.1073_pnas.2305556120	RESEARCH ARTICLE \| BIOCHEMISTRY OPEN ACCESS Changing protein–DNA interactions promote ORC binding- site exchange during replication origin licensing Annie Zhanga,b , and Stephen P. Bella,b,1 , Larry J. Friedmanc , Jeff Gellesc,1 Edited by James Berger, Johns Hopkins University, Baltimore, MD; received April 17, 2023; accepted June 21, 2023 During origin licensing, the eukaryotic replicative helicase Mcm2- 7 forms head- to- head double hexamers to prime origins for bidirectional replication. Recent single- molecule and structural studies revealed that one molecule of the helicase loader ORC (origin recognition complex) can sequentially load two Mcm2- 7 hexamers to ensure proper head- to- head helicase alignment. To perform this task, ORC must release from its initial high- affinity DNA- binding site and “flip” to bind a weaker, inverted DNA site. However, the mechanism of this binding- site switch remains unclear. In this study, we used single- molecule Förster resonance energy transfer to study the changing interac- tions between DNA and ORC or Mcm2- 7. We found that the loss of DNA bending that occurs during DNA deposition into the Mcm2- 7 central channel increases the rate of ORC dissociation from DNA. Further studies revealed temporally controlled DNA sliding of helicase- loading intermediates and that the first sliding complex includes ORC, Mcm2- 7, and Cdt1. We demonstrate that sequential events of DNA unbending, Cdc6 release, and sliding lead to a stepwise decrease in ORC stability on DNA, facili- tating ORC dissociation from its strong binding site during site switching. In addition, the controlled sliding we observed provides insight into how ORC accesses secondary DNA- binding sites at different locations relative to the initial binding site. Our study highlights the importance of dynamic protein–DNA interactions in the loading of two oppositely oriented Mcm2- 7 helicases to ensure bidirectional DNA replication. Mcm2- 7 helicase \| origin recognition complex (ORC) \| origin licensing \| DNA replication initiation \| single- molecule FRET Eukaryotic chromosomal replication initiates by building two oppositely oriented repli- cation forks at origins of replication. During G1, the core component of the replicative helicase, the Mcm2- 7 complex, is assembled around origins in a process termed origin licensing or helicase loading. This process marks all potential origins of replication, and a subset of the loaded helicases are activated in S phase to form the core of the replication machinery (1). To ensure bidirectional replication, each pair of Mcm2- 7 helicases must be assembled around origin DNA in a head- to- head orientation (2–6). In budding yeast, a “one- loader” helicase loading mechanism has been proposed in which one molecule of the helicase- loader, the origin recognition complex (ORC), coor- dinates the loading of both Mcm2- 7 hexamers (7–9). In this model, ORC begins helicase loading by binding origin DNA and recruiting Cdc6 (Fig. 1A, step 1) (10, 11). The ORC–Cdc6–DNA complex recruits a Mcm2- 7–Cdt1 complex (Fig. 1A, step 2) to form the transient ORC–Cdc6–Cdt1–Mcm2- 7 (OCCM) intermediate (Fig. 1A, step 3). Cdc6 and then Cdt1 are subsequently released in a strict sequential order (Fig. 1A, steps 4 to 5) (7). Upon Cdt1 dissociation, the initial ORC- Mcm2- 7 interaction is broken, allowing this ORC molecule to “flip” to the opposite side of the helicase and form a distinct Mcm2- 7- ORC (MO) intermediate (Fig. 1A, step 6) (8, 9). MO formation mediates stable closing of the Mcm2/5 gate, a gap between subunits Mcm2 and Mcm5 that allows access of DNA to the Mcm2- 7 central channel, such that the helicase stably encircles DNA (8, 12). The resulting MO complex is required to recruit and load a second helicase in the opposite orientation (8, 9). Once recruited, the second helicase rapidly forms the head- to- head interactions with the first helicase. In addition to protein–protein interactions, ORC forms extensive but changing inter- actions with DNA (Fig. 1A, red dashed boxes). In budding yeast, the primary ORC binding site is defined by the ARS consensus sequence (ACS) and B1 element (11, 13, 14). ORC encircles the ACS as an open ring, forming additional interactions with DNA including and beyond the B1 element to create a strong ~80° bend in DNA (15). Structural studies have captured a “preinsertion OCCM” intermediate with bent DNA that contains all helicase- loading proteins (Fig. 1A) (16). However, the bent DNA straightens and is inserted in the Mcm2- 7 open ring in the later OCCM intermediate (17). These intermediates Significance Bidirectional DNA replication, in which two replication forks travel in opposite directions from each origin of replication, is required for complete genome duplication. To prepare for this event, two copies of the Mcm2- 7 replicative helicase are loaded in opposite orientations at each origin. Using single- molecule assays, we identified a sequence of changing protein–DNA interactions involved in this process. These stepwise changes gradually reduce the DNA- binding strength of ORC (origin recognition complex), the primary DNA- binding protein involved in this event. This reduced affinity promotes ORC dissociation and rebinding in the opposite orientation on the DNA, facilitating the sequential assembly of two Mcm2- 7 molecules in opposite orientations. Our findings identify a coordinated series of events that drive proper DNA replication initiation. Author affiliations: aHHMI, Cambridge, MA 02139; bDepartment of Biology, Massachusetts Institute of Technology, Cambridge, MA 02139; and cDepartment of Biochemistry, Brandeis University, Waltham, MA 02454 Author contributions: A.Z., L.J.F., J.G., and S.P.B. designed research; A.Z. performed research; A.Z., L.J.F., J.G., and S.P.B. analyzed data; and A.Z., L.J.F., J.G., and S.P.B. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2305556120/- /DCSupplemental. Published July 18, 2023. PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 1 of 12 together suggest that a segment of DNA flanking the bend, which we refer to as the bend- proximal region (BPR) (Fig. 1A), is trans- ferred from ORC to Mcm2- 7, allowing the DNA bend to straighten as the BPR enters the Mcm2- 7 central channel. Although molecular dynamics models have simulated the DNA insertion process (16), DNA unbending and deposition into Mcm2- 7 have yet to be mon- itored in real time. In addition, the trigger(s) for this change in ORC- DNA interactions has not been defined. For one ORC to load both Mcm2- 7 helicases at an origin, ORC has to switch between two oppositely oriented DNA- binding sites. Natural yeast origins include at least one B2 element, each of which includes a partial match to the ACS but in the inverted orientation (Fig. 1A, purple arrow) (18–21). As a result, the B2 element is proposed to serve as a weaker secondary binding site for ORC. Studies with artificial origins showed that a second inverted binding site is essential for robust helicase loading (22). A ORC-DNA (PDB: 5ZR1) Pre-insertion OCCM (PDB: 6WGG) OCCM (PDB: 5V8F) MO (PDB: 6RQC) ORC-BPR ORC ORC-ACS ACS Cdc6 ① ② Mcm2-7/ Cdt1 ③ ⑥ ORC Flip Cdc6 ④ ⑤ MCM-BPR Cdt1 No ORC High ORCC•+51 FRET Low ORCC•+51 FRET ARS1+51-Cy5 ACS B C ) . U . A ( y t i s n e t n I e c n e c s e r o u F l 4 2 0 4 2 0 4 2 0 1 ORCC Dex , Dem Dex , Aem Total emission Dex , (Dem+Aem) T E R F E ) . U . A ( y t i s n e t n I e c n e c s e r o u F l ORCC 4 2 0 4 2 0 4 2 0 1 T E R F E 0.5 0 Dex , Aem/(Dem+Aem) 25 50 75 100 125 Time (s) 0.5 0 75 100 125 150 Time (s) ORC-B2 B2 0.62 D y t i s n e D y t i l i b a b o r P 5 4 3 2 1 0 0.04 0.02 0 0 0.2 0.28 0 0.2 0.6 0.4 EFRET 0.8 1 E ×10-2 ) 1 - s ( p p a • f f o k 15 10 5 0 EFRET < 0.28 EFRET ≥ 0.28 Fig. 1. ORC introduces a stable bend in DNA that increases its stability on DNA. (A) Model of helicase- loading from initial ORC recruitment onto origin DNA to MO formation. Illustrations based on cryo- EM structures of helicase- loading intermediates are labeled with PDB identifiers. Pink dashed boxes: protein–DNA interactions. Colored arrows on the bottom show the relative orientation of the ACS and B2 elements. Further details are given in the text. (B) Schematic of ORCC•+51 sm- FRET experiment that monitors ORC- BPR interaction. ORCC•+51 FRET (curved arrow) is high when the complex is in the bent state, and low in the unbent state. (C) ORCC•+51 FRET assay records at two individual DNA molecules. Top plot: donor emission during donor excitation (green, Dex, Dem). The green arrow indicates ORCC arrival on DNA. Second plot: acceptor emission during donor excitation (purple, Dex, Aem). Third plot: total emission during donor excitation [black, Dex, (Dem+Aem)]. Bottom plot: effective FRET efficiency (EFRET) [blue, Dex, Aem/(Dem+Aem)]. Gray segments represent background signal when no fluorescent ORC is colocalized with DNA; segments of the intensity line plot during ORC colocalization are colored. The colored background indicates ORC- DNA in the high EFRET bent state (blue, as in B) or in the low EFRET unbent state (orange, as in B). A red dashed line at EFRET = 0.28 separates the high and low EFRET states. Additional traces are shown in SI Appendix, Fig S1B. (D) Histogram of EFRET values during each frame when ORC was present on a DNA. Center EFRET value from a one- component Gaussian fit (red curve; SI Appendix, Table S1) is 0.62 ± 0.001 (red vertical line). Inset: Magnified histogram. EFRET values above (blue) or below (yellow) the 0.28 threshold (red dashed line) are shown. Data are from 25,388 acquired frames during 186 ORC colocalization events. (E) Apparent dissociation rate of ORC from DNA (koff•app) when ORC is in the low EFRET state (<0.28) or the high EFRET state (≥0.28). Error bars indicate the SE. Number of video frames in low EFRET state = 141; number of frames in high EFRET state = 25,247. 2 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org Interestingly, B2 elements are found at variable distances from the ACS (18, 20, 23). A key question raised by the “one- loader” hypothesis for helicase loading is how ORC releases from its strong initial binding site to bind a weaker, oppositely oriented secondary binding site a variable distance away. Combining colocalization single- molecule spectroscopy (CoSMoS) (24, 25) and single- molecule Förster resonance energy transfer (sm- FRET) (26), we monitored a variety of interactions between ORC and Mcm2- 7 with the origin DNA and identified key triggers for changes in the observed protein–DNA interactions. We demonstrated that the ORC–DNA complex predominantly contains bent DNA but infrequently transitions to a less stable con- formation with straightened DNA. Cdc6 binding does not change the bent state of the ORC- bound DNA but does decrease the dis- sociation of ORC from DNA. Although the OCCM is static on the DNA, upon loss of Cdc6 (Fig. 1A, step 4) we observe changes in sm- FRET supporting sliding of the ORC–Cdt1–Mcm2- 7 (OC1M) complex. Based on our data, we propose that the sequential events of DNA unbending, Cdc6 release, and DNA sliding progressively reduce ORC stability on DNA to facilitate ORC dissociation from DNA and flipping to bind the inverted B2 element. Results The ORC–DNA Complex Predominantly Exists in a Bent DNA State. To understand how and when origin DNA changes its conformation during helicase loading, we developed a sm- FRET assay to monitor the dynamics of ORC- induced DNA bending. Previous cryo- EM structures showed that the C- terminal face of ORC, particularly the C- terminal domain of Orc6, is in close proximity to the BPR of origin DNA (15). To position a donor fluorophore close to the BPR, we generated an ORC fluorescently labeled on its C- terminal face (ORCC). To this end, we deleted residues 1 to 266 of Orc6 and attached a fluorophore to the truncated N terminus. Although defective in recruiting a second Mcm2- 7 (27), truncation of this region in Orc6 does not compromise OCCM formation in ensemble helicase- loading assays (SI Appendix, Fig. S1A). An acceptor fluorophore was coupled to DNA at an internal position 51 bp from the ARS1 origin ACS (ARS1+51- Cy5). This position was chosen to maximize proximity to the Orc6 label when ORC is bound to the BPR and minimize their proximity when the DNA is straight (Fig. 1B). We refer to the FRET between this donor–acceptor pair as “ORCC•+51” FRET (Fig. 1B), and the corresponding interaction between ORC and the BPR as the “ORC- BPR” interaction (Fig. 1A, first panel, dashed box). Using total internal reflection fluorescence microscopy, we monitored the colocalization of ORCC in solution with surface- tethered origin DNA (24). Throughout each ORC- DNA colocalization event, we measured effective ORCC•+51 FRET efficiency (EFRET) to examine ORC- induced DNA bending. When ORC arrived on DNA, we consistently observed a high ORCC•+51 EFRET state, indicating that ORC was bound to the ACS and BPR and that the bound DNA was bent (blue back- ground in Fig. 1 C and D). The majority of ORCC•+51 EFRET values were centered at 0.62 (Fig. 1D). Although they represented less than 0.5% of the time of ORC- DNA colocalization, we also observed short periods of lower EFRET values (<0.28, orange back- ground in Fig. 1 C and D and SI Appendix, Fig. S1B). These low EFRET values were not caused by photobleaching, as we examined the DNA- coupled fluorophore by acceptor excitation after the experiment and restricted analysis to DNA molecules that main- tained fluorescence to the end of the experiment. In addition, control experiments showed that these low EFRET values were not caused by ORC sliding along the DNA to move away from the ACS (SI Appendix, Fig. S2). We excluded a small fraction (approx- imately 2%) of ORC molecules that did not show a high EFRET signal at any time during the DNA colocalization, as these likely represent nonspecific ORC binding. The poor fit of a single Gaussian model at low EFRET values (Fig. 1D, red curve in Inset) suggest the presence of distinct populations at the bent or unbent conformational states. We conclude that the low EFRET values reflect an unbent state in which the ORC remains bound at the ACS but the ORC- BPR interaction is lost (Fig. 1B). Addition of Cdc6 increases ORC stability on DNA, leading to a near twofold decrease in its apparent dissociation rate from 0.0131 s−1 to 0.0073 s−1 (SI Appendix, Fig. S1C and Table S1). However, Cdc6 did not shift the ORCC•+51 EFRET distribution between the bent and unbent states, consistent with Cdc6 not significantly altering the ORC- BPR interaction (SI Appendix, Fig. S1C) (16, 28, 29). Thus, ORC- bound DNA is predominantly in the bent state with infre- quent transitions to the unbent state. We frequently observed the low EFRET unbent state immediately prior to ORC dissociation from DNA (Fig. 1C and SI Appendix, Fig. S1B). Thus, we asked whether DNA unbending increases the rate of ORC dissociation from DNA. Indeed, ORC in the low EFRET state (<0.28) dissociates more than 40 times faster from DNA than ORC in the high EFRET state (≥0.28) (Fig. 1E). These data indicate that the ORC- BPR interaction increases ORC sta- bility on DNA. As a further test of the ORC- BPR contribution to the kinetic stability of the ORC–DNA complex, we mutated 9 amino acids in the ORC region that interacts with the phosphate backbone of the BPR (15) (SI Appendix, Fig. S1D). When these mutations are incor- porated into labeled ORC (ORCC- BPR), we observed a limited decrease in EFRET (centered at 0.52, SI Appendix, Fig. S1E), suggest- ing that these interactions contribute to but are not solely responsible for ORC- induced DNA bending. Nevertheless, ORCC- BPR stability on DNA is notably reduced (SI Appendix, Fig. S1E), reflected by a significant increase in its apparent dissociation rate (0.0849 s−1) compared to WT (0.0131 s−1, SI Appendix, Table S1). We conclude that the kinetic stability of ORC on DNA is strongly dependent on the ORC- BPR interaction, as a partial loss of this interaction results in a strong reduction in ORC- DNA stability. Mcm4 and Mcm6 Winged- Helix Domains (WHDs) Trigger DNA Unbending. After Mcm2- 7 recruitment, the BPR disengages from ORC and is inserted into the helicase central channel (17). To understand the transition of the BPR DNA from being bound to ORC- Cdc6 to being deposited into the helicase, we investigated the kinetics of DNA unbending during helicase loading. We included Cdc6 and labeled Mcm2- 74N- 650- Cdt1 in our experiment to simultaneously monitor ORC- BPR interactions and Mcm2- 7 recruitment. To avoid using red- excited fluorophores on both DNA and Mcm2- 7, we replaced the DNA- coupled acceptor with the fluorescence quencher Black Hole Quencher- 2 (BHQ- 2) at the same DNA site (ARS1+51- BHQ2). In this experiment, DNA bending induced by ORC- BPR interaction results in quenching of green- excited ORCC fluorescence, and DNA unbending due to loss of the ORC- BPR interaction leads to unquenching (Fig. 2A). Using increase in green- excited fluorescence as a marker for DNA unbending, we consistently detected a temporal delay between Mcm2- 7 arrival (Fig. 1A, step 2) and DNA unbending (Fig. 2B and SI Appendix, Fig. S3). Unbending occurs 3.9 ± 0.3 s (mean ± SEM) after Mcm2- 7 arrival, which is significantly earlier (Fig. 2C) than Cdc6 release (Fig. 1A, step 4) (7). Given the temporal delay between Mcm2- 7 recruit.ment and DNA unbending, we explored the possibility of additional regulatory steps between these two events. PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 3 of 12 A C Quenched Unquenched i i g n v v r u s n o i t c a r F D n o i t c a r F Low ORCC Fluorescence High ORCC Fluorescence B ) . U . A ( y t i s n e n t I e c n e c s e r o u F l 2 1 0 2 1 0 o6d266cbhq2m4n9 AOI 287 GexRex , Gem Unquench ORCC GexRex , Rem Mcm2-74N-650 400 410 420 430 440 450 Time (s) 1 0.8 0.6 0.4 0.2 0 0 1 0.8 0.6 0.4 0.2 0 Unquenching Cdc6 release 5 10 20 15 t - t arrival (s) Mcm2-7 25 30 0.83 0.65 0.33 0.022 WT 4∆WHD 6∆WHD 4∆6∆WHD Fig. 2. DNA unbending occurs rapidly after Mcm2- 7 recruitment and is driven by WHDs of Mcm4 and Mcm6. (A) Schematic of fluorescence quenching experiment monitoring ORC- BPR interaction during the loading of the first Mcm2- 7 helicase. Mcm2- 7 is labeled with a red- excited fluorophore at the unstructured Mcm4 N- terminal tail (Mcm2- 74N- 650). The black dot represents the quencher placed at the +51 position on ARS1. (B) Representative record from the experiment in (A). Top plot: green emission from green and red simultaneous excitation (GexRex, Gem). The green arrow marks ORCC association with DNA, and the black arrow marks unquenching of ORC- associated fluorophore. Bottom plot: red emission from green and red simultaneous excitation (GexRex, Rem). The dashed line and red arrow mark Mcm2- 74N- 650 arrival. Time resolution = 0.25 s. Additional examples are shown in SI Appendix, Fig. S3. (C) Survival functions comparing the interval between Mcm2- 7 arrival and unquenching (i.e., DNA unbending, black) and the interval between Mcm2- 7 arrival and Cdc6 release (purple). Shading: 95% CI. 228 DNA unbending and 82 Cdc6 release events are plotted. The Cdc6 data were originally reported in Ticau et al. (7, figure 4D). (D) Fraction of Mcm2- 7 colocalization events during which DNA unbent. Error bars: SE. Previous biochemical and structural studies suggest that Mcm2- 7 association with the ORC–Cdc6–DNA complex is a multistep pro- cess. After ORC- Cdc6 initially recruits Mcm2- 7 via the WHDs of Mcm3 and Mcm7, additional interactions are formed between ORC and Mcm2- 7 (16, 30, 31). In particular, the C- terminal WHDs of Mcm4 and Mcm6 establish additional interactions with ORC between initial helicase recruitment and OCCM formation. To address if these new interactions promote DNA unbending, we created red- excited Mcm2- 7 mutants labeled at the Mcm4 N ter- minus that lack the WHDs in Mcm4 (4ΔWHD), Mcm6 (6ΔWHD), or both subunits (4Δ6ΔWHD). We tested each of these mutants in the ORCC•+51 unquenching assay. Using unquenching as a marker for DNA unbending, we determined the fraction of Mcm2- 7- DNA colocalization events that resulted in DNA unbending. Deletion of either the Mcm4 or Mcm6 WHDs showed reduced fractions of Mcm2- 7 colocalization events that displayed unbending, with the Mcm6 WHD deletion having the stronger effect (Fig. 2D). Simultaneous deletion of the Mcm4 and Mcm6 WHDs almost completely abolished DNA unbending (Fig. 2D). Although these mutations could delay DNA unbending or lead to Mcm2- 7 dissociation before DNA unbending occurs, the time from Mcm2- 7 arrival to DNA unbending was not changed by the mutations, and this time is significantly faster than Mcm2- 7 dissociation for each mutant (SI Appendix, Fig. S4A). Thus, these findings strongly suggest that interactions made by ORC with the Mcm4 and Mcm6 WHDs trigger DNA unbending, coordinating this event with proper ORC- Mcm2- 7 positioning. DNA Is Deposited Rapidly into the Central Channel of Mcm2- 7 after DNA Unbending. Structural studies of the preinsertion OCCM intermediate show that ORC holds the BPR directly above the Mcm2/5 gate to position the BPR for entry into the Mcm2- 7 ring (Fig. 1A) (16). This finding suggests that DNA unbending would lead to rapid DNA deposition into the Mcm2- 7 central channel. To monitor DNA deposition and test this hypothesis, we developed a FRET- based assay to detect MCM- BPR interactions (Fig. 1A, green dashed box) using an Mcm2- 7- coupled donor at the N terminus of Mcm3 (Mcm2- 73N- 550) and the same DNA- coupled acceptor at the +51 position within BPR (ARS1+51- Cy5). Instead of ORCC, unlabeled ORC with full- length Orc6 was used, allowing for complete helicase loading. In this assay, a high “MCM•+51” EFRET is expected following DNA insertion into the central channel of the first Mcm2- 7 (Fig. 3A). Consistent with this prediction, we observed periods of high EFRET shortly after first Mcm2- 7 arrival (Fig. 3B). In 135 Mcm2- 7 colocalization events with DNA, most of the complexes initially had ORC- engaged BPR, as evidenced by a low MCM•+51 EFRET distribution (centered at 0.12) 1 to 2 s after Mcm2- 7 association (Fig. 3C). In contrast, 10 to 11 s after Mcm2- 7 arrival, the MCM•+51 EFRET values were higher (centered at 0.55), implying that DNA had been deposited into the Mcm2- 7 channel (Fig. 3C). Using the increase to high EFRET (>0.35, see Materials and Methods) as a marker for DNA deposition, the average time between Mcm2- 7 arrival and DNA deposition was 4.9 ± 0.7 s (Fig. 3D). Consistent with DNA deposition being dependent on DNA unbending, the Mcm2- 7 WHD mutants that exhibited reduced DNA unbending (Fig. 2D) also exhibited reduced DNA deposition (SI Appendix, Fig. S4B). Similar distributions are observed for both time to unbending and time to deposition after Mcm2- 7 arrival (Fig. 3D), suggesting a rapid transfer of the BPR DNA from ORC to the Mcm2- 7 central channel. Shortly after the transition to high EFRET upon Mcm2- 7 arrival, we usually observed a decrease of EFRET (e.g., Fig. 3B at 4 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org A B ) . U . A ( y t i s n e n t I e c n e c s e r o u F l 1 0 1 0 2 1 0 0.8 0.6 0.4 0.2 0 T E R F E Low MCM•+51 FRET High MCM•+51 FRET Mcm2-73N-550 Dex , Dem Dex , Aem Total emission Dex , (Dem+Aem) Dex , Aem/(Dem+Aem) 0.12 0.55 1 (cid:31) 2 s N = 135 10 (cid:31)11 s -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 EFRET C y t i s n e D y t i l i b a b o r P D 3 2 1 3 2 1 0 1 i i g n v v r u s n o i t c a r F 0.8 0.6 0.4 0.2 Unbending (N = 228) mean = 3.9 ± 0.3 s Deposition (N = 135) mean = 4.9 ± 0.7 s 0 20 40 t - t arrival (s) 1st Mcm2-7 60 80 100 0 0 5 15 10 t - t arrival (s) 1st Mcm2-7 20 25 Fig. 3. DNA deposition into the Mcm2- 7 central channel exhibits a similar temporal distribution to DNA unbending. (A) Schematic of the experiment monitoring DNA deposition into the Mcm2- 7 central channel, which is indicated by high MCM•+51 EFRET. The FRET probes were Mcm2- 73N- 550 and ARS1+51- Cy5. (B) Example record from the experiment described in (A). Figure descriptions are as Fig. 1C, except that green arrow marks arrival of Mcm2- 7 on DNA. Gray shading over the plots: 1 to 2 s after Mcm2- 7 arrival; yellow shading: 10 to 11 s after Mcm2- 7 arrival. Time resolution = 0.25 s. (C) MCM•+51 EFRET distribution during 1 to 2 s (gray, top plot) and 10 to 11 s after first Mcm2- 7 arrival (yellow, bottom plot). Data were fitted by a global two- component Gaussian mixture model (solid red lines, SI Appendix, Table S1) with global component centers at 0.12 and 0.55 (red dashed lines). The time intervals correspond to the gray and yellow shadings in (B). N: number of Mcm2- 7 colocalization events. (D) Cumulative distribution of time intervals separating the first Mcm2- 7 arrival and the subsequent DNA unbending (black, same as Fig. 2C) or DNA deposition (red). The mean time to reach unbending or deposition ±SE of the mean are shown. Shaded areas: 95% CI; N: number of DNA- bound ORC or Mcm2- 7 molecules that exhibited intensity or EFRET change. 1stMcm2−7 t − t arrival = ∼ 40s ). To ensure that this EFRET decrease is associated with successful helicase loading, we limited our analysis to Mcm2- 7- DNA colocalization events that resulted in high- salt- resistant, loaded Mcm2- 7 hexamers (7, 32). Using data from the experimental setup described in Fig. 3A, we plotted a two- dimensional heat map for these productive loading events to visualize MCM•+51 EFRET values 0 to 100 s after the first Mcm2- 7 arrival (Fig. 4A). We divided the heat map into four time windows (TW, Fig. 4A) and fit a two- component Gaussian model to the EFRET distribution in each TW (Fig. 4B, red curves, SI Appendix, Table S1). The centers of the low and high Gaussian components in TW1 and TW2 were not significantly different (Fig. 4B and SI Appendix, Table S1). Consistent with Fig. 3B, the mixture of two Gaussian components in TW1 and TW2 suggests that most helicase- loading intermediates transitioned from a low- EFRET DNA- bent state to a high- EFRET DNA- deposited state between TW1 and TW2. After DNA deposition, EFRET values decreased in TW3 to a lower level maintained in TW4 (Fig. 4A). Although TW3 shows a gradual decrease in the aggregate heat map (Fig. 4A), single- molecule records show examples of both gradual and sharp EFRET decline (SI Appendix, Fig. S5). Notably, the peak centers of TW3 and TW4 did not over- lap with those of TW1 and TW2 (Fig. 4B), suggesting the presence of new conformational states distinct from the DNA- bent and the DNA- deposited states. Mcm2- 7 Hexamers Slide on dsDNA after DNA Deposition. Based on previous studies indicating that both Mcm2- 7 double hexamers (DHs) and ORC can slide on DNA (5, 6, 33–35), we hypothesized that the EFRET changes after DNA deposition are the result of PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 5 of 12 A TW1 TW2 TW3 TW4 0.7 0.6 0.5 0.4 0.3 0.2 0.1 T E R F E N = 130 20 40 80 t - t arrival (s) 1st Mcm2-7 60 ×10-3 15 y t i s n e D y t i l i b a b o r P 10 5 0 100 i i n o g n 1 n 0.5 a m 0F e 0 r i t c a r B 2 2 y 0 t i s n e D y t i l i 0 b a b o r P 2 0 2 0 0 (cid:31) 4 s TW1 4 (cid:31) 15 s TW2 15 (cid:31) 50 s TW3 50 (cid:31) 100 s TW4 C D 2 0 1 0 2 0 ) . U . A ( y t i s n e n t I e c n e c s e r o u F l T E R F E 0.6 0.4 0.2 0 ARS1 -4 ACS B1 +51 B2 +82 MCM•-4 FRET MCM•+82 FRET Mcm2-73N-550 Dex , Dem Dex , Aem Total Emission Dex , (Dem+Aem) Dex , Aem/(Dem+Aem) 2 0 1 0 2 0 0.6 0.4 0.2 0 E n o i t c a r F F i i g n v v r u s 1 0.8 0.6 0.4 0.2 0 1 0.8 0.6 -4 +51 +82 Acceptor position -4 (N = 43) +51 (N = 106) +82 (N = 121) n o i t c a r F 0.4 0.2 0 0 20 40 60 80 100 t start - t arrival (s) High FRET 1st Mcm2-7 -0.2 0 0.2 0.4 0.6 0.8 1 EFRET 0 50 100 t - t arrival (s) 1st Mcm2-7 0 50 100 t - t arrival (s) 1st Mcm2-7 Fig. 4. Single Mcm2- 7 hexamers slide on DNA after DNA insertion into the central channel. (A) Heat map showing MCM•+51 EFRET distribution changing with time after first Mcm2- 7 arrival (Materials and Methods). Data were from 130 salt- stable helicase loading events taken from experiments described in Fig. 3A. The EFRET probability density (color scale) was calculated using Mcm2- 7 molecules still visible at each time point. The fraction of Mcm2- 7 that remain bound (black curve) and the 95% CI (black shading) are shown in the bottom plot. Because all Mcm2- 7 molecules represent successfully loaded helicases, the decrease in fraction bound Mcm2- 7 was due to termination of the experiment rather than dissociation. The heat map is divided into four time windows (TW) as defined by the dashed line boundaries. TW1: 0 to 4 s after the first Mcm2- 7 arrival; TW2: 4 to 15 s; TW3: 15 to 50 s; and TW4: 50 to 100 s. Time resolution = 1 s. (B) MCM•+51 EFRET in each TW plotted with bin width 0.02. The EFRET distributions of each TW were separately fit to two- component Gaussian mixture models (red curves, SI Appendix, Table S1). The center values of each Gaussian component (red dashed lines) and the SEs of these values (red shadings) are shown. (C) Diagram of ARS1 origin showing the acceptor modification sites used to detect sliding in the MCM•- 4 and MCM•+82 FRET experiments. The +51 site of modification used in the MCM•+51 FRET experiment to detect DNA deposition and sliding is also indicated. (D) Representative intensity records from FRET experiments with Mcm2- 73N- 550 and DNA- coupled acceptors at −4 (Left) or +82 (Right). Green arrow: Mcm2- 73N- 550 association; gray arrows: the first EFRET peak observed; orange arrow: second EFRET peak. (E) Fraction of DNA- bound Mcm2- 7 that exhibited at least one high FRET peak in the MCM•- 4 FRET, MCM•+51 FRET, and MCM•+82 FRET experiments. Error bars: SE. (F) Cumulative distribution of time intervals separating the first Mcm2- 7 arrival and the first observation of high EFRET, including only Mcm2- 7 colocalization events that showed high EFRET. Time resolution of the MCM•+51 experiment (red) is 1 s to allow for direct comparison with the other experiments. Shaded areas: 95% CI; N: the number of Mcm2- 7 colocalization events. Mcm2- 7 sliding on DNA (SI Appendix, Fig. S6A). To confirm Mcm2- 7 sliding, we created two additional DNA substrates, each with a single DNA- coupled acceptor at a position located more than 30 bp (>10 nm) away from the +51 position used to detect DNA deposition (Fig. 4C). The −4 dye (ARS1- 4- Cy5) is intended to detect leftward Mcm2- 73N- 550 sliding, and the +82 dye (ARS1+82- Cy5) rightward sliding (Fig. 4C and SI Appendix, Fig. S6B). In experiments with DNA- coupled acceptors at either the −4 or +82 positions, more than half of the DNA- bound Mcm2- 7 molecules exhibited one or more EFRET peaks (Fig. 4 D and E), implying that the Mcm2- 7 molecules can slide leftward or right- ward along the DNA. Because some molecules displayed more than one EFRET peak (Fig. 4D, orange arrow), we infer that the sliding movement can change direction. For Mcm2- 7 colocaliza- tion events that displayed high EFRET, the time to sliding detected by the −4 and +82 dyes is significantly longer than time to depo- sition after Mcm2- 7 arrival (Fig. 4F), which is consistent with sliding occurring only after initial DNA deposition. Together, these findings suggest that after DNA deposition, Mcm2- 7 single hexamers can slide back and forth on DNA with no preferential direction. Regulation of Mcm2- 7 Sliding. To determine whether Mcm2- 7 ATP- hydrolysis impacts sliding, we performed MCM•+51 FRET experiments with six Mcm2- 7 ATPase mutants (36), each containing a single R→A mutation in one of the six ATPase arginine- finger motifs present in Mcm2- 7. The resulting EFRET heat maps show that each of the Mcm2- 7 ATPase mutants reached a high EFRET state that indicates successful DNA deposition (Fig. 5A). Although we observed a spectrum of different EFRET patterns in the mutants (SI Appendix, Fig. S7), we consistently observed a prolonged duration of the high EFRET state (Fig. 5A) compared to WT Mcm2- 7 (Fig. 4A). Thus, efficient Mcm2- 7 sliding is initiated by at least one round of ATP hydrolysis at each Mcm2- 7 subunit interface. We next investigated the temporal relationship between the start of Mcm2- 7 sliding and the sequential releases of Cdc6 and Cdt1 from the OCCM (Fig. 5 B–F). We performed MCM•+51 FRET experiments in the presence of Cdc6 or Cdt1 labeled with a red- excited dye to monitor Cdc6 or Cdt1 release simultaneously with MCM•+51 EFRET (Fig. 5 B and E). We labeled the N termi- nus of Cdc6 (Cdc6N- 649), which is distant from the Mcm3 N- terminal FRET donor based on structural studies (17) and 6 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org placed the Cdt1- coupled fluorophore on a rigid, N- terminal linker (Cdt1NL- 650) to separate it from the Mcm3 donor fluorophore (Materials and Methods). Control experiments confirmed that minimal FRET was observed between Mcm2- 73N- 550 and Cdc6N- 649 or Cdt1NL- 650 (SI Appendix, Fig. S8). We used alternat- ing red and green laser excitation to monitor protein binding to DNA and FRET: Green excitation enabled monitoring of FRET and DNA association of green- excited Mcm2- 7, while red excita- tion monitored DNA association of red- excited Cdc6 or Cdt1. In these MCM•+51 FRET experiments, Cdc6 or Cdt1 release is represented by a decrease in red- excited fluorescence (Fig. 5 C and F, red arrows). The initiation of Mcm2- 7 sliding is indicated by EFRET decrease below the defined threshold (Fig. 5 C and F, red dotted line, SI Appendix, Fig. S8 B and E). More than 90% of DNA- bound Mcm2- 7 molecules exhibited EFRET decrease only after Cdc6 release (Fig. 5D, positive values, SI Appendix, Fig. S8C). In contrast, we found that 80% of Mcm2- 7 exhibited the EFRET decrease before Cdt1 release (Fig. 5G, negative values, SI Appendix, A 0.8 0.6 T E R F E 0.4 0.2 2RA 0.8 0.6 0.4 0.2 B 3RA Cdc6N-649 n o i t c a r F i i g n n a m e r 0 1 0.5 0 0 N = 151 40 60 20 80 100 0 1 0.5 0 0 N = 66 40 60 20 0.8 0.6 T E R F E 0.4 0.2 4RA 0.8 0.6 0.4 0.2 n o i t c a r F i i g n n a m e r 0 1 0.5 0 0 N = 78 40 60 20 80 100 0 1 0.5 0 0 N = 102 40 60 20 0.8 0.6 T E R F E 0.4 0.2 6RA 0.8 0.6 0.4 0.2 80 100 5RA 80 100 7RA MCM•+51 0.04 0.02 0 0 10 20 30 ×10-2 2 0 -50 0 50 100 150 t drop - t release (s) MCM•+51 Cdc6 Cdt1NL-650 MCM•+51 ×10-2 D ) 1 - s ( y t i s n e D y t i l i b a b o r P E G ) 1 - s ( y t i s n e D y t i l i C 4 2 0 1 0 1 0 2 0 ) . U . A ( y t i s n e t n I e c n e c s e r o u F l T E R F E 0.6 0.3 0 F 6 3 0 1 0 1 0 2 1 0 ) . U . A ( y t i s n e t n I e c n e c s e r o u F l T E R F E 0.6 0.3 0 Cdc6 release Aex , Aem Dex , Dem Dex , Aem Dex , (Dem+Aem) Dex , Aem/(Dem+Aem) 100 50 0 t - t arrival (s) Mcm2-7 Cdt1 release 0 100 50 t - t arrival (s) Mcm2-7 n o i t c a r F i i g n n a m e r 0 1 0.5 0 0 0 1 0.5 0 N = 116 40 60 20 t - t arrival (s) 1st Mcm2-7 80 100 0 2 0.1 1 0 -10 -5 0 5 N = 124 40 60 20 t - t arrival (s) 1st Mcm2-7 80 100 b a b o r 0P ×10-3 15 0 5 10 Probability Density -200 -100 100 t drop - t release (s) MCM•+51 Cdt1 0 Fig. 5. Mcm2- 7 sliding requires ATP- hydrolysis and generally begins after Cdc6 release but before Cdt1 release. (A) MCM•+51 EFRET heat maps of each of the six Mcm2- 7 arginine- finger ATPase mutants. Plots are as described in Fig. 4A. N: the number of Mcm2- 7 colocalization events. Example records are shown in SI Appendix, Fig. S7. (B) Schematic of the MCM•+51 FRET experiment in the presence of Cdc6N- 649. (C) Representative record from the experiment in (B). The red arrow and purple dashed line mark Cdc6N- 649 release. Top plot: acceptor emission during acceptor excitation (Aex, Aem). Fluorescence in light red is contributed by labeled ARS1+51- Cy5. Darker red interval indicates Cdc6N- 649 colocalization with DNA. Second plot: donor emission during donor excitation (Dex, Dem). Third plot: acceptor emission during donor excitation (Dex, Aem). Fourth plot: total emission during donor excitation [Dex, (Dem+Aem)]. Bottom plot: effective FRET efficiency (EFRET) values [Dex, Aem/(Dem+Aem)]. The red dotted line marks the threshold (0.34) separating the high and low EFRET states. Time resolution = 2.7 s. Additional records are shown in SI Appendix, Fig. S8C. (D) Distribution of time intervals from Cdc6N- 649 release to the decrease in MCM•+51 EFRET. A vertical red dashed line marks time interval = 0. Histogram bin width = 10.8 s (4 frames). 42 out of 46 time intervals are positive (Cdc6N- 649 releases before EFRET decrease). Inset: magnified view with bin width = 5.4 (2 frames). E.Schematic of the MCM•+51 FRET experiment in the presence of Cdt1NL- 650. (F) Representative record from the experiment in (E). Figure descriptions are as (C), except the red arrow and green dashed line mark Cdt1NL- 650 release, and the darker red interval in the top plot indicates Cdt1NL- 650 colocalization with DNA. The red dotted line marks the threshold (0.37) separating the high and low EFRET states. Additional records are shown in SI Appendix, Fig. S8F. (G) Distribution of time intervals from Cdt1NL- 650 release to the decrease in MCM•+51 EFRET. A vertical red dashed line marks time interval = 0. Histogram bin width = 21.6 s (8 frames). 46 out of 58 time intervals are negative (Cdt1NL- 650 releases after EFRET decrease). Inset: magnified view with bin width = 2.7 (1 frame). PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 7 of 12 Fig. S8F). Taken together, these findings show that Mcm2- 7 slid- ing primarily occurs after Cdc6 release but before Cdt1 release. The Initial Sliding Helicase- Loading Intermediate Includes ORC, Mcm2- 7, and Cdt1. Because the initial interactions between ORC and Mcm2- 7 remain stable until Cdt1 release (9), our observation that Mcm2- 7 sliding typically begins before Cdt1 release strongly suggests that the initial sliding complex also includes Cdt1 and ORC. To confirm this hypothesis and characterize the timing of ORC sliding relative to other steps in helicase loading, we developed a FRET assay that monitors ORC- ACS interaction. We placed a donor fluorophore on the N- terminal face of ORC (ORCN, see Materials and Methods) and an acceptor fluorophore adjacent to the ACS (ARS1- 4- Cy5) (Fig. 6A). To monitor Mcm2- 7 DNA association simultaneously with this ORCN•- 4 FRET pair, we included Mcm2- 7 labeled with a red- excited fluorophore at the N terminus of Mcm4 (Mcm2- 74N- 650), which displayed minimal FRET with ORCN due to its long unstructured N- terminal tail (SI Appendix, Fig. S9A). In this ORCN•- 4 FRET experiment with labeled Mcm2- 7, we observed high EFRET before Mcm2- 7 arrival, indicating stable ORC binding at the ACS prior to Mcm2- 7 recruitment (Fig. 6B and SI Appendix, Fig. S9B). We consistently observe the loss of high ORCN•- 4 EFRET after Mcm2- 7 arrival (Fig. 6B, dashed line, SI Appendix, Fig. S9B), indicating ORC displacement from the ACS. Importantly, we consistently detected a decrease in ORCN•- 4 FRET for events that led to salt- stable DH formation, indicating that ORC displacement represents an intermediate step on path- way to successful helicase loading (SI Appendix, Fig. S9B). By determining the time interval between Mcm2- 7 arrival and ORCN•- 4 EFRET decrease, we found that ORC leaving the ACS occurs significantly after DNA unbending or deposition into the Mcm2- 7 central channel (Fig. 6C). The decrease in ORCN•- 4 EFRET could be due to ORC sliding away from ACS or ORC flipping associated with MO formation. If ORC begins sliding simultaneously with Mcm2- 7, ORCN•- 4 EFRET decrease would fall between Cdc6 and Cdt1 release (Fig. 5 B–G). However, if the EFRET decrease corresponds to ORC flipping, which occurs after Cdt1 release (9), we would expect ORCN•- 4 EFRET decrease to follow Cdt1 release. To distinguish between these two possibilities, we determined when ORC leaves the ACS relative to Cdc6 and to Cdt1 release. To this end, we performed ORCN•- 4 EFRET experiments in the presence of red- excited Cdc6N- 649 or Cdt1N- 649. Consistent with the predicted distances from structural studies (17), neither Cdc6N- 649 nor Cdt1N- 649 exhibits significant EFRET to ORCN (SI Appendix, Fig. S9A). ORCN•- 4 EFRET consistently decreased after Cdc6 release (80/84 positive values in Fig. 6D and SI Appendix, Fig. S10A). In contrast, ORCN•- 4 EFRET almost always decreased before Cdt1 release (89/91 negative values in Fig. 6E and SI Appendix, Fig. S10B). These distributions show that ORC movement away from ACS predominantly begins between Cdc6 release and Cdt1 release, matching the time interval during which Mcm2- 7 starts sliding. Importantly, we detected negligible background sliding of ORC in the absence of other helicase- loading factors (SI Appendix, Fig. S2), consistent with ORC sliding being temporally controlled during helicase loading. To detect ORC sliding, we examined EFRET between ORC and a DNA probe distant from the ACS. By placing the ARS1 DNA acceptor at the +51 position, a high ORCN•+51 EFRET is only expected if ORC moves to the proximity of the +51 dye (SI Appendix, Fig. S6C). Importantly, ORC bound at B2 would place the donor fluorophore too far (>8.2 nm) from the +51 DNA fluorophore to exhibit high EFRET (SI Appendix, Fig. S6C). In the presence of labeled Cdt1N- 649, we observed high ORCN•+51 EFRET peaks that occurred before Cdt1 release (Fig. 6F and SI Appendix, Fig. S10C). These EFRET peaks indicate that ORC slides away from the ACS while still bound to the C- terminal face of Mcm2- 7 and before flipping to form the MO complex. Given our previous observation that OC1M sliding is bidirec- tional (Fig. 4E), we would expect ORC to revisit the ACS during sliding. If true, then in the ORCN•- 4 EFRET experimental setup, after an initial decrease in EFRET, we would sometimes observe increases that represent ORC reengagement with the ACS. Indeed, 42.9% of ORC molecules exhibited a return to high ORCN•- 4 EFRET signal after the initiation of sliding (SI Appendix, Fig. S11A). Consistent with ORC having low affinity for the ACS under these conditions, these rebinding events were short- lived relative to ORC independently binding the ACS (SI Appendix, Fig. S11B). Interestingly, only 3.3% of ORC molecules showed a high ORCN•- 4 EFRET signal after Cdt1 release, suggesting that Cdt1 release inhibits ORC rebinding to the ACS (SI Appendix, Fig. S11A). Together, our studies show that sliding of ORC and Mcm2- 7 on the DNA is temporally controlled. This process is triggered by the release of Cdc6 from the OCCM but does not require Cdt1 release. Because the initial ORC- Mcm2- 7 interactions remain stable until Cdt1 release (9), we conclude that the OC1M com- plex, consisting of ORC, Cdt1, and Mcm2- 7, exhibits bidirec- tional sliding on dsDNA. ORC sliding away from the ACS represents a third mechanism to reduce ORC affinity for DNA, preparing it to rapidly release from the DNA upon Cdt1 release and disruption of the initial ORC- Mcm2- 7 interface. Together, these events drive ORC release from the ACS and facilitate its subsequent flipping, B2 element binding, and MO complex formation. Discussion Our previous studies showed that a single ORC can mediate load- ing of both helicases found at licensed origins (7, 9). This process requires the same ORC to sequentially bind two distinct and oppositely oriented sites on the DNA. This model raises the ques- tion of how ORC is released from its initial high- affinity binding site ACS. Using sm- FRET assays to examine ORC and Mcm2- 7 inter- actions with DNA during recruitment of the first Mcm2- 7, our studies revealed a coordinated series of changes that promote ORC site exchange. We propose an updated helicase loading model that incorporates the findings from our study (Fig. 7). DNA- bound ORC consistently induces DNA bending by simultaneously inter- acting with the ACS and the BPR (Fig. 1D). Shortly after recruit- ment of the first Mcm2- 7 onto ORC- Cdc6 bound to bent DNA (Fig. 7, 1st Mcm2- 7 recruitment), DNA unbending is triggered by interactions formed during OCCM assembly, such as between ORC and the WHDs of Mcm4 and Mcm6 (Fig. 2D). This leads to the rapid DNA deposition into the Mcm2- 7 central channel (Figs. 3D and 7, circled 1). Loss of the ORC- BPR interaction destabilizes ORC on DNA (Fig. 1E), and subsequent Cdc6 release (Fig. 7, circled 2) further weakens ORC- DNA interactions (SI Appendix, Fig. S1C). Mcm2- 7 ATP hydrolysis activity initiates ORC- Cdt1- Mcm2- 7 (OC1M) sliding on DNA (Figs. 5A and 7, circled 3), which both fully releases ORC from the ACS and enables access to B2 elements located at variable distances from the ACS. The sequential events of DNA unbending, Cdc6 release, and OC1M sliding progressively reduce ORC stability on DNA (Fig. 7, gray wedge) to promote ORC dissociation from its strong site, allowing for its subsequent rebinding to the inverted B2 ele- ment (Fig. 7, ORC Flip). After ORC flipping, B2- bound ORC 8 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org A B ORCN•-4 + Mcm2-74N-650 F ORCN•+51 + Cdt1N-649 (E) Cdt1 (B) Mcm2-7 (D) Cdc6 -4 ORC at ACS: High ORCN•-4 FRET DNA unbending (N = 402) DNA deposition (N = 121) ORC leaving ACS (N = 113) C 1 n o i t c a r F e v i t a u m u C l 0.8 0.6 0.4 0.2 0 0 150 100 50 t - t arrival (s) 1st Mcm2-7 200 3 ) . U 1.5 . A ( y t i s n e t n I e c n e c s e r o u F l 0 1 0 2 1 0 2 1 0 1 T E R F E 0.5 0 ORCN Mcm2-74N-650 Aex , Aem Dex , Dem Dex , Aem Total emission Dex , (Dem+Aem) Dex , Aem/(Dem+Aem) 6 3 0 2 1 0 1 0 2 ) . U . A ( y t i s n e t n I e c n e c s e r o u F l T E R F E 0.5 0 Cdt1N-649 8 4 0 Aex , Aem Dex , Dem 1 0 ORCN Dex , Aem 2 0 2 0 1 0.5 0 Total emission Dex , (Dem+Aem) 0 1 Dex , Aem/(Dem+Aem) 250 300 350 400 450 500 Time (s) E D 950 1050 200 300 Time (s) ORCN•-4 + Cdc6N-649 ORCN•-4 + Cdt1N-649 ) 1 - s ( y t i s n e D y t i l i b a b o r P ×10-2 3 2 1 0 -20 0 20 40 60 80 100 t drop - t release (s) ORC •-4N Cdc6 ×10-2 2 1 0 ×10-2 -40 -20 0 2 1.5 1 0.5 ) 1 - s ( y t i s n e D y t i l i b a b o r P 0 [-420, -300] -250-200-150-100-50 0 t drop - t release (s) ORC •-4N Cdt1 Fig. 6. ORC slides away from the ACS as a complex with Mcm2- 7 and Cdt1. (A) Schematic of experiments that monitor ORC- ACS interaction. Black lines indicate the proteins that are labeled with a red- excited fluorophore in the ORCN•- 4 FRET experiments presented in the corresponding figures. (B) Example record of the ORCN•- 4 experiment in the presence of Mcm2- 74N- 550. Top plot: acceptor emission during acceptor excitation (Aex, Aem). Fluorescence in light red is contributed by DNA. Red arrow and darker red segments of the intensity plot indicate Mcm2- 74N- 650 colocalization. Second plot: donor emission during donor excitation (Dex, Dem). Green arrow marks ORCN arrival. Third plot: acceptor emission during donor excitation (Dex, Aem). Fourth plot: Total emission during donor excitation [Dex, (Dem+Aem)]. Bottom plot: EFRET values [Dex, Aem/(Dem+Aem)]. The dashed line marks the decrease of EFRET, indicating ORC movement away from ACS. More records are shown in SI Appendix, Fig. S9B. Time resolution = 2.7 s. (C) Cumulative distribution for the time interval separating the first Mcm2- 7 arrival and one of the following three events: DNA unbending (black), DNA deposition into Mcm2- 7 central channel (red), and ORC movement away from ACS (blue). The data from DNA unbending (black) and DNA deposition (red) are from the same experiments as shown in Fig. 3D but with 1 s time resolution to allow direct comparison with the ORC movement data (blue). Shadings: 95% CI; N: number of colocalization events. (D) Distribution of the time intervals separating ORCN•- 4 EFRET decrease and Cdc6 release, bin width = 5.4 s (2 frames). 80 out of 84 time intervals are positive (EFRET decreases after Cdc6 release). Example intensity traces are shown in SI Appendix, Fig. S10A. (E) Distribution of the time intervals separating ORCN•- 4 EFRET decrease and Cdt1 release, bin width = 13.5 s (5 frames). Outliers of values from the range [−420, −300] are grouped in the red bar. Inset: magnified histogram. 89 out of 91 time intervals values are negative (EFRET decreases before Cdt1 release). Example records are shown in SI Appendix, Fig. S10B. (F) Two example records of ORCN•+51 FRET experiments in the presence of red- labeled Cdt1N- 649. A schematic of the FRET probes used in this experiment is shown in SI Appendix, Fig. S6C. Gray arrows indicate FRET peaks of ORC sliding to the proximity of the +51 DNA dye. The green dashed line marks Cdt1 release. More records are shown in SI Appendix, Fig. S10C. in the context of the MO complex recruits a second Mcm2- 7 in the inverted orientation, completing helicase loading (Fig. 7, 2nd Mcm2- 7 recruitment and DH). Stepwise Reduction in ORC Stability on DNA. Our finding that ORC is progressively destabilized on DNA is consistent with previous structural studies. ORC makes numerous contacts with the BPR when the DNA is in a bent state (15), and we show that mutation of the amino acids responsible for a subset of these interactions significantly reduces ORC- DNA stability (SI Appendix, Fig. S1D). Similar ORC- BPR interactions observed in metazoan ORC suggest that DNA bending is a common mechanism for regulating ORC–DNA complex stability (37). A role for the Mcm4 and Mcm6 WHDs in triggering DNA unbending is consistent with structural studies showing that these domains orient the Mcm2/5 gate next to the BPR for subsequent deposition (38). Notably, the Orc1 basic patch region that interacts with the ACS in the ORC- ACS structure becomes disordered in the OCCM structure, which may further contribute to the lowered ACS affinity of ORC (15, 17). Following DNA unbending, Cdc6 release serves a dual purpose. Structural studies of the ORC–Cdc6–DNA complex show that Cdc6 binding to ORC completes an ORC- Cdc6 protein ring around the DNA (17, 28). Cdc6 dissociation is thus required to reopen the ORC ring for DNA release from its central channel during ORC flipping. Moreover, the WHD and initiator- specific motif of Cdc6 form extensive specific contacts with the ACS (28). Both of these functions are consistent with previous data showing PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 9 of 12 DNA binding strength of ORC ① DNA unbending and deposition ② Cdc6 ORC/Cdc6 Mcm2-7/ Cdt1 ACS 1st Mcm2-7 recruitment OCCM ③ Sliding OC1M Cdt1 ORC Flip DH 2nd Mcm2-7 recruitment MO Fig. 7. A helicase loading model. We propose that DNA unbending and deposition, Cdc6 release, and OC1M sliding (red circled numbers) sequentially lower the DNA binding strength of ORC (gray wedge). These events facilitate ORC to release DNA from its central channel and rebind to an inverted binding site (ORC Flip), which enables loading of the second Mcm2- 7 in the correct orientation and completes formation of the Mcm2- 7 DH. that Cdc6 enhances ORC binding to the ACS (39) and our data that Cdc6 stabilizes DNA- bound ORC (SI Appendix, Fig. S1C). The decreased affinity of ORC for the ACS after Cdc6 release would lower the energy barrier for OC1M sliding on DNA, which would further reduce specific ORC- DNA interactions. Even upon reencountering the ACS site during sliding, ORC in this destabi- lized state fails to maintain stable interactions with the ACS, resulting in its rapid disengagement from the site (SI Appendix, Fig. S11). The combined effects of DNA unbending, Cdc6 release, and OC1M sliding promote ORC dissociation from DNA neces- sary for ORC flipping. Our studies also provide insights into the importance of the ordered release of Cdc6 and Cdt1 from the OCCM. Cdt1 release is associated with the disruption of the interaction between ORC and the C- terminal domain of Mcm2- 7 involved in initial Mcm2- 7 recruitment (9). Thus, the retention of Cdt1 as sliding commences means that both ORC and Mcm2- 7 remain as a com- plex during initial sliding. Once ORC associates with nonspecific DNA, we predict that the stability of ORC on DNA is highly dependent on its interaction with the C- terminal domain of Mcm2- 7. When this ORC- Mcm2- 7 interaction is disrupted dur- ing Cdt1 release, ORC is poised to rapidly dissociate from DNA and seek a new binding site. Two key questions remain to be addressed by future studies. First, how does ORC remain tethered to the first Mcm2- 7 after disen- gaging from the Mcm2- 7 C- terminal domains and DNA? One possibility is that ORC is tethered to the N- terminal domain of Mcm2- 7 via Orc6, potentially using the same Orc6- Mcm2 inter- action observed in the MO complex (8). Given the long unstruc- tured domain between the Orc6 N- terminal domain and the rest of ORC (27), such an interaction would allow ORC to identify a second binding site without releasing into solution. Second, what causes ORC to preferentially bind to a secondary site in the inverted orientation? Biochemical studies have found that Cdt1 binding to Mcm2- 7 relieves the autoinhibitory activity of Mcm6 WHD that blocks Mcm2- 7 binding to ORC (40). It is likely that Cdt1 release before MO formation results in the Mcm6 WHD to return to the autoinhibitory state. This change would prevent ORC reassociation with the Mcm2- 7 C terminus and reduce the likelihood of ORC rebinding to the ACS. Consistent with this hypothesis, we see that release of Cdt1 from the OC1M complex significantly reduces ORC rebinding to the ACS (SI Appendix, Fig. S11A). Future experiments will be necessary to determine the precise mechanism of ORC bind- ing to the inverted B2 site. Energy Requirements that Lead to ORC Flipping. Because the DNA binding strength of ORC is sequentially decreased to facilitate ORC flip and MO formation (Fig. 7, gray wedge), we propose that the resulting increase in the free energy of ORC- DNA interactions is compensated by additional stabilizing interactions or by energy derived from ATP hydrolysis. First, destabilization caused by the loss of ORC- BPR interaction during DNA unbending is offset by stabilizing interactions formed between DNA and the central channel of Mcm2- 7 during DNA deposition (MCM- BPR in Figs. 1 and 3), as well as interactions between ORC and Mcm2- 7 that involve Mcm4 and Mcm6 WHDs (Fig. 2D). Second, ATP hydrolysis by Cdc6 has been reported to drive Cdc6 release (41). Third, consistent with previous studies showing that Mcm2- 7 ATP hydrolysis is required for helicase loading (36, 42), our study demonstrated that Mcm2- 7 ATP hydrolysis is required for the onset of OC1M sliding (Fig. 5A). A likely explanation for this requirement is the ATP control of a transition of the Mcm2- 7 β- hairpin loops from a DNA- engaged (as seen in the OCCM) (3, 17) to a weakly DNA- bound open conformation (as seen in the DH) that favors DNA sliding (2, 43). Although ATP hydrolysis is important for initiating sliding, our data suggest that OC1M movement involves passive diffusion instead of ATP- powered translocation. We observed evidence for movement in both directions from the initial ORC- Mcm2- 7 bind- ing sites (Fig. 4 D and E), which differs from the biased random walk exhibited by the CMG helicase during active translocation along ssDNA (43). In addition, ORC, single or double Mcm2- 7 hexamers, and potential loading intermediates including both ORC and Mcm2- 7 have been shown to passively slide on dsDNA (5, 6, 8, 35, 44, 45). Analogous sliding has been observed in many different protein complexes including MutS (46, 47) and Type III restriction enzymes (48). Role of Sliding during Helicase Loading. Due to the variable distance between the ACS and B2 elements in different origins, the first Mcm2- 7 hexamer must slide a variable distance in either direction for ORC to gain access to the inverted B2 binding site(s). In the case of ARS1, in which the OCCM would occupy the ACS and B2 elements simultaneously, OC1M sliding would overcome 10 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org steric obstruction of B2 by the first Mcm2- 7. At other origins, such sliding would similarly provide ORC and the first Mcm2- 7 access to one or more potential B2 elements. Our data also agrees with previous studies that single roadblocks on either side of the origin do not impact helicase loading (49) because the OC1M complex is able to slide away from the roadblocks to access distant B2 elements. The lack of sliding directionality is also consistent with observations that artificial origins with B2 positioned on either side of ACS can support helicase loading (22). A strong requirement for a specific B2 element is observed when flanking roadblocks limit OC1M sliding such that only a single B2 element is accessible (49), or in artificial origins where only a single B2- like sequence is present (22). As such, sliding enables ORC to search for another near- ACS sequence in the inverted orientation. Though any ACS- related sequence on either side can serve as a secondary ORC landing site, the distance between the two binding sites is likely confined in vivo by origin- flanking nucleosomes that limit sliding (50, 51). As a result, the B2 elements that are found to be important in vivo (18, 20) are likely the most accessible secondary sites relative to the ACS. Our study identifies the OC1M as the first sliding intermediate and establishes that sliding is prevented before this intermediate is formed, but it is likely that later helicase- loading intermediates also slide on DNA to search for a stable secondary binding site. Because previous studies proposed that ORC remains tethered to the helicase to prevent its dissociation while exchanging binding sites (8, 9), it is possible that the DNA- bound Mcm2- 7 single hexamers travel along DNA while the tethered ORC actively searches for a suitable landing site. Moreover, given that ORC can search DNA via diffusion (34, 35), and the MO complex has been found to occupy DNA sites other than the initial recruitment site (8), the MO complex can likely diffuse on DNA prior to ORC identifying a stable binding site in the inverted orientation. Future experiments will be required to demonstrate and characterize slid- ing of other helicase- loading intermediates. Materials and Methods Nomenclature for Fluorescently Modified Proteins and DNAs. We use a superscript notation to describe proteins and DNA with fluorescent or dark quencher dye modifications. The superscript notation consists of two parts sepa- rated by a dash: the site of modification and the type of modification. If the type of modification is denoted by a number, it refers to the Dylight™ fluorophore conjugated. For example, ARS1+51- BHQ2 refers to ARS1 origin labeled at the +51 nucleotide position relative to ACS with a BHQ- 2; Mcm2- 74N- 550 refers to Mcm2- 7 labeled with Dylight 550 at the Mcm4 N terminus. We used abbreviations for two specific ORC constructs: ORCC for ORC labeled at the C- terminal face with Dylight 550 and ORCN for ORC labeled at the N- terminal face with Dylight 550 (see subsections below). SI Appendix, Table S2 summarizes all the fluorescently- labeled proteins and DNA used in this study. +51- Cy5 +51- BHQ2 - 4- Cy5 +82- Cy5 , ARS1 , ARS1 Preparation of Internally Modified DNA (ARS1 , ). Internally modified ARS1 DNAs were assembled ARS1 by ligating two overlapping PCR fragments. A list of oligos (Integrated DNA Technologies) used in this study is provided in SI Appendix, Table S3. The inter- nal Cy5 (iCy5) and internal BHQ2 (iBHQ2) fluorophores were directly attached through the DNA phosphate backbone. The biotinylated PCR fragments were generated with 5′ biotin- labeled oligos, and the dye- coupled PCR fragments were generated using internally labeled oligos that anneal to the overlap region. The two PCR fragments were ligated as described previously (49) to yield the final 1.4- kb DNA construct containing a biotin end modification and an internal dye modification near the point of ligation. Protein Purification. Wild- type (WT) Mcm2- 7 and WT ORC were purified as described previously (36). WT Cdc6 and Cdc6N- 649 were purified as described (7, 31). N . An Saccharomyces cerevisiae strain coexpressing all Preparation of ORC codon- optimized ORC subunits (52) was used to prepare internally labeled ORCN. A 3xFLAG tag was inserted at the ORC1 N terminus, and an s6 tag (GDSLSWLLRLLN) replaced Orc1 residues 314 to 325, an unstructured region of Orc1 (2, 15–17). ORC was purified using FLAG affinity resin (Sigma- Aldrich) as previously described (9). ORC, SFP synthase (NEB), and Dy550- CoA (Thermo Scientific) were incubated at a 1:3:30 ratio for 30 min at room temperature. Labeled ORC was then purified on a Superdex 200 Increase 10/300 gel filtration column (Cytiva). Sortase- Mediated Protein Labeling and Purification. Fluorescence labeling at the N or C terminus of proteins was performed via sortase- mediated coupling and purified as described (9). C C- BPR and ORC . S. cerevisiae strains overexpressing the Preparation of ORC codon- optimized ORC subunits with the indicated mutations were grown, arrested, induced, harvested, and lysed as described above. Following elution from the FLAG resin, peak fractions were coupled to Dylight 550 via sortase as above. 3N- 550 4N- 650 (WT and RA Mutants) and Mcm2- 7 Preparation of Mcm2- 7 (WT and WHD Mutants). S. cerevisiae strains coexpressing all Mcm2- 7 subunits with the appropriate modifications were grown, arrested, induced, and harvested. Mcm2- 7 protein was purified using FLAG resin and labeled via sortase- mediated conjugation as above. For the Mcm2- 7 WHD mutants, 4ΔWHD is a C- terminal truncation mutant lacking residues 854 to 933; 6ΔWHD lacks residues 839 to 1,017; 4Δ6ΔWHD combines both truncations. N- 649 N- 649 NL- 650 , Cdt1 , and Cdt1 . Cdc6N- 649 and Cdt1N- 649 Preparation of Cdc6 were purified as described (7). To separate the Cdt1- conjugated fluorophore from the donor fluorophore on Mcm2- 73N- 550, a 3× FLAG tag and a rigid alpha- helical (EAAAK)10 linker (53) was placed at the N terminus of Cdt1 after the sortase- recognition motif GGG. Cdt1 was purified via FLAG resin and labeled at the N terminus via sortase- mediated conjugation as described above. Single- Molecule Assays. A micromirror total internal reflection microscope was used to perform multiwavelength single- molecule imaging (25). Glass slides were functionalized with PEG and Biotin- PEG; fiducial markers and biotinylated DNA were coupled to the slide as described (9). All reactions were performed in buffer con- taining 25 mM HEPES- KOH pH 7.6, 0.3 M potassium glutamate, 5 mM Mg(OAc)2, 3 mM ATP, 1 mM dithiothreitol, 1 mg/mL bovine serum albumin, with an oxygen scavenging system (glucose oxidase/catalase), and 2 mM Trolox (54). ORCC•+51 and ORCN•- 4 experiments contained 1 nM ORC, 1 µM 60- bp nonspecific DNA gen- erated by annealing two oligonucleotides (SI Appendix, Table S3), and 3 nM Cdc6 if specified. For helicase loading assays, 5 to 10 nM Mcm2- 7/Cdt1 was added. DNA molecules labeled with Alexa488 or Cy5 were identified before the experiments using 488- nm or 633- nm excitation, respectively. Three different protocols were used for experimental acquisition: 1) reactions containing only green- excited proteins: continuous 532- nm excitation of specified exposure time; 2) reactions containing green- excited ORC, red- excited Mcm2- 7, and quencher- labeled DNA: simultaneous 532- nm and 633- nm excitation; 3) reactions containing green- excited ORC, red- excited Mcm2- 7, and red- excited DNA: alternating 532- nm and 633- nm excitation. Cy5 internally labeled DNA that did not photobleach was identified by 633- nm excitation after the experiments. To identify salt- resistant helicases, a high- salt buffer containing 25 mM HEPES- KOH pH 7.6, 0.5 M KCl, and 5 mM Mg(OAc)2 was applied to wash away loading intermediates at the end of the experiments. FRET Data Analysis. Data were analyzed as described previously (7) and fluores- cence intensity values were corrected for background fluorescence as described (55). Apparent FRET efficiency (EFRET) calculations were performed as described (9). EFRET values were maximum likelihood fit using the models and yielding the fit parameters given in SI Appendix, Table S1. The EFRET threshold used to differentiate the two EFRET states was defined as the position of the trough in the fit. An EFRET transition was considered to have taken place once the EFRET value crossed the threshold for at least two consecutive frames. ORC Apparent Dissociation Rate in High and Low EFRET States. In the ORCC•+51 FRET experiment, we used the threshold 0.28 to distinguish between the high and low EFRET states. The frequency of ORC release (Fig. 1E) in its high and low EFRET states was determined by first identifying the EFRET state of ORC in PNAS 2023 Vol. 120 No. 30 e2305556120 https://doi.org/10.1073/pnas.2305556120 11 of 12 the current frame n. The next frame (frame n+1) was then assessed to determine whether ORC was released into solution, which is indicated by the decrease of green- excited fluorescence back to background levels. The frequency of release was calculated by dividing the number of ORC dissociation events by the total number of frames in the corresponding EFRET state. EFRET Heat Maps. The EFRET heat maps in Figs. 4A and 5A were generated using MATLAB code adapted from (56), in the folder: /plot/plot_2d_histo/. The code used two- dimensional kernel density estimation, with a normal kernel function (SD of time axis = 5 s and EFRET axis 0.05) to generate a heat map of EFRET values against time (0 to 100 s after first Mcm2- 7 arrival). Each vertical slice represents the probability density function of EFRET values at the particular time interval. The time axis has a resolution of 1 s to match the time resolution of the datasets, and the EFRET axis has a resolution of 0.005. Normalization was performed such that the density estimate at each time slice integrates to one. Ensemble OCCM Formation Assays. Ensemble OCCM formation assays (SI Appendix, Fig. S1A) were done as described previously (36). Data, Materials, and Software Availability. Source data for the single- molecule experiments are provided as Matlab “intervals” files that can be read and manipulated by the program imscroll (https://github.com/gelles- brandeis/CoSMoS_Analysis) (57). The source data are archived at doi: 10.5281/ zenodo.7814499 (58). ACKNOWLEDGMENTS. This work was supported by NIH grants R01 GM147960 (S.P.B. and J.G.) and R01 GM81648 (J.G.). A.Z. was supported by an Angela Leong Fellowship. S.P.B. is an investigator with the Howard Hughes Medical Institute. This work was supported in part by the Koch Institute Support Grant P30- CA14051 from the NCI. We thank the Koch Institute Swanson Biotechnology Center for technical support. 1. 2. A. Costa, J. F. X. Diffley, The initiation of eukaryotic DNA replication. Annu. Rev. Biochem. 91, 107–131 (2022). Y. Noguchi et al., Cryo- EM structure of Mcm2- 7 double hexamer on DNA suggests a lagging- strand DNA extrusion model. Proc. Natl. Acad. Sci. U.S.A. 114, E9529–E9538 (2017). 3. N. Li et al., Structure of the eukaryotic MCM complex at 3.8 Å. Nature 524, 186–191 (2015). 4. F. Abid Ali et al., Cryo- EM structure of a licensed DNA replication origin. Nat. Commun. 8, 2241 (2017). C. Evrin et al., A double- hexameric MCM2- 7 complex is loaded onto origin DNA during licensing of eukaryotic DNA replication. Proc. Natl. Acad. Sci. U.S.A. 106, 20240–20245 (2009). D. Remus et al., Concerted loading of Mcm2–7 double hexamers around DNA during DNA replication origin licensing. Cell 139, 719–730 (2009). S. Ticau, L. J. Friedman, N. A. Ivica, J. Gelles, S. P. Bell, Single- molecule studies of origin licensing reveal mechanisms ensuring bidirectional helicase loading. Cell 161, 513–525 (2015). T. C. R. Miller, J. Locke, J. F. Greiwe, J. F. X. Diffley, A. Costa, Mechanism of head- to- head MCM double- hexamer formation revealed by cryo- EM. Nature 575, 704–710 (2019). S. Gupta, L. J. Friedman, J. Gelles, S. P. Bell, A helicase- tethered ORC flip enables bidirectional helicase loading. Elife 10, e74282 (2021). 5. 6. 7. 8. 9. 30. M. Guerrero- Puigdevall, N. Fernandez- Fuentes, J. Frigola, Stabilisation of half MCM ring by Cdt1 during DNA insertion. Nat. Commun. 12, 1746 (2021). 31. J. Frigola, D. Remus, A. Mehanna, J. F. X. Diffley, ATPase- dependent quality control of DNA replication origin licensing. Nature 495, 339–343 (2013). 32. S. Donovan, J. Harwood, L. S. Drury, J. F. X. Diffley, Cdc6p- dependent loading of Mcm proteins onto pre- replicative chromatin in budding yeast. Proc. Natl. Acad. Sci. U.S.A. 94, 5611–5616 (1997). 33. M. J. Scherr, S. A. Wahab, D. Remus, K. E. Duderstadt, Mobile origin- licensing factors confer resistance to conflicts with RNA polymerase. Cell Rep. 38, 110531 (2022). 34. D. Duzdevich et al., The dynamics of eukaryotic replication initiation: Origin specificity, licensing, and firing at the single- molecule level. Mol. Cell 58, 483–494 (2015). 35. H. Sánchez et al., DNA replication origins retain mobile licensing proteins. Nat. Commun. 12, 1908 (2021). 36. S. Kang, M. D. Warner, S. P. Bell, Multiple functions for Mcm2–7 ATPase motifs during replication initiation. Mol. Cell 55, 655–665 (2014). 37. F. Bleichert, A. Leitner, R. Aebersold, M. R. Botchan, J. M. Berger, Conformational control and DNA- binding mechanism of the metazoan origin recognition complex. Proc. Natl. Acad. Sci. U.S.A. 115, E5906–E5915 (2018). 10. C. Liang, M. Weinreich, B. Stillman, ORC and Cdc6p interact and determine the frequency of 38. Y. Zhai et al., Unique roles of the non- identical MCM subunits in DNA replication licensing. Mol. Cell initiation of DNA replication in the genome. Cell 81, 667–676 (1995). 67, 168–179 (2017). 11. S. P. Bell, B. Stillman, ATP- dependent recognition of eukaryotic origins of DNA replication by a 39. C. Speck, Z. Chen, H. Li, B. Stillman, ATPase- dependent cooperative binding of ORC and Cdc6 to multiprotein complex. Nature 357, 128–134 (1992). origin DNA. Nat. Struct. Mol. Biol. 12, 965–971 (2005). 12. A. Amasino, S. Gupta, L. J. Friedman, J. Gelles, S. P. Bell, Regulation of replication origin licensing by 40. A. Fernández- Cid et al., An ORC/Cdc6/MCM2- 7 complex is formed in a multistep reaction to serve as ORC phosphorylation reveals a two- step mechanism for Mcm2- 7 ring closing. Proc. Natl. Acad. Sci. U.S.A., in press (2023). https://doi.org/10.1101/2023.01.02.522488 a platform for MCM double- hexamer assembly. Mol. Cell 50, 577–588 (2013). 41. F. Chang et al., Cdc6 ATPase activity disengages Cdc6 from the pre- replicative complex to promote 13. H. Rao, B. Stillman, The origin recognition complex interacts with a bipartite DNA binding site within DNA replication. Elife 4, e05795 (2015). yeast replicators. Proc. Natl. Acad. Sci. U.S.A. 92, 2224–2228 (1995). 42. G. Coster, J. Frigola, F. Beuron, E. P. Morris, J. F. X. Diffley, Origin licensing requires ATP binding and 14. A. Rowley, J. H. Cocker, J. Harwood, J. F. Diffley, Initiation complex assembly at budding yeast hydrolysis by the MCM replicative helicase. Mol. Cell 55, 666–677 (2014). replication origins begins with the recognition of a bipartite sequence by limiting amounts of the initiator, ORC. The EMBO Journal 14, 2631–2641 (1995). 43. D. R. Burnham, H. B. Kose, R. B. Hoyle, H. Yardimci, The mechanism of DNA unwinding by the eukaryotic replicative helicase. Nat. Commun. 10, 2159 (2019). 15. N. Li et al., Structure of the origin recognition complex bound to DNA replication origin. Nature 559, 44. J. Gros et al., Post- licensing specification of eukaryotic replication origins by facilitated Mcm2- 7 217–222 (2018). sliding along DNA. Mol. Cell 60, 797–807 (2015). 16. Z. Yuan et al., Structural mechanism of helicase loading onto replication origin DNA by ORC- Cdc6. 45. P. Goswami et al., Structure of DNA- CMG- Pol epsilon elucidates the roles of the non- catalytic Proc. Natl. Acad. Sci. U.S.A. 117, 17747–17756 (2020). polymerase modules in the eukaryotic replisome. Nat. Commun. 9, 5061 (2018). 17. Z. Yuan et al., Structural basis of Mcm2–7 replicative helicase loading by ORC–Cdc6 and Cdt1. Nat. 46. R. R. Iyer, A. Pluciennik, V. Burdett, P. L. Modrich, DNA mismatch repair: Functions and mechanisms. Struct. Mol. Biol. 24, 316–324 (2017). Chem. Rev. 106, 302–323 (2006). 18. Y. Marahrens, B. Stillman, A yeast chromosomal origin of DNA replication defined by multiple 47. R. Qiu et al., Large conformational changes in MutS during DNA scanning, mismatch recognition functional elements. Science 255, 817–823 (1992). and repair signalling. EMBO J. 31, 2528–2540 (2012). 19. G. M. Wilmes, S. P. Bell, The B2 element of the Saccharomyces cerevisiae ARS1 origin of replication requires specific sequences to facilitate pre- RC formation. Proc. Natl. Acad. Sci. U.S.A. 99, 101–106 (2002). 20. F. Chang et al., High- resolution analysis of four efficient yeast replication origins reveals new insights into the ORC and putative MCM binding elements. Nucleic Acids Res. 39, 6523–6535 (2011). 21. J. F. Theis, C. S. Newlon, Two compound replication origins in Saccharomyces cerevisiae contain redundant origin recognition complex binding sites. Mol. Cell Biol. 21, 2790–2801 (2001). 48. F. W. Schwarz et al., The helicase- like domains of type III restriction enzymes trigger long- range diffusion along DNA. Science 340, 353–356 (2013). 49. M. D. Warner, I. F. Azmi, S. Kang, Y. Zhao, S. P. Bell, Replication origin–flanking roadblocks reveal origin- licensing dynamics and altered sequence dependence. J. Biol. Chem. 292, 21417–21430 (2017). 50. M. L. Eaton, K. Galani, S. Kang, S. P. Bell, D. M. MacAlpine, Conserved nucleosome positioning defines replication origins. Genes Dev. 24, 748–753 (2010). 51. N. M. Berbenetz, C. Nislow, G. W. Brown, Diversity of eukaryotic DNA replication origins revealed by 22. G. Coster, J. F. X. Diffley, Bidirectional eukaryotic DNA replication is established by quasi- symmetrical genome- wide analysis of chromatin structure. PLoS Genet. 6, e1001092 (2010). helicase loading. Science 357, 314–318 (2017). 52. V. Posse, E. Johansson, J. F. X. Diffley, Eukaryotic DNA replication with purified budding yeast 23. H. Rao, Y. Marahrens, B. Stillman, Functional conservation of multiple elements in yeast proteins. Methods Enzymol. 661, 1–33 (2021). chromosomal replicators. Mol. Cell Biol. 14, 7643–7651 (1994). 53. R. Arai, H. Ueda, A. Kitayama, N. Kamiya, T. Nagamune, Design of the linkers which effectively 24. L. J. Friedman, J. Gelles, Multi- wavelength single- molecule fluorescence analysis of transcription separate domains of a bifunctional fusion protein. Protein Eng. 14, 529–532 (2001). mechanisms. Methods 86, 27–36 (2015). 54. D. J. Crawford, A. A. Hoskins, L. J. Friedman, J. Gelles, M. J. Moore, Visualizing the splicing of single 25. L. J. Friedman, J. Chung, J. Gelles, Viewing dynamic assembly of molecular complexes by multi- pre- mRNA molecules in whole cell extract. RNA 14, 170–179 (2008). wavelength single- molecule fluorescence. Biophys. J. 91, 1023–1031 (2006). 55. L. de Jesús- Kim et al., DDK regulates replication initiation by controlling the multiplicity of Cdc45- 26. R. Roy, S. Hohng, T. Ha, A practical guide to single- molecule FRET. Nat. Methods 5, 507–516 (2008). 27. S. Chen, S. P. Bell, CDK prevents Mcm2–7 helicase loading by inhibiting Cdt1 interaction with Orc6. GINS binding to Mcm2- 7. Elife 10, e65471 (2021). 56. J. Gelles, Single- molecule data analysis algorithms. GitHub. https://github.com/gelles- brandeis/ Genes Dev. 25, 363–372 (2011). jganalyze. Accessed 9 November 2022. 28. X. Feng et al., The structure of ORC–Cdc6 on an origin DNA reveals the mechanism of ORC activation 57. L. J. Friedman, J. Gelles, Tools for analyzing CoSMoS image data. GitHub. https://github.com/gelles- by the replication initiator Cdc6. Nat. Commun. 12, 3883 (2021). brandeis/CoSMoS_Analysis. Accessed 4 August 2016. 29. J. M. Schmidt et al., A mechanism of origin licensing control through autoinhibition of S. cerevisiae 58. A. Zhang, Changing protein- DNA interactions promote ORC binding site exchange during ORC·DNA·Cdc6. Nat. Commun. 13, 1059 (2022). replication origin licensing. Zenodo. https://zenodo.org/record/7814499. Accessed 19 June 2023. 12 of 12 https://doi.org/10.1073/pnas.2305556120 pnas.org
10.1073_pnas.2220576120	RESEARCH ARTICLE \| GENETICS OPEN ACCESS Derepression of Y-linked multicopy protamine-like genes interferes with sperm nuclear compaction in D. melanogaster Jun I. Parka,b,c , and Yukiko M. Yamashitad,e,f,1 , George W. Belld Edited by Mariana Wolfner, Cornell University, Ithaca, NY; received December 3, 2022; accepted March 17, 2023 Across species, sperm maturation involves the dramatic reconfiguration of chromatin into highly compact nuclei that enhance hydrodynamic ability and ensure paternal genomic integrity. This process is mediated by the replacement of histones by sperm nuclear basic proteins, also referred to as protamines. In humans, a carefully bal- anced dosage between two known protamine genes is required for optimal fertility. However, it remains unknown how their proper balance is regulated and how defects in balance may lead to compromised fertility. Here, we show that a nucleolar pro- tein, modulo, a homolog of nucleolin, mediates the histone-to-protamine transition during Drosophila spermatogenesis. We find that modulo mutants display nuclear compaction defects during late spermatogenesis due to decreased expression of auto- somal protamine genes (including Mst77F) and derepression of Y-linked multicopy Mst77F homologs (Mst77Y), leading to the mutant’s known sterility. Overexpression of Mst77Y in a wild-type background is sufficient to cause nuclear compaction defects, similar to modulo mutant, indicating that Mst77Y is a dominant-negative variant interfering with the process of histone-to-protamine transition. Interestingly, ectopic overexpression of Mst77Y caused decompaction of X-bearing spermatids nuclei more frequently than Y-bearing spermatid nuclei, although this did not greatly affect the sex ratio of offspring. We further show that modulo regulates these protamine genes at the step of transcript polyadenylation. We conclude that the regulation of protamines mediated by modulo, ensuring the expression of functional ones while repressing dominant-negative ones, is critical for male fertility. protamine \| spermatogenesis \| Drosophila In many species, spermatids undergo the process of nuclear compaction, an essential process to produce sperm that are capable of fertilization (1–3). Nuclear compaction is critical for the sperm’s hydrodynamic performance and protecting the paternal genome against mutagens (4–6). Nuclear compaction involves the dramatic chromatin reorgan- ization mediated by the histone-to-protamine transition (1–5, 7, 8). Sperm nuclear basic proteins, also referred to as protamines, are small, positively charged proteins that replace histone-based nucleosomes to achieve the extreme degree of DNA compaction often seen in sperm (2). As such, these protamines are required for fertility across many different species (4). Although protamines are essential for fertility, they are rapidly evolving across species (4, 9, 10), where the primary sequence, the number, and the functionality of protamine genes are not well conserved. For example, human and mouse protamine genes, PRM1 and PRM2, are required for fertility (4, 6, 7), while PRM2 has become nonfunctional in bulls and boars (4, 11). Closely related Drosophila species have independently evolved many different protamine-like genes (10): Drosophila melanogaster has Mst35Ba and Mst35Bb (also known as ProtA and ProtB), which are the most similar to mammalian PRM1 and PRM2 (3, 12, 13), as well as Mst77F, Prtl99C, and Y-linked multicopy Mst77Y, with evidence that several more uncharacterized genes may also be involved (14). In contrast, in Drosophila simulans, there is just one orthologous copy of the ProtA/B gene (Prot) as well as one ortholog each for Mst77F (GD12157) and Prtl99c (GD21472). D. simulans lacks Mst77Y (10, 14), but have evolved their own clade-specific genes that contain large regions of protamine sequences (Dox family genes), which are not present in D. melanogaster (15, 16). Surprisingly, while ProtA and ProtB are most similar to their mammalian counterparts, they are not required for fertility in D. melanogaster (12); instead, more divergent genes Mst77F and Prtl99C are essential (17–19). The potential function of the D. melanogaster-specific multicopy locus of Mst77F homologs (the Mst77Y genes) is unknown (20, 21). Interestingly, it has been observed that mammals appear to feature their own species-specific ratios of protamine dosage (2, 11, 22, 23), and in humans, even small alterations in the ratio of PRM1 and PRM2 are associated with infertility (2, 23–26), suggesting that a specific balance of protamines is important for sperm DNA packaging. However, it remains unknown Significance Protamines are small, highly positively charged proteins that are required for packaging DNA to produce mature sperm with highly condensed nuclei capable of fertilization. Even small changes in the dosage of protamines in humans is associated with infertility. Our work reveals the presence of dominant-negative protamine genes on the Y chromosome of Drosophila melanogaster and shows that the precise expression of functional protamines and repression of dominant-negative protamines is a critical process to ensure male fertility. Author affiliations: aLife Sciences Institute, University of Michigan, Ann Arbor, MI 48109; bDepartment of Cell and Developmental Biology, University of Michigan cMedical Medical School, Ann Arbor, MI 48109; Scientist Training Program, University of Michigan Medical School, Ann Arbor, MI 48109; dWhitehead Institute for Biomedical Research, Cambridge, MA 02142; eDepartment of Biology, School of Science, Massachusetts Institute of Technology, Cambridge, MA 02142; and fHHMI, Cambridge, MA 02142 Author contributions: J.I.P. and Y.M.Y. designed research; J.I.P. performed J.I.P. contributed new reagents/analytic tools; J.I.P., G.W.B., and Y.M.Y. analyzed data; and J.I.P. and Y.M.Y. wrote the paper. research; The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2220576120/-/DCSupplemental. Published April 10, 2023. PNAS 2023 Vol. 120 No. 16 e2220576120 https://doi.org/10.1073/pnas.2220576120 1 of 8 why carefully balanced protamine expression is important and how it is achieved to support fertility. While studying D. melanogaster modulo mutants, we discovered that modulo transheterozygotic mutant causes misregulation of protamine genes. modulo mutant spermatids display decreased nuclear incorporation of protamine-like protein Mst77F and increased incorporation of its Y-linked homolog, Mst77Y, which is barely incorporated in the wild type, leading to a DNA com- paction defect that explains the reported sterility of modulo mutant. Our data indicate that Mst77Y likely acts as a dominant-negative form of Mst77F, interfering with the process of histone-to-protamine transition. Interestingly, Mst77Y has disproportionate effects on spermatids carrying an X chromosome, leading to biased decom- paction of X-bearing spermatid nuclei, although it does not lead to large effects on the sex ratio of offspring. We further find that modulo is involved in safeguarding polyadenylation of Mst77F transcript over that of the Y-linked Mst77Y. Our study reveals a mechanism of protamine gene expression mediated by modulo, balancing the correct ratio of protamine gene expression to ensure male fertility. Results modulo Mutant Is Defective in Sperm Nuclear Compaction. Modulo is the Drosophila homolog of Nucleolin, a nucleolar protein implicated in RNA processing (27, 28). Although modulo-mutant males have been known to be sterile (27, 29), the cytological defects that lead to their sterility have not been characterized. We find that the modulo transheterozygote mutant (mod L8/mod 07570) exhibits defects in nuclear morphology transformation during late spermiogenesis. In wild-type males, postmeiotic spermatid nuclei undergo well-documented morphological changes (1), from round spermatid stage, to “leaf ” stage, to “canoe” stage, resulting in highly condensed “needle-” stage nuclei, which is accompanied by the histone-to-protamine transition (Fig. 1A). Although modulo- mutant germ cells proceeded through spermatogenesis normally, including early nuclear compaction (Fig. 1 B and C), the modulo mutant exhibited striking “decompaction” of the nuclei after reaching the canoe stage, coinciding with the individualization of spermatids (Fig. 1 D and E). Immunofluorescence (IF) staining using anti-dsDNA, which has been previously used to assess the compaction status of spermatid nuclei (30), revealed that defective spermatid nuclei of modulo mutant are indeed decompacted (Fig. 1 F and G). Decompacting nuclei are initially negative via Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL), a method used to identify DNA breaks that occur during apoptosis (Fig. 1 H and I), then become TUNEL positive (Fig. 1J), suggesting that decompaction is not the result of cell death, but may rather be a cause. Overall, 100% of the modulo-mutant testes exhibited a nuclear decompaction phenotype (Fig. 1K), and it appeared that all nuclei eventually become decompacted and die, filling the distal end of the testis with cellular debris (SI Appendix, Fig. S1 A and B). The eventual death of all sperm nuclei likely results in the entire lack of mature sperm in the seminal vesicles (SI Appendix, Fig. S1 C and D) and the modulo mutant’s known sterility. modulo Mutant Fails in Histone-to-Protamine Transition. Because nuclear decompaction in the modulo mutant occurs at stages when sperm chromatin is known to undergo reorganization through the histone-to-protamine transition, we explored whether the modulo mutant is defective in this process. Histone-to- protamine transition occurs step wise: 1) histone modification and removal, 2) transition protein incorporation then removal, and 3) protamine incorporation (1). IF staining revealed that modulo-mutant spermatids undergo proper histone removal and transient transition protein incorporation (SI Appendix, Fig. S2 A– F), but fail to properly accumulate ProtA/B and Mst77F (Fig. 2 A and B). Moreover, using a specific antibody (SI Appendix, Fig. S3 A and B), we found that Mst77Y, Y-linked multicopy homologs of Mst77F (20, 21) (SI Appendix, Fig. S4A), strongly accumulated in modulo-mutant spermatid nuclei, whereas it was barely detectable in control (Fig. 2 C–G), suggesting that Mst77Y is aberrantly expressed in the modulo mutant. As the deletion of Mst77F and ProtA/B does not cause nuclear decompaction as severe as that of the modulo mutant (18), we infer that the incorporation of Mst77Y (in addition to the depletion of Mst77F and ProtA/B) causes the observed catastrophic nuclear decompaction seen in the modulo- mutant spermatids. Ectopic Expression of Mst77Y Alone Is Sufficient to Cause Nuclear Decompaction. The Mst77Y genes have several interesting features. First, the gene locus contains 18 copies of Mst77F homolog (SI Appendix, Fig. S4 A and B), which originated from a single event of Mst77F translocation to the Y chromosome, followed by gene amplification (20, 21). Second, many of the Mst77Y genes have mutations, which have resulted in changes in the position and number of critical arginine, lysine, and cysteine residues believed to be important for protamine function (4, 31). Other mutations have resulted in premature truncations (SI Appendix, Fig. S4B) (21). Note that anti-Mst77Y antibody was generated by using multiple peptides from Mst77Y that are distinct from Mst77F to increase specificity. The antibodies were also designed to be able to identify all copies of Mst77Y, which feature similar mutations and were tested to be able to identify both full-length Mst77Y (Mst77Y-12) and normally truncated Mst77Y (Mst77Y-3) (SI Appendix, Figs. S4 B and S5 A–C). Because Mst77Y’s mutations likely alter Mst77F’s normal function, we hypothesized that Mst77Y genes may function as a dominant-negative form of Mst77F. Accordingly, Mst77Y’s aberrantly high expression in the modulo mutant may interfere with the process of normal histone- to-protamine transition. To test the possibility that the expression of Mst77Y causes the nuclear decompaction phenotype, we generated transgenic lines that express Mst77Y under a male germline-specific tubulin pro- moter (β2-tubulin promoter) (32–34). From the 18 copies of Mst77Y homologs present on the Y chromosome (20, 21) we generated two lines expressing either Mst77Y-12 (a full-length version) or Mst77Y-3 (a truncated version due to premature stop codon) (SI Appendix, Figs. S4B and S6), as the transcripts of these two genes have been previously detected by qRT-PCR (21). Strikingly, expression of either Mst77Y-3 or Mst77Y-12 recapitu- lated a nuclear decompaction phenotype similar to that seen in modulo mutant (Fig. 3 A–D): 45.7% and 43.2% of testes exam- ined exhibited nuclear decompaction upon expression of Mst77Y-3 or Mst77Y-12, respectively (Fig. 3E), suggesting that high Mst77Y expression is sufficient to cause nuclear decompaction in a subset of spermatids. Notably, in contrast to the eventual decompaction of all spermatids seen in the modulo mutant, Mst77Y overexpres- sion alone does not cause sterility. We speculate that this might be due to the added effect of the decreased incorporation of Mst77F and ProtA/B, in addition to high Mst77Y incorporation, seen in the modulo mutant. Given that Barckmann et al. utilized the same promoter to overexpress autosomal Mst77F and did not observe such nuclear compaction defects during spermiogenesis (32) as we observed with Mst77Y overexpression, we infer that Mst77Y may act as a dominant-negative form, perhaps interfering with the function of Mst77F (Discussion). This notion is further supported by the 2 of 8 https://doi.org/10.1073/pnas.2220576120 pnas.org A Histone : Protamine exchange round spermatid Elongation leaf stage spermatocyte growth/ maturation mitoses GSCs meiotic divisions Seminal Vesicle canoe stage needle stage Nuclear Shaping, nuclear compaction, and individualization ) + / 0 7 5 7 0 d o m ( l o r t n o C ) 0 7 5 7 0 d o m / 8 L d o m ( t n a t u M B DAPI D DAPI phalloidin F N DAPI dsDNA H C DAPI TUNEL mod 07570/+ mod 07570/+ mod 07570/+ mod 07570/+ f o e g a t n e c r e P C E G I DAPI TUNEL J K g n i y a l p s i d s e t s e t e p y t o n e h p n o i t c a p m o c e d 100 50 0 * 100% (105/105) 0% (0/67) control mod 07570/+ mutant modL8/mod 07570 DAPI TUNEL mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 mod L8/mod 07570 Fig. 1. Sterility of modulo mutant is accompanied by defective spermatid chromatin compaction. (A) Schematic of spermatogenesis in Drosophila proceeding from germline stem cells to mature sperm. Proceeding from meiotic divisions onward, only nuclei are depicted. (B and C) Representative images of canoe-stage nuclei stained with DAPI (gray) in control males (mod07570/+) (B), and modulo-mutant males (modL8/mod07570) (C). (D and E) Representative images at the stage shortly before individualization stained with DAPI (gray) and phalloidin (cyan, to visualize the individualization complex) in control males (mod07570/+) (D) and modulo- mutant males (modL8/mod07570) (E). Although all nuclei eventually become decompacted in modulo-mutant males, individualization complex (marked by phalloidin staining) appears to be normally formed. Yellow arrowheads indicating decompacted nuclei. (F and G) Representative images of late canoe to needle stages stained with anti-dsDNA (red) and DAPI (gray) in control (mod07570/+) (F) and modulo-mutant males (modL8/mod07570) (G). N: needle-stage spermatids that do not stain for dsDNA due to advanced DNA compaction, C: canoe-stage spermatids that are less compact and positive for anti-dsDNA staining. (H–J) Staining via Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) (magenta) of needle-shaped spermatid cysts in control (mod07570/+) (H) and mutant (modL8/mod07570) males without (I) or with (J) TUNEL signal. (K) Percentage of decompaction phenotype in modulo-mutant vs. wild-type males. * indicates P < 0.001 (unpaired Student’s t test assuming unequal variances in five independent experiments). n (total number of testes counted per genotype) is presented on the bar graph. fact that a truncated version (Mst77Y-3) also causes the decom- paction phenotype. Indeed, spermatid cysts of transgenic males expressing Mst77Y-3 exhibited uneven Mst77F staining, sug- gesting that some nuclei fail to accomplish proper Mst77F incor- poration (SI Appendix, Fig. S5 D and E). It is important to note that the nuclear decompaction was most prominently observed when males were raised in 25 °C after their parents were raised at 18 °C (Methods). Interestingly, using DNA Fluorescence in situ hybridization (FISH) to distinguish X- vs. Y-bearing sper- matids, we found that overexpression of Mst77Y results in biased demise of X-bearing spermatids, where 61.8% of decompacting nuclei were X-bearing, compared to only 38.2% being Y-bearing (Fig. 3 F and G). It is important to note that this bias is not due to differential efficiency of hybridization of X chromosome vs. Y chromosome DNA FISH probes: Leaf to canoe stage sperma- tids of control males (SI Appendix, Fig. S7 A and B), as well as leaf to canoe stage spermatids of modulo-mutant males (before they exhibit decompaction defects), exhibited ~50:50 ratio of X:Y signal (SI Appendix, Fig. S7 C and D), further suggesting that decompaction is biased toward X-bearing spermatids. However, a fertility assay revealed only a minor increase in the male progeny compared to sex chromosome–matched controls (51.8% vs. 47.8%, P = 0.0005) (SI Appendix, Fig. S8A). Likewise, only a small degree of sex ratio distortion was observed in modulo heterozygous mutant, compared to sex chromosome– matched control (SI Appendix, Fig. S8B) (Discussion). Together, these results suggest that Mst77Y acts as a dominant-negative form of Mst77F, interfering with the incorpo- ration of normal protamines and leading to spermatid nuclear decompaction. Modulo Promotes Polyadenylation of Autosomal Mst77F Transcript. How does modulo regulate the expression of Mst77F and Mst77Y? Modulo protein is expressed in the nucleolus of spermatogonia and spermatocytes, but is down-regulated prior to the meiotic divisions (Fig. 4 A and B), days earlier than the stages in which its mutant phenotype manifests. Protamine genes are known to be transcribed many days prior to the sperm nuclear compaction process in both flies and mice (3, 32, 35). Interestingly, we found that Mst77F transcripts colocalize with Modulo in the spermatocyte nucleolus, while Mst77Y transcripts localize adjacent to the nucleolus (Fig. 4C). These results prompted us to examine whether Mst77F and/or Mst77Y transcripts may be deregulated in modulo mutant. Indeed, we found that Mst77F messenger RNA (mRNA) is dramatically reduced in modulo mutant, whereas Mst77Y mRNA was increased approximately threefold using qRT-PCR of polyA-selected RNA (Fig. 4D). However, when total RNA was used for qRT-PCR or total RNA sequencing, we found that both Mst77F and Mst77Y transcripts were increased in modulo mutant (Fig. 4D and SI Appendix, Fig. S9A). RNA FISH, which visualizes total RNA, also confirmed the increase of both Mst77F and Mst77Y transcripts in modulo mutant (SI Appendix, PNAS 2023 Vol. 120 No. 16 e2220576120 https://doi.org/10.1073/pnas.2220576120 3 of 8 DAPI Prot A/B Mst77F DAPI Mst77F Prot A/B A B A′ A″ A‴ B′ B″ B‴ Control (mod 07570/+) Mutant (mod L8/mod 07570) G DAPI Mst77Y Basal End E Basal End ) + / 0 7 5 7 0 d o m ( l o r t n o C ) 0 7 5 7 0 d o m / 8 L d o m ( t n a t u M C 1 0.5 0 e g a t s e o n a c - e t a l f o n o i t r o p o r P Y 7 7 t s M r o f g n n i a t s i s t s y c d e p a h s - e l d e e n r o control mod 07570/+ mutant modL8/mod 07570 D D′ F F′ Fig. 2. Nuclear decompaction in modulo-mutant spermatids is associated with decreased incorporation of Mst77F and increased incorporation of Mst77Y. (A and B) Representative images of late canoe-stage nuclei stained with DAPI (gray), anti-Prot A/B (cyan), and anti-Mst77F (magenta) in control (mod07570/+) (A) and mutant (modL8/mod07570) (B) males. Split channel view of DAPI (A' and B'), anti-Mst77F (A'' and B''), and anti-Prot A/B (A''' and B''') in control (A) and mutant (B) males. (C–F) Representative images of canoe-stage and needle-stage spermatids at the basal end of testis stained with DAPI (gray) and anti-Mst77Y (green) in control (mod07570/+) (C and D) and mutant (modL8/mod07570) (E and F) males. Split channel view of anti-Mst77Y in control (D') and mutant (F') males. Dotted lines outline the testis. (G) Proportion of canoe-stage cysts with nuclei positive for Mst77Y staining in mutant (modL8/mod07570) vs. control (mod07570/+) males. * indicates P ≤ 0.001 (unpaired Student’s t test assuming unequal variances) with n=10 testes in control and n=9 testes in mutant males from 2 independent experiments. Exact P-values are listed SI Appendix, Table S1. Fig. S9B). Furthermore, total RNA-Seq and qRT-PCR did not detect any splicing defects of Mst77F or Mst77Y in modulo mutant (SI Appendix, Fig. S10). Collectively, these results suggest that Modulo specifically regulates transcripts of Mst77F and Mst77Y at the step of polyadenylation. Given that Modulo protein and Mst77F transcript colocalize in the nucleolus, we speculate that Mst77F is directly regulated by Modulo, whereas increased mRNA level of Mst77Y may be an indi- rect consequence of reduced functional Mst77F mRNA. Interestingly, RNA FISH using poly(T) probes revealed that poly(A) signal encir- cles the nucleolus in wild-type spermatocytes, whereas markedly less poly(A) was detected around the nucleolus in the modulo mutant (Fig. 4 E and F), further suggesting that modulo may function to facilitate polyadenylation of transcripts localized to the nucleolus. Our findings are consistent with the known importance of polyade- nylation to sperm-specific transcripts, such as protamines, which must be translationally repressed for long periods(36–39). Taken together, these results suggest that modulo plays an essential role in sperm nuclear compaction by facilitating maturation of canonical Mst77F transcript over that of Y-linked Mst77Y (Fig. 4G). Discussion The present study reveals a regulatory mechanism mediated by a nucleolar protein Modulo that balances the expression of protamine subtypes in D. melanogaster. This finding may represent a similar theme to what is seen in the fragile balance of PRM1 and PRM2 in mammalian fertility (2, 7, 24, 25). In the case of Mst77Y, Y-linked 4 of 8 https://doi.org/10.1073/pnas.2220576120 pnas.org A y w DAPI C 2-Tub-Mst77Y3/+ DAPI E i l g n y a p s d s i t s e t i B mod L8/mod 07570 DAPI D 2-Tub-Mst77Y12/+ DAPI f o n o i t r o p o r P F G X * 95 (38.2%) Y 154 (61.8%) e p y t o n e h p n o i t c a p m o c e d 1.0 0.5 ns * * 0 0 UAS- Mst77Y-12/+ (no driver) 2-Tub- Mst77Y-3/+ 2-Tub- Mst77Y-12/+ n = 56 61 40 DAPI X Y X (TAGA) Y (AATAC) F′ F″ Fig. 3. Mst77Y overexpression is sufficient to cause nuclear decompaction and causes biased decompaction of X chromosome-bearing spermatids. (A–D) Representative images of needle-stage nuclei stained with DAPI (gray) showing normal morphology in control (y w) (A), decompaction phenotype in modulo- mutant males (modL8/mod07570) (B), transgenic males expressing Mst77Y-3 (truncated copy) (C) or Mst77Y-12 (full-length copy) (D) driven by β2-tubulin promoter. IF confirming overexpression shown in SI Appendix, Fig. S5 A–C. (E) Proportion of testes displaying decompaction phenotype in transgenic Mst77Y males. Control (UAS–Mst77Y-12/+) does not express Mst77Y-12 due to the absence of driver. * indicates P ≤ 0.001 (unpaired Student’s t test assuming unequal variance), ns indicates P > 0.05, n = 56 testes in control, n = 61 testes in β2-tub–Mst77Y-3/+ condition, n = 40 in β2-tub–Mst77Y-12/+ condition from three independent experiments. (F) Representative images of DNA Fluorescence in situ hybridization of decompacted spermatids in Mst77Y-3-expressing males using TAGA-Cy3 (magenta, X-specific probe) (F') and AATAC-Cy5 (cyan, Y-specific probe) (F''). (G) Percentage of decompacted haploid nuclei containing X chromosome vs. Y chromosome in Mst77Y-3-expressing males. * indicates P(X ≥ 154) < 0.001 (exact binomial distribution) assuming P = 0.5 with n = 249 nuclei counted from three independent experiments. Exact P-values listed in Table S1. multicopy Mst77F homologs, our study suggests that they have the ability to act as dominant-negative protamines and thus must be carefully regulated/repressed. The present study also confirmed that Mst77Y genes are expressed as proteins as suggested previously by the finding that several of the copies contain complete open-reading frames (21) and is also consistent with small RNA sequencing reveal- ing that the Mst77Y locus is not a source of small RNAs (40). We showed that overexpression of Mst77Y dominantly inter- feres with Mst77F incorporation, leading to decompaction of sperm nuclei and their demise. Mst77Y genes feature differences from their autosomal homolog that further support the idea that they are dominant-negative versions of Mst77F and interfere with sperm chromatin compaction. Mst77Y-12, which retains the full ORF of Mst77F (SI Appendix, Fig. S4), exhibits 87% overall sequence homology to autosomal Mst77F. At the domain/motif level, the MST-HMG-box domain, suggested to be important for DNA binding (14), maintains 100% homology, while the coiled-coil motif and C-terminal domain maintain only ~79.5% and ~85% homology, respectively (SI Appendix, Fig. S6B). It has been shown that the N-terminal domain of Mst77F, which con- tains the coiled-coil motif, interacts with the C-terminal domain to induce multimerization to mediate DNA compaction (41). The changes to Mst77Y at important regions may thus influence the multimerization of protamines and the formation of proper sperm chromatin structure, by interfering with the ability of the canon- ical version to multimerize. The notion that Mst77Y behaves as a dominant-negative version of Mst77F is further supported by the fact that overexpression of Mst77Y-3, a truncated version which does not contain the C-terminal domain (SI Appendix, Fig. S6B), is still sufficient to cause defects in nuclear compaction (Fig. 3C). What is the potential “function” of dominant-negative pro- tamines? We propose a few nonmutually exclusive possibilities. First, dominant-negative protamines may participate in meiotic drive, as suggested by recent works in D. simulans (15, 16) as well as D. melanogaster (10). Indeed, our data suggest that Mst77Y has the ability to disproportionally affect X-bearing spermatids. While this did not result in a large sex ratio distortion in offspring (SI Appendix, Fig. S8), this ability to harm a subset of developing spermatids during postmeiotic development may indicate the pos- sibility that these protamine variants could be exploited by a meiotic drive system to unleash its own selfish purpose. Intriguingly, studies on the Winters sex-ratio meiotic drive system in D. simulans revealed that the driver, Dox, contains a large portion of the Protamine gene (15, 16). While it has not been confirmed whether this protamine-like region is essential for sex ratio distortion, the derepression of Dox does seem to cause nuclear defects during sper- miogenesis (42). We propose that a drive system that would be able to localize a dominant-negative protamine such as Mst77Y to a subset of spermatids containing one homolog over another could be quite successful at achieving bias. Alternatively, the dominant-negative version of a protamine may be utilized when spermatogenesis needs to be aborted (similar to the concept of “programmed cell death”), for example under stress conditions. In such a case, dominant-negative protamines (such as Mst77Y) can be up-regulated to lead to abortive spermatogenesis. In such a sce- nario, a dominant-negative protamine may have a beneficial func- tion for the organism. Yet another possibility that may contribute toward the rapid divergence of protamines is that protamine genes evolve to optimally package the genome, which may be greatly influenced by the composition of the most abundant sequences in a given genome, i.e., repetitive DNA such as satellite DNA. As these repetitive sequences are known to rapidly diverge across species (43), protamine genes may have to adapt to accommodate diverged repetitive DNA sequences, leading to rapid divergence and/or emer- gence of multiple protamine genes to optimally package different repetitive DNA with distinct structure/sequence. In such a scenario, PNAS 2023 Vol. 120 No. 16 e2220576120 https://doi.org/10.1073/pnas.2220576120 5 of 8 A A′ n i r a l l i r b i f p f g - d o M I P A D p f g - d o M B B′ B″ e g r e M p f g - d o M n i r a l l i r b i f DAPI Mod-gfp Mst77F Mst77Y Mod-gfp C C′ Mst77F Mst77Y C″ C‴ D n i e g n a h C d o F d e z i l l a m r o N l o r t n o c o t . l e r t n a t u m o u d o m l 5 4 3 2 1 0 RT-qPCR method: F 7 7 t s M A y o P l I P A D A y o p l F 7 7 t s M G p = 0.0166 p = 0.0166 p = 0.0248 p = 0.0248 p = 0.0021 p = 0.0021 p < 0.001 p < 0.001 0 0 Mst77F Mst77Y Mst77F Mst77Y Poly(A) selection Total RNA y w mod L8/mod 07570 E E′ F F′ E″ F″ Modulo AAAAA Mst77F Nucleolus Mst77Y AAAAA Fig. 4. Modulo localizes to the nucleolus and functions to promote polyadenylation of Mst77F. (A and B) Localization of Modulo to the nucleolus in the apical tip of the testis (A) and in the spermatocyte nuclei (B). Males expressing Modulotagged with Green fluorescent protein (gfp) at the C-terminus (yellow) stained with anti-fibrillarin (magenta), a nucleolar marker, and DAPI (gray). Dotted lines outline the testis (A) and nucleus (B). (C) RNA FISH for Mst77F and Mst77Y transcripts in wild-type spermatocyte nucleus. DAPI (gray), Modulo–gfp (yellow), Mst77F (magenta), and Mst77Y (cyan). Split channel view of Modulo-gfp (C'), Mst77F transcript (C''), and Mst77Y transcript (C''') in spermatocytes. Dotted lines outline the nucleus. (D) qRT-PCR following polyA selection (dark gray) or using total RNA qRT-PCR (light gray) in modulo-mutant males (modL8/mod07570) vs. sibling control males (mod07570/+) assessing levels of Mst77F (magenta) and Mst77Y (cyan). Data were normalized to Rp49 and control. Mean ± SD from three technical replicates is shown. P-values are listed (unpaired Student’s t test assuming unequal variance on untransformed ddct values). Similar results were obtained from two biological replicates. Primer locations are shown in SI Appendix, Fig. S11A. (E and F) RNA FISH for polyA (magenta) and Mst77F transcript (cyan) in control (y w) (F) vs. modulo-mutant males (modL8/mod07570) (E), counterstained with DAPI (gray). Split channel view of polyA-containing transcripts (E' and F') and Mst77F transcripts (E'' and F'') in control (E) and mutant males (F). Mst77F probe was used to identify nucleolus. Yellow arrowhead indicates polyA-containing RNA-encircling nucleolus. (G) Model for Modulo function in the nucleolus. fine-tuning the expression of different protamine genes may be critical. Additionally, if any protamine genes have evolved to opti- mally package certain satellite DNA, conversion of such protamine into a dominant-negative version can immediately target the chro- mosome that harbors the given satellite DNA, leading to meiotic drive that selectively harms the specific chromosome. This possibility 6 of 8 https://doi.org/10.1073/pnas.2220576120 pnas.org is intriguing as the Segregation Distorter (SD) meiotic drive system in D. melanogaster is known to target Responder satellite DNA repeats (44–46) and exhibits sperm nuclei decompaction similar to what is observed in this study (30). The possibility that dominant- negative protamines are involved in the decompaction of spermatid nuclei in the SD drive system remains to be studied. Taken together, our study identified a mechanism by which var- ious protamine variants are coordinately regulated at the posttran- scriptional level, possibly to achieve balanced expression of multiple protamine variants. A similar mechanism may be at play to fine-tune the expression levels of protamine variants in human and mouse, disruption of which is associated with compromised fertility. Methods Fly Husbandry and Strains. All fly stocks were raised on standard Bloomington medium at 25 °C, and young flies (1- to 3-d-old adults) were used for all exper- iments unless otherwise specified. Flies used for wild-type experiments were the standard laboratory wild-type strain y w (y1w1). The following fly stocks were used: modulo07570/TM3 [Bloomington Drosophila Stock Center (BDSC): 11795], moduloL8/TM3 (BDSC: 38432), and C(1)RM/C(1;Y)6, y1w1f1/0 (BDSC: 9460). The β2-tubulin promoter sequence used for producing Mst77Y overexpression was generously provided by Peiwei Chen and Alexei Aravin. The Mst77Y transgenic flies were generated by phiC31 site–directed integra- tion into the Drosophila genome. For UAS–Mst77Y-12, β2-tubulin–Mst77Y-3, and β2-tubulin–Mst77Y-12 transgenic lines, the Mst77Y overexpression sequences in D. melanogaster were synthesized by gene synthesis service from Thermo Fisher Scientific (GeneArt Gene Synthesis) and were cloned into pattB vector to insert into specific integration site on second chromosome (attP40) (SI Appendix, Fig. S5C and Table S2). All injection and selection of flies containing integrated transgene were performed by BestGene Inc. Because UAS–Mst77Y-12 transgene was injected to the same host fly strain as β2-tubulin–Mst77Y-3, and β2-tubulin–Mst77Y-12 trans- genic lines, we used this (without gal4 driver) as a “background-matched control.” Modulo–gfp strain was constructed using CRISPR-mediated knock-in of a Green fluorescent protein (gfp)-tag at the C terminus of Modulo (Beijing Fungene Biotechnology Co.) (SI Appendix, Table S3). Sex Ratio Assay. Individual 1-d-old males raised for at least one generation at 18 °C were crossed with 3× 1- to 3-d-old virgin y w females at 25 °C. After 1 d, males were removed. This was done to maximize the proportion of males exhib- iting decompaction phenotype described in Fig. 3. Females were left to produce embryos for a total of 5 d before cleared. Following the onset of eclosion, sex of offspring was scored for 10 consecutive days. RNA Fluorescent In Situ Hybridization. All solutions used were Rnase free. Testes from 1- to 3-d-old flies were dissected in 1X phosphate buffered saline (PBS) and fixed in 4% formaldehyde in 1X PBS for 30 min. Then, the testes were washed briefly in PBS and permeabilized in 70% ethanol overnight at 4 °C. For strains expressing gfp (i.e., Modulo–gfp), the overnight permeabilization in 70% ethanol was omitted. The testes were briefly rinsed with wash buffer (2X saline-so- dium citrate (SSC), 10% formamide) and then hybridized overnight at 37 °C with fluorescently labeled probes in hybridization buffer [2X SSC, 10% dextran sulfate (sigma, D8906), 1 mg/mL E. coli transfer RNA (sigma, R8759), 2 mM vanadyl ribonucleoside complex (NEB S142), 0.5% Bovine serum albumin (BSA) (Ambion, AM2618), 10% formamide]. Following hybridization, samples were washed two times in wash buffer for 30 min each at 37 °C and mounted in VECTASHIELD with DAPI (Vector Labs). Fluorescently labeled probes were added to the hybridization buffer to a final concentration of 100 nM. Poly(T) probes for recognizing Poly(A) sequence were from Integrated DNA Technologies. Probes against Mst77F and Mst77Y were designed using the Stellaris1 RNA FISH Probe Designer (Biosearch Technologies, Inc.) available online at www.biosearchtech.com/stellarisdesigner. Each set of custom Stellaris1 RNA FISH probes was labeled with Quasar 670 or Quasar 570 (SI Appendix, Table S4). Images were acquired using an upright Leica TCS SP8 confocal microscope with a 63× oil immersion objective lens (NA = 1.4) and processed using Adobe Photoshop and ImageJ software. DNA Fluorescence In Situ Hybridization. Testes from 1- to 3-d-old flies were rapidly dissected in 4% formaldehyde and 1mM Ethylenediaminetetraacetic acid (EDTA) in 1X PBS and then nutated for 30 min. Then, the testes were washed three times in 1X PBS containing 0.1% Triton-X (PBST) +1 mM EDTA for 30 min each. The testes were briefly rinsed with 1X PBST and then incubated at 37 °C for 10 min with 2 mg/mL Rnase A in PBST. Following Rnase treatment, samples were washed once in 1X PBST + 1 mM EDTA for 10 min. The samples were then briefly rinsed with 2X SSC + 1 mM EDTA + 0.1% Tween-20, and then washed three times in 2X SSC + 0.1% Tween-20 + formamide (20% for first wash, 40% for second, 50% for third) for 15 min each. The samples were then washed with 2X SSC + 0.1% Tween-20 + 50% formamide for 30 min. The samples were then incubated for 5 min at 95 °C with fluorescently labeled probes in hybridization buffer (2X SSC, 10% dextran sulfate, 50% formamide, 1 mM EDTA) and then transferred to 37 °C overnight. Following hybridization, the samples were washed three times in 2X SSC + 1 mM EDTA + 0.1% Tween-20 for 20 min each and then mounted in VECTASHIELD with DAPI (Vector Labs). Fluorescently labeled probes were added to the hybridization buffer to a final concentration of 500 nM. Satellite DNA probes distinguishing X and Y chromosomes (AATAC)6-Cy5 for Y and (TAGA)8-Cy3 were from Integrated DNA Technologies. IF Staining. Testes were dissected in 1X PBS, transferred to 4% formaldehyde in 1X PBS, and fixed for 30 min. The testes were then washed three times in 1X PBST for 20 min each followed by incubation with primary antibodies diluted in 1X PBST with 3% BSA at 4 °C overnight. Samples were washed three times in 1X PBST for 20 min each and incubated with secondary antibody in 1X PBST with 3% BSA for 2 h at room temperature. The samples were then washed three times in 1X PBST for 20 min each and mounted in VECTASHIELD with DAPI (Vector Labs). The following primary antibodies were used: anti-fibrillarin (1:200, mouse; Abcam; ab5812), anti-Modulo (1:1,000, guinea pig; this study), anti-protamine A/B [1:200, guinea pig, gift of Elaine Dunleavy, Centre for Chromosome Biology, National University of Ireland, Galway, Ireland (47), anti-dsDNA (1:500; mouse,; Abcam; ab27156), anti-histone H3 (1:200, rabbit; Abcam; ab1791), anti-Mst77F (1:1,000; guinea pig, this study), anti-Mst77Y (1:500; rabbit, this study), anti-Tpl94D (1:500; rabbit, this study), and phalloidin-Alexa Fluor 546 or 488 (1:200; Thermo Fisher Scientific; A22283 or A12379). The Modulo antibody was generated by injecting a peptide sequence CRKQPVKEVPQFSEED[48-62] targeting the N-terminal end of Modulo in guinea pigs (Covance). The Tpl94D antibody was generated by injecting a peptide DKGSAYKPLTLNRSYVIRKC[96-114] in rabbits (Covance). The Mst77F antibody was generated by injecting multiple peptides (SKPEVAVTC[9-16], YKKSIEYVNC[22-30], CRSSEGEHR[112-119], LQRSSEGEHRMHSEC[110-123], RSSGKPKPKGARPRKC[169-183]) targeting sites in Mst77F, as indicated, differen- tiating it from Mst77Y in guinea pigs (Covance). The Mst77Y antibody was gen- erated by injecting multiple peptides (IKPDVAVSC[9-16], SRKAIEYVKC[22-30], CRSIEAELR[112-119], KTSRKAIEYVKSD[20-32], CVSSLQRSIEAELR[107-119]) target- ing sites of varying aa length in Mst77Y, differentiating it from Mst77F in rabbits (Covance). Alexa Fluor–conjugated secondary antibodies (Life Technologies) were used at a dilution of 1:200. qRT-PCR. Total RNA was purified from D. melanogaster adult testes (75 pairs/sam- ple) by Direct-zol RNA Miniprep (Zymo Research), with DNase treatment according to manufacturer’s protocol. One microgram total RNA was reverse transcribed following priming with either random hexamers or polyT using SuperScript III® Reverse Transcriptase (Invitrogen) followed by qPCR using Power SYBR Green rea- gent (Applied Biosystems). Primers for qPCR were designed to amplify mRNA or intron-containing transcript as indicated. Relative expression levels were normal- ized to Rp49 and control siblings. All reactions were done in technical triplicates with at least two biological replicates. Graphical representation was inclusive of all replicates and P-values were calculated using a t test performed on untransformed average ddct values. Primers used are described in SI Appendix, Fig. S11 A and B. Total RNA-Seq. Total RNA was purified from D. melanogaster adult testes by Direct-zol RNA Miniprep (Zymo Research), with Dnase treatment. Quality of the indexed libraries was confirmed using an Agilent Fragment Analyzer and qPCR. Sequencing libraries were prepared with the KAPA RNA HyperPrep Kit with RiboErase. Samples were sequenced on a HiSeq 2500, producing 100 × 100 nt paired-end reads. The read pairs were mapped to the canonical chromosomes of the D. melanogaster genome (assembly BDGP6/dm6) using STAR 2.7.1a (48); PNAS 2023 Vol. 120 No. 16 e2220576120 https://doi.org/10.1073/pnas.2220576120 7 of 8 default parameters, except “—alignIntronMax 25000,” indexed with all FlyBase genes (FB2020_06 Dmel Release 6.37) and the option “—sjdbOverhang 100.” Gene counts were obtained using featureCounts (49); v 2.0.1, with “-M –fraction -p -s 2.” After summing gene counts for technical replicates, differential expres- sion was assayed using DESeq2 v1.26.0 (50), with lfcShrink(type=”ashr”)). RNA coverage across genes at nucleotide resolution was quantified with “bedtools coverage” (51) and scaled by the total number of reads mapped to genes. All the statistical details of the experiments are provided in the main text and leg- ends. P-values are listed either in figure, figure legends, or SI Appendix, Table S1. Data, Materials, and Software Availability. Sequencing data is available at National Center for Biotechnology Information Gene Expression Omnibus under accession GSE214456 (52). All other data are included in the manuscript and/ or SI Appendix. Statistics and Reproducibility. Data are presented as mean ± SD unless oth- erwise indicated. The sample number (n) indicates the number of testes, nuclei, or male flies in each experiment as specified in the figure legends. We utilized two-sided Student’s t test to compare paired or independent samples, as applica- ble and is specified in the figure legends. We calculated probability using exact binomial distribution with parameters specified in Fig. 3G legend. No statistical methods were used to predetermine sample sizes. The experimenters were not blinded to the experimental conditions, and no randomization was performed. ACKNOWLEDGMENTS. We thank the Bloomington Drosophila Stock Center and Dr. Elaine Dunleavy for reagents. We thank the Data Science, Bioinformatics, and Informatics Core at the University of Michigan for consulting and Dr. Bing Ye for advice and support. We thank the Yamashita, Lehmann, and Ye lab members, Drs. Daven Presgraves and Eric Lai for discussions, and Yamashita Lab members for comments on the manuscript. The research was supported by the Eunice Kennedy Shriver Institute of Child Health and Development of the NIH (to J.I.P., F30HD105324), HHMI (to Y.M.Y.), and Whitehead Institute for Biological Research. 1. 2. 3. 4. 5. 6. 7. 8. 9. C. Rathke et al., Transition from a nucleosome-based to a protamine-based chromatin configuration during spermiogenesis in Drosophila. J. Cell Sci. 120, 1689–1700 (2007). D. T. Carrell, B. R. Emery, S. Hammoud, Altered protamine expression and diminished spermatogenesis: What is the link? Hum. Reprod. Update 13, 313–327 (2007). C. Rathke, W. M. Baarends, S. Awe, R. Renkawitz-Pohl, Chromatin dynamics during spermiogenesis. Biochim. Biophys. Acta (BBA) - Gene Regul. Mechanisms 1839, 155–168 (2014). R. Balhorn, The protamine family of sperm nuclear proteins. Genome Biol. 8, 227 (2007). R. Oliva, Protamines and male infertility. Hum. Reprod. Update 12, 417–435 (2006). C. Cho et al., Protamine 2 deficiency leads to sperm DNA damage and embryo death in Mice1. Biol. Reprod. 69, 211–217 (2003). C. Cho et al., Haploinsufficiency of protamine-1 or -2 causes infertility in mice. Nat. Genet. 28, 82–86 (2001). D. Miller, M. Brinkworth, D. Iles, Paternal DNA packaging in spermatozoa: More than the sum of its parts? DNA, histones, protamines and epigenetics. Reproduction 139, 287–301 (2010). L. Lüke, A. Vicens, M. Tourmente, E. R. S. Roldan, Evolution of protamine genes and changes in sperm head phenotype in Rodents1. Biol. Reprod. 90, 67 (2014). 28. K. Abdelmohsen, M. Gorospe, RNA-binding protein nucleolin in disease. RNA Biol. 9, 799–808 (2012). 29. D. H. Castrillon et al., Toward a molecular genetic analysis of spermatogenesis in Drosophila melanogaster: Characterization of male-sterile mutants generated by single P element mutagenesis. Genetics 135, 489–505 (1993). 30. M. Herbette et al., Distinct spermiogenic phenotypes underlie sperm elimination in the Segregation Distorter meiotic drive system. PLoS Genet. 17, e1009662 (2021). 31. J. D. Lewis et al., Histone H1 and the origin of protamines. Proc. Natl. Acad. Sci. U.S.A. 101, 4148–4152 (2004). 32. B. Barckmann et al., Three levels of regulation lead to protamine and Mst77F expression in Drosophila. Dev. Biol. 377, 33–45 (2013). 33. F. Michiels, D. Buttgereit, R. Renkawitz-Pohl, An 18-bp element in the 5’ untranslated region of the Drosophila beta 2 tubulin mRNA regulates the mRNA level during postmeiotic stages of spermatogenesis. Eur. J. Cell Biol. 62, 66–74 (1993). 34. F. Michiels, A. Gasch, B. Kaltschmidt, R. Renkawitz-Pohl, A 14 bp promoter element directs the testis specificity of the Drosophila beta 2 tubulin gene. EMBO J. 8, 1559–1565 (1989). 10. C.-H. Chang, I. Mejia Natividad, H. S. Malik, Expansion and loss of sperm nuclear basic protein genes in Drosophila correspond with genetic conflicts between sex chromosomes. Elife 12, e85249 (2023). 11. W.-M. Maier, G. Nussbaum, L. Domenjoud, U. Klemm, W. Engel, The lack of protamine 2 (P2) in boar and bull spermatozoa is due to mutations within the P2 gene. Nucleic Acids Res. 18, 1249–1254 (1990). 35. R. E. Braun, J. J. Peschon, R. R. Behringer, R. L. Brinster, R. D. Palmiter, Protamine 3’-untranslated sequences regulate temporal translational control and subcellular localization of growth hormone in spermatids of transgenic mice. Genes Dev. 3, 793–802 (1989). 36. D. Elliott, Pathways of post-transcriptional gene regulation in mammalian germ cell development. 12. S. Tirmarche et al., Drosophila protamine-like Mst35ba and Mst35bb are required for proper Cytogenet. Genome Res. 103, 210–216 (2003). sperm nuclear morphology but are dispensable for male fertility. G3: Genes, Genomes, Genetics 4, 2241–2245 (2014). 37. G. Dreyf, Structure and function of nuclear and cytoplasmic ribonucleoprotein particles. Annu. Rev. Cell Biol. 2, 459–498 (1986). 13. S. Jayaramaiah Raja, R. Renkawitz-Pohl, Replacement by Drosophila melanogaster Protamines and 38. S. Ozturk, F. Uysal, Potential roles of the poly(A)-binding proteins in translational regulation during Mst77F of histones during chromatin condensation in late spermatids and role of sesame in the removal of these proteins from the male pronucleus. Mol. Cell Biol. 25, 6165–6177 (2005). spermatogenesis. J. Reprod. Dev. 64, 289–296 (2018). 39. H. K. Kini, M. R. Vishnu, S. A. Liebhaber, Too much PABP, too little translation. J. Clin. Invest. 120, 14. C. M. Doyen et al., A testis-specific chaperone and the chromatin remodeler ISWI mediate 3090–3093 (2010). repackaging of the paternal genome. Cell Rep. 13, 1310–1318 (2015). 40. P. Chen et al., piRNA-mediated gene regulation and adaptation to sex-specific transposon 15. C. A. Muirhead, D. C. Presgraves, Satellite DNA-mediated diversification of a sex-ratio meiotic drive expression in D. melanogaster male germline. Genes. Dev. 35, 914–935 (2021). gene family in Drosophila. Nat. Ecol. Evol. 5, 1604–1612 (2021). 41. N. Kost et al., Multimerization of Drosophila sperm protein Mst77F causes a unique condensed 16. J. Vedanayagam, C. J. Lin, E. C. Lai, Rapid evolutionary dynamics of an expanding family of meiotic chromatin structure. Nucleic Acids Res. 43, 3033–3045 (2015). drive factors and their hpRNA suppressors. Nat. Ecol. Evol. 5, 1613–1623 (2021). 42. Y. Tao, J. P. Masly, L. Araripe, Y. Ke, D. L. Hartl, A sex-ratio meiotic drive System in Drosophila simulans 17. C. Rathke et al., Distinct functions of Mst77F and protamines in nuclear shaping and chromatin condensation during Drosophila spermiogenesis. Eur. J. Cell Biol. 89, 326–338 (2010). 18. S. Kimura, B. Loppin, The Drosophila chromosomal protein Mst77F is processed to generate an I: An Autosomal Suppressor. PLoS Biol. 5, e292 (2007). 43. M. Jagannathan, Y. M. Yamashita, Function of Junk: Pericentromeric satellite DNA in chromosome maintenance. Cold Spring Harb. Symp. Quant. Biol. 82, 319–327 (2017). essential component of mature sperm chromatin. Open Biol. 6, 160207 (2016). 44. A. M. Larracuente, The organization and evolution of the Responder satellite in species of the 19. Z. Eren-Ghiani, C. Rathke, I. Theofel, R. Renkawitz-Pohl, Prtl99C acts together with protamines and safeguards male fertility in drosophila. Cell Rep. 13, 2327–2335 (2015). 20. F. J. Krsticevic, C. G. Schrago, A. B. Carvalho, Long-read single molecule sequencing to resolve tandem gene copies: The Mst77Y region on the drosophila melanogaster Y chromosome. G3: Genes, Genomes, Genetics 5, 1145–1150 (2015). Drosophila melanogaster group: dynamic evolution of a target of meiotic drive. BMC Evol. Biol. 14, 233 (2014). 45. K. Houtchens, T. W. Lyttle, Responder (Rsp) alleles in the segregation distorter (SD) system of meiotic drive in Drosophila may represent a complex family of satellite repeat sequences. Genetica 117, 291–302 (2003). 21. F. J. Krsticevic, H. L. Santos, S. Januário, C. G. Schrago, A. B. Carvalho, Functional copies of the 46. Y. Hiraizumi, L. Sandler, J. F. Crow, Meiotic drive in natural populations of Drosophila Mst77F gene on the Y chromosome of Drosophila melanogaster. Genetics 184, 295–307 (2010). 22. M. Corzett, J. Mazrimas, R. Balhorn, Protamine 1: Protamine 2 stoichiometry in the sperm of eutherian mammals. Mol. Reprod. Dev. 61, 519–527 (2002). 23. S. Hammoud, L. Liu, D. T. Carrell, Protamine ratio and the level of histone retention in sperm selected from a density gradient preparation. Andrologia 41, 88–94 (2009). 24. K. Steger et al., Both protamine-1 to protamine-2 mRNA ratio and Bcl2 mRNA content in testicular spermatids and ejaculated spermatozoa discriminate between fertile and infertile men. Hum. Reprod. 23, 11–16 (2007). melanogaster. III. Populational implications of the Segregation-Distorter Locus. Evolution (N Y) 14, 433 (1960). 47. W. K. Mills, Y. C. G. Lee, A. M. Kochendoerfer, E. M. Dunleavy, G. H. Karpen, RNA from a simple- tandem repeat is required for sperm maturation and male fertility in Drosophila melanogaster. Elife 8, e48940 (2019). 48. A. Dobin et al., STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013). 49. Y. Liao, G. K. Smyth, W. Shi, featureCounts: An efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014). 25. S. Amjad et al., Protamine 1/Protamine 2 mRNA ratio in nonobstructive azoospermic patients. 50. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion for RNA-seq Andrologia 53, e13936 (2021). data with DESeq2. Genome Biol. 15, 550 (2014). 26. V. W. Aoki, L. Liu, D. T. Carrell, Identification and evaluation of a novel sperm protamine abnormality 51. A. R. Quinlan, I. M. Hall, BEDTools: A flexible suite of utilities for comparing genomic features. in a population of infertile males. Hum. Reprod. 20, 1298–1306 (2005). Bioinformatics 26, 841–842 (2010). 27. L. M. Mikhaylova, A. M. Boutanaev, D. I. Nurminsky, Transcriptional regulation by Modulo integrates 52. J. I. Park, G. W. Bell, Y. M. Yamashita, Total RNA-seq in the modulo mutant reveals broad changes meiosis and spermatid differentiation in male germ line. Proc. Natl. Acad. Sci. U.S.A. 103, 11975–11980 (2006). to the transcriptome. Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?acc=GSE214456. Deposited 29 September 2022. 8 of 8 https://doi.org/10.1073/pnas.2220576120 pnas.org
10.1016_j.jmb.2023.168145	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript J Mol Biol. Author manuscript; available in PMC 2023 July 01. Published in final edited form as: J Mol Biol. 2023 July 01; 435(13): 168145. doi:10.1016/j.jmb.2023.168145. APEX3 – an optimized tool for rapid and unbiased proximity labeling Jordan T. Becker1,2,3,, Ashley A. Auerbach4, Reuben S. Harris1,4,5, 1Department of Biochemistry, Molecular Biology, and Biophysics, University of Minnesota Twin Cities, Minneapolis, MN, USA 55455 2Department of Microbiology and Immunology, University of Minnesota Twin Cities, Minneapolis, MN, USA 55455 3Institute for Molecular Virology, University of Minnesota Twin Cities, Minneapolis, MN, USA 55455 4Department of Biochemistry and Structural Biology, University of Texas Health San Antonio, San Antonio, TX, USA 78229 5Howard Hughes Medical Institute, University of Texas Health San Antonio, San Antonio, TX, USA 78229 Abstract Macromolecular interactions regulate all aspects of biology. The identification of interacting partners and complexes is important for understanding cellular processes, host-pathogen conflicts, and organismal development. Multiple methods exist to label and enrich interacting proteins in living cells. Notably, the soybean ascorbate peroxidase, APEX2, rapidly biotinylates adjacent biomolecules in the presence of biotin-phenol and hydrogen peroxide. However, during initial experiments with this system, we found that APEX2 exhibits a cytoplasmic-biased localization and is sensitive to the nuclear export inhibitor leptomycin B (LMB). This led us to identify a putative nuclear export signal (NES) at the carboxy-terminus of APEX2 (NESAPEX2), structurally This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Address correspondence to: [email protected] or [email protected]. AUTHOR CONTRIBUTIONS JTB and RSH conceptualized the study. JTB performed all experiments, generated plasmid constructs, curated the data, performed formal data analysis, generated figures, and drafted the manuscript. AAA repeated the observation of APEX2 cytoplasmic-biased localization in preliminary experiments and generated plasmid constructs. RSH provided funding and resources. RSH and AAA contributed to review and editing of the manuscript. The authors have no competing interests. Jordan T. Becker: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review and editing. Ashley A. Auerbach: Investigation, Writing – review and editing. Reuben S. Harris: Conceptualization, Funding acquisition, Writing – review and editing. Co-corresponding authors Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Declaration of interests The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 2 adjacent to the conserved heme binding site. This putative NES is functional as evidenced by cytoplasmic localization and LMB sensitivity of a mCherry-NESAPEX2 chimeric construct. Single amino acid substitutions of multiple hydrophobic residues within NESAPEX2 eliminate cytoplasm- biased localization of both mCherry-NESAPEX2 as well as full-length APEX2. However, all but one of these NES substitutions also compromises peroxide-dependent labeling. This unique separation-of-function mutant, APEX2-L242A, is termed APEX3. Localization and functionality of APEX3 are confirmed by fusion to the nucleocytoplasmic shuttling transcriptional factor, RELA. APEX3 is therefore an optimized tool for unbiased proximity labeling of cellular proteins and interacting factors. Graphical Abstract Keywords APEX2/3; nuclear export signal (NES); nucleocytoplasmic shuttling; proximity labeling technology; subcellular localization INTRODUCTION The subcellular localization of biological macromolecules is a key determinant of function [1–6]. Notably, RNA-binding proteins (RBPs) and their preferred RNA substrates must occupy or transit to the same subcellular location in order to interact [7–10]. In homeostatic circumstances, colocalization is important for regulating key biological processes including RNA trafficking in neurons, mRNA translation, splicing, RNA export, and non-sense mediated decay [11–15]. In the context of host-pathogen conflicts, compartmentalized interactions are required for multiple antiviral mechanisms such as APOBEC3 packaging into retroviral virions [16,17] and ZAP binding to CpG-rich RNAs [18–20]. Finally, these interactions determine successful or abortive virus replication (e.g., viral structural proteins binding genomic RNA or RNA polymerase with template viral RNA [21,22]). Furthermore, purposeful disruption of biological localization is a pharmaceutical target (e.g., the XPO1 inhibitor Selinexor disrupts nuclear export) [23–27]. Understanding the diverse interactions of a particular protein of interest (POI) is important and, accordingly, multiple methods have been developed to interrogate these interactions. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 3 Protein-protein interactions (PPIs) and RNA-protein interactions (RPIs) are commonly investigated by immunoprecipitation (IP) of a known target using a specific antibody coupled to downstream mass spectrometry or RNA sequencing, respectively [28]. These methods are effective but may sometimes miss relevant interactions. For example, a particular protein may only interact with some partners transiently, weakly, or indirectly (i.e., through interaction with other molecules). Recent advances for in cellulo proximity labeling allow for tunable labeling and enrichment of interacting molecular complexes as well as capturing of transient interactions. Notably, the engineered soybean ascorbate peroxidase (APEX2) catalyzes the conjugation of biotin-phenol to adjacent proteins and RNA in the presence of hydrogen peroxide [29–31]. This reaction occurs rapidly (seconds to minutes) and allows for flexible experimental modification. Indeed, APEX2 has been used to identify the spatiotemporal PPI network of GPCR signaling [32], the spatiotemporal PPI and RPI networks of RNA granule formation [33], and a transcriptome-wide atlas of subcellular RNA localization [34]. Notably, these studies used proteins with strong subcellular localization determinants fused to APEX2. We initially intended to use APEX2 to identify the PPI and RPI networks and shuttling mechanisms of a family of related cellular proteins with disparate nuclear and cytoplasmic localizations. However, during experiments designed to validate the localization of APEX2 fusion proteins we found that APEX2 itself exhibits a cytoplasm-biased localization rather than a diffuse, whole-cell localization that is characteristic of most fluorescent and non-compartmentalized proteins. This observation led to the identification of a putative nuclear export signal (NES) in APEX2 that is conserved in the parental soybean ascorbate peroxidase 1 (APX1) protein as well other plant homologs. The functionality of this putative NES is demonstrated by cytoplasmic localization of a heterologous fusion to mCherry and responsiveness to the XPO1 inhibitor leptomycin B (LMB). In addition, mutations to hydrophobic residues in this NES motif compromise the cytoplasm-biased localization of APEX2, however most of these mutants also lack peroxidase activity. We identified one APEX2 mutant (L242A) that exhibits whole-cell localization and retains peroxidase activity. This separation-of-function mutant, termed APEX3, is further validated by fusion to a well-characterized nucleocytoplasmic shuttling protein, the RELA component of NF-κB. Overall, we describe an optimized proximity labeling technology, APEX3, that expands the utility of this technology and may help answer a broader number of important questions. RESULTS APEX2 exhibits a cytoplasm-biased localization We were initially interested in using APEX2 to identify interacting factors for nucleocytoplasmic shuttling proteins. In the early stages of these experiments, we generated control constructs for labeling different subcellular compartments (i.e., nucleus, cytoplasm, and whole cell; construct schematics in Figure 1A): APEX2 alone, APEX2 fused to mNeonGreen (mNG) at either its amino- or carboxy-terminus (mNG-APEX2 or APEX2- mNG), a strong NLS [35] fused to mNG-APEX2 (NLS-mNG-APEX2), and a strong NES [36–38] fused to mNG-APEX2 (NES-mNG-APEX2). A V5 epitope tag was also added to the carboxy-terminus of APEX2 in all these constructs for detection. HeLa cell lines J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 4 were generated by transduction to stably express each of these constructs for imaging by fluorescence microscopy. To our surprise, we noticed that APEX2 alone as well as mNG- APEX2 and APEX2-mNG exhibit greater cytoplasmic than nuclear fluorescence intensity (Figure 1B). This localization bias was evident in fixed cells using direct mNG fluorescent signal (green) as well as by indirect V5 epitope staining and immunofluorescent microscopy (rose). As expected, the NES-mNG-APEX2 control construct with a heterologous NES exhibits increased nuclear localization in HeLa cells following leptomycin B (LMB) treatment to inhibit XPO1-dependent nuclear export [39]. To our surprise, the mNG-APEX2 and APEX2-mNG constructs (without a heterologous NES) also exhibit increased nuclear localization following LMB treatment (Figure 1C, quantified in 1D). As additional controls, mNG alone and NLS-mNG-APEX2 are unaffected by LMB. To demonstrate that this apparent LMB-sensitive (XPO1-dependent) nuclear export activity is intrinsic to APEX2 and not to other features of the mNG-APEX2-V5 fusion constructs, we created a mNG- APEX2 “Stop” construct lacking the V5 epitope and found that it still exhibits cytoplasmic localization and sensitivity to LMB treatment (Figure 1C, quantified in 1D). These constructs also display similar subcellular localization properties in 293T cells (Figure S1). Thus, these results led us to hypothesize that APEX2 has a signal or motif responsible for the observed cytoplasm-biased localization. APEX2 encodes an amino acid motif that acts as a heterologous NES Many XPO1-dependent nucleocytoplasmic cargo proteins have an amino acid motif with 3–4 hydrophobic residues [40–42], which is often flanked by acidic residues [43] (e.g., DxLxxxLxxLxLxD; alignment in Figure 2A). Submission of the APEX2 amino acid sequence (including mNG, peptide linkers, and V5 epitope) to multiple NES prediction servers [44–46] did not identify such a motif. However, manual inspection of the primary APEX2 amino acid sequence revealed an NES-like region at the carboxy-terminus of the protein that is shared with the parental soybean (Glycine hispida) ascorbate peroxidase, ghAPX1 (Figure 2A). Alignment of hundreds of available plant APX1 homologs indicated that this putative NES is highly conserved with the central leucine residues demonstrating 100% conservation (Figure S2). This putative NES is also found in split APEX2 [47] as well as the original APEX [48] derived from Pisum sativum APX. To test whether this candidate NES is responsible for APEX2 cytoplasmic localization, this motif (NESAPEX2) was fused to the carboxy-terminus of mCherry and examined in HeLa cells by fluorescent microscopy. In comparison to mCherry alone, which has a cell-wide distribution, mCherry-NESAPEX2 exhibits a distinctly cytoplasmic localization (Figure 2B– C; quantification in Figure 2D). Moreover, LMB treatment of cells causes an accumulation of this construct in the nuclear compartment and yields an overall cell wide appearance. As a positive control, an analogous mCherry fusion construct with the well-characterized NES of HIV-1 Rev behaves similarly (Figure 2B–C; quantified in 2D). In contrast, an mCherry construct fused with the NLS of SV40 shows predominantly nuclear localization and is unaffected by LMB (Figure 2B–C; quantified in 2D). J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 5 We next generated a panel of single amino acid mCherry-NESAPEX2 mutant constructs to determine the residues responsible for nuclear export. NES-containing cargos are exported from the nucleus by XPO1, and this interaction is typically mediated by hydrophobic residues within the NES and can be further influenced by flanking acidic residues [40,43]. Therefore, these conserved residues were changed to alanine, glycine, or serine and studied by fluorescence microscopy in HeLa cells. Mutation of the acidic residues of the NES motif resulted in proteins that maintain export function as evidenced by predominantly cytoplasmic localization (E1A or D13A in Figure 2E). In contrast, changes to hydrophobic residues L6, L9, and F11 compromised export activity, evidenced by whole-cell mCherry localization (L6A/G/S, L9A/G/S, and F11A/G/S in Figure 2E). Taken together, these results demonstrate that the putative NES of APEX2 is both portable and functional and therefore a robust export motif. Most APEX2 NES mutants also lack peroxidase activity Upon inspection of the structure of soybean ascorbate peroxidase (PDB: 1OAG) [49], we noticed that the putative NES is part of an alpha-helix typical of many NES cargos (Figure 3A). This putative NES may have eluded prediction algorithms (e.g., LocNES and NESsential) because it may have been deemed structurally inaccessible and unlikely to achieve the disorder/flexibility required to engage export machinery [44,46]. Indeed, the critical hydrophobic residues required for export are positioned toward the globular center of the protein and therefore are likely to contribute to structural stability and function of the overall enzyme as a peroxidase (Figure 3A). To address this possibility and ask whether the putative APEX2 NES is active in the context of its native protein structure, single amino acid substitution mutations were introduced into the NES motif of the full-length mNG-APEX2- V5 construct (Figure 3B). Most of these mutations were selected to reduce hydrophobic interactions with XPO1 [40] and to overlap with the set described above. These constructs were expressed stably in HeLa cells and compared using fluorescence microscopy. As expected, most substitutions that eliminate NES activity in the heterologous mCherry- NESAPEX2 construct described above also cause analogous reductions in the NES activity of full-length APEX2 (Figure 3C, quantified in 3D). Specifically, L242A/G/S/N or L245A/G/S single amino acid mutants show cell-wide localization indicative of compromised NES- like function. More conservative changes at position 242 (L242V and L242I) have no effect on NES function and, as above, the acidic residues at positions 237 and 249 are dispensable for function (E237A and D249A). An exception appears to be F247S which maintains NES function in the full-length protein despite losing it above in the heterologous mCherry-NESAPEX2 context. Nevertheless, taken together, these results further confirm the functionality of the putative APEX2 NES. However, when the same cell lines were subjected to hydrogen peroxide-dependent labeling with biotin-phenol, we found that only the L242A mutant of mNG-APEX2-V5 retains labeling activity as detected by streptavidin immunofluorescence (Figure 3E and Figure S3, quantified in S3C). All other mutants showed one of three different phenotypes. The first group retained wild type-like peroxide-dependent labeling activity, as well as NES-like function as evidenced by cytoplasmic localization (e.g., E237A, L242I/V, D249A). The J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 6 second group lost both peroxide-dependent labeling activity as well as NES function (e.g., L242G/N/S and L245A/G/S). The third group lost peroxide-dependent labeling activity but retained NES-like function (e.g., F247S as well as F231A, F232A, and Y235A that are located amino-terminal relative to the putative NES motif). We speculate that this null phenotype might be due to weakened heme binding and/or a compromised core structure (despite maintaining near wildtype expression levels; Figure 3E and Figure S3). Ultimately, only one single amino acid substitution mutant, L242A, exhibited whole cell localization while retaining peroxidase-dependent labeling activity and, thus, this protein is named APEX3. APEX3 localizes faithfully upon fusion to nucleocytoplasmic shuttling factor RELA Finally, the functionality of APEX3 was tested by fusing it to RELA, which is a well- characterized nucleocytoplasmic shuttling protein (reviewed by [50,51]). RELA is a key component of the NF-κB transcription factor complex that exhibits cytoplasmic localization at steady state when bound by its inhibitor IκBα [52–54]. However, upon receiving a relevant stimulus (e.g., LPS, IL-1, or TNF), IκBα releases from RELA which unveils a NLS that allows translocation to the nucleus, DNA binding, and interaction with transcription elongation factors, CCNT1 and CDK9 (schematic in Figure 4A). Therefore, human RELA was fused to APEX2 or APEX3 with a V5 epitope to observe localization and labeling at steady state and following stimulation with TNF for multiple time points in addition to RELA-mNG as a control (Figure 4B; with quantification in Figure 4C). As expected, RELA-mNG, RELA-APEX, and RELA-APEX3 exhibit cytoplasmic localization at steady state. Moreover, all 3 constructs fully relocalize to the nucleus upon stimulation with TNF after 60 minutes (min). However, by performing a time-course of TNF treatment, we observed that RELA-mNG and RELA-APEX3 exhibit faster nuclear translocation relative to RELA-APEX2 (Figure 4B; quantified in Figure 4C; p < 0.001 by two-way ANOVA). Together these results support the use of APEX3 as an optimized tool for rapid in cellulo proximity labeling, particularly for nucleocytoplasmic shuttling proteins or proteins with uncharacterized localization and regulatory mechanisms. DISCUSSION Proximity labeling is now a widely used experimental technique for identifying molecular interactions across multiple time scales, subcellular locales, and organismal systems [28]. Selecting the appropriate enzymatic labeling tool can dramatically affect the nature and specificity of interactions identified. In trial experiments with the engineered ascorbate peroxidase, APEX2, we observed a cytoplasmic-biased localization that was sensitive to the XPO1 inhibitor, LMB (Figure 1). We identified an amino acid motif at the carboxy-terminus of APEX2 that can act heterologously as an NES when fused to mCherry (Figure 2). Mutations at hydrophobic residues of the NES-like motif in the mCherry fusion as well as in the context of full-length APEX2 eliminated NES-like activity, however only L242A retained the peroxide-dependent labeling activity (Figure 3). We called the L242A mutant, APEX3 – a localization-optimized ascorbate peroxidase. Finally, we confirmed that APEX3 faithfully localizes as a fusion to RELA during NF-κB signaling following TNF stimulation J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 7 (Figure 4). APEX3 is therefore anticipated to enable a wider range of proximity labeling applications. Previous work used APEX2 largely in the context of fusions to very strong subcellular- targeting amino acid motifs or to proteins with strong subcellular localizations. Notably, APEX2 fused to potent subcellular localization signals was used to determine the subcellular distribution of cellular RNA molecules [34]. Others used APEX2 fused to eIF4A1 to identify RNA and protein interactions during translation initiation and within stress granules [33]. As the primary helicase acting during translation initiation, eIF4A1 has a well characterized and strong cytoplasmic localization. In this study, all the APEX2 fusion constructs also included strong subcellular localization signals in addition to proteins of interest: NES derived from HIV-1 Rev on eIF4A1, eIF4E, and a GFP control, SV40 NLS on CBX1, and an endoplasmic reticulum signal from CYP2C1. Another group used G-protein coupled receptors (GPCRs) fused to APEX2 to identify protein-protein interactions that occur during signaling cascades with high spatiotemporal resolution [32]. As GPCRs are membrane-associated proteins, the NESAPEX2 was unlikely to have biased their results. Others used APEX2 fused to two RNA-binding proteins (MS2 coat protein or dCas13a) engineered to include strong nuclear localization signals to tether APEX2 to RNAs of interest (bound by MS2 or dCas13a) and label endogenous proteins bound to those RNAs of interest [55]. Finally, two groups used APEX2 fused to a catalytically inactive Cas9 to label proteins associated with specific genomic loci [56,57]. Altogether, the interactomes identified in these reports using APEX2 are unlikely to have been markedly affected by the cytoplasmic-biased localization of APEX2 due to the strong localization determinants that effectively “over-ride” the putative NES of APEX2. The subcellular localization of RNA and proteins determines their functions, interactions, and regulation. Methods that can identify the complex networks of interacting RNA and proteins within cells are crucial in understanding development, disease, and degeneration as well as finding therapeutic strategies to extend and improve health-spans and lifespans. Here, we have identified and systematically confirmed a putative NES within the APEX2 proximity labeling peroxidase that may complicate its use for identifying interaction networks for several different types of proteins (e.g., nuclear proteins and nucleocytoplasmic shuttling proteins). While many plant APX1 proteins are named “ascorbate peroxidase, cytosolic” as determined by cell fractionation followed by activity assay or western blot [58–62], our studies here are the first to identify and characterize this conserved putative NES, which we speculate is perhaps necessary for preventing promiscuous activity in the plant cell nucleus. APEX2 has been used over other in cellulo biotin labeling methods such as BirA/BioID [63] or TurboID [64] due to its rapid labeling activity. However, we found that the NES-like activity within APEX2 noticeably slows nuclear translocation on a time-scale of minutes, compared to APEX3. Importantly, we recommend that researchers validate the subcellular localization of proteins of interest with minimal tags relative to their APEX2 fusion proteins, as is common for fluorescent fusion proteins. We encourage those using APEX, APEX2, or split-APEX2 to test APEX3 in their systems to reduce potential unforeseen complications due to active nuclear export. In addition, APEX2 and APEX3 could be used in parallel to generate comparative interactomes of a particular protein of interest. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 8 It is remarkable how mutationally sensitive the APEX2 enzyme is regarding amino acid changes in the conserved NES-containing alpha-helix region. In addition, it is curious that the hydrophobic NESAPEX2 residues that may interact with the nuclear export factor XPO1 are structurally occluded, which suggests dynamic flexibility in the structure of APEX2 throughout the cell. This carboxy-terminal alpha-helix of APEX2 could hinge away from the globular domain, which would allow binding to XPO1 and nuclear export. We would further predict that APEX2 would become transiently inactive during export but regain ascorbate peroxidase activity upon NES helix reorganization in the cytosol. Furthermore, it is possible that (1) post-translational modifications or (2) heme occupancy may modulate the accessibility of this region, or (3) a co-factor may facilitate an indirect interaction between APEX2 and XPO1. However, such post-translational modification machinery and substrate preferences or unknown co-factor for indirect interactions would need to be conserved in plants as well as in humans that lack an APX1 homolog. Finally, we note that while we have not shown a direct interaction between APEX2 and XPO1 here, our data strongly support this region of APEX2 exhibiting NES-like activity and, importantly, led to the rational discovery of APEX3. MATERIALS AND METHODS Plasmids APEX2 cDNA was ordered as a gBlock from IDT based on sequence available from GFP-APEX2-NIK3x [34] (Addgene #129274) with synonymous codon optimization to remove restriction enzyme recognition sites and cloned in-frame to the V5 epitope (GKPIPNPLLGLDST) [65]. All APEX2-encoding (and mNeonGreen control) plasmid DNA constructs were cloned using conventional restriction enzymes and T4 DNA ligation (New England Biosciences, #M0202L) cloning into a bespoke MIGR1-derived simple retroviral vector [66] encoding blasticidin resistance gene downstream of an IRES. mNeonGreen was ordered as a codon optimized gBlock from IDT based on the published amino acid sequence of mNeonGreen [67]. Codon-optimized cDNA encoding human RELA (NCBI GenBank accession: NP_068810.3) was ordered as a gBlock from IDT. mCherry expression constructs were cloned by PCR amplification with primers encoding amino acid motifs at the carboxy-terminus of mCherry. All amino acid substitutions were generated by site-directed mutagenesis using Phusion DNA Polymerase. Amino acid sequences of mNG-APEX3-V5 and mCherry-NESAPEX2 are provided in Figure S4. Cell culture, transfections, and transductions HeLa and 293T cells were obtained from ATCC and cultured in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin at 37°C/5%CO2/50%H2O. All transfections were performed using TransIT-LT1 (Mirus Bio, #MIR-2306) with OptiMEM serum-free media at the following ratio: 100 μL OptiMEM, 3 μg LT1, and 1 μg DNA. To generate retrovirus for transducing APEX2 expressing vectors, pre-adhered 293T cells in 6-well plates were transfected with 1 μg APEX2 or control package plasmid, 1 μg pMD.Gag/GagPol [68] plasmid, and 200 ng VSV-G [69] plasmid. Media was replaced at 24 hrs post-transfection. Virus-containing supernatant was harvested at 48 hrs post-transfection, 0.45 μm syringe-filtered, and stored at −20°C. Stable cells were generated as described J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 9 [70]. Briefly, approximately 2,500 target cells (HeLa or 293T) were seeded into a 96-well flat bottom plate, allowed to adhere overnight at 37°C/5%CO2/50%H2O, and 50–200 μL of transducing viral supernatant with 10 μg/mL polybrene added to each well. Transduced cells were selected at 48 hrs post-transduction with 2 μg/mL Blasticidin S (GoldBio, #B-800-100), expanded, and maintained in culture in the presence of drug. Leptomycin B (LMB; Sigma; #L2913, dissolved in methanol) was used at a final concentration of 12.5 nM for two hrs. RELA-APEX/mNG expressing cells were stimulated with TNF (R&D Systems, #210-TA-020; 30 ng/mL) for up to 60 minutes. Peroxide-dependent proximity labeling For APEX2 labeling experiments, cells were plated in culture media for 24 hrs, then replaced with culture media containing 500 μM biotin-phenol (Iris Biotech GMBH; #LS3500) and incubated for at least 60 minutes. For peroxide-dependent labeling, H2O2 (Sigma; #H1009) was added to culture media at 100 mM concentration for 60 seconds while gently shaking/swirling. Peroxidase labeling reaction was stopped by removing media and replacing with “quenching solution” containing 10 mM sodium ascorbate (Sigma; #A7631), 10 mM sodium azide (Sigma; #S2002), and 5 mM Trolox (Sigma; #238813) in PBS. Cells were washed twice more with quenching solution. For fluorescence-based detection of labeling, cells were washed with PBS prior to fixation with 4% paraformaldehyde for 20 minutes and washed again with PBS. Immunostaining and fluorescence microscopy Fixed cells were washed with PBS three times prior to permeabilization in PBS containing 0.25% Triton X-100 for 15 minutes. After washing three times with PBS, cells with blocked with PBS containing 3% bovine serum albumin (blocking buffer) for one hour. Cells were incubated with primary antibodies detecting V5 epitope (mouse anti-V5; Invitrogen, #R960-25) in blocking buffer for one hour at room temperature or overnight at 4°C at a 1:1000 dilution. Cells were washed thrice with PBS and incubated with secondary antibodies in blocking buffer for one hour at room temperatures at the following dilutions: anti-mouse AlexaFluor-568 and streptavidin AlexaFluor-647 each at 1:1000 (Invitrogen; #A11032 and #S32357). Cells were washed once with PBS, incubated with PBS containing DAPI (Sigma; #D9542; 0.1μg/mL) for 15 mins, then washed twice with PBS. We note that imaging fixed and permeabilized cells expressing mNeonGreen exhibited a different distribution than live cells, yet other cells did not exhibit remarkable differences in fixed versus live cells. Cells were imaged using a Nikon Ti-2E widefield microscope using a 20X objective (NA 0.75). Live cell imaging of mNeonGreen-tagged constructs was performed similarly on a Nikon Ti-2E widefield microscope. LMB treatment was performed on live cells for 2 hrs at 12.5 nM alongside vehicle treated cells and/or cells imaged prior to treatment. Images were processed and analyzed using FIJI [71]. Cytoplasmic-to-nuclear ratio was quantified using integrated fluorescence intensity from each compartment for at least 50 cells per condition. Whole cell streptavidin integrated fluorescence intensity was quantified for at least 50 cells per condition and normalized to the maximum value quantified for wild-type APEX2. All constructs were tested in biological triplicate. Representative images are displayed with mNeonGreen in green, V5 staining in rose, streptavidin staining in blue, and mCherry in magenta. All scale bars represent 10 μm. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Bioinformatic analysis Page 10 Sequence logo of NES peptides from UniRef50 for UniProt P48534 containing 266 trimmed UniProt entries with ≥50% identity to Pisum sativum APX1 amino acid sequence (GenBank accession: AAA33645). Amino acid sequences were aligned using ClustalOmega [72] in SeaView5 [73]. Aligned sequences corresponding to residues 237 to 249 from Glycine hispida APX1 (GenBank accession: AAD20022) were used to generate the sequence logo using WebLogo [74] (version 2.8.2). Statistical analyses GraphPad Prism 9.0 was used for statistical analysis (one-way or two-way ANOVA with corrections for multiple comparisons) of quantitative data presented as mean +/− 95% confidence interval for cytoplasmic-to-nuclear fluorescence intensity ratio or streptavidin intensity. Supplementary Material Refer to Web version on PubMed Central for supplementary material. ACKNOWLEDGEMENTS We wish to acknowledge members of the Harris lab for helpful conversations. We appreciate feedback and critical reading of the manuscript from Arad Moghadasi, Sofia Moraes, and Ryan Langlois. This work was supported by NIAID R37-AI064046, NCI P01-CA234228, and a Recruitment of Established Investigators Award from the Cancer Prevention and Research Institute of Texas (CPRIT RR220053). Salary support for JTB was provided by NIAID F32-AI147813. RSH is an Investigator of the Howard Hughes Medical Institute and the Ewing Halsell President’s Council Distinguished Chair. ABBREVIATIONS: APEX2/3 ascorbate peroxidases IP LMB mNG NES NLS POI PPI RBP RPI immunoprecipitation leptomycin B mNeonGreen nuclear export signal nuclear localization signal protein of interest protein-protein interaction RNA binding protein RNA-protein interaction XPO1 Exportin-1 J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. REFERENCES Page 11 [1]. Khong A, Parker R, The landscape of eukaryotic mRNPs, RNA. 26 (2020) 229–239. 10.1261/ rna.073601.119. [PubMed: 31879280] [2]. Ivanov P, Kedersha N, Anderson P, Stress Granules and Processing Bodies in Translational Control, Cold Spring Harb Perspect Biol. 11 (2019) a032813. 10.1101/cshperspect.a032813. [PubMed: 30082464] [3]. Castello A, Fischer B, Eichelbaum K, Horos R, Beckmann BM, Strein C, Davey NE, Humphreys DT, Preiss T, Steinmetz LM, Krijgsveld J, Hentze MW, Insights into RNA Biology from an Atlas of Mammalian mRNA-Binding Proteins, Cell. 149 (2012) 1393–1406. 10.1016/ j.cell.2012.04.031. [PubMed: 22658674] [4]. Hentze MW, Castello A, Schwarzl T, Preiss T, A brave new world of RNA-binding proteins, Nat Rev Mol Cell Biol. 19 (2018) 327–341. 10.1038/nrm.2017.130. [PubMed: 29339797] [5]. Keskin O, Tuncbag N, Gursoy A, Predicting Protein-Protein Interactions from the Molecular to the Proteome Level, Chem Rev. 116 (2016) 4884–4909. 10.1021/acs.chemrev.5b00683. [PubMed: 27074302] [6]. Lorenzi L, Chiu H-S, Avila Cobos F, Gross S, Volders P-J, Cannoodt R, Nuytens J, Vanderheyden K, Anckaert J, Lefever S, Tay AP, de Bony EJ, Trypsteen W, Gysens F, Vromman M, Goovaerts T, Hansen TB, Kuersten S, Nijs N, Taghon T, Vermaelen K, Bracke KR, Saeys Y, De Meyer T, Deshpande NP, Anande G, Chen T-W, Wilkins MR, Unnikrishnan A, De Preter K, Kjems J, Koster J, Schroth GP, Vandesompele J, Sumazin P, Mestdagh P, The RNA Atlas expands the catalog of human non-coding RNAs, Nat Biotechnol. 39 (2021) 1453–1465. 10.1038/s41587-021-00936-1. [PubMed: 34140680] [7]. Van Treeck B, Parker R, Principles of Stress Granules Revealed by Imaging Approaches, Cold Spring Harb Perspect Biol. 11 (2019). 10.1101/cshperspect.a033068. [8]. Das S, Vera M, Gandin V, Singer RH, Tutucci E, Intracellular mRNA transport and localized translation, Nat Rev Mol Cell Biol. 22 (2021) 483–504. 10.1038/s41580-021-00356-8. [PubMed: 33837370] [9]. Sato H, Das S, Singer RH, Vera M, Imaging of DNA and RNA in Living Eukaryotic Cells to Reveal Spatiotemporal Dynamics of Gene Expression, Annu Rev Biochem. 89 (2020) 159–187. 10.1146/annurev-biochem-011520-104955. [PubMed: 32176523] [10]. Calabretta S, Richard S, Emerging Roles of Disordered Sequences in RNA-Binding Proteins, Trends Biochem. Sci 40 (2015) 662–672. 10.1016/j.tibs.2015.08.012. [PubMed: 26481498] [11]. Buxbaum AR, Wu B, Singer RH, Single β-actin mRNA detection in neurons reveals a mechanism for regulating its translatability, Science. 343 (2014) 419–422. 10.1126/ science.1242939. [PubMed: 24458642] [12]. Chao JA, Yoon YJ, Singer RH, Imaging Translation in Single Cells Using Fluorescent Microscopy, Cold Spring Harb Perspect Biol. 4 (2012) a012310. 10.1101/cshperspect.a012310. [PubMed: 22960595] [13]. Cullen BR, Nuclear mRNA export: insights from virology, Trends Biochem. Sci 28 (2003) 419– 424. 10.1016/S0968-0004(03)00142-7. [PubMed: 12932730] [14]. Hammarskjöld ML, Constitutive transport element-mediated nuclear export, Curr. Top. Microbiol. Immunol 259 (2001) 77–93. [PubMed: 11417128] [15]. Isken O, Maquat LE, The multiple lives of NMD factors: balancing roles in gene and genome regulation, Nat Rev Genet. 9 (2008) 699–712. 10.1038/nrg2402. [PubMed: 18679436] [16]. York A, Kutluay SB, Errando M, Bieniasz PD, The RNA Binding Specificity of Human APOBEC3 Proteins Resembles That of HIV-1 Nucleocapsid, PLoS Pathog. 12 (2016) e1005833. 10.1371/journal.ppat.1005833. [PubMed: 27541140] [17]. Apolonia L, Schulz R, Curk T, Rocha P, Swanson CM, Schaller T, Ule J, Malim MH, Promiscuous RNA Binding Ensures Effective Encapsidation of APOBEC3 Proteins by HIV-1, PLoS Pathog. 11 (2015) e1004609. 10.1371/journal.ppat.1004609. [PubMed: 25590131] [18]. Takata MA, Gonçalves-Carneiro D, Zang TM, Soll SJ, York A, Blanco-Melo D, Bieniasz PD, CG dinucleotide suppression enables antiviral defence targeting non-self RNA, Nature. (2017). 10.1038/nature24039. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 12 [19]. Meagher JL, Takata M, Gonçalves-Carneiro D, Keane SC, Rebendenne A, Ong H, Orr VK, MacDonald MR, Stuckey JA, Bieniasz PD, Smith JL, Structure of the zinc-finger antiviral protein in complex with RNA reveals a mechanism for selective targeting of CG-rich viral sequences, Proc. Natl. Acad. Sci. U.S.A 116 (2019) 24303–24309. 10.1073/pnas.1913232116. [PubMed: 31719195] [20]. Luo X, Wang X, Gao Y, Zhu J, Liu S, Gao G, Gao P, Molecular Mechanism of RNA Recognition by Zinc-Finger Antiviral Protein, Cell Rep. 30 (2020) 46–52.e4. 10.1016/j.celrep.2019.11.116. [PubMed: 31914396] [21]. Korboukh VK, Lee CA, Acevedo A, Vignuzzi M, Xiao Y, Arnold JJ, Hemperly S, Graci JD, August A, Andino R, Cameron CE, RNA virus population diversity, an optimum for maximal fitness and virulence, J. Biol. Chem 289 (2014) 29531–29544. 10.1074/jbc.M114.592303. [PubMed: 25213864] [22]. Te Velthuis AJW, Grimes JM, Fodor E, Structural insights into RNA polymerases of negative- sense RNA viruses, Nat Rev Microbiol. 19 (2021) 303–318. 10.1038/s41579-020-00501-8. [PubMed: 33495561] [23]. Xu D, Farmer A, Chook YM, Recognition of nuclear targeting signals by Karyopherin-β proteins, Curr Opin Struct Biol. 20 (2010) 782–790. 10.1016/j.sbi.2010.09.008. [PubMed: 20951026] [24]. Fung HYJ, Chook YM, Atomic basis of CRM1-cargo recognition, release and inhibition, Semin Cancer Biol. 27 (2014) 52–61. 10.1016/j.semcancer.2014.03.002. [PubMed: 24631835] [25]. Wagstaff KM, Rawlinson SM, Hearps AC, Jans DA, An AlphaScreen®-based assay for high- throughput screening for specific inhibitors of nuclear import, J Biomol Screen. 16 (2011) 192– 200. 10.1177/1087057110390360. [PubMed: 21297106] [26]. Wagstaff KM, Sivakumaran H, Heaton SM, Harrich D, Jans DA, Ivermectin is a specific inhibitor of importin α/β-mediated nuclear import able to inhibit replication of HIV-1 and dengue virus, Biochem J. 443 (2012) 851–856. 10.1042/BJ20120150. [PubMed: 22417684] [27]. Grosicki S, Simonova M, Spicka I, Pour L, Kriachok I, Gavriatopoulou M, Pylypenko H, Auner HW, Leleu X, Doronin V, Usenko G, Bahlis NJ, Hajek R, Benjamin R, Dolai TK, Sinha DK, Venner CP, Garg M, Gironella M, Jurczyszyn A, Robak P, Galli M, Wallington-Beddoe C, Radinoff A, Salogub G, Stevens DA, Basu S, Liberati AM, Quach H, Goranova-Marinova VS, Bila J, Katodritou E, Oliynyk H, Korenkova S, Kumar J, Jagannath S, Moreau P, Levy M, White D, Gatt ME, Facon T, Mateos MV, Cavo M, Reece D, Anderson LD, Saint-Martin J-R, Jeha J, Joshi AA, Chai Y, Li L, Peddagali V, Arazy M, Shah J, Shacham S, Kauffman MG, Dimopoulos MA, Richardson PG, Delimpasi S, Once-per-week selinexor, bortezomib, and dexamethasone versus twice-per-week bortezomib and dexamethasone in patients with multiple myeloma (BOSTON): a randomised, open-label, phase 3 trial, The Lancet. 396 (2020) 1563– 1573. 10.1016/S0140-6736(20)32292-3. [28]. Qin W, Cho KF, Cavanagh PE, Ting AY, Deciphering molecular interactions by proximity labeling, Nat Methods. 18 (2021) 133–143. 10.1038/s41592-020-01010-5. [PubMed: 33432242] [29]. Hung V, Zou P, Rhee H-W, Udeshi ND, Cracan V, Svinkina T, Carr SA, Mootha VK, Ting AY, Proteomic mapping of the human mitochondrial intermembrane space in live cells via ratiometric APEX tagging, Mol Cell. 55 (2014) 332–341. 10.1016/j.molcel.2014.06.003. [PubMed: 25002142] [30]. Lam SS, Martell JD, Kamer KJ, Deerinck TJ, Ellisman MH, Mootha VK, Ting AY, Directed evolution of APEX2 for electron microscopy and proximity labeling, Nat Methods. 12 (2015) 51–54. 10.1038/nmeth.3179. [PubMed: 25419960] [31]. Zhou Y, Wang G, Wang P, Li Z, Yue T, Wang J, Zou P, Expanding APEX2 Substrates for Proximity-Dependent Labeling of Nucleic Acids and Proteins in Living Cells, Angew Chem Int Ed Engl. 58 (2019) 11763–11767. 10.1002/anie.201905949. [PubMed: 31240809] [32]. Lobingier BT, Hüttenhain R, Eichel K, Miller KB, Ting AY, von Zastrow M, Krogan NJ, An Approach to Spatiotemporally Resolve Protein Interaction Networks in Living Cells, Cell. 169 (2017) 350–360.e12. 10.1016/j.cell.2017.03.022. [PubMed: 28388416] [33]. Padrón A, Iwasaki S, Ingolia NT, Proximity RNA Labeling by APEX-Seq Reveals the Organization of Translation Initiation Complexes and Repressive RNA Granules, Mol. Cell 75 (2019) 875–887.e5. 10.1016/j.molcel.2019.07.030. [PubMed: 31442426] J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 13 [34]. Fazal FM, Han S, Parker KR, Kaewsapsak P, Xu J, Boettiger AN, Chang HY, Ting AY, Atlas of Subcellular RNA Localization Revealed by APEX-Seq, Cell. 178 (2019) 473–490.e26. 10.1016/ j.cell.2019.05.027. [PubMed: 31230715] [35]. Kalderon D, Roberts BL, Richardson WD, Smith AE, A short amino acid sequence able to specify nuclear location, Cell. 39 (1984) 499–509. 10.1016/0092-8674(84)90457-4. [PubMed: 6096007] [36]. Malim MH, Böhnlein S, Hauber J, Cullen BR, Functional dissection of the HIV-1 Rev trans- activator--derivation of a trans-dominant repressor of Rev function, Cell. 58 (1989) 205–214. [PubMed: 2752419] [37]. Fischer U, Huber J, Boelens WC, Mattaj IW, Lührmann R, The HIV-1 Rev activation domain is a nuclear export signal that accesses an export pathway used by specific cellular RNAs, Cell. 82 (1995) 475–483. 10.1016/0092-8674(95)90436-0. [PubMed: 7543368] [38]. Fridell RA, Bogerd HP, Cullen BR, Nuclear export of late HIV-1 mRNAs occurs via a cellular protein export pathway, Proc Natl Acad Sci U S A. 93 (1996) 4421–4424. 10.1073/ pnas.93.9.4421. [PubMed: 8633082] [39]. Nishi K, Yoshida M, Fujiwara D, Nishikawa M, Horinouchi S, Beppu T, Leptomycin B targets a regulatory cascade of crm1, a fission yeast nuclear protein, involved in control of higher order chromosome structure and gene expression, J Biol Chem. 269 (1994) 6320–6324. [PubMed: 8119981] [40]. Güttler T, Madl T, Neumann P, Deichsel D, Corsini L, Monecke T, Ficner R, Sattler M, Görlich D, NES consensus redefined by structures of PKI-type and Rev-type nuclear export signals bound to CRM1, Nat Struct Mol Biol. 17 (2010) 1367–1376. 10.1038/nsmb.1931. [PubMed: 20972448] [41]. Kosugi S, Hasebe M, Tomita M, Yanagawa H, Nuclear export signal consensus sequences defined using a localization-based yeast selection system, Traffic. 9 (2008) 2053–2062. 10.1111/ j.1600-0854.2008.00825.x. [PubMed: 18817528] [42]. Xu D, Farmer A, Collett G, Grishin NV, Chook YM, Sequence and structural analyses of nuclear export signals in the NESdb database, Mol Biol Cell. 23 (2012) 3677–3693. 10.1091/ mbc.E12-01-0046. [PubMed: 22833565] [43]. Patenaude A-M, Orthwein A, Hu Y, Campo VA, Kavli B, Buschiazzo A, Di Noia JM, Active nuclear import and cytoplasmic retention of activation-induced deaminase, Nat Struct Mol Biol. 16 (2009) 517–527. 10.1038/nsmb.1598. [PubMed: 19412186] [44]. Xu D, Marquis K, Pei J, Fu S-C, Cağatay T, Grishin NV, Chook YM, LocNES: a computational tool for locating classical NESs in CRM1 cargo proteins, Bioinformatics. 31 (2015) 1357–1365. 10.1093/bioinformatics/btu826. [PubMed: 25515756] [45]. Bernhofer M, Goldberg T, Wolf S, Ahmed M, Zaugg J, Boden M, Rost B, NLSdb-major update for database of nuclear localization signals and nuclear export signals, Nucleic Acids Res. 46 (2018) D503–D508. 10.1093/nar/gkx1021. [PubMed: 29106588] [46]. Fu S-C, Imai K, Horton P, Prediction of leucine-rich nuclear export signal containing proteins with NESsential, Nucleic Acids Res. 39 (2011) e111. 10.1093/nar/gkr493. [PubMed: 21705415] [47]. Han Y, Branon TC, Martell JD, Boassa D, Shechner D, Ellisman MH, Ting A, Directed Evolution of Split APEX2 Peroxidase, ACS Chem Biol. 14 (2019) 619–635. 10.1021/acschembio.8b00919. [PubMed: 30848125] [48]. Martell JD, Deerinck TJ, Sancak Y, Poulos TL, Mootha VK, Sosinsky GE, Ellisman MH, Ting AY, Engineered ascorbate peroxidase as a genetically encoded reporter for electron microscopy, Nat Biotechnol. 30 (2012) 1143–1148. 10.1038/nbt.2375. [PubMed: 23086203] [49]. Sharp KH, Mewies M, Moody PCE, Raven EL, Crystal structure of the ascorbate peroxidase- ascorbate complex, Nat Struct Biol. 10 (2003) 303–307. 10.1038/nsb913. [PubMed: 12640445] [50]. Li Q, Verma IM, NF-kappaB regulation in the immune system, Nat Rev Immunol. 2 (2002) 725–734. 10.1038/nri910. [PubMed: 12360211] [51]. Aqdas M, Sung M-H, NF-κB dynamics in the language of immune cells, Trends Immunol. 44 (2023) 32–43. 10.1016/j.it.2022.11.005. [PubMed: 36473794] J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 14 [52]. Johnson C, Van Antwerp D, Hope TJ, An N-terminal nuclear export signal is required for the nucleocytoplasmic shuttling of IkappaBalpha, EMBO J. 18 (1999) 6682–6693. 10.1093/emboj/ 18.23.6682. [PubMed: 10581242] [53]. Huang TT, Kudo N, Yoshida M, Miyamoto S, A nuclear export signal in the N-terminal regulatory domain of IkappaBalpha controls cytoplasmic localization of inactive NF-kappaB/ IkappaBalpha complexes, Proc Natl Acad Sci U S A. 97 (2000) 1014–1019. 10.1073/ pnas.97.3.1014. [PubMed: 10655476] [54]. Harhaj EW, Sun S-C, Regulation of RelA Subcellular Localization by a Putative Nuclear Export Signal and p50, Mol Cell Biol. 19 (1999) 7088–7095. [PubMed: 10490645] [55]. Han S, Zhao BS, Myers SA, Carr SA, He C, Ting AY, RNA-protein interaction mapping via MS2- or Cas13-based APEX targeting, Proc Natl Acad Sci U S A. 117 (2020) 22068–22079. 10.1073/pnas.2006617117. [PubMed: 32839320] [56]. Myers SA, Wright J, Peckner R, Kalish BT, Zhang F, Carr SA, Discovery of proteins associated with a predefined genomic locus via dCas9-APEX-mediated proximity labeling, Nat Methods. 15 (2018) 437–439. 10.1038/s41592-018-0007-1. [PubMed: 29735997] [57]. Gao XD, Tu L-C, Mir A, Rodriguez T, Ding Y, Leszyk J, Dekker J, Shaffer SA, Zhu LJ, Wolfe SA, Sontheimer EJ, C-BERST: defining subnuclear proteomic landscapes at genomic elements with dCas9-APEX2, Nat Methods. 15 (2018) 433–436. 10.1038/s41592-018-0006-2. [PubMed: 29735996] [58]. Dalton DA, Russell SA, Hanus FJ, Pascoe GA, Evans HJ, Enzymatic reactions of ascorbate and glutathione that prevent peroxide damage in soybean root nodules, Proc Natl Acad Sci U S A. 83 (1986) 3811–3815. 10.1073/pnas.83.11.3811. [PubMed: 16593704] [59]. Klapheck S, Zimmer I, Cosse H, Scavenging of Hydrogen Peroxide in the Endosperm of Ricinus communis by Ascorbate Peroxidase, Plant and Cell Physiology. 31 (1990) 1005–1013. 10.1093/ oxfordjournals.pcp.a077996. [60]. Mittler R, Zilinskas BA, Purification and characterization of pea cytosolic ascorbate peroxidase, Plant Physiol. 97 (1991) 962–968. 10.1104/pp.97.3.962. [PubMed: 16668537] [61]. Patterson WR, Poulos TL, Crystal structure of recombinant pea cytosolic ascorbate peroxidase, Biochemistry. 34 (1995) 4331–4341. 10.1021/bi00013a023. [PubMed: 7703247] [62]. Maruta T, Inoue T, Noshi M, Tamoi M, Yabuta Y, Yoshimura K, Ishikawa T, Shigeoka S, Cytosolic ascorbate peroxidase 1 protects organelles against oxidative stress by wounding- and jasmonate-induced H(2)O(2) in Arabidopsis plants, Biochim Biophys Acta. 1820 (2012) 1901– 1907. 10.1016/j.bbagen.2012.08.003. [PubMed: 22921811] [63]. Choi-Rhee E, Schulman H, Cronan JE, Promiscuous protein biotinylation by Escherichia coli biotin protein ligase, Protein Sci. 13 (2004) 3043–3050. 10.1110/ps.04911804. [PubMed: 15459338] [64]. Branon TC, Bosch JA, Sanchez AD, Udeshi ND, Svinkina T, Carr SA, Feldman JL, Perrimon N, Ting AY, Efficient proximity labeling in living cells and organisms with TurboID, Nat Biotechnol. 36 (2018) 880–887. 10.1038/nbt.4201. [PubMed: 30125270] [65]. Hanke T, Szawlowski P, Randall RE, Construction of solid matrix-antibody-antigen complexes containing simian immunodeficiency virus p27 using tag-specific monoclonal antibody and tag- linked antigen, J Gen Virol. 73 (Pt 3) (1992) 653–660. 10.1099/0022-1317-73-3-653. [PubMed: 1372038] [66]. Evans EL, Becker JT, Fricke SL, Patel K, Sherer NM, HIV-1 Vif’s Capacity To Manipulate the Cell Cycle Is Species Specific, J. Virol 92 (2018). 10.1128/JVI.02102-17. [67]. Shaner NC, Lambert GG, Chammas A, Ni Y, Cranfill PJ, Baird MA, Sell BR, Allen JR, Day RN, Israelsson M, Davidson MW, Wang J, A bright monomeric green fluorescent protein derived from Branchiostoma lanceolatum, Nat. Methods 10 (2013) 407–409. 10.1038/nmeth.2413. [PubMed: 23524392] [68]. Ory DS, Neugeboren BA, Mulligan RC, A stable human-derived packaging cell line for production of high titer retrovirus/vesicular stomatitis virus G pseudotypes, Proc Natl Acad Sci U S A. 93 (1996) 11400–11406. 10.1073/pnas.93.21.11400. [PubMed: 8876147] J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 15 [69]. Naldini L, Blömer U, Gallay P, Ory D, Mulligan R, Gage FH, Verma IM, Trono D, In vivo gene delivery and stable transduction of nondividing cells by a lentiviral vector, Science. 272 (1996) 263–267. 10.1126/science.272.5259.263. [PubMed: 8602510] [70]. Becker JT, Sherer NM, Subcellular Localization of HIV-1 gag-pol mRNAs Regulates Sites of Virion Assembly, J. Virol 91 (2017). 10.1128/JVI.02315-16. [71]. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, Tinevez J-Y, White DJ, Hartenstein V, Eliceiri K, Tomancak P, Cardona A, Fiji: an open-source platform for biological-image analysis, Nat. Methods 9 (2012) 676–682. 10.1038/nmeth.2019. [PubMed: 22743772] [72]. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, Lopez R, McWilliam H, Remmert M, Söding J, Thompson JD, Higgins DG, Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega, Mol Syst Biol. 7 (2011) 539. 10.1038/msb.2011.75. [PubMed: 21988835] [73]. Gouy M, Tannier E, Comte N, Parsons DP, Seaview Version 5: A Multiplatform Software for Multiple Sequence Alignment, Molecular Phylogenetic Analyses, and Tree Reconciliation, Methods Mol Biol. 2231 (2021) 241–260. 10.1007/978-1-0716-1036-7_15. [PubMed: 33289897] [74]. Crooks GE, Hon G, Chandonia J-M, Brenner SE, WebLogo: a sequence logo generator, Genome Res. 14 (2004) 1188–1190. 10.1101/gr.849004. [PubMed: 15173120] J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 16 HIGHLIGHTS • • • • APEX2 peroxidase has cytoplasmic localization due to a putative NES Most mutations that eliminate NES activity also disrupt peroxidase function One separation-of-function mutant, L242A, shows unbiased cell-wide localization APEX3 (APEX2-L242A) is a more versatile proximity labeling enzyme J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 17 Figure 1. APEX2 is cytoplasmic and leptomycin B sensitive. (A) Schematic of APEX2-dependent biotin labeling of a protein-of-interest (POI). Illustrations of constructs used in Figure 1 and Figure S1 studies. (B) Representative fixed images of the indicated APEX2 constructs expressed stably in HeLa cells (green, mNeonGreen fluorescence; rose, anti-V5 immunostaining; scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. (C) Representative live-cell images of the indicated APEX2 constructs expressed stably in HeLa cells following mock (top) or 2 hrs LMB treatment (bottom) (green, mNeonGreen fluorescence; scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 18 (D) Quantification of cytoplasmic-to-nuclear mNeonGreen fluorescence ratio in experiments from panel C (N=50 cells per condition; ns, not significant; , p < 0.05; *, p < 0.0001 by one-way ANOVA; scale = 10 μm). J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 19 Figure 2. Putative APEX2 NES functions autonomously upon fusion to mCherry. (A) Amino acid alignment of established NES peptides, NESAPEX2, and motif found in Glycine hispida apx1. (B) Schematic of constructs used in Figure 2. (C) Representative live-cell images of the indicated mCherry constructs expressed transiently in HeLa cells following mock (top) or 2 hrs LMB treatment (bottom) (pink, mCherry fluorescence; scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. (D) Quantification of cytoplasmic-to-nuclear mCherry fluorescence ratio in experiments from panel C (N=50 cells per condition; ns, not significant; *, p < 0.0001 by one-way ANOVA; scale = 10 μm). J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 20 (E) Representative fixed images of HeLa cells transfected with indicated mCherry or mCherry-NES constructs. The single amino acid substitution mutants (sequence in panel A) are derivatives of mCherry-NESAPEX2 (scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 21 Figure 3. Putative APEX2 NES residues are mostly required for peroxidase activity. (A) Ribbon schematic of the Glycine hispida APX1 crystal structure (PDB: 1OAG) with NES highlighted in magenta and the heme in red. The NES in APEX2 is identical (Figure 2A). (B) Schematic of the mNG-APEX2-V5 construct and the approximate position of NESAPEX2 residues. (C) Representative live-cell images of the indicated mNG-APEX2-V5 constructs expressed stably in HeLa cells (green, mNG fluorescence; scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. (D) Quantification of cytoplasmic-to-nuclear mNeonGreen fluorescence ratio in experiments from panel C (N=50 cells per condition; ns, not significant; , p < 0.0001 by one-way J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 22 ANOVA compared to APEX2; scale = 10 μm). Gray dashed lines highlight values for APEX2 (top) and L242A (bottom). (E) Representative fixed images of mNG-APEX2-V5 and the indicated L242 mutants expressed stably in HeLa cells showing mNeonGreen fluorescence (green), anti-V5 staining (rose), and streptavidin staining to detect peroxide-dependent biotinylation (blue) (scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. J Mol Biol. Author manuscript; available in PMC 2023 July 01. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Becker et al. Page 23 Figure 4. APEX3 localizes and functions appropriately as fusion to RELA. (A) Schematic of RELA relocalization following TNF stimulation. (B) Representative images of fixed HeLa cells stably expressing RELA-mNG, RELA- APEX2-V5, or RELA-APEX3-V5 following vehicle control or 30 ng/mL TNF treatment for the indicated times (green, mNG fluorescence; rose, anti-V5 immunostaining; scale = 10 μm). Cell and nuclear boundaries outlined with white dashed lines. (C) Quantification of cytoplasmic-to-nuclear fluorescence ratio of cells in panel B (N=50 cells per condition; ns, not significant; *, p < 0.001 by two-way ANOVA). J Mol Biol. Author manuscript; available in PMC 2023 July 01.
10.1084_jem.20221816	ARTICLE Dietary protein shapes the profile and repertoire of intestinal CD4+ T cells Ainsley Lockhart1, Aubrey Reed1, Tiago Rezende de Castro1, Calvin Herman1, Maria Cecilia Campos Canesso1, and Daniel Mucida1,2 The intestinal immune system must tolerate food antigens to avoid allergy, a process requiring CD4+ T cells. Combining antigenically defined diets with gnotobiotic models, we show that food and microbiota distinctly influence the profile and T cell receptor repertoire of intestinal CD4+ T cells. Independent of the microbiota, dietary proteins contributed to accumulation and clonal selection of antigen-experienced CD4+ T cells at the intestinal epithelium, imprinting a tissue-specialized transcriptional program including cytotoxic genes on both conventional and regulatory CD4+ T cells (Tregs). This steady state CD4+ T cell response to food was disrupted by inflammatory challenge, and protection against food allergy in this context was associated with Treg clonal expansion and decreased proinflammatory gene expression. Finally, we identified both steady-state epithelium-adapted CD4+ T cells and tolerance-induced Tregs that recognize dietary antigens, suggesting that both cell types may be critical for preventing inappropriate immune responses to food. Introduction Large quantities of food-derived antigens are absorbed through the intestine each day which must be tolerated by the immune system to avoid food allergies. Oral tolerance, a key mechanism whereby oral administration of antigen results in both local and systemic tolerance to that antigen, requires CD4+ T cells, in- cluding specifically regulatory T cells (Tregs; Garside et al., 1995; Hadis et al., 2011; Josefowicz et al., 2012; Mucida et al., 2005; Pabst and Mowat, 2012). However, T cell responses to dietary antigens have primarily been characterized using monoclonal T cell receptor (TCR) transgenic systems which do not represent a physiological immune response. Polyclonal CD4+ T cell re- sponses to food, including TCR-specific selection and functional differentiation, remain largely uncharacterized and are critical for understanding mechanisms of tolerance and allergy. CD4+ T cells occupy two major adjacent tissue compartments in the intestine, the lamina propria (LP) and epithelium (IE), which are segregated by a basement membrane and are im- munologically distinct. Tregs, which are critical mediators of intestinal inflammation, are enriched in the LP but relatively rare in the highly selective IE (Sujino et al., 2016). Upon mi- gration to the IE, both conventional CD4+ T cells and Tregs can undergo stepwise acquisition of a specialized transcriptional program, upregulating genes associated with tissue residency (CD103, CD69), cytotoxicity (granzymes), natural killer function (NKG7), and CD8+ T cell lineage (Runx3; London et al., 2021; Mucida et al., 2013; Reis et al., 2013; Sujino et al., 2016). At the terminal point of this differentiation process, IE-adapted CD4+ T cells upregulate the CD8αα homodimer, which is proposed to dampen TCR signaling resulting in a high activation threshold. (Cheroutre and Lambolez, 2008). Our recent findings suggest that IE-adapted CD4+ T cells play a complementary anti- inflammatory role to Tregs, providing an important regulatory mechanism in the gut epithelium where Tregs are rare (Bilate et al., 2016, 2020; Bousbaine et al., 2022; Sujino et al., 2016). Disruption of Treg generation (Bouziat et al., 2017; Josefowicz et al., 2012; Torgerson et al., 2007) or epithelial T cell pro- gramming (Reis et al., 2013; Sujino et al., 2016) leads to intestinal inflammation. Additionally, IE-adapted CD4+ T cells can con- tribute to immune regulation toward dietary antigens (Sujino et al., 2016). However, IE-adapted CD4+ T cells also have proin- flammatory potential and can play a pathological role in a dys- regulated response to dietary antigens, as is seen in Celiac disease (Abadie et al., 2012, 2020; Costes et al., 2019; Fina et al., 2008). Here, we characterize polyclonal intestinal CD4+ T cell re- sponses to dietary protein and demonstrate that their prevalent steady-state fate in the intestine is the acquisition of an epi- thelium residency-associated transcriptional profile including expression of cytotoxicity-associated genes. We further dem- onstrate how intestinal T cell responses to food are altered in active tolerance or allergy to favor tissue influx of Tregs or proinflammatory T helper cells, respectively. These findings suggest that epithelium-adapted CD4+ T cells in addition to Tregs contribute to homeostatic immune responses to food. ............................................................................................................................................................................. 2Howard Hughes Medical Institute, The Rockefeller University, New York, NY, USA. 1Laboratory of Mucosal Immunology, The Rockefeller University, New York, NY, USA; Correspondence to Ainsley Lockhart: [email protected]; Daniel Mucida: [email protected]. © 2023 Lockhart et al. This article is available under a Creative Commons License (Attribution 4.0 International, as described at https://creativecommons.org/licenses/by/4.0/). Rockefeller University Press J. Exp. Med. 2023 Vol. 220 No. 8 e20221816 https://doi.org/10.1084/jem.20221816 1 of 18 Results Dietary signals promote accumulation of mature CD4+ T cells in the gut epithelium To characterize intestinal T cell responses to food protein, we developed a protein antigen-free solid diet (AA) containing free amino acids. AA diet lacks polypeptides but does contain other dietary macromolecules including simple and complex carbo- hydrates (sucrose, corn starch), lipids (corn oil), and fiber (cel- lulose; Table S1). Standard chow diet by contrast is highly complex, comprised of whole food ingredients (e.g., wheat, corn, soybeans) which contain hundreds of distinct proteins and di- verse dietary metabolites. Chow diet can therefore impact in- testinal T cells through dietary antigen-specific TCR stimulation, immune costimulation by dietary metabolites (some in a microbiota-dependent manner), and secondary stimulation via diet-induced changes to the microbiota. Specific pathogen-free (SPF) C57BL/6 mice weaned onto the AA diet were similar to standard chow diet mice in weight gain, intestinal inflammation measured by fecal lipocalin-2, and se- rum nutritional biomarkers (Fig. S1, A and B). Although the length of the small intestine was reduced in 8-wk-old AA diet mice (Fig. S1 C), histological examination revealed no evidence of tissue damage or inflammation and minimal differences in tissue architecture (Fig. S1, D and E). Finally, we found no sig- nificant differences in intestinal myeloid cell frequencies in SPF AA vs. chow diet mice (Fig. S1 F). Altogether, AA diet mice ap- pear heathy and display no evidence of intestinal damage or inflammation. We assessed the two major intestinal T cell compartments, the IE and LP, by flow cytometry and found increased TCRαβ+ CD4+ T cells in the small intestine IE of 8-wk-old mice weaned onto chow died compared with AA diet (Fig. 1 A). Total TCRαβ+ T cells were also increased in the small intestine LP of chow diet mice; however, chow diet did not promote increased frequency of CD4+ T cells in this compartment (Fig. S2 A). Antigen- experienced CD44+ CD62L− cells, including tissue-adapted pre- CD8αα+ (CD103+ CD8αα−) and CD8αα+ subsets (London et al., 2021), accounted for the vast majority of IE CD4+ T cells in- duced by exposure to chow diet (Fig. 1, B and C). Whereas the absolute number of IE Tregs was also slightly increased in chow diet, they were reduced by relative frequency out of CD4+, suggesting that exposure to dietary signals favors IE-adapted subsets over Tregs (Fig. 1, B and C). Additionally, this may suggest that the epithelial conversion of Tregs into CD4+ CD8αα+ T cells may be impaired in AA-diet mice (Sujino et al., 2016). Among LP CD4+ T cells, chow diet mice had a slight increase in the absolute number of CD44+ cells but no increase in Tregs compared with AA diet (Fig. S2 B). Additionally, we did not find increased Rorγt− peripherally induced Tregs (pTregs), which reportedly respond to a chow diet in a germ-free setting (Kim et al., 2016), in either the IE or LP (Fig. S2 C). We expected to find the greatest impact of dietary protein in the small intestine, the primary site of food absorption. Consistent with this, we found no chow-associated increase in large intestine IE CD4+ T cells, including CD44+ CD4+, pre-CD8αα+ (CD103+ CD8αα−), or Tregs, and only a small increase in CD4+ CD8αα+ T cells (Fig. S2 D). Altogether, these data demonstrate that a complex diet promotes maturation and epithelial adaptation of small intestine CD4+ T cells, whereas the large intestine and LP are relatively less affected, indicative of a distinct and localized T cell response to food. To assess the impact of dietary antigen on T cell phenotypes in a highly controlled manner, we next supplemented AA diet with a low dose of OVA (0.1% in drinking water, ∼5 mg/d) in- tended to model the amount of a single food protein present in a protein-diverse diet. OVA supplied in drinking water at this dose for 2 d was sufficient to stimulate OVA-specific TCR transgenic CD4+ T cells in vivo in gut-draining lymph nodes (Fig. S2 E). SPF C57BL/6 mice fed AA + OVA had slightly yet significantly ele- vated absolute counts of total IE CD4+ T cells, CD44+ CD4+ T cells, and CD4+ CD8αα+ T cells relative to AA diet mice, though the frequency was unchanged (Fig. S2 F). We found no difference in pre-CD8αα+ (CD103+ CD8αα−), Tregs from the IE, or any tested LP T cell subsets (Fig. S2, F and G). To address whether different or more diverse food proteins could further impact intestinal CD4+ T cells, we weaned SPF mice onto AA diets supplemented with either cow milk casein or a mix of casein, gluten, and soy protein. Comparing frequencies of IE CD4+ CD8αα+ T cells and Tregs in the proximal small intestine, we found no difference between AA and casein, but increased CD4+ CD8αα+ T cells and decreased Tregs in casein–gluten–soy diet, suggesting that the Treg to CD4+ CD8αα+ differentiation pathway (Sujino et al., 2016) was induced in mice fed more diverse proteins (Fig. S2 H). These results indicate that dietary protein can promote IE CD4+ T cell accumulation and maturation in SPF mice and that increased diversity of dietary protein such as in casein– gluten–soy or standard chow diet may have an additive effect. Dietary signals promote microbiota-independent epithelial and cytotoxic programming of CD4+ T cells Diet highly influences the gut microbiota, which can lead to indirect downstream effects on local immune cells (Sonnenburg and B¨ackhed, 2016). Indeed, 16S rRNA sequencing of SPF AA versus chow diet mice revealed distinct gut microbiome compo- sition, with chow diet promoting increased microbial diversity in both the small intestine and cecum (Fig. 2, A–C; Fig. S2, I and J; and Table S2). To better characterize the microbiota-independent di- etary impact on immune cells, we established our dietary models in germ-free (GF) mice and gnotobiotic mice colonized with Oligo- MM12, a stable, vertically transmissible consortium of 12 com- mensal strains representing members of the major bacterial phyla in the murine gut (Brugiroux et al., 2016). GF AA-diet mice lack exposure to foreign protein antigens and provide an ideal model to assess the impact of diet independent of the microbiota, while Oligo-MM12 provides a more physiological model with some co- stimulation from commensal bacteria while still limiting the di- versity and complexity of intestinal antigen. Similar to SPF, intestinal length was reduced in GF AA-diet mice compared with chow diet (see Fig. S1 C), and the tissue showed no evidence of damage, inflammation, or major mor- phological changes (see Fig. S1, D and E). GF chow diet mice had elevated frequencies of LP macrophages but no other differences between the diets were observed within the myeloid compart- ment (see Fig. S1 F). Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 2 of 18 Figure 1. Dietary signals promote accumula- tion and adaptation of intestinal CD4+ T cells in the small intestine epithelium of specific pathogen-free mice. (A–C) Flow cytometry from the IE of SPF mice fed AA or standard chow diet measuring frequency or absolute count of the indicated cell subsets (A–C) or showing rep- resentative flow plots pre-gated on CD4+ T cells (B). (A–C) Mean ± SD from three to five inde- pendent experiments with 14–18 mice per con- dition. Unpaired t tests, P < 0.01, P < 0.0001. GF and Oligo-MM12 mice had highly reduced intestinal CD4+ T cells, similar to levels seen in SPF AA diet mice (Fig. 2 D; and Fig. S2, K and L), suggesting that both complex diet and a complex microbiota are required for steady-state gut T cell ac- cumulation. Within GF or Oligo-MM12 mice, chow diet once again increased IE TCRαβ+ T cells relative to the AA diet (Fig. 2 D and Fig. S2 K). Although CD4+ T cell frequency was not im- pacted, the chow diet increased frequencies of CD44+, pre- CD8αα+ (CD103+ CD8αα−), and CD8αα+ CD4+ T cell subsets (Fig. 2 D and Fig. S2 K), demonstrating that dietary signals promote IE CD4+ T cell maturation and tissue adaptation inde- pendent of the microbiota. To address how diet impacts functional gene expression pathways of intestinal CD4+ T cells in greater detail, we performed single-cell RNA sequencing (scRNAseq) of CD4+ T cells from the IE and LP of GF or Oligo-MM12 mice fed AA, AA + OVA, or chow using the Chromium 10X (10X Genomics) platform (Fig. S3 A). Sequenced cells were assigned to 11 major unbiased clusters which we defined based on their top differentially expressed genes (Fig. 2 E, Fig. S3 B, and Table S3). We observed high frequencies of na¨ıve and NKT cells, particularly in GF or AA diet mice (Fig. 2 F). Conversely, mature cells, which formed three clusters in the IE and a single major cluster defined by Th1-type genes in the LP, Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 3 of 18 Figure 2. Chow diet promotes microbiota-independent epithelial adaptation and cytotoxic transcriptional programming of intestinal CD4+ T cells. (A–C) 16S rRNA sequencing of small intestine (SI) or cecum contents of 8-wk-old SPF mice fed AA or standard chow diet represented by detrended cor- respondence analysis (DCA; A), relative SI phyla abundance (B), and SI Chao1 alpha diversity with mean ± SD and unpaired t test, *P < 0.0001 (C). Data is from four independent experiments using 11–15 mice per condition. (D) Flow cytometry from the IE of GF or Oligo-MM12 mice fed AA or standard chow diet measuring frequency of the indicated cell subsets. Dashed lines show mean value from SPF Chow (red) or SPF AA (blue). Mean + SD from three to five ˇ independent experiments with 7–16 mice per condition. Two-way ANOVA P values beneath each plot, and Holm- Sid´ak multiple comparison test between diets within each colonization within each plot, P < 0.05, P < 0.01 *P < 0.0001. (E–J) scRNAseq of 12,139 IE and LP CD4+ T cells from GF or Oligo-MM12 mice fed AA, AA + OVA, or standard chow diet with two to four mice per condition. (E) UMAP visualization of sequenced cells positioned by gene expression similarity and colored by gene expression cluster. (F) Frequency of cells within each cluster from the IE (top) or LP (bottom). (G and H) Expression (Pearson Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 4 of 18 residuals) of IE signature genes within the three IE mature clusters (G) or within all IE CD4+ T cells (H). For H, Wilcoxon rank sum test with Bonferroni correction for multiple comparison, P-adj < 1e−5 were considered statistically significant. Groups labeled with asterisk () are significantly higher than AA diet mice within the same colonization group. Groups labeled with a circle (+) are significantly higher than GF mice from the same dietary group. (I) IE gene signature score grouped by condition. Each data point contributing to the violin plots represents a single sequenced cell. Wilcoxon rank sum test within each colonization group, P-adj < 1e−5. (J) Three-way volcano plot showing differential gene expression between diets in all sequenced IE CD4+ T cells. Colored genes are differentially expressed (P-adj < 0.05 from FDR-corrected Kruskal–Wallis Test and log2 fold change > 0.5), colored by the diet(s) in which they are upregulated. Select genes of interest are labeled on each plot. were increasingly frequent Oligo-MM12 or chow diet mice (Fig. 2 F and Fig. S3 C). We assessed expression of signature genes associated with stepwise CD4+ T cell transcriptional adaptation to the IE (London et al., 2021) and found that our IE mature clusters represented three stages of signature gene acquisition (Fig. 2 G, Fig. S3 D, and Table S3). IE 1 cells were least adapted to the epithelium, ex- pressing Nkg7, Ccl5, and Cd160, which were expressed in all three clusters (Fig. 2 G). IE 2 matched our pre-CD8αα+ population, expressing CD103 (Itgae), and granzymes (Gzma, Gzmb), and were enriched in the chow diet relative to AA or AA + OVA regardless of colonization (Fig. 2, F and G; and Fig. S3 E). IE 3, representing fully IE adapted CD4+ CD8αα+ T cells, additionally expressed Cd8a, Lag3, Prf1 (Perforin), and Cd7, and were rare in all sequenced groups, confirming their dependence on complex microbiota (Fig. 2, F and G; and Fig. S3 E). Comparing the ex- pression of IE signature genes between conditions, we found that IE1 genes (Nkg7, Ccl5, Cd160) were more uniformly ex- pressed whereas IE2 genes (Itgae, Gzma, Gzmb) were expressed almost exclusively in chow diet mice from both GF and Oligo- MM12 (Fig. 2 H, genes significantly upregulated by diet within each colonization are indicated by ). Indeed, when we created a gene signature using all 10 IE hallmark genes, CD4+ T cells from chow diet mice scored higher than AA or AA + OVA diet mice in both GF and Oligo-MM12 (Fig. 2 I). Unbiased three-way com- parison of differential gene expression between AA, AA + OVA, or chow diet mice across GF and Oligo-MM12 further confirmed that IE signature genes including Itgae and granzymes were among the top upregulated genes in chow diet mice (Fig. 2 J and Table S4). Although feeding mice a single food protein (AA + OVA) did not alter frequencies of intestinal CD4+ T cell subsets either by flow cytometry (data not shown) or scRNAseq (see Fig. 2 F; and Fig. S3, C–E), IE CD4+ T cells from Oligo-MM12 AA + OVA mice significantly upregulated Itgae compared with AA and had a trending increase in Gzma (P = 0.007; Fig. 2 H). This difference was not seen in GF, suggesting that a low dose of single food protein is sufficient to promote some epithelial adaptation of CD4+ T cells in a manner that may require costimulation from the microbiota. Total Treg frequency did not vary greatly between diets in GF or Oligo-MM12 (see Fig. 2 F and Fig. S3 F). However, Rorγt− pTregs were increased in chow diet relative to AA diet in GF and Oligo-MM12 (Fig. S3 F), supporting previous reports that these cells respond to food (Kim et al., 2016). We next assessed Treg transcriptional programs within our scRNAseq dataset, identi- fying eight subclusters including one that upregulated IE sig- nature genes (Nkg7, Gzma, Gzmb; Fig. S3, G and I; and Table S5). IE signature Tregs were expanded in the chow diet regardless of colonization, and trended toward an increase in AA + OVA mice from Oligo-MM12 (Fig. 3, A and B). These cells were transcrip- tionally similar to previously identified pTregs on a trajectory toward CD4+ CD8αα+ differentiation (Bilate et al., 2020). In contrast, Il10-high Tregs were almost exclusively found in Oligo- MM12 (Fig. 3 A and Fig. S3 J), and Rorc expression was confined to this population among Tregs (data not shown). This popula- tion may therefore represent Rorγt+ pTregs known to depend on the microbiota (Sefik et al., 2015; Yang et al., 2016) while the IE signature pTregs may represent Rorγt− pTregs reported to de- pend on diet (Kim et al., 2016). Indeed, regardless of colonization status, the chow diet Tregs upregulated IE2 signature genes (Itgae, Gzma, and Gzmb) and scored higher for total IE gene signature (Fig. 3, C and D). Unbiased three-way comparison of differential gene expression between AA, AA + OVA, or chow diet Tregs across GF and Oligo-MM12 further revealed that chow diet led to upregulation of IE signature genes (Itgae, Gzma, Gzmb, and Lag3) and core Treg suppressive function genes (Il10, Ctla4), while a distinct natural Treg-associated transcriptional profile (Gata3, Nrp1, Cd81, and Klrg1) was enriched in AA or AA + OVA diet (Fig. 3 E and Table S6). Similar to our findings in total IE CD4+ T cells, a low dose of OVA did not impact Treg transcriptional profile in GF mice but was sufficient to promote Itgae expression in Oligo-MM12 (Fig. 3 C). Correspondingly, in both total CD4+ T cells and Tregs, Oligo- MM12 promoted expression of IE signature genes including Gzma and Itgae over GF mice with the same diet (Fig. 2 H and Fig. 3 C; indicated by +), but this depended on the presence of food protein. Exposure to food protein therefore recruits antigen- experienced CD4+ T cells to the intestinal epithelium, imprint- ing a tissue-resident cytotoxic gene expression signature in a manner that is amplified by microbial costimulation. Complex diet induces Granzyme B expression in intestinal T cells To further validate our findings from scRNAseq, we assessed intestinal T cell Granzyme B expression by flow cytometry. In- deed, the chow diet promoted Granzyme B expression among total IE CD4+ T cells and Tregs in both GF and SPF (Fig. 4 A). Additionally, Granzyme B expression was higher among CD8αβ+ T cells, and thymic TCRγδ+ and CD8αα+ TCRαβ+ T cells (Fig. 4 B). Whereas diet-induced Granzyme B expression among CD4+ and CD8αβ+ T cells was boosted in the presence of microbiota, the impact on thymic-derived T cell subsets was microbiota inde- pendent. Complex diet therefore broadly induces Granzyme B expression in intestinal T cells, although the pathway of in- duction may be distinct depending on the subset. To characterize the kinetics of dietary imprinting on intesti- nal CD4+ T cells, SPF mice were analyzed at 3 wk (pre-weaning), Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 5 of 18 Figure 3. Chow diet imprints an epithelial transcriptional signature on intestinal Tregs. scRNAseq of 1,183 IE and LP Tregs from GF or Oligo-MM12 mice fed AA, AA + OVA, or standard chow diet with two to four mice per condition. (A and B) Frequency of cells in each Treg subcluster (A) or in the IE signature subcluster (B). One-way ANOVA displaying with Tukey’s multiple comparisons test, *P < 0.01. (C) Treg expression (Pearson residuals) of IE2 signature genes, grouped by condition. Wilcoxon rank sum test with Bonferroni correction for multiple comparison, P-adj < 1e−5 were considered statistically significant. Groups labeled with a triangle (Δ) are significantly higher than AA diet mice within the same colonization group. Groups labeled with a circle (+) are significantly higher than GF mice from the same dietary group. (D) Treg IE gene signature score grouped by condition. Each data point contributing to the violin plots represents a single sequenced cell. Wilcoxon rank sum test with P-adj < 1e−5 within each colonization group displayed on the plot. (E) Three-way volcano plot showing differential gene expression between diets in all Tregs. Colored genes are differentially expressed (P-adj < 0.05 from FDR-corrected Kruskal–Wallis Test and log2 fold change > 0.5), colored by the diet(s) in which they are upregulated. Select genes of interest are labeled on each plot. 8 wk, or 12 wk. Some mice that were fed AA or chow diet until 8 wk old (i.e., following the standard protocol for prior ex- periments) were switched to the opposite diet until endpoint analysis at 12 wk old (Fig. 4 C). This experimental setup enabled us to assess intestinal CD4+ T cell fate if dietary signals were removed from adult mice (Chow to AA) or if exposure to a complex diet was delayed until adulthood (AA to Chow). Pre- weaning (3-wk-old) mice had very low frequencies of CD8αα+ or Granzyme B+ CD4+ T cells (Fig. 4 D). Mice that were subse- quently weaned onto chow diet had progressively increased frequencies of these cells by 8 and then 12 wk old, whereas AA diet mice maintained a weanling-like immature IE CD4+ T cell profile (Fig. 4 D). Switching AA to chow diet at 8 wk did not recover CD4+ CD8αα+ T cell frequency but did rescue Granzyme B expression (Fig. 4 D, red squares). Conversely, chow-induced CD4+ CD8αα+ T cells persisted in mice that were switched to AA diet at 8 wk, whereas Granzyme B expression declined (Fig. 4 D, blue squares). This may indicate a critical window during weaning and development during which diet imprints an epi- thelial signature on intestinal CD4+ T cells, whereas once CD4+ CD8αα+ develop, they can persist at least 4 wk without a chow diet. By contrast, Granzyme B expression is dependent on re- cent exposure to a complex diet. Our previous work has demonstrated that IE adaptation by CD4+ T cells is linked to Granzyme B upregulation (Bilate et al., 2020; London et al., 2021; Mucida et al., 2013; Reis et al., 2013). However, in these diet switch experiments, Granzyme B ex- pression and IE adaptation were uncoupled such that Chow-to- AA mice have IE-adapting CD4+ T cells that no longer express Granzyme B (Fig. 4 E). Altogether, these data demonstrate that dietary signals in conjunction with the microbiota promote both CD4+ CD8αα+ differentiation and Granzyme B expression in intestinal CD4+ T cells, though the temporal dynamics and therefore pathways of induction appear distinct. Exposure to dietary protein drives clonal selection of intestinal CD4+ T cells We next assessed how routine exposure to dietary protein shapes the TCR repertoire of intestinal CD4+ T cells by analyzing TCRs from our scRNAseq of GF and Oligo-MM12 mice fed AA, AA + OVA, or chow. IE and LP mature CD4+ T cells were overall highly clonally expanded despite the absence of complex mi- crobiota, whereas na¨ıve cells and NKT cells (which use an in- variant TCRα but diverse TCRβ) were highly diverse as expected (Fig. 5 A). We estimated repertoire diversity across sample groups using D50 in which repertoires are scored from 0 (least diverse) to 0.5 (most diverse) and found a trend toward higher clonal diversity in GF AA or AA + OVA compared with the other groups (Fig. 5 B). However, within mature (IE1, IE2, IE3, LP Th1) CD4+ T cells or Tregs, we did not observe major differences in repertoire diversity across sample groups (Fig. 5 B), demon- strating that while few mature CD4+ T cells accumulate in the Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 6 of 18 Figure 4. Chow diet induces Granzyme B expression in intestinal T cells. (A and B) Flow cytometry analysis of Granzyme B expression within IE T cell subsets from 8-wk-old SPF or GF mice fed AA or standard chow diet. Mean ± SD from two to three independent experiments with 7–11 mice per condition. ˇ Unpaired t tests (A) or two-way ANOVA P values beneath each plot, and P < 0.05 from Holm- Sid´ak multiple comparison test between diets within each colonization displayed on each plot (B). (C) Schematic of weaning and diet switch experiments. (D) Flow cytometry analysis of 3-, 8-, and 12-wk-old SPF mice fed chow or AA diets according to schematic C. Mean ± SEM from two to five independent experiments with 5–18 mice per condition. One-way ANOVA comparing conditions in 12-wk-old mice displaying P < 0.05 from Tukey’s multiple comparison test to the right of each plot. Unpaired t tests comparing ˇ Sid´ak correction for multiple comparisons displayed on plot. (E) Flow cytometry of Granzyme B ex- consecutive timepoints between conditions with Holm- pression within IE T cell subsets from 12-wk-old SPF mice fed chow or AA diets according to schematic C. Mean ± SD from two independent experiments with five to seven mice per condition. Two-way ANOVA displaying P < 0.05 from Dunnett’s multiple comparison test comparing each group against chow only. (A–E) P < 0.05, P < 0.01, P < 0.001, *P < 0.0001. intestine in absence of major foreign antigen exposure, those that do are still clonally expanded. We recently reported that gut-associated germinal centers are highly reduced in GF AA diet mice, although B cells from these germinal centers still exhibit clonal selection and expansion (Nowosad et al., 2020), suggest- ing a parallel mechanism for intestinal B cells. To determine whether food antigens can drive the clonal selection of intestinal CD4+ T cells, we examined shared (public) clones between mice, as defined by identical paired TCRα and TCRβ CDR3 amino acid sequences (Fig. 5, C and D). Among GF mice where the primary source of foreign antigen is food, we found no clonal overlap between AA diet mice, a low level of clonal sharing between AA + OVA mice, and extensive clonal sharing between chow diet mice, raising the possibility that food antigen is driving clonal selection. We additionally found a small amount of clonal overlap between AA + OVA and chow diet mice, which may be driven by self or unaccounted environ- mental antigens. There was more clonal overlap in general be- tween Oligo-MM12 mice, likely due to recognition of microbial antigen and/or a co-stimulatory effect from microbiota. Never- theless, between diets in Oligo-MM12, we found similar trends to GF, with no clonal sharing between AA diet mice, low level sharing between AA + OVA mice, and a higher degree of sharing between chow diet mice. When we compared clonal overlap across GF and Oligo-MM12 mice, we found some additional sharing between AA + OVA mice, whereas clones were highly shared between chow diet mice. These data point to clonal se- lection of intestinal CD4+ T cells by dietary antigen, where ex- posure to a low dose of a single food protein leads to a low level of clonal selection, while exposure to diverse food proteins in the context of complex chow results in a high level of selection. In this case, presence of diverse dietary metabolites may have an additional adjuvant effect, promoting differentiation and ex- pansion of food antigen-specific T cells. Context of exposure shapes intestinal CD4+ T cell responses to food protein To better understand how food-induced gut CD4+ T cells con- tribute to tolerance and how these responses are perturbed in the context of food allergy, we combined a na¨ıve T cell fate- mapping mouse model (iSellTomato; Merkenschlager et al., 2021) with a cholera toxin (CT) mouse model of food allergy (Jimenez-Saiz et al., 2017). iSellTomato, in which CD62L+ na¨ıve cells are permanently labeled with Tomato fluorescence upon tamoxifen administration, enables identification of “ex-na¨ıve” (Tomato+ CD62L–) T cells that matured and entered the gut since the time of tamoxifen labeling (Parsa et al., 2022). We utilized this strategy to enrich for intestinal CD4+ T cells re- sponding to dietary protein in three contexts: (1) feeding, where mice are fed OVA, representing the steady-state re- sponse to food; (2) allergy, where primary exposure to OVA is with CT, resulting in allergic sensitization; and (3) tolerance, where mice are fed OVA alone prior to OVA/CT, resulting in oral tolerance and protection against allergy (Fig. 6 A). Tomato+ CD62L– cells analyzed on day 26 of our treatment protocol therefore include all cells that matured and entered the gut during the 4-wk treatment protocol, which will be enriched for cells responding to OVA but may also include cells responding Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 7 of 18 Figure 5. Exposure to dietary protein drives clonal selection of intestinal CD4+ T cells. scTCRseq of 12,139 IE and LP CD4+ T cells from GF or Oligo-MM12 mice fed AA, AA + OVA, or standard chow diet using two to four mice per condition. (A) Clonal expansion (by TCR nucleotide sequence) of cells visualized by UMAP (left) and bar plot of gene expression clusters (right). (B) D50 in which repertoires are scored from 0 (least diverse) to 0.5 (most diverse) within all cells (top left), mature clusters IE1, IE2, IE3, and LP Th1 combined (top right), and Tregs (bottom left). (C and D) Clonal sharing between mice defined by paired TCRα and TCRβ CDR3 amino acid sequence. NKT cells were discarded from analysis. (C) Circos plots in which each segment represents a mouse, colored by diet and sized by cell count. Links between segments represent public clones which are colored by diet if shared between mice of the same diet or uncolored if shared between mice of different diets. (D) Morisita overlap index heatmaps where each square represents the mean overlap between each mouse in the indicated conditions (left and center) or scatter plot where each dot represents overlap between mice in the same diet (right). Kruskal–Wallis test with Dunn’s multiple comparisons, P < 0.001. to CT, microbiota, and other food antigens since CD4+ T cells are continuously recruited to the intestine. At the day 26 time- point, mice in the allergy group have elevated serum total IgE and OVA-specific IgG1, demonstrating allergic sensitization (Fig. 6 B), but have no observable increase in intestinal tissue damage or inflammation (Fig. S4 A). Upon continuation of this protocol for 4 wk doses of OVA/CT, systemic OVA challenge results in anaphylaxis in allergy mice (Jimenez-Saiz et al., 2017), while mice in the tolerance group are protected (Fig. 6 C). 4 wk after initial tamoxifen administration, ∼15% of small intestine IE or LP CD4+ T cells from SPF mice in the OVA feeding protocol were Tomato+ CD62L– (Fig. 6 D), representing the baseline for CD4+ T cell influx to the intestine at steady state. In both tissues, CD4+ T cell influx was increased over twofold in OVA allergy mice, while OVA tolerance fell between feeding and allergy (Fig. 6 D). By contrast, in the large intestine, there was no increase in CD4+ T cell influx in tolerance or allergy mice (Fig. S4 B), suggesting that responses to food in inflammatory contexts remain localized to the small intestine. Among Tomato− (pre- Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 8 of 18 of Schematic experimental Figure 6. Tracking intestinal CD4+ T cell re- sponses during OVA feeding, tolerance, or allergy. iSellTomato mice were analyzed on day 26 after treatment with tamoxifen to permanently label na¨ıve T cells and then exposure to OVA in the context of feeding, tolerance, or allergy. (A) protocol. (B) Total serum IgE (left) or OVA specific IgG1 (right) as measured by ELISA. One-way ANOVA displaying P < 0.05 from Tukey multiple com- parison test. Data is representative of two–three independent experiments with 4–13 mice per group. (C) Anaphylaxis as measured by body temperature of mice at the indicated times after intraperitoneal OVA injection, following four weekly doses of OVA/CT. Mean ± SEM represen- tative of two independent experiments with five to six mice per group. Unpaired t tests with Holm- ˇSid´ak multiple comparison test. (D and E) Flow cytometry measuring frequency of Tomato+ out of total CD4+ T cells in the IE or LP (D) or of the in- dicated CD4+ T cell subsets out of Tomato+ or Tomato− CD4+ T cells in the IE (E) with repre- sentative flow cytometry plots shown on the right. Mean from four (D) or two (E) independent ex- periments with 10–12 (D) or 5–8 (E) mice per group. One-way ANOVA with Tukey’s multiple comparison test, showing P values < 0.05. (B−E) P < 0.05, P < 0.01, *P < 0.0001. existing non-na¨ıve) CD4+ T cells, mice from each treatment group had similar frequencies of IE2, IE3, Tregs, and cells pro- ducing IL-17A or IL-4 based on flow cytometry analysis (Fig. 6 E; and Fig. S4, C and D). However, profiles of infiltrating Tomato+ CD4+ T cells were distinct in OVA feeding, tolerance, and allergy conditions, suggesting that iSellTomato is an effective system for enriching and characterizing polyclonal intestinal CD4+ T cells responding to food protein in different contexts. In OVA toler- ance, Treg frequency was higher among Tomato+ CD4+ T cells, whereas in OVA allergy, there was increased IL-17A expression (Fig. 6 E and Fig. S4 D), consistent with prior reports that CT induces intestinal Th17 in a microbiota-dependent manner (Zhao et al., 2017). C57BL/6 mice do not typically have a strong Th2 response in the CT allergy model and we did not see sig- nificant increases in IL-4 production among Tomato+ CD4+ T cells from allergy mice (Fig. S4, C and D). IE Tomato+ CD4+ T cells from all three treatment groups demonstrated only partial acquisition of an epithelial profile in the course of 4 wk, with ∼25% expressing CD103 (IE2) but few cells expressing CD8αα (IE3; Fig. S4 C). To transcriptionally characterize the CD4+ T cell response to OVA in the context of feeding, tolerance, and allergy, we per- formed scRNAseq on Tomato+ CD62L– (fate-mapped ex-na¨ıve) and Tomato– CD62L– (pre-existing non-na¨ıve) CD4+ T cells (pooled at a ratio of ∼1:2 Tomato+:Tomato– to evenly enrich for Tomato+ cells) from the IE and LP of SPF iSellTomato mice on day 26 of the treatment protocols (see Fig. 6 A). scRNAseq was performed on the 10X Genomics platform with four to five mice per condition pooled across two independent experiments and sequencing runs (Fig. S4, E and F). Tomato− and Tomato+ populations defined in the scRNAseq data enrich for true pre- existing mature versus ex-na¨ıve cells, whereas there may be some contaminating cells in each population due to incomplete efficiency of the fate mapping mouse model, low detection of Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 9 of 18 Figure 7. Distinct intestinal CD4+ T cell responses to OVA feeding, tolerance, and allergy. scRNAseq of 11,217 Tomato+ and Tomato− CD4+ T cells from the IE and LP of mice on day 26 of OVA feeding, tolerance, or allergy protocols using four to five mice per condition pooled across two independent ex- periments. (A) UMAP visualization of sequenced cells positioned by gene expression similarity and colored by gene expression cluster. (B) Frequency of cells within each cluster from the IE (left) or LP (right) within each sample group. (C) Three-way volcano plots showing differential gene expression between conditions in Tomato+ CD4+ T cells from the IE (top) or LP (bottom). Colored genes are differentially expressed (P-adj < 0.05 from FDR-corrected Kruskal–Wallis Test and log2 fold change > 0.5), colored by the condition(s) in which they are upregulated. Select genes of interest are labeled on each plot. (D) Analysis of Treg subclusters among Tomato+ Tregs (496 total cells) showing frequency of all subclusters (above) or Il10+ or pooled Helios+ subsets (below). (E) Differentially expressed Treg functional genes between Tomato+ Tregs (496 total cells) in different conditions. (D and E) One-way ANOVA with Tukey’s multiple comparison test (D) or Wilcoxon rank sum test corrected with FDR (E), P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. TdTomato/Stop, or labeling of Sell+ central memory cells which could potentially differentiate and enter the gut. Sequenced cells were assigned to 11 major unbiased clusters, which we defined based on their top differentially expressed genes (Fig. 7 A; Fig. S4, G and H; and Table S7). Tomato+ cells were present in all clusters, though they were particularly en- riched among CCR7+ migratory cells and Tregs, and were lowest in IE-adapted clusters (discussed below; Fig. S4 F). Consistent with a weak Th2 response in the C57BL/6 CT allergy model, we did not identify gene expression clusters defined by hallmark Th2 genes (Fig. S5 A). Based on the expression of IE signature genes, we identified IE2 and IE3 clusters comparable with our GF/Oligo-MM12 dataset, but no IE1 cluster, further reinforcing the notion of an additive effect of chow diet and full colonization on CD4+ T cell IE adaptation (Fig. 7 A, Fig. S5 B, and Table S7). An additional IE cluster was identified (IE4), which expressed the full IE3 signature as well as additional genes associated with NK function (Fcer1g, Tyrobp, Klrd1), Treg function (Ikzf2), and apop- tosis (Tox, Bcl2), while Cd4 levels were reduced (Fig. S5 B and Table S7). Tomato labeling frequency progressively decreased from IE2 (40%) to IE3 (15%) to IE4 (5%), confirming that only partial acquisition of an epithelial profile occurs in the course of 4 wk (Fig. S5 C). Comparison of cluster distributions between groups demonstrated that IE4 cells were almost exclusively Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 10 of 18 Tomato+ and from either tolerance or allergy mice (Fig. S5 D and Fig. 7 B). Indeed, three-way differential gene expression analysis among IE Tomato− cells showed that IE4 genes were highly enriched in tolerance and allergy groups compared with the feeding group (Fig. S5 E and Table S8). These data suggest that the IE4 gene expression program was turned on in pre-existing tissue resident IE3 cells in inflammatory (CT-exposed) con- ditions. IE4 may therefore represent an activated state of IE-adapted CD4+ T cells in which they can function in an antigen-independent manner through innate receptors, as pre- viously suggested (Bilate et al., 2020). Of note, the IE4 tran- scriptional program is similar to thymic-derived (CD4−) CD8αα+ TCRαβ+ cells found both in tumors and the steady-state intes- tinal epithelium (Chou et al., 2022; Denning et al., 2007), though with the addition of Ikzf2, which may suggest some regulatory capacity. To assess whether CD4+ T cells that mature and enter the gut during OVA feeding, tolerance, or allergy have distinct charac- teristics, we compared cluster distributions of Tomato+ cells between these groups. In agreement with our observations from flow cytometry, OVA tolerance mice had increased frequencies of Tregs whereas allergy mice had increased Th17 among IE Tomato+ cells; a similar though nonsignificant trend was ob- served in the LP (Fig. 7 B and Fig. S5 F). Three-way differential gene expression found that IE and LP Tomato+ cells from OVA- feeding mice upregulated genes associated with tissue residency and homing (Zfp683, Ccl5), memory (Ccr7, Satb1, Bach2), and type I immune responses (Ly6a, Ifi47). By contrast, OVA allergy To- mato+ cells upregulated Th17-associated genes (Il17a, Il22, Rorc, Them176a; Fig. 7 C, Tables S9, and S10). OVA tolerance mice shared the majority of their upregulated genes, with feeding or allergy mice demonstrating their intermediate inflammatory phenotype. However, in the LP, tolerance alone upregulated Treg-associated genes (Foxp3, Ikzf2, Itgav, Cd83; Fig. 7 C). Finally, LP Tomato+ cells from both tolerance and allergy upregulated Gata3, which was primarily expressed in Tregs in this se- quencing dataset (Fig. 7 C and Fig. S5 A). We next assessed Treg transcriptional programs in detail, identifying five subclusters based on top differentially expressed genes (Fig. S5 G and Table S11). Among Tomato+ Tregs, sub- cluster frequencies varied between conditions. At steady state (OVA feeding), ∼50% of incoming Tregs were Il10 high pTregs, whereas this dropped to ∼20% in either OVA tolerance or allergy conditions (Fig. 7 D). Incoming Tregs in these settings instead trended for the enrichment of Helios+ (Ikzf2) cells (feeding vs. allergy P = 0.059), which were primarily either Gata3-high in allergy or Gata3-intermediate in tolerance (Fig. 7 D). Tomato+ Tregs also differentially expressed key functional molecules between conditions (Fig. 7 E). Compared with steady state, both tolerance and allergy conditions suppressed Ctla4 and upregu- lated Ikzf2. Tolerance Tomato+ Tregs downregulated Il10 and Gzmb relative to the other two conditions, whereas allergy promoted Gzmb. Allergy was additionally characterized by de- creased Cd83 and increased Socs2. Tregs that infiltrate the in- testine during active tolerance or allergy, therefore, bear distinct transcriptional profiles which may contribute to the respective functional outcomes. Altogether, these findings demonstrate how steady-state CD4+ T cell gut infiltration and tissue adaptation, imprinted by signals from both diet and the microbiota, is disrupted by an inflammatory challenge resulting in an effector CD4+ T cell re- sponse. Maintenance of immune tolerance to food protein dur- ing the inflammatory challenge is associated with an increased influx of Tregs, whereas in the OVA/CT allergy model, we ob- served an increased influx of proinflammatory Th17. In both tolerance and allergy, we found altered functional gene ex- pression programs in incoming Tregs, as well as altered tran- scriptional programs in resident IE CD4+ T cells. These data demonstrate that distinct intestinal immune responses underly steady-state exposure to food versus active tolerance to food upon inflammatory challenge. Clonal dynamics and antigen-specificity of tolerogenic and inflammatory CD4+ T cell responses to food To gain insight into the polyclonal repertoire dynamics of CD4+ T cell responses to food protein in the context of feeding, tol- erance, or allergy, we next assessed TCRs from our scRNAseq dataset. Under steady-state conditions, we found a high degree of clonal expansion among most clusters, although Tregs and FR4 high cells were more clonally diverse (Fig. 8 A). The TCR repertoires of Tomato− cells were equally diverse across con- ditions, whereas Tomato+ cells trended toward increased clonal expansion in OVA tolerance or allergy (Fig. 8, A and B). Finally, we found increased clonal expansion of Tregs in the tolerance group and a trend toward the increased expansion of Th17 cells in the allergy group (Fig. 8 C), suggesting that these subsets are simultaneously recruited and expanded in each respective condition. Finally, we assessed whether TCRs identified in our scRNA- seq datasets were specific for dietary protein. We selected 31 TCRs from GF mice fed AA + OVA diet (i.e., where OVA is the only source of foreign antigen) or from Tomato+ cells in OVA feeding or tolerance conditions (Table S12) and expressed them in NFAT-GFP hybridomas (Ise et al., 2010). Two TCRs responded to OVA in the NFAT-GFP assay—one from OVA feeding (TCR 18) and one from OVA tolerance (TCR 316; Fig. 8 D). We further tested these two TCRs for reactivity in an overlapping OVA peptide library and found that they both bind within the epitope OVA 26:40 (Fig. S5 H), which is within the protein’s non-cleaved signal peptide. By contrast, the OT-II TCR, which was generated in response to OVA through immunization, binds OVA 323:339 (Fig. S5 H). This may suggest that T cell responses generated upon oral exposure to food proteins preferentially use different epitopes than immunizing responses. TCR 18, which was derived from a Tomato+ OVA feeding mouse, was highly expanded and found primarily within IE2, IE3, and Th1 clusters (Fig. 8 D). TCR 316, derived from Tomato+ OVA tolerance, was moderately expanded and found only within Treg and cycling clusters (Fig. 8 D). Further, these Tregs were primarily from the Il10+ subcluster, suggesting that although infiltration of Il10+ Tregs decreased in tolerance compared with steady state, these cells still have an antigen-specific functional role in maintaining oral tolerance. These findings demonstrate that in a fully polyclonal system, food-antigen-specific T cell Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 11 of 18 Figure 8. Clonal dynamics and antigen- specificity of tolerogenic and inflammatory CD4+ T cell responses to food. (A–C) scTCRseq of 11,217 Tomato+ and Tomato− CD4+ T cells from the IE and LP of mice on day 26 of OVA feeding, tolerance, or allergy protocols using four to five mice per condition pooled across two in- dependent experiments. (A) Clonal expansion size (by TCR nucleotide sequence) plotted by gene expression cluster within all OVA feeding cells (top), and by mouse within all Tomato+ cells (bottom). (B) D50 in which repertoires are scored from 0 (least diverse) to 0.5 (completely diverse) within Tomato− or Tomato+ cells from each mouse. (C) Clonal expansion size (by TCR nucle- otide sequence) among Tomato+ Tregs (above) or Th17 (below) with corresponding Shannon di- versity scores to the right. (D) NFAT-GFP assay to determine TCR recognition of OVA relative to a-CD3 (positive control) or unloaded DCs (nega- tive control). Heatmap indicates percent NFAT- GFP expression out of TCR + NFAT hybridoma cells. Mouse experimental group and scRNAseq cluster of cells from which each TCR was iden- tified are indicated to the right. (B and C) One- way ANOVA with Tukey’s multiple comparison test, P < 0.05, *P < 0.01. responses are generated not only in active oral tolerance but also in the course of natural feeding. Discussion Here, we provide a comprehensive analysis of intestinal CD4+ T cell clonal dynamics and functional differentiation in response to food. We found that at steady state, signals from food promote clonal selection, epithelial adaptation, and cytotoxic program- ming of intestinal CD4+ T cells, a pathway which is further boosted by signals from the microbiota. The requirement for Tregs in controlling inflammatory responses to food is well es- tablished (Mucida et al., 2005; Torgerson et al., 2007) and an LP Treg response to dietary protein has been described using TCR monoclonal models (Hadis et al., 2011), MHC class II tetramers (Hong et al., 2022), and antigen-free dietary models (Kim et al., 2016). Our data further demonstrate that diet-induced Tregs in the gut upregulate genes associated with cytotoxicity and epi- thelial residency at steady state. Previously, we reported that IE Tregs can differentiate into CD4+ CD8αα+ T cells in a microbiota- dependent manner (Sujino et al., 2016), and that Tregs expressing IE signature genes are likely on a trajectory towards this fate (Bilate et al., 2020). These data suggest that dietary signals in the intestine promote a cytotoxic IE-adapted phe- notype not only in conventional CD4+ T cells but also in Tregs. While steady-state dietary signals promoted Granzyme B expression in Tregs, we showed that incoming Tregs in the al- lergy model further upregulated Granzyme B, whereas tolerant mice suppressed it, indicating that regulation of this pathway for Treg-mediated immune control. may be important Granzyme-dependent cytolysis of activated effector cells has been described as one mechanism of Treg suppressive function (Grossman et al., 2004), suggesting that cytotoxicity can con- tribute to both pro- and anti-inflammatory responses. CD4+ T cells with cytotoxic properties have also been described in the setting of chronic immune stimulation from viral infection, autoimmunity, or cancer (Cenerenti et al., 2022). Our work suggests that continuous homeostatic exposure to food and commensals drives a similar CD4+ T cell phenotype in the gut. One essential difference between cytotoxic CD4+ T cells de- scribed in disease states and cytotoxic CD4+ T cells induced by steady-state intestinal stimulation is the expression of CD8αα, Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 12 of 18 which has been proposed to raise the threshold for TCR acti- vation (Cheroutre and Lambolez, 2008; Mucida et al., 2013). Therefore, although IE-adapted CD4+ CD8αα+ T cells are equip- ped with cytotoxic machinery, they may require stimulation above steady-state levels to activate cytotoxic mechanisms. Our data show that exposure to food protein in tandem with an in- flammatory signal (cholera toxin) drives a hyperactive state in tissue-resident IE-adapted CD4+ T cells, including upregulation of cytotoxic genes. In Celiac disease, IE-adapted CD4+ T cells exhibit highly inflammatory Th1-like responses and contribute to tissue pathology, which is abrogated upon the removal of gluten from the diet (Abadie et al., 2012; Costes et al., 2019; Fina et al., 2008). Thus, dysregulation or dysfunction of IE CD4+ T cells in addition to Tregs may be one mechanism by which inflammatory responses to harmless gut antigens emerge. How- ever, the current functional understanding of cytotoxic CD4+ T cells is limited. For example, the extent to which CD4+ T cell cytotoxicity depends on cognate antigen presentation via MHC II, as well as the precise cellular targets of CD4-mediated cytolysis, remains unclear (Takeuchi and Saito, 2017). Further work is needed to define whether and how CD4+ T cell cytotoxicity con- tributes to tolerance in the gut or conversely can become dysre- gulated and contribute to tissue damage. We show that inflammatory signals disrupt steady-state CD4+ T cell intestinal adaptation, leading to an increased influx of cells skewed toward effector phenotypes. If oral tolerance to a food protein was established prior to inflammatory challenge, then exposure to the same food protein during inflammatory chal- lenge results in clonal expansion of gut Tregs associated with reduced proinflammatory CD4+ T cell gene expression and protection against food allergy. However, if the primary expo- sure to a food protein occurs at the same time as the in- flammatory stimulus, impaired Treg response and increased pro-inflammatory CD4+ T cell gene expression are observed in the gut, associated with the development of food allergy. Al- though pTregs are known to be required for oral tolerance (Garside et al., 1995; Hadis et al., 2011; Mucida et al., 2005), we found that Treg influx during oral tolerance was characterized by increased expression of Helios, a marker typically associated with thymic Tregs which do not respond to intestinal antigen, and reduced Il10. Nevertheless, OVA-specific Tregs identified in the tolerance condition were from the Il10+ cluster, demon- strating that Il10+ pTregs contribute to antigen-specific tolerance in a polyclonal setting whereas the increased Helios + Tregs may contribute broadly to tissue protection in an antigen-nonspecific manner. While human allergy is typically associated with a Th2 response toward allergens, we and others find the intestinal CD4+ T cell response in the OVA/CT mouse model to be skewed toward proinflammatory Th17, an effect that has been shown to depend on the microbiota (Zhao et al., 2017). Whether Th17 are effector cells contributing to allergy in C57BL/6 mice or are merely a parallel response to microbiota in the presence of mucosal adjuvant remains unclear. Our iSellTomato data demonstrate that important repertoire dynamics accompany T cell differentiation in response to die- tary antigens, conclusions previously precluded by the wide- spread use of TCR monoclonal systems (Mucida et al., 2005; Weiner et al., 2011). Additionally, we provide evidence that steady-state exposure to dietary proteins contributes to clonal selection of intestinal CD4+ T cells and identify TCR clones from the tissue that recognize dietary protein. Our findings comple- ment a recent study that used MHC class II tetramers to show generation of a dietary antigen-specific CD4+ T cell response in a natural polyclonal setting, where the primary fate in the LP was Treg differentiation (Hong et al., 2022). Our approach does not rely on epitope-specific discovery, and the clones we identify do not bind the immunodominant OVA epitope previously identi- fied via immunization (Robertson et al., 2000), but rather both bind OVA 26:40. This may suggest distinct dynamics of clonal selection by dietary antigen in the natural oral route. The food- antigen-specific clone we identified from steady-state feeding was derived from IE-adapted or Th1 cells whereas the oral tol- erance clone was derived from Tregs. Thus, we show that while oral tolerance is characterized by an antigen-specific Treg re- sponse, the steady-state response can include differentiation to IE-adapted CD4+ CD8αα+ cells. Food protein can therefore drive diverse phenotypic outcomes in antigen-specific intestinal CD4+ T cells, including adaptation to the intestinal epithelium. Altogether, our findings suggest that the prevalent fate of food-responsive CD4+ T cells in the steady-state intestine is epithelial adaptation, whereas maintenance of oral tolerance in an inflammatory setting correlates with the increased influx and clonal expansion of Tregs. We and others have demonstrated an important regulatory activity of epithelium-adapted CD4+ T cells in the context of response to diet (Sujino et al., 2016), or in the context of colitis or infection (Basu et al., 2021; Reis et al., 2013). Further, dysregulation of IE CD4+ T cells may be one mechanism by which inflammatory responses to harmless gut antigens emerge. Therefore, highly regulated maintenance of epithelium- adapted CD4+ T cells in addition to Tregs may be critical for preventing inappropriate immune responses to food and sub- sequent disease. Materials and methods Animals Animal care and experimentation were consistent with the National Institutes of Health guidelines and were approved by the Institutional Animal Care and Use Committee at The Rockefeller University. All mice were maintained at The Rockefeller University animal facilities. Germ-free C57BL/6J mice were obtained from Sarkis Mazmanian (California Insti- tute of Technology, Pasadena, CA, USA) and bred and main- tained in germ-free isolators. SPF C57BL/6J mice were recolonized from GF with a single gavage of feces and bred and maintained in SPF conditions. Vertically colonized ex-GF off- spring were used for SPF experiments to control for genetic drift in our GF isolators. The Oligo-MM12 consortium was a gift from K. McCoy (University of Calgary, Calgary, Canada). We colonized GF C57BL/6J breeders with a single gavage of Oligo- MM12 and monitored colonization (including the presence of the entire consortium in successive generations) by specific ampli- fication of individual bacterial members by quantitative polymerase chain reaction (qPCR; see below). Oligo-MM12 mice Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 13 of 18 were subsequently bred and maintained in isolators, and vertically colonized offspring were used for all experiments. SellCre-ERT2 mice were provided by M. Nussenzweig (Merkenschlager et al., 2021), crossed with Rosa26CAG-LSL-tdTomato-WPRE (007914) mice from Jackson Laboratory, and maintained under SPF con- ditions. CD45.1 OT-II TCR-transgenic mice were originally purchased from Taconic Farms and maintained in our facili- ties. Mice were used at 8 wk of age for most experiments, except iSellTomato where experimental protocols were initi- ated at 7 wk of age and endpoint analysis was performed at 11 wk. Both male and female mice were used for all experiments, except scRNAseq, which used exclusively female (for diet) or male (for iSellTomato) mice to avoid sex effects on gene expression. Germ-free and Oligo-MM12 monitoring Germ-free status was confirmed by qPCR analysis using uni- versal 16S rRNA primers (fwd: 59-ACTCCTACGGGAGGCAGC AGT-39; rev: 59-ATTACCGCGGCTGCTGGC-39). Colonization of mice by the Oligo-MM12 consortium was confirmed and moni- tored over generations by qPCR by using primer pairs specific to each species as previously described (Nowosad et al., 2020). DNA was extracted from fecal samples using the ZR Fecal DNA kit (Zymo Research) according to the manufacturer’s in- structions. Quantitative PCR was performed with the Power SYBR Green master mix (Applied Biosystems). The average cycle threshold (Ct) value of two technical replicates was used to quantify the relative abundance of each species’ 16S ribo- somal RNA using the ΔΔCt method, with the universal 16S rRNA primers serving as controls between samples. Relative abundance was corrected according to the genome copy num- ber of 16S rRNA for each species. Experimental diets For all dietary experiments, breeders were maintained on a standard chow diet, and breeding cages were switched to an amino acid diet when pups were 1-wk-old to prevent early ex- posure to food proteins. Mice were subsequently weaned onto experimental diets at 3 wk and maintained on that diet until endpoint analysis at 8 wk old unless otherwise indicated. Pro- tein-antigen-free solid diet containing free amino acids (Modi- fied TestDiet 9GCV with 5% cellulose; composition details are in Table S1) was irradiated at >45 kGy to ensure sterility for germ- free conditions. GF, Oligo-MM12, and ex-GF SPF chow diet control mice were fed autoclaved standard chow diets fortified with extra nutrients to compensate for losses during autoclaving (5K54; LabDiet). For AA + OVA, Ovalbumin (OVA) grade III (A5378; Sigma-Aldrich) was provided at 1 mg/ml in drinking water and autoclaved for sterility. Casein and Casein–gluten–soy diets were modified from the AA diet to contain 50% less amino acid and 50% even mix of casein or casein, gluten, and soy protein (9GU1, 9GU2; TestDiet). Serum biochemistry Serum was collected from 8-wk-old mice fed AA or chow diet since weaning after 2 h of fasting. Serum analysis was per- formed by IDEXX (USA) using standard protocols. Histology Representative portions of duodenum, ileum, and colon were fixed in 4% paraformaldehyde and embedded in paraffin ac- cording to standard protocols. 5-µm sections were mounted on glass slides and stained with hematoxylin and eosin (H&E). Images were acquired on a Keyence BZ-X800 inverted micro- scope using a 10 × 0.3/14.50 mm objective lens with brightfield illumination (Keyence). Blinded quantitative evaluation of intestinal pathology was performed according to established methods (Erben et al., 2014). Briefly, each tissue section was microscopically assessed for the extent of inflammatory cell infiltrate (0–7) and changes to the epithelial (0–23) or mucosal (0–15) architecture. The sum of these scores represents the combined pathological score reported for each tissue, with 40 being the maximum score and a score <10 indicating normal tissue with minimal to mild inflammation. Further quantifi- cation of tissue metrics was performed in ImageJ. Isolation of intestinal T cells Intraepithelial and LP lymphocytes were isolated as previously described (Bilate et al., 2016; Reis et al., 2013). Briefly, small intestines or large intestines (cecum and colon) were harvested and washed in PBS and 1 mM dithiothreitol (DTT), followed by 30 mM EDTA. Intraepithelial cells were recovered from the supernatant of DTT and EDTA washes and mononuclear cells were isolated by gradient centrifugation using Percoll. LP lym- phocytes were obtained after collagenase digestion of the tissue. Antibodies and flow cytometry analysis Fluorescent dye-conjugated antibodies were purchased from BD Biosciences, BioLegend, Ebioscience (Thermo Fisher Scientific), or R&D Biosciences. The following clones were used: anti-CD4 RM4-5; anti-CD8α 53–6.7; anti-CD8β YTS 156.7.7; anti-CD11b M1/ 70; anti-CD11c N418; anti-CD44 IM7; anti-CD45 30-F11; anti- CD45.1 A20; anti-CD62L MEL-14, G8.8; anti-CD69 H1.2F3; anti- CD103 2E7; anti-CD117 (cKit) 2B8; anti-F480 BM8; anti-FceR1 MAR-1; anti-Foxp3 FJK-16 s; anti-Granzyme B GB11; anti-IL-4 11B11, anti-IL-17A TC11-18H10; anti-Ly6C AL-21; anti-Ly6G RB6- 8C5; anti-Nrp1 BAF566; anti-Rorγt Q31-378; anti-Siglec F E50- 2440; anti-TCRβ H57-597; anti-TCRγδ GL3; and anti-TCR Vα2 B20.1. Live/dead fixable dye Aqua (Thermo Fisher Scientific) was used according to manufacturer’s instructions. Intranuclear staining of Foxp3 and intracellular staining of Granzyme B was conducted using Foxp3 Mouse Regulatory T Cell Staining Kit according to kit instructions (eBioscience). For analysis of IL-4 and IL17-A production, cells were incubated at 37°C with 100 ng/ ml phorbol 12-myristate 13-acetate (PMA, Sigma-Aldrich) 200 ng/ml ionomycin (Sigma-Aldrich), and Golgi stop solution containing Monensin (2 mM, BD Biosciences) for 4 h. Intra- cellular staining for cytokines was conducted in Perm/Wash buffer after fixation and permeabilization in Fix/Perm buffer (BD Biosciences) according to kit instructions. Flow cytometry data weres acquired on an LSR-II or Symphony flow cytometer (Becton Dickinson) and analyzed using FlowJo software pack- age (Tri-Star). For cell sorting experiments, lymphocytes were sorted on a FACS Aria II instrument as indicated in the figure legends. AccuCheck Counting Beads (Thermo Fisher Scientific) Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 14 of 18 live, were used for counting absolute cell numbers. For flow cyto- metric analysis, the following gating strategies were used to lymphocytes identify cell populations. T cells: single, (based on FSC, SSC, and live/dead fixable dye Aqua stain), CD45+, TCRgδ+ (for gδ T cells), or TCRβ+ (for all other T cells); CD4+ T cells: CD4+, CD8β−. CD8αβ+ T cells: CD4−, CD8β+. CD8αα+ T cells: CD4−, CD8β−, CD8α+. Myeloid cells: single, live, CD45+; eosinophils: Siglec F+, CD11b+; Mast cells: cKit+, FCer1+; mac- rophages: CD11b+, CD11c+, F480+; monocytes: CD11b+ CD11c− Ly6c+; neutrophils: Ly6G+ CD11b+. OTII transfer experiment Na¨ıve CD4 T cells from the spleen and lymph nodes of CD45.1 OTII TCR transgenic mice were isolated by negative selection using biotinylated antibodies against CD8α, CD25, CD11c, CD11b, TER-119, NK1.1, and B220 and anti-biotin MACS beads (Miltenyi Biotec). 1 × 106 cells were transferred by retro-orbital injection to CD45.2 hosts under isoflurane gas anesthesia. Host mice were provided regular drinking water or drinking water supplemented with 1 mg/ml OVA grade III (A5378; Sigma-Aldrich). After 48 h, mesenteric lymph nodes were collected, and CD45.1+ TCRVα2+ CD4+ OTII cells were analyzed for activation by expression of CD69. Tamoxifen treatment Tamoxifen (Sigma-Aldrich) was dissolved in corn oil (Sigma- Aldrich) and 10% ethanol, shaking at 37°C for 30 min–1 h. Two doses of Tamoxifen (5 mg/dose) were administered to mice via oral gavage at 50 mg/ml, 3 d and 1 d before start of treatment protocol. OVA/cholera toxin allergy model OVA grade III (A5378; Sigma-Aldrich) was provided at 0.1% in drinking water, autoclaved for sterility for 3 d (Feeding, Toler- ance) or not (Allergy) to initialize tolerance. All mice were then provided with regular drinking water for 1 wk. 1 mg OVA in 0.2 M sodium bicarbonate (Feeding) or 1 mg OVA + 20 µg cholera toxin (100B; List Biological) in 0.2 M sodium bicarbonate (Tol- erance, Allergy) were provided once per week for 3 wk, followed by endpoint analysis 2 d after the final dose. Serum was har- vested for ELISA 1 d prior to endpoint analysis. For anaphylaxis experiments, the above protocol was followed except for the following modifications: OVA or OVA/CT was provided once per week for 4 wk, followed by a challenge 7 d after the final dose. Implantable electronic temperature probes (Avidity IPTT-300) were injected s.c. 1 d prior to the challenge. Mice were challenged with 5 mg OVA i.p., and body temperature was measured every 10 min for 50 min. Mice used in anaphylaxis experiments were not used for downstream sequencing analysis. ELISA Lipocalin-2 was analyzed by using Lcn-2 ELISA kit (R&D Bio- sciences) as described by Chassaing et al. (2012). IgE and OVA specific IgG1 ELISAs were performed as described previously (Atarashi et al., 2011). 16S rRNA sequencing Intestinal contents were collected fresh from the whole small intestine or cecum at endpoint analysis. Littermate controls were used for all 16S experiments to control for maternal effects. Samples were collected from at least four different cages per experimental group to control for cage effect. Samples were prepared for 16S rRNA sequencing following the 16S Illumina Amplicon protocol from the Earth Microbiome Project (Caporaso et al., 2011). Libraries were sequenced using Miseq 2 × 150 using a 15% PhiX spike. Sequence processing and analysis were per- formed in R. Briefly, read assembly into amplicon sequence variants (ASVs) and taxonomic assignment on the Silva database were performed using Dada2 (v.1.2.6). Taxa not seen >10 times in at least 20% of samples were removed from the analysis. ASV quantification and analyses Tawere performed using Phyloseq (v1.42; McMurdie and Holmes, 2013). scRNAseq library preparation Lymphocytes isolated from the small intestine epithelium or LP were isolated as described above and indexed with TotalSeqC Hashtag (BioLegend) cell surface antibodies, with two barcodes used per sample for deeper multiplexing. Total CD4+ T cells were sorted, pooled, and immediately loaded onto a Chromium Con- troller (10× Genomics). For SellTomato sorted Tomato+ and To- mato− (CD62L−, CD4+ T) cells were pooled at a ∼1:3 ratio. 59 Gene expression, VDJ, and Feature Barcode libraries were prepared using the Chromium Single Cell 59 v2 Reagent Kit (10× Ge- nomics) according to the manufacturer’s protocol at the Ge- nomics core of The Rockefeller University. Libraries were sequenced on Illumina NextSeq500 or NovaSeq 6000. Hashtag indexing was used to demultiplex the sequencing data and generate gene-barcode matrices. Data processing of scRNAseq and single-cell TCRseq libraries Raw .FASTQ files from our 10X libraries were processed with Cellranger count (v6.2.0) using the 10X Genomics prebuilt mouse reference (v3.0.0 mm10; for diets) or a customized mouse genome (mm10) that included the Ai9Tomato sequence plasmid as an artificial chromosome with the Ai9Tomato and STOP se- quences annotated as features (for SellTomato). Analyses were performed in R 4.2.2. Quality control was performed by re- moving cells with high (>10% for the antigen-free diet libraries, >5% for SellTomato) mitochondrial unique molecular identifier (UMI) content. Cells not expressing Trac or Cd4 were excluded from our analysis. We defined Tomato+ and Tomato– cells post- sequencing using normalized UMI counts for tdTomato and Stop (Tomato+: tdTomato > Stop; Tomato–: tdTomato < Stop), discard- ing ambiguous cells where tdTomato = Stop. The matrix of UMI counts was normalized by applying a regression model with a negative binomial distribution, available through the SCTrans- form function in the Seurat (v41.-4.3.) package (Hafemeister and Satija, 2019). The top 3,000 variable genes were first used for dimensional reduction by PCA using the scaled data. The first 30 principal components were further used for visualization using the Manifold Approximation and Projection (UMAP) and cell clustering (Hafemeister and Satija, 2019; Stuart et al., 2019). TCR contigs and annotation were performed with the Cellranger vdj workflow from 10X Genomics and the prebuild mouse reference (v3.1.0 mm10). Contigs filtering, clonotype calling, and down- stream TCR analysis were performed using scRepertoire (v1.5.2; Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 15 of 18 Borcherding et al., 2020). Further processing, statistical analy- sis, and visualizations were performed using ggplot2 (v3.4.1; Wickham, 2016) and rstatix (v0.7.2). Signature scoring IE signature scores were calculated using UCell (v2.2; Andreatta and Carmona, 2021) using the following genes as input: Nkg7, Ccl5, Cd160, Itgae, Gzma, Gzmb, Cd7, Prf1, Lag3, and Cd8a. was changed at 24 h and harvested at 48 h for viral transduction. NFAT-GFP hybridoma cells were plated 1 × 106/ml in viral su- pernatant with 5 µg/ml polybrene (TR-1003-G; Sigma-Aldrich) in six-well plates precoated with 15 µg retronectin (T100A; Ta- kara). Plates were spinfected by centrifugation at 2,500 rpm 32°C for 90 min. Cells were then cultured for 2–7 d in IMDM + 10% FBS + Pen/Strep/L-glutamine + 50 µM β-me + 1 mM sodium pyruvate before FACS selection for TCR+ CD3+ cells. 3D volcano 3D volcano plots were generated using volcano3D (v1.2–2.0.8; Lewis et al., 2019). Briefly, genes differentially expressed across three groups were identified by FDR-adjusted Kruskal–Wallis test (P-adj < 0.05) considering only genes expressed in at least 10% of cells in any group. Upregulation in each group was fur- ther determined by FDR-adjusted Wilcoxon rank sum test (P-adj <0.05, Log2FC > 0.5). Each point on the plot represents a gene colored by the group(s) in which they are significantly upre- gulated and uncolored for nonsignificant genes. Distance from the origin for each gene represents a z-score calculated based on −log10 P values from Kruskal-Wallis comparisons between all three groups, with genes further from the origin having higher significance. The degree of upregulation of a gene in a given condition is indicated by its angle on the plot relative to condition-labeled axes. Scaled gene expression data was used for all calculations except fold change. Select genes of interest are labeled on each plot. Circos plots TCR sharing (clonal overlap) was visualized using Circos to create circular plots aesthetics (Krzywinski et al., 2009). Each segment denotes a mouse as indicated in figures, with bands representing clonal sharing based on paired TCRα and TCRβ amino acid sequences. NKT cells and cells without paired TCRα and TCRβ sequenced were not considered in the analysis. Statistical analyses Statistical analysis was carried out using GraphPad Prism v.9. Flow cytometry analysis was carried out using FlowJo software. Data in the graphs show mean ± SEM, and P values < 0.05 were considered significant. Repertoire diversity was analyzed by Diversity 50 (D50) calculated in R as the fraction of dominant clones that account for the cumulative 50% of the total paired CDR3s. GraphPadPrism v.9 was used for graphs and Adobe Il- lustrator 2021 was used to assemble and edit figures. TCR hybridoma generation Select TCRs from the scRNAseq datasets were synthesized (TCRα and TCRβ linked by P2A) and cloned into pMSCV-mCD4 vectors (Twist Bioscience). Phoenix-Eco cells (CRL-3214; Addg- ene) were used for retrovirus production. Phoenix were grown to 60–80% confluence in DMEM in 10-cm culture dishes pre- coated 2 ml 0.01% poly-L-lysine (A-005-M; Sigma-Aldrich) in 8 ml 0.1% gelatine (1040700500; Sigma-Aldrich) and then transfected with a mix containing 5 µg pMSCV-mCD4-TCR plasmid, 2 µg pEco (12371; Addgene), 72 µl 1 mg/ml poly- ethylenamine (23966; Polysciences), and 600 µl DMEM. Media TCR testing for OVA specificity Dendritic cells were collected from mouse spleens (130-092-465; Miltenyi Biotec) and co-cultured 5 × 105/ml with 50 µg OVA grade III (A5378; Sigma-Aldrich) for 4 h at 37 C in 96-well U-bottom plates to load with antigen or left unloaded for nega- tive control. 2.5 × 104 TCR-expressing NFAT-GFP cells were added to each well and incubated overnight. For positive control, 0.2 µl anti-CD3 (553057; BD) was added to a well with NFAT and unloaded DCs. NFAT-GFP response to TCR stimulation was measured by flow cytometry. To test epitope specificity, crude 15-aa peptides covering the length of OVA with a 5-aa shift were synthesized (LifeTein) and used in the assay as above at 1 µM per well. TCR stimulation was measured by IL-2 ELISA on culture supernatant (555148; BD). Figure preparation Figures were made in Prism or R and compiled and formatted in Adobe Illustrator. The graphical abstract was made using Bio- Render (graphical license LO25BBHSCS). Online supplemenal material Fig. S1 shows validation of the health of AA-diet mice including body weight, fecal lipocalin, serum nutritional biomarkers, in- testine length, intestinal histology, and intestinal myeloid cell quantification. Fig. S2 shows additional analysis of intestinal CD4+ T cell accumulation in response to dietary protein and 16S rRNA sequencing supporting Figs. 1 and 2; and Fig. S3 shows additional scRNAseq analysis of GF and OligoMM12 mice fed AA, AA + OVA, or chow diet supporting Figs. 2 and 3; and Fig. S4, A–D shows additional histological and flow cytometry charac- terization of OVA feeding, tolerance, and allergy models in iSellTomato mice. Fig. S4, E–H shows supporting scRNAseq analysis of iSellTomato mice. Fig. S5, A–G shows additional scRNAseq analysis of iSellTomato mice in OVA feeding, tolerance, and allergy. Fig. S5 H shows OVA overlapping peptide reactivity testing for candidate TCRs. Table S1 shows the composition of the AA diet. Table S2 shows 16S rRNA sequencing results for SPF mice fed AA or chow diet. Table S3 shows differentially ex- pressed genes across total CD4+ T cell UMAP clusters from GF and OligoMM12 mice fed AA, AA + OVA, or chow diet. Table S4 shows results from three-way differential gene expression test- ing between IE CD4+ T cells from AA, AA + OVA, or chow diet mice. Table S5 shows differentially expressed genes across Treg UMAP clusters from GF and OligoMM12 mice fed AA, AA + OVA, or chow diet. Table S6 shows results from three-way differential gene expression testing between Tregs from AA, AA + OVA, or chow diet mice. Table S7 shows differentially expressed genes across total CD4+ T cell UMAP clusters from iSellTomato mice in Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 16 of 18 OVA feeding, tolerance, and allergy. Tables S8 shows results from three-way differential gene expression testing between Tomato− IE CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S9 shows results from three-way differential gene expression testing between Tomato+ IE CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and al- lergy. Table S10 shows results from three-way differential gene expression testing between Tomato+ LP CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S11 shows differentially expressed genes across Treg UMAP clusters from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S12 shows supporting information about TCRs identified in scRNAseq datasets tested for OVA specificity. Data availability scRNAseq data of intestinal CD4+ T cells from GF or Oligo-MM12 mice fed AA, AA + OVA, or chow diet, and SPF iSellTomato mice in the OVA feeding, tolerance, or allergy protocols are publicly available under Gene Expression Omnibus accession number GSE231351. Other data are available in the published article and online supplemental material. Acknowledgments We thank all Mucida Lab members and Rockefeller University employees for their continuous assistance, particularly A. Rogoz and S. Gonzalez for the maintenance of mice, RU Genomics core for sequencing, Tri-I Laboratory of Comparative Pathology for histology preparation, K. Gordon, K. Chhosphel, and J.P. Truman for sorting, B. Reis for assistance with figures, and A. Bilate, G. Donaldson, and R. Parsa for critical reading of the manuscript. We also thank the Victora and Lafaille labs for fruitful discussions. This work was supported by The Howard Hughes Medical Institute, grants R01DK093674, R01DK113375, and R21AI144827, and Food Allergy FARE/FASI Consortium. Author contributions: A. Lockhart initiated, designed, per- formed, and analyzed experiments and wrote the manuscript. A. Reed, C. Herman, and M.C. Campos Canesso designed and per- formed experiments. T. Rezende de Castro performed analysis. D. Mucida conceived, initiated, designed, and supervised the research, and wrote the manuscript. All authors revised and edited the manuscript and figures. Disclosures: The authors declare no competing interests exist. Submitted: 22 October 2022 Revised: 11 April 2023 Accepted: 3 May 2023 References Abadie, V., V. Discepolo, and B. Jabri. 2012. Intraepithelial lymphocytes in celiac disease immunopathology. Semin. Immunopathol. 34:551–566. https://doi.org/10.1007/s00281-012-0316-x Abadie, V., S.M. Kim, T. Lejeune, B.A. Palanski, J.D. Ernest, O. Tastet, J. Voisine, V. Discepolo, E.V. Marietta, M.B.F. Hawash, et al. 2020. IL-15, gluten and HLA-DQ8 drive tissue destruction in coeliac disease. Nature. 578:600–604. https://doi.org/10.1038/s41586-020-2003-8 Andreatta, M., and S.J. Carmona. 2021. UCell: Robust and scalable single-cell gene signature scoring. Comput. Struct. Biotechnol. J. 19:3796–3798. https://doi.org/10.1016/j.csbj.2021.06.043 Atarashi, K., T. Tanoue, T. Shima, A. Imaoka, T. Kuwahara, Y. Momose, G. Cheng, S. Yamasaki, T. Saito, Y. Ohba, et al. 2011. Induction of colonic regulatory T cells by indigenous Clostridium species. Science. 331: 337–341. https://doi.org/10.1126/science.1198469 Basu, J., B.S. Reis, S. Peri, J. Zha, X. Hua, L. Ge, K. Ferchen, E. Nicolas, P. Czyzewicz, K.Q. Cai, et al. 2021. Essential role of a ThPOK autor- egulatory loop in the maintenance of mature CD4+ T cell identity and function. Nat. Immunol. 22:969–982. https://doi.org/10.1038/s41590 -021-00980-8 Bilate, A.M., D. Bousbaine, L. Mesin, M. Agudelo, J. Leube, A. Kratzert, S.K. Dougan, G.D. Victora, and H.L. Ploegh. 2016. Tissue-specific emergence of regulatory and intraepithelial T cells from a clonal T cell precursor. Sci. Immunol. 1:eaaf7471. https://doi.org/10.1126/sciimmunol.aaf7471 Bilate, A.M., M. London, T.B.R. Castro, L. Mesin, J. Bortolatto, S. Kongthong, A. Harnagel, G.D. Victora, and D. Mucida. 2020. T cell receptor is re- quired for differentiation, but not maintenance, of intestinal CD4+ in- traepithelial lymphocytes. Immunity. 53:1001–1014.e20. https://doi.org/ 10.1016/j.immuni.2020.09.003 Borcherding, N., N.L. Bormann, and G. Kraus. 2020. scRepertoire: An R-based toolkit for single-cell immune receptor analysis. F1000 Res. 9:47. https:// doi.org/10.12688/f1000research.22139.1 Bousbaine, D., L.I. Fisch, M. London, P. Bhagchandani, T.B. Rezende de Cas- tro, M. Mimee, S. Olesen, B.S. Reis, D. VanInsberghe, J. Bortolatto, et al. 2022. A conserved Bacteroidetes antigen induces anti-inflammatory intestinal T lymphocytes. Science. 377:660–666. https://doi.org/10 .1126/science.abg5645 Bouziat, R., R. Hinterleitner, J.J. Brown, J.E. Stencel-Baerenwald, M. Ikizler, T. Mayassi, M. Meisel, S.M. Kim, V. Discepolo, A.J. Pruijssers, et al. 2017. Reovirus infection triggers inflammatory responses to dietary antigens and development of celiac disease. Science. 356:44–50. https://doi.org/ 10.1126/science.aah5298 Brugiroux, S., M. Beutler, C. Pfann, D. Garzetti, H.J. Ruscheweyh, D. Ring, M. Diehl, S. Herp, Y. L¨otscher, S. Hussain, et al. 2016. Genome-guided design of a defined mouse microbiota that confers colonization resis- tance against Salmonella enterica serovar Typhimurium. Nat. Microbiol. 2:16215. https://doi.org/10.1038/nmicrobiol.2016.215 Caporaso, J.G., C.L. Lauber, W.A. Walters, D. Berg-Lyons, C.A. Lozupone, P.J. Turnbaugh, N. Fierer, and R. Knight. 2011. Global patterns of 16S rRNA diversity at a depth of millions of sequences per sample. Proc. Natl. Acad. Sci. USA. 108:4516–4522. https://doi.org/10.1073/pnas.1000080107 Cenerenti, M., M. Saillard, P. Romero, and C. Jandus. 2022. The era of cyto- toxic CD4 T cells. Front. Immunol. 13:867189. https://doi.org/10.3389/ fimmu.2022.867189 Chassaing, B., G. Srinivasan, M.A. Delgado, A.N. Young, A.T. Gewirtz, and M. Vijay-Kumar. 2012. Fecal lipocalin 2, a sensitive and broadly dynamic inflammation. PLoS One. 7: non-invasive biomarker for intestinal e44328. https://doi.org/10.1371/journal.pone.0044328 Cheroutre, H., and F. Lambolez. 2008. Doubting the TCR coreceptor function of CD8αα. Immunity. 28:149–159. https://doi.org/10.1016/j.immuni.2008 .01.005 Chou, C., X. Zhang, C. Krishna, B.G. Nixon, S. Dadi, K.J. Capistrano, E.R. Kansler, M. Steele, J. Han, A. Shyu, et al. 2022. Programme of self- reactive innate-like T cell-mediated cancer immunity. Nature. 605: 139–145. https://doi.org/10.1038/s41586-022-04632-1 Costes, L.M.M., D.J. Lindenbergh-Kortleve, L.A. van Berkel, S. Veenbergen, H.R.C. Raatgeep, Y. Simons-Oosterhuis, D.H. van Haaften, J.J. Karrich, J.C. Escher, M. Groeneweg, et al. 2019. IL-10 signaling prevents gluten- dependent intraepithelial CD4+ cytotoxic T lymphocyte infiltration and epithelial damage in the small intestine. Mucosal Immunol. 12:479–490. https://doi.org/10.1038/s41385-018-0118-0 Denning, T.L., S.W. Granger, D. Mucida, R. Graddy, G. Leclercq, W. Zhang, K. Honey, J.P. Rasmussen, H. Cheroutre, A.Y. Rudensky, and M. Kronen- berg. 2007. Mouse TCRαβ+CD8αα intraepithelial lymphocytes express genes that down-regulate their antigen reactivity and suppress immune J. Immunol. 178:4230–4239. https://doi.org/10.4049/ responses. jimmunol.178.7.4230 Erben, U., C. Loddenkemper, K. Doerfel, S. Spieckermann, D. Haller, M.M. Heimesaat, M. Zeitz, B. Siegmund, and A.A. Kühl. 2014. A guide to histomorphological evaluation of intestinal inflammation in mouse models. Int. J. Clin. Exp. Pathol. 7:4557–4576. Fina, D., M. Sarra, R. Caruso, G. Del Vecchio Blanco, F. Pallone, T.T. Mac- Donald, and G. Monteleone. 2008. Interleukin 21 contributes to the Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 17 of 18 mucosal T helper cell type 1 response in coeliac disease. Gut. 57:887–892. https://doi.org/10.1136/gut.2007.129882 Garside, P., M. Steel, F.Y. Liew, and A.M. Mowat. 1995. CD4+ but not CD8+ T cells are required for the induction of oral tolerance. Int. Immunol. 7: 501–504. https://doi.org/10.1093/intimm/7.3.501 Grossman, W.J., J.W. Verbsky, W. Barchet, M. Colonna, J.P. Atkinson, and T.J. Ley. 2004. Human T regulatory cells can use the perforin pathway to cause autologous target cell death. Immunity. 21:589–601. https://doi .org/10.1016/j.immuni.2004.09.002 Hadis, U., B. Wahl, O. Schulz, M. Hardtke-Wolenski, A. Schippers, N. Wagner, W. Müller, T. Sparwasser, R. F¨orster, and O. Pabst. 2011. Intestinal tolerance requires gut homing and expansion of FoxP3+ regulatory T cells in the lamina propria. Immunity. 34:237–246. https://doi.org/10 .1016/j.immuni.2011.01.016 Hafemeister, C., and R. Satija. 2019. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial re- gression. Genome Biol. 20:296. https://doi.org/10.1186/s13059-019-1874 -1 Hong, S.W., P.D. Krueger, K.C. Osum, T. Dileepan, A. Herman, D.L. Mueller, and M.K. Jenkins. 2022. Immune tolerance of food is mediated by layers of CD4+ T cell dysfunction. Nature. 607:762–768. https://doi.org/10 .1038/s41586-022-04916-6 Ise, W., M. Kohyama, K.M. Nutsch, H.M. Lee, A. Suri, E.R. Unanue, T.L. Murphy, and K.M. Murphy. 2010. CTLA-4 suppresses the pathogenicity of self antigen-specific T cells by cell-intrinsic and cell-extrinsic mechanisms. Nat. Immunol. 11:129–135. https://doi.org/10.1038/ni.1835 Jimenez-Saiz, R., D.K. Chu, T.S. Mandur, T.D. Walker, M.E. Gordon, R. Chaudhary, J. Koenig, S. Saliba, H.J. Galipeau, A. Utley, et al. 2017. Lifelong memory responses perpetuate humoral TH2 immunity and anaphylaxis in food allergy. JAllergy Clin Immunol. 140:1604–1615 e1605. https://doi.org/10.1016/j.jaci.2017.01.018 Josefowicz, S.Z., R.E. Niec, H.Y. Kim, P. Treuting, T. Chinen, Y. Zheng, D.T. Umetsu, and A.Y. Rudensky. 2012. Extrathymically generated regula- tory T cells control mucosal TH2 inflammation. Nature. 482:395–399. https://doi.org/10.1038/nature10772 Kim, K.S., S.W. Hong, D. Han, J. Yi, J. Jung, B.G. Yang, J.Y. Lee, M. Lee, and C.D. Surh. 2016. Dietary antigens limit mucosal immunity by inducing regulatory T cells in the small intestine. Science. 351:858–863. https:// doi.org/10.1126/science.aac5560 Krzywinski, M., J. Schein, I. Birol, J. Connors, R. Gascoyne, D. Horsman, S.J. Jones, and M.A. Marra. 2009. Circos: An information aesthetic for comparative genomics. Genome Res. 19:1639–1645. https://doi.org/10 .1101/gr.092759.109 Lewis, M.J., M.R. Barnes, K. Blighe, K. Goldmann, S. Rana, J.A. Hackney, N. Ramamoorthi, C.R. John, D.S. Watson, S.K. Kummerfeld, et al. 2019. Molecular portraits of early rheumatoid arthritis identify clinical and treatment response phenotypes. Cell Rep. 28:2455–2470.e5. https://doi .org/10.1016/j.celrep.2019.07.091 London, M., A.M. Bilate, T.B.R. Castro, T. Sujino, and D. Mucida. 2021. Stepwise chromatin and transcriptional acquisition of an intraepithelial lymphocyte program. Nat. Immunol. 22:449–459. https://doi.org/10 .1038/s41590-021-00883-8 McMurdie, P.J., and S. Holmes. 2013. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One. 8:e61217. https://doi.org/10.1371/journal.pone.0061217 Merkenschlager, J., S. Finkin, V. Ramos, J. Kraft, M. Cipolla, C.R. Nowosad, H. Hartweger, W. Zhang, P.D.B. Olinares, A. Gazumyan, et al. 2021. Dy- namic regulation of TFH selection during the germinal centre reaction. Nature. 591:458–463. https://doi.org/10.1038/s41586-021-03187-x Mucida, D., M.M. Husain, S. Muroi, F. van Wijk, R. Shinnakasu, Y. Naoe, B.S. Reis, Y. Huang, F. Lambolez, M. Docherty, et al. 2013. Transcriptional reprogramming of mature CD4⁺ helper T cells generates distinct MHC class II-restricted cytotoxic T lymphocytes. Nat. Immunol. 14:281–289. https://doi.org/10.1038/ni.2523 Mucida, D., N. Kutchukhidze, A. Erazo, M. Russo, J.J. Lafaille, and M.A. Curotto de Lafaille. 2005. Oral tolerance in the absence of naturally occurring Tregs. J. Clin. Invest. 115:1923–1933. https://doi.org/10.1172/ JCI24487 Nowosad, C.R., L. Mesin, T.B.R. Castro, C. Wichmann, G.P. Donaldson, T. Araki, A. Schiepers, A.A.K. Lockhart, A.M. Bilate, D. Mucida, and G.D. Victora. 2020. Tunable dynamics of B cell selection in gut germinal centres. Nature. 588:321–326. https://doi.org/10.1038/s41586-020-2865 -9 Pabst, O., and A.M. Mowat. 2012. Oral tolerance to food protein. Mucosal Immunol. 5:232–239. https://doi.org/10.1038/mi.2012.4 Parsa, R., M. London, T.B. Rezende de Castro, B. Reis, J. Buissant des Amorie, J.G. Smith, and D. Mucida. 2022. Newly recruited intraepithelial Ly6A+CCR9+CD4+ T cells protect against enteric viral infection. Immu- nity. 55:1234–1249.e6. https://doi.org/10.1016/j.immuni.2022.05.001 Reis, B.S., A. Rogoz, F.A. Costa-Pinto, I. Taniuchi, and D. Mucida. 2013. Mu- tual expression of the transcription factors Runx3 and ThPOK regulates intestinal CD4⁺ T cell immunity. Nat. Immunol. 14:271–280. https://doi .org/10.1038/ni.2518 Robertson, J.M., P.E. Jensen, and B.D. Evavold. 2000. DO11.10 and OT-II T cells recognize a C-terminal ovalbumin 323-339 epitope. J. Immunol. 164: 4706–4712. https://doi.org/10.4049/jimmunol.164.9.4706 Sefik, E., N. Geva-Zatorsky, S. Oh, L. Konnikova, D. Zemmour, A.M. McGuire, D. Burzyn, A. Ortiz-Lopez, M. Lobera, J. Yang, et al. 2015. Individual intestinal symbionts induce a distinct population of RORγ⁺ regulatory T cells. Science. 349:993–997. https://doi.org/10.1126/science.aaa9420 Sonnenburg, J.L., and F. B¨ackhed. 2016. Diet-microbiota interactions as moderators of human metabolism. Nature. 535:56–64. https://doi.org/ 10.1038/nature18846 Stuart, T., A. Butler, P. Hoffman, C. Hafemeister, E. Papalexi, W.M. Mauck, 3rd, Y. Hao, M. Stoeckius, P. Smibert, and R. Satija. 2019. Compre- hensive integration of single-cell data. Cell. 177:1888–1902 e1821. https:// doi.org/10.1016/j.cell.2019.05.031 Sujino, T., M. London, D.P. Hoytema van Konijnenburg, T. Rendon, T. Buch, H.M. Silva, J.J. Lafaille, B.S. Reis, and D. Mucida. 2016. Tissue adaptation of regulatory and intraepithelial CD4⁺ T cells controls gut inflammation. Science. 352:1581–1586. https://doi.org/10.1126/science.aaf3892 Takeuchi, A., and T. Saito. 2017. CD4 CTL, a cytotoxic subset of CD4+ T cells, their differentiation and function. Front. Immunol. 8:194. https://doi .org/10.3389/fimmu.2017.00194 Torgerson, T.R., A. Linane, N. Moes, S. Anover, V. Mateo, F. Rieux-Laucat, O. Hermine, S. Vijay, E. Gambineri, N. Cerf-Bensussan, et al. 2007. Severe food allergy as a variant of IPEX syndrome caused by a deletion in a noncoding region of the FOXP3 gene. Gastroenterology. 132:1705–1717. https://doi.org/10.1053/j.gastro.2007.02.044 Weiner, H.L., A.P. da Cunha, F. Quintana, and H. Wu. 2011. Oral tolerance. Immunol. Rev. 241:241–259. https://doi.org/10.1111/j.1600-065X.2011 .01017.x Wickham, H. 2016. ggplot2: Elegant Graphics for Data Analysis. Springer- Verlag. New York Yang, B.H., S. Hagemann, P. Mamareli, U. Lauer, U. Hoffmann, M. Beckstette, L. F¨ohse, I. Prinz, J. Pezoldt, S. Suerbaum, et al. 2016. Foxp3+ T cells expressing RORγt represent a stable regulatory T-cell effector lineage with enhanced suppressive capacity during intestinal inflammation. Mucosal Immunol. 9:444–457. https://doi.org/10.1038/mi.2015.74 Zhao, Q., S.N. Harbour, R. Kolde, I.J. Latorre, H.M. Tun, T.R. Schoeb, H. Turner, J.J. Moon, E. Khafipour, R.J. Xavier, et al. 2017. Selective in- duction of homeostatic Th17 cells in the murine intestine by cholera toxin interacting with the microbiota. J. Immunol. 199:312–322. https:// doi.org/10.4049/jimmunol.1700171 Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 18 of 18 Supplemental material Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S1 Figure S1. AA diet mice are heathy and display no evidence of intestinal damage or inflammation. (A) Percent of original body weight (top) or fecal lipocalin-2 levels measured by ELISA (bottom) from SPF mice at indicated weeks after weaning onto AA or standard chow diet. Mean ± SEM representative of two independent experiments using 19–22 mice per group. Comparisons between diets within timepoints are not significant as calculated by unpaired t test. (B) Serum nutritional biomarkers measured in 8-wk-old SPF mice fed AA or chow diet since weaning. ALP—alkaline phosphatase, AST—aspartate amino- transferase, BUN—blood urea nitrogen, HDL—high density lipoprotein, LDL—low density lipoprotein, LDH—lactose dehydrogenase. Two independent ex- periments, each point represents pooled serum from four to six mice. (C–F) SPF or GF mice were fed AA or standard chow diet from weaning until analysis at 8 wk old. (C) Small intestine length. (D) Representative H&E histology images with 200 μm scale bar. (E) H&E pathology scores based on one image per tissue per mouse, where 40 is the maximum score (top left), or tissue morphology measures, where each dot represents the average of four measurements per tissue per mouse. (F) Flow cytometry analysis of myeloid cells in the small intestine IE and LP. (C–F) Mean ± SD representative of two independent experiments with six to nine mice per group. Unpaired t tests with Holm- ˇ Sid´ak multiple comparison test, P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S2 Figure S2. Dietary signals promote accumulation and adaptation of intestinal CD4+ T cells in the small intestine epithelium. Additional data sup- porting Figs. 1 and 2. (A–D) Flow cytometry from the small intestine (A–C) or large intestine (D) IE or LP of 8-wk-old SPF mice weaned onto AA or standard chow diet measuring frequency or absolute count of the indicated cell subsets. Mean ± SD representative of three to five independent experiments using 7–18 mice per group. (E) Flow cytometry of transferred OTII CD4+ T cells from the mesenteric lymph nodes (mLN) after 48 h of OVA supplied 1 mg/ml in drinking water as indicated. Data is representative of two independent experiments with three to four mice per group. (F–H) Flow cytometry from the small intestine IE or LP of 8-wk-old SPF mice weaned onto AA diet with or without 1 mg/ml OVA supplied in drinking water (F–G) or AA, casein, or casein–gluten–soy diet (H) measuring frequency or absolute count of the indicated cell subsets. Mean + SD representative of two to three independent experiments using 3–11 mice per group. (I and J) 16S rRNA sequencing of cecum contents of 8-wk-old SPF mice fed AA or standard chow diet represented by relative phyla abundance (I), and SI Chao1 alpha diversity with mean ± SD (J). Data are from four independent experiments using 11–15 mice per condition. (K and L) Flow cytometry from the IE or LP of 8-wk-old GF or Oligo-MM12 mice weaned onto AA or standard chow diet measuring frequency or absolute count of the indicated cell subsets. Dashed lines show mean value from SPF Chow (red) or SPF AA (blue). Bar plots show mean + SD representative of two to three independent experiments using 6–12 mice per group. (A–L) Unpaired t tests (A–D, F, G, and J) or one-way ANOVA with Tukey’s multiple comparison test (E–H), or two-way ANOVA with P values beneath each plot and Holm-ˇSid´ak multiple comparison test between diets within each colonization within each plot (K–L), P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S3 Figure S3. Chow diet promotes microbiota-independent epithelial adaptation and cytotoxic transcriptional programming of intestinal CD4+ T cells. Additional data supporting Figs. 2 and 3. (A–E and G–J) CD4+ T cells were sorted from the IE or LP of 8-wk-old GF or Oligo-MM12 mice weaned onto AA, AA + OVA, or standard chow diet and scRNAseq was performed using the 10X Genomics platform, pooling two to four mice per diet/colonization group. The data shown is for all sequenced CD4+ T cells (A–E) or subclustered Tregs (G–J). (A) Number of cells sequenced per indicated sample, colored by sequencing batch (left), and violin plots showing number of detected RNA molecules, number of sequenced genes, or percentage of mitochondrial DNA per cell per sequencing batch (right). (B–G) Top five differentially expressed genes (ranked by fold change) in each UMAP gene expression cluster from total CD4+ T cells (B) or Tregs (G). Wilcoxon rank sum test (P < 0.01). (C–J) Frequency of mature clusters (IE1, IE2, IE3, and Th1 combined; C) or Il10 high Tregs (J). (D–I) IE signature score of total CD4+ T cell (E) or Treg (I) gene expression clusters. (E) Frequency of IE2 or IE3 out of total IE CD4+ T cells. (F) Flow cytometry from the IE or LP of 8-wk- old GF or Oligo-MM12 mice weaned onto AA or standard chow diet measuring frequency or absolute count of the indicated cell subsets. Dashed lines show mean value from SPF Chow (red) or SPF AA (blue). Bar plots show mean + SD representative of two to three independent experiments using three to nine mice per group. (H) Frequency of IE subclusters in IE or LP. (K) Flow cytometry from the IE of 8-wk-old SPF mice weaned onto standard chow diet measuring frequency of Granzyme B out of the indicated cell subsets. Mean + SEM representative of three independent experiments using eight mice per group. (C, E, F, J, and K) Two-way ANOVA with P values beneath each plot and Holm-ˇSid´ak multiple comparison test between diets within each colonization within each plot (C, F, and J) or one-way ANOVA with Tukey’s multiple comparison test (E–K), P < 0.05, P < 0.01, P < 0.001, **P < 0.0001. Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S4 Intestinal CD4+ T cell responses during OVA feeding, tolerance, or allergy. Additional data supporting Figs. 6 and 7. iSellTomato mice were Figure S4. analyzed on day 26 after treatment with tamoxifen to permanently label na¨ıve T cells and then exposure to OVA in the context of feeding, tolerance, or allergy. (A) Representative H&E histology images with 200 μm scale bar (left) and pathology scores based on one image per tissue per mouse, where 40 is the maximum score (right). Mean + SD representative of two independent experiments using four to five mice per group. (B–D) Flow cytometry measuring frequency of Tomato+ CD4+ T cells in the large intestine (B) or of the indicated CD4+ T cell subsets out of Tomato+ or Tomato− CD4+ T cells in the IE (C) or LP (D). Data are representative of two independent experiments with five to eight mice per group. (E–I) scRNAseq of 11,217 Tomato+ and Tomato− CD4+ T cells from the IE and LP using four to five mice per condition pooled across two independent experiments and sequencing runs. (E) Captured cells per sample in 10X sequencing experiment with Tomato+ or Tomato– assignments. (F) Violin plots showing number of detected RNA molecules, number of sequenced genes, or percent mitochondrial DNA per cell per sequencing run. (G) Top five differentially expressed genes (ranked by fold change) in each UMAP gene expression cluster from total CD4+ T cells. Wilcoxon rank sum test (P < 0.01). (H) UMAP visualization of sequenced cells positioned by gene expression similarity and colored by tissue (top left) or treatment group (top right) or Tomato assignment (bottom left). (A–D) One-way ANOVA with Tukey’s multiple comparison test, P < 0.05. Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S5 Intestinal CD4+ T cell responses and antigen specificity during OVA feeding, tolerance, or allergy. Additional data supporting Figs. 7 and 8. Figure S5. (A–G) scRNAseq of 11,217 Tomato+ and Tomato− CD4+ T cells from the IE and LP of mice on day 26 of OVA feeding, tolerance, or allergy protocols using four to five mice per condition pooled across two independent experiments and sequencing runs. (A and B) Expression (Pearson residuals) of hallmark Th2 genes (A) or IE or IE4 signature genes (B) in the indicated cell clusters. (C) Frequency of Tomato labeling within each gene expression cluster. (D−F) Frequency of the indicated cell subsets within each group. (E) Three-way volcano plot showing differential gene expression between conditions in Tomato− CD4+ T cells from the IE. Colored genes are differentially expressed (P-adj < 0.05 from FDR-corrected Kruskal–Wallis Test and log2 fold change > 0.5), colored by the condition(s) in which they are upregulated. Select genes of interest are labeled on each plot. (G) Top five differentially expressed genes (ranked by fold change) in each Treg subcluster. Wilcoxon rank sum test (P < 0.01). (H) Overlapping OVA peptide library to determine epitope specificity of OVA-responsive TCRs. A library of 15 aa OVA peptides with 10 aa overlap and 5 aa shifts covering the full length of OVA were tested for TCR response in NFAT hybridomas expressing candidate TCRs. Response was measured with an IL-2 ELISA and data is represented as fold increase in IL-2 production compared to the positive control (a-CD3). (D–F) One- way ANOVA with Tukey multiple comparison test, *P < 0.05. 12 tables are provided online. Table S1 provides the composition of protein antigen-free solid diet (AA). Table S2 shows 16S rRNA sequencing results for SPF mice fed AA or chow diet. Table S3 shows differentially expressed genes across total CD4+ T cell UMAP Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S6 clusters from GF and Oligo-MM12 mice fed AA, AA + OVA, or chow diet. Table S4 shows results from three-way differential gene expression testing between IE CD4+ T cells from AA, AA + OVA, or chow diet mice. Table S5 shows differentially expressed genes across Treg UMAP clusters from GF and Oligo-MM12 mice fed AA, AA + OVA, or chow diet. Table S6 shows results from three-way differential gene expression testing between Tregs from AA, AA + OVA, or chow diet mice. Table S7 shows differentially expressed genes across total CD4+ T cell UMAP clusters from iSellTomato mice in OVA feeding, tolerance, and allergy. Tables S8 shows results from three-way differential gene expression testing between Tomato− IE CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S9 shows results from three-way differential gene expression testing between Tomato+ IE CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S10 shows results from three-way differential gene expression testing between Tomato+ LP CD4+ T cells from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S11 shows differentially expressed genes across Treg UMAP clusters from iSellTomato mice in OVA feeding, tolerance, and allergy. Table S12 shows supporting information about TCRs identified in scRNAseq datasets tested for OVA specificity used in this study. Lockhart et al. Dietary protein shapes intestinal CD4+ T cells Journal of Experimental Medicine https://doi.org/10.1084/jem.20221816 S7
10.1021_jacs.3c01003	pubs.acs.org/JACS Article Environmentally Ultrasensitive Fluorine Probe to Resolve Protein Conformational Ensembles by 19F NMR and Cryo-EM Yun Huang,* Krishna D. Reddy, Clay Bracken, Biao Qiu, Wenhu Zhan, David Eliezer,* and Olga Boudker* Cite This: J. Am. Chem. Soc. 2023, 145, 8583−8592 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information to studying multistate equilibria of ABSTRACT: Limited chemical shift dispersion represents a significant barrier large membrane proteins by 19F NMR. We describe a novel monofluoroethyl 19F probe that dramatically increases the chemical shift dispersion. The improved conformational sensitivity and line shape enable the detection of previously unresolved states in one- dimensional (1D) 19F NMR spectra of a 134 kDa membrane transporter. Changes in the populations of these states in response to ligand binding, mutations, and temperature correlate with population changes of distinct conformations in structural ensembles determined by single-particle cryo-electron microscopy (cryo-EM). Thus, 19F NMR can guide sample preparation to discover and visualize novel conformational states and facilitate image analysis and three-dimensional (3D) classification. ■ INTRODUCTION including The functions of numerous membrane proteins, transporters, channels, and receptors, require conformational transitions. Insights into the structure and dynamics of the functional states are crucial for mechanistic understanding and therapeutic development.1−3 Crystallography and single- particle cryogenic electron microscopy (cryo-EM) can provide structural snapshots of different states in a protein conforma- tional ensemble, while various bulk4 and single-molecule5−7 approaches can report on dynamics. Nuclear magnetic resonance (NMR) spectroscopy is a powerful tool for studying the dynamics of small proteins. However, NMR sensitivity and resolution decline with the increase of protein size, and probing the dynamics of proteins over 100 kDa, including membrane proteins, by NMR remains challenging. One-dimensional (1D) 19F NMR using site-specific probes can be effective for this purpose because of its high sensitivity, the absence of background signals, the exquisite responsiveness of 19F chemical shifts to conformational changes, and its compatibility with inexpensive low-field instruments.8−10 Furthermore, 1D 19F NMR is a single-pulse experiment with no heteronuclear magnetization transfer. It can therefore quantify state populations of even weak or broad resonances. The unique advantages of 19F NMR have led to wide-ranging applications in studying membrane protein dynamics,9,11 protein aggregation,12 protein behavior in cells,13 and drug screening.14 Commonly used 19F probes include aromatic fluorine and trifluoromethyl groups. The latter shows faster longitudinal and slower transverse relaxation rates due to fast leading to higher sensitivity.15 Manglik et al. rotation, developed a trifluoromethyl phenyl group, where the phenyl ring functions as a chemical shift dispersion amplifier.16,17 However, the chemical shift dispersion of trifluoromethyl- based probes generally does not exceed 2 ppm,16,18−20 resulting in severe resonance overlap in large proteins. Recently, Boeszoermenyi et al. reported 19F−13C TROSY improving resonance dispersion via a second dimension21 but sacrificing the advantages of the 1D experiment. Therefore, developing 19F probes with improved dispersion, i.e., increased environmental sensitivity, remains a challenge and is essential for expanding 19F NMR applications.17,22,23 Herein, we report a novel cysteine-conjugated monofluor- oethyl (mFE) probe exhibiting narrow linewidths and ultrahigh sensitivity to conformational changes with chemical shift dispersion reaching 9 ppm. 19F chemical shifts of aliphatic monofluorides depend on C− C−C−F dihedral angles in addition to local electric fields and van der Waals interactions.24,25 In a trifluoroalkyl probe, all rotamers are equivalent, but in the monofluoroalkyl probe, the gauche- and anti-rotamers, which can exhibit chemical shift differences of up to 14 ppm,25 are not equivalent. Because of Received: Published: April 6, 2023 January 27, 2023 © 2023 The Authors. Published by American Chemical Society 8583 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article Figure 1. Synthesis of the mFE probes and site-specific protein labeling. Synthesis of deuterated and protonated mFE probes (a) TsSCD2CD2F and (b) TsSCH2CH2F. (c) Cysteine-specific labeling. (d) Structure of the GltPh trimer with scaffold domains colored tan and the transport domains colored blue. Colored spheres show the sites of single-cysteine mutations A380C (cyan), A381C (magenta), and M385C (yellow). Dashed line indicates the distance between the 19F labels. (e) Elevator transition of the GltPh transport domain from the OFS (left) to IFS (right). Single protomers are shown in the membrane plane, represented by the dotted lines. The bound substrate aspartate is shown as lime spheres. ACN: acetonitrile; DAST: diethylaminosulfur trifluoride; DCM: dichloromethane; X: hydrogen or deuterium; and Ts: toluenesulfonyl. the low energy barrier separating the rotamers, they exchange much faster than the NMR time scale,26 giving a population- weighted chemical shift. We hypothesized that for an mFE probe, weak 19F interactions with the local environment of distinct protein conformations might affect the equilibrium of the gauche- and anti-isomers, giving rise to different population-weighted chemical shifts and larger chemical shift dispersions than for a trifluoroethyl (tFE) probe. Such weak interactions should not significantly alter the kinetics of rotamer exchange, which would remain fast on the NMR time scale. Additionally, fluorine chemical shift anisotropy (CSA) is approximately two times lower in mFE than in the tFE group.27 Because 19F transverse relaxation depends quadratically on the CSA, mFE might feature slower relaxation and narrower linewidths. To test the environmental sensitivity of the novel mFE probe, we used the aspartate transporter GltPh, an excellent model protein because it samples several functional con- formations with known structures. GltPh is similar in size to or larger than many membrane receptors, channels, and trans- porters. As an archaeal homologue of human glutamate it harnesses energy from sodium gradients to transporters, transport aspartate via an “elevator” mechanism.28 It is an obligate homotrimer of 134 kDa molecular weight that can be reconstituted as a protein−micelle particle of ∼300 kDa.29 function During transport, independently,30 undergoing conformational transitions be- tween the outward- and inward-facing states (OFS and IFS), where the substrate-binding site can open to the extracellular solution and cytoplasm, respectively. 19F NMR measurements using mFE-labeled GltPh resolve an unexpectedly complex conformational landscape, demonstrating the power of the new probe and prompting concordant cryo-EM studies that confirm multiple coexisting states. The response of the dominant NMR signals to environmental changes, including temperature, correlates with changes in the ensemble of the three GltPh protomers structures determined by cryo-EM, allowing for provisional assignments of some of the NMR signals to specific structural states. Importantly, our data also suggest that cryo-EM can temperature and other accurately reflect conditions on conformational equilibria in solution, a subject of considerable controversy. the effects of ■ RESULTS High Environmental Sensitivity of the mFE Probe. We designed and synthesized deuterated 19F labeling compounds S-(2-fluoroethyl-1,1,2,2-D 4) p-toluenethiosulfonate (TsSCD2CD2F) using commercially available reagents (Figure 1a,b). Deuterated probes provide sharper lines resulting from reduced 19F−2H scalar couplings compared to 19F−1H couplings and from the reduced 19F−1H dipole relaxation at the cost of reduced sensitivity due to longer longitudinal relaxation times (T1) and relaxation delays. We then prepared single-cysteine mutants, A380C, A381C, and M385C, of a previously characterized fully functional cysteine-free GltPh variant (termed wild-type, WT, for brevity; see the Materials and Methods Section for details and ref 20) and a variant with additional gain-of-function mutations R276S/M395R (termed RSMR). The RSMR mutations accelerate transitions between the OFS and IFS20,31 and substrate uptake by ∼8-fold.32 Thiosulfonates are known to react with cysteine thiols selectively, rapidly, and quantitatively (Figure 1c),33 and the observed efficiency of TsSCD2CD2F reacting with a single cysteine of GltPh was near 100% (Figure S1a). All mFE-labeled mutants retained transport activity when reconstituted into liposomes (Figure S1b). We next compared the chemical shift dispersion of the deuterated mFE (mFED) and its predecessor tFE probe. We recorded 1D 19F spectra of mFED- and tFE-labeled WT- M385C and RSMR-M385C GltPh in the presence of Na+ ions and aspartate or a competitive blocker (3S)-3-[[3-[[4- (trifluoromethyl)benzoyl]amino]phenyl]methoxy]-L-aspartic 8584 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article acid (TFB-TBOA). As previously reported, the tFE probe exhibited three peaks with a chemical shift dispersion of 1.1 ppm (Figures 2a and S2a).20 In contrast, the mFED spectra Figure 2. mFED-labeled GltPh variants show wide chemical shift dispersion. 19F NMR spectra of (a, b) GltPh-M385C and (c, d) GltPh- A380C labeled with (a, c) tFE and (b, d) mFED. From top to bottom, (a, b) WT and RSMR mutants in 100 mM Na+ and 2 mM aspartate (Asp) and the RSMR mutant in 100 mM Na+ and 0.6 mM TFB- TBOA, and (c, d) WT in 400 mM Na+ and 100 mM Na+ and 2 mM aspartate. All spectra were recorded at 25 °C. Raw data are black, fitted spectra are pink, and deconvoluted Lorentzian peaks are blue. The asterisk denotes a signal from the CF3 group of TFB-TBOA. S1− S5 denote resolved resonances of GltPh-M385C-mFED. Δδ is the largest chemical shift difference observed for the labeling site. All 19F NMR spectra here and elsewhere were recorded at least twice on independently prepared protein samples, producing similar results. collectively featured five peaks, S1−S5, at 25 °C, with respective chemical shifts of ∼ −214.8, 216.0, 216.6, 217.5, and 218.1 ppm (Figures 2b and S2b,c). An analysis of three independently prepared samples showed that peak identi- fication is highly reproducible, with only small deviations in the fitted peak positions, linewidths, and populations (Figure S3). We ascribe a slight upfield shift in the S4 peak in the presence of TFB-TBOA compared with the aspartate-bound spectrum (Figure 2b, compare bottom and top spectra) to a structural difference between the aspartate- and TFB-TBOA-bound transporters.34 When we lowered the temperature below 15 °C to slow down conformational transitions, we observed an additional S6 peak at ∼ −218.5 ppm for RSMR-M385C-mFED (Figure S2). Overall, the mFED resonances covered a range of 3.6 ppm. Ligands, mutations, and temperature affect the S1− S6 peak populations, suggesting that the resonances corre- spond to distinct structural states of the transporter and that their populations reflect the kinetic and equilibrium properties of the conformational ensemble. The three GltPh protomers function independently,30,35 and the labeling positions on the adjacent protomers are too distant to affect each other (Figure 1d). Therefore, the 19F resonances report on the structural states of individual protomers. We also compared tFE- and mFED-labeled WT-A380C and WT-A381C GltPh mutants bound to Na+ ions only or Na+ ions and aspartate (Figures 2c,d and S4). For both labeling sites, the mFED probe resulted in wider chemical shift dispersion. Strikingly, the spectra of WT-A380C-mFED in the presence and absence of Asp feature seven peaks distributed over 8.9 ppm (Figure 2d), demonstrating the tremendous potential of the probe. To our knowledge, this is the broadest chemical shift dispersion observed and the greatest number of protein states simultaneously resolved using 19F NMR. The ultrahigh environmental sensitivity of the mFED probe enables the monitoring of conformational equilibria as a function of the physicochemical environment, ligands, and mutations. Moreover, it can reveal hitherto uncharacterized structural states. For example, the multiple peaks observed for the TFB-TBOA-bound RSMR mutant were unexpected, given that it blocks substrate transport. Similarly, 19F NMR spectra of WT-M385C-mFED and RSMR-M385C-mFED bound to the competitive blocker DL-threo-β-benzyloxyaspartic acid (TBOA), from which TFB-TBOA was originally derived, are dominated by two distinct peaks, S4 for WT and S2 for RSMR (TBOA, Figure 3a). We pursued the simpler TBOA-bound spectrum to assess whether these resonances correspond to an IFS by probing their solvent accessibility using OFS or paramagnetic relaxation enhancement (PRE). M385 is solvent- exposed in the OFS but buried on the interface between the transport and scaffold domains in the IFS (Figure 1e).34,36 The addition of the soluble paramagnetic reagent Gd-DTPT-BMA Figure 3. Identification and structural elucidation of new transporter conformations. (a) 19F NMR spectra of WT-M385C-mFED (upper panel) and RSMR-M385C-mFED (bottom panel) in the presence of 100 mM NaCl and 2 mM TBOA. The resonances occurring with similar chemical shifts to those in Figure 2 are labeled S2, S3, and S4. (b) Cryo-EM density maps of TBOA-bound RSMR GltPh mutant protomers in the IFS (left) and OFS (right) structural classes. The corresponding populations are below the maps. 8585 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article Figure 4. Assignment of IFS and OFS conformations based on solvent PRE effects. 19F NMR spectra (left) were recorded, and T1 relaxation times (right) were measured in 300 mM NaCl and 2 mM Asp at 15 °C. Relaxation data were fitted to monoexponential functions (solid lines). The fitted T1 values are (a) 1.36 ± 0.20, 1.41 ± 0.02, 1.26 ± 0.33, 1.56 ± 0.16, and 1.26 ± 0.09 s for S2, S3, S4, S5, and S6, respectively, in the absence of Gd- DPTA−BMA and (b) 1.15 ± 0.20, 1.04 ± 0.08, 0.39 ± 0.08, 0.42 ± 0.06, and 1.01 ± 0.06 s in the presence of 20 mM Gd-DPTA−BMA. The error bars, estimated as described in the Materials and Methods Section, are too small to see. Figure 5. State populations observed in 3D classifications of protomers in cryo-EM parallel NMR measurements. (a) NMR spectra of RSMR- M385C-mFED were recorded in the presence of 500 mM NaCl and 2 mM Asp at 4 (top), 15 (middle), and 30 °C (bottom). (b) Representative maps of the four structural classes identified during cryo-EM particle 3D classifications of RSMR-M385C-mFED prepared at 4 °C. (c) Populations of protomers’ 3D classes in samples preincubated at 4, 15, and 30 °C (left). Open circles are results obtained using different tau values and numbers of classes during 3D classifications in RELION (see the Materials and Methods Section for details). The state populations measured in 19F NMR experiments at the same temperatures are shown in the right panel. Errors are from multiple-peak deconvolutions of spectra in OriginPro 2019. to the TBOA-bound RSMR mutant significantly broadened the S4 but only weakly affected the S2 peak (Figure S5a). Therefore, we assigned S4 to an OFS and S2 to an IFS. Similarly, for the Na+-bound WT, the paramagnetic reagent broadened S4 but not S2 or S3 (Figure S5b), confirming the assignments of S4 and indicating that S3 is also IFS. The observation that the S2 and S3 peaks are intrinsically broader than S4 is also consistent with the faster transverse relaxation expected for a buried M385 site in these states. 19F NMR-Guided Cryo-EM Imaging to Discover New Conformational States. Our 19F NMR data revealed that the blocker TBOA stabilized the WT transporter in the OFS, consistent with the published cryo-EM structure of this state.34 Unexpectedly, however, the blocker appeared to stabilize the RSMR mutant in an unknown IFS state. To understand the structural origin of this state, we imaged TBOA-bound RSMR by cryo-EM. Following particle alignment with imposed C3 symmetry, we performed symmetry expansion and three- dimensional (3D) classification to sort multiple conformations of the transporter protomers.34,37 We observed 91% protomers in the IFS with a wide-open substrate-binding site occupied by TBOA (Figures 3b and S6, and Table S1; see the Materials and Methods Section for data processing details). The remaining protomers were in the OFS, similar to the TBOA-bound WT (RMSD of 0.646 Å). Because S2 and S4 are the only peaks in the RSMR-M385C-mFED NMR spectrum, we ascribe S2 to the highly populated open IFS and S4 to the minor OFS population observed by cryo-EM. Interestingly, the TBOA- bound WT populated a distinct S3 IFS (Figure 3a). Consistently, a cryo-EM structure of the TBOA-bound WT 8586 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article transporter constrained in the IFS by crosslinking showed a different, more closed conformation34 than the structure we assigned to S2. Thus, the previously unobserved S2 state reveals a new modality of the ligand interaction with the transporter. While the physiological relevance of the S2 state is beyond the scope of this paper, it could represent an open-gate intermediate in the substrate release process. Synergistic Use of 19F NMR and Cryo-EM in Exploring the Conformation Landscape. Under transport conditions in the presence of sodium and aspartate, 19F NMR spectra of RSMR-M385C-mFED showed that it populates a surprising number of conformational states, manifesting as resonances S1−S6 (Figures 2b and S2b,c). Temperature modulated their populations so that, at 4 °C, we observed all states, with S6 being the most populated, while above 15 °C, S3 became dominant and S6 invisible (Figure S2). Notably, the S6 peak decreased in amplitude and shifted to the left with increasing temperature, suggesting that it exchanges with another peak with rates approaching the NMR time scale of high μs to low ms. Solvent PRE at 15 °C, where S2−S6 are well resolved, resulted in faster T1 relaxations of S4 and S5 but not S2, S3, and S6 resonances (Figures 4 and S7), suggesting that the mFED probe is exposed to the solvent for the former states and buried in the latter. The S1 peak is too small to evaluate by PRE. These results suggest that the RSMR mutant visits at two OFS-like conformations, S4 and S5, and three least different IFSs, S2, S3, and S6, exchanging slower than the NMR time scale. These assignments are consistent with the S2, S3, and S4 assignments described above for the TBOA-bound and Na+-bound transporters (Figures 3 and S6). The temperature-dependent population changes observed in 19F NMR experiments prompted us to ask whether cryo-EM can recapitulate them and inform the assignment of resonances to structural states. We flash-froze cryo-EM grids of RSMR- M385C-mET preincubated in 500 mM NaCl and 2 mM aspartate at 4, 15, and 30 °C. After data processing, 3D classification of protomers imaged at 4 °C revealed that ∼87% of them fell into IFS classes with variable orientations of the transport domain relative to the scaffold (Figure S8). The remaining ∼13% were in a previously described intermediate OFS (iOFS), in which the transport domain moves inward slightly compared to the OFS. Based on similar transport domain positions, we grouped the IFS classes into IFS-A and IFS-B ensembles with populations of 19 and 67%, respectively (Figures 5 and S8a). The main structural difference between them is that the transport domain packs more tightly against the scaffold in IFS-A but leans away and leaves a detergent- filled gap in IFS-B (Figure S8e). Similar structures were observed crystallographically and termed “locked” and “unlocked”, respectively.31 The increasing temperature had little effect on the population of the iOFS but led to a dramatic increase in IFS-A to ∼82% at 30 °C and a corresponding decrease in IFS-B populations to ∼9% (Figures 5c and S8b). from IFS-B to IFS-A at higher temperatures observed in cryo-EM parallels the increase of S3 and decrease of S6 in 19F NMR (Figure 5a), strongly suggesting that S3 corresponds to IFS-A and S6 to IFS-B (Figure 5c). Notably, the RSMR trimer crystallized at 4 °C with two protomers in the “unlocked” IFS-B and one protomer in the “locked” IFS-A,31 consistent with the higher IFS-B population observed by NMR and cryo-EM at 4 °C. S4, which corresponds to the OFS, is weakly populated in RSMR but in the WT transporter (Figures 2b and 3a). dominant The population shift Consistently, cryo-EM did not identify the OFS in the 4 °C data set. Instead, we only found ∼2% of particles in the OFS in the 15 °C cryo-EM data set, reflecting the challenge in cryo- EM of detecting lowly populated states. S5 likely corresponds to iOFS because it is the only other observed structural state with the solvent-exposed 385C residue. While these assign- ments are plausible based on solvent exposure and the correlated population changes in the NMR and cryo-EM data, they remain speculative in the absence of direct structural data, such as we previously used to assign NMR signals to the OFS or IFS.20 The identities of the S1 and S2 resonances remain ambiguous. In particular, we did not observe a state in the cryo-EM classifications resembling the TBOA-bound RSMR structure, which populates the S2 peak (Figure 3). Therefore, it is possible that one of the IFS conformations in the ensembles serendipitously shows a chemical shift similar to that of the TBOA-bound RSMR structure. The decrease in the S2 population at higher thermodynamically similar to S6, suggests that it might correspond to one of the “unlocked” IFS-B subclasses. temperatures, ■ DISCUSSION AND CONCLUSIONS Despite decades of efforts to develop biophysical methods to study protein dynamics, monitoring changes in membrane protein samples with more than two states remains challenging. The model protein in this study is a 134 kDa membrane aspartate/sodium symporter, GltPh, which undergoes complex conformational transitions to transport aspartate and coupled sodium ions across the membrane. Resolving multiconforma- tional ensembles of such large membrane proteins using existing NMR methods is difficult. The new mFE 19F probe takes advantage of the unique electronic properties of the monofluoroalkyl group leading to ultrahigh environmental sensitivity and achieving a chemical shift dispersion of ∼9 ppm for GltPh. However, a reduction of the number of chemically equivalent fluorine atoms and an increase in T1 relaxation time make the sensitivity of the mFEH probe only one-sixth that of the tFE probe. The sensitivity of the deuterated mFED version is approximately two-fold lower still. This new 19F probe constitutes a straightforward and inexpensive tool enabling 19F NMR-guided high-resolution structural determinations by crystallography or cryo-EM. For example, 19F NMR showed that the transport blocker TBOA stabilizes the gain-of-function RSMR mutant of GltPh in a different conformation than the WT. Consequent cryo-EM imaging led to the discovery of a novel conformation of the transporter in the IFS featuring an open substrate gate and a new mode of inhibitor binding. 19F NMR resonances are difficult to assign to specific structural states because their chemical shifts cannot be routinely predicted from even high-resolution structural models and are exquisitely sensitive to the local environment, indirectly reflecting global protein conformations. Previously, we showed that transition-metal-mediated longitudinal PRE could assist with state assignments by evaluating the distance between the 19F label and a double-histidine-coordinated metal ion. Here, we take advantage of enhanced line broadening and T1 relaxation due to the PRE effects of gadolinium compounds in solution to distinguish states with buried and exposed mFE probes. These measurements discriminate between the OFS- and IFS-like states in WT GltPh and its RSMR mutant. Protein solutions are plunge-frozen for cryo-EM imaging in a process thought to preserve protein conformations, which can 8587 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article then be visualized by 3D classifications of the particles. Factors such as protein adsorption to the grids, surface tension, and temperature changes during freezing may shift the conforma- tion equilibrium. In addition, data processing and parameters used in classifications can affect the obtained populations of 3D classes.38 Therefore, whether the class distributions reflect state populations in solutions is unclear. Here, for the first time, we evaluated the reliability of the state populations obtained from 3D classifications of EM-imaged particles. We found that the state populations determined by cryo-EM match well with those measured by 19F NMR. For example, the IFS and OFS populations of TBOA-bound RSMR were 95 and 5% in 19F NMR and 91 and 9% in cryo-EM, respectively. There was also good correspondence in complex ensembles of the aspartate-bound RSMR mutant preincubated at different temperatures. We observed a temperature-dependent corre- lated population increase of the main S3 resonance and the cryo-EM IFS-A structural ensemble and a corresponding decrease in the S6 resonance and the IFS-B ensemble. Structural determination by cryo-EM and accurate state populations measured by 19F NMR provide a novel means to correlate the structural and thermodynamic properties of membrane proteins. For example, our results suggest that IFS- B has a dramatically lower enthalpy than IFS-A, resulting in the the IFS-B to IFS-A steep temperature dependence of transition. The lower IFS-B enthalpy must be associated with breaking the protein interface between the transport and scaffold domains and inserting detergent moieties into the gap. The detailed analysis of the phenomenon is beyond the scope of the paper. Nevertheless, our results show that even modest temperature changes can have profound effects on the observed conformational ensemble of a membrane protein, a feature characteristic, for example, of temperature-sensing ion channels.39 19F NMR is ideally suited to probe the temperature dependence of protein equilibria and might facilitate the study of the thermodynamics of protein-bilayer interactions. 19F NMR can detect conformations of RSMR-M385C-mEF present at only a few percent, such as the S4 peak, attributed to the OFS. Such weakly populated states might be important functional intermediates, yet they are difficult to isolate during the 3D classification of cryo-EM-imaged particles, especially when several conformations coexist. Thus, we only found the OFS class in the 15 °C cryo-EM data set, where 19F NMR suggests its population is slightly higher than at 4 or 30 °C and where we collected the largest number of particles. Therefore, the ability of 19F NMR to visualize rare states provides a means to assess their populations and optimize conditions that can enrich them for high-resolution structural elucidation. In summary, mFE chemical shift dispersion significantly exceeds that of tFE, enabling a more detailed and quantitative description of protein conformational ensembles and guiding Improved peak separation should also cryo-EM imaging. facilitate measurements of chemical exchange rates by methods such as EXSY20 and saturation transfer, including CEST.40 Continuously expanding structural methodologies reveal that proteins populate diverse conformational ensembles. For example, a G-protein-coupled adenosine A2A receptor samples at least five distinguishable states during activation.41 Our model protein is similar in size to or larger than many physiologically important transporters, and ion channels. The new 19F probe should greatly facilitate mechanistic studies of such proteins. receptors, ■ MATERIALS AND METHODS Synthesis of S-(2-Fluoroethyl-1,1,2,2-D4) P-Toluenethiosul- fonate. To a solution of potassium p-toluenethiosulfonate (678 mg, TCI Chemical) in dry acetonitrile (10 mL) was added 2- bromoethanol-1,1,2,2-d4 (240 μL, Cambridge Isotope Laboratories) under argon. The mixture was stirred at 60 °C overnight. The solvent was evaporated under reduced pressure, and the residue was dissolved in 50 mL of ethyl acetate and washed with 0.5 M HCl aqueous solution (2 × 20 mL) and brine. The organic layer was dried over anhydrous Na2SO4 and then evaporated under reduced pressure to give S-(2-hydroxyethyl-1,1,2,2-D4) p-toluenethiosulfonate (600 mg, 85% yield) as an oil. The product was directly used for the next step without further purification. To a solution of S-(2-hydroxyethyl-1,1,2,2-D4) p-toluenethiosulfo- nate (480 mg, 2 mmol) in dry dichloromethane (5 mL) precooled to −70 °C 1.5 equiv of diethylaminosulfur trifluoride (3 mL of 1 M solution in dichloromethane, Sigma-Aldrich) was added dropwise under argon. After 10 min of stirring at −70 °C, the flask was warmed to room temperature for 30 min. The reaction was quenched with 0.5 mL of methanol. The reaction mixture was evaporated to dryness and then purified using a flash column (RediSep Rf) using 0−60% petroleum ether/ethyl acetate. The major fraction was collected and evaporated to obtain the pure title compound (220 mg, 45% yield). The identity of the compound was confirmed by NMR. 1H NMR (500 MHz, CDCl3) δ 7.82 (d, J = 8.4 Hz, 2H), δ 7.36 (d, J = 7.7 Hz, 2H), δ 2.46 (s, 3H); 13C NMR (125 MHz, CDCl3) δ 145.40, 141.85, 130.18, 127.24, 81.44−79.32 (J = 171.19 Hz), 34.88 (m), 31.84; 19F NMR (470 MHz, CDCl3) δ −215.25 (m). Synthesis of S-(2-Fluoroethyl) P-Toluenethiosulfonate. To a solution of potassium p-toluenethiosulfonate (620 mg, 2.7 mmol, TCI Chemical) in dry acetonitrile (10 mL) 1-bromo-2-fluoroethane (203 μL, Accela ChemBio Inc.) was added under argon. The mixture was stirred at 60 °C overnight. The solvent was removed under reduced pressure, and the residue was dissolved in 50 mL of ethyl acetate and washed with 0.5 M HCl aqueous solution (2 × 20 mL) and brine. The organic layer was dried over anhydrous Na2SO4 and evaporated under reduced pressure. The residue was purified by flash column chromatography to give the pure title compound (503 mg, 86% yield) as an oil. The identity of the compound was confirmed by NMR. 1H NMR (500 MHz, CDCl3) δ 7.82 (d, J = 8.2 Hz, 1H), 7.36 (d, J = 8.1 Hz, 1H), 4.61 (t, J = 6.3 Hz, 1H), 4.52 (t, J = 6.3 Hz, 1H), 3.30 (t, J = 6.3 Hz, 1H), 3.26 (t, J = 6.3 Hz, 1H), 2.46 (s, 3H); 13C NMR (126 MHz, CDCl3) δ 145.22, 141.72, 130.01, 127.11, 81.01 (d, J = 172.6 Hz), 35.38 (d, J = 22.7 Hz), 21.70; 19F NMR (471 MHz, CDCl3) δ −213.51 (tt, J = 46.9, 20.4 Hz). Protein Expression, Purification, and Labeling. The single- cysteine mutations A380C, A381C, and M385C were introduced into the Y215H/E219H cysteine-free GltPh mutant (termed WT here for brevity), where the introduction of double histidine has been shown to not perturb the transport activity significantly.20 The M385C mutation was also introduced into the gain-of-function variant generated by additional R276S/M395R mutations (RSMR for brevity). All constructs were expressed, purified, and labeled as described, with modifications.20,36 In brief, pBAD plasmids encoding constructs with C-terminal thrombin cleavage sites followed by (His)8-tag were transformed into E. coli DH10-B cells (Invitrogen). Cells were grown in LB media at 37 °C to an OD600 of 1.1. The temperature was then set to 30 °C and protein expression was induced by adding 0.2% arabinose (Goldbio). Cells were grown for additional 16 h. The cells were harvested by centrifugation and resuspended in 20 mM HEPES, pH 7.4, 200 mM NaCl, 1 mM L-asp, and 1 mM EDTA. The suspended cells were broken by an Emulsiflex C3 high-pressure homogenizer (Avestin Inc.) in the presence of 0.5 mg/mL lysozyme (Goldbio) and 1 mM phenylmethanesulfonyl fluoride (PMSF, MP Biomedicals). After centrifugation for 15 min at 5000g, the supernatant was centrifuged at 125,000g for 50 min. For mFE labeling, the membrane pellets were collected and solubilized in a buffer containing 20 mM HEPES, pH 7.4, 200 mM NaCl, 1 mM Asp, 10 mM 2-mercaptoethanol, 40 mM n-dodecyl-β-D- 8588 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article maltopyranoside (DDM, Anatrace, Inc.) for 2 h at 4 °C. The mixture was centrifuged for 50 min at 125,000g. The supernatant was diluted three times with buffer A (20 mM HEPES, pH 7.4, 200 mM NaCl, 1 mM Asp) and incubated with Ni-NTA resin (Qiagen) for 1 h at 4 °C. The resin was washed with six volumes of buffer A supplemented with 1 mM DDM and 25 mM imidazole, and proteins were eluted in buffer A supplemented with 1 mM DDM and 300 mM imidazole. EDTA was added to the collected protein fractions to a final concentration of 0.5 mM. The protein was concentrated to ∼10 mg/mL using concentrators with 100 kDa MW cutoff (Amicon). Protein concentration was determined by ultraviolet (UV) absorbance at 280 nm using the extinction coefficient of 57,400 M−1 cm−1 and MW of 44.7 kDa (protomer). Two molar equivalents of TsSCD2CD2F or TsSCH2CH2F, prepared as stock solutions in DMSO, were added to the protein samples, followed by incubation at 4 °C for 2 h or room temperature for 1 h. Thrombin was added and incubated overnight at room temperature to cleave the (His)8-tag. For tFE labeling, the membrane pellet was solubilized in a buffer containing 20 mM HEPES, pH 7.4, 200 mM NaCl, 1 mM Asp, 40 mM DDM, and 2 mM 4,4'-dithiodipyridine (DTDP, Sigma-Aldrich). After binding to Ni-NTA resin, the protein−resin slurry was first washed with five volumes of buffer A supplemented with 1 mM DDM. Trifluoroethanethiol (2 mM, Sigma-Aldrich) was added, and the slurry was incubated with mixing at 4 °C overnight. The resin was washed with buffer A supplemented with 1 mM DDM and 25 mM imidazole and then eluted with buffer A supplemented with 1 mM DDM and 300 mM imidazole. The (His)8-tag was cleaved using thrombin at room temperature overnight. Both mFE- and tFE-labeled proteins were further purified by size exclusion chromatography (SEC) using a Superdex 200 Increase 10/ 300 GL column (GE Healthcare Life Sciences) in a buffer containing 20 mM HEPES, pH 7.4, 50 mM KCl, and 1 mM DDM. NaCl (100 mM) was added to the protein fractions immediately. The protein was concentrated and supplemented with additional NaCl and ligands as needed. The labeling efficiency was quantified using 1D NMR and found to be quantitative (Figure S1a) Transport Activity Assay. Unlabeled and mFE-labeled WT GltPh were concurrently reconstituted into liposomes, and initial rates of 3H Asp uptake were measured as previously described.42 Briefly, liposomes were prepared from a 3:1 (w/w) mixture of E. coli polar lipids and egg yolk phosphatidylcholine (Avanti Polar Lipids) in 20 mM HEPES/Tris, pH 7.4, containing 200 mM KCl and 100 mM choline chloride. Liposomes were destabilized by the addition of Triton X-100 at a detergent-to-lipid ratio of 0.5:1 (w/w). GltPh proteins were added at a final protein-to-lipid ratio of 1:2,000 (w/w) and incubated for 30 min at room temperature. Detergents were removed with Bio-Beads SM-2 resin (Bio-Rad) via two incubations at room temperature, one overnight incubation at 4 °C, and two more room temperature. The proteoliposomes were incubations at concentrated to 50 mg/mL and flash-frozen in liquid N2. On the day of the experiment, liposomes were thawed and extruded through 400 nm filters. The uptake reaction was started by diluting the proteoliposomes 100-fold into a reaction buffer containing 20 mM HEPES/Tris pH 7.4, 200 mM KCl, 100 mM NaCl, 1 μM 3H-L-Asp (PerkinElmer), and 0.5 μM valinomycin. Uptake was measured at 2 min at 35 °C. Reactions were quenched by adding 10 volumes of cold buffer containing 20 mM HEPES/Tris pH 7.4, 200 mM KCl, and 100 mM LiCl. Uptakes of the mutants were normalized to the WT protein in each experiment. Each data point is an average of at least two technical replicates, and the data are composed of the results from three independent liposome reconstitutions. 19F NMR Spectroscopy. 19F NMR spectra were collected on a Bruker Avance III HD 500 MHz spectrometer equipped with a TCI 1H-19F/13C/15N triple resonance cryogenic probe (Bruker Instru- 19F. For mFED-labeled proteins, 50 μM 2- ments) tuned for fluoroethanol and 10% D2O were added to the sample and used as reference (−224.22 ppm) and a lock signal, a chemical shift respectively. Typically, 160 μL of the transporter solution at a final concentration of 100−250 μM (protomer) was loaded into a 4 mm Shigemi tube (Shigemi Co., Ltd). 1D 19F NMR spectra were recorded using the standard ZG pulse in the Bruker pulse library, with 2096 points recorded and a spectral width (SW) of 40 ppm. The carrier frequency was set at −220 ppm. The recycle delay was set to 1.5 and 0.9 s for mFED and mFEH, respectively, except when otherwise indicated. The scan numbers were set according to protein and salt concentrations and temperatures between 2 and 30 K (acquisition times ranged from 1 to 14 h) to achieve a satisfactory signal-to-noise ratio. For tFE-labeled proteins, 50 μM trifluoroacetic acid was added to the samples and used as a chemical shift reference (−76.55 ppm), the carrier frequency was set at −70 ppm, and the recycle delay was set to 0.6 s. All of the spectra were recorded without decoupling. All 1D 19F NMR spectra were processed using MestReNova 12.0.0 software (Mestrelab Research). The free induction decay signals were zero-filled to 8192 points and Fourier transformed after applying a 20 Hz exponential window function. The spectra were baseline- corrected, and the peaks were manually picked. The initial values of peak heights and linewidths were manually set such that their sum approximated the original spectrum. Simulated annealing was used for iterative fitting. During fitting, linewidths were constrained between 20 and 800 Hz, peak positions were constrained within 5% variation, and the peak shapes were set to Lorentzian. The reported linewidth values were obtained by subtracting the line-broadening value of 20 Hz from the fitted linewidth. To test the reliability of the fitting procedure and the reproducibility of the resulting parameters, we recorded spectra of three independently prepared RSMR-M385C- mFED samples in 300 mM NaCl and 2 mM Asp. The obtained fitted values of chemical shifts, linewidths, and state populations showed small standard deviations (Figure S3a,c,e). Additionally, we exported MestReNova-processed spectra and fitted them using the Multiple Peak Fit tool in OriginPro 2019 software (OriginLab), which also reported small fitting errors (Figure S3d). 19F longitudinal relaxation times (T1) were measured by the inversion recovery method. The recycle delays were set to 5 and 2.5 s for mFED and mFEH, respectively. The spectra were processed in MestReNova, imported into OriginPro (OriginLab), and globally fitted using the Global Fit Tool patch. The peaks were picked, and initial estimates of their chemical shifts, linewidths, and heights were set manually. The chemical shift and linewidth values for each peak were held fixed between all spectra in the relaxation series but varied relative to other peaks during fitting. The peak intensities and areas were normalized using values obtained after the longest delay. T1 relaxation times were estimated by fitting the relaxation plots to single exponential functions I = I0 (1 − 2 exp(−t/T1)), where I is the normalized peak intensity and t is the relaxation time. To explore whether probe deuteration sharpens the resonance peaks as expected, we compared the spectra of deuterated (mFED) and protonated mFE (mFEH) attached to WT-M385C in the presence of saturating Na+ ions, where we observed well-resolved S3 and S4 peaks (Figure S9). Consistent with deuterium reducing the 19F−1H dipole relaxation and peak splitting due to 19F−1H spin−spin coupling, mFED produced sharper peaks than mFEH with a more pronounced effect on the intrinsically sharper S4 peak (Figure S9). However, the longer T1 relaxation times (Figure S9), increasing the the benefits of recycle delay and reducing sensitivity, offset deuteration. If hardware enabling 1H decoupling during 19F detection is available, the protonated mFE probe might be preferable. Cryo-EM Data Collection. To prepare cryo-grids, 3.5 μL of labeled GltPh protein (4.5 mg/mL) was applied to a glow-discharged QuantiFoil R1.2/1.3 300-mesh gold grid (Quantifoil Micro Tools GmbH) and incubated for 2 min under 100% humidity at the desired temperatures. Following incubation, grids were blotted for 3 s at 0 blot force and plunge-frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Cryo-EM imaging data were acquired on a Titan Krios microscope (Thermo Fisher Scientific) operated at 300 kV with a K3 Summit direct electron detector (Gatan, Inc.). Automated data collection was carried out in super-resolution mode with a magnification of 105,000×, which corresponds to a calibrated pixel size of 0.852 Å on the specimen and 0.426 Å for super-resolution images. An energy slit width of 20 eV was used throughout the the TBOA-bound RSMR sample, movies were collection. For 8589 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society collected using Leginon43 at a total dose of 52.88 e−/Å2 distributed over 48 frames (1.102 e−/Å2/frame) with an exposure time of 2.40 s (50 ms/frame) and a defocus range of 1.3 −2.0 μm. A total of 8100 movies were collected. For the Asp-bound RSMR grids made at 4 °C, movies had a total dose of 50.94 e−/Å2 distributed over 48 frames (1.061 e−/Å2/frame) with an exposure time of 2.40 s (50 ms/frame) and a defocus range of 1.3−1.5 μm. A total of 5267 movies were collected. For the Asp-bound RSMR grids prepared at 15 °C, movies had a total dose of 58.57 e−/Å2 distributed over 50 frames (1.171 e−/ Å2/frame) with an exposure time of 2 s (40 ms/frame) and a defocus range of 0.9−1.9 μm. A total of 14,132 movies were collected. For the Asp-bound RSMR grids prepared at 30 °C, movies had a total dose of 56.04 e−/Å2 distributed over 40 frames (1.401 e−/Å2/frame) with an exposure time of 1.6 s (40 ms/frame) and a defocus range of 1.3−1.6 μm. A total of 5823 movies were collected. Image Processing. The frame stacks were motion corrected using MotionCor244 with 2× binning, and contrast transfer function (CTF) estimation was performed using CTFFIND4.1.45 Further processing steps were carried out using RELION (REgularized LIkelihood OptimizatioN) 3.0.8 or 3.1.0 and cryoSPARC 3.0 or 3.2.46,47 Particles were picked from micrographs using the Laplacian-of-Gaussian (LoG) picker, aiming for ∼2000 picks per micrograph. These particles were extracted using a box size of 300 pixels with 4× binning and imported into cryoSPARC. Following one round of two-dimensional (2D) classification to remove artifacts, the particles underwent three rounds of heterogeneous refinement in C1 using eight classes. Seven classes were noise volumes created by one iteration of ab initio, and one was an unmasked 3D volume obtained from a complete ab initio run. The particles were converted to the RELION format using PyEM48 and re- extracted at full box size. These particles were reimported into cryoSPARC and underwent one round of nonuniform (NU) refinement using C3 symmetry.49 These particles were converted back to the RELION format for Bayesian polishing with parameters obtained using 5000 random particles.50 The polished particles were reimported into cryoSPARC and subjected to one round of NU refinement in C3 with both local and global CTF refinement options turned on. The particles were again polished in RELION using an expanded box size of 384 pixels, reimported into cryoSPARC, and subjected to one round of NU refinement with C3 symmetry and local and global CTF refinement options turned on. The particles were then C3 symmetry-expanded and subjected to a focused 3D classification in RELION. The local mask was generated by UCSF Chimera51 using a combination of OFS (chain A of PDB model 2NWX) and IFS (chain A of PDB model 3KBC). The tau value was set between 10 and 40 depending on the job. The 3D class populations were calculated by dividing the particle number in the individual class or a group of similar classes by the total number of C3 symmetry-expanded particles. The particles from the individual classes or combined similar classes were imported separately into cryoSPARC and subjected to local refinement using the mask and map obtained from the most recent NU refinement. Model Building and Refinement. Crystal or cryo-EM structures were used as initial models and docked into the density maps using UCSF Chimera. The models were first real-space refined in PHENIX.52 Misaligned regions were manually rebuilt, and missing side chains and residues were added in COOT.53 Models were iteratively refined with applied secondary structure restraints and validated using MolProbity.54 To cross-validate models, all atoms in the refined models were randomly displaced by an average of 0.3 Å, and each resulting model was refined against the first half-map obtained from processing. The FSC between the refined models and the first half-maps was calculated and compared to the FSC of the other half-maps. The resulting FSC curves were similar, showing no evidence of overfitting (Figure S10). The structural figures were prepared in UCSF Chimera and PyMOL (DeLano Scientific). pubs.acs.org/JACS Article ■ ASSOCIATED CONTENT sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.3c01003. Characterization of labeling efficiency and transport activity assay (Figure S1); additional spectral and results (Figures S2−S10); cryo-EM data structural collection, map and model validation statistics for PDB entry 7UG0, 7UGJ, 7UGD, 7UGV, 7UGX, 7UH6, and 7UH3 (Figure S10 and Table S1); and NMR character- ization of synthesized labeling compounds (PDF) ■ AUTHOR INFORMATION Corresponding Authors Yun Huang − Department of Physiology & Biophysics, Weill Cornell Medicine, New York, New York 10021, United States; Howard Hughes Medical Institute, Chevy Chase, Maryland 20815, United States; 2991-6122; Email: [email protected] orcid.org/0000-0002- David Eliezer − Department of Biochemistry, Weill Cornell Medicine, New York, New York 10021, United States; orcid.org/0000-0002-1311-7537; Email: dae2005@ med.cornell.edu Olga Boudker − Department of Physiology & Biophysics, Weill Cornell Medicine, New York, New York 10021, United States; Howard Hughes Medical Institute, Chevy Chase, Maryland 20815, United States; Email: olb2003@ med.cornell.edu Authors Krishna D. Reddy − Department of Physiology & Biophysics, Weill Cornell Medicine, New York, New York 10021, United States; Howard Hughes Medical Institute, Chevy Chase, Maryland 20815, United States Clay Bracken − Department of Biochemistry, Weill Cornell Medicine, New York, New York 10021, United States Biao Qiu − Department of Physiology & Biophysics, Weill Cornell Medicine, New York, New York 10021, United States Wenhu Zhan − Department of Microbiology & Immunology, Weill Cornell Medicine, New York, New York 10021, United States; Present Address: iCarbonX (Shenzhen) Co., Ltd., Shenzhen, 518000, China; orcid.org/0000-0002-7095- 7081 Complete contact information is available at: https://pubs.acs.org/10.1021/jacs.3c01003 Notes The authors declare no competing financial interest. ■ ACKNOWLEDGMENTS The authors thank Dr. M. Goger and Dr. S. Bhattacharya for their help with setting up NMR. The authors thank Dr. B. Wang, Dr. A. Paquette, and Dr. W. Rice at the NYU cryo-EM facility for help with electron microscopy data collection. The authors also thank W. Eng and E. Wu for assistance with protein expression and Dr. Z. Shen for assistance with chemical supported by NIH grants synthesis. This work was R37NS085318 and R01NS064357 (O.B.), RF1AG066493 and R35 GM136686 (D.E.), F32 NS102325 (K.R.), and S10OD016320 and S10OD028556 (C.B.). O.B., D.E., and C.B. are members of the New York Structural Biology Center (NYSBC), which is supported in part by NIH Grant P41 8590 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article GM118302 (CoMD/NMR: Center on Macromolecular Dynamics by NMR Spectroscopy), ORIP/NIH facility improvement grant CO6RR015495, and NIH grant S10OD018509. ■ REFERENCES (1) Lee, Y.; Lazim, R.; Macalino, S. J. Y.; Choi, S. Importance of protein dynamics in the structure-based drug discovery of class A G protein-coupled receptors (GPCRs). Curr. Opin. Struct. Biol. 2019, 55, 147−153. (2) Peng, J. W. Communication Breakdown: Protein Dynamics and Drug Design. Structure 2009, 17, 319−320. (3) Sekhar, A.; Kay, L. E. An NMR View of Protein Dynamics in Health and Disease. Annu. Rev. Biophys. 2019, 48, 297−319. (4) Palmer, A. G. NMR Characterization of the Dynamics of Biomacromolecules. Chem. Rev. 2004, 104, 3623−3640. (5) Lerner, E.; Cordes, T.; Ingargiola, A.; Alhadid, Y.; Chung, S.; Michalet, X.; Weiss, S. Toward dynamic structural biology: Two decades of single-molecule Förster resonance energy transfer. Science 2018, 359, No. 288. (6) Lewis, J. H.; Lu, Z. Resolution of ångström-scale protein conformational changes by analyzing fluorescence anisotropy. Nat. Struct. Mol. Biol. 2019, 26, 802−807. (7) Heath, G. R.; Scheuring, S. Advances in high-speed atomic force microscopy (HS-AFM) reveal dynamics of transmembrane channels and transporters. Curr. Opin. Struct. Biol. 2019, 57, 93−102. (8) Danielson, M. A.; Falke, J. J. Use of 19F NMR to probe protein structure and conformational changes. Annu. Rev. Biophys. Biomol. Struct. 1996, 25, 163−195. (9) Didenko, T.; Liu, J. J.; Horst, R.; Stevens, R. C.; Wüthrich, K. Fluorine-19 NMR of integral membrane proteins illustrated with studies of GPCRs. Curr. Opin. Struct. Biol. 2013, 23, 740−747. (10) Di Pietrantonio, C.; Pandey, A.; Gould, J.; Hasabnis, A.; Prosser, R. S.Understanding Protein Function Through an Ensemble Description: Characterization of Functional States by 19F NMR. In Methods in Enzymology; Elsevier, 2019; Vol. 615, pp 103−130. (11) Picard, L.-P.; Prosser, R. S. Advances in the study of GPCRs by 19F NMR. Curr. Opin. Struct. Biol. 2021, 69, 169−176. (12) Suzuki, Y.; Brender, J. R.; Soper, M. T.; Krishnamoorthy, J.; Zhou, Y.; Ruotolo, B. T.; Kotov, N. A.; Ramamoorthy, A.; Marsh, E. N. G. Resolution of Oligomeric Species during the Aggregation of Aβ1−40 Using 19F NMR. Biochemistry 2013, 52, 1903−1912. (13) Zhu, W.; Guseman, A. J.; Bhinderwala, F.; Lu, M.; Su, X. C.; Gronenborn, A. M. Visualizing Proteins in Mammalian Cells by 19F NMR Spectroscopy. Angew. Chem., Int. Ed. 2022, 61, No. e202201097. (14) Buchholz, C. R.; Pomerantz, W. C. K. 19F NMR viewed through two different lenses: ligand-observed and protein-observed 19F NMR applications for fragment-based drug discovery. RSC Chem. Biol. 2021, 2, 1312−1330. (15) Rashid, S.; Lee, B. L.; Wajda, B.; Spyracopoulos, L. Side-Chain Dynamics of the Trifluoroacetone Cysteine Derivative Characterized by 19F NMR Relaxation and Molecular Dynamics Simulations. J. Phys. Chem. B 2019, 123, 3665−3671. (16) Manglik, A.; Kim, T. H.; Masureel, M.; Altenbach, C.; Yang, Z.; Hilger, D.; Lerch, M. T.; Kobilka, T. S.; Thian, F. S.; Hubbell, W. L.; Prosser, R. S.; Kobilka, B. K. Structural Insights into the Dynamic Process of β2-Adrenergic Receptor Signaling. Cell 2015, 161, 1101− 1111. (17) Ye, L.; Larda, S. T.; Frank Li, Y. F.; Manglik, A.; Prosser, R. S. A comparison of chemical shift sensitivity of tags: optimizing resolution in 19F NMR studies of proteins. J. Biomol. NMR 2015, 62, 97−103. (18) Liu, J. J.; Horst, R.; Katritch, V.; Stevens, R. C.; Wuthrich, K. Biased signaling pathways in β2-adrenergic receptor characterized by 19F-NMR. Science 2012, 335, 1106−1110. (19) Frei, J. N.; Broadhurst, R. W.; Bostock, M. J.; Solt, A.; Jones, A. J. Y.; Gabriel, F.; Tandale, A.; Shrestha, B.; Nietlispach, D. trifluoromethyl Conformational plasticity of ligand-bound and ternary GPCR complexes studied by 19F NMR of the beta1-adrenergic receptor. Nat. Commun. 2020, 11, No. 669. (20) Huang, Y.; Wang, X.; Lv, G.; Razavi, A. M.; Huysmans, G. H. M.; Weinstein, H.; Bracken, C.; Eliezer, D.; Boudker, O. Use of paramagnetic 19F NMR to monitor domain movement in a glutamate transporter homolog. Nat. Chem. Biol. 2020, 16, 1006−1012. (21) Boeszoermenyi, A.; Chhabra, S.; Dubey, A.; Radeva, D. L.; Burdzhiev, N. T.; Chanev, C. D.; Petrov, O. I.; Gelev, V. M.; Zhang, M.; Anklin, C.; Kovacs, H.; Wagner, G.; Kuprov, I.; Takeuchi, K.; Arthanari, H. Aromatic 19F-13C TROSY: a background-free approach to probe biomolecular structure, function, and dynamics. Nat. Methods 2019, 16, 333−340. (22) Ycas, P. D.; Wagner, N.; Olsen, N. M.; Fu, R.; Pomerantz, W. C. K. 2-Fluorotyrosine is a valuable but understudied amino acid for protein-observed 19F NMR. J. Biomol. NMR 2020, 74, 61−69. (23) Wang, X.; Zhao, W.; Al-Abdul-Wahid, S.; Lu, Y.; Cheng, T.; Madsen, J. J.; Ye, L. Trifluorinated Keto-Enol Tautomeric Switch in Probing Domain Rotation of a G Protein-Coupled Receptor. Bioconjugate Chem. 2021, 32, 99−105. (24) de Dios, A. C.; Oldfield, E. Evaluating 19F Chemical Shielding in Fluorobenzenes: Implications for Chemical Shifts in Proteins. J. Am. Chem. Soc. 1994, 116, 7453−7454. (25) Feeney, J.; McCormick, J. E.; Bauer, C. J.; Birdsall, B.; Moody, C. M.; Starkmann, B. A.; Young, D. W.; Francis, P.; Havlin, R. H.; 19F Nuclear Magnetic Resonance Arnold, W. D.; Oldfield, E. Chemical Shifts of Fluorine Containing Aliphatic Amino Acids in Proteins: Studies on Lactobacillus casei Dihydrofolate Reductase Containing (2S,4S)-5-Fluoroleucine. J. Am. Chem. Soc. 1996, 118, 8700−8706. (26) Beguin, C. G.; Dupeyre, R. Nuclear spin relaxation in benzyl fluoride. I. 2H relaxation for internal rotational barrier determination and 1H and 19F intra- and intermolecular relaxation in pure benzyl fluoride. J. Magn. Reson. 1981, 44, 294−313. (27) Grage, S. L.; Dürr, U. H. N.; Afonin, S.; Mikhailiuk, P. K.; Komarov, I. V.; Ulrich, A. S. Solid state 19F NMR parameters of fluorine-labeled amino acids. Part II: aliphatic substituents. J. Magn. Reson. 2008, 191, 16−23. (28) Drew, D.; Boudker, O. Shared Molecular Mechanisms of Membrane Transporters. Annu. Rev. Biochem. 2016, 85, 543−572. (29) Yernool, D.; Boudker, O.; Folta-Stogniew, E.; Gouaux, E. Trimeric Subunit Stoichiometry of the Glutamate Transporters from Bacillus caldotenax and Bacillus stearothermophilus. Biochemistry 2003, 42, 12981−12988. (30) Ruan, Y.; Miyagi, A.; Wang, X.; Chami, M.; Boudker, O.; Scheuring, S. Direct visualization of glutamate transporter elevator mechanism by high-speed AFM. Proc. Natl. Acad. Sci. U.S.A. 2017, 114, 1584−1588. (31) Akyuz, N.; Georgieva, E. R.; Zhou, Z.; Stolzenberg, S.; Cuendet, M. A.; Khelashvili, G.; Altman, R. B.; Terry, D. S.; Freed, J. H.; Weinstein, H.; Boudker, O.; Blanchard, S. C. Transport domain unlocking sets the uptake rate of an aspartate transporter. Nature 2015, 518, 68−73. (32) Huysmans, G. H. M.; Ciftci, D.; Wang, X.; Blanchard, S. C.; Boudker, O. The high-energy transition state of the glutamate transporter homologue GltPh. EMBO J. 2021, 40, No. e105415. (33) Kenyon, G. L.; Bruice, T. W.Novel Sulfhydryl Reagents. In Methods in Enzymology; Elsevier, 1977; Vol. 47, pp 407−430. (34) Wang, X.; Boudker, O. Large domain movements through the lipid bilayer mediate substrate release and inhibition of glutamate transporters. eLife 2020, 9, No. e58417. (35) Erkens, G. B.; Hänelt, I.; Goudsmits, J. M. H.; Slotboom, D. J.; van Oijen, A. M. Unsynchronised subunit motion in single trimeric sodium-coupled aspartate transporters. Nature 2013, 502, 119−123. (36) Yernool, D.; Boudker, O.; Jin, Y.; Gouaux, E. Structure of a glutamate transporter homologue from Pyrococcus horikoshii. Nature 2004, 431, 811−818. 8591 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592 Journal of the American Chemical Society pubs.acs.org/JACS Article (37) Reddy, K. D.; Ciftci, D.; Scopelliti, A.; Boudker, O. The archaeal glutamate transporter homologue GltPh shows heteroge- neous substrate binding. J. Gen. Physiol. 2021, 154, No. e202213131. (38) Geraets, J. A.; Pothula, K. R.; Schröder, G. F. Integrating cryo- EM and NMR data. Curr. Opin. Struct. Biol. 2020, 61, 173−181. (39) Kwon, D. H.; Zhang, F.; Suo, Y.; Bouvette, J.; Borgnia, M. J.; Lee, S.-Y. Heat-dependent opening of TRPV1 in the presence of capsaicin. Nat. Struct. Mol. Biol. 2021, 28, 554−563. (40) Gao, J.; Liang, E.; Ma, R.; Li, F.; Liu, Y.; Liu, J.; Jiang, L.; Li, C.; Dai, H.; Wu, J.; Su, X.; He, W.; Ruan, K. Fluorine Pseudocontact Shifts Used for Characterizing the Protein-Ligand Interaction Mode in the Limit of NMR Intermediate Exchange. Angew. Chem. Int., Ed. 2017, 56, 12982−12986. (41) Huang, S. K.; Pandey, A.; Tran, D. P.; Villanueva, N. L.; Kitao, A.; Sunahara, R. K.; Sljoka, A.; Prosser, R. S. Delineating the conformational landscape of the adenosine A2A receptor during G protein coupling. Cell 2021, 184, 1884−1894. (42) Boudker, O.; Ryan, R. M.; Yernool, D.; Shimamoto, K.; Gouaux, E. Coupling substrate and ion binding to extracellular gate of a sodium-dependent aspartate transporter. Nature 2007, 445, 387− 393. (43) Suloway, C.; Pulokas, J.; Fellmann, D.; Cheng, A.; Guerra, F.; Quispe, J.; Stagg, S.; Potter, C. S.; Carragher, B. Automated molecular microscopy: the new Leginon system. J. Struct. Biol. 2005, 151, 41− 60. (44) Zheng, S. Q.; Palovcak, E.; Armache, J. P.; Verba, K. A.; Cheng, Y.; Agard, D. A. MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat. Methods 2017, 14, 331−332. (45) Rohou, A.; Grigorieff, N. CTFFIND4: Fast and accurate defocus estimation from electron micrographs. J. Struct. Biol. 2015, 192, 216−221. (46) Zivanov, J.; Nakane, T.; Forsberg, B. O.; Kimanius, D.; Hagen, W. J. H.; Lindahl, E.; Scheres, S. H. W. New tools for automated high- resolution cryo-EM structure determination in RELION-3. eLife 2018, 7, No. e42166. (47) Punjani, A.; Rubinstein, J. L.; Fleet, D. J.; Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 2017, 14, 290−296. (48) Asarnow, D.; Palovcak, E.; Cheng, Y. UCSF pyem v0.5; Zenodo. 2019. (49) Punjani, A.; Zhang, H.; Fleet, D. J. Non-uniform refinement: adaptive regularization improves single-particle cryo-EM reconstruc- tion. Nat. Methods 2020, 17, 1214−1221. (50) Zivanov, J.; Nakane, T.; Scheres, S. H. W. A Bayesian approach to beam-induced motion correction in cryo-EM single-particle analysis. IUCrJ 2019, 6, 5−17. (51) Pettersen, E. F.; Goddard, T. D.; Huang, C. C.; Couch, G. S.; Greenblatt, D. M.; Meng, E. C.; Ferrin, T. E. UCSF Chimera-A visualization system for exploratory research and analysis. J. Comput. Chem. 2004, 25, 1605−1612. (52) Afonine, P. V.; Grosse-Kunstleve, R. W.; Chen, V. B.; Headd, J. J.; Moriarty, N. W.; Richardson, J. S.; Richardson, D. C.; Urzhumtsev, A.; Zwart, P. H.; Adams, P. D. Phenix.model_vs_data: a high-level tool for the calculation of crystallographic model and data statistics. J. Appl. Crystallogr. 2010, 43, 669−676. (53) Emsley, P.; Lohkamp, B.; Scott, W. G.; Cowtan, K. Features and development of Coot. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 486−501. (54) Chen, V. B.; Arendall, W. B., 3rd; Headd, J. J.; Keedy, D. A.; Immormino, R. M.; Kapral, G. J.; Murray, L. W.; Richardson, J. S.; Richardson, D. C. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr., Sect. D: Biol. Crystallogr. 2010, 66, 12−21. 8592 https://doi.org/10.1021/jacs.3c01003 J. Am. Chem. Soc. 2023, 145, 8583−8592
10.1073_pnas.2306564120	RESEARCH ARTICLE \| IMMUNOLOGY AND INFLAMMATION OPEN ACCESS Contribution of the IGCR1 regulatory element and the 3′Igh CTCF- binding elements to regulation of Igh V(D)J recombination Zhuoyi Lianga,b,c,1,2, Lijuan Zhaoa,b,c,1, Adam Yongxin Yea,b,c,1, Sherry G. Lina,b,c,1, Yiwen Zhanga,b,c, Chunguang Guoa,b,c, Hai- Qiang Daia,b,c,3, Zhaoqing Baa,b,c,2,4 and Frederick W. Alta,b,c,2 , Contributed by Frederick W. Alt; received April 21, 2023; accepted May 12, 2023; reviewed by Cornelis Murre and Andre Nussenzweig Immunoglobulin heavy chain variable region exons are assembled in progenitor- B cells, from VH, D, and JH gene segments located in separate clusters across the Igh locus. RAG endonuclease initiates V(D)J recombination from a JH- based recombination center (RC). Cohesin- mediated extrusion of upstream chromatin past RC- bound RAG presents Ds for joining to JHs to form a DJH- RC. Igh has a provocative number and organization of CTCF- binding elements (CBEs) that can impede loop extrusion. Thus, Igh has two divergently oriented CBEs (CBE1 and CBE2) in the IGCR1 element between the VH and D/JH domains, over 100 CBEs across the VH domain convergent to CBE1, and 10 clustered 3′Igh- CBEs convergent to CBE2 and VH CBEs. IGCR1 CBEs seg- regate D/JH and VH domains by impeding loop extrusion- mediated RAG- scanning. Downregulation of WAPL, a cohesin unloader, in progenitor- B cells neutralizes CBEs, allowing DJH- RC- bound RAG to scan the VH domain and perform VH- to- DJH rearrange- ments. To elucidate potential roles of IGCR1- based CBEs and 3′Igh- CBEs in regulating RAG- scanning and elucidate the mechanism of the ordered transition from D- to- JH to VH- to- DJH recombination, we tested effects of inverting and/or deleting IGCR1 or 3′Igh- CBEs in mice and/or progenitor- B cell lines. These studies revealed that nor- mal IGCR1 CBE orientation augments RAG- scanning impediment activity and suggest that 3′Igh- CBEs reinforce ability of the RC to function as a dynamic loop extrusion impediment to promote optimal RAG scanning activity. Finally, our findings indicate that ordered V(D)J recombination can be explained by a gradual WAPL downregulation mechanism in progenitor- B cells as opposed to a strict developmental switch. V(D)J recombination \| CTCF- binding elements (CBEs) \| CTCF \| antibody repertoires \| chromatin 3D structure Variable region exons that encode antigen- binding sites of antibodies are assembled in progenitor (“pro”)- B cells from germline VH, D, and JH gene segments (1). V(D)J recom- bination is initiated by RAG1/2 endonuclease (RAG) (2). RAG introduces DNA double- stranded breaks (DSBs) between VH, D, and JH coding segments and flanking recombination signal sequences (RSSs) (2). RSSs comprise a conserved heptamer, a spacer of 12 or 23 base pairs, and an AT- rich nonamer. To robustly initiate V(D)J recombination, RAG must bind and cleave a pairs of gene segments flanked by RSSs with complementary 12- and 23- bp spacers (termed 12- RSSs and 23- RSSs, respectively) (2). After RAG cleav- age, 12/23- RSS matched gene segment ends and, separately, their corresponding RSS ends are fused by the classical nonhomologous end- joining (3). The mouse IgH locus (Igh) spans 2.7 megabases (Mbs) on chromosome 12 with over 100 VHs interspersed within a several Mb distal portion (1). This VH domain lies 100 kb upstream of a 50- kb region containing up to 13 Ds, with 4 JHs embedded within a 2- kb region just downstream of the most proximal D (DQ52) (1). RAG initiates Igh V(D)J recombination from a recombination center (RC) formed within highly transcribed chromatin that spans DQ52, the four JHs, and the intronic enhancer (iEμ) (4, 5). The VHs and JHs have 23RSSs and cannot be directly joined. Ds are flanked on either side by 12RSSs, allowing them to join to a downstream JH and an upstream VH to form a V(D)J exon (3). V(D)J recombination is developmentally ordered with Ds joined to a JH to form a DJH RC, after which VHs are joined to the upstream D12RSS of the DJH RC (3). The CTCF chromatin looping factor binds target DNA sequences, termed CTCF- binding elements (“CBEs” or “CTCF sites”), in an orientation- specific manner (6, 7). In this regard, the numerous genomic CBEs, when in adjacent regions, can occur in the same, divergent, or in convergent orientations (8). The cohesin complex mediates extrusion of chromatin loops genome- wide, forming contact loops when extrusion in each direction reaches Significance To counteract diverse pathogens, vertebrates evolved adaptive immunity to generate diverse antibody repertoires through a B lymphocyte- specific somatic gene rearrangement process termed V(D)J recombination. Tight regulation of the V(D)J recombination process is vital to generating antibody diversity and preventing off- target activities that can predispose the oncogenic translocations. Recent studies have demonstrated V(D)J rearrangement is driven by cohesin- mediated chromatin loop extrusion, a process that establishes genomic loop domains by extruding chromatin, predominantly, between convergently oriented CTCF looping factor- binding elements (CBEs). By deleting and inverting CBEs within a critical antibody heavy chain gene locus developmental control region and a loop extrusion chromatin- anchor at the downstream end of this locus, we reveal how these elements developmentally contribute to generation of diverse antibody repertoires. 1Z.L., L.Z., A.Y.Y., and S.G.L. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected], bazhaoqing@nibs. ac.cn, or [email protected]. 3Present address: Center for Excellence in Molecular Cell Science, Chinese Academy of Sciences, Shanghai 200031, China. 4Present address: National Sciences, Beijing 102206, China. Institute of Biological This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2306564120/- /DCSupplemental. Published June 20, 2023. PNAS 2023 Vol. 120 No. 26 e2306564120 https://doi.org/10.1073/pnas.2306564120 1 of 10 CTCF- bound CBEs that impede extrusion (9, 10). Such CBE- anchored chromatin loops occur most dominantly between CBEs in convergent orientation (8, 11), an orientation that forms the most stable anchor (9, 10). Convergent CBE orientation has been implicated in mediating physiological functions (12–17). However, CTCF- bound CBEs can impede loop extrusion regard- less of orientation (18), and non- CBE- based impediments, for example highly transcribed chromatin, can impede extrusion and contribute to developmental or tissue- specific regulation of loop extrusion (18–24). The Igh contains numerous CBEs with remark- able relative orientation. The IGCR1 element, which lies just upstream of the most distal D segment (DFL16.1) has two diver- gently oriented CBEs, namely upstream CBE1 and downstream CBE2 (25). The long upstream VH- containing domain has over 100 CBEs, with most lying in convergent orientation to IGCR1 CBE1 (26). Ten consecutive CBEs, termed 3′Igh CBEs (27, 28) lie just downstream of Igh in convergent orientation to IGCR1 CBE2 and VH domain CBEs (25, 26). This organization has been proposed to have various functions in regulating V(D)J recombi- nation (25, 29–31). Cohesin- mediated loop extrusion provides the mechanistic underpinnings for RC- bound RAG to scan chromatin across the Igh locus for substrates (18, 19, 22, 24, 32). In pro- B cells, RC- bound RAG initiates scanning upon binding a JH- 23RSS into one of its two active sites (1, 18, 24). For this scanning process, the active RC, which lacks CBEs, serves as a transcription- based dynamic down- stream loop extrusion anchor, while IGCR1 serves as a CBE- based upstream anchor that terminates RAG scanning, preventing scan- ning from entering the VH- containing domain (18, 24, 33). The upstream orientation of the RAG- bound JH programs RAG scan- ning of upstream D- containing chromatin extruded past the RC (24). During RAG scanning of the D locus, only downstream D- 12RSSs, convergently oriented to the JH23RSSs, are used, result- ing in D- to- JH joining that deletes all sequences between the par- ticipating D and JH (24). Predominant utilization of RSSs, or cryptic RSSs, in convergent orientation to the initiating RC RSS is a mech- anistic property of the linear RAG scanning process (22, 32). While VHs- 23RSSs are compatible for joining to D12- RSSs, the IGCR1 impedes access of VHs to the D locus during the D- to- JH rearrangement and, thereby, enforces ordered D- to- JH rearrangement before VH- to- DJH rearrangement in pro- B cells (18, 25, 32). Mutational inactivation of IGCR1 CBEs allows RC- bound RAG to scan directly into the proximal VH locus, causing the most proximal functional VH5- 2 to robustly rearrange and dominate the VH repertoire, with little rearrangement of more distal VHs (18, 25, 32). The mechanism by which VH5- 2 dominates rearrangement when IGCR1 CBEs are inactivated is based on its RSS- associated CBE that impedes extrusion past the RC, making it accessible for rearrangement (18). Indeed, dozens of the proximal VHs have RSS- associated CBEs. Thus, while deletion of the VH5- 2 CBE results in a 50- fold reduction of VH5- 2 rearrangement along with greatly reduced RC interaction, the next upstream VH becomes dominantly rearranged based on its RSS- associated CBE (18). Because dozens of the D- proximal VHs have RSS- associated CBEs, this portion of the VH locus is a major barrier to RAG scanning to further upstream VHs when IGCR1 CBEs are inactivated (18). Fluorescent in situ hybridization and chromosome conforma- tion capture (3C)- based studies revealed the VH domain to undergo large- scale contraction in pro- B cells (34–42). VH locus contraction was proposed to bring distal VHs into proximity with the DJH RC for recombinational access (43). Recent studies revealed that locus contraction is mediated by loop extrusion, which extends across the several Mb VH locus due to approxi- mately fourfold downregulation of the WAPL cohesin unloading factor in pro- B cells (44). In this context, depletion of WAPL in nonlymphoid cells extends genome- wide loop extrusion by increasing cohesin density, allowing it to bypass CBEs and poten- tially other impediments (45–47). WAPL downregulation in pro- B cells reduces loop extrusion impediment activity of IGCR1 CBEs, proximal VH- associated CBEs, and others, allowing RAG scanning from a DJH RC to extend linearly across the VH domain (22). Abelson murine leukemia virus- transformed pro- B cell lines (“v- Abl lines”) can be viably arrested in the G1- cell cycle phase in which V(D)J recombination occurs (48). Introduction of RAG into G1- arrested RAG- deficient v- Abl lines activates robust RAG scanning across the D domain (19, 22, 24). However, scanning is impeded at IGCR1, and there is little VH- to- DJH joining (19, 22, 24). Inactivation of IGCR1 CBEs leads to dominant rear- rangement of VH5- 2 in v- Abl lines due to its RSS- associated CBE (18). In this regard, v- Abl lines have high WAPL levels and do not neutralize VH locus CBEs (19, 22). Neutralization of CBE imped- iments by depletion of CTCF or WAPL extends cohesin loop extrusion and RAG- scanning past IGCR1 and proximal VHs to the most distal VHs (19, 22). Thus, v- Abl pro- B cell studies have provided substantial mechanistic insights into VH- locus contrac- tion and the RAG chromatin scanning process (1). Despite recent advances in elucidating functions of Igh CBEs during V(D)J recombination, many questions remain. While an early study indicated that IGCR1 CBEs act synergistically to seg- regate the VH domain from the D/JH domain (49), the question of whether orientation of these CBEs is critical to their function remained open. Likewise, 3′Igh CBEs confine Igh class switch recombination (CSR) activity to the Igh (50). However, their roles in V(D)J recombination have not been resolved (1). Finally, a long- standing question is how the developmental transition from D- to- JH joining to VH- to- DJH joining is regulated. While a role for the DJH intermediate in signaling the transition (51, 52) has been considered, such a transition could, in theory, be mediated by gradual WAPL- downregulation during pro- B cell development (1). We now describe studies that address these questions. Results Role of IGCR1 CBEs in D- to- JH and Proximal VH- to- DJH Rearrange- ments. To gain insight into factors that determine primary bone marrow (BM) pro- B cell VH repertoires, we assessed effects of IGCR1 CBE1 and CBE2 inactivation via high- resolution HTGTS- V(D) J- Seq (24). For these assays, we purified B220+CD43highIgM− pro- B cells from BM of wild- type (WT) 129SV controls and previously generated IGCR1/CBE1&2−/− mice (25). We performed HTGTS- V(D)J- Seq on DNA from these samples using a JH4 bait primer to compare their levels of D- to- JH and VH- to- DJH rearrangements (Fig. 1A and SI Appendix, Table S1) (32). A JH4 bait primer was used for HTGTS- V(D)J- Seq analyses in this study to eliminate potential confounding effects of rearrangements within extrachromosomal deletion products (24). For these analyses, individual peaks found in HTGTS libraries can be normalized as a fraction of total HTGTS reads for a given experiment, which reveals absolute V(D)J levels of each rearranging gene segment. Alternatively, peaks can be normalized as a fraction of total recovered junctions across a given locus or section of a locus to reveal the relative utilization of a given gene segments to each other as a percentage of all junctions in the analyzed region (SI Appendix, Fig. S1). Such analyses are useful for examining effects of potential regulatory element mutations. For example, finding a decrease in total reads between two samples in the absence of differences in the junction profile, reflects decreased RAG or RC activity without changes in long range scanning patterns (22). 2 of 10 https://doi.org/10.1073/pnas.2306564120 pnas.org Fig. 1. Role of IGCR1/CBE1 and CBE2 in D- to- JH and proximal VH- to- DJH rearrangements. (A) Schematic of the murine Igh locus showing VHs, Ds, JHs, CHs, and IGCR1. The red arrow indicates the JH4 coding end (CE) bait primer. (B–E) Utilization of VHs across the entire Igh locus in WT (B and C) and CBE1&2−/− (D and E) pro- B cells. The VHDJH and DJH junctions are shown in the Insets (n = 3 mice, mean±SEM; all HTGTS libraries are normalized to 78,091 total reads; see SI Appendix, Table S1). (F and G) Proximal VH usage in VHDJH junctions and D usage DJH junctions of WT (F) and CBE1&2−/− (G) pro- B cells (n = 3 mice, mean ± SEM). (H and I) Relative percentage of upstream VHs beyond the five most proximal VHs normalized to the indicated VHDJH junction number in WT (H) and CBE1&2−/− (I) pro- B cells. Upstream VHs junctions are extracted from the data of B and D (n = 3 mice, mean ± SEM). WAPL downregulation in mouse pro- B cells (22, 44) allows RAG to scan the entire VH locus, which results in highly repro- ducible utilization of the different VHs (Fig. 1 B and C). However, in WT pro- B cells, VH5- 2 (“VH81X”) and three immediately upstream VHs with RSS- associated CBEs are much more highly utilized than any VHs further upstream (Fig. 1 B and C). In such steady- state BM pro- B cell populations, DJH rearranged alleles are more prevalent than VH(D)JH rearranged alleles (Fig. 1 B, Inset), likely reflecting steady- state distributions within pro- B cells enter- ing the compartment and successively generating DJH and VH- to- DJH rearrangements before leaving the compartment. Inactivation of IGCR1 by mutational inactivation of CBE1 and CBE2 (18, 25, 32), allows VH5- 2 to dominate the VHDJH reper- toire (Fig. 1 D and E). Moreover, the great majority of VH5- 2 rearrangements are nonproductive (SI Appendix, Fig. S2 and Table S2), consistent with selection against productive VH5- 2 rearrangements (53). This finding also confirms that dominant VH5- 2 rearrangements in pro- B cells do not result from cellular selection (25). Rearrangement frequencies of more distal VHs are dramatically decreased upon IGCR1 inactivation (Compare Fig. 1 B and C with Fig. 1 D and E; also see SI Appendix, Table S1). Strikingly, however, in CBE1&2−/− pro- B cell populations as com- pared to WT pro- B cell populations, the absolute level of VHDJH rearrangements is greatly increased with the vast majority utilizing VH5- 2, while the absolute level of DJH rearrangements and upstream VH rearrangements is, correspondingly, decreased (Fig. 1 E–G and SI Appendix, Table S1). The above findings support the notion that, in the absence of IGCR1 CBE activity in normal pro- B cells, RAG scanning contin- ues between the DJH RC and VH5- 2, which allows VH5- 2 to dom- inate VH- to- DJH rearrangements due to its robust CBE- mediated interaction with the DJH RC (18). Moreover, the finding that the increased frequency of VH- to- DJH rearrangements in the steady- state CBE1&2−/− pro- B results almost totally from increased VH5- 2 rear- rangements indicates that these dominant rearrangements occur at the D- to- JH scanning stage before sufficient WAPL downregulation neutralizes proximal VH- RSS- associated CBE impediments to allow upstream scanning. These high- resolution HTGTS- V(D)J- seq stud- ies also revealed another notable finding. Despite the dramatically reduced levels of upstream VH rearrangements in CBE1&2−/− pro- B cells, the upstream VHs beyond the several most proximal VHs, relative to each other, have rearrangement junction patterns (i.e. relative levels compared to each other), that were nearly identical to those of WT pro- B cells (Fig. 1, Compare panels H and I). Thus, VH5- 2, and to a lesser extent immediately upstream VHs, dominate initial rearrangements in the absence of IGCR1 CBE activity, and, in doing so, terminate most RAG upstream scanning. However, RAG scanning that does proceed beyond VH5- 2 continues through the remainder of the VH locus, with similar VH usage patterns as those of WT cells. Overall, these findings indicate that, while most PNAS 2023 Vol. 120 No. 26 e2306564120 https://doi.org/10.1073/pnas.2306564120 3 of 10 VH5- 2 rearrangements in CBE1&2−/− pro- B cells occur before WAPL downregulation, a small fraction of CBE1&2−/− pro- B that do not form VH5- 2 rearrangements on one or both Igh alleles undergo upstream VH- to- DJH recombination events at normal fre- quencies when WAPL- downregulation reaches appropriate levels. Role of IGCR1/CBE1 and CBE2 in Regulating RAG Scanning into the VH Locus. To further assess potential mechanisms by which IGCR1 CBE impediment activity is modulated to promote RAG scanning of the VH domain, we applied the highly sensitive 3C- HTGTS chromatin interaction assay to explore interactions of the RC- based iEμ enhancer element with upstream and downstream Igh locus chromatin domains in WT rag2−/− and IGCR1/CBE1&2−/−rag2−/− cultured pro- B cells derived from the corresponding mouse lines. RAG- deficient cells must be used for such assays to eliminate confounding effects of V(D)J recombination events on such interactions (18, 19, 22, 24). These studies revealed that the iEμ/RC interacts robustly with 15 highly focused regions across the 2.4 Mb VH locus in WT rag2−/−pro- B cells (Fig. 2A and SI Appendix, Fig. S3) (19, 22). Among the most robust of these iEμ/RC interacting peaks are peaks associated with robustly transcribed PAX5- activated intergenic repeat (PAIR) elements (38, 54, 55) in the J558/3609, and J558 VH- containing regions in the distal portion of the VH locus (Fig. 2A and SI Appendix, Fig. S3; Peaks 1, 4, 6, 8–10). RC interactions with transcribed PAIR element- associated sequences are considered a hallmark of loop extrusion- mediated VH locus contraction in pro- B cells (38). In this regard, locus contraction results from an approximately fourfold developmental downregulation of WAPL in pro- B cells (44), which at least partially neutralizes IGCR1- CBEs, proximal VH CBEs, and likely other VH locus CBEs, and potentially transcription- associated impediments to RC- based RAG linear scanning (22, 44). Although VH5- 2 and proximal VHs are the most dominantly utilized VHs in normal pro- B cells (Fig. 1B), they show only low- level interactions with the RC in RAG2- deficient WT pro- B cells at steady state (Fig. 2 A and B), consistent with most CBE- based interactions being diminished by WAPL- downregulation in a large fraction of the pro- B cells (22). In this regard, some WT rag2−/− pro- B cells retain interactions between the RC and IGCR1(Fig. 2B, upper track), indicating that some cells in the population have not fully down- regulated WAPL and/ or that IGCR1 impediment activity is not completely neutralized by physiological levels of WAPL downregulation (Fig. 2B; upper track). Notably, CBE1&2−/−rag2−/− cultured pro- B cells exhibit greatly increased RC interaction with proximal VH5- 2 and the 3 proximal VHs immediately upstream, which, based on prior studies (18), is dependent on their RSS- associated CBEs (Fig. 2 A and B and SI Appendix, Fig. S3; Peaks 16–18). Strikingly, nearly all major further upstream interaction peaks were also present in chromatin from CBE1&2−/−rag2−/− pro- B cells, mostly at similar relative levels to those in WT rag2−/− pro- B cells (Fig. 2A). Given that the chro- matin interactions are investigated in RAG- deficient cells and upstream interactions are not impacted by proximal VH- to DJH recombination events, it is not unexpected that a large number of IGCR1/CBE- mutated pro- B cells would have robust interactions between the RC to the upstream VHs after developmental WAPL downregulation. As previously described (18), the RC also robustly interacts downstream with the transcribed enhancer- like element between Cγ1 and Cγ2b (38) and with the 3′Igh CBEs in pro- B cells (Fig. 2B, upper track); these interactions were not altered by IGCR1 CBE1 and CBE2 deletion (Fig. 2B, lower track). Influence of IGCR1 CBEs and Their Orientation on VH- Utilization in pro- B Cells. We have previously generated CBE1−/− and CBE2−/− mice (49). To assess whether orientation of IGCR1/CBEs is critical for Igh V(D)J recombination control, we replaced CBE1 or CBE2 with their inverted sequences to generate a CBE1inv or CBE2inv allele in mouse 129SV ES cells (SI Appendix, Fig. S4 A and B) (25, 49) (Methods). We employed HTGTS- V(D)J- Seq to assay JH4- based utilization of the various VHs in pro- B populations harboring IGCR1 CBE deletion or inversion mutations (Fig. 3). CBE1−/− pro- B cell populations have markedly increased VH5- 2 rearrangements and markedly decreased rearrangements of more distal VHs, with the degree of increases and decreases modestly, but significantly, less than those observed for CBE1&2−/− pro- B cells (Fig. 3 A and E and Fig. 2. Role of IGCR1/CBE1 and CBE2 in regulating RAG scanning into the VH locus. (A) 3C- HTGTS signal counts of all VHs in WT (red) and CBE1&2−/− (blue) RAG2- deficient pro- B cells baiting from iEμ/RC (). Each library was normalized to 160,314 total junctions (n = 3 mice, mean ± SEM). 18 peaks across VHs region are called by MACS2 pipeline and highlighted in gray (peaks 1–9, 12–15 are called in both conditions), orange (peaks 10–11 are called only in WT), green (peaks 16–18 are called only in CBE1&2−/−) (SI Appendix, Fig. S3). (B) Zoom- in 3C- HTGTS profiles of Igh locus from proximal VHS to 3′Igh CBEs. 4 of 10 https://doi.org/10.1073/pnas.2306564120 pnas.org SI Appendix, Table S1). In contrast, CBE2−/− pro- B cells had very modestly increased VH5- 2 rearrangements (2.5- fold; Fig. 3 B and E) and normal patterns of distal VH rearrangements (Fig. 3 B and E and SI Appendix, Table S1). These findings indicate that CBE1 and CBE2 cooperatively provide the full impact of IGCR1 scanning impediment activities and unequivocally demonstrate that CBE1 plays a much more dominant role. CBE1inv/inv BM pro- B cells also have significantly increased levels of proximal VH5- 2 utilization relative to WT; but to a much lower extent than CBE1−/− pro- B cells, indicating that ability of CBE1 to impede RAG scanning is dampened, but not abrogated, when inverted (Fig. 3 C and E and SI Appendix, Table S1). In contrast, CBE2inv/inv pro- B cells were very similar to WT pro- B cells with respect to utilization of VH5- 2 and upstream VHs (Fig. 3 D and E and SI Appendix, Table S1). Consistent with our findings for IGCR1/CBE1&2−/− pro- B cells (Fig. 1), the rearrangement pattern of upstream VHs, relative to each other, was not markedly impacted by CBE1 or CBE2 deletions or inversions (SI Appendix, Fig. S5 A–F and Table S1). We performed 3C- HTGTS on RAG- deficient IGCR1/WT, IGCR1/CBE1&2−/−, IGCR1/CBE1inv/inv and IGCR1/CBE2inv/ inv v- Abl lines with bait primers to the iEμ in the RC (SI Appendix, Fig. S6A) and VH5- 2 (SI Appendix, Fig. S6B) locales. In CBE1−/− pro- B cell populations, we observed robust, albeit somewhat diminished, interactions between the iEμ/RC bait and the prox- imal VH- CBEs compared to interactions in CBE1&2−/− pro- B cells (SI Appendix, Fig. S6A, CBE1−/− vs. CBE1&2−/−). In CBE2−/− pro- B cells, we observed a modest increase in the inter- actions between the iEμ/RC bait and the proximal VH- CBEs (SI Appendix, Fig. S6A, CBE2−/− vs. CBE1&2−/−). These findings indicate that IGCR1 CBEs play a cooperative role in impeding loop extrusion- mediated proximal VH- CBEs and RC interactions and, again, that CBE1 has a more dominant role. In contrast, when CBE1 or CBE2 was inverted, interactions between proximal VH- CBEs and iEμ/RC showed only very modest changes com- pared to those when IGCR1 CBEs are in normal orientation (SI Appendix, Fig. S6 A and B). Fig. 3. The mutation of IGCR1/CBEs alters VHs utilization in pro- B cells. (A–D) Each panel shows the utilization of VHs across the entire Igh locus in indicated IGCR1/CBE- mutated pro- B cells. The VHDJH and DJH junctions are shown in Insets (n = 3 mice, mean ± SEM; All HTGTS libraries are normalized to 78,091 total reads; see SI Appendix, Table S1). (E) VHs usage in WT and indicated IGCR1/CBE- mutated pro- B cells (n = 3 mice, mean ± SEM). PNAS 2023 Vol. 120 No. 26 e2306564120 https://doi.org/10.1073/pnas.2306564120 5 of 10 Influence of 3′Igh CBEs on D- to- JH and VH- to- DJH Rearrangement Patterns and Levels. As mentioned above, v- Abl pro- B cell lines provide a very useful system for studies of RAG chromatin scanning process (1). In particular, the ability to study RAG scanning in a system in which WAPL can be completely depleted allows experiments to test the proposal that 3′Igh CBEs reinforce the loop anchor provided by the RC when juxtaposed via loop extrusion (1). To investigate potential contributions of 3′Igh CBEs on RAG scanning, we deleted all 3′CBEs in RAG1- deficient C57BL/6 WAPL- degron v- Abl cells that contain a single copy of the Igh locus (22). For analyses, WT or 3′CBE- deleted (3′CBE−) WAPL- degron v- Abl cells were arrested in G1 and then left untreated or treated with Dox plus IAA to degrade WAPL (22). Subsequent introduction of RAG into untreated WT WAPL- degron v- Abl cells activated V(D)J recombination leading to robust D- to- JH recombination, very low- level VH5- 2 to DJH recombination, and extremely low- level upstream VH- to- DJH recombination (Fig. 4A and SI Appendix, Table S3), as v- Abl lines have high WAPL levels and RAG scanning is impeded at IGCR1 (22). Dox plus IAA treatment completely depletes WAPL in these G1- arrested v- Abl lines, with little effect on viability (22). Introduction of RAG into WAPL- depleted WT v- Abl cells activated D- to- JH rearrangement, but the level of DJH rearrangements was reduced 6.5- fold compared to that of untreated cells (Fig. 4I and SI Appendix, Table S3). This reduction was associated with dramatically decreased distal DFL16.1- JH and, to a lesser degree, most other DJH absolute rearrangement levels (Fig. 4 M and N). Notably, however, DQ52 rearrangement levels showed little change (Fig. 4M). WAPL- depletion also led to RAG- scanning and utilization of VHs across the 2.4 Mb VH locus (Fig. 4 A and B and SI Appendix, Table S3) as expected (22). The absolute level of VH DJH rear- rangements in WAPL- depleted WT v- Abl cells was similar to that of the low level of proximal VH rearrangements in untreated cells; but represented a 5.3- fold increase with respect to their fraction of DJH rearrangements (Fig. 4 I and J and Discussion). In both untreated and WAPL- depleted 3′CBE− lines, we observed a fur- ther 25% decrease in DJH rearrangement levels from their baseline levels. We observed a similar 25% decrease in VHDJH rearrange- ments in untreated 3′CBE− lines and a nearly 50% decrease in WAPL- depleted 3′CBE lines (compare Fig. 4 panels B and D, Fig. 4 I–L). Despite the further reduction of VHDJH rearrange- ment levels in WAPL- depleted 3′CBE− v- Abl cells, their relative VH usage pattern was very similar to that of WAPL- depleted WT lines (compare Fig. 4, panel F and H). We performed 3C- HTGTS in RAG- deficient WT and 3′CBE− v- Abl cells baiting from iEμ/RC (Fig. 5 A and B and SI Appendix, Figs. S7 B and C and S8). These studies revealed, as previously described (22), that complete depletion of WAPL substantially increased a number of interaction peaks in the J558/3609, J558 and middle VH regions (Fig. 5A and SI Appendix, Fig. S8 Dox+IAA vs untreated; Peaks 1–13). Many of the sequences that contribute to these peaks were highly transcribed including the well- characterized PAIR elements (Fig. 5A and SI Appendix, Fig. S8; Peaks 1–8). In contrast, peaks in the proximal 7183/Q52 region that are dominant in untreated v- Abl cells are mainly associated with proximal VH RSS- CBEs and were significantly diminished by WAPL depletion (Fig. 5A and SI Appendix, Fig. S8, Dox+IAA vs untreated; Peaks 14–16). Notably, in 3′CBE− cells, these same major interaction peaks, including those associated with proximal VHRSS- CBEs in untreated and those associated with the upstream VH regions in treated and untreated cells, remain robust and largely correspond to those in the same locations as in the WT line (Fig. 5A and SI Appendix, Fig. S8, 3′CBE− vs. WT). While the intensity of some peaks in the untreated or WAPL- depleted 3′CBE− v- Abl cells were somewhat diminished compared those of the untreated or WAPL- depleted WT v- Abl cells, when viewed at high resolution they are clearly still associated with same transcriptional or CBE impediments (Fig. 5A and SI Appendix, Fig. S8). Upon the deletion of 3′Igh CBEs, multiple CBEs downstream of the 3′Igh CBEs appear to gain robust interactions with the iEμ/RC (Fig. 5B). Finally, WAPL- depletion also substantially diminished interactions of IGCR1 CBE1 with both CBE- based (proximal VHs and 3′Igh CBEs and transcription- based (RC and γ1- γ2b enhancer) loop extrusion impediments (Fig. 5C). Discussion HTGTS- V(D)J- Seq analyses provided a deep analysis of VH rep- ertoires in primary WT and IGCR1/CBE1&2−/− pro- B cells (Fig. 1). In addition, 3C- HTGTS- Seq analyses of iEμ/RC interac- tions across the Igh locus in RAG2- deficient primary WT and IGCR1/CBE1&2−/− pro- B cells complemented the HTGTS- V(D) J- Seq findings to reveal a likely mechanism by which the ordered transition from D- to- JH versus VH- to- DJH rearrangement is regu- lated (Fig. 2). Our overall findings indicate that this developmental transition can be explained in the context of gradual WAPL down- regulation in pro- B cells (Outlined in SI Appendix, Fig. S9). Our findings indicate that robust D- to- JH rearrangements occur in WT pro- B cells before WAPL is sufficiently down- regulated to allow scanning to pass IGCR1 or proximal VH- associated CBEs. In CBE1&2−/− pro- B cells, robust iEμ/RC interactions with proximal VHRSS- associated CBEs promote their robust rearrangements and suppress scanning to upstream VHs. Our finding that low- level rearrangements of upstream VHs in CBE1&2−/− pro- B cells have normal RAG- scanning patterns is consistent with these rearrange- ments occurring after WAPL- downregulation in the cells that have not formed proximal VHDJH rearrangements on both alleles. In CBE1&2−/−rag2−/− pro- B cells, lack of V(D)J recombination upon WAPL downregulation allows the iEμ/RC to scan into the upstream VH domain where it reaches normal impediments in a substantial fraction of the cells. In this context, the relatively robust contribu- tion of VH5- 2 and immediately upstream VHs to the WT pro- B repertoire indicates that these VHs are dominantly utilized in normal pro- B cells until WAPL levels are sufficiently down- regulated Finally, in support of this model, proximal VHs are poorly utilized in v- Abl cells in which RAG is introduced after complete WAPL- depletion (22) (Fig. 4). Our current studies confirm unequivocally that CBE1 and CBE2 function synergistically to provide the full RAG scanning impediment activity of IGCR1 and that CBE1 provides the major portion of this activity (Fig. 3). We have previously shown that the linear scanning process from the JH or DJH recombination centers is strongly impeded by CTCF bound IGCR1 CBEs until they are neutralized by WAPL downregulation (19, 22). Moreover, based on CTCF depletion studies, we found CBE1 retains bound CTCF under conditions in which CBE2 completely loses bound CTCF, which implies CBE1 more strongly binds CTCF (19). Such stronger CTCF- binding activity may form a basis for the stronger RAG- scanning impediment activity of CBE1 versus CBE2. Finally, reminiscent of the effects of VH5- 2 RSS- associated CBE inversion as compared to complete inactivation on proximal VH5- 2 rearrangements (18), normal orientation of IGCR1 CBEs is required to provide physiological levels of RAG- scanning impediment activity, but both retain substantial activity when inverted (Fig. 3). 6 of 10 https://doi.org/10.1073/pnas.2306564120 pnas.org Fig. 4. Role of 3′Igh CBEs in RC activity during loop extrusion. (A–D) Utilization of VHs across the entire Igh locus in WT (A and B) and 3′CBE− (C and D) v- Abl cells with or without Dox/IAA treatments. The VHDJH and DJH junctions are shown in Insets. (n = 3 repeats from three independent clones, mean ± SEM; all HTGTS libraries are normalized to 1,964,102 total reads; see SI Appendix, Table S3). (E–H) Relative percentage of VHs utilization normalized to the indicated VHDJH junction number (n = 3 repeats, mean ± SEM; percentages are plotted from the data of Fig. 4 A–D). (I and J) Absolute level of VHDJH and DJH rearrangements in WT and 3′CBE− lines (n = 3 repeats, mean ± SEM; t test, P < 0.01, ). (K and L) Relative percentages of VHDJH and DJH normalized to untreated or treated WT conditions. (M and N) Absolute (M) and relative (N) D usage in DJH rearrangements in untreated and WAPL- depleted WT v- Abl cells (relative percentage was normalized to 106,936 DJH junctions in untreated or 16,565 DJH junctions in WAPL- depleted cells; see SI Appendix, Table S3). Our findings on the impact of WAPL- depletion on chromatin interactions and RAG scanning activity support a model in which 3′Igh CBEs reinforce RC activity during Igh V(D)J recombination (SI Appendix, Fig. S10A) (1). In WT v- Abl cells with high WAPL expression, the RC robustly interacts with the downstream γ1- γ2b enhancer and the 3′Igh CBEs, which, as proposed (1), could reinforce its loop- extrusion impediment activity (Fig. 5B). Complete WAPL depletion in v- Abl cells substantially diminishes downstream RC interactions (Fig. 5B). In addition, interactions of the IGCR1/CBE1 with the RC, as well as with the γ1- γ2b enhancer and 3′Igh CBEs, are also greatly diminished in WAPL- depleted v- Abl cells (Fig. 5C), consistent with WAPL- depletion diminishing transcription- based RC impediment activity. In contrast, in WT pro- B cells, in which WAPL levels are modestly reduced (44), these interactions are rela- tively robust (Fig. 2B). We propose that the 6.5- fold decrease in DJH rearrangements in WAPL- depleted v- Abl cells results from decreased RC activity (Figs. 4I and 5B), as previously proposed for a similar reduction in Vκ- to- Jκ rearrangements upon WAPL depletion in this PNAS 2023 Vol. 120 No. 26 e2306564120 https://doi.org/10.1073/pnas.2306564120 7 of 10 Fig. 5. 3C- HTGTS profiles at Igh locus bating from RC in WT and 3′CBE− WAPL- degron v- Abl cells. (A) 3C- HTGTS signal counts at VHs domains of WT and 3′CBE− RAG1- deficient v- Abl lines baiting from iEμ/RC () with (red) or without (blue) Dox/IAA treatment. Each library was normalized to 160,314 total junctions (n = 3 repeats from three independent clones, mean ± SEM). Sixteen peaks across VHs region are called by MACS2 pipeline and highlighted in gray (peak 1, 6–9, 12–13 are called in WAPL- depleted WT v- Abl cells and WT pro- B cells), red (peaks 2–5, 10–11 are called in WAPL- depleted v- Abl cells), orange (peak1 is called only in WAPL- depleted WT v- Abl cells), and green (peaks are called in 3′CBE− v- Abl cells); peaks 14–16 are present in untreated WT and 3′CBE− v- Abl lines (SI Appendix, Fig. S8). (B) Zoom- in 3C- HTGTS profiles of Igh locus from proximal VHS to 3′Igh CBEs. (C) 3C- HTGTS signal counts of Igh locus from proximal VHs to 3′Igh CBEs in RAG- deficient WT v- Abl cells (red), WAPL- depleted v- Abl cells (blue) and pro- B cells (red) baiting from IGCR1/CBE1 (*). Each library was normalized to 112,525 total junctions (n = 3 repeats from three independent clones or 3 mice, mean ± SEM). v- Abl line (22). Notably, WAPL- depletion reduced rearrangement levels of all Ds, other than DQ52, leading to DQ52 contributing more substantially to residual DJH rearrangements (Fig. 4N). A sim- ilar overall trend was obtained when previously reported data that employed JH 1- 4 baits (22) were analyzed for absolute levels, as well as relative percentages (SI Appendix, Fig. S7 D and E and Table S4). We propose that DQ52 recombination, after WAPL depletion, may be less affected, because it accesses RAG by diffusion (versus scan- ning) from its RC location (24). Finally, diverse VHDJH rearrange- ments in WAPL- depleted cells occurred at a similarly low, absolute level to those of proximal VHs in untreated cells. However, VH rear- rangements in WAPL- depleted cells contributed to a 5.3- fold increase in the proportion of VHDJH/DJH rearrangements (Fig. 4 I and J). This latter finding may reflect increased levels of VHDJH recombina- tion in WAPL- depleted v- Abl cells compensating for reduced RC activity (Fig. 4). However, the net effect is that overall V(D)J recom- bination levels are much lower in WAPL- depleted v- Abl lines than in BM pro- B cells as reported (22). Deletion of 3′Igh CBEs decreased DJH and VHDJH rearrangement levels in both untreated and WAPL- depleted v- Abl cells. Yet, despite the nearly 50% decrease in V(D)J junctions in WAPL- depleted 3′CBE− versus WAPL depleted WT v- Abl cells, their VH utilization patterns across the VH locus were nearly identical. These findings support the proposal that the 3′Igh CBEs help maintain RC activity by reinforcing its impediment activity for the RAG scanning process (SI Appendix, Fig. S10) (1). We note that the extent to which the 8 of 10 https://doi.org/10.1073/pnas.2306564120 pnas.org 3′Igh CBEs reinforce RC activity may be compensated, in its absence, by interactions of the RC with downstream CBEs; it is also notable that these downstream CBE interactions are diminished by complete WAPL downregulation (Fig. 5B). Such compensatory activity of downstream CBEs, in the absence of 3′Igh CBEs, was also implicated in the context of Igh class switch recombination (50). Finally, normal pro- B cells do not completely down- regulate WAPL levels (44), which may contribute to preserving 3′Igh CBEs/RC interactions and RC activity in these cells. In this regard, WT and 3′CBE− pro- B cells have indistinguishable Igh V(D)J recombination patterns (22). Comparison of the RC- interacting peaks from 129SV pro- B cells (Fig. 2A), and C57BL/6 v- Abl cells (Fig. 5A) shows that a number of the peaks C57BL/6 Peaks 6–9, 12–13 (Fig. 5A) are shared, while others are unique due to the significant differences in the VH loci in these two strains. Major peaks in J558/3609, distal VHs region, are often associated with PAIR elements (Peaks 1, 4, 6, 8–10 in Fig. 2A and peaks 1- 8 in Fig. 5A), while other peaks in J558/3609, J558, and middle VH regions (Peaks 2, 3, 5, 7, 11–15 in Fig. 2A and peaks 9–13 in Fig. 5A) are associated with transcrip- tion or CBE- binding motifs. These observations support the notion that when WAPL is down- regulated, upon the neutralization of IGCR1, various transcription sites and CBEs still form sufficiently active loop extrusion impediments to promote interactions with the RC during RAG scanning of upstream VHs locus sequences. Methods Mice. Wild- type 129SV mice were purchased from Charles River Laboratories International. RAG2- deficient mice in 129SV background were purchased from Taconic. All animal experiments were performed under protocols approved by the Institutional Animal Care and Use Committee of Boston Children’s Hospital. Generation of IGCR1 CBE- Inversion Mice. A previously described pLNTK target- ing vector (49) containing inversion mutations of the 20- bp CBE1 and correspond- ing upstream activating sequence (WT sequence: 5′- TGCTTCCCCCTTGTGGCCATGA GCATTACTGCA- 3′; inverted: 5′- TGCAGTAATGCTCATGGCCACAAGGGGGAAGCA- 3′); or the 19- bp CBE2 (WT sequence: 5′- TCTCCACAAGAGGGCAGAA- 3′; inverted sequence: 5′- TTCTGCCCTCTTGTGGAGA- 3′) sites within IGCR1 were electroporated into TC1 ES cells. Successfully targeted clones with CBE1 or CBE2 inversion integration were assessed by Southern blot analyses using StuI- digested (13.9- kb untargeted; 10- kb targeted) or SpeI- digested (16.3- kb untargeted; 12.7- kb targeted) genomic DNA with appropriate probes. Two independently targeted clones containing each inversion mutation were subjected to adenovirus mediated Cre deletion to remove the NeoR gene, karyotyped, and injected for germline transmission. Homozygous mice were generated through breeding and genotype was confirmed by PCR gen- otyping (Primer sequences are listed in SI Appendix, Table S5). Generation of v- Abl Cell Lines. The WT v- Abl- kinase- transformed pro- B cell line (v- Abl pro- B cells) was derived by retroviral infection of BM pro- B cells derived from rag2−/− mice as described (48). IGCR1- mutated RAG2- deficient v- Abl lines were established by breeding each IGCR1 mutant mice with rag2−/− germline mice to generate RAG2- deficient homozygous IGCR1- mutated mice (i.e., IGCR1/ CBE1−/−rag2−/−), and deriving v- Abl lines as described (48). All these mutations were confirmed by PCR genotyping. 3′Igh CBE deletion in single Igh WAPL- degron v- Abl (22) were generated by designed sgRNAs and screened by PCR. The sequence of all sgRNAs and oligos used is listed in SI Appendix, Table S5. HTGTS- V(D)J- seq and Data Analyses. Pro- B cells used in HTGTS- V(D)J- seq exper- iments were purified from WT or IGCR1- mutated mice as described (53). 2ug pro- B cell genomic DNA were used to generate each library. The sequence of the JH4 coding end primer (129SV background) used to generate HTGTS- V(D)J- seq librar- ies is listed in SI Appendix, Table S5. HTGTS- V(D)J- seq libraries were prepared as described (24). HTGTS- V(D)J- seq libraries were sequenced using paired- end 300- bp sequencing on a Mi- Seq (Illumina) machine. The WT 129SV pro- B cell data shown in Fig. 1 and SI Appendix, Figs. S1 and S2 were extracted from a prior publication (GSM2183881- GSM2183883) (53). All libraries were normalized to total reads (junctions+germine reads) or junctions across a given locus, and the VHDJH and DJH junctions are described in SI Appendix, Tables S1 and S2. When normalized to total reads, all libraires were normalized to the smallest libraries from the same batch of experiments. The number of normalized reads or junctions is indicated in the figure and figure legends. D usage from the VHDJH joins was analyzed via the VDJ_annotation pipeline. Productive and nonproductive VHDJH joins were analyzed via VDJ_productivity_annotation pipeline (Data, Materials, and Software Availability). RAG recombination and treatment of WAPL- degron v- Abl cells was performed as described (22), and the JH4 coding end primer (C57BL/6 background) used to generate HTGTS- V(D)J- seq libraries. All libraries were normalized to total reads or junctions, and the VHDJH and DJH junctions are described in SI Appendix, Table S3. D usage from the VHDJH joins was analyzed by the VDJ_annotation pipeline. 3C- HTGTS and Data Analyses. RAG2- deficient pro- B cells for 3C- HTGTS were purified and cultured as described (19). Cycling or G1- arrested RAG2- deficient v- Abl pro- B cells for 3C- HTGTS were prepared as described (18). Treatment of WAPL- degron v- Abl cells was performed as described (22). 3C- HTGTS was per- formed as described (18). Briefly, 10 million cells were cross- linked with 2% (v/v) formaldehyde for 10 min at RT. Cells were lysed in 50 mM Tris- HCl, pH 7.5, containing 150 mM NaCl, 5 mM EDTA, 0.5% NP- 40, 1% Triton X- 100 and protease inhibitors (Roche, #11836153001). Nuclei were digested with 700 units of NlaIII (NEB, #R0125) restriction enzyme at 37°C overnight, followed by ligation (T4 DNA ligase NEB M0202L) at 16°C overnight. Cross- links were reversed and samples were treated with Proteinase K (Roche, #03115852001) and RNase A (Invitrogen, #8003089) prior to DNA precipitation. 3C- HTGTS libraries were generated using LAM- HTGTS (56), and primers are listed in SI Appendix, Table S5. 3C- HTGTS libraries were sequenced using paired- end 150- bp sequencing on a Next- seq550 (Illumina) or paired- end 300- bp sequencing on a Mi- Seq (Illumina) machine. Data were processed as described previously (18). In addition, the PCR artificial junctions at Chr12: 114,692,680 in SI Appendix, Fig. S6 A were removed from the total junctions. The junctions from Chr12 were extracted and counted for normalization. All 3C- libraires were normalized to the smallest libraries from the same batch of experiments. The number of normalized junctions is indicated in the figure legends. For peak analysis, 3C- HTGTS profiles were analyzed by MACS2 pipeline to call robust interaction peaks (macs2 bdgpeakcall - c20 - l400 - g1000 was used for pro- B 3C- HTGTS in Fig. 2 and macs2 bdgpeakcall - c30 - l400 - g1000 was used for v- Abl 3C- HTGTS in Fig. 5). The peaks that showed >twofold intensity change were annotated as unique peaks of indicated condition. CBE motifs were called by the MEME- FIMO scanning CTCF motif (MA0139.1 in JASPAR database) (57) and validated by previously published ChIP- seq data (19, 22). Data, Materials, and Software Availability. 3C- HTGTS and HTGTS- V(D)J- seq data were processed through published pipelines (http://robinmeyers.github.io/ transloc_pipeline/) as described (18). D usage in VHDJH joins was processed via a custom pipeline (https://github.com/Yyx2626/VDJ_annotation/) (19). Productive and non- productive junctions were processed via another pipeline (https://github. com/Yyx2626/VDJ_annotation/) (19). HTGTS- V(D)J- seq, 3C- HTGTS and GRO- seq sequencing data reported in this study are available through GEO (GSE230605) (58). All study data are included in the article and/or SI Appendix. Previously published data were used for this work [GSE151910 (22) and GSE821126 (53)]. ACKNOWLEDGMENTS. We thank members of the Alt laboratory for stimulating discussions. This work was supported by the NIH Grant R01AI020047 to F.W.A. and Grant F31- AI117920 to S.G.L.; Z.B. and H.- Q.D. were supported in part by Cancer Research Institute Irvington Fellowships. F.W.A. is an investigator of the Howard Hughes Medical Institute. Author affiliations: aHHMI, Boston Children’s Hospital, Boston, MA 02115; bProgram in Cellular and Molecular Medicine, Boston Children’s Hospital, Harvard Medical School, Boston, MA 02115; and cDepartment of Genetics, Harvard Medical School, Boston, MA 02115 Author contributions: Z.L., L.Z., S.G.L., Z.B., and F.W.A. designed research; Z.L., L.Z., S.G.L., Y.Z., H.- Q.D., and Z.B. performed research; Z.L., S.G.L., C.G., and Z.B. contributed new reagents/analytic tools; Z.L., A.Y.Y., S.G.L., Z.B., and F.W.A. analyzed data; and Z.L., L.Z., Z.B., and F.W.A. wrote the paper. Reviewers: C.M., University of California, San Diego; and A.N., National Cancer Institute. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). PNAS 2023 Vol. 120 No. 26 e2306564120 https://doi.org/10.1073/pnas.2306564120 9 of 10 1. 2. 3. 4. 5. Y. Zhang, X. Zhang, H.- Q. Dai, H. Hu, F. W. Alt, The role of chromatin loop extrusion in antibody diversification. Nat. Rev. Immunol. 9, 1–17 (2022). C. Liu, Y. Zhang, C. C. Liu, D. G. Schatz, Structural insights into the evolution of the RAG recombinase. Nat. Rev. Immunol. 22, 353–370 (2022). F. W. Alt, Y. Zhang, F.- L. Meng, C. Guo, B. Schwer, Mechanisms of programmed DNA lesions and genomic instability in the immune system. Cell 152, 417–429 (2013). Y. Ji et al., The in vivo pattern of binding of RAG1 and RAG2 to antigen receptor loci. Cell 141, 419–431 (2010). G. Teng, D. G. Schatz, Regulation and evolution of the RAG recombinase. Adv. Immunol. 128, 1–39 (2015). 32. J. Hu et al., Chromosomal loop domains direct the recombination of antigen receptor genes. Cell 163, 947–959 (2015). 33. S. G. Lin, Z. Ba, F. W. Alt, Y. Zhang, RAG chromatin scanning during V(D)J recombination and chromatin loop extrusion are related processes in Advances in Immunology (Elsevier, 2018), pp. 93–135. 34. C. Benner, T. Isoda, C. Murre, New roles for DNA cytosine modification, eRNA, anchors, and superanchors in developing B cell progenitors. Proc. Natl. Acad. Sci. U.S.A. 112, 12776–12781 (2015). 35. M. Fuxa et al., Pax5 induces V- to- DJ rearrangements and locus contraction of the immunoglobulin heavy- chain gene. Genes Dev. 18, 411–422 (2004). 6. M. J. MacPherson, P. D. Sadowski, The CTCF insulator protein forms an unusual DNA structure. BMC 36. S. Jhunjhunwala et al., The 3D structure of the immunoglobulin heavy- chain locus: Implications for Mol. Biol. 11, 101 (2010). long- range genomic interactions. Cell 133, 265–279 (2008). 7. H. Nakahashi et al., A genome- wide map of CTCF multivalency redefines the CTCF code. Cell Rep. 3, 37. S. T. Kosak et al., Subnuclear compartmentalization of immunoglobulin loci during lymphocyte 8. 9. 1678–1689 (2013). S. S. P. Rao et al., A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell 159, 1665–1680 (2014). G. Fudenberg et al., Formation of chromosomal domains by loop extrusion. Cell Rep. 15, 2038–2049 (2016). 10. A. L. Sanborn et al., Chromatin extrusion explains key features of loop and domain formation in wild- type and engineered genomes. Proc. Natl. Acad. Sci. U.S.A. 112, E6456–E6465 (2015). 11. M. Vietri Rudan et al., Comparative Hi- C reveals that CTCF underlies evolution of chromosomal domain architecture. Cell Rep. 10, 1297–1309 (2015). 12. K. Kraft et al., Serial genomic inversions induce tissue- specific architectural stripes, gene misexpression and congenital malformations. Nat. Cell Biol. 21, 305–310 (2019). 13. T.- H.S. Hsieh et al., Resolving the 3D landscape of transcription- linked mammalian chromatin folding. Mol. Cell 78, 539–553.e8 (2020). development. Science 296, 158–162 (2002). 38. J. Medvedovic et al., Flexible long- range loops in the VH gene region of the Igh locus facilitate the generation of a diverse antibody repertoire. Immunity 39, 229–244 (2013). 39. L. Montefiori et al., Extremely long- range chromatin loops link topological domains to facilitate a diverse antibody repertoire. Cell Rep. 14, 896–906 (2016). 40. E. Roldán et al., Locus “decontraction” and centromeric recruitment contribute to allelic exclusion of the immunoglobulin heavy- chain gene. Nat. Immunol. 6, 31–41 (2005). 41. M. B. Rother et al., Nuclear positioning rather than contraction controls ordered rearrangements of immunoglobulin loci. Nucleic Acids Res. 44, 175–186 (2016). 42. C. Sayegh, S. Jhunjhunwala, R. Riblet, C. Murre, Visualization of looping involving the immunoglobulin heavy- chain locus in developing B cells. Genes Dev. 19, 322–327 (2005). 43. J. S. Lucas, Y. Zhang, O. K. Dudko, C. Murre, 3D trajectories adopted by coding and regulatory DNA elements: First- passage times for genomic interactions. Cell 158, 339–352 (2014). 14. Y. Guo et al., CRISPR inversion of CTCF sites alters genome topology and enhancer/promoter 44. L. Hill et al., Wapl repression by Pax5 promotes V gene recombination by Igh loop extrusion. Nature function. Cell 162, 900–910 (2015). 584, 142–147 (2020). 15. E. de Wit et al., CTCF binding polarity determines chromatin looping. Mol. Cell 60, 676–684 (2015). 16. Z. Jia et al., Tandem CTCF sites function as insulators to balance spatial chromatin contacts and 45. J. Gassler et al., A mechanism of cohesin- dependent loop extrusion organizes zygotic genome architecture. EMBO J. 36, 3600–3618 (2017). topological enhancer- promoter selection. Genome Biol. 21, 75 (2020). 46. J. H. I. Haarhuis et al., The cohesin release factor WAPL restricts chromatin loop extension. Cell 169, 17. M. Ruiz- Velasco et al., CTCF- mediated chromatin loops between promoter and gene body regulate 693–707.e14 (2017). alternative splicing across individuals. Cell Syst. 5, 628–637.e6 (2017). 47. G. Wutz et al., Topologically associating domains and chromatin loops depend on cohesin and are 18. S. Jain, Z. Ba, Y. Zhang, H.- Q. Dai, F. W. Alt, CTCF- binding elements mediate accessibility of RAG regulated by CTCF, WAPL, and PDS5 proteins. EMBO J. 36, 3573–3599 (2017). substrates during chromatin scanning. Cell 174, 102–116.e14 (2018). 48. A. L. Bredemeyer et al., ATM stabilizes DNA double- strand- break complexes during V(D)J 19. Z. Ba et al., CTCF orchestrates long- range cohesin- driven V(D)J recombinational scanning. Nature recombination. Nature 442, 466–470 (2006). 586, 305–310 (2020). 20. E. J. Banigan et al., Transcription shapes 3D chromatin organization by interacting with loop extrusion. Proc. Natl. Acad. Sci. U.S.A. 120, e2210480120 (2023). 21. H. B. Brandão et al., RNA polymerases as moving barriers to condensin loop extrusion. Proc. Natl. Acad. Sci. U.S.A. 116, 20489–20499 (2019). 22. H.- Q. Dai et al., Loop extrusion mediates physiological Igh locus contraction for RAG scanning. Nature 590, 338–343 (2021). 49. S. G. Lin, C. Guo, A. Su, Y. Zhang, F. W. Alt, CTCF- binding elements 1 and 2 in the Igh intergenic control region cooperatively regulate V(D)J recombination. Proc. Natl. Acad. Sci. U.S.A. 112, 1815–1820 (2015). 50. X. Zhang, H. S. Yoon, A. M. Chapdelaine- Williams, N. Kyritsis, F. W. Alt, Physiological role of the 3′IgH CBEs super- anchor in antibody class switching. Proc. Natl. Acad. Sci. U.S.A. 118, e2024392118 (2021). 51. F. W. Alt et al., Ordered rearrangement of immunoglobulin heavy chain variable region segments. 23. M. V. Neguembor et al., Transcription- mediated supercoiling regulates genome folding and loop EMBO J. 3, 1209–1219 (1984). formation. Mol. Cell 81, 3065–3081.e12 (2021). 24. Y. Zhang et al., The fundamental role of chromatin loop extrusion in physiological V(D)J recombination. Nature 573, 600–604 (2019). 25. C. Guo et al., CTCF- binding elements mediate control of V(D)J recombination. Nature 477, 424–430 (2011). 26. S. C. Degner, T. P. Wong, G. Jankevicius, A. J. Feeney, Cutting edge: Developmental stage- specific recruitment of cohesin to CTCF sites throughout immunoglobulin loci during B lymphocyte development. J. Immunol. 182, 44–48 (2009). 27. B. Birshtein, The role of CTCF binding sites in the 3′ immunoglobulin heavy chain regulatory region. Front. Genet. 3, 251 (2012). 52. G. Kumari, R. Sen, Chapter two–Chromatin interactions in the control of immunoglobulin heavy chain gene assembly in Advances in Immunology, Molecular Mechanisms that Orchestrate the Assembly of Antigen Receptor Loci, C. Murre, Ed. (Academic Press, 2015), pp. 41–92. 53. S. G. Lin et al., Highly sensitive and unbiased approach for elucidating antibody repertoires. Proc. Natl. Acad. Sci. U.S.A. 113, 7846–7851 (2016). 54. A. Ebert et al., The distal VH gene cluster of the Igh locus contains distinct regulatory elements with Pax5 transcription factor- dependent activity in Pro- B cells. Immunity 34, 175–187 (2011). 55. J. Verma- Gaur et al., Noncoding transcription within the Igh distal VH region at PAIR elements 28. F. E. Garrett et al., Chromatin architecture near a potential 3′ End of the Igh locus involves modular regulation of histone modifications during B- cell development and in vivo occupancy at CTCF sites. Mol. Cell Biol. 25, 1511–1525 (2005). affects the 3D structure of the Igh locus in pro- B cells. Proc. Natl. Acad. Sci. U.S.A. 109, 17004–17009 (2012). 56. R. L. Frock et al., Genome- wide detection of DNA double- stranded breaks induced by engineered 29. C. Bossen, R. Mansson, C. Murre, Chromatin topology and the regulation of antigen receptor nucleases. Nat. Biotechnol. 33, 179–186 (2015). assembly. Annu. Rev. Immunol. 30, 337–356 (2012). 57. T. L. Bailey, J. Johnson, C. E. Grant, W. S. Noble, The MEME suite. Nucleic Acids Res. 43, W39–W49 30. C. Proudhon, B. Hao, R. Raviram, J. Chaumeil, J. A. Skok, Long- range regulation of V(D)J (2015). recombination. Adv. Immunol. 128, 123–182 (2015). 31. L. Vian et al., The energetics and physiological impact of cohesin extrusion. Cell 173, 1165–1178. e20 (2018). 58. A. Y. Ye, Contribution of IGCR1 and 3′ CBE super anchor to developmental regulation of Igh V(D)J recombination. Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc. cgi?acc=GSE230605. Deposited 25 April 2023. 10 of 10 https://doi.org/10.1073/pnas.2306564120 pnas.org
10.1073_pnas.2300052120	RESEARCH ARTICLE \| NEUROSCIENCE OPEN ACCESS Glial dysregulation in the human brain in fragile X-associated tremor/ataxia syndrome Caroline M. Diasa,b,c,d,1, Biju Issace, Liang Sune, Abigail Lukowiczd, Maya Talukdarb,f Shira Rockowitzb,e, and Christopher A. Walshb,c,i,j,1 , Michael B. Millerb,h , Shyam K. Akulab,g , Katherine Walshb, Edited by Mary Hatten, Rockefeller University, New York, NY; received January 5, 2023; accepted April 3, 2023 Short trinucleotide expansions at the FMR1 locus are associated with the late-onset condition fragile X-associated tremor/ataxia syndrome (FXTAS), which shows very different clinical and pathological features from fragile X syndrome (associated with longer expansions), with no clear molecular explanation for these marked differences. One prevailing theory posits that the shorter, premutation expansion uniquely causes extreme neurotoxic increases in FMR1 mRNA (i.e., four to eightfold increases), but evidence to support this hypothesis is largely derived from analysis of peripheral blood. We applied single-nucleus RNA sequencing to postmortem frontal cortex and cerebellum from 7 individuals with premutation and matched controls (n = 6) to assess cell type–specific molecular neuropathology. We found only modest upreg- ulation (~1.3-fold) of FMR1 in some glial populations associated with premutation expansions. In premutation cases, we also identified decreased astrocyte proportions in the cortex. Differential expression and gene ontology analysis demonstrated altered neuroregulatory roles of glia. Using network analyses, we identified cell type–specific and region-specific patterns of FMR1 protein target gene dysregulation unique to premutation cases, with notable network dysregulation in the cortical oligodendro- cyte lineage. We used pseudotime trajectory analysis to determine how oligoden- drocyte development was altered and identified differences in early gene expression in oligodendrocyte trajectories in premutation cases specifically, implicating early cortical glial developmental perturbations. These findings challenge dogma regarding extremely elevated FMR1 increases in FXTAS and implicate glial dysregulation as a critical facet of premutation pathophysiology, representing potential unique thera- peutic targets directly derived from the human condition. FMR1 \| FXTAS \| human brain \| glia \| snRNA-seq FMR1-related disorders contribute to neurologic dysfunction across the lifespan (1, 2). Long trinucleotide (CGG) expansion (i.e., full mutations) in the 5′ UTR of the FMR1 gene are associated with the neurodevelopmental disorder fragile X syndrome (FXS), while short, “premutations” are associated with fragile X-associated tremor and ataxia syndrome (FXTAS), a late-onset condition characterized by executive functioning decline and pro- gressive cerebellar ataxia, presenting in a subset of premutation carriers (3–7). In the latter, neuropathological and imaging studies have identified intranuclear neuronal and astrocytic inclusions, prominent white matter abnormalities including myelin pallor and spongiosis, and characteristic T2 white matter hyperintensities on MRI (3, 8–11). In contrast, in FXS, an early-onset neurodevelopmental disorder characterized by intellectual disability and characteristic facial features (12, 13), only subtle functional changes in white matter in humans have been identified on imaging (14–16). The molecular correlates of these findings in both conditions are unknown. The full mutation is associated with hypermethylation and transcriptional silencing of the FMR1 locus, and absent FMR1 protein (FRMP), while the premutation has been reported to be paradoxically associated with increases in FMR1 mRNA, particularly in blood, with variable reductions in FMRP levels (5–7, 17–21). FMRP is a critical RNA-binding and regulatory protein that acts as a central hub in brain function (22–25). Although individuals with the premutation may present with alterations in typical neu- rodevelopment, FXS patients generally do not present with features of FXTAS. These divergent clinical and molecular phenotypes have led to the hypothesis that the clinical symptomatology associated with FXTAS is related to a neurotoxic effect of increased levels of FMR1 mRNA in the nervous system. This argument is bolstered by findings of a four to eightfold increase of FMR1 mRNA in peripheral blood cells of individuals with the premutation (19). However, prior bulk studies of human postmortem brain tissue from individuals with the premutation have revealed more modest, ~0.9- to 1.5-fold, changes in FMR1 mRNA (9, 26). Significance Genetic variation at the FMR1 locus confers risk for both the neurodevelopmental disorder fragile X syndrome and the neurodegenerative condition fragile X-associated tremor/ ataxia syndrome. Although animal models have been critical in elucidating molecular mechanisms of cellular dysfunction in fragile X-related disorders, understanding how the human brain is directly impacted remains unresolved. We conducted a cell type–specific transcriptomic analysis of postmortem human brains from individuals with fragile X mutations and matched controls, sequencing over 120,000 nuclei from the frontal cortex and cerebellum. We find evidence for cell type–specific, disease- specific, and regional-specific patterns of transcriptional and FMR1 protein (FMRP) network perturbations, providing a foundation for therapeutic development directly derived from the human condition. Author contributions: C.M.D., S.R., and C.A.W. designed research; C.M.D. and S.K.A. performed research; C.M.D., B.I., L.S., A.L., M.T., M.B.M., K.W., and S.R. analyzed data; and C.M.D. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected] or christopher. [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2300052120/-/DCSupplemental. Published May 30, 2023. PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 1 of 12 Prior studies in postmortem human brain in both FXS and FXTAS have focused on bulk cellular analysis, which does not resolve cell type–specific molecular alterations. Whereas it is pos- sible that the cellular heterogeneity of the human CNS may mask toxic levels of FMR1 mRNA, other hypotheses, including an inappropriate DNA damage response, mitochondrial stress, and polyglycine-containing peptide accumulation, have been put forth as additional hypotheses to explain the pathophysiology of FXTAS (11, 27, 28). It is also possible that reduced FMRP contributes to premutation pathology in a developmentally distinct manner from the total loss that occurs in FXS. Finally, while studies of the impact of FMR1 disruption have been focused on postmitotic neurons, there is increasing evidence implicating important roles for FMR1 in a diversity of cellular subtypes at multiple points in nervous system development, including in glia (29–33). Despite these gaps in knowledge, there has been no global cell type–specific analysis of transcriptional changes related to FMR1 expansions in the human brain to date. To understand the molecular and cellular perturbations associ- ated with fragile X expansion in the human brain, we applied single-nuclei RNA-sequencing (snRNA-seq) to postmortem fron- tal cortex and cerebellar hemisphere of individuals with FMR1 premutations and controls. We identified changes in FMR1 expres- sion, cellular proportion, global gene expression, and oligoden- drocyte cortical development that challenge current assumptions about molecular mechanisms underlying FXTAS pathogenesis and specifically implicate glial dysregulation as critical in fragile X molecular neuropathology. Results Prior to tissue processing, we reviewed available medical records to ensure that clinical and neuropathological data were consistent with genetic diagnoses (Table 1). We used tissue samples from primarily BA10 and lateral cerebellar hemisphere. All cases here have been previously presented in prior published work (34–36). Although we focus on the premutation, we also studied two cases of FXS, to assess whether well-known effects on FMR1 expression were present in our dataset. The majority of premutation cases had either clinical and/or neuropathological evidence of FXTAS (Table 1). We identified one case that in the past was mistakenly categorized as FXS, but whose clinical records and genetic testing revealed it to be a premu- tation (Table 1). One case of FXS due to a deletion of FMR1 was included given the known shared molecular consequences of FMR1 deletion and trinucleotide expansion (37) and neither FXS case had neuropathological abnormalities noted. We used dounce homogenization followed by sucrose gradient centrifugation to isolate nuclei (Fig. 1A), fluorescent nuclear sorting, and bioinfor- matic processing to ensure only high-quality nuclei were evaluated, Table 1. Demographic information for postmortem samples used ID 1793 5408 5497 5541 5657 AN10723 4555 4664 4751 5006 5212 5529 5746‡ 4806 5319 Sample* Region Age Race Sex PMI RIN Notes Prior validation CON CON CON CON CON CON FC, CBL FC FC, CBL FC, CBL FC, CBL CBL 11 6 Black M Black M 68 White M 84 White M 82 White M 60 M FXPM (67) FC, CBL 80 White M FXPM (100) FC, CBL 71 White M FXPM (88) FC, CBL 21 White M FXPM (150) FC, CBL 85 White M 19 16 13 12 22 24 12 3 5 5 FXPM CBL 80 White M 12 FXPM (58) FC 89+† White M FXPM (116) CBL, BA22 80 White M FXS FC, CBL 9 White M FXS (450+) FC, CBL 71 White M 16 22 22 17 7.6 8 8.3 8.4 Clinical diagnosis of Parkinson’s disease Progressive neurological decline, history of hydrocephalus Perinatal ischemic event, seizures 7.1 FXTAS neuropathology FXTAS neuropathology, progressive neurologi- cal decline 7.1 FXTAS neuropathology CGG repeat-primed PCR (Esanov 2016) §CGG repeat-primed PCR (D’Gama 2015, Esanov 2016) CGG repeat-primed PCR (Esanov 2016) §CGG repeat-primed PCR (D’Gama 2015, Esanov 2016) CGG repeat-primed PCR (Esanov 2016), western blotting (Tran 2019) western blotting (Tran 2019) Progressive neurological western blotting (Tran decline, middle cerebellar peduncle sign 2019) 4.3 6.6 FXS, gene deletion, no neuropathological findings FXS, full mutation, no neuropathological changes CGG repeat-primed PCR (Esanov 2016) (approximate FMR1 repeat size if applicable) †age deidentified ‡misclassified as FXS in Tran 2019 despite only partially reduced FMR1 protein; BA22 used only for SI Appendix, Fig. S8, sample not included in frontal cortex analysis §in addition to premutation, mosaic full mutation was identified in small percent of cells, records consistent with premutation-associated pathology and clinical presentation Repeat size if applicable ascertained from clinical records and prior published work. Validation summary describes published reports of cases in addition to clinical records. CON: non- disease control, FXPM: fragile X premutation, FXS: fragile X syndrome. FC: frontal cortex (BA10: Brodmann area 10; 4555 and 5529 listed as prefrontal cortex), CBL: lateral cerebellar hem- isphere [(section 5); 1793, 5212, 5746, 4555 listed as cerebellum], BA22: Brodmann area 22 (posterior superior temporal cortex), RIN: RNA integrity number, PMI: postmortem interval. 2 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org Fig. 1. Cell type–specific analysis of frontal cortex and cerebellum. (A) Sample preparation included dounce homogenization, sucrose centrifugation, fluorescent nuclear sorting, and nuclear encapsulation. (B) Summary of sample size and final filtered nuclei number per condition and region. (C) Cerebellar UMAP plot and dot plot of cell type–specific markers. (D) Frontal cortex UMAP plot and dot plot of cell type–specific markers in the frontal cortex. Abbreviations- FXRD: Fragile X-related disorders, Endo: endothelial, Astro: astrocyte, MOL: mature cortical oligodendrocyte, Oligo: mature cerebellar oligodendrocyte, OPC: oligodendrocyte progenitor, IN: interneuron, OL: oligodendrocyte lineage, Neu: excitatory neuron, Inh: inhibitory neuron, L: layer-specific excitatory neuron clusters. PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 3 of 12 Table 2. Filtered nuclei number Cluster Cerebellum Granule Oligo Bergmann glia Interneuron Interneuron II Microglia Astrocyte OPC Endothelial Purkinje Cerebellar Total Cortex Inh Neu Exc Neu OPC OL I OL II MOL Astro I Astro II Microglia Endo CON FXPM FXS Total 25,067 24,237 9,831 366 643 497 395 219 263 157 87 59 542 1276 848 930 268 276 327 63 21 130 334 155 97 85 104 79 20 23 27,753 28,788 10,858 67,399 3,916 2,776 2,411 3,173 250 4,381 4,264 1,719 3,348 138 2,049 2,689 1,467 1,058 81 6,890 914 468 1,518 26 1,081 868 1,591 522 630 1,767 1,577 614 1,092 63 Cortex total 26,376 17,160 9,805 53,341 Inhibitory neuron represents sum of all inhibitory neuron subclasses, and excitatory neu- ron represents sum of all excitatory neuron subclasses. leaving ~120,000 nuclei for downstream analysis (Fig. 1B and Table 2 and SI Appendix, Fig. S1 A–C). There were no differences between premutation and control or FXS groups in age, postmor- tem interval (PMI), or RNA integrity number (RIN) (SI Appendix, Fig. S2A). There was a reduction in RIN in FXS compared to controls, which may be related to increased metabolic stress that has been reported in FXS (38). There was no association between RIN and PMI and age (SI Appendix, Fig. S2B). We additionally validated predicted functional changes in FMRP expression with western blotting of frontal cortex samples (SI Appendix, Fig. S2C). Cell Type Annotation. We applied known cell type–specific markers to assess the specificity and accuracy of unsupervised clustering (Fig. 1 C and D). For both prefrontal cortex and cerebellar hemisphere, we identified specific classification of cellular subtypes for both neurons and glia. Layer-specific excitatory neuron and inhibitory neuron subclusters in the frontal cortex were consistent with broader prior published annotations (SI Appendix, Fig. S3) (39–42). There were distinctions between the cerebellum and cortex in overall cell composition. In the frontal cortex, we identified several distinct clusters appearing to reflect different states of oligoden- drocyte development, including PDGFRA + oligodendrocyte progenitor cells (OPCs), two intermediate clusters (OLI-PTPRZ1+ and OLII-TFC7L2/ENPP6+), and a mature myelinating oligo- dendrocyte (MOL) cluster (SI Appendix, Figs. S3 and S4). We compared the transcriptional profile of OLI and OLII to oligo- dendrocyte lineage clusters identified in mouse (43) and found that OLI gene expression resembled mouse committed oligoden- drocyte progenitors (COPs) and OLII resembled immature, newly formed, nonmyelinating oligodendrocytes. We also identified astrocyte groups similar to the protoplasmic astrocytes (astrocyte I) and fibrous astrocytes (astrocyte II) previously described in snRNA-seq studies of postmortem brain (39). On the contrary, in the cerebellum, although granule cells accounted for most nuclei captured, as expected, we also identified a cerebellar-specific Bergmann glia cluster, as well as interneuron and interneuron II that resembled molecular layer interneurons I and II, respectively, as described in ref. 44. OLI and OLII clusters were not identified in cerebellar samples. Given our sample size, we may be inade- quately powered to detect less frequent cellular subtypes known to reside within these brain regions, including functionally heter- ogeneous oligodendrocyte subtypes. FMR1 Expression. Analysis of individual cluster FMR1 expression revealed cell type–specific effects of fragile X status on FMR1 transcription in the premutation cases; specifically, modest but significant upregulation in only a few glial populations, as well as cell type–specific heterogeneity (Fig. 2). In fact, in premutation cases, the only clusters that demonstrated significantly increased FMR1 mRNA expression in either the frontal cortex or cerebellum were nonneuronal, including cerebellar Bergmann glia and cortical microglia (Fig. 2A). On the contrary, in fragile X syndrome, despite a smaller n, we identified total abrogation of FMR1 expression, as expected, in both cases of the full mutation and gene deletion (Fig. 2A). This provides important proof of principle that expected transcriptional signatures are present within the snRNA-seq data. Indeed, despite the smaller sample size and nuclei number, reduced FMR1 expression was robust among different clusters in Fig. 2. Modest FMR1 changes in premutation postmortem brain. (A) snRNA-seq of frontal cortex and cerebellar changes in FMR1 expression in premutation cases and FXS cases vs control. Cluster abbreviations as in Fig. 1. In premutation cases, only cerebellar Bergmann glia and cortical microglia demonstrated significant increases in FMR1 mRNA expression. There was widespread FMR1 reduction in FXS cases despite the smaller sample size. (Asterisk above the dot indicates padj < 0.05 for condition vs control comparison.) Expression level indicates average of scaled log-normalized expression. (B) Representative image from fluorescent in-situ hybridization demonstrating comparable FMR1 expression in excitatory neurons (SLC17A7 marker) in premutation vs control in the frontal cortex. (Scale bar, 30 μm). (C) No significant difference in FMR1 mRNA was seen in SLC17A7+ nuclei, or, using an extended boundary, SLC17A7+ cells (two-tailed t test, P = 0.91 (nuclei), P = 0.46 (cells)). n = 2 control, n = 3 premutation. 4 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org FXS in both neuronal and glial subpopulations across the brain, consistent with the large effect size of this genetic driver. To rule out inadequate power as a reason for lack of significant FMR1 regulation in cortical neuronal premutation populations, we also grouped inhibitory and excitatory neuronal subclusters and similarly identified no significant upregulation in this pseudo-bulk analysis (SI Appendix, Fig. S5). Thus, in general, the lack of significant upregulation of FMR1 mRNA in most neuronal subclusters in the premutation cases is not due to a lack of power. Rather, it suggests that overall, the increase in FMR1 expression in the brain caused by the premutation is far more modest than the four to eightfold increase observed in blood, and shows a preferential impact on glia, in the regions assessed here. Given our sample size, we are underpowered to detect changes in very rare cell types, such as Purkinje cells, and cannot rule out significant changes in those clusters. To validate these findings, we conducted fluorescent in-situ hybridization (RNAscope) on a small subset of premutation and control samples. Like the snRNA-seq findings, we found no evi- dence for significant upregulation of FMR1 in excitatory neurons. (Fig. 2 B and C). In RNAscope, we have the advantage of defining a larger border around the nuclei to incorporate an estimation of cellular expression outside of just the nucleus defined by DAPI expression. Using either the nuclear or extended “cellular” border, we found the same result (Fig. 2C). We also found no evidence for increases in FMR1 expression overall in all cortical nuclei/cells, and there were no major outliers in the distribution of FMR1 expression within samples that drove these findings. (SI Appendix, Fig. S6 A and B). Individual donors demonstrated heterogeneity with respect to FMR1 expression changes (SI Appendix, Fig. S7A-C). Premutation repeat size has been found to be a critical factor on the molecular and clinical level in FXTAS, with a significant correlation between repeat size and age of symptom onset as well as reduced FMRP and increased FMR1 transcription (17, 45). We interestingly found a significant positive correlation with cortical microglia FMR1 expression and repeat size (SI Appendix, Fig. S7B), in agree- ment with past work and suggesting that the modest cell type– specific increases in FMR1 expression observed may have clinical relevance. However, it was not generally a specific donor that drove FMR1 changes. Thus, the findings of changes in FMR1 expression are robust across samples. Changes in Cellular Abundance in FXTAS. Given that we used an unbiased nuclear collection, we can use nuclei number as a proxy for cellular composition (Fig. 3). We identified changes in relative cell types in the frontal cortex (BA10) in association with the premutation that were unexpected. Premutation cases demonstrated fewer-than-expected astrocytes, observations not due to the effect of age, and not observed in the cerebellum (Fig. 3 A and B, Supplemental Information File 1). To determine whether this was specific to BA10, we assessed BA22 in one premutation sample and identified astrocyte levels to be similar to control, suggesting that these findings may reflect a subcortical specific finding (SI Appendix, Fig. S8). We were surprised to see no significant changes in neuronal proportions. We wondered whether age-associated changes in neuronal composition might mask subtle effects of the premutation on neuronal clusters, given that age-associated changes in inhibitory density have also been observed (46–48). In this case, we identified a significant age- related decline in inhibitory neuron/total neuron composition in the frontal cortex as previously reported (49), although there was no detectable effect of fragile X mutation status on this decline (SI Appendix, Fig. S9 and Supplemental Information File 1). Note, Fig. 3. Change in cellular abundance in postmortem brain. (A) Average percentage of nuclear composition in the frontal cortex and cerebellum. (B) For premutation and control samples, linear regression analysis was conducted to determine the effect of condition and age on cellular abundance, using the equation cluster % = β0 + β1condition + β2*age. There was a significant effect of premutation condition on both cortical astrocyte groups (β1 P < 0.05), but not age (β2 P > 0.05). 89+ aged individual is omitted from graph but was included in regression analysis. although the changes in inhibitory neuron density have been previously reported with respect to age, our experimental design cannot distinguish between aging and maturational changes during development. We also identified age-related changes in cortical OPCs and microglia, but there was no impact of FMR1 status in those cluster proportions. Thus, we identified alterations in astrocyte number specific to the frontal cortex in premutation cases. In the cerebellum, changes in abundance in premutation cases also recapitulated past neuropathological studies, specifically previ- ous work demonstrating cerebellar Purkinje cell loss and Bergmann cell gliosis in individuals with the premutation (50) (Fig. 3A). Consistent with this, we identified a trend toward relatively fewer Purkinje cell nuclei, and greater Bergmann glial cell nuclei, in the cerebellum of premutation carriers versus controls (SI Appendix, Supplemental Information File 1), suggesting that loss of Purkinje cells may contribute to FXTAS signs and symptoms. Differential Gene Expression & Gene Ontology. We next interrogated global patterns of differentially expressed genes (DEG) (Datasets S1–S6 and SI Appendix, Tables S1–S7) in each cluster subtype between conditions. We observed that many neuronal clusters in FXS demonstrated a marked tendency toward upregulation of gene expression, reflecting a derepressed state, an effect not observed to the same extent in neuronal premutation comparisons. Additionally, premutation vs FXS comparison DEG lists were larger than premutation or FXS vs control comparisons, suggestive of divergent gene expression regulation between these two closely related conditions. Although there were no significant differences in age or PMI between control and premutation groups, these variables are impor- tant to consider given the wide age range we include and develop- mentally dependent clinical phenotypes. To explore the effects of age and PMI, we used PCA to determine whether these variables impacted gene expression variability and reran differential expression analysis in MAST (SI Appendix, Fig. S10) including these variables independently. Most comparisons demonstrated high overlap in DEG significant findings, and changes in FMR1 expression were equivalent (SI Appendix, Fig. S10 and Datasets S7–S18). It was only PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 5 of 12 for primarily FXS comparisons, in which inclusion of age rarely altered the DEG list substantially. This suggests that our main find- ings of FMR1 changes in expression, and global DEG in premuta- tion cases, are not due to confounding effects of these variables. We conducted gene ontology (GO) analysis to identify per- turbed biological processes. In premutation cases relative to con- trols, we identified evidence of disrupted neuroregulatory roles of glia, in both cortex and cerebellum (SI Appendix, Supplemental Information Files 2 and 3). For example, in multiple glial clusters, biological process terms including synaptic functioning, axon guidance, and neurotransmitters were enriched. On the contrary, classical glial terms were not ubiquitously found in different con- dition comparisons. For example, myelination terms were uniquely enriched in the OLI population in premutation vs control com- parisons, but in the OPC population, in FXS vs control compar- isons. There was also evidence of neuronal dysfunction: neurons in premutation cases demonstrated evidence of altered neuronal function and structure, such as synaptic signaling and neuronal arborization. Other notable processes that were revealed through GO analysis included widespread enrichment of protein folding and prion dis- ease terms, as well as signaling cascades implicated in FMR1 patho- physiology [including Wnt signaling, phosphoinositide 3 kinase (PI3K) signaling, and MAPK signaling (33, 51, 52)] in both neu- rons and glia in both premutation and FXS comparisons in multiple populations (SI Appendix, Supplemental Information Files 1 and 2). Interestingly, however, in any individual cell type, GO terms in FXS vs CON and PM vs CON comparisons were distinct, demonstrat- ing the unique biological changes occurring in these conditions. The observed increase in FMR1 expression in premutation cases that has been observed in the past has been ascribed to increased transcription as opposed to changes in mRNA stability (53). Our findings here of isolated increases of FMR1 in glial subtypes sug- gest that this may reflect cell type–specific transcriptional mech- anisms. To assess this further, we inspected the DEG lists from the cell clusters that demonstrated significant FMR1 upregulation in cortical microglia and cerebellar Bergmann glia and used the gene lists to identify core transcriptional regulators (SI Appendix, Tables S8 and S9). We identified transcriptional regulators, includ- ing IRF1 and STAT2, that have not been previously implicated in FXTAS pathogenesis, that warrant future study, particularly given some reports of immune-mediated disorders preceding FXTAS symptoms in individuals with the premutation (54). FMRP Network Dysregulation. We wondered whether the patterns observed in DEG lists were reflective of altered FMRP network functioning or more downstream, nonspecific effects. We noticed that known FMRP targets were present in both neuronal and glial differentially expressed gene lists in both FXS and premutation cases, so we next used network analysis and visualization to understand changes in FMRP network function. To do this, we generated a list of FMRP targets previously functionally validated in human cells and used DiVenn to visualize cell type– and network-specific patterns of regulation of these target genes among significantly differentially expressed gene lists (Dataset S19) (35, 55–57). We identified both expected and unexpected network perturbations. For example, cortical inhibitory neurons in FXS demonstrated a hub of shared, derepressed FMRP target genes, consistent with loss of FMRP’s role as a transcriptional repressor (Fig. 4A). This network convergence was absent in inhibitory cortical neurons in premutation cases, but present in cerebellar neurons in FXS (Fig. 4B), demonstrating that there are distinct effects on FMRP target dysregulation in FXS and FXTAS. Importantly, it suggests that despite variable FMRP loss A B Fig. 4. Network analysis of neuronal FMRP target dysregulation. Frontal cortex inhibitory neurons (A) demonstrate a hub of common derepressed FMRP target genes in FXS (Right), which is not observed in premutation cases (Left). (B) Cerebellar neurons in FXS also demonstrate a shared derepressed hub, with an opposite pattern in premutation cases. Different cell types demonstrate disproportionate effects of FMRP dysregulation depending on mutation status. Cluster abbreviations as in Fig. 1. Red, upregulation; blue, downregulation; yellow, opposite regulation in different cell types. in premutation cases, neuronal transcriptional derepression is not the main biological consequence, consistent with global DEG analysis. Intriguingly, in cerebellar premutation neurons, frequent downregulation of FMRP targets was observed. Additionally, there was differential impact on different cell types. For example, in FXS cases, granule cells demonstrated a disproportionate burden of the FMRP dysregulation not seen in premutation cases (Fig. 4B). Examination of FMRP target dysregulation in glia in premuta- tion cases also revealed an intriguing example of network discord- ance (Fig. 5 A and B). Unlike the majority of FMRP targets that were dysregulated in multiple cell types in the same direction, in premutation cases, in OPCs and MOLs, there was notable opposite regulation of the same FMRP target genes. (Fig. 5B). This was not observed to the same extent in FXS and highlights a disease-specific “switch” in gene expression regulation in two closely related oli- godendrocyte cell clusters. Additionally, in several glial subtypes in FXS, such as astrocyte II, there was a preponderance of unique FMRP targets which demonstrated downregulation (Fig. 5A). This is contrary to the known role of FMRP as a repressor and poten- tially highlights alternative cell type contexts of FMRP in vivo, such as in mRNA stabilization (58). (29, 30, 32, 59), we hypothesized Pseudotime Analysis Implicates Abnormal Oligodendrocyte Development in FXTAS. Given the observed changes in oligodendrocyte FMRP network function in premutation cases, and because FMRP is known to be critical in oligodendrocyte development that oligodendrocyte development would be uniquely perturbed in premutation cases. To assess this, we conducted a pseudotime trajectory analysis of oligodendrocyte clusters in the frontal cortex. Following reclustering with Monocle3 (Fig. 6A and SI Appendix, Fig. S11 A–B), we identified 2 distinct trajectories, one that went from OPC →OLI→MOL (branch 1), and another that went from OPC →OLI → OLII (branch 2). As expected, immature and mature markers tracked as expected with pseudotime (SI Appendix, Fig. S11C); however, there were marked differences across 6 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org A B Fig. 5. Network analysis of cortical glial FMRP target dysregulation. (A) Frontal cortex astrocytes and microglia demonstrate cell type–specific effects of FMRP network dysregulation in premutation cases (Left) and FXS (Right) (B) Oligodendrocyte lineage in premutation cases demonstrate uniquely divergent regulation of FMRP targets between OPCs and MOLs (red box); this pattern is not seen in FXS (black box). Cluster abbreviations as in Fig. 1. Red, upregulation; blue, downregulation; yellow, opposite regulation in different cell types. pseudotime in glial developmental markers (Fig. 6B), suggesting that FMR1 disruption impacts key processes in oligodendrocyte development in a temporally dependent manner. Indeed, the distribution of cell types along pseudotime trajectories revealed noticeable shifts in the distribution density in premutation cases (SI Appendix, Fig. S11D). We conducted differential expression analysis both along the pseu- dotime trajectory and between conditions (SI Appendix, Table S7 and Dataset S20). Using Moran’s I test to assess spatial autocorrelation of gene expression along the two trajectories, we identified several genes (NRXN1, CSMD1, for example) that have been implicated in cognitive function and aging, but not to our knowledge specifically through oligodendrocyte function (60–66) (Fig. 7). Additionally, assessment of top differentially expressed genes between conditions in pseudotime revealed an intriguing phenomenon in frontal cortex premutation cases: Genes that were expressed very early in narrow pseudotime windows in control cases were shifted later and with widened expression windows in premutation cases in both trajecto- ries. (Fig. 8A) This was not the case in cortical FXS vs control com- (SI Appendix, Fig. S12 A and B), or cerebellar parisons oligodendrocyte pseudotime analysis of premutation vs control comparisons (SI Appendix, Fig. S12 C and D), which suggests unique perturbation of early oligodendrocyte development in the frontal cortex, in premutation cases. GO analysis of biological process dys- regulation identified from these significantly differentially expressed genes between different conditions in pseudotime revealed a similar pattern in the top 20 most significant biological processes––prom- inent inclusion of multiple terms implicating neuroregulatory roles of glia (Fig. 8B). Additionally, in branch 1, myelination terms were uniquely identified in premutation comparisons (Fig. 8B). Discussion We present a cell type–specific analysis of gene expression of fragile X-related disorders in the human brain. We identified changes in FMR1 mRNA expression and cell type–specific gene expression that sheds light on molecular perturbations associated with FMR1 and specifically highlights an important role for glial molecular dysregulation in premutation pathology. impact RNA-binding protein Modest Glial FMR1 Upregulation in Premutation Cases. It has been shown that the FXTAS CGG repeat expansion leads to increased FMR1 expression and mRNA-rich intranuclear inclusions that function, supportive of an RNA gain-of-function hypothesis (9, 19, 67, 68); however, our data suggest that FMR1 mRNA expression in premutation cases, at least in the brain regions analyzed here, is more modestly affected than has been observed in peripheral blood cells and furthermore that it preferentially affects glial cells more than neurons. Given the robust elimination of FMR1 mRNA expression observed here in association with FXS, regardless of the genetic driver (trinucleotide expansion vs gene deletion), nuclei cluster, or brain region, we are confident in the validity of our approach. In the premutation cases, we identified modest upregulation of FMR1 expression, limited to some glial subclusters in both cerebellum and cortex demonstrating significant increases. Rather than extreme neurotoxic increases in FMR1 mRNA, our findings suggest a modest, ~1.3 fold, increase in FMR1 transcript levels, paralleling past studies of brain homogenate (9, 26). Findings of FMR1 upregulation in cortical microglia are intriguing given past reports of altered microglial activation in FXTAS human cases (31). It opens up the possibility that these prior reported changes may at least be in part directly driven by the effects of the premutation within microglia, as opposed to solely an inflammatory response to the microenvironment. Although we cannot rule out that neural cells expressing toxic levels of FMR1 transcript are selectively vulnerable and preferen- tially lost with time, our cellular proportion analysis (see below) does not support this interpretation, as one would expect remaining cells in populations that are disproportionately lost to have rela- tively higher increases in FMR1 mRNA. Finally, changes in FMR1 mRNA expression were comparable between clusters known to be vulnerable to premutation-associated intranuclear inclusions (neu- rons, astrocytes) and those known to be spared (oligodendrocytes), arguing against inclusion presence/nuclear measurement as being a confounding factor in FMR1 mRNA measurement. Our in-situ analysis using an extended cellular border also argues against nuclear measurement as being the reason for our findings for mod- est FMR1 changes in premutation cases. Finally, past work suggests that the FMR1 canonical poly(A) site is used comparably in normal and premutation cases (69), suggesting that alternative 3′ variants that might be underrepresented in our sequencing approach are an unlikely source for the differences observed in premutation cases here. These findings challenge the RNA toxicity hypothesis that dominates the literature, at least at the levels extrapolated from studies of peripheral blood. Given the ongoing controversy around the relative importance of FMR1 mRNA toxicity and other hypothesized mechanisms in in vitro and animal models (27, 70–72), our work is particularly relevant. Alterations in Cellular Abundance Implicate Cortical Glia. Changes in cell proportions in premutation cases in both cerebellum and cortex implicate glial dysregulation. In premutation cases, we identified a proportional decrease in cortical astrocytes, findings not explained by age. The relationship of basal FMR1 mRNA expression, change in FMR1 expression, and cellular proportion was not straightforward, arguing against a simplistic relationship between cellular proportion and PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 7 of 12 Fig. 6. Pseudotime analysis of cortical oligodendrocytes. (A) Reclustering of major oligodendrocyte clusters identifies two major trajectory branches identified in the frontal cortex. (A) Expression of oligodendrocyte markers track in both trajectories as expected, with significant differences between conditions across pseudotime. Orange asterisk: significant difference between premutation and control; purple: significant difference between FXS and control; red: significant difference between premutation and FXS, see SI Appendix, Methods for details on the Wald test. FMR1 toxicity. For example, glial cells in the frontal cortex that demonstrated modest differential expression of FMR1 also demonstrated the most marked changes in cellular proportion. This may be related to earlier developmental time points that are impacted, cellular extrinsic effects on survival and proliferation, or both. Regardless, given the findings of global brain atrophy and the reported decline in executive functioning reported in FXTAS (8, 73–75), these changes in cellular proportion warrant further exploration of the role of glia in FXTAS-associated cognitive symptoms and in other cortical areas. 8 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org it has recently been shown that the polyglycine region of FMRpolyG (the toxic protein generated from repeat-associated non-AUG translation of FMR1, a posited contributor to neu- rotoxicity in FXTAS) has a prion-like domain with low com- plexity similar to the so-called prion proteins in yeast, and that it may propagate cell to cell in a prion-like manner (70). Although none of the cases showed the rapid progression or typical clinical features of actual human prion protein (PrP)-related prion disease, such as Creutzfeldt–Jakob disease, our enrichment analysis is consistent with protein processing dysregulation being broadly relevant to FMR1 dysregulation in both premutation and FXS cases. For example, a heterogeneous increase in de novo protein synthesis in FXS cases is well described and is posited to contribute to challenges in drug development (23, 85). Other processes previously reported to be involved in FMR1 pathophysiology also appeared in GO lists, including Wnt, MAPK, and PI3K signaling. However, these signaling pathways did not appear ubiquitously, and our unbiased approach may thus inform future research on unex- pected critical cellular targets in which these pathways may play outsized roles, such as cerebellar Bergmann glia. Given that cell type–specific transcriptional regulatory mechanisms may con- tribute to differences in FMR1 regulation, we also identified transcription factor motifs that may provide critical information for future mechanistic work into the molecular steps that lead from the premutation expansion to changes in FMR1 mRNA. Unexpected Glial FMRP Network Functioning in Premutation Cases. FMRP reduction may contribute to clinical symptoms in premutation carriers in a developmentally distinct manner compared to FXS. We observed marked patterns of FMRP network dysregulation in both FXS and in cases with premutations, but these patterns were quite distinctive. For example, neurons, but not glia, demonstrated evidence of coherent FMRP network derepression in FXS. Furthermore, this phenomenon was notably absent in premutation cases, with certain networks demonstrating increased repression or incoherent regulation between cell types of a similar lineage. This analysis suggests that although FMRP loss may be present in both FXS and FXTAS, FXTAS does not represent simply a “milder” hit on FMRP network dysregulation. Rather, there are cell type–specific, region-specific, and developmental-specific factors which fundamentally alter FMRP network functioning. Specific Perturbations in Cortical Oligodendrocyte Development in Premutation Cases. Finally, our pseudotime analysis identified specific alterations in oligodendrocyte developmental trajectories in premutation cases in the frontal cortex and identified potential targets and pathways that may mediate these effects. For example, neurexin/neuroligin signaling has been implicated in neuronal signaling-dependent glioma growth (86). Premutation status specifically perturbed a narrow window of very early gene expression in oligodendrocyte pseudotime; this opens up intriguing lines of inquiry as to how this developmental hit may relate to a primarily degenerative phenotype. It also warrants follow-up given the plethora of developmental phenotypes observed in individuals with the premutation (87). Additionally, although myelination- related genes are known FMRP targets (23, 32, 59), these were not ubiquitously impacted in the oligodendrocyte lineage in different conditions or pseudotime (Fig. 8B), revealing the importance of cellular and developmental context. Unexpectedly, we did not identify a branch including both intermediate oligodendrocyte stages and MOLs. It is possible, given our small sample size, that additional rare oligodendrocyte states were missed, given the high Fig. 7. Spatial autocorrelation analysis of differentially expressed genes along pseudotime in the frontal cortex. Top 20 genes that change with pseudotime in branch 1 (Top) and branch 2 (Bottom). Genes were selected by q value and Moran’s I statistic. Alterations in Known and Unique Cellular Functions Revealed with Differential Gene Expression. We identified global transcriptional alterations associated with premutation status that support the now well-established principle that glia play central roles in neurodevelopment and disease (76–79). For example, OPCs are known to form synaptic- like structures and respond to neuronal activity (80) and glia more generally are critical in neuronal development, axonal integrity, and behavior (81–83). Differential gene regulation in premutation glia frequently identified perturbations in the maintenance of synaptic structure and function and altered neurotransmission in multiple glial lineage clusters. Indeed, white matter abnormalities in fragile X-related disorders more broadly may reflect subtle disruption of glial regulatory roles in neuronal homeostasis. Determination of whether these glial abnormalities contribute causally to clinical symptomatology or represent a secondary response to neuronal dysfunction will require further work in human model systems. Classical glial function was also uniquely disrupted in premutation cases. For example, in premutation cases, we identified upregulation and enrichment of myelination terms in the intermediate committed oligodendrocyte progenitor OLI cluster, a finding also identified in pseudotime analysis of branch 1, the branch that uniquely contains the OLI–MOL transition (78, 84). GO analysis also revealed widespread enrichment of terms implicated in protein processing and prion disease. Interestingly, PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 9 of 12 Fig. 8. Differentially expressed genes along oligodendrocyte pseudotime in the frontal cortex. (A) Heatmap of top 50 genes with strongest difference in expression (pair-wise comparison, Wald statistic) in branch 1 (Left) and branch 2 (Right) between premutation cases and controls. (B) Top 20 gene ontology biological processes enriched in branch 1 (Left) and branch 2 (Right) in the premutation vs control comparison functional heterogeneity of the oligodendrocyte lineage within the brain (43, 80). Given the small sample size of FXS cases, we use it primarily as a comparison for premutation biology and proof of principle that expected changes in biology are observed. However, replica- tion of past findings including absent FMR1 expression (5, 88) and evidence of metabolic stress (89, 90) corroborate known molecular neuropathology of the disorder (38). Thus, our findings on FXS, representing over 20,000 nuclei, serve as proof of prin- ciple that expected changes in FMR1 biology are present in these nuclear transcriptome datasets. Our work demonstrates the need for more comprehensive study of fragile X in human tissue directly in a variety of different cell types, particularly given that our sam- ple size is inadequate to identify meaningful changes in rare but potentially relevant cell types. In conclusion, we provide compelling evidence from the human brain regarding cell type–specific molecular neuropathology that helps contextualize the clinical heterogeneity associated with genetic variation at the FMR1 locus in neurodevelopment and neurodegeneration and specifically implicates glial dysregulation in premutation pathology. Our findings in premutation postmor- tem brain, in light of known neuropathological and imaging abnormalities, support the interpretation of FXTAS as a disorder defined by glial dysfunction and warrant consideration of FXTAS as a “gliodegenerative” disorder (79, 91). Materials and Methods Samples. Postmortem human tissues were obtained from the NIH NeuroBioBank, University of Maryland Brain and Tissue Bank, and the Autism BrainNet accord- ing to their institutional review board approvals and following written informed consent. Initial dissection of tissue for brain bank specimens was done under standardized procedures using sequential sectioning. Research on these dei- dentified specimens and data was performed at Boston Children’s Hospital with approval from the Committee on Clinical Investigation. Fragile X mutation status/ repeat size was verified through direct review of deidentified clinical records and crossreferenced with prior published validation of the same cases (Table 1). Most of the premutation cases had clinical symptomatology or neuropathological evidence of FXTAS (Table 1). Samples were group matched for age and sex, but no cutoffs were utilized to exclude any cases for PMI and RIN (SI Appendix, Fig. S2). Western Blotting. Approximately 25 to 50 mg frozen frontal cortex was homog- enized in RIPA buffer + protease inhibitors and centrifuged, and total protein content was then quantified. Laemmli sample buffer was added to the protein supernatant and boiled for 5 min. Equal amounts of protein (10 μg) were loaded onto precast SDS-Page gels with molecular weight ladders. Samples were trans- ferred to membranes, blocked with Licor block (Lincoln, NE), cut along molecular 10 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org weight markers, and incubated in primary antibody overnight diluted in block at four degrees. Following four washes in tris-buffered saline + tween (TBS + T), blots were incubated with LI-COR secondary fluorescent antibodies in the dark at room temperature for 1 h. After further washing including a final wash of TBS, the blots were scanned on a LI-COR Odyssey imager. The following primary antibodies and dilutions were used: GAPDH (Cell Signaling # 2118S) 1:15,000; FMRP (Cell Signaling 4317S) 1:1,000. RNAscope Multiplex Fluorescent V2 Assay. Frozen tissue blocks were mounted in prechilled Optimal Cutting Temperature compound media, and then placed on dry ice and stored at −80 °C. Ten micrometer sections were cut on a Leica CM1520 cryostat and mounted on charged slides. Fluorescent RNAscope was conducted as per Advanced Cell Diagnostics’ (ACD) instructions, including 30 min postfixation in chilled 10% NBF. A mixture of probes used (diluted 50:1 C1:C2 per ACD’s protocol) included Hs-FMR1-C1 (Cat #590731) and Hs-SLC17A7-C2 (Cat# 415611-C2). Fluorophores used were Akoya Biosciences Opal 520 (Cat #FP1487001KT, 1:1,000, FMR1) and Opal 570 (Cat #FP1488001KT, 1:1,000 to 1,500, SLC17A7). Four z-stacks were taken with identical settings from each slide on a Zeiss LSM 780 confocal. A maximum projection was generated in Fiji and then run through a semi-automated pipeline in Cell Profiler to count nuclei (defined by DAPI), cells (set boundary beyond DAPI), and mRNA puncta within nuclei and cells and for colocalization. The average of four stacks was taken for each sample/slide. Isolation of Postmortem Nuclei. Frozen tissue (~25 mg) from either frontal cortex or cerebellar hemisphere (see table of demographics) was dissected at −20 and subjected to dounce homogenization followed by sucrose gradient centrifugation as previously described. Tissue was primarily BA10 or lateral cerebellar hemisphere although some samples were named only as frontal cortex or cerebellum. Nuclei were filtered and incubated for 5 min in 1:1,000 Hoechst (Invitrogen H3569, Waltham MA). A total of 10,000 Hoechst + nuclei from the suspension were then sorted directly into 10× Genomics RT buffer (Pleasanton, CA) on a chilled plate holder to remove doublets, debris, and dying nuclei on a FACS Aria (BD Biosciences, Franklin Lakes, NJ) with a low-pressure nozzle (SI Appendix, Fig. S2). Following sorting, reverse transcriptase enzyme was added on ice, and nuclei were immediately processed for encapsulation in the 10× Chromium controller. cDNA and libraries were prepared according to the 10× documentation protocol for 3′ gene expression v3 chemistry. Sequencing and Quality Control. Samples were prepared and sequenced in matched groups to avoid confounding batch effects. The samples were sequenced on a NovaSeq 6000 (Illumina, San Diego, CA) to obtain high coverage and saturation and demultiplexed with bcl2fastq. CellRanger Count was utilized to generate count matrices with introns included, given intronic information is known to be informative for nuclear preparations. Sequencing metrics for each sample, in the output from CellRanger, are provided in SI Appendix, Table S10. To obtain a final high-quality nuclei set, filtering metrics were applied to nuclei in Seurat including: # UMIs> 500, # Genes > 250, log10GenesPerUMI (complexity measure) >0.8, and mitoRatio <0.1. Datasets were processed with SCTransform and integrated, and potential sources of variation were assessed with princi- pal component analysis, with mitochondrial gene expression regressed out. SCTransform conducts normalization, variance stabilization, and regression of unwanted variation; it removes variation due to cellular sequencing depth. Unsupervised clustering was performed with different resolutions followed by application of known cell type markers. For cerebellar Purkinje and endothelial cells, which represented a very small percentage of the total nuclei sample, cell type markers (CALB1; CLDN5) were used to manually select cell clusters using the SelectCells feature in Seurat. Analysis. Computationally intensive work was conducted on the Harvard Computing Cluster, O2, and Boston Children’s Hospital’s High-Performance Computing Cluster, E2. For detailed analysis methods, please see SI Appendix Methods. Data, Materials, and Software Availability. Data have been deposited in the controlled access section of DbGaP consistent with informed consent of tissue donors and can be accessed through the direct link https://www.ncbi.nlm.nih. gov/projects/gap/cgi-bin/study.cgi?study_id=phs000639.v2.p1 (92). ACKNOWLEDGMENTS. We thank Jennifer E. Neil for assistance with human sam- ples and documentation; Robert Sean Hill, Dilenny Gonzalez, and Sattar Khoshkhoo for assistance in reagent ordering and sample sequencing; Sara Bizzotto and Sattar Khoshkhoo for discussion of cell type–specific markers in cortex, Ronald Mathieu and the Boston Children’s Hematology/Oncology Flow Cytometry Research Facility for assistance with cell sorting; and the Engle lab and the Harvard Biopolymers Facility for assistance with Chromium Controller use and sequencing. Molecular genetics library quantification services were provided by the Boston Children’s Hospital Intellectual and Developmental Disabilities Research Center Molecular Genetics Core Facility supported by U54HD090255 from the NIH Eunice Kennedy Shriver National Institute of Child Health and Human Development. We acknowl- edge Boston Children’s Hospital’s High-Performance Computing Resources BCH HPC Cluster Enkefalos 2 (E2) which was made available for conducting the research reported in this publication. Portions of this research were conducted on the O2 High-Performance Compute Cluster, supported by the Research Computing Group, at Harvard Medical School. Analysis of transcriptional regulatory motifs was supported by the NIH P30CA046934 Bioinformatics and Biostatistics Shared Resource Core at the University of Colorado. We are grateful and indebted to the individuals and families who donated brain tissue for research purposes. This human tissue was obtained from the NIH NeuroBioBank at the University of Maryland, Baltimore, MD, and the Autism BrainNet. Autism BrainNet is a resource of the Simons Foundation Autism Research Initiative and includes the Autism Tissue Program collection, previ- ously funded by Autism Speaks. C.A.W. is supported by the Simons Foundation, the John Templeton Foundation, the NIA (R01AG070921), the NIMH (U01MH106883), and the Tan Yang Center for Autism Research at Harvard Medical School. C.A.W. is an investigator of the Howard Hughes Medical Institute. S.K.A. is supported by a Paul and Daisy Soros Fellowship for New Americans. S.K.A. and M.T were supported by T32GM007753 and T32GM144273. C.M.D. was supported in part by NIMH T32MH112510. M.B.M. is supported by K08AG065502. Some figures were gen- erated at Biorender.com. Author affiliations: aDivision of Developmental Medicine, Boston Children’s Hospital, Boston, MA 02115; bDivision of Genetics and Genomics, Manton Center for Orphan Disease Research, Boston Children’s Hospital, Boston, MA 02115; cDepartment of Pediatrics, Harvard Medical School, Boston, MA 02115; dDepartment of Pediatrics, Section of Developmental Pediatrics, Section of Genetics and Metabolism, and Denver Fragile X Clinic and Research Center, Children’s Hospital Colorado, University of Colorado Anschutz Medical Campus, Aurora, CO 80045; eResearch Computing, Department of Information Technology, Boston Children’s Hospital, Boston, MA 02115; fHarvard-Massachusetts Institute of Technology MD/PhD Program, Program in Bioinformatics & Integrative Genomics, Harvard Medical School, Boston, MA 02115; gHarvard-Massachusetts Institute of Technology MD/PhD Program, Program in Neuroscience, Harvard Medical School, Boston, MA 02115; hDepartment of Pathology, Brigham and Women’s Hospital, Boston, MA 02115; iHHMI, Boston Children’s Hospital, Boston, MA 02115; and jDepartment of Neurology, Harvard Medical School, Boston, MA 02115 1. R. J. Hagerman, “Fragile X syndrome” in Nature Reviews Disease Primers 2017 3:1 (Nature Publishing Group, 2017), pp 1–19. 2. M. A. Leehey, Fragile X-associated tremor/ataxia syndrome (FXTAS): Clinical phenotype, diagnosis 3. 4. and treatment. J. Invest. Med. 57, 830–836 (2009). S. Jacquemont et al., Fragile X premutation tremor/ataxia syndrome: Molecular, clinical, and neuroimaging correlates. Am. J. Hum. Genet. 72, 869–878 (2003). J. Grigsby, Clinically significant psychiatric symptoms among male carriers of the fragile X premutation, with and without FXTAS, and the mediating influence of executive functioning. Clin. Neuropsychol. 30, 944–959 (2016). 6. 7. 8. 9. A. J. M. H. Verkerk, Identification of a gene (FMR-1) containing a CGG repeat coincident with a breakpoint cluster region exhibiting length variation in fragile X syndrome. Cell 65, 905–914 (1991). Y. H. Fu, Variation of the CGG repeat at the fragile X site results in genetic instability: Resolution of the Sherman paradox. Cell 67, 1047–1058 (1991). C. M. Greco, Neuropathology of fragile X-associated tremor/ataxia syndrome (FXTAS). Brain 129, 243–255 (2006). F. Tassone, Intranuclear inclusions in neural cells with premutation alleles in fragile X associated tremor/ataxia syndrome. J. Med. Genet. 41, e43 (2004). 5. M. Pieretti, Absence of expression of the FMR-1 gene in fragile X syndrome. Cell 66, 817–822 10. S. Cohen et al., Molecular and imaging correlates of the fragile X-associated tremor/ataxia (1991). syndrome. Neurology 67, 1426–1431 (2006). PNAS 2023 Vol. 120 No. 23 e2300052120 https://doi.org/10.1073/pnas.2300052120 11 of 12 11. J. L. Schwartz, K. L. Jones, G. W. Yeo, Repeat RNA expansion disorders of the nervous system: 52. S. H. Kim, J. A. Markham, I. J. Weiler, W. T. Greenough, Aberrant early-phase ERK inactivation Post-transcriptional mechanisms and therapeutic strategies. Crit. Rev. Biochem. Mol. Biol. 56, 31–53 (2021). impedes neuronal function in fragile X syndrome. Proc. Natl. Acad. Sci. U.S.A. 105, 4429–4434 (2008). 12. G. Turner et al., X-linked mental retardation associated with macro-orchidism. J. Med. Genet. 12, 53. F. Tassone et al., Elevated FMR1 mRNA in premutation carriers is due to increased transcription. RNA 367–371 (1975). 13. J. P. Martin, J. Bell, A pedigree of mental defect showing sex-linkage. J. Neurol. Psychiatry 6, 154–157 (1943). 54. 13, 555–562 (2007). I. Jalnapurkar, N. Rafika, F. Tassone, R. Hagerman, Immune mediated disorders in women with a fragile X expansion and FXTAS. Am. J. Med. Genet. A. 167A, 190–197 (2015). 14. M. R. Swanson et al., Development of white matter circuitry in infants with fragile X syndrome. JAMA 55. L. Sun et al., DiVenn: An interactive and integrated web-based visualization tool for comparing gene Psychiatry 75, 505–513 (2018). lists. Front. Genet. 10, 421 (2019). 15. B. P. Hallahan et al., In vivo brain anatomy of adult males with Fragile X syndrome: An MRI study. 56. M. Li et al., Identification of FMR1-regulated molecular networks in human neurodevelopment. Neuroimage 54, 16–24 (2011). Genome Res. 30, 361–374 (2020). 16. G. M. Sandoval et al., Neuroanatomical abnormalities in fragile X syndrome during the adolescent 57. M. Ascano Jr., et al., FMRP targets distinct mRNA sequence elements to regulate protein expression. and young adult years. J. Psychiatr. Res. 107, 138–144 (2018). Nature 492, 382–386 (2012). 17. A. Kenneson, F. Zhang, C. H. Hagedorn, S. T. Warren, Reduced FMRP and increased FMR1 58. C. R. Hale et al., FMRP regulates mRNAs encoding distinct functions in the cell body and dendrites transcription is proportionally associated with CGG repeat number in intermediate-length and premutation carriers. Hum. Mol. Genet. 10, 1449–1454 (2001). 18. F. Tassone, R. J. Hagerman, W. D. Chamberlain, P. J. Hagerman, Transcription of the FMR1 gene in individuals with fragile X syndrome. Am. J. Med. Genet. 97, 195–203 (2000). 19. F. Tassone et al., Elevated levels of FMR1 mRNA in carrier males: A new mechanism of involvement in the fragile-X syndrome. Am. J. Hum. Genet. 66, 6–15 (2000). 20. F. Tassone et al., FMRP expression as a potential prognostic indicator in fragile X syndrome. Am. J. 21. Med. Genet. 84, 250–261 (1999). I. Oberle, Instability of a 550-base pair DNA segment and abnormal methylation in fragile X syndrome. Science 252, 1097–1102 (1991). 22. L. Ceolin, Cell type-specific mRNA dysregulation in hippocampal CA1 pyramidal neurons of the fragile X syndrome mouse model. Front. Mol. Neurosci. 10, 340 (2017). of CA1 pyramidal neurons. Elife 10, e71892 (2021). 59. H. Wang et al., Developmentally-programmed FMRP expression in oligodendrocytes: A potential role of FMRP in regulating translation in oligodendroglia progenitors. Hum. Mol. Genet. 13, 79–89 (2004). 60. V. Stepanov, A. Marusin, K. Vagaitseva, A. Bocharova, O. Makeeva, Genetic variants in CSMD1 gene are associated with cognitive performance in normal elderly population. Genet. Res. Int. 2017, 6293826 (2017). 61. M. Patel, Parkinson disease: CSMD1 gene mutations can lead to familial Parkinson disease. Nat. Rev. Neurol. 13, 641 (2017). 62. J. Ruiz-Martinez, L. J. Azcona, A. Bergareche, J. F. Marti-Masso, C. Paisan-Ruiz, Whole-exome sequencing associates novel CSMD1 gene mutations with familial Parkinson disease. Neurol. Genet. 3, e177 (2017). 23. J. C. Darnell et al., FMRP stalls ribosomal translocation on mRNAs linked to synaptic function and 63. C. Montani et al., The X-linked intellectual disability protein IL1RAPL1 regulates dendrite complexity. autism. Cell 146, 247–261 (2011). J. Neurosci. 37, 6606–6627 (2017). 24. Y. Q. Zhang et al., Drosophila fragile X-related gene regulates the MAP1B homolog Futsch to control 64. M. Ramos-Brossier et al., Novel IL1RAPL1 mutations associated with intellectual disability impair synaptic structure and function. Cell 107, 591–603 (2001). synaptogenesis. Hum. Mol. Genet. 24, 1106–1118 (2015). 25. F. Zalfa et al., The fragile X syndrome protein FMRP associates with BC1 RNA and regulates the 65. E. L. Youngs, R. Henkhaus, J. A. Hellings, M. G. Butler, IL1RAPL1 gene deletion as a cause of X-linked translation of specific mRNAs at synapses. Cell 112, 317–327 (2003). intellectual disability and dysmorphic features. Eur. J. Med. Genet. 55, 32–36 (2012). 26. D. I. Pretto, Reduced EAAT1 and mGluR5 expression in the cerebellum of FMR1 premutation carriers 66. M. Nawara et al., Novel mutation of IL1RAPL1 gene in a nonspecific X-linked mental retardation with FXTAS. Neurobiol. Aging 35, 1189–1197 (2014). (MRX) family. Am. J. Med. Genet. A. 146A, 3167–3172 (2008). 27. C. Sellier, Translation of expanded CGG repeats into FMRpolyG is pathogenic and may contribute to 67. C. Sellier et al., Sam68 sequestration and partial loss of function are associated with splicing fragile X tremor ataxia syndrome. Neuron 93, 331–347 (2017). alterations in FXTAS patients. EMBO J. 29, 1248–1261 (2010). 28. D. Garcia-Arocena, P. J. Hagerman, Advances in understanding the molecular basis of FXTAS. Hum 68. D. Hessl et al., Abnormal elevation of FMR1 mRNA is associated with psychological symptoms in Mol Genet 19, R83–89 (2010). 29. C. A. Doll, K. Scott, B. Appel, Fmrp regulates oligodendrocyte lineage cell specification and differentiation. Glia 69, 2349–2361 (2021). individuals with the fragile X premutation. Am. J. Med. Genet. B Neuropsychiatr. Genet. 139B, 115–121 (2005). 69. F. Tassone et al., Differential usage of transcriptional start sites and polyadenylation sites in FMR1 30. C. A. Doll, K. M. Yergert, B. H. Appel, The RNA binding protein fragile X mental retardation protein premutation alleles. Nucleic Acids Res. 39, 6172–6185 (2011). promotes myelin sheath growth. Glia 68, 495–508 (2020). 31. V. Martínez Cerdeño, Microglial cell activation and senescence are characteristic of the pathology FXTAS. Mov. Disord. 33, 1887–1894 (2018). 32. A. Giampetruzzi, J. H. Carson, E. Barbarese, FMRP and myelin protein expression in oligodendrocytes. Mol. Cell Neurosci. 56, 333–341 (2013). 70. S. Asamitsu et al., CGG repeat RNA G-quadruplexes interact with FMRpolyG to cause neuronal dysfunction in fragile X-related tremor/ataxia syndrome. Sci. Adv. 7, eabd9440 (2021). 71. S. N. Haify et al., Lack of a clear behavioral phenotype in an inducible FXTAS mouse model despite the presence of neuronal FMRpolyG-positive aggregates. Front. Mol. Biosci. 7, 599101 (2020). 72. P. K. Todd, CGG repeat-associated translation mediates neurodegeneration in fragile X tremor ataxia 33. N. Raj et al., Cell-type-specific profiling of human cellular models of fragile X syndrome reveal PI3K- syndrome. Neuron 78, 440–455 (2013). dependent defects in translation and neurogenesis. Cell Rep. 13, 108991 (2021). 73. J. Grigsby et al., Cognitive profile of fragile X premutation carriers with and without fragile 34. T. G. Lohith et al., Is metabotropic glutamate receptor 5 upregulated in prefrontal cortex in fragile X X-associated tremor/ataxia syndrome. Neuropsychology 22, 48–60 (2008). syndrome? Mol. Autism. 4, 15 (2013). 35. S. S. Tran et al., Widespread RNA editing dysregulation in brains from autistic individuals. Nat. Neurosci. 22, 25–36 (2019). 36. R. Esanov, N. S. Andrade, S. Bennison, C. Wahlestedt, Z. Zeier, The FMR1 promoter is selectively hydroxymethylated in primary neurons of fragile X syndrome patients. Hum. Mol. Genet. 25, 4870–4880 (2016). 74. A. G. Brega et al., The primary cognitive deficit among males with fragile X-associated tremor/ataxia syndrome (FXTAS) is a dysexecutive syndrome. J. Clin. Exp. Neuropsychol. 30, 853–869 (2008). 75. J. A. Brunberg et al., Fragile X premutation carriers: Characteristic MR imaging findings of adult male patients with progressive cerebellar and cognitive dysfunction. AJNR Am. J. Neuroradiol. 23, 1757–1766 (2002). 76. P. Teismann et al., Pathogenic role of glial cells in Parkinson’s disease. Mov. Disord. 18, 121–129 37. A. K. Gedeon et al., Fragile X syndrome without CCG amplification has an FMR1 deletion. Nat. Genet. (2003). 1, 341–344 (1992). 77. L. Liu et al., Glial lipid droplets and ROS induced by mitochondrial defects promote 38. P. Licznerski et al., ATP synthase c-subunit leak causes aberrant cellular metabolism in fragile X neurodegeneration. Cell 160, 177–190 (2015). syndrome. Cell 182, 1170–1185.e1179 (2020). 78. Y. Lee et al., Oligodendroglia metabolically support axons and contribute to neurodegeneration. 39. D. Velmeshev et al., Single-cell genomics identifies cell type-specific molecular changes in autism. Nature 487, 443–448 (2012). Science 364, 685–689 (2019). 40. B. B. Lake et al., Neuronal subtypes and diversity revealed by single-nucleus RNA sequencing of the human brain. Science 352, 1586–1590 (2016). 79. E. Croisier, M. B. Graeber, Glial degeneration and reactive gliosis in alpha-synucleinopathies: The emerging concept of primary gliodegeneration. Acta Neuropathol. 112, 517–530 (2006). 80. D. E. Bergles, W. D. Richardson, Oligodendrocyte development and plasticity. Cold Spring Harb. 41. C. M. Langseth et al., Comprehensive in situ mapping of human cortical transcriptomic cell types. Perspect. Biol. 8, a020453 (2015). Commun. Biol. 4, 998 (2021). 81. V. M. Fernandes, Z. Chen, A. M. Rossi, J. Zipfel, C. Desplan, Glia relay differentiation cues to 42. R. D. Hodge et al., Conserved cell types with divergent features in human versus mouse cortex. coordinate neuronal development in Drosophila. Science 357, 886–891 (2017). Nature 573, 61–68 (2019). 82. J. Nagai et al., Behaviorally consequential astrocytic regulation of neural circuits. Neuron 109, 43. S. Marques et al., Oligodendrocyte heterogeneity in the mouse juvenile and adult central nervous 576–596 (2021). system. Science 352, 1326–1329 (2016). 44. V. Kozareva et al., A transcriptomic atlas of mouse cerebellar cortex comprehensively defines cell types. Nature 598, 214–219 (2021). 83. K. A. Nave, Myelination and support of axonal integrity by glia. Nature 468, 244–252 (2010). 84. D. Lecca, S. Raffaele, M. P. Abbracchio, M. Fumagalli, Regulation and signaling of the GPR17 receptor in oligodendroglial cells. Glia 68, 1957–1967 (2020). 45. F. Tassone et al., CGG repeat length correlates with age of onset of motor signs of the fragile 85. S. Jacquemont et al., Protein synthesis levels are increased in a subset of individuals with fragile X X-associated tremor/ataxia syndrome (FXTAS). Am. J. Med. Genet. B. Neuropsychiatr. Genet. 144B, 566–569 (2007). syndrome. Hum. Mol. Genet 27, 3825 (2018). 86. H. S. Venkatesh et al., Neuronal activity promotes glioma growth through neuroligin-3 secretion. 46. E. M. Stanley, J. R. Fadel, D. D. Mott, Interneuron loss reduces dendritic inhibition and GABA release Cell 161, 803–816 (2015). in hippocampus of aged rats. Neurobiol. Aging 33, 431.e1-13 (2012). 47. T. Hua, C. Kao, Q. Sun, X. Li, Y. Zhou, Decreased proportion of GABA neurons accompanies age- related degradation of neuronal function in cat striate cortex. Brain Res. Bull. 75, 119–125 (2008). 48. A. Rozycka, M. Liguz-Lecznar, The space where aging acts: Focus on the GABAergic synapse. Aging Cell 16, 634–643 (2017). 49. M. Majdi, A. Ribeiro-da-Silva, A. C. Cuello, Cognitive impairment and transmitter-specific pre- and 87. R. J. Hagerman, Fragile X-associated neuropsychiatric disorders (FXAND). Front. Psychiat. 9, 3389 (2018). 88. A. Bhattacharyya, X. Zhao, Human pluripotent stem cell models of Fragile X syndrome. Mol. Cell Neurosci. 73, 43–51 (2016). 89. E. Donnard, H. Shu, M. Garber, Single cell transcriptomics reveals dysregulated cellular and molecular networks in a fragile x syndrome model. bioRxiv [Preprint] (2020). https://doi. org/10.1101/2020.02.12.946780. postsynaptic changes in the rat cerebral cortex during ageing. Eur. J. Neurosci. 26, 3583–3596 (2007). 90. Y. Kang et al., A human forebrain organoid model of fragile X syndrome exhibits altered 50. P. Hagerman, Fragile X-associated tremor/ataxia syndrome (FXTAS): Pathology and mechanisms. Acta Neuropathol. 126, 1–19 (2013). 51. C. R. Casingal, T. Kikkawa, H. Inada, Y. Sasaki, N. Osumi, Identification of FMRP target mRNAs in the developmental brain: FMRP might coordinate Ras/MAPK, Wnt/beta-catenin, and mTOR signaling during corticogenesis. Mol. Brain 13, 167 (2020). neurogenesis and highlights new treatment strategies. Nat. Neurosci. 24, 1377–1391 (2021). 91. M. T. Heneka, J. J. Rodriguez, A. Verkhratsky, Neuroglia in neurodegeneration. Brain Res. Rev. 63, 189–211 (2010). 92. C. A. Walsh, Human Autism Genetics. dbGaPDirect. https://www.ncbi.nlm.nih.gov/projects/gap/cgi- bin/study.cgi?study_id=phs000639.v2.p1. Deposited 12 May 2023. 12 of 12 https://doi.org/10.1073/pnas.2300052120 pnas.org
10.1016_j.cell.2023.06.002	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Cell. Author manuscript; available in PMC 2023 July 27. Published in final edited form as: Cell. 2023 July 20; 186(15): 3148–3165.e20. doi:10.1016/j.cell.2023.06.002. Vaccine-boosted CAR T crosstalk with host immunity to reject tumors with antigen heterogeneity Leyuan Ma1,8,9,, Alexander Hostetler1,#, Duncan M. Morgan1,4,#, Laura Maiorino1,#, Ina Sulkaj1, Charles A. Whittaker1, Alexandra Neeser10, Ivan Susin Pires1, Parisa Yousefpour1, Justin Gregory1, Kashif Qureshi1, Jonathan Dye1, Wuhbet Abraham1, Heikyung Suh1, Na Li1, J. Christopher Love1,4,5,6, Darrell J. Irvine1,2,3,5,7,11, 1David H. Koch Institute for Integrative Cancer Research, MIT, Cambridge, MA, 02139, United States 2Department of Materials Science and Engineering, MIT, Cambridge, MA, 02139, United States 3Department of Biological Engineering, MIT, Cambridge, MA, 02139, United States 4Department of Chemical Engineering, MIT, Cambridge, MA, USA 5Ragon Institute of Massachusetts General Hospital, Cambridge, MA, 02139, United States 6Broad Institute of MIT and Harvard, Cambridge, MA, 02142, United States 7Howard Hughes Medical Institute, Chevy Chase, MD, 20815, United States 8Department of Pathology and Laboratory Medicine, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104, USA 9The Raymond G. Perelman Center for Cellular and Molecular Therapeutics, Children’s Hospital of Philadelphia, Philadelphia, PA 19104, USA 10Department of Bioengineering, School of Engineering and Applied Science, University of Pennsylvania, Philadelphia, PA 19104, USA This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Correspondence: [email protected] (L.M.), [email protected] (D.J.I). #These authors contributed equally to this work AUTHOR CONTRIBUTIONS L.M., D.J.I. and J.C.L designed the studies. L.M., D.M.M, D.J.I. analyzed and interpreted the data and wrote the manuscript. L.M. performed the experiments. L.M. and I.S.P. carried out amphiphile-peptide vaccine synthesis. I.S. and A.H. assisted with CAR T production, sample preparation, and ELISPOT assays. D.M.M. and C.W. performed single cell and bulk RNA-sequencing analysis. L.M., A.H. assisted with necropsy and flow cytometry, P.Y assisted with Env/p15E antigen validation. A.N., J.G., J.D., K.Q., H.S., A.L., W.A., and N.L. assisted in assay preparation. Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. DECLARATION OF INTERESTS L.M. and D.J.I. are inventors on patents filed related to the amphiphile-vaccine technology. D.J.I. is a co-founder, shareholder, and consultant for Elicio Therapeutics, which has licensed patents related to the amphiphile- vaccine technology. INCLUSION AND DIVERSITY We support inclusive, diverse, and equitable conduct of research. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. 11Lead contact SUMMARY Page 2 Chimeric Antigen Receptor (CAR) T-cell therapy effectively treats human cancer, but loss of the antigen recognized by the CAR poses a major obstacle. We found that in vivo vaccine boosting of CAR T-cells triggers engagement of the endogenous immune system to circumvent antigen-negative tumor escape. Vaccine-boosted CAR-T promoted dendritic cell (DC) recruitment to tumors, increased tumor antigen uptake by DCs, and elicited priming of endogenous anti- tumor T-cells. This process was accompanied by shifts in CAR-T metabolism toward oxidative phosphorylation and was critically dependent on CAR T-derived IFN-γ. Antigen spreading induced by vaccine-boosted CAR T enabled a proportion of complete responses even when the initial tumor was 50% CAR-antigen-negative, and heterogenous tumor control was further enhanced by genetically amplifying CAR T IFN-γ expression. Thus, CAR T-cell-derived IFN-γ plays a critical role in promoting antigen spreading, and vaccine boosting provides a clinically- translatable strategy to drive such responses against solid tumors. Graphical Abstract In-brief Vaccine boosting modifies CAR T cell metabolism and promotes crosstalk between CAR T cells and endogenous immunity to elicit and sustain antigen spreading, thereby effectively treating tumors with antigen heterogeneity. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. INTRODUCTION Page 3 Adoptive cell therapy (ACT) using chimeric antigen receptor (CAR) T-cells has revolutionized the treatment of relapsed/refractory CD19+ B-cell acute lymphoblastic leukemia and lymphomas1–5. In the setting of solid tumors, CAR T therapy has been less successful so far, though progress is being made to address issues such as limited tumor infiltration, poor CAR T functionality and persistence1,6–8. However, two key challenges in the treatment of tumors with CAR T-cells are pre-existing antigenic heterogeneity, where not all tumor cells express the antigen targeted by the CAR, and antigen loss occurs during treatment. For example, a recent first-in-human clinical trial assessing CAR T-cells targeting mutant EGFRvIII in glioblastoma resulted in the emergence of EGFRvIIInull tumors9. Even in leukemia patients initially responding to CD19 CAR T therapy, loss or downregulation of the CD19 antigen has been frequently observed and often results in disease relapse10. An additional mechanism of antigen loss is via inflammation-induced dedifferentiation in melanomas11. These observations highlight the need for novel approaches to address antigen-loss-mediated tumor escape. Antigen spreading (AS) is the induction and amplification of immune responses to secondary antigens distinct from the original therapeutic target12. In the setting of adoptive cell therapy, strategies to target one surface-expressed antigen using CAR T-cells while inducing endogenous T-cell responses against additional tumor antigens would be an attractive approach to overcome tumor heterogeneity and antigen loss-mediated escape. Accumulating evidence suggests that AS can be elicited and may contribute to the overall therapeutic outcome during cancer immunotherapy. For example, recruitment and expansion of tumor-specific T-cells that were undetectable prior to therapy was found in patients receiving Ipilimumab13. Some cancer patients treated with neoantigen vaccines also exhibited AS towards shared neoantigens or cancer testis antigens14,15. In addition, increased anti-tumor antibody responses or weak T-cell responses were documented in a few cases of pre-clinical and clinical CAR T-cell therapy16–18. Nonetheless, to date there is limited evidence of CAR T-cell therapy itself inducing therapeutically meaningful AS. Preclinically, a majority of CAR-T studies employ immunodeficient mice that by definition exclude endogenous T-cell responses. In immunocompetent mouse models, CAR T therapy itself seems to have limited ability to trigger AS especially in solid tumors21. By contrast, CAR T-cells engineered with additional immune response-provoking molecules, including FLT3L22, CD40L23, IL-1224,25, IL-1826, IL-7/CCL1927, or when used in combination with oncolytic viruses28,29, have been reported to exhibit increased anti-tumor activity as well as evidence for AS. However, introduction of such additional effector functions to CAR T-cells with uniform activity across patients can be challenging and lead to new safety risks30,31. More importantly, irrespective of the CAR T-cell modality, mechanisms by which AS is promoted during adoptive cell therapy remain poorly understood. We recently described an approach to amplify CAR-T activity in solid tumors by vaccine- like boosting of CAR T-cells via their chimeric antigen receptor in lymph nodes32. This was accomplished by the synthesis of CAR ligands conjugated to an amphiphilic polymer-lipid tail, which following parenteral injection, efficiently traffic to draining lymph nodes and decorate the surfaces of macrophages and dendritic cells (DCs) with CAR-T ligands. CAR Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 4 T-cells encountering ligand-decorated DCs in the lymph node receive stimulation through the CAR in tandem with native costimulatory receptor signals and cytokine stimulation from the ligand-presenting cell, leading to CAR T-cell expansion and enhanced functionality. Vaccine boosting of CAR T-cells via administration of these “amph-ligands” together with vaccine adjuvants substantially enhanced tumor rejection by CAR T-cell therapy, and unexpectedly, was accompanied by the development of endogenous anti-tumor T-cell responses32. Here we used this approach of CAR-T therapy in tandem with vaccine boosting as a model setting to understand the role of antigen spreading in the clearance of antigenically heterogenous solid tumors, and to define mechanisms underlying AS. In multiple murine syngeneic tumor models, we found that AS elicited by CAR T-cell therapy using second- generation CARs was negligible. However, endogenous T-cell priming could be markedly induced by vaccine boosting of CAR T-cells, even in the context of lymphodepletion preconditioning. This process was critically dependent on IFN-γ, and enhanced IFN-γ expression induced either by vaccine boosting or genetic engineering enabled CAR T-cells to control solid tumors with preexisting antigen heterogeneity. RESULTS Vaccine boosting enables CAR T-cells to elicit endogenous CD4+ and CD8+ T-cell responses in multiple tumor models The amph-ligand-based vaccine boosting approach is illustrated schematically in Figure 1A: Amph-ligands are comprised of a ligand for a selected CAR linked to a hydrophobic phospholipid tail via a poly(ethylene glycol) (PEG) spacer. Upon co-injection with a suitable vaccine adjuvant at a site distal from the tumor, amph-ligands bind to albumin present in the interstitial fluid and are efficiently transported to the downstream draining lymph nodes (dLNs)33. Within the densely packed LN parenchyma, the amph-ligand transfers into cell membranes, decorating primarily the surface of macrophages and dendritic cells that line the subcapsular sinus and collagen conduits carrying lymph into the T-cell paracortex32. The co- administered adjuvant simultaneously activates DCs in the dLN to upregulate expression of costimulatory receptors and produce cytokines. CAR T-cells encountering ligand-decorated, activated DCs are stimulated in a manner mimicking natural T-cell priming, leading to CAR T-cell expansion and enhanced effector functions. Unexpectedly, we found that vaccine-boosted CAR T-cells also induce the expansion of endogenous anti-tumor T-cell responses32. We first assessed how the composition of the boosting vaccine impacts this antigen spreading response in a syngeneic murine EGFRvIII+CT-2A glioblastoma model. In this model, CAR T-cells targeting mutant EGFR (mEGFRvIII) are vaccine boosted using an amph-ligand comprised of an mEGFRvIII-derived peptide epitope recognized by the CAR T-cells32 (Figure S1A) combined with the potent STING agonist vaccine adjuvant cyclic di-GMP. Animals received lymphodepletion, followed one day later by s.c. injection of amph-ligand alone, adjuvant alone, or the full vaccine (amph-ligand + adjuvant). Amph- ligand/adjuvant was administered again 7 days later as a second boost, and then splenocytes were isolated at day 21 and co-cultured with irradiated EGFRvIII− CT-2A cells in an IFN-γ Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 5 ELISPOT assay to detect endogenous T-cell responses against non-CAR T-targeted antigens (Figure 1B). Endogenous lymphocyte and dendritic cell numbers were still recovering across the time course of these experiments following lymphodepletion (Figure S1B), However, their recovery was sufficiently rapid to permit robust de novo endogenous T-cell priming, consistent with prior preclinical studies reporting antigen spreading following lymphodepleting therapy22. Injection of the amph-ligand alone without adjuvant failed to initiate endogenous T-cell priming, while CAR T treatment in tandem with vaccine adjuvant alone elicited low but detectable endogenous T-cell responses (Figure 1B). However, the full vaccine (amph-ligand + adjuvant) led to 6-fold greater endogenous T-cell priming. This antigen spreading response did not reflect a direct effect of the vaccine on tumors, as inoculating tumors distal from the vaccine injection site did not change the antigen spreading response (Figure S1C). Similar magnitudes of endogenous T-cell priming were also observed with alternative adjuvants (TLR7/8 agonist Resiquimod, or the TLR9 agonist CpG, Figure S1D). Further analysis revealed that while CAR T therapy elicited no statistically significant endogenous anti-tumor CD8+ T-cell response and only a weak (but detectable) CD4+ T-cell response compared to untreated tumors, CAR-T combined with amph-ligand vaccination (hereafter, CAR T-vax) primed robust responses from both the CD4+ and CD8+ T-cell compartments (Figure 1C). To evaluate AS in a tumor model carrying a defined T-cell antigen, we assessed vaccine- boosted CAR-T treatment in a second model of B16F10 murine melanoma expressing the surrogate antigen ovalbumin (OVA), treated with bispecific FITC/TA99 CAR T-cells recognizing FITC and the melanoma-associated antigen Trp1 (Figure S1E). In this model, CAR T-cells are boosted by vaccination with amph-FITC and attack the tumor through Trp1 recognition. By ELISPOT, we observed host T-cell responses to both the model antigen OVA (Figure 1D) and B16F10 neoantigens (Figure 1E), but only when mice received both CAR T-cells and vaccine boosting. As shown in Figure 1F–G, quantifying CD8+ T-cells targeting the immunodominant OVA epitope SIINFEKL by peptide-MHC tetramer staining, no OVA-specific T-cells were detected in mice receiving CAR T-cells alone, but CAR T-vax therapy elicited a readily detectable SIINFEKL-specific T-cell response. Finally, to evaluate whether vaccine boosting could promote antigen spreading in a setting of a CAR T-cell targeting an endogenous tumor-associated antigen without the presence of an overexpressed neoantigen, we treated parental B16F10 tumors with FITC/TA99 CAR T-cells (Figure 1H). FITC vaccine alone or FITC/TA99 CAR T-cells alone elicited no AS above baseline. Vaccine boosting of a CAR that cannot recognize the tumor (FITC CAR T-vax) also failed to elicit antigen spreading, but vaccine boosting of FITC/TA99 CAR T-cells led to readily detectable host T-cell responses directed against non-Trp1 tumor antigens (Figure 1H). Thus, in three different tumor models using two different CARs, CAR T-cell treatment combined with amph-vax boosting promoted antigen spreading. Vaccine-boosted CAR T-cells drive functional and phenotypic changes in endogenous T-cells We analyzed tumor-infiltrating lymphocytes (TILs) by flow cytometry on day 7 post CAR T-vax treatment and observed substantially increased endogenous CD8+ TILs and a trend toward increased CD4+ cells (Figure 2A). A similar increase of host T-cell infiltration was Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 6 found by adding vaccine boosting to CAR-T therapy treatment of OVA-expressing CT-2A tumors, including a 3-fold increase in bona fide tumor-antigen (OVA)-specific TILs (Figure S1F). We isolated host CD4+ and CD8+ TILs 7 or 14 days after treatment and carried out single-cell RNA-seq and paired α/β TCR sequencing on the recovered host lymphocytes (Figure 2B). Quality single-cell transcriptomes were obtained for 21,835 T-cells (Figure 2C–D). Unsupervised clustering of the transcriptome data revealed five major endogenous T-cell subsets: CD8+ cytotoxic T lymphocytes (CTLs, expressing Cd8a, Ccl5, Pdcd1), CD4+ T helper cells (Cd4, Cd40lg), Tregs (Foxp3, Il2ra, Ikzf2), a proliferating Ki-67+ population that included both CD4+ and CD8+ cells (Mki67, Top2a), and a small population of IFN- stimulated T-cells (characterized by expression of Ifit1, Ifit3, Isg15) (Figure 2D–E, S2A, Supplemental Table 1), as has been described previously34,35. We observed an increase in the frequency of the CD8+ CTL population in mice treated with CAR T-vax at both day 7 and day 14. Interestingly, we also observed a transient decrease in the frequency of Tregs at day 7 in mice treated with CAR T-vax compared to those treated with CAR-T alone (Figure 2E). We computed differentially expressed genes between CD8+ CTLs recovered from mice treated with CAR T-vax vs. CAR T alone. At day 14, CD8+ T-cells from CAR T-vax-treated mice upregulated transcripts associated with both cytotoxicity (Gzmb, Gzmk) and T-cell activation (Havcr2) relative to the CΑR-T alone group (Figure 2F, S2B–C, Supplemental Table 2). We validated these findings at the protein level by carrying out flow cytometry analysis of endogenous TILs. Compared to CAR T only therapy, vaccine boosting did not change the proportion of PD-1+TIM-3+ or PD-1+TIM-3− endogenous TILs (Figure S3A), but did enhance IFN-γ, TNF-α, and granzyme B expression in both populations (Figure S3B–C). Among CD4+ cells, we found an elevation of transcripts associated with Th17 function (Rorc, Il17a, Il17re) among mice treated with CAR-T alone at day 14 compared to day 7 (Figure 2F, Supplemental Table 2). By contrast, CD4+ Th cells from CAR T-vax-treated mice upregulated genes associated with Th1 function (Ifng, Cxcr3) and self-renewal (Slamf6, Tcf7) (Figure 2F, S2D–E), suggesting that the vaccine may also promote anti-tumor phenotypes among CD4+ TILs. Next, we sought to assess how CAR T-vax affects TILs according to their antigen specificities. Using data generated in a recent study defining TCR sequences specific for a common murine endogenous retroviral antigen p15E (Grace et al., 2022) that is also expressed by CT-2A cells (Figure S3D–E), we assessed the transcriptional state of tumor-specific endogenous TILs. At day 7, both p15E-specific T- cells and TILs of unknown specificity from CAR T-vax-treated mice exhibited significantly higher cytotoxicity than TILs from animals treated with CAR-T alone (Figure 2G, S2F–G, S3F, Supplemental Table 3). Overall, this analysis suggests that the addition of the vaccine to CAR-T therapy increases the anti-tumor potential of tumor-infiltrating CD8+ T-cells and skews the differentiation of tumor-infiltrating CD4+ T-cells to a Th1 phenotype. Vaccine-driven antigen spreading prevents relapse of antigen-loss variants and enables control of antigenically heterogenous tumors To determine if endogenous T-cells impact the outcome of CAR T-vax treatment, we treated wildtype (WT) or RAG1−/− mice bearing mEGFRvIII+CT-2A tumors with CAR T-cells ± vaccine boosting. CAR T-vax therapy in WT mice led to much greater tumor control Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 7 compared to CAR T-cells alone (Fig. 3A–C). In RAG−/− animals, CAR T-vax treatment also elicited a high frequency of initial tumor regressions, but a majority of tumors relapsed 20–50 days post treatment (Figure 3A–C). Analysis of relapsed tumors revealed that loss or down-regulation of EGFRvIII on tumor cells was a major escape mechanism in the RAG−/− animals (Figure 3D–E). These data suggested that endogenous lymphocytes are critical for the high frequency of complete responses observed in WT animals. Given the substantial effect of CAR T-vax treatment on cytotoxic effector gene expression in endogenous CD8+ T cells (Figure 2F), we evaluated the importance of endogenous CD8+ T cells in tumor control, by comparing CAR T-vax treatment in WT versus CD8α−/− tumor-bearing mice. Early tumor growth control was only modestly affected in the absence of endogenous CD8 T cells (Figure 3F), but long-term survival was almost completely abolished (Figure 3G). Encouraged by these findings, we tested whether endogenous T-cell priming could enable CAR T-cells to eliminate tumors with pre-existing antigenic heterogeneity. To this end, we inoculated a mixture of EGFRvIII+ CT-2A cells and parental EGFRvIII− CT-2A cells at defined ratios into both WT and RAG1−/− mice (Figure 3H). We previously showed that these two CT-2A variants have similar growth rates in WT mice32. When 100% of the tumor cells express EGFRvIII, CAR T-vax therapy elicited comparable initial tumor regressions in both WT and RAG1−/− mice, but long-term remission was only achieved in WT animals (Figure S3G). More strikingly, in heterogeneous tumors comprised of as little as 10% EGFRvIII− cells, CAR T-vax therapy delayed tumor progression but induced no actual regressions in RAG1−/− mice. By contrast, CAR T-vax treatment cured ~50% animals bearing tumors with up to 20% EGFRvIII− cells and could still achieve complete responses in a small proportion of animals when the EGFRvIII− population was 50% of the tumor mass at time zero. To confirm that vaccine boosting of CAR T-cell therapy could augment heterogeneous tumor control in the setting of a non-overexpressed tumor antigen, we also treated melanoma tumors comprised of a mixture of 80% parental and 20% Trp1−/− B16F10 tumor cells with bivalent FITC/TA99 CAR T-cells and amph-FITC vaccine. Treatment of this mixed tumor elicited readily detectable antigen spreading to non-Trp1 antigens (Figure S3H) and controlled tumor growth (Figure S3I). The drastic difference of therapeutic outcome in WT vs RAG1−/− mice demonstrates the pivotal role endogenous T-cells and AS can play in controlling tumors with pre-existing antigenic heterogeneity. Vaccine boosting induces cell-intrinsic enhancements in CAR T-cell function We next sought to understand how amph-vax boosting promotes endogenous T-cell priming. We first tested whether the anti-tumor efficacy of vaccine boosting was simply driven by increased numbers of CAR T-cells, vs. a change in CAR T function. CAR T-cells were transferred into non-tumor bearing mice, vaccine boosted (or not as controls), and then isolated 7 days later from the two groups and transferred at equal numbers into new tumor-bearing recipient mice (Figure 4A). This approach revealed that even when the same number of CAR T-cells were present, vaccine-boosted CAR T still exhibited enhanced tumor control and long-term animal survival, suggesting that vaccine boosting enhances the intrinsic per-cell functionality of CAR T-cells (Figure 4A). Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 8 To gain an unbiased view of changes in CAR-T function, we carried out bulk RNA-seq on CAR T-cells 8 days after adoptive transfer, with or without vaccine boosting. Vaccination increased the expression of genes associated with effector function and cell trafficking (e.g, FasL, Gzma, Gzmk, Ccl5, Itgb1) in CAR T-cells recovered from the spleen (Figure 4B, Supplemental Table 4); we confirmed the expression of several of these genes by quantitative PCR (Figure S4A). Gene set enrichment analysis (GSEA) of tumor-infiltrating cells further revealed that vaccine-boosted CAR T-cells maintained a high proliferative potential, as evidenced by elevated Myc and E2F target genes (Figure 4C). Vaccine-boosted cells also showed a significant upregulation of metabolic pathways, including oxidative phosphorylation (OXPHOS), MTORC1 signaling, fatty acid metabolism, and peroxisome signaling (Figure 4C). Prompted by these transcriptional signatures, we analyzed the intracellular expression of PGC-1α, a master transcription factor controlling many genes and pathways involved in OXPHOS36, and found that vaccine boosting increased PGC-1α levels in CAR T-cells (Figure 4D). PGC-1α is involved in mitochondria generation and maintenance37, and we noted increased mitochondria levels in vaccine-boosted CAR T-cells (Figure 4E). Notably, endogenous T-cell priming was significantly reduced ~50% following CAR T-vax treatment with PGC-1α−/− CAR T-cells compared to WT CAR T (Figure 4F). Hence, metabolic reprogramming in vaccine-boosted CAR T-cells is one factor promoting antigen spreading. Enhanced IFN-γ production by vaccine-boosted CAR T-cells is critical for induction of antigen spreading OXPHOS has been shown to be critical for maintaining the polyfunctionality of T-cells within the TME38, and we previously observed that vaccine-boosted CAR T-cells recovered from the peripheral blood showed increased cytokine production32. To determine if this enhanced effector function was maintained in tumors and impacted antigen spreading, we analyzed IFN-γ and TNF-α expression in TILs and found that both cytokines were markedly upregulated in vaccine-boosted CAR T-cells (Figure 5A). This enhanced cytokine production is partially linked to vaccine-induced metabolic changes, because IFN- γ expression was reduced in PGC-1α-deficient CAR T-cells (Figure 5B). Interestingly, although CAR T-cells ± vaccine exhibited comparable levels of PD-1 and Tim-3 expression, high-level cytokine production was maintained in both PD-1+Tim-3− and PD-1+Tim-3+ CAR T-cells that received vaccine boosting (Figure S4B–D). To assess the role of these cytokines in AS, we treated tumor-bearing mice with CAR T-vax therapy in the presence of neutralizing antibodies against IFN-γ or TNF-α. Therapy in the presence of isotype control or TNF-α-blocking antibodies had no impact on endogenous T-cell priming, but IFN-γ blockade completely abrogated AS, including both CD4+ and CD8+ T cell responses (Figure 5C, Figure S4E–F). To confirm this result, we repeated IFN-γ blockade experiments in a second model of OVA+EGFRvIII+CT-2A cells. CAR T-vax treatment expanded OVA- specific T-cells and induced IFN-γ-producing T-cells recognizing SIINFEKL, as determined by peptide-MHC tetramer staining and ELISPOT, respectively (Figure 5D–E). However, IFN-γ neutralization during treatment eliminated the OVA-specific T-cell response (Figure 5D–E). Further, endogenous T-cell infiltration and functional enhancement were also repressed by IFN-γ blockade (Figure S5, Supplemental Table 5). Administration of blocking antibodies at different time points during therapy revealed that IFN-γ was most critical for Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 9 promoting AS during the first week of treatment (Figure 5F). Blockade of IFN-γ using neutralizing antibodies also greatly reduced the efficacy of the treatment (Figure 5G–H). To determine what cells were the key producers of IFN-γ, we tested CAR T-vax therapy employing IFN-γ-deficient CAR T-cells; this treatment elicited no antigen spreading (Figure 5I) and tumor control was lost, demonstrating an important role for CAR T-derived cytokine (Figure 5J). Early tumor control trended toward lower efficacy when CAR T-vax therapy was applied to tumor-bearing IFN-γ-deficient hosts, but this did not reach statistical significance (Figure 5K). However, long-term tumor control and overall survival was strongly reduced in IFN-γ−/− mice (Figure 5L). Thus, CAR T-vax therapy amplifies CAR T-cell-derived IFN-γ that is critical for initial tumor control and antigen spreading, but also requires host-derived IFN-γ at later time points in the treatment, consistent with the important role for endogenous T cells in preventing tumor relapse. IFN-γ sustains vaccine-boosted CAR T effector functions, promotes DC recruitment and antigen uptake, and triggers IL-12-mediated CAR T-DC crosstalk Autocrine signaling from IFN-γ has been found to support the cytotoxicity of conventional T-cells39. To test if IFN-γ also promotes CAR T killing in a similar manner, we evaluated the cytotoxicity of IFN-γ−/− and IFNGR1−/− CAR T-cells against EGFRvIII+CT-2A cells in vitro and found that lack of IFN-γ or IFNGR1 expression by the CAR T-cells reduced cytotoxicity by ~50% (Figure 6A). Consistent with this finding, vaccine-boosted CAR-T with elevated IFN-γ expression also exhibited increased granzyme B levels in tumors (Figure S6A) and tumor cells exhibited increased signatures of immunogenic cell death, such as upregulated cell surface calreticulin expression (Figure S6B). We next examined the DC and macrophage compartment of treated tumors, as tumor antigen released by CAR T-mediated tumor killing must be acquired by antigen presenting cells to drive T-cell priming. Vaccine boosting of CAR-T led to substantial increases in macrophages and multiple DC populations infiltrating treated tumors, including plasmacytoid DCs (pDCs), CD8+ DCs, CD103+ cDC1s (10-fold increase), and CD11b+ cDC2s (11-fold increase) (Figure 6B). Intratumoral macrophages also showed a shift in phenotype with upregulation of costimulatory receptors and a reduction in CD206+ macrophages (Figure S6C–E). However, AS induced by CAR T-vax treatment was greatly reduced in Batf3−/− animals lacking cross-presenting DCs40,41 (Figure 6C), and hence we focused our attention on the DC compartment. DC recruitment to tumors relies on chemokines such as CCL3, CCL4 and CCL542,43, and intratumoral expression of these chemokines was reduced when treating with IFN-γ−/− CAR T-cells (Figure 6D). Ki67 expression was upregulated in CD11b+ and CD103+ DCs, suggesting a role for local expansion of intratumoral DCs in addition to recruitment from the circulation (Figure 6E). Using an EGFRvIII+CT-2A tumor line expressing ZsGreen as a traceable antigen, we found that vaccine boosting triggered DC activation as evidenced by upregulation of the lymph node homing marker CCR7, costimulatory receptors, and MHC-II (Figure 6F, S6F–K), and increased tumor antigen uptake by both cDC1 and cDC2 populations (Figure 6G–H). Consistent with the observed loss of AS with IFN-γ-deficient CAR T-cells, DC activation and tumor antigen uptake were lost if treatment was applied using IFN-γ−/− CAR T-cells (Figure S6H–K). We also Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 10 assayed for potential changes in the tumor vasculature following CAR T-vax treatment, but found it was not significantly affected (Figure S6L–N). Thus vaccine-boosting CAR T-cells amplified multiple prerequisite steps for antigen spreading. Our in vitro analysis suggested a role for autocrine CAR T-cell-derived IFN-γ in sustaining CAR T cytotoxicity, but the target cells responding to IFN-γ in vivo remained unclear. We first tested if host cells were important responders, by transferring CAR T-cells into tumor-bearing WT or IFNGR1−/− mice, followed by vaccine boosting. Endogenous T-cell priming and tumor control were completely lost in IFNGR1−/− mice (Figure 6I–J). Next, we generated mice with specific deletion of IFNGR1 in CD11c+ DCs by crossing CD11c-cre and IFNGR-floxed animals to generate CD11cΔIFNGR1 mice. As shown in Figure 6K, CAR T-vax treatment of tumor-bearing CD11cΔIFNGR1 mice led to reduced but not fully ablated endogenous T-cell priming, suggesting that DCs are important responders but not the sole host cell population stimulated by IFN-γ. Activation of dendritic cells by T-cell-derived IFN-γ has been shown to trigger production of IL-12 by DCs, which in turn acts as positive feedback signal reinforcing T-cell IFN-γ expression and cytotoxic activity during checkpoint blockade immunotherapy44. Strikingly, antibody-mediated neutralization of IL-12 during CAR T-vax therapy or treatment of IL-12-deficient mice eliminated antigen spreading comparably to IFN-γ blockade (Figure 6L–M). The CAR T-cells themselves are important responders to IL-12, as therapy with IL-12Rb2−/− CAR T-cells elicited nearly baseline endogenous T-cell priming in 4 of 5 animals (Figure 6M). IL-12 drives sustained/elevated autocrine IFN-γ expression by T-cells. In vivo, vaccine- boosted IFNGR1-deficient CAR T-cells showed reduced production of IFN-γ, granzyme B and a trend toward reduced levels of TNF-α (Figure S7A–D). Blunted effector functions of IFNGR1-deficient CAR T-cells correlated with reduced induction of immunogenic cell death markers on tumor cells (Figure S7E), decreased tumor antigen uptake by intratumoral DCs (Figure S7F–G), and reduced tumor antigen acquisition by lymph node-resident CD8α+ cDC1 (Figure S7H); tumor antigen uptake by LN cDC2 was low and unaffected (Figure S7I). These changes in CAR-T function, tumor killing, and tumor antigen release correlated with complete loss of endogenous T-cell priming and tumor control when tumor- bearing animals were treated with CAR T-vax therapy using IFNGR1−/− CAR T-cells (Figure 6N–O). Altogether, vaccine boosting enables CAR T-cells to sustain cytotoxicity in the TME and drive key events required for antigen spreading, dependent both on the ability of host DCs and the CAR T-cells themselves to respond IFN-γ. Robust IFN-γ production is essential for CAR T-vax therapy to control tumors with pre- existing antigen heterogeneity Based on our collective mechanistic findings regarding the importance of IFN-γ in AS, we finally assessed the role of IFN-γ in promoting control of antigenically heterogeneous tumors. Using mixed tumors comprising 80% EGFRvIII+ and 20% EGFRvIII− tumor cells, CAR T-vax therapy in the presence of IFN-γ blockade led to loss of survival extension and elicited no complete responses (Figure 7A–B); similar results were obtained when IL-12 was blocked (Figure 7C–D). We hypothesized that enforced expression of IFN-γ might further enhance endogenous T-cell priming elicited by CAR T-vax therapy. To test this Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 11 idea, we transduced CAR T-cells with retroviral constructs bearing an NFAT-driven IFN-γ expression cassette, to obtain elevated IFN-γ production following CAR activation45. We confirmed that NFAT-IFN-γ CAR T-cells produced nearly twice as much of IFN-γ as WT CAR T-cells upon stimulation in vitro (Figure 7E). Non-vaccine boosted NFAT-IFN-γ CAR T therapy elicited a significant level of endogenous T-cell priming, consistent with a critical role for sustained CAR T-cell IFN-γ in AS generally (Figure 7F). AS was further increased when NFAT-IFN-γ CAR T were used in combination with vaccine boosting, reaching 50% higher levels than treatment with WT CAR T-cells (Figure 7F). Vaccine boosting of NFAT-IFN-γ CAR T-cells led to slight trends toward increased CAR T-cell numbers in the tumor and increased IFN-γ and granzyme expression, but these did not reach statistical significance (Figure 7G–I). By contrast, endogenous T cell infiltration and granzyme expression were enhanced for NFAT-IFN-γ CAR T-vax therapy compared to CAR T-vax treatment, and IFN-γ showed a trend toward increased expression (Figure 7J–L). Vaccine- boosted WT CAR T-cells were able to reject 25–50% of 80:20 EGFRvIII+:EGFRvIII− mixed tumors (Figure 7A–B, M–N). NFAT-IFN-γ CAR T-cells achieved similar complete response rates in the absence of vaccine boosting, and strikingly, this complete response rate increased to 80% when vaccine boosting was added to the treatment (Figure 7M– N). Importantly, vaccine boosting of NFAT-IFN-γ CAR T-cells was accompanied by mild elevations in systemic IFN-γ following the first vaccine boost, and only mild transient weight loss in animals that rapidly recovered after each vaccine boost (Figure S7J–K). Thus, strategies to enhance IFN-γ production and favorable CAR T-cell metabolism appear promising to increase the efficacy of CAR T-cell therapy against antigenically heterogenous solid tumors. Discussion Antigenic heterogeneity and antigen loss play important roles in tumor escape from immune surveillance and resistance to CAR-T therapies46–48. The induction of antigen spreading by CAR-T therapy could address this challenge, but evidence for AS during ACT in humans remains limited. Preclinical studies using combination therapies or CAR T-cells transduced with one or more supporting genes have reported induction of AS, but mechanisms governing these responses remain poorly understood. Here we found that T-cells bearing second-generation CARs, which receive in vivo restimulation via a vaccine activating the CAR in lymph nodes, are capable of promoting robust host CD4+ and CD8+ T-cell responses against non-CAR-related tumor antigens. This endogenous T-cell response has significant consequences for the outcome of CAR T therapy: (1) long-term tumor regressions and complete responses are achieved against tumors that otherwise undergo antigen loss-based relapse; (2) control of antigenically heterogeneous tumors can be achieved; and (3) long term protection against tumor rechallenge is achieved. Mechanistically, we found that enhanced production of IFN-γ by vaccine-boosted CAR T-cells was a major contributor to antigen spreading. In natural immune responses, IFN-γ promotes the activation of both innate and adaptive immunity49, maintenance of T-cell cytotoxicity and mobility39, polarization of T helper cells to Th1 cells50, reduction of Treg-mediated suppression51 and sensitization of tumors to T-cell-mediated cytotoxicity50. However, the role of IFN-γ in the function of CAR T-cells remains poorly defined. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 12 Recently, IFN-γ was shown to regulate the expression of cell adhesion molecules on solid tumor cells, but not leukemic cells, and subsequently enhance CAR T-cell cytotoxicity by stabilizing CAR T-tumor cell engagement52. Alizadeh et al. have also demonstrated that CAR T-cell-derived IFN-γ can promote recruitment of endogenous immune cells to tumors and shift the phenotype of intratumoral myeloid cells toward anti-tumor phenotypes20. Here, although both host T cells and CAR-T cells are IFN-γ producers in the TME, we found that IFN-γ production by CAR T-cells was most critical to enable an antigen spreading response. IFN-γ sustained high levels of cytotoxicity and effector cytokine expression in vaccine-boosted CAR T-cells in a cell-intrinsic manner. These enhanced CAR T-cell effector functions in turn correlated with increased expression of DC-recruiting chemokines in tumors, increased DC infiltration, tumor antigen uptake, and activation of intratumoral DCs. These effects of CAR T-derived IFN-γ were propagated via a positive feedback loop involving DC-derived IL-12. Such IFN-γ-IL-12 crosstalk has proven to underlie a number of successful immunotherapies, including checkpoint blockade therapy44 and CAR T-cell therapy in lymphoma53. Our data do not exclude potential contributions of other cytokines or immune cell types, such as tumor-resident macrophages, which might also play a role in the endogenous immune response20. IFN-γ production is tightly regulated at both the transcriptional level by transcription factors (TFs)54,55 including CREB, AP-1, T-bet, NFAT, and at the post-transcriptional level by various miRNAs, ARE or GAPDH binding to its 3’UTR56,57. Although IFN-γ synthesis has been proposed to be predominantly associated with glycolysis due to its regulation by GAPDH56, both glycolysis and oxidative phosphorylation have been shown to control IFN-γ production in NK cells58,59, consistent with previous reports that elevated OXPHOS and mitochondria integrity was required to support IFN-γ production58,60,61. These findings align with our observation that genetic deletion of PGC-1α, a key transcription factor regulating OXPHOS, resulted in reduced expression of IFN-γ and a significant reduction in AS. OXPHOS is often an important feature of memory-like T-cells62, and enforced expression of PGC-1α endows T-cells with superior anti-tumor activity63. The extent to which other metabolic pathways and which gene(s), including GAPDH, are responsible for IFN-γ production by vaccine-boosted intratumoral CAR-T cells will require future investigation. In summary, we have shown that vaccine boosting through the chimeric receptor triggers markedly enhanced CAR-T polyfunctionality and metabolic reprogramming (Figure 7O). Vaccine-boosted CAR-T cells trigger robust recruitment and activation of DCs in the tumor, which in turn secrete IL-12 that, together with the autocrine effect of IFN-γ, enhances CAR T-cell anti-tumor activity (Figure 7O), leading to pronounced endogenous T-cell priming and induction of enhanced effector programs in endogenous T-cells that infiltrate tumors. In our models, we find that such antigen spreading is critical for avoidance of antigen loss-mediated tumor escape and control of antigenically heterogenous tumors. As few solid tumors express target antigens on >90% of tumor cells, these findings provide guidance for engineering more effective CAR-T therapies. Notably, vaccines for CAR T-cells are already being explored clinically64–66, suggesting this approach can be readily translated to CAR-T cell clinical trials. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Limitations of the Study Page 13 We elected to use a glioblastoma model (CT-2A) transduced to express the GBM mutant antigen EGFRvIII implanted in the flank for many of our studies, which could be mixed with parental EGFRvIII-CT-2A cells in distinct ratios to quantify the impact of antigen spreading. This provided a model system where antigen heterogeneity was well defined and allowed experimental throughput for mechanistic studies that would not be possible in an orthotopic GBM model, but does not model the orthotopic GBM microenvironment or natural EGFRvIII expression heterogeneity. We did however evaluate CAR T-vax therapy targeting endogenous tumor-associated antigens to confirm the key findings of antigen spreading and heterogenous tumor control in a model lacking artificially introduced antigens. We also focused our studies on syngeneic mouse models, as immunodeficient mouse hosts used for preclinical human CAR T-cell therapy lack proper lymphatic and lymph node formation, which is problematic for the vaccine boosting treatment. STAR Methods RESOURCE AVAILABILITY Lead Contact—Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact Darrell Irvine ([email protected]). Materials Availability—New plasmids from this paper are available from the lead contact upon request. Data and Code Availability • • • Bulk-RNA seq and single cell RNA-seq data have been deposited at GEO (GSE211938, GSE212453) and are publicly available as of the date of publication. Codes used to process and analyze single-cell RNA-seq data are available at github.com/duncanmorgan/CAR_AgSpreading or Zenodo (10.5281/ zenodo.7939518). Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request. EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS Cell line and Constructs—B16F10 and 293 phoenix cells were obtained from ATCC. B16F10-OVA cells were a gift from Dr. Glen Dranoff at the Dana Farber Cancer Institute. TRP1−/− B16F10 cells were generated previously using CRIPSR 70. The mouse CT-2A glioma cell line was kindly provided by Dr. Thomas Seyfried from Boston College. mEGFRvIII-expressing CT-2A cells were generated by lentiviral transduction of CT-2A cells with a murine version of EGFRvIII and stably selected with puromycin. ZsGreen+ mEGFRvIII-CT-2A cells were generated by transducing mEGFRvIII-CT-2A cells with ZsGreen-expressing lentivirus and subsequent flow cytometry enrichment. mEGFRvIII- CT-2A-OVA cells were generated by transducing mEGFRvIII-CT-2A cells with PLKO- based lentivirus expressing Thy1.1-IRES-OVA (aa251–388). MHCII+ CT-2A cells were Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 14 generated by transducing CT-2A cells with lentivirus expressing CIITA (Class II Major Histocompatibility Complex Transactivator). Animals—Female mice (6–8 week old) were used for all studies. Wildtype female C57BL/6J mice (B6, CD45.2+), CD45.1+ congenic mice (B6.SJL-Ptprca Pepcb/BoyJ), Rag1−/− (B6.129S7-Rag1tm1Mom/J, B6 background), IFN-γ−/− (B6.129S7-Ifngtm1Ts/J, congenic with B6, backcrossed for at least 8 generations ), IFNGR1−/− (B6.129S7- Ifngr1tm1Agt/J, B6 background), Batf3−/− (B6.129S(C)-Batf3tm1Kmm/J, B6 background), PGC-1α-flox (B6N.129(FVB)-Ppargc1atm2.1Brsp/J), LCK-cre (B6.Cg-Tg(Lck-cre)548Jxm/J, Hemizygous), IL12rb2−/− (B6;129S1-Il12rb2tm1Jm/J, B6 background), IL12p40−/− (B6.129S1-Il12btm1Jm/J, congenic with B6, backcrossed for at least 9 generations), CD11c- cre (C57BL/6J-Tg(Itgax-cre,-EGFP)4097Ach/J, Hemizygous)，IFNGR1-flox (C57BL/6N- Ifngr1tm1.1Rds/J) mice and CD8α−/− (B6.129S2-Cd8atm1Mak/J) mice were purchased from the Jackson Laboratory. To avoid neonatal lethality caused by whole body KO of PGC-1α, T cell-specific PGC-1α KO mice were created by crossing LCK-cre mice with PGC-1α-flox mice; cre+ F1 offspring have T cell-specific PGC-1α KO while the cre− F1 offspring have a wildtype phenotype and were used as donor control T cells for Fig 3F. CD11cΔIFNGR1 mice were generated by crossing CD11c-cre mice with IFNGR1-flox mice, cre+ F1 offspring are IFNGR1-deficient in CD11c+ cells while the cre− F1 offspring have a wildtype phenotype and were used as control recipients in Fig. 6H. For all studies, 6–8 weeks old mice were used. All animal studies were carried out following an IACUC-approved protocol following local, state, and federal guidelines. METHOD DETAILS Cloning and constructs—The murine EGFRvIII CAR (28z) and FITC/TA99 bispecific CAR (28z) were cloned into an MSCV retroviral vector as previously described32. The NFAT-IFN-γ cassette was constructed in a self-inactivating (SIN)-retroviral vector with 6xNFAT binding sites71 upstream of the minimal IL2 promoter driving murine IFN-γ expression. Primary mouse T cell isolation and CAR T-cell production—For T cell activation, 6-well plates were pre-coated with 5 ml of anti-CD3 (0.5 μg/ml, Clone: 2C11) and anti- CD28 (5 μg/ml, Clone: 37.51) per well at 4°C for 18 hr. CD8 + T cells were isolated using a negative selection kit (Stem Cell Technology), and seeded onto pre-coated 6-well plates at 5 ×106 cells/well in 5 ml of complete medium (RPMI + penicillin/streptomycin + 10% FBS + 1x NEAA + 1x Sodium pyruvate + 1x 2-mercaptoethanol + 1x ITS [Insulin-Transferrin- Selenium, Thermo Fisher]). Cells were cultured at 37°C for 48 hr without disturbance. Twenty-four hr before transduction, non-TC treated plates were coated with 15 μg/ml of retronectin (Clonetech). On day 2, cells were collected, counted and resuspended at 2×106 cells/ml in complete medium supplemented with 20 μg/ml of polybrene and 40 IU/mL of mIL-2. Retronectin-coated plates were blocked with 0.05% FBS containing PBS for 30 min before use. 1 ml of virus supernatant was first added into each well of the blocked retronectin plate, then 1 mL of the above cell suspension was added and mixed well by gentle shaking to reach the working concentration of polybrene at 10 μg/ml and mIL-2 at 20 IU/ml. Spin infection was carried out at 2000×g for 120 min at 32°C. Plates were then Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 15 carefully transferred to an incubator and maintained overnight. On day 3, plates were briefly centrifuged at 1,000×g for 1 min, and virus-containing supernatants were carefully removed. 3 mL of fresh complete medium containing 20IU/ml of mIL-2 were then added into each well. Cells were passaged 1:2 every 12 hr with fresh complete medium containing 20IU/mL of mIL-2. Transduction efficiency was evaluated by surface staining of a c-Myc tag included in the CAR construct32 using an anti-Myc antibody (Cell signaling, Clone:9B11) ~30 hr after transduction. If needed, CAR T-cells on day 3, after flow cytometry analysis of virus transduction, could be frozen down and stored for assays at a later time. For in vivo experiments, CAR T-cells were used on day 4. For in vitro experiments, CAR T-cells were cultured till day 5. Virus production and transduction evaluation—For optimal retrovirus production, 293 phoenix cells were cultured till 80% confluence, then split at 1:2 for further expansion. 24 hr later, 5.6×106 cells were seeded in a 10 cm dish and cultured for 16 hr till the confluency reached 70%. 30 min – 1 hr before transfection, each 10 cm dish was replenished with 10 ml pre-warmed medium. Transfection was carried out using the calcium phosphate method following the manufacturer’s protocol (Clonetech). Briefly, for each transfection, 18 μg of plasmid (16.2 μg of CAR plasmid plus 1.8 μg of Eco packaging plasmid) was added to 610 μl of ddH 2O, followed by addition of 87 μl of 2 M CaCl 2. 700 μl of 2x HBS was then added in a dropwise manner with gentle vortexing. After a 10 min incubation at 25°C, the transfection mixture was gently added to phoenix cells. After 30 min incubation at 37°C, the plate was checked for the formation of fine particles, as a sign of successful transfection. The next day, old medium was removed and replenished with 8 ml of pre-warmed medium without disturbing the cells. Virus-containing supernatant was collected 36 hr later and passed through a 0.45 um filter to remove cell debris, designated as the “24hr” batch. Dishes were refilled with 10ml of fresh medium and cultured for another 24 hr to collect viruses again, designated as the “48hr” batch, this process can be repeated for another two days to collect a “72hr” batch and “96hr” batch. All virus supernatant was aliquoted and stored at −80°C. Virus transduction rate was evaluated in a 12-well format by mixing 0.5 million activated T cells with 0.5ml of viruses from each batch. Plate coating, spin infection and FACS analysis of CAR expression were carried out as described above. In the majority of experiments, the “48hr” and “72hr” batches yielded viruses that transduced T cells at 90–95% efficiency, the “24hr” and “96hr” batch viruses led to >80% transduction. Only viruses with >90% transduction rate were used for animal studies. Amphiphile-ligand production and vaccination—DSPE-PEG-FITC was purchased from Avanti. Amph-pepvIII was produced as previously described72. Briefly, pepvIII peptides (LEEKKGNYVVTDHC) were dissolved in dimethylformamide at 10 mg/mL and mixed with 2.5 equivalents of 1,2-distearoyl-sn-glycero-3-phosphoethanolamine- N-[maleimide(polyethylene glycol)-2000] (Laysan Bio, Inc), 1 equivalent of tris(2- carboxyethyl)phosphine hydrochloride (Sigma), and a catalytic amount (~10ul) of triethylamine. The mixture was agitated at 25°C for 24 hr. Unconjugated peptides were removed using HPLC. Amph-pepvIII concentration was determined using nanodrop. The resulting products were lyophilized, re-dissolved in PBS and stored at −20°C. For vaccination, unless otherwise stated, mice received weekly s.c injection of 10 μg peptide Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 16 equivalent of amph-pepvIII mixed with 25 μg of Cyclic-di-GMP (CDG, Invivogen) in 100 μl 1x PBS, administered 50 μl to each side at the tail base. To compare the effect of adjuvants on vaccination, 1.24 nmol lipo-CpG72 or 10μg R848 (TLR7/8 agonist, Resiquimod [Invivogen]) was used per mouse. ELISPOT—To evaluate epitope spreading, the spleen was harvested from individual mice for total T cell isolation using a CD3+ T cell isolation kit (Stem Cell Technology). For most experiments, CAR T-cells were prepared using T cells isolated from CD45.1+ mice, transferred into tumor-bearing CD45.2+ recipients, enabling magnetic depletion of adoptively transferred CAR T-cells during endogenous T cell isolation using negative selection. For this purpose, anti-CD45.1 antibody (Clone A20, Stem Cell Technology) were added to whole splenocytes at 1ug/ml together with the T cell isolation cocktail. The day before T cell isolation, 2×106 tumor cells (CT-2A, MHCII+CT-2A or B16F10 cells) were seeded in a T75 flask in the presence of 100 IU of murine IFN-γ [PeproTech] and subjected to 120Gy of irradiation the next morning. Tumor cells were then trypsinized into single cell suspension using TrypLE Express (Gibco) to avoid removal of surface proteins and washed twice with 1x PBS to remove residual IFN-γ. 4×105 CD3+ T cells were mixed with 25,000 irradiated tumor cells in 200 µL complete medium and seeded in a 96-well ELISPOT plate (BD) that was pre-coated with IFN-γ capture antibody (BD IFN-γ ELISPOT kit). Plates were wrapped in foil and cultured for 24hr in 37°C incubator, then developed according to the manufacturer’s protocol. Plates were scanned using a CTL-ImmunoSpot Plate Reader, and data were analyzed using CTL ImmunoSpot Software. CAR T functionality assay—The functionality of WT, IFN-γ−/−, IFNGR1−/− or NFAT- IFNγ CAR T-cells was assessed by co-coculturing with EGFRvIII-CT2A cells in 96-well flat-bottom plates. Unless otherwise stated, 1×105 CAR T-cells were mixed with 1×104 target cells in a total volume of 200 μl complete medium containing 20IU/ml of mIL-2. After 6 hr co-culture, cells were resuspended by vigorous pipetting, transferred to a U- bottom plate, and pelleted at 2,000×g for 5 min. The supernatant was saved for ELISA following the manufacturer’s protocol (Mouse IFN-γ Duo set, R&D systems). Cells were stained with anti-CD45 and anti-CD8α for 20 min on ice and resuspended in flow cytometry buffer with 1x SYTOX Red (Thermo Fisher) for flow analysis. Dead tumor cells were gated as CD8− CD45− SYTOX RED+ population. IFN-γ ELISAs were performed following the manufacturer’s protocol. P15E antigen and Env protein detection—Env protein expression on CT-2A cell surface was monitored using flow cytometry and staining with 1E4.2.1 anti-Env antibody as previously described67 (Wittrup lab). The presentation of Env antigen p15E on CT-2A cells were assessed by co-culturing IFN-γ-treated CT-2A cells with a 58−/− T cell hybridoma cell line expressing a p15E-specific TCR 7PPG-2 (Birnbaum lab) and monitoring T cell activation using mouse IL-2 ELISA (Invitrogen) as previously described32,68. A 58−/−hybridoma cells expressing an irrelevant 2C TCR (Birnbaum lab) were included as negative control. TC-1 cells and MC38 cells were included as negative control and positive control of ENV/p15E expression, respectively. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 17 Secondary transplantation study—To evaluate the qualitative anti-tumor activity of CAR T-cells, 10 million CD45.1+ donor CAR T-cells were i.v. infused to lymphodepleted CD45.2+ recipients (500cGy sublethal irradiation) followed 24hr later by a single dose of amph-pepVIII vaccination or mock vaccination with PBS. Seven days later, mice were euthanized, and spleens were harvested and combined for each group for total T cell isolation using a modified pan-T cell negative selection protocol. Briefly, total splenocytes were stained with a pan-T cell isolation cocktail (Stem Cell Technology) plus 1:500 dilution of biotinylated anti-CD45.2 antibody (Clone 104, 0.5mg/ml, Stem Cell Technology). Negative selection was performed following the same downstream procedures as listed in the manufacture’s protocol to obtain untouched vaccine-boosted CD45.1 CAR T-cells. Immediately after isolation, ~8×106 CD45.1 T cells from either mock or vaccine-treated groups were adoptively transferred to secondary recipients bearing ~25 mm2 EGFRvIII-CT-2A tumors that had been lymphodepleted the day before, followed by periodic monitoring of tumor growth and animal survival. Note: throughout this study, the retroviral transduction efficiency and subsequent CAR+ T cells was constantly >90%, therefore, the total number of transferred CAR+ CD45.1 T cells are ~7×106. Luminex assay—EGFRvIII-CT-2A tumor-bearing C57BL/6 mice received lymphodepletion followed by adoptive transfer of either WT or IFN-γ−/− CAR T-cells plus a single dose of vaccination. Mice were euthanized and tumors isolated at day 7 post vaccination. Tumors were weighted, cut using a razor blade into small pieces and dounced to generate tumor homogenate in tissue protein extraction buffer (T-PERTM, Thermo Fisher Scientific, cat. no. 78510) in the presence of 1% proteinase and phosphatase inhibitors (Thermo Fisher Scientific, cat. no. 78442). The lysates were incubated at 4°C for 30 min with slow rotation followed by top-speed centrifugation to remove debris. The supernatants were transferred to a clean tube and stored at −80°C. Part of the samples were subjected to Luminex analysis using a Mouse Cytokine 32-Plex panel analysis at Eve Technology. Tumor sectioning and vasculature staining—C57BL/6 mice bearing EGFRvIII+CT-2A tumors received lymphodepletion (LD), followed by no treatment, or were treated with WT CAR T or IFNγ−/− CAR T in the presence or absence of vaccination as in Fig 1C. Seven days post vaccination, mice were euthanized 5 minutes after intravenous injection with 0.2 mg Hoechst 33342 (Thermofisher) and 0.2 mg Dextran Tetramethylrhodamine 70,000 MW (Thermofisher). Tumors were harvested and fixed with 4% PFA at 4 °C for 18 h. Next, isolated tumors were washed in PBS and embedded in 3% (wt/vol) low-melting agarose at 37 °C. The agarose was allowed to solidify on ice for 15 min before sectioning on a vibratome (Leica VT1000S). 150-μm tissue sections were incubated with Fc Receptor Blocker (Innovex Bioscience) for 30 minutes and then blocked with 2% bovine serum albumin in PBS for 1 h at room temperature. Tumor vessel staining with primary antibodies (1:100) was performed overnight at 4 °C in blocking buffer using Alexa Fluor® 647 anti-mouse CD31 Antibody (BioLegend, #102516). After three washes with PBS, the sections were mounted onto glass slides using mounting media (ProLong Diamond Antifade Mountant, Thermo Fisher Scientific). Images were acquired using a Leica SP8 laser-scanning confocal microscope with a 25× objective. Image processing was performed with Fiji 69 and Imaris v10. The surface tracing algorithm was used to trace and mask Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 18 the anti-CD31 channel. Total vessel diameter and vessel volume was calculated using the Filament tracing algorithm on the masked anti-CD31 channel. Hoechst area were calculated using the Analyze particles function in Fiji as % of tumor area perfused. Bulk RNA-sequencing for CAR T characterization—EGFRvIII-CT-2A tumor- bearing CD45.2+ mice were treated with CD45.1+ CAR T-cells and mock (PBS) or amph- pepVIII vaccination. 7 days later, mice were euthanized to harvest spleens and tumors. Total splenic T cells were isolated using the pan-T cell isolation kit and stained with anti-CD8α, anti-CD4, anti-CD45.1 and 7AAD for flow sorting. 5×104 CD45.1+ CAR T-cells were directly sorted into Trizol. For intratumoral CAR T isolation, tumors were cut into 1–2 mm2 pieces using razor blades, placed in 1.5ml or 5ml tubes (depending on tumor size) and digested (2 mg/ml Collagenase IV [Worthington], 0.1mg/ml of DNAse I [Sigma], and 10% of TrypLE [Thermo Fisher] in 1xRPMI) for 20 min on a rotator at 37°C. Digested tumors were then mushed through a 70um cell strainer using a blunt non-rubber end of the a 1ml syringe plunger, washed 1x with 1xRPMI. Intratumoral T cells were enriched using mouse CD4/CD8 (TIL) MicroBeads (Miltenyi), stained, and sorted into Trizol as above. The total number of sorted CD45.1+ CAR T-cells from tumors ranged from 6×103 to 5×104 per sample. Total RNA was isolated using the RNeasy Micro kit (Qiagen). Samples were submitted to the BioMicro center at MIT for library construction and sequencing. Bulk RNA-sequencing data was analyzed with the help from the bioinformatics core at the Koch Institute. Briefly, paired-end RNA-seq data was used to quantify transcripts from the mm10 mouse assembly with the Ensembl version 100 annotation using Salmon version 1.2.173. Gene level summaries for were prepared using tximport version 1.16.074. running under R version 4.0.0 (https://www.R-project.org). Differential expression analysis was performed using DESeq2 version 1.28.175,76 and differentially expressed genes were defined as those having an absolute apeglm77 log2 fold change greater than 1 and an adjusted p-value less than 0.05. Data parsing and some visualizations were carried out using Tibco Spotfire Analyst 7.6.1. Mouse genes were mapped to human orthologs using Mouse Genome Informatics (http://www.informatics.jax.org/) orthology report. Preranked GSEA78 was run using javaGSEA version 4.0.3 for gene sets from MSigDB version 7.179. Preranked GSEA for custom mouse gene sets was run with javaGSEA version 4.1.0 Seq-Well Single cell RNA-sequencing to profile AS in intratumoral T cells— Tumors were digested and tumor-infiltrating lymphocytes enriched as described above for bulk RNA-seq. Enriched TILs from individual mice were first labeled with Total-seq A anti- mouse hashing antibodies (BioLegend) and washed 2x in flow cytometry buffer. Samples from the same group were then combined and stained with the same surface staining antibody cocktail. Endogenous CD45.2+ CD4 and CD8 T cells were sorted collectively into 1x RPMI +10%FBS, 2–5×104 total T cells were obtained for each group. Cells were pelleted at 1000×g for 5min, resuspended in 1xRPMI at 20,000 cells per 200μl and then processed for scRNA-seq using the Seq-Well platform with second strand chemistry, as previously described80. Whole transcriptome libraries were barcoded and amplified using the Nextera XT kit (Illumina) and were sequenced on a Novaseq 6000 (Illumina). Hashtag oligo libraries were amplified as described previously81 and were sequenced on a Nextseq 550. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 19 Processing of single cell hashing data—Cell hashing data was aligned to HTO barcodes using CITE-seq-Count v1.4.2 (https://zenodo.org/badge/latestdoi/99617772). To establish thresholds for positivity for each HTO barcode, we first performed centered log- ratio normalization of the HTO matrix and then performed k-medoids clustering with k=5 (one for each HTO). This produced consistently five clusters, each dominated by one of the 5 barcodes. For each cluster, we first identified the HTO barcode that was dominant in that cluster. We then considered the threshold to be the lowest value for that HTO barcode among the cells classified in that cluster. To account for the scenario in which this value was substantially lower than the rest of the values in the cluster, we used Grubbs’ test to determine whether this threshold was statistically an outlier relative to the rest of the cluster. If the lower bound was determined to be an outlier at p=0.05, it was removed from the cluster, and the next lowest value was used as the new threshold. This procedure was iteratively applied until the lowest value in the cluster was no longer considered an outlier at p=0.05. Cells were then determined to be “positive” or “negative” for each HTO barcode based on these thresholds. HTO thresholds were examined and manually adjusted if necessary. Cells that were positive for multiple HTOs (doublets) or were negative for all HTOs were excluded from downstream analysis. To account for differences in sequencing depth between samples, these steps were performed separately for each Seq-Well array that was processed. scRNA-seq data processing and visualization—Raw read processing of scRNA-seq reads was performed as previously described82. Briefly, reads were aligned to the mm10 reference genome and collapsed by cell barcode and unique molecular identifier (UMI). Then, cells with less than 500 unique genes detected and genes detected in fewer than 5 cells were filtered out, and the data for each cell was log-normalized to account for library size. Genes with log-mean expression values greater than 0.1 and a dispersion of greater than 1 were selected as variable genes, and the ScaleData function in Seurat was used to regress out the number of UMI and percentage of mitochondrial genes in each cell. Principal components analysis was performed. The number of principal components used for visualization was determined by examination of the elbow plot, and two-dimensional embeddings were generated using uniform manifold approximation and projection (UMAP). Clusters were determined using Louvain clustering, as implemented in the FindClusters function in Seurat, and clusters that contained activated T cells were selected for further analysis. These cells were reprocessed with the same processing and clustering steps described above. DEG analysis was performed for each cluster and between indicated cell populations using the FindMarkers function. qPCR to validate differentially expressed genes—Splenic CD45.1+ CAR T-cells isolated from CAR T only or CAR T-vax treated mice 7 days post the first vaccine. Briefly, total T cells were first enriched using pan T cell isolation kit followed by staining of CD8, Myc and CD45.1 surface markers. CD45.1+ Myc+ cells were FACS-sorted into Trizol. The total number of sorted CD45.1+ CAR T-cells from spleens range from 4×104 – 6×104 per sample. Total RNA was isolated using the RNeasy Micro kit (Qiagen) and subjected to cDNA synthesis using iScript Reverse Transcription Supermix in 20ul reaction. qPCR primers were designed and qPCR reactions as carried out previously described 83. Actin was Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 20 used as the internal control, and genes with CT value lower than 32 were considered as detectable and the corresponding sample was included for statistically analysis. Paired single-cell TCR sequencing and analysis—Paired TCR sequencing and read alignment was performed as previously described84. Briefly, whole transcriptome amplification product from each single-cell library was enriched for TCR transcripts using biotinylated Tcrb and Tcra probes and magnetic streptavidin beads. The enrichment product was further amplified using V-region primers and Nextera sequencing handles, and the resulting libraries were sequenced on an Illumina Novaseq 6000. Processing of reads was performed using the Immcantation software suite85,86. Briefly, reads were aggregated by cell barcode and UMI, and UMI with under 5 reads were discarded. ClusterSets.py was used to divide sequences for each UMI into sets of similar sequences. Only sets of sequences that comprised greater than 90% of the sequences obtained for that UMI were considered further. Consensus sequences for each UMI were determined using the BuildConsensus.py function. Consensus sequences were then mapped against TCRV and TCRJ IMGT references sequences with IgBlast. Sequences for which a CDR3 sequence could not be unambiguously determined were discarded. UMI for consensus sequences were corrected using a directional UMI collapse, as implemented in UMI-tools87. TCR sequences were then mapped to single cell transcriptomes by matching cell barcodes. If multiple Tcra or Tcrb sequences were detected for a single cell barcode, then the corresponding sequence with the highest number of UMI and raw reads was retained. TCR data for p15E tetramer-sorted CD8+ T cells was obtained from Grace et al68. Using this data, we defined high-confidence p15E-specific Tcrb and Tcra CDR3 amino acid sequences as sequences that were detected in more than one cell and for which greater than 80% of total sequences recovered were in the tetramer-positive fraction. Using this set of sequences as a reference, we defined likely p15E-specific clonotypes in our sequencing of TIL from CAR T and CAR T-Vax treated mice as clonotypes that utilized either one of these Tcrb or Tcra amino acid sequences or utilized the Tcra motif “DYSNNRLT”, which was strongly implicated in the recognition of the p15E epitope by Grace et al. To define a cytotoxicity score for each CD8+ T cell in our single-cell sequencing data, we utilized the AddModuleScore function in Seurat using the following genes as a signature: Gzma, Gzmb, Gzmc, Gzmd, Gzme, Gzmf, Gzmg, Gzmk, Gzmm. Single T cells for which neither a Tcrb or Tcra sequence were recovered were excluded from this analysis. Phenotyping of immune cells in peripheral blood, lymph nodes, and tumors—Peripheral blood (PB) was collected via retro-orbital bleeding. 50–100 μl PB (lymphodepleted mice) was processed in ACK lysis buffer twice (3–5min the 1st time till all RBCs were lysed followed by centrifugation at 1000×g for 5min, decant, resuspend in 200ul ACK and spin again), immediately after spin the 2nd time, instead of decanting, RBC debris from each well was carefully removed by vacuuming in a circular motion without touching the center of the pellet. Lymph nodes (LNs) were placed in a 5ml flow cytometry tube with a 70 µM cell strainer cap and smashed through with the rubber end of a 1ml syringe plunger with frequent addition of flow cytometry buffer. Dissociated LN cells were pelleted and transferred to a 96 well U-bottom plate for further analysis. EGFRvIII-CT-2A tumors from mice receiving CAR T or CAR T-vax therapy Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 21 were surgically removed and weighed and dissociated into single cell suspension using enzyme digestion as described above. Single cell suspensions were obtained by passing tumors through a 70 µM cell strainer with a 1 ml syringe plunger. Cells were pelleted and resuspended with 100 μl of FACS buffer per 100 mg tumor. For immunophenotyping analysis, PBMCs or lymph node cell suspensions were pelleted in a 96 well U-bottom plate and stained with desired antibody cocktails at a 1:200 dilution for CD4+ T cells, CD8+ T cells, Tregs (FoxP3+), B cells (B220+), CD103+ cDC1 (CD24+CD11c+F4/80−CD103+), CD11b+ cDC2 (CD24+CD11c+F4/80−CD11b+), pDCs (CD24+CD11c+F4/80−CD317+), M1 (CD11b+F4/80+CD206−) and M2 (CD11b+F4/80+CD206+) macrophages. For intracellular cytokine staining (ICS) analysis, 50–75 μl of the above tumor cell suspension was pelleted in 96 well U-bottom plates and directly resuspended in RPMI1640 with 10% FBS plus 1x Golgi plug and 1x cell stimulation cocktail (Thermo Fisher) for 6 hr at 37°C. Cells were then pelleted at 1000×g for 5 min and washed once with PBS, then stained with live/dead aqua for 15 min in the dark at 25°C. Cells were pelleted again, surface stained with desired antibody cocktail for ~20 min on ice followed by 1x wash with flow cytometry buffer. Cells were resuspended in 75 ml of BD Fix/Perm and kept at 4°C for 15 min, then washed once by directly adding 200 μl 1x Perm/Wash. The pellet was resuspended in 50 μl of cytokine antibody cocktail (IFN- γ at 1:100, TNF-α at 1:100, and granzyme B at 1:100) pre-diluted in 1x Perm/Wash buffer, 30 min on ice, then washed once with 1x Perm/Wash buffer and resuspended in 1x flow cytometry buffer for analysis immediately or kept at 4°C for FACS analysis the next day. For FoxP3 or PGC-1α staining, 50–75μl of the above cell suspension was pelleted and processed using a FoxP3 staining kit (Thermo Fisher) according to the manufacturer’s instructions. For tetramer staining, PBMCs or tumor suspensions were stained with 50 µl of SIINFEKL- Tetramer (PE conjugate) plus Fc block at a 1:50 dilution for 30 min at room temperature in the dark, followed by mixing with a pre-made 50 μl cocktail of the remaining surface antibodies (1:50 dilution), 20min on ice kept from light. Then cells were washed twice for flow analysis. CAR T-vax therapy in solid tumor models—In the EGFRvIII-CT-2A mouse glioblastoma model, unless otherwise stated, 5×106 EGFRvIII-CT-2A cells were injected into the right flank of recipient mice in 50 μl saline and allowed to establish palpable tumors ~25 mm2 in size at day 6. Lymphodepletion was carried out using 500 cGy sublethal irradiation, mice were then randomly allocated into each group. 10×106 CAR T-cells from mice with the desired background were i.v. infused via the tail vein into recipient mice followed by weekly s.c. immunization with amph-pepvIII vaccine (10 μg amph-pepvIII, 25 μg CDG in 100μl PBS) or PBS alone. For consistency, Rag1 −/− mice were also subjected to the same lymphodepletion preconditioning. For experiments involving cytokine blockade, unless otherwise stated, anti-IFN-γ (BioXcell) was administered i.p. at 200 μg per mouse every three days, anti-TNF-α (BioXcell) was administered i.p. at 300 μg per mouse every two days, anti-IL12(p75) was administered i.p. at 1mg per mouse for the initial dose followed by 500 μg per mouse every three days. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 22 For the mixed tumor studies, each mouse was inoculated in the right flank with 5×106 EGFRvIII-CT-2A cells and WT CT-2A cells mixed at pre-defined ratios (100:0, 90:10, 80:20, 50:50, 25:75, 0:100). 5 days later, when the tumors reached ~25mm2, mice were subjected to lymphodepletion and adoptive transfer of 10×106 EGFRvIII CAR T-cells, followed 24hr later with weekly vaccination. In the B16F10-based mouse melanoma model, B16F10-OVA tumors were established by s.c injection of 1×106 Ova+B16F10 cells into the right flank of C57BL/6 recipient mice in 50 μl saline. For the mixed B16F10 tumor studies, each mouse was inoculated in the right flank with 4 ×105 WT B16F10 cells and Trp1−/− B16F10 cells70 mixed at pre-defined ratios (80:20). Mice received lymphodepletion preconditioning with 500 cGy sublethal irradiation at day 5, and the i.v. infusion of PBS, 10×106 CD45.1+ FITC-CAR T, FITC/TRP1 bispecific CAR T-cells on day 6, followed with or without two weekly amph-FITC immunizations (10nmol amph-FITC, 25 μg CDG in 100μl PBS). NFAT-IFNγ CAR-T vax toxicity analysis—Serum was collected 24 hr before and after the 1st and 2nd vaccination of tumor-bearing animals. Serum cytokine levels were quantified using Legendplex beads following the manufacturer’s protocol. Vaccine and CAR- T therapy-induced body weight (BW) fluctuations in each group were calculated with the following equation: [BW (Day x) / BW (Day 0)] / [BWcontrol (Day x) / BWcontrol (Day 0)]. QUANTIFICATION AND STATISTICAL ANALYSIS Statistical analyses were performed using GraphPad Prism 8. All values and error bars are shown as mean ± 95% CI (confidence interval). Animal survival was analyzed using Log-rank (Mantel-Cox) test. All pair-wise comparisons were analyzed by student’s t-test. Multi-group comparisons was carried out using one-way ANOVA with Tukey’s multiple comparisons test. Experiments that involved repeated measures over a time course, such as tumor growth, were analyzed using a RM (repeated measures) two-way ANOVA based on a general linear model (GLM). The RM design included factors for time, treatment and their interaction. Tukey’s multiple comparisons test was carried out for the main treatment effect. P-values are adjusted to account for multiple comparisons in both one-way ANOVA, and RM two-way ANOVA. For all animal experiments, 6–8-week-old female C57BL/6 were used. At this age, mice have developed a mature immune system, thus ideal for evaluating immunomodulating therapies. We determined the size of samples for experiments involving either quantitative or qualitative data as previously reported88. Based on our previous experience with the animal models and as reported by others27,70,89, we consider the CAR T-vax therapy as significant if it increases the survival of animals up to 100% within 4 weeks, and we need >=5 animals per group to achieve this goal with 95% confidence interval and at 80% power. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 23 ACKNOWLEDGMENTS We thank the Koch Institute Swanson Biotechnology Center for technical support, specifically the flow cytometry core facility. We thank Dr. Thomas Seyfried for providing CT-2A cell line. Funding: D.J.I was supported by the NIH (award EB022433), the Marble Center for Nanomedicine, and the Mark Foundation, and L.M was supported by an American Cancer Society postdoctoral fellowship, the Cell and Gene Therapy Collaborative at CHOP, and the W.W. Smith Charitable Trust. This work is also partially supported by Cancer Center Support (core) Grant P30-CA14051 from the NCI to the Barbara K. Ostrom (1978) Bioinformatics and Computing Core Facility of the Swanson Biotechnology Center. D.J.I. is an investigator of the Howard Hughes Medical Institute. We thank the Wittrup lab at MIT for sharing the 1E4.2.1 anti-Env antibody and MC38 cells, and the Birnbaum lab at MIT for sharing 7PPG-2 or a 2C TCR-expressing 58−/− T cell hybridoma cell lines. References 1. Irvine DJ, Maus MV, Mooney DJ, and Wong WW (2022). The future of engineered immune cell therapies. Science 378, 853–858. 10.1126/science.abq6990. [PubMed: 36423279] 2. Labanieh L, and Mackall CL (2022). CAR immune cells: design principles, resistance and the next generation. Nature 614, 635–648. 10.1038/s41586-023-05707-3. 3. Wang M, Munoz J, Goy A, Locke FL, Jacobson CA, Hill BT, Timmerman JM, Holmes H, Jaglowski S, Flinn IW, et al. (2020). KTE-X19 CAR T-Cell Therapy in Relapsed or Refractory Mantle-Cell Lymphoma. New Engl J Med 382, 1331–1342. 10.1056/nejmoa1914347. [PubMed: 32242358] 4. Maude SL, Laetsch TW, Buechner J, Rives S, Boyer M, Bittencourt H, Bader P, Verneris MR, Stefanski HE, Myers GD, et al. (2018). Tisagenlecleucel in Children and Young Adults with B-Cell Lymphoblastic Leukemia. New Engl J Medicine 378, 439–448. 10.1056/nejmoa1709866. 5. Abramson JS, Palomba ML, Gordon LI, Lunning MA, Wang M, Arnason J, Mehta A, Purev E, Maloney DG, Andreadis C, et al. (2020). Lisocabtagene maraleucel for patients with relapsed or refractory large B-cell lymphomas (TRANSCEND NHL 001): a multicentre seamless design study. Lancet 396, 839–852. 10.1016/s0140-6736(20)31366-0. [PubMed: 32888407] 6. Rafiq S, Hackett CS, and Brentjens RJ (2020). Engineering strategies to overcome the current roadblocks in CAR T cell therapy. Nat Rev Clin Oncol 17, 147–167. 10.1038/s41571-019-0297-y. [PubMed: 31848460] 7. Roselli E, Faramand R, and Davila ML (2021). Insight into next-generation CAR therapeutics: designing CAR T cells to improve clinical outcomes. J Clin Invest 131, e142030. 10.1172/ jci142030. [PubMed: 33463538] 8. Hou AJ, Chen LC, and Chen YY (2021). Navigating CAR-T cells through the solid-tumour microenvironment. Nat Rev Drug Discov 20, 531–550. 10.1038/s41573-021-00189-2. [PubMed: 33972771] 9. O’Rourke DM, Nasrallah MP, Desai A, Melenhorst JJ, Mansfield K, Morrissette JJD, Martinez- Lage M, Brem S, Maloney E, Shen A, et al. (2017). A single dose of peripherally infused EGFRvIII-directed CAR T cells mediates antigen loss and induces adaptive resistance in patients with recurrent glioblastoma. Sci Transl Med 9, eaaa0984. 10.1126/scitranslmed.aaa0984. [PubMed: 28724573] 10. Shah NN, and Fry TJ (2019). Mechanisms of resistance to CAR T cell therapy. Nature Reviews Clinical Oncology, 1–14. 10.1038/s41571-019-0184-6. 11. Landsberg J, Kohlmeyer J, Renn M, Bald T, Rogava M, Cron M, Fatho M, Lennerz V, Wölfel T, Hölzel M, et al. (2012). Melanomas resist T-cell therapy through inflammation-induced reversible dedifferentiation. Nature 490, 412–416. 10.1038/nature11538. [PubMed: 23051752] 12. Gulley JL, Madan RA, Pachynski R, Mulders P, Sheikh NA, Trager J, and Drake CG (2017). Role of Antigen Spread and Distinctive Characteristics of Immunotherapy in Cancer Treatment. Jnci J National Cancer Inst 109, djw261. 10.1093/jnci/djw261. 13. Kvistborg P, Philips D, Kelderman S, Hageman L, Ottensmeier C, Joseph-Pietras D, Welters MJP, Burg S van der, Kapiteijn, E., Michielin, O., et al. (2014). Anti–CTLA-4 therapy Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 24 broadens the melanoma-reactive CD8+ T cell response. Sci Transl Med 6, 254ra128. 10.1126/ scitranslmed.3008918. 14. Brossart P (2020). The role of antigen-spreading in the efficacy of immunotherapies. Clin Cancer Res, clincanres.0305.2020. 10.1158/1078-0432.ccr-20-0305. 15. Awad MM, Govindan R, Balogh KN, Spigel DR, Garon EB, Bushway ME, Poran A, Sheen JH, Kohler V, Esaulova E, et al. (2022). Personalized neoantigen vaccine NEO-PV-01 with chemotherapy and anti-PD-1 as first-line treatment for non-squamous non-small cell lung cancer. Cancer Cell 40, 1010–1026.e11. 10.1016/j.ccell.2022.08.003. [PubMed: 36027916] 16. Beatty GL, Haas AR, Maus MV, Torigian DA, Soulen MC, Plesa G, Chew A, Zhao Y, Levine BL, Albelda SM, et al. (2014). Mesothelin-Specific Chimeric Antigen Receptor mRNA-Engineered T Cells Induce Antitumor Activity in Solid Malignancies. Cancer Immunol Res 2, 112–120. 10.1158/2326-6066.cir-13-0170. [PubMed: 24579088] 17. Kim RH, Plesa G, Gladney W, Kulikovskaya I, Levine BL, Lacey SF, June CH, Melenhorst JJ, and Beatty GL (2017). Effect of chimeric antigen receptor (CAR) T cells on clonal expansion of endogenous non-CAR T cells in patients (pts) with advanced solid cancer. J Clin Oncol 35, 3011–3011. 10.1200/jco.2017.35.15_suppl.3011. 18. Hegde M, Joseph SK, Pashankar F, DeRenzo C, Sanber K, Navai S, Byrd TT, Hicks J, Xu ML, Gerken C, et al. (2020). Tumor response and endogenous immune reactivity after administration of HER2 CAR T cells in a child with metastatic rhabdomyosarcoma. Nat Commun 11, 3549. 10.1038/s41467-020-17175-8. [PubMed: 32669548] 19. Hegde M, Joseph SK, Pashankar F, DeRenzo C, Sanber K, Navai S, Byrd TT, Hicks J, Xu ML, Gerken C, et al. (2020). Tumor response and endogenous immune reactivity after administration of HER2 CAR T cells in a child with metastatic rhabdomyosarcoma. Nat Commun 11, 3549. 10.1038/s41467-020-17175-8. [PubMed: 32669548] 20. Alizadeh D, Wong RA, Gholamin S, Maker M, Aftabizadeh M, Yang X, Pecoraro JR, Jeppson JD, Wang D, Aguilar B, et al. (2021). IFNγ is Critical for CAR T Cell–mediated Myeloid Activation and Induction of Endogenous Immunity. Cancer Discov 11, 2248–2265. 10.1158/2159-8290.cd-20-1661. [PubMed: 33837065] 21. Klampatsa A, Leibowitz MS, Sun J, Liousia M, Arguiri E, and Albelda SM (2020). Analysis and Augmentation of the Immunologic Bystander Effects of CAR T Cell Therapy in a Syngeneic Mouse Cancer Model. Mol Ther Oncolytics 18, 360–371. 10.1016/j.omto.2020.07.005. [PubMed: 32802940] 22. Lai J, Mardiana S, House IG, Sek K, Henderson MA, Giuffrida L, Chen AXY, Todd KL, Petley EV, Chan JD, et al. (2020). Adoptive cellular therapy with T cells expressing the dendritic cell growth factor Flt3L drives epitope spreading and antitumor immunity. Nat Immunol, 1–13. 10.1038/s41590-020-0676-7. [PubMed: 31831887] 23. Kuhn NF, Lopez AV, Li X, Cai W, Daniyan AF, and Brentjens RJ (2020). CD103+ cDC1 and endogenous CD8+ T cells are necessary for improved CD40L-overexpressing CAR T cell antitumor function. Nat Commun 11, 6171. 10.1038/s41467-020-19833-3. [PubMed: 33268774] 24. Kueberuwa G, Kalaitsidou M, Cheadle E, Hawkins RE, and Gilham DE (2018). CD19 CAR T Cells Expressing IL-12 Eradicate Lymphoma in Fully Lymphoreplete Mice through Induction of Host Immunity. Molecular Therapy: Oncolytics 8, 41–51. 10.1016/j.omto.2017.12.003. [PubMed: 29367945] 25. Etxeberria I, Bolaños E, Quetglas JI, Gros A, Villanueva A, Palomero J, Sánchez-Paulete AR, Piulats JM, Matias-Guiu X, Olivera I, et al. (2019). Intratumor Adoptive Transfer of IL-12 mRNA Transiently Engineered Antitumor CD8+ T Cells. Cancer Cell 36, 613–629.e7. 10.1016/ j.ccell.2019.10.006. [PubMed: 31761658] 26. Chmielewski M, and Abken H (2017). CAR T Cells Releasing IL-18 Convert to T-Bethigh FoxO1low Effectors that Exhibit Augmented Activity against Advanced Solid Tumors. Cell Reports 21, 3205–3219. 10.1016/j.celrep.2017.11.063. [PubMed: 29241547] 27. Adachi K, Kano Y, Nagai T, Okuyama N, Sakoda Y, and Tamada K (2018). IL-7 and CCL19 expression in CAR-T cells improves immune cell infiltration and CAR-T cell survival in the tumor. Nature Biotechnology 36, 346–351. 10.1038/nbt.4086. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 25 28. Park AK, Fong Y, Kim S-I, Yang J, Murad JP, Lu J, Jeang B, Chang W-C, Chen NG, Thomas SH, et al. (2020). Effective combination immunotherapy using oncolytic viruses to deliver CAR targets to solid tumors. Sci Transl Med 12. 10.1126/scitranslmed.aaz1863. 29. Walsh SR, Simovic B, Chen L, Bastin D, Nguyen A, Stephenson K, Mandur TS, Bramson JL, Lichty BD, and Wan Y (2019). Endogenous T cells prevent tumor immune escape following adoptive T cell therapy. J Clin Invest 129, 5400–5410. 10.1172/jci126199. [PubMed: 31682239] 30. Zhang L, Morgan RA, Beane JD, Zheng Z, Dudley ME, Kassim SH, Nahvi AV, Ngo LT, Sherry RM, Phan GQ, et al. (2015). Tumor-Infiltrating Lymphocytes Genetically Engineered with an Inducible Gene Encoding Interleukin-12 for the Immunotherapy of Metastatic Melanoma. Clin Cancer Res 21, 2278–2288. 10.1158/1078-0432.ccr-14-2085. [PubMed: 25695689] 31. Kerkar SP, Muranski P, Kaiser A, Boni A, Sanchez-Perez L, Yu Z, Palmer DC, Reger RN, Borman ZA, Zhang L, et al. (2010). Tumor-Specific CD8+ T Cells Expressing Interleukin-12 Eradicate Established Cancers in Lymphodepleted Hosts. Cancer Res 70, 6725– 6734. 10.1158/0008-5472.can-10-0735. [PubMed: 20647327] 32. Ma L, Dichwalkar T, Chang JYH, Cossette B, Garafola D, Zhang AQ, Fichter M, Wang C, Liang S, Silva M, et al. (2019). Enhanced CAR–T cell activity against solid tumors by vaccine boosting through the chimeric receptor. Science 365, 162–168. 10.1126/science.aav8692. [PubMed: 31296767] 33. Liu H, Moynihan KD, Zheng Y, Szeto GL, Li AV, Huang B, Egeren DSV, Park C, and Irvine DJ (2014). Structure-based programming of lymph-node targeting in molecular vaccines. Nature 507, 519–522. 10.1038/nature12978. [PubMed: 24531764] 34. Szabo PA, Levitin HM, Miron M, Snyder ME, Senda T, Yuan J, Cheng YL, Bush EC, Dogra P, Thapa P, et al. (2019). Single-cell transcriptomics of human T cells reveals tissue and activation signatures in health and disease. Nat Commun 10, 4706. 10.1038/s41467-019-12464-3. [PubMed: 31624246] 35. Tibbitt CA, Stark JM, Martens L, Ma J, Mold JE, Deswarte K, Oliynyk G, Feng X, Lambrecht BN, Bleser PD, et al. (2019). Single-Cell RNA Sequencing of the T Helper Cell Response to House Dust Mites Defines a Distinct Gene Expression Signature in Airway Th2 Cells. Immunity 51, 169–184.e5. 10.1016/j.immuni.2019.05.014. [PubMed: 31231035] 36. LeBleu VS, O’Connell JT, Herrera KNG, Wikman H, Pantel K, Haigis MC, Carvalho F.M. de, Damascena A, Chinen LTD, Rocha RM, et al. (2014). PGC-1α mediates mitochondrial biogenesis and oxidative phosphorylation in cancer cells to promote metastasis. Nat Cell Biol 16, 992–1003. 10.1038/ncb3039. [PubMed: 25241037] 37. Fernandez-Marcos PJ, and Auwerx J (2011). Regulation of PGC-1α, a nodal regulator of mitochondrial biogenesis. Am J Clin Nutrition 93, 884S–890S. 10.3945/ajcn.110.001917. [PubMed: 21289221] 38. Vardhana SA, Hwee MA, Berisa M, Wells DK, Yost KE, King B, Smith M, Herrera PS, Chang HY, Satpathy AT, et al. (2020). Impaired mitochondrial oxidative phosphorylation limits the self-renewal of T cells exposed to persistent antigen. Nat Immunol 21, 1022–1033. 10.1038/ s41590-020-0725-2. [PubMed: 32661364] 39. Bhat P, Leggatt G, Waterhouse N, and Frazer IH (2017). Interferon-γ derived from cytotoxic lymphocytes directly enhances their motility and cytotoxicity. Cell Death Dis 8, e2836–e2836. 10.1038/cddis.2017.67. [PubMed: 28569770] 40. Böttcher JP, and Sousa CR e (2018). The Role of Type 1 Conventional Dendritic Cells in Cancer Immunity. Trends Cancer 4, 784–792. 10.1016/j.trecan.2018.09.001. [PubMed: 30352680] 41. Murphy TL, and Murphy KM (2022). Dendritic cells in cancer immunology. Cell Mol Immunol 19, 3–13. 10.1038/s41423-021-00741-5. [PubMed: 34480145] 42. Böttcher JP, Bonavita E, Chakravarty P, Blees H, Cabeza-Cabrerizo M, Sammicheli S, Rogers NC, Sahai E, Zelenay S, and Sousa CR e (2018). NK Cells Stimulate Recruitment of cDC1 into the Tumor Microenvironment Promoting Cancer Immune Control. Cell 172, 1022–1037.e14. 10.1016/ j.cell.2018.01.004. [PubMed: 29429633] 43. Vilgelm AE, and Richmond A (2019). Chemokines Modulate Immune Surveillance in Tumorigenesis, Metastasis, and Response to Immunotherapy. Front Immunol 10, 333. 10.3389/ fimmu.2019.00333. [PubMed: 30873179] Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 26 44. Garris CS, Arlauckas SP, Kohler RH, Trefny MP, Garren S, Piot C, Engblom C, Pfirschke C, Siwicki M, Gungabeesoon J, et al. (2018). Successful Anti-PD-1 Cancer Immunotherapy Requires T Cell-Dendritic Cell Crosstalk Involving the Cytokines IFN-γ and IL-12. Immunity 49, 1148– 1161.e7. 10.1016/j.immuni.2018.09.024. [PubMed: 30552023] 45. Zhang L, Kerkar SP, Yu Z, Zheng Z, Yang S, Restifo NP, Rosenberg SA, and Morgan RA (2011). Improving Adoptive T Cell Therapy by Targeting and Controlling IL-12 Expression to the Tumor Environment. Mol Ther 19, 751–759. 10.1038/mt.2010.313. [PubMed: 21285960] 46. Majzner RG, and Mackall CL (2018). Tumor Antigen Escape from CAR T-cell Therapy. Cancer Discov 8, 1219–1226. 10.1158/2159-8290.cd-18-0442. [PubMed: 30135176] 47. Guedan S, Calderon H Jr., A.D.P., and Maus MV (2019). Engineering and Design of Chimeric Antigen Receptors. Molecular Therapy - Methods & Clinical Development 12, 145–156. 10.1016/ j.omtm.2018.12.009. [PubMed: 30666307] 48. Lim WA, and June CH (2017). The Principles of Engineering Immune Cells to Treat Cancer. Cell 168, 724–740. 10.1016/j.cell.2017.01.016. [PubMed: 28187291] 49. Schroder K, Hertzog PJ, Ravasi T, and Hume DA (2004). Interferon-γ: an overview of signals, mechanisms and functions. J Leukocyte Biol 75, 163–189. 10.1189/jlb.0603252. [PubMed: 14525967] 50. Castro F, Cardoso AP, Gonçalves RM, Serre K, and Oliveira MJ (2018). Interferon-Gamma at the Crossroads of Tumor Immune Surveillance or Evasion. Front Immunol 9, 847. 10.3389/ fimmu.2018.00847. [PubMed: 29780381] 51. Overacre-Delgoffe AE, Chikina M, Dadey RE, Yano H, Brunazzi EA, Shayan G, Horne W, Moskovitz JM, Kolls JK, Sander C, et al. (2017). Interferon-γ Drives Treg Fragility to Promote Anti-tumor Immunity. Cell 169, 1130–1141.e11. 10.1016/j.cell.2017.05.005. [PubMed: 28552348] 52. Larson RC, Kann MC, Bailey SR, Haradhvala NJ, Llopis PM, Bouffard AA, Scarfó I, Leick MB, Grauwet K, Berger TR, et al. (2022). CAR T cell killing requires the IFNγR pathway in solid but not liquid tumours. Nature 604, 563–570. 10.1038/s41586-022-04585-5. [PubMed: 35418687] 53. Boulch M, Cazaux M, Loe-Mie Y, Thibaut R, Corre B, Lemaître F, Grandjean CL, Garcia Z, and Bousso P (2021). A cross-talk between CAR T cell subsets and the tumor microenvironment is essential for sustained cytotoxic activity. Sci Immunol 6. 10.1126/sciimmunol.abd4344. 54. Xu T, Keller A, and Martinez GJ (2019). NFAT1 and NFAT2 Differentially Regulate CTL Differentiation Upon Acute Viral Infection. Front Immunol 10, 184. 10.3389/fimmu.2019.00184. [PubMed: 30828328] 55. Samten B, Townsend JC, Weis SE, Bhoumik A, Klucar P, Shams H, and Barnes PF (2008). CREB, ATF, and AP-1 Transcription Factors Regulate IFN-γ Secretion by Human T Cells in Response to Mycobacterial Antigen. J Immunol 181, 2056–2064. 10.4049/jimmunol.181.3.2056. [PubMed: 18641343] 56. Chang C-H, Curtis JD, Maggi LB, Faubert B, Villarino AV, O’Sullivan D, Huang SC-C, van der Windt GJW, Blagih J, Qiu J, et al. (2013). Posttranscriptional Control of T Cell Effector Function by Aerobic Glycolysis. Cell 153, 1239–1251. 10.1016/j.cell.2013.05.016. [PubMed: 23746840] 57. Savan R (2014). Post-Transcriptional Regulation of Interferons and Their Signaling Pathways. J Interf Cytokine Res 34, 318–329. 10.1089/jir.2013.0117. 58. Keating SE, Zaiatz-Bittencourt V, Loftus RM, Keane C, Brennan K, Finlay DK, and Gardiner CM (2016). Metabolic Reprogramming Supports IFN-γ Production by CD56bright NK Cells. J Immunol 196, 2552–2560. 10.4049/jimmunol.1501783. [PubMed: 26873994] 59. Donnelly RP, Loftus RM, Keating SE, Liou KT, Biron CA, Gardiner CM, and Finlay DK (2014). mTORC1-Dependent Metabolic Reprogramming Is a Prerequisite for NK Cell Effector Function. J Immunol 193, 4477–4484. 10.4049/jimmunol.1401558. [PubMed: 25261477] 60. Gerbec ZJ, Hashemi E, Nanbakhsh A, Holzhauer S, Yang C, Mei A, Tsaih S-W, Lemke A, Flister MJ, Riese MJ, et al. (2020). Conditional Deletion of PGC-1α Results in Energetic and Functional Defects in NK Cells. Iscience 23, 101454. 10.1016/j.isci.2020.101454. [PubMed: 32858341] 61. Lisci M, Barton PR, Randzavola LO, Ma CY, Marchingo JM, Cantrell DA, Paupe V, Prudent J, Stinchcombe JC, and Griffiths GM (2021). Mitochondrial translation is required for sustained killing by cytotoxic T cells. Science 374, eabe9977. 10.1126/science.abe9977. [PubMed: 34648346] Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 27 62. Windt GJW, and Pearce EL (2012). Metabolic switching and fuel choice during T-cell differentiation and memory development. Immunol Rev 249, 27–42. 10.1111/ j.1600-065x.2012.01150.x. [PubMed: 22889213] 63. Scharping NE, Menk AV, Moreci RS, Whetstone RD, Dadey RE, Watkins SC, Ferris RL, and Delgoffe GM (2016). The Tumor Microenvironment Represses T Cell Mitochondrial Biogenesis to Drive Intratumoral T Cell Metabolic Insufficiency and Dysfunction. Immunity 45, 374–388. 10.1016/j.immuni.2016.07.009. [PubMed: 27496732] 64. Mackensen A, Haanen JBAG, Koenecke C, Alsdorf W, Wagner-Drouet E, Heudobler D, Borchmann P, Bokemeyer C, Klobuch S, Smit E, et al. (2022). LBA38 BNT211–01: A phase I trial to evaluate safety and efficacy of CLDN6 CAR T cells and CLDN6-encoding mRNA vaccine-mediated in vivo expansion in patients with CLDN6-positive advanced solid tumours. Ann Oncol 33, S1404–S1405. 10.1016/j.annonc.2022.08.035. 65. Haanen J, Mackensen A, Koenecke C, Alsdorf W, Desuki A, Wagner-Drouet E, Heudobler D, Borchmann P, Wiegert E, Schulz C, et al. (2021). LBA1 BNT211: A phase I/II trial to evaluate safety and efficacy of CLDN6 CAR-T cells and CARVac-mediated in vivo expansion in patients with CLDN6+ advanced solid tumors. Ann Oncol 32, S1392. 10.1016/j.annonc.2021.10.216. 66. Snook AE (2020). Companion vaccines for CAR T-cell therapy: applying basic immunology to enhance therapeutic efficacy. Future Med Chem 12, 1359–1362. 10.4155/fmc-2020-0081. [PubMed: 32597219] 67. Kang BH, Momin N, Moynihan KD, Silva M, Li Y, Irvine DJ, and Wittrup KD (2021). Immunotherapy-induced antibodies to endogenous retroviral envelope glycoprotein confer tumor protection in mice. Plos One 16, e0248903. 10.1371/journal.pone.0248903. [PubMed: 33857179] 68. Grace BE, Backlund CM, Morgan DM, Kang BH, Singh NK, Huisman BD, Rappazzo CG, Moynihan KD, Maiorino L, Dobson CS, et al. (2022). Identification of Highly Cross-Reactive Mimotopes for a Public T Cell Response in Murine Melanoma. Front Immunol 13, 886683. 10.3389/fimmu.2022.886683. [PubMed: 35812387] 69. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, et al. (2012). Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676–682. 10.1038/nmeth.2019. [PubMed: 22743772] 70. Moynihan KD, Opel CF, Szeto GL, Tzeng A, Zhu EF, Engreitz JM, Williams RT, Rakhra K, Zhang MH, Rothschilds AM, et al. (2016). Eradication of large established tumors in mice by combination immunotherapy that engages innate and adaptive immune responses. Nature Medicine 22, 1402–1410. 10.1038/nm.4200. 71. Clipstone NA, and Crabtree GR (1992). Identification of calcineurin as a key signalling enzyme in T-lymphocyte activation. Nature 357, 695–697. 10.1038/357695a0. [PubMed: 1377362] 72. Liu H, Moynihan KD, Zheng Y, Szeto GL, Li AV, Huang B, Egeren DSV, Park C, and Irvine DJ (2014). Structure-based programming of lymph-node targeting in molecular vaccines. Nature 507, 519–522. 10.1038/nature12978. [PubMed: 24531764] 73. Patro R, Duggal G, Love MI, Irizarry RA, and Kingsford C (2017). Salmon provides fast and bias- aware quantification of transcript expression. Nat Methods 14, 417–419. 10.1038/nmeth.4197. [PubMed: 28263959] 74. Soneson C, Love MI, and Robinson MD (2016). Differential analyses for RNA-seq: transcript-level estimates improve gene-level inferences. F1000research 4, 1521. 10.12688/f1000research.7563.2. 75. Love MI, Huber W, and Anders S (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550. 10.1186/s13059-014-0550-8. [PubMed: 25516281] 76. Anders S, and Huber W (2010). Differential expression analysis for sequence count data. Genome Biol 11, R106–R106. 10.1186/gb-2010-11-10-r106. [PubMed: 20979621] 77. Zhu A, Ibrahim JG, and Love MI (2019). Heavy-tailed prior distributions for sequence count data: removing the noise and preserving large differences. Bioinformatics 35, 2084–2092. 10.1093/ bioinformatics/bty895. [PubMed: 30395178] 78. Mootha VK, Lindgren CM, Eriksson K-F, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstråle M, Laurila E, et al. (2003). PGC-1α-responsive genes involved in oxidative Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 28 phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34, 267–273. 10.1038/ng1180. [PubMed: 12808457] 79. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al. (2005). Gene set enrichment analysis: A knowledge- based approach for interpreting genome-wide expression profiles. Proc National Acad Sci 102, 15545–15550. 10.1073/pnas.0506580102. 80. Hughes TK, Wadsworth MH, Gierahn TM, Do T, Weiss D, Andrade PR, Ma F, Silva B.J. de A., Shao S, Tsoi LC, et al. (2020). Second-Strand Synthesis-Based Massively Parallel scRNA- Seq Reveals Cellular States and Molecular Features of Human Inflammatory Skin Pathologies. Immunity 53, 878–894.e7. 10.1016/j.immuni.2020.09.015. [PubMed: 33053333] 81. Stoeckius M, Zheng S, Houck-Loomis B, Hao S, Yeung BZ, Mauck WM, Smibert P, and Satija R (2018). Cell Hashing with barcoded antibodies enables multiplexing and doublet detection for single cell genomics. Genome Biol 19, 224. 10.1186/s13059-018-1603-1. [PubMed: 30567574] 82. Macosko EZ, Basu A, Satija R, Nemesh J, Shekhar K, Goldman M, Tirosh I, Bialas AR, Kamitaki N, Martersteck EM, et al. (2015). Highly Parallel Genome-wide Expression Profiling of Individual Cells Using Nanoliter Droplets. Cell 161, 1202–1214. 10.1016/j.cell.2015.05.002. [PubMed: 26000488] 83. Ma L, Shan Y, Bai R, Xue L, Eide CA, Ou J, Zhu LJ, Hutchinson L, Cerny J, Khoury HJ, et al. (2014). A therapeutically targetable mechanism of BCR-ABL-independent imatinib resistance in chronic myeloid leukemia. Science translational medicine 6, 252ra121–252ra121. 10.1126/ scitranslmed.3009073. 84. Tu AA, Gierahn TM, Monian B, Morgan DM, Mehta NK, Ruiter B, Shreffler WG, Shalek AK, and Love JC (2019). TCR sequencing paired with massively parallel 3′ RNA-seq reveals clonotypic T cell signatures. Nat Immunol 20, 1692–1699. 10.1038/s41590-019-0544-5. [PubMed: 31745340] 85. Gupta NT, Heiden JAV, Uduman M, Gadala-Maria D, Yaari G, and Kleinstein SH (2015). Change-O: a toolkit for analyzing large-scale B cell immunoglobulin repertoire sequencing data. Bioinformatics 31, 3356–3358. 10.1093/bioinformatics/btv359. [PubMed: 26069265] 86. Heiden JAV, Yaari G, Uduman M, Stern JNH, O’Connor KC, Hafler DA, Vigneault F, and Kleinstein SH (2014). pRESTO: a toolkit for processing high-throughput sequencing raw reads of lymphocyte receptor repertoires. Bioinformatics 30, 1930–1932. 10.1093/bioinformatics/btu138. [PubMed: 24618469] 87. Smith T, Heger A, and Sudbery I (2017). UMI-tools: modeling sequencing errors in Unique Molecular Identifiers to improve quantification accuracy. Genome Res 27, 491–499. 10.1101/ gr.209601.116. [PubMed: 28100584] 88. Charan J, and Kantharia ND (2013). How to calculate sample size in animal studies? J Pharmacol Pharmacother 4, 303–306. 10.4103/0976-500x.119726. [PubMed: 24250214] 89. Davila ML, Kloss CC, Gunset G, and Sadelain M (2013). CD19 CAR-Targeted T Cells Induce Long-Term Remission and B Cell Aplasia in an Immunocompetent Mouse Model of B Cell Acute Lymphoblastic Leukemia. PLOS ONE 8, e61338–14. 10.1371/journal.pone.0061338. [PubMed: 23585892] Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 29 Highlights Vaccine boosting enhances CAR T cell metabolism and polyfunctionality Vaccine-boosted CAR T therapy elicits robust and potent antigen spreading Antigen spreading supports CAR T therapy to treat antigenically heterogeneous tumors CAR T-derived IFNγ and DC-derived IL12 are critical for sustaining antigen spreading Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 30 Figure 1. Vaccine boosting enables CAR T-cells to elicit endogenous T-cell responses in multiple tumor models. (A) Schematic of CAR T-vax therapy. Created with BioRender.com. (B) IFN-γ ELISPOT. Mice bearing EGFRvIII+CT-2A tumors (n=5) treated with or without CAR T + various combinations of vaccine components. (C) Priming of endogenous CD8+ and CD4+ T-cells in EGFRvIII+CT-2A tumor-bearing mice (n=5–6) following CAR-T ± vax as measured by IFN-γ ELISPOT. (D-G) Mice (n=5–6) bearing OVA+ B16F10 tumors received FITC/TA99 CAR T-vax. (D) IFN-γ ELISPOT measuring OVA-specific endogenous T-cell responses. (F) IFN-γ ELISPOT measuring endogenous T-cell responses against Trp1−/− B16F10 cells. (F-G)Tetramer-staining showing representative flow cytometry staining (F) and mean percentages of SIINFEKL tetramer+ endogenous T cells (G). (H) IFN-γ ELISPOT. Mice (n=4–5) bearing B16F10 tumors were treated with vax only, FITC-CAR T, or FITC/TA99 CAR T ± vax. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 31 Error bars show mean ± 95% CI. , p<0.0001; , p<0.01; , p<0.05; n.s., not significant by one-way ANOVA with Tukey’s post-test. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 32 Figure 2. Endogenous tumor-infiltrating T cells show transcriptional changes associated with enhanced anti-tumor activity in response to CAR T-vax therapy. (A) Enumeration of intratumoral host T-cells in tumor-bearing mice (n=5) post CAR T ± vax treatment. (B-G) Tumor-bearing mice were treated with CAR T ± vax, TILs were isolated for scRNA- seq. (B) Experimental setup/timeline. Created with BioRender.com. (C) UMAP of endogenous T-cells obtained from tumors. (D) Curated clusters based on signature gene expression. (E) Stacked charts showing proportions of each T-cell cluster. (F) Dot plots showing differential expression of signature genes in endogenous CD8+ CTLs or CD4+ Th cells. (G) Cytotoxicity score of endogenous p15E-specific TILs and TILs of unknown specificity. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 33 All mice bear EGFRvIII+CT-2A tumors. Error bars are mean ± 95% CI, *p<0.0001; , p<0.01; n.s., not significant by Student’s t-test for A, by two-sided Wilcoxon rank-sum test for G. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 34 Figure 3. Vaccine-driven antigen spreading is required for long-tumor tumor control in immunocompetent mice. (A-E) Treatment of tumor-bearing WT or Rag1−/− mice with WT CAR-T ± vax. (A) Tumor growth in individual mice. Untreated, n = 5; CAR-T in WT mice, n = 10; CAR T-vax, n = 15 and 10 in WT and Rag1−/− mice, respectively. (B) Percentage of mice that completely rejected tumors or experienced tumor relapse. (C) Overall survival. (D-E) Surface EGFRvIII expression (D) and mean expression normalized to untreated tumors (E) on parental or representative relapsed tumors from WT and Rag1−/− mice following CAR T-vax treatment. (F-G) Tumor-bearing WT or CD8α−/− mice (n=5–8) ± CAR T-vax treatment. (F) Tumor growth. (G) Overall survival. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 35 (H) Individual tumor growth and overall survival of WT (n=10) or Rag1−/− mice (n=5) bearing heterogeneous CT-2A tumors upon CAR T-vax treatment. EGFRvIII+:EGFRvIII− cells were pre-mixed at the indicated ratios. All mice in A-G bear EGFRvIII+CT-2A tumors. Error bars are mean ± 95% CI, *, p<0.0001; , p<0.01; , p<0.05 by Student’s t-test for E, by Log-rank (Mantel-Cox) test for C,G-H, by two-way ANOVA with Tukey’s post-test for F. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 36 Figure. 4. Vaccine boosting induces cell-intrinsic enhancements in CAR T-cell function that include metabolic reprogramming. (A) Tumor growth (left) and overall survival (right) of tumor-bearing mice (n=8) after receiving vaccine-boosted or non-boosted CAR T cells. Created with BioRender.com. (B-C) Tumor-bearing mice received WT CAR T ± vax treatment, and CAR T-cells were isolated from spleens and tumors for RNA-seq. (B) Volcano plot showing differential gene expression in splenic CAR T-cells. (C) GSEA showing enriched pathways in intratumoral CAR T-cells. (D-E) Intracellular PGC-1α expression (D) and mitochondrial mass (E) in intratumoral CAR T cells from mice (n=5) 7 days post treatment with WT CAR T ± vax. (F) IFN-γ ELISPOT. Tumor-bearing mice (n=5) treated with WT CAR T ± vax or PGC-1α−/− CAR T-vax. All mice bear EGFRvIII+CT-2A tumors. Error bars are mean ± 95% CI, , p<0.01; , p<0.05 by Student’s t-test for D-E, and one-way ANOVA with Tukey’s post-test for F. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 37 Figure. 5. Enhanced IFN-γ production by vaccine-boosted CAR T-cells is critical for antigen spreading. (A) IFN-γ and TNF-α expression in intratumoral CAR T-cells from mice (n=5) treated with WT CAR T ± vax. (B) IFN-γ expression in intratumoral CAR T-cells from mice (n=5) 7 days post treatment with WT or PGC-1α−/− CAR T-vax. (C) IFN-γ ELISPOT. Tumor-bearing mice (n=5) treated with WT CAR T or WT CAR T-vax + isotype control antibody (IgG), anti-TNF-α or anti-IFN-γ. (D-E) OVA+EGFRvIII+CT-2A tumor-bearing mice(n=5–10) treated by WT CAR T or WT CAR T-vax ± anti-IFN-γ. Endogenous OVA-specific T-cell responses detected by SIINFEKL-tetramer staining (D) and IFN-γ ELISPOT (E). (F) IFN-γ ELISPOT. Tumor-bearing mice (n=5) treated with WT CAR T-vax ± anti-IFN-γ at indicated time points. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 38 (G-H) Tumor growth (G) and overall survival (H) of mice left untreated (n=5) or treated (n=10) with WT CAR T-vax ± anti-IFN-γ. (I) IFN-γ ELISPOT. Tumor-bearing mice (n=5) treated with WT or IFN-γ−/− CAR T ± vax. (J) Tumor growth in mice (n=5) left untreated or treated with WT or IFN-γ−/− CAR T-vax. (K-L) Tumor growth (K) and overall survival (L) of WT or IFN-γ−/− mice (n=5–8) treated with or without WT CAR T-vax therapy. All mice bear EGFRvIII+CT-2A tumors. Error bars are mean ± 95% CI, *, p<0.0001; p<0.01; , p<0.05, ns, not significant by Student’s t-test for A-B, by one-way ANOVA with Tukey’s post-test for C-F, I, by two-way ANOVA with Tukey’s post-test for J-K, and by Log-rank (Mantel-Cox) test for H and L. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 39 Figure. 6. DCs regulate CAR T-cell-induced antigen spreading through enhanced tumor antigen acquisition and IFN-γ-IL-12 crosstalk. (A) EGFRvIII+ CT-2A cell killing by WT, IFN-γ−/−, or IFNGR1−/− CAR T-cells in vitro (n=3). (B) Enumeration of tumor-infiltrating immune cells in mice (n = 4–5) receiving WT CAR T ± vax. See supplemental methods for phenotyping details. (C) IFN-γ ELISPOT. Tumor-bearing WT or Batf3−/− mice (n=5) treated with WT CAR T ± vax. (D) Tumor-bearing mice were left untreated (n=4) or treated with WT or IFN-γ−/− CAR T-vax (n=5). Shown are chemokine expression in tumors 7 days post treatment. (E-F) Ki67 (E) and CCR7(F) expression in intratumoral CD103+ DCs and CD11b+ DCs from mice (n=5) treated with WT CAR T ± vax. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 40 (G-H) Mice bearing ZsGreen+EGFRvIII+CT-2A tumors were treated with WT CAR T, WT CAR T-vax or IFN-γ−/− CAR T-vax (n=5), shown are tumor antigen (ZsGreen) uptake by intratumoral CD103+ DCs (G) and CD11b+ DCs (H). (I-J) IFN-γ ELISPOT (I) and tumor growth (J) in mice (n = 5) treated with WT or IFNGR1−/− CAR T ± vax. (K-N) IFN-γ ELISPOT. (K) WT vs. CD11c-specific IFNGR1 KO tumor-bearing mice (n=5) following WT CAR T ± vax. (L) Tumor-bearing mice (n=5) following WT CAR T or WT CAR T-vax + anti-IFN-γ or anti-IL12(p70). (M) Tumor-bearing WT mice (n=5) following WT or IL12rb2−/− CAR T-vax therapy or in IL12p40−/− mice following WT CAR T-vax. (N) Tumor-bearing WT mice (n=7) following WT CAR T-vax or IFNGR1−/− CAR T ± vax. (O) Tumor growth in mice (n=5–7) left untreated or treated with WT or IFNGR1−/− CAR T ± vax. All mice except those in G-H bear EGFRvIII+CT-2A tumors. Error bars are mean ± 95% CI, , p<0.001; , p<0.01; , p<0.05; ns, not significant by Student’s t-test for B and E-F, by one-way ANOVA with Tukey’s post-test for A, C-D, G-I and K-N, by two-way ANOVA with Tukey’s post-test for J and O. Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 41 Figure. 7. Engineering CAR T-cells for increased IFN-γ expression synergizes with vaccine boosting to enhance antigen spreading and rejection of solid tumors with pre-existing antigen heterogeneity. (A-D). Heterogenous CT-2A tumors were established in C57BL/6 mice. (A)Tumor growth and (B)survival of mice (n=10) after treatment with WT CAR T-vax therapy ± anti-IFN-γ. (C)Tumor growth and (D)survival of mice (n=8) left untreated, receiving WT CAR T, or WT CAR T-vax ± anti-IL12 (p75). (E) IFN-γ secretion from WT or NFAT-IFN-γ CAR T-cells ± anti-CD3/CD28 beads (n=3). (F) IFN-γ ELISPOT. EGFRvIII+ CT-2A tumor-bearing mice (n=6) treated with WT or NFAT-IFN-γ CAR T ± vax. (G-L) Mice bearing heterogenous CT-2A tumors (n=5) treated with WT or NFAT-IFN-γ CAR T ± vax therapy. Enumeration of CAR T (G) and endogenous CD8+ T cells (J) Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 42 infiltrated into tumors as well as the expression of IFN-γ (H for CAR T, K for host CD8 T) and granzyme B (I for CAR T, L for host CD8 T). (M-N) Tumor growth (M) and overall survival (N) of mice bearing heterogenous CT-2A tumors (n=10) treated with WT or NFAT-IFN-γ CAR T ± vax. (O) Schematic overview of CAR T-vax therapy triggered antigen spreading. Created with BioRender.com. Heterogenous CT-2A tumors are EGFRvIII+:EGFRvIII− cells mixed at 80:20 ratio. Error bars are mean ± 95% CI. , p<0.001; , p<0.01; *, p<0.05; ns not significant by one-way ANOVA with Tukey’s post-test for E-L, by two-way ANOVA with Tukey’s post-test for C, and Log-rank (Mantel-Cox) test for B, D and N. Cell. Author manuscript; available in PMC 2023 July 27. Ma et al. Page 43 KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Anti-mouse CD3 (17A2) Alex488 BioLegend 100220; RRID:AB_1732057 Anti-mouse CD8a (53–6.7) BUV395 BD Biosciences 563786; RRID:AB_2732919 Anti-mouse CD8a (53–6.7) BV421 Anti-mouse CD4 (RM4–5) FITC Anti-mouse CD25 (PC61) APC-Cy7 Anti-mouse B220 (RA3–6B2) PE-cy7 Anti-mouse PD-1 (29F.1A12) BV421 Anti-mouse TIM3 (RMT3–23) APC Anti-mouse CD45 (30-F11) Percp-cy5.5 Anti-mouse CD45.1 (A20) BV421 Anti-mouse CD45.2 (104) BUV737 Anti-mouse CD317 (927) Alex488 Anti-mouse CD11c (N418) FITC Anti-mouse CD11b (M1/70) APC-Cy7 Anti-mouse CD24 (M1/69) BV711 Anti-mouse MHC II (M5/114.15.2) BV605 BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend 100738; RRID:AB_11204079 100510; RRID:AB_312713 102026; RRID:AB_830745 103222; RRID: AB_313005 135218; RRID:AB_2561447 119706; RRID:AB_2561656 103132 RRID: AB_893340 110732 BRID: AB_2562563 BD Biosciences 612778; RRID:AB_2870107 BioLegend BioLegend BioLegend BioLegend BioLegend 127012 RRID: AB_1953287 117306; RRID:AB_313775 101226; RRID:AB_830642 563450; RRID:AB_2738213 107639; RRID:AB_2565894 Anti-mouse F4/80 (T45–2342) BUV395 BD Biosciences 565614; RRID:AB_2739304 Anti-mouse CD86 (GL-1) PE-Dazzle 594 Anti-mouse CD103 (2E7) PE Anti-mouse CD45.1 (A20) APC Anti-mouse IFN-γ (XMG1.2) PE Anti-mouse TNF-α (MP6-XT22) APC Anti-mouse Granzyme B (QA16A02) APC Anti-mouse FoxP3 (150D) PE Anti-mouse PGC-1a (D-5) PE Anti-mouse Ki67 (11F6) BV421 Anti-mouse CD206 PE 1E4.2.1 anti-Env antibody Anti-mouse IFN-γ (XMG1.2) Anti-mouse TNF-α (XT3.11) Anti-mouse CD3ε (2C11) Anti-mouse CD28 (37.51) Anti-CD45.1 (A20) Anti-CD45.2 (104) Bacterial and virus strains 5-alpha Competent E. coli BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend Santa Cruz BioLegend BioLegend 105042; RRID:AB_2566409 121406; RRID:AB_1133989 110714; RRID:AB_313503 505808; RRID:AB_315402 506308; AB_315429 372204; RRID:AB_2687028 320007 AB_492981 sc-518025 PE 151208 RRID: AB_2629748 141706 RRID: AB_10895754 Wittrup lab at MIT N/A BioXCell BioXCell BioXCell BioXCell Stem cell Tech Stem cell Tech BE0055; RRID:AB_1107694 BE0058; RRID:AB_1107764 BE0001–1 BRID:AB_1107634 BE0015–1 BRID:AB_1107624 60117BT 60118BT New England Biolabs C2987U Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 44 REAGENT or RESOURCE SOURCE IDENTIFIER A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Chemicals, peptides and recombinant proteins 1,2-distearoyl-sn-glycero-3-phosphoethanolamine-N-[maleimide (polyethylene glycol)-2000] Layson Bio DSPE-PEG-FITC Cyclic-di-GMP Resiquimod Avanti invivogen invivogen GolgiPlug™ Protein Transport Inhibitor (containing Brefeldin A) BD Biosciences Cell Stimulation Cocktail Protease Inhibitor Cocktail Recombinant murine IL-2 Recombinant murine IFN-γ DNase I Collagenase IV CalPhos™ Mammalian Transfection Kit Sytox Red Retronectin TRIzol™ Reagent eBioscience Roche Biolegend Peprotech Sigma Aldrich Worthington Takara Thermo Fisher Takara Thermo Fisher iTAg Tetramer/PE – H-2 Kb OVA (SIINFEKL) MBL international Critical commercial assays NucleoSpin® Plasmid TrypLE™ Express Enzyme Gibco ACK Lysing Buffer CellTrace Violet LIVE/DEAD™ Fixable Aqua Dead Cell Stain Kit, for 405 nm excitation FITC Annexin V Apoptosis Detection Kit Fixation/Permeabilization Solution Kit Foxp3 / Transcription Factor Staining Buffer Set Mouse CD45 microbeads Takara Thermo Fisher Thermo Fisher Thermo Fisher Thermo Fisher BD Biosciences BD Biosciences eBioscience Miltenyi Biotec EasySep™ Mouse CD8+ T Cell Isolation Kit Stemcell Technologies Mouse IFN-γ ELISA kit Mouse IL-2 ELISA kit Mouse IFN-γ ELISPOT Kit RNeasy Micro Kit T-PERTM Proteinase and phosphatase inhibitors iScript™ Reverse Transcription Supermix LEGENDplex™ assays Deposited data Bulk RNA-seq data R&D systems Invitrogen BD Biosciences Qiagen Thermo Fisher Thermo Fisher Biorad BioLegend 100220 810120 tlrl-nacdg tlrl-r848 BDB555029 00–4970-93 5892970001 575408 315–05 10104159001 LS004188 631312 S34859 T100B 15596018 TB-5001–1 740588.250 12605036 A10492–01 C34557 L34966 556547 554714 00–5523-00 130–052-301 19853 DY485 88–7024 551083 74004 78510 78442 1708841 740621 GEO GSE211938 Cell. Author manuscript; available in PMC 2023 July 27. Ma et al. Page 45 REAGENT or RESOURCE sc RNA-seq data Experimental Models: Cell Lines B16F10 cells CT-2A cells MHCII+ CT-2A cells mEGFRvIII-CT-2A cells ZsGreen+ mEGFRvIII-CT-2A cells mEGFRvIII-CT-2A-OVA cells 293 phoenix cells B16F10-OVA cells 2C TCR-58−/− T cell hybridoma cells 7PPG2 TCR-58−/− T cell hybridoma cells MC38 cells TC-1 cells TRP1−/− B16F10 cells Experimental Models: Organism/Strains C57BL/6J mice, CD45.2+ C57BL/6J mice, CD45.1+ Rag1−/− (B6.129S7-Rag1tm1Mom/J) IFNGR1−/− (B6.129S7-Ifngr1tm1Agt/J) Batf3−/− (B6.129S(C)-Batf3tm1Kmm/J) SOURCE GEO IDENTIFIER GSE212453 ATCC CRL-6475; RRID:CVCL_0159 T. Seyfried Lab at Boston college Generated in the Irvine lab Generated in the Irvine lab Generated in the Irvine lab Generated in the Irvine lab N/A N/A N/A N/A N/A ATCC CRL-3214 G. Dranoff Lab at DFCI Birnbaum lab at MIT Birnbaum lab at MIT Wittrup lab at MIT N/A N/A N/A N/A ATCC CRL-2493 Generated in the Irvine lab N/A Jackson Laboratory 000624; RRID:IMSR_JAX:000624 Jackson Laboratory 002014; RRID:IMSR_JAX:002014 Jackson Laboratory 013755; RRID:IMSR_JAX:013755 Jackson Laboratory 003288; RRID:IMSR_JAX:00328 Jackson Laboratory 013755; RRID:IMSR_JAX:013755 CD11c-cre (C57BL/6J-Tg(Itgax-cre,-EGFP)4097Ach/J ) Jackson Laboratory 007567; RRID:IMSR_JAX:007567 IFNGR1-flox (C57BL/6N-Ifngr1tm1.1Rds/J) Jackson Laboratory 025394; RRID:IMSR_JAX:025394 PGC-1α-flox (B6N.129(FVB)-Ppargc1atm2.1Brsp/J) Jackson Laboratory 009666 RRID:IMSR_JAX:009666 LCK-cre (B6.Cg-Tg(Lck-cre)548Jxm/J) IL12rb2−/− (B6;129S1-Il12rb2tm1Jm/J) IL12p40−/− (B6.129S1-Il12btm1Jm/J IFNγ −/− (B6.129S7-Ifngtm1Ts/J) CD8α −/− (B6.129S2-Cd8atm1Mak/J) Oligonucleotides Ccl4 qPCR primers For: CCAAGCCAGCTGTGGTATTCC Rev: GAGCTGCTCAGTTCAACTCC Ccl5 qPCR primers For: GCTGCTTTGCCTACCTCTCC Rev: TCGAGTGACAAACACGACTGC Itgb1 qPCR primers For: ATGCCAAATCTTGCGGAGAAT Rev: TTTGCTGCGATTGGTGACATT Jackson Laboratory 003802 RRID:IMSR_JAX:003802 Jackson Laboratory 003248 RRID:IMSR_JAX:003248 Jackson Laboratory 002693 RRID:IMSR_JAX:002693 Jackson Laboratory 002287; RRID:IMSR_JAX:002287 Jackson Laboratory 002665 RRID:IMSR_JAX:002665 Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich N/A N/A N/A Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ma et al. Page 46 REAGENT or RESOURCE Itga4 qPCR primers For: GATGCTGTTGTTGTACTTCGGG Rev: ACCACTGAGGCATTAGAGAGC Cx3cr1 qPCR primers For: CCCATCTGCTCAGGACCTC Rev: ATGGTTCCAAAGGCCACAATG SOURCE Sigma-Aldrich Sigma-Aldrich IDENTIFIER N/A N/A Cell. Author manuscript; available in PMC 2023 July 27. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1016_j.cell.2023.03.031	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Cell. Author manuscript; available in PMC 2023 July 05. Published in final edited form as: Cell. 2023 May 11; 186(10): 2127–2143.e22. doi:10.1016/j.cell.2023.03.031. A tissue injury sensing and repair pathway distinct from host pathogen defense Siqi Liu1,7, Yun Ha Hur1,7, Xin Cai2, Qian Cong3, Yihao Yang1, Chiwei Xu1, Angelina M. Bilate4, Kevin Andrew Uy Gonzales1, S. Martina Parigi1, Christopher J. Cowley1, Brian Hurwitz1, Ji-Dung Luo5, Tiffany Tseng1, Shiri Gur-Cohen1, Megan Sribour1, Tatiana Omelchenko1, John Levorse1, Hilda Amalia Pasolli6, Craig B. Thompson2, Daniel Mucida4, Elaine Fuchs1,8,* 1Robin Chemers Neustein Laboratory of Mammalian Development and Cell Biology, Howard Hughes Medical Institute, The Rockefeller University, New York, NY 10065, USA 2Cancer Biology and Genetics Program, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 3McDermott Center for Human Growth and Development, Department of Biophysics, and Harold C. Simmons Comprehensive Cancer Center, University of Texas Southwestern Medical Center, Dallas, TX 75390, USA 4Laboratory of Mucosal Immunology, Howard Hughes Medical Institute, The Rockefeller University, New York, NY 10065, USA 5Bioinformatics Resource Center, The Rockefeller University, New York, NY 10065, USA 6Electron Microscopy Resource Center, The Rockefeller University, New York, NY 10065, USA 7These authors contributed equally 8Lead contact This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Correspondence: [email protected]. AUTHOR CONTRIBUTIONS Conceptualization, S.L. and E.F.; methodology, S.L., Y.H.H., E.F., X.C., Q.C., C.X., A.M.B., B.H., C.J.C., K.A.U.G., Y.Y., T.T., S.G.-C., M.S., and H.A.P.; investigation: S.L., Y.H.H., B.H., and M.S. (CRISPR-CAS-mediated ablations in vitro and in mice); S.L., Y.H.H., C.X., and J.L. (sleeping beauty shIl24 mice); S.L., X.C., and C.B.T. (hypoxia and metabolic experiments in vitro); Q.C. (evolutionary tree analyses); S.L., C.X. (PLISH); S.L., Y.H.H., and T.O. (Tre-IL-24); S.L., Y.Y., C.J.C., K.A.U.G., and J.L. (high- throughput studies/analyses); S.L. and H.A.P. (ultrastructural studies); S.L. and S.G.-C. (whole- mount 3D clearance of wounds); S.L., A.M.B., and D.M. (germ-free facility studies); S.M.P. (immune profiling studies); S.L., Y.H.H., and T.T. (all other experiments); visualization, S.L., Y.H.H., and E.F.; funding acquisition, E.F. and S.L; supervision, S.L. and E.F; writing – original draft, S.L., Y.H.H., and E.F. DECLARATION OF INTERESTS S.L. is now an Asst. Prof. in Pharmacology at UT Southwestern Medical Center; X.C. is now an Asst. Prof. in Radiation Oncology at UT Southwestern Medical Center; K.A.U.G. is currently at Novo Nordisk, Research Center, Oxford, England; C.J.C. is now a postdoctoral fellow at NYU; T.T. is now a graduate student at Yale Univ.; B.H. is now a medical student at Weill Cornell Medical College; S.G.-C. is now an Asst. Prof. in Stem Cells and Regenerative Medicine at UCSD; M.S. is now an embryologist at Tennessee Reproductive Medicine in Chattanooga, TN; J.L. is currently at Temple Univ. C.B.T is a founder of Agios Pharmaceuticals. He is on the board of directors of Regeneron and Charles River Laboratories. E.F. is a member of the editorial board of Cell. She is also a former member of the scientific advisory boards of L’Oré al and Arsenal Biosciences and owns stock futures with Arsenal Biosciences. SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.cell.2023.03.031. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. SUMMARY Page 2 Pathogen infection and tissue injury are universal insults that disrupt homeostasis. Innate immunity senses microbial infections and induces cytokines/chemokines to activate resistance mechanisms. Here, we show that, in contrast to most pathogen-induced cytokines, interleukin-24 (IL-24) is predominately induced by barrier epithelial progenitors after tissue injury and is independent of microbiome or adaptive immunity. Moreover, Il24 ablation in mice impedes not only epidermal proliferation and re-epithelialization but also capillary and fibroblast regeneration within the dermal wound bed. Conversely, ectopic IL-24 induction in the homeostatic epidermis triggers global epithelial-mesenchymal tissue repair responses. Mechanistically, Il24 expression depends upon both epithelial IL24-receptor/STAT3 signaling and hypoxia-stabilized HIF1α, which converge following injury to trigger autocrine and paracrine signaling involving IL-24-mediated receptor signaling and metabolic regulation. Thus, parallel to innate immune sensing of pathogens to resolve infections, epithelial stem cells sense injury signals to orchestrate IL-24-mediated tissue repair. In brief Epithelial stem cells sense injury signals to activate an IL-24-mediated tissue repair pathway that is molecularly distinct but functionally parallel to pathogen-induced IFN signaling in innate immunity. Graphical Abstract Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. INTRODUCTION Page 3 Maintaining homeostasis is a hallmark of biological systems, from unicellular organisms to mammals, and is exemplified by our ability to resolve disruptions, including pathogen infection and tissue injury.1 Barrier epithelial tissues of skin, lung, and intestine are the first line of defense against external assaults. Upon infection, these epithelia often sense pathogen-associated molecular pat- terns (PAMPs), such as “non-self” bacterial DNA or viral RNA, which activate pattern recognition receptors and downstream interferon response transcription factors (IRFs) to promote induction and secretion of type-I and -III interferons (IFNs).2,3 Upon IFN engagement, receptor Janus tyrosine kinases (JAKs) become activated, phosphorylating transcription factors STAT1/2 and orchestrating the cell-, tissue-, and organismal-level defense that resists and eliminates pathogens and restores homeostasis.4 Injury is another acute tissue-level insult that multicellular organisms must confront and respond to.1,5 Following injury, hemostasis initiates eschar (scab) formation, while neutrophils and macrophages enter damaged tissue to launch inflammation and clear debris (Figure 1A). Skin heals through re-epithelialization and dermal remodeling. This includes the tightly coordinated migration of epidermal progenitors (epidermal stem cells [EpdSCs]),7–11 followed by proliferation and regeneration of both epidermal and dermal components to restore skin organ homeostasis (Figure 1A).5,12–16 The molecular details underlying the complex wound repair process are still unfolding. Recent studies begin to reveal how tissue damage triggers immediate inflammatory responses.17–20 However, it remains poorly understood, especially in mammals, how injury is sensed by the host to coordinate progressive tissue-/organ-level repair. As a consequence, it is still largely unknown whether responses to tissue damage resemble the innate immune response to infection and, if so, how. Exposed at the body surface, skin is ideal to interrogate how hosts sense and respond to tissue damage (Figure 1A). Here, we identify a wound-induced signaling pathway that can be triggered independently of microbes or adaptive immunity. We show that it is molecularly distinct but functionally similar to pathogen-induced IFN signaling in innate immunity. In this pathway, EpdSCs within the innermost (basal) layer at the wound edge sense wound-induced hypoxia as a damage signal to induce activation and signaling of IFN homolog interleukin-24 (IL-24). Despite being linked previously to injury,21–24 IL-24’s origins, mechanism of activation, and functions remain elusive. We now provide compelling evidence that in hypoxic conditions, an autocrine IL-24/IL-24-receptor signaling/STAT3 loop is induced, which then sustains the HIF1α-mediated expression of epidermal IL-24. In turn, IL-24 acts in an autocrine and paracrine fashion to coordinate re-epithelialization, re-vascularization, dermal fibroblast proliferation, and collagen deposition to restore the damaged tissue to homeostasis. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. RESULTS IL-24 is specifically expressed by EpdSCs near the wound site Page 4 Upon skin wounding, EpdSCs activate (phosphorylate) transcription factor STAT3 (p- STAT3), which is essential for their proliferation and migration at the wound edge.13,25 STAT3 is also activated in nearby dermal cells and remains high in both compartments until healing nears completion (~day-7; Figure 1B). The vital importance of STAT3 in tissue repair led us to wonder whether STAT3’s functional roles in tissue repair might be analogous to those played by STAT1/2 in pathogen resistance.26 To further probe this relation, we compiled a list of signaling factors reported to activate STAT3 (Table S1).27–29 To evaluate their early response to skin injury, we introduced a 6 mm full-thickness wound, and then at day-0 and day-1 post-injury, we microdissected an ~0.5-mm skin region surrounding the wound site and analyzed mRNAs from enzymatically separated dermal and epidermal fractions by quantitative reverse transcription polymerase chain reaction (qRT-PCR). Among factors capable of activating STAT3, only a few exhibited a wound-induced expression pattern. Il24 stood out as a cytokine induced after injury and largely, if not exclusively, in the epidermal fraction (Figure 1C). IL-24 is a conserved member of the IL-10 family, which includes IL-10, IL-22, IL-19, IL-20, and IL-2429,30 (Table S2). Unbiased phylogenetic analyses indicated that this family and its receptors28 share greater sequence/structure homology to IFN and IFN-receptors than other cytokines/cytokine-receptors (Figures S1A and S1B; Table S3). Notably, the heterodimeric receptor subunits of the IFN and IL-10 families also sub-clustered, suggestive of a common ancestral heterodimeric receptor specific to these two families (Figure S1B). In contrast to IFNs, however, the IL-10 cytokine family has not been as clearly linked to pathogens/danger signals. This raises the tantalizing possibility that, during evolution, these pathways may have bifurcated from a common ancestor to cope with the increasing diversity of pathogens and injuries. Most studies on IL-24 center on cultured cells.22–24,31 IL-24’s expression, regulation, and functions in natural physiological settings remain elusive, with both positive and negative effects on wound repair described. To pinpoint the cells expressing Il24 in skin wounds and assess IL-24’s possible importance in repair, we performed fluorescence-activated cell sorting (FACS) and purified the major cellular constituents in and within 0.5–1 mm of the wound bed at times during re-epithelialization (Figure S1C). Il24 was induced primarily within the EpdSC fraction (integrin-α6hiSCA1hiCD34negCD45neg CD31negPDGFR⍺neg CD117neg) at the wound site (Figure 1D). Among other IL-10 family members, only Il19 showed weak induction in EpdSCs following injury (Figure 1E). Probing deeper, we performed 10x single-cell RNA sequencing (scRNA-seq) of skin wounds. Il24 mRNA was predominantly within the epithelial cell cluster (Krt14+) co- expressing basal EpdSC marker integrin-α6 (Itga6) (Figure S1D; red arrows). Analysis of additional 10x scRNA-seq data on wounds32 was consistent with these findings. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 5 We next combined immunofluorescence microscopy and proximity-ligation-based fluorescent in situ hybridization (PLISH)33 to localize Il24. While Krt14 PLISH marked the epidermis of both homeostatic and wound-induced tissue, Il24 PLISH was only detected following injury, where it appeared within 24 h in EpdSCs near the wound site (Figure 1F). As wound-edge EpdSCs migrate into the wound bed, they induce integrin-α5+.7,10 By day-3, the Il24 PLISH signal had intensified within integrin-α5+ basal EpdSCs of the re-epithelializing tongue (Figure 1F). This finding corroborated both our bulk RNA-seq and qPCR results of Il24 mRNA enrichment in the integrin-α5+ migrating EpdSCs (Figures S1E and S1F). Together, these data pointed to the view that an as yet undetermined injury signal(s) is received by nearby EpdSCs, causing them to produce IL-24 predominantly at the migrating epidermal front of the wound bed. Injury-induced IL-24 signaling resembles infection-induced IFN innate immune signaling During infection, pathogen-derived signals trigger a host innate immune response, which frequently leads to IFN production and pathogen clearance.2 Given that in wounds EpdSCs are exposed to microbes, we first tested whether commensal bacteria/microbes at the skin surface are responsible for inducing Il24 following injury. Intriguingly, mice raised under completely sterile (germ-free) conditions still robustly induced Il24 in EpdSCs at the wound edge (Figure 2A). Consistently, Il24 was also induced following wounding of Myd88−/−Trif−/− mice, which lack Tolllike receptor (TLR) signaling essential for many microbial responses (Figure 2B). This was notable, as TLR-signaling functions in the production of some other IL-10 family members.34,35 Together, these results provided compelling evidence that distinct from pathogens/danger signals, which trigger type-I IFNs, a microbe-independent tissue damage signal induces Il24 at the wound edge. Type-I IFNs are induced by the activation of innate immune pathways, whereas type- II IFN (IFN‐γ) is predominantly induced by lymphocytes.36 Recent studies show that adaptive immune cells involving regulatory T cells and IL-17A-expressing Rorγt+ T cells are important for wound repair.11,37 We thus examined whether the adaptive immune system might be responsible for inducing IL-24 in wound-edge EpdSCs. However, when compared against wild-type (WT) mice, Rag2/Il2rg double knockout (DKO) mice, which lack functional lymphocytes alto- gether,38,39 still temporally induced Il24 in EpdSCs at the wound site (Figure 2C). These data point to an upstream damage signal(s) that induces Il24 at the wound site and is independent of adaptive immune cells. Based upon our collective evidence, we hypothesized that, analogous to the sensing of pathogen-derived non-self patterns that prompt somatic cells to activate type-I IFN-receptor- STAT1/2 signaling in defense against microbial infections, injury-induced signals that do not exist in homeostatic conditions (“non-homeostatic” patterns) may be sensed directly by EpdSCs at the wound edge to trigger the activation of IL-24-receptor-p-STAT3 signaling and initiate tissue-damage-mediated repair (Figure 2D). STAT3 activation and epithelial proliferation rely upon IL-24 in wound repair If this IL-24-mediated tissue injury response is analogous to pathogen infection where IFNs are upstream of STAT1/2, then IL-24 should be important for STAT3 activation in wounds. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 6 To test this hypothesis and further interrogate the physiological significance of IL-24 in wound repair, we engineered Il24−/− mice by directly injecting Il24 guide RNA and CAS9 protein into fertilized embryos. Two independent CRISPR-Cas9-generated Il24−/− lines were generated that harbored loss-of-function frameshift mutations within exon 2 (Figures 3A and S2A). Adult Il24−/− mice were healthy, fertile, and indistinguishable from WT littermates at baseline. Upon challenge, however, the wounded Il24−/− epidermis displayed a markedly reduced ability to activate STAT3 specifically near the wound edge where IL-24 was normally expressed (Figures 3B and S2B). In marked contrast, despite IL-6 being oft- considered the major STAT3-activating cytokine in skin,40 Il6 ablation showed little effect on p-STAT3 in wound-induced skin (Figure S2C). Further consistent with reduced p-STAT3 in the Il24−/− migrating epithelial tongue, the thickness of KRT14+ progenitor layers at the wound edge was markedly reduced compared with WT wounded skin (Figure 3B). These data highlight parallels between pathogen and damage response pathways and suggest that IL-24 acts directly on the wound-edge epithelium to sustain p-STAT3 and promote repair. Deletion of IL-20RB, the pan subunit for IL-24-receptor signaling, also displayed defects in p-STAT3 and re-epithelialization, setting IL-24 apart from IL-22 and IL-10, which have been implicated in wound repair but use different heterodimeric receptors.41,42 However, the response to IL-20RB loss was even more robust than IL-24 alone (Figures 3B, S2D, and S2E). This accentuated phenotype is likely attributable to redundancy with IL-19, which is the only other IL-20 subfamily member that both utilizes IL-24 receptors43 and was wound-induced, albeit at lower levels than Il24 (Figures 1C and 1E). RNA-seq analysis confirmed that the shared IL-24/IL-19 receptor subunit IL-20RB, as well as the other two co-receptors, was highly expressed in EpdSCs, indicative of the importance of epithelial IL-24/IL-19 signaling in STAT3 activation and wound re-epithelialization (Figures 3C and S2F). Epithelial IL-24 coordinates dermal repair and re-epithelialization The robust epidermal expression of both IL-24 and its receptor was consistent with autocrine IL-24 action, as discussed above. Interestingly, however, despite lower IL-24- receptor expression in mesenchymal cells, Il24−/− wounded skin dermis displayed marked proliferation defects (Figures 3D and S3A). Seeking the source of these dermal defects, we first co-immunolabeled for markers of proliferation and endothelial cells (CD31, endomucin), where IL-24-receptor expression was appreciable. Notably, in the absence of IL-24, a striking impairment arose in the sprouting of regenerating blood capillaries that normally account for ~50% of proliferating dermal cells in day-5 post-injured skin (Figures S3B–S3D). Consistently, a recently developed clearing method44 in conjunction with whole- mount immunofluorescence and 3D image reconstruction of day-5 wounded skin revealed a marked paucity of dermal blood vessel angiogenesis, which normally closely associates with the overlying migrating epithelial tongue (Figures 3E and S3E). 70% of the epidermis that migrated into the Il24−/− wound bed lacked underlying vascular support, without which epidermal proliferation plummeted (Figures 3D and 3E). Consistent with the importance of IL-24-receptor signaling, Il20rb−/− mice exhibited a similar paucity of proliferating blood Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 7 vessels migrating into the wound bed, a defect still evident even at day-7 after wounding (Figures 3E and S3F). The remaining proliferating dermal cells in WT day-5 wounds were mostly PDGFRα+ fibroblasts, but these too were largely absent in the Il24−/− day-5 wound bed (Figures 3F and S3C). Consistently, the Il24−/− wound bed displayed a paucity of type-I collagen, an essential extracellular matrix (ECM) component secreted by mature fibroblasts to provide structural support for vasculature and the overlying epidermis. Although IL-24 induction did not require pathogens nor adaptive immunity, innate immune cells are involved in tissue damage responses, and hence we examined whether they responded to IL-24 loss. Consistent with their paucity of IL-24 receptors (Figures 3C and S2F), innate immune cell numbers were largely insensitive to IL-24 status (Figure S4A). Besides neutrophils, macrophages were the most abundant immune cells in the wound bed. Although their total cellularity was similar, macrophage distribution and maturation were noticeably perturbed in Il24−/− wounded skin. In day-5 wounds, Arg1+ and MHCII (H2-aa)+ cells were the two major subpopulations of macrophages/monocytes (Figure S4B). In WT wounds, ARG1+ macrophages appeared underneath the migrating epithelial tongue by day-3, and by day-5, as dermal proliferation and angiogenesis populated the region, ARG1+ cells retreated deeper into the wound bed where re-epithelialization and angiogenesis had not yet taken place. In striking contrast, ARG1+ cells in Il24−/− day-5 wounds persisted underneath the migrating epithelium and erroneously overlapped with dermal proliferating cells (Figure S4C). Additional perturbations were noted in MHCII+ cells, which normally tracked with proliferating dermal cells migrating into the wound bed. In the Il24−/− wound bed, they failed to do so (Figure S4C, middle). Given that MHCII+ and ARG1+ macrophages strongly expressed Vegfa (Figure S4B), they likely both contribute to angiogenesis, providing an avenue for why angiogenesis may have been altered in wounds of our IL-24-deficient mice. Indeed, VEGFA immunofluorescence was considerably stronger in the dermal wound bed of WT versus Il24−/− mice (Figure S4C, right). Thus, despite not responding directly, macrophages were nonetheless sensitive to IL-24-dependent changes in the wound bed. Given the known impact of fibroblasts on macrophages,45 the paucity of fibroblasts in the Il24-deficient wound bed may further contribute indirectly to these perturbations. Toluidine blue staining of semithin tissue sections and transmission electron microscopy further substantiated these defects in restoring dermal cellularity (Figures S5A–S5C). The paucity of both mature dermal fibroblasts and abundant collagen deposition, coupled with the persistence of fibrin clots (pseudo-colored in green), left the migrating Il24−/− epithelial tongue atop a fibrin clot rather than collagen-based ECM. The failure to efficiently clear dermal fibrin and cell debris, including red blood cells (RBCs), further underscored the disorganization of macro- phages. These findings underscored an overall decoupling of the normal repair process. Visually, in comparison with WT, the Il24−/− wound healing course was delayed by ~4 days, while hair re-growth, which relies upon proper epithelial-mesenchymal signaling, exhibited Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 8 delays of up to 2 weeks post-injury, also seen at the histological level (Figures S5D and S5E). Wounds eventually healed and hairs regrew. This did not appear to involve obvious compensatory action, as later stage induction of other IL-20 family members—other than a transient increase of Il19—was not observed (Figure S5F). Rather, the results further reflected the dispensability of IL-24 for skin homeostasis. Epithelial-specific depletion of IL-24 recapitulates proangiogenic defects in Il24−/− wounds Although our data showed that, in skin, IL-24 is predominantly produced by wound-edge epithelial cells, IL-24 had previously been reported in other cell types and tissues.46–50 The broad range of wound-related defects upon whole body loss of IL-24 function coupled with a general decline in p-STAT3 signal within the wound bed (Figure 3G) mandated the need to know whether these defects originated specifically from the inability to induce IL-24 in the skin epithelium following injury. To this end, we generated inducible, skin-epithelium- specific Il24-mRNA knockdown mice by directly injecting Krt14-rtTAfertilized mouse eggs with a sleeping beauty system, including two plasmids encoding (1) transposase and (2) transposable elements, including H2BGFP, followed by shIl24 (miRE-shIl24) driven by a TRE regulatory element activatable by the doxycycline (Dox)-sensitive transactivator rtTA (Figure 3H). The majority of skin epithelial progenitors of both founder and F1 offspring mice efficiently and stably integrated the transposon, as indicated by H2BGFP in >90% of epidermal cells following Dox administration. In these shIl24 animals, Dox also efficiently silenced wound-induced Il24 mRNA. Importantly, and as we had observed with full-body Il24−/− wounds, epidermal-specific shIl24 wounds failed to properly coordinate re-epithelialization and dermal angiogenesis (Figure 3H). The expression of IL-24 receptors by endothelial cells and fibroblasts suggested that wound- induced epidermal IL-24 was triggering paracrine effects (Figures 3C and S2F). The paucity of p-STAT3 in both dermis and epidermis of Il24−/− skin added fuel to this fire (Figure 3G). Indeed, upon treating primary endothelial and fibroblast cultures with recombinant IL-24, we observed robust p-STAT3 activation and cell proliferation (Figure S3G). Ectopic IL-24 induction in homeostatic skin epithelium elicits a wound-like response in the absence of injury As IL-24 is specifically activated following injury, we asked whether its ectopic activation might be sufficient to elicit a wound-like response in the absence of injury. A prior study in which IL-24 was constitutively ectopically expressed in skin, starting in embryogenesis, led to epidermal hyperplasia but also neonatal lethality,51 emphasizing the necessity of an inducible approach to unravel the deeper complexities underlying IL-24’s actions. Using our powerful in utero lentiviral delivery method,52 we transduced the skin of mice genetic for an EpdSC (Krt14) specific, Dox-inducible rtTA with Il24 driven by an rtTA-regulated enhancer (TRE) (Figure 4A). Within 48 h of Dox-induction, radical changes arose, marked by enhanced epidermal thickness, elevated dermal collagen deposition, and local vascular remodeling directly beneath the EpdSC layer (Figures 4B–4D). These features were accompanied by marked Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 9 increases in epidermal and dermal proliferation and, a few days thereafter, overt gross phenotypic features of a hyperproliferative skin state appeared (Figures 4B and 4C). In wounded WT skin, the strongest p-STAT3 signal was in epidermal cells, which also expressed the highest level of IL-24 and IL-24 receptors, suggestive of autocrine signaling (Figure 4E). Despite lower levels of IL-24 receptors, endothelial cells and fibroblasts also displayed p-STAT3 in induced IL-24 skin (Figure 4E). Thus, even in the absence of injury, epidermal-specific IL-24 induction was sufficient to elicit a tissue-level wound-like response with both autocrine (epidermal) and paracrine (dermal) IL-24-receptor activation. Tissue-damage-associated hypoxia and HIF1α in wounds are important for robust Il24 expression We next searched for upstream signals that lead to Il24 induction. Our data thus far indicated that the injury signal(s) must be a non-homeostatic pattern that is independent of microbes or adaptive immune cells and only unleashed after wounding. Further corroborating this point, this signal was independent of TNF signaling (Figure S6A), indicating that the mechanism that induces Il24 in a physiological wound is distinct from the patho- logical scenario where the inhibitor of nuclear factor kappa-B kinase subunit beta (IKKβ) is deleted from skin.53 In WT mice, epidermal proliferation during wound repair paralleled newly sprouting blood capillaries (Figure S6B). In Il24-null mice, a deficiency in dermal angiogenesis following injury was among the most notable defects (Figures 3D and 3E). Hence, we posited that the non-homeostatic pattern(s) sensed by EpdSCs following injury may emanate from severed blood vessels. Turning to tissue hypoxia as a top candidate, we began by verifying that the early wound bed of WT skin is hypoxic.54,55 Indeed, hypoxia probe pimonidazole56 strongly labeled the wound bed and, correspondingly, hypoxia-stabilizing transcription factor HIF1α was nuclear, beginning at the immediate WT wound edge following injury and extending to the migrating (IL-24-expressing) epithelial tongue (Figures 5A, 5B, and S6C). Additionally, the intensity of nuclear HIF1α in EpdSCs correlated with distance from blood capillaries, with the most robust signal always in the epithelial tongue at least 100 μm ahead of regener- ating (day-3) blood capillaries. In contrast to day-5 WT wounds, where HIF1α had waned in epidermis concomitant with newly sprouted underlying blood capillaries (Figure S6C), day-5 Il24−/− wounds resembled that of WT day-3 wounds, displaying strong nuclear HIF1α in overlying epidermis that still lacked close contact with blood capillaries (Figure 5C). These data placed hypoxia and HIF1α upstream of IL-24. If hypoxia regulates Il24 expression, the loss of HIF1α might be expected to deleteriously affect wound-stimulated Il24 induc- tion. Indeed, this was the case, as parallel to the well-established HIF1α target gene, Vegfa, Il24 mRNA levels plummeted when HIF1α was conditionally ablated within epidermis prior to wounding (Figures 5D and S6D). Together with IL-24’s importance for dermal blood capillary regeneration and EpdSC proliferation, Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 10 these results suggested that following EpdSC-sensing of wound-generated hypoxia, IL-24 was induced in order to promote revascularization and proper re-epithelialization. Critical roles for both hypoxia/HIF1α and IL-24-receptor/STAT3 signaling in governing robust Il24 expression We next explored whether additional possible non-homeostatic patterns associated with blood vessel disruption could induce IL-24. To this end, we established an in vitro primary EpdSC culture system and tested a panel of conditions pertinent to blood vessel disruption, including not only hypoxia but also nutrient deprivation (e.g., essential amino acids, glucose, and glutamine), alternative ECM (fibrin clots, collagen), and lactate, a major product of anaerobic glycolysis (Figure 6A). We also tested H2O2, as it induces oxidative stress, a first signal induced by the wound for immune cell recruitment.17 Unexpectedly, none of these in vitro conditions, including hypoxia, had a robust effect on Il24 induction (Figure 6A). This was not because of a culture-related impairment in hypoxia- stabilized HIF1α, as traditional HIF1α targets, Pgk1 and Pdk1,57 were induced (Figure S6E). Rather, these results suggested that Il24 induction after injury requires not only hypoxia and HIF1α but also some additional factor(s). Digging deeper, we learned that despite high expression in vivo, IL-24 receptors were silenced in vitro (Figure S6F). Upon reconstitution, IL-24-receptor positive keratinocytes responded to hypoxia, but not to the other conditions, in eliciting Il24 transcription (Figure 6A). Intriguingly, activating Il24 relied upon not only HIF1α but also IL-24-receptor signaling (Figures 6B, 6C, and S7A). The downregulation of IL-24-receptor signaling in vitro provided a likely explanation for why studies based largely on in vitro data have dispensed with IL-24 as either unimportant or counterproductive for epidermal hyperproliferation and wound repair.21,24 The existence of a positive receptor signaling feedback loop for Il24 was reminiscent of that seen for Ifn,36 and shed light on why following tissue damage, only EpdSCs showed robust Il24 induction even though many skin cells experienced acute hypoxia and also stabilized HIF1α (Figures 5A and 5B). Because STAT3 was downstream of IL-24-receptor signaling, we posited that STAT3 might function in concert with HIF1α to regulate Il24. Indeed, when we conditionally targeted epidermal Stat3 and subjected mice to wounding,13 Il24 induction at the wound edge was markedly diminished (Figure 6D). These findings underscored the importance of STAT3 as a major effector of Il24 in tissue injury and placed IL-24 both upstream and downstream of STAT3. In this regard, Il24 also differed from classical HIF1α targets, e.g., Vegfa and Ldha (encoding lactate dehydrogenase A), which showed hypoxia sensitivity and functional HIF1α dependency, but did not rely upon STAT3 for their induction (Figures 6B, S7A, and S7B). Further addressing the importance for hypoxia/HIF1α on Il24 expression specifically, we interrogated the effects of IL-17A produced by wound-activated RORC+ lymphocytes and recently reported to promote HIF1α stabilization after prolonged hypoxia later in the repair process.11 Adaptive immune cells were dispensable for Il24 induction in vivo especially early in the repair process (Figure 2C), and Rag2/Il2rg null mice lack IL-17A-producing Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 11 cells (Figure S7C). That said, under hypoxic conditions in vitro, IL-17A boosted Il24 expression (Figure S7D), revealing an additive, albeit not essential, effect of IL-17A and further underscoring the importance of hypoxia in regulating Il24. Probing deeper, we next examined the wound-induced dynamics of transcription and chromatin accessibility58 at the Il24 locus. Several ATAC (Assay for Transposase-Accessible Chromatin using sequencing)-peaks associated with HIF1α and STAT3 motifs were induced concomitantly with Il24 transcription at wound-edge EpdSCs (Figures 6E and S7E). Cut&Run sequencing59 showed that HIF1α and STAT3 each bound at their cognate sites and in a hypoxia and IL-24 receptor-dependent manner (Figure 6E). In contrast, only HIF1α bound to the Pgk1 locus, and this canonical hypoxia-induced gene was largely refractile to the status of STAT3 (Figures S7E and S7F). We posit that the dual dependency of Il24 on both hypoxia and IL-24-receptor signaling ensures specificity and affords fine-tuning in response to tissue damage. Additional insights into the role of IL-24 in orchestrating wound repair Finally, we returned to how the HIF1α-IL-24-STAT3 axis orchestrates the collective involvement of different cells in repairing damaged tissue, this time focusing on downstream transcriptional targets of the axis and their impact on tissue repair. Upon analyzing known hypoxia-induced HIF1α targets for their sensitivity to IL-24-receptor-dependent expression, Slc2a1, encoding glucose transporter protein type 1 (GLUT1), stood out (Figures 7A, S6E, and S7B). Moreover, of the glucose transporter family of genes, only Slc2a1 was expressed strongly in migrating EpdSCs at the wound edge (Figure 7B). If GLUT1 expression is dependent upon IL-24, then it should show sensitivity to IL-24- receptor activity in vivo as well as in vitro. Indeed, in both wounded Il20rb-null and Il24-deficient mice, GLUT1 was diminished (Figure 7C). Moreover, Glut1 was sensitive to STAT3, as its expression was abolished in Stat3-null epidermal cells at the wound edge (Figure 7D). GLUT1 regulates glucose uptake, leading to elevated lactate production and secretion. We corroborated this effect in our cultured EpdSCs, where the most potent effects on glucose uptake and lactate production were seen under hypoxic conditions and when IL-24- receptor was present (Figure 7E). In contrast to IL-24, lactate can have paracrine effects that don’t require IL-24-receptor signaling, which could explain why macrophages showed positional defects upon IL-24 loss, even though they appeared to lack IL-24-receptor/p- STAT3-signaling. Lactate can also have a proangiogenic effect on macrophages,60,61 raising the possibility of additional signaling circuits unleashed downstream of IL-24-receptor- signaling. That said, conditional ablation of Glut1 in EpdSCs in vivo had on its own a hitherto unappreciated impact on both paracrine effect on angiogenesis and fibroblasts close to the epidermis at the wound edge (Figure 7F). These evidences, combined with our observation that IL-24 directly signals to dermal endothelial cells and fibroblasts (Figure S3G), suggest that by inducing IL-24 in response to injury, EpdSCs orchestrate both autocrine and paracrine cascades of events involving Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 12 proliferation and metabolic changes that together trigger a joint collaboration among tissue cells to orchestrate coordinated repair after injury. DISCUSSION Injury and infection are universal insults to living organisms throughout evolution. The ability to properly sense and respond to acute insults for timely resolution is essential for organismal survival. Numerous PAMPs are known to stimulate IFN signaling to resist infection.2 Here, we uncovered a previously elusive molecular pathway that is induced upon tissue damage, independent of microbes and the adaptive immune system (Figure 7G). At the root of this tissue damage pathway is an IFN homolog, IL-24, which while not expressed in homeostasis, is specifically induced by EpdSCs at the hypoxic wound edge region. The ability to sense tissue damage such as hypoxia in a microbe- independent manner distinguishes IL-24 from PAMP-induced signaling. However, analogous to the role of IFN in resisting pathogen infection, IL-24 coordinates a pro-angiogenetic repair and proliferation program to restore tissue integrity and homeostasis. IFN production must be tightly regulated to prevent inflammation and autoimmunity.36,62 We learned that IL-24 production is similarly tightly regulated and occurs only at the wound site. Although the damaged blood vessels generate a hypoxic state, hypoxia alone was not sufficient for Il24 activation, which also relied upon autocrine IL-24-receptor signaling and STAT3 activation. The feedback loop that we exposed here provides an interesting insight into how the epithelial tongue progresses specifically at the wound site and how it is able to simultaneously coordinate dermal repair in proximity. In the end, the repair process becomes naturally autoregulated at the back end in that as the vasculature is re-established, both the hypoxia-induced signaling and Il24 expression wane. Our data revealed that as an epithelial-derived cytokine induced at the wound site, IL-24 is poised to unleash a multifaceted cascade of paracrine and autocrine effects in coordinating tissue repair. Although IL-24-receptor expression is highest in EpdSCs, nearby dermal endothelial cells and fibroblasts also express the receptor and directly proliferate in response to IL-24. Additionally, however, IL-24 also alters gene expression through its ability to activate STAT3 signaling, and downstream effectors such as the glucose transporter GLUT1. Although GLUT1 has been shown to impact epidermal proliferation and wound re-epithelialization,11,63 we discovered that GLUT1 is highly upregulated in the wound edge epithelium, where it is impacted directly by autocrine IL-24-receptor signaling. IL-24’s ability to alter epithelial metabolic processes, including lactate production to impact mesenchymal repair response within the wound bed takes on newfound importance, as it suggests that IL-24’s paracrine effects may extend beyond whether a cell within the injured skin expresses the IL-24-receptor. In closing, the mechanistic insights we have unraveled here strongly suggest that by sensing injury signals such as hypoxia and autocrine IL-24-receptor/STAT3 signaling to maximize IL-24 production, EpdSCs not only choreograph their own proliferation and re-epithelialization to seal wounds but also coordinate the requisite dermal repair responses that involve blood vessel sprouting and fibroblast reconstruction of the ECM. Our findings Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 13 also offer insights into complex infectious and inflammatory diseases, which can cause secondary tissue damage, the proper repair of which is essential for disease tolerance and host survival.64,65 In this regard, it is intriguing that in severe COVID-19 cases, patients with damaged lungs display prominent IL-24,66 and the colons from patients with ulcerative colitis also express IL-24.49 Taken together, the implications of our findings here are likely to extend broadly to many conditions of tissue damage. Limitations of the study Further investigations will be needed to fully dissect the myriad of possible secondary effects that are likely to be triggered downstream of IL-24 signaling. Given the lack of Il20rb-floxed mice and the complexity of cell types involved, a comprehensive study of IL-24 signaling in each cell type within the wound bed was beyond the scope of the current study. Methodology is currently limited for measuring the in vivo levels of lactate and other metabolites in homeostasis and at wound sites. We mostly limited our studies to female animals, as males tend to fight and introduce wounds that might preclude accurate analyses. STAR★METHODS RESOURCE AVAILABILITY Lead contact—Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Elaine Fuchs ([email protected]). Materials availability—Materials used in this study will be provided upon request and available upon publication. Data and code availability • • • Bulk RNA-, 10x singe-cell RNA-, ATAC-sequencing data and Cut-and- Run sequencing data from this study have been deposited in the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/sra) under accession codes PRJNA731164, PRJNA885018, and PRJNA731304. All other data in the manuscript, supplementary materials and source data are available from the corresponding author upon request. All original code is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS Animals—C57BL/6 and B6.129X1-Gt(ROSA)26Sortm1(EYFP)Cos/J (Rosa26-stop-lox-stop YFP) mice were purchased from The Jackson Laboratory. Krt14-Cre and Krt14-CreER mice were previously generated in the Fuchs laboratory. Il20rb−/− mice were obtained from Genentech, which was previously used in a skin wound healing study.82 Hif1α null mice were obtained by crossing Hif1α floxed animals from The Jackson Laboratory (Stock No: 007561) to Krt14-CreER/Rosa26-YFP (Fuchs Lab) animals. Glut1 null mice were obtained by crossing Glut1 floxed animals from The Jackson Laboratory (Stock No: 031871) to Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 14 Krt14-CreER/Rosa26-YFP (Fuchs Lab) animals. Stat3 cKO mice were obtained by crossing Stat3 floxed animals from The Jackson Laboratory (Stock No:016923) to K14-Cre/ Rosa26- YFP (Fuchs Lab) animals. Myd88−/−(Stock No: 009088) and Trif−/− (Stock No: 005037) mice were obtained from Jackson Laboratories and crossed into Myd88−/−Trif−/− in-house. Rag2−/−Il2rg−/− (Stock No. 4111-F) and control wildtype C57BL/6NTac (Stock No. B6-F) females were purchased from Taconic. TNFR1/TNFR2 DKO mice were purchased from The Jackson Laboratory (Stock No. 003243). In order to generate Il24−/− mice using the CRISPR-Cas9 method, we used the Alt- R CRISPR-Cas9 system from IDTdna. Il24 gRNA (GGAGAACCACCCCTGTCACT) targeting its exon 2 was selected using guidescan (http://www.guidescan.com/). crRNA (containing Il24 gRNA sequence), tracrRNA (IDT cat. #1072533), and recombinant Cas9 (IDT cat. #1081058) were purchased from IDTdna, and crRNA:trRNA:Cas9 RNP particles were assembled in vitro as described by the manufacturer and suspended in injection buffer (1 mM Tris-HCl pH 7.5, 0.1 mM EDTA) at a final RNP concentration of 0.122 μM. The mixture was then injected into the pronucleus of fertil- ized single-cell mouse embryos, and embryos were implanted into the oviducts of pseudo-pregnant wild-type C57BL/6 female mice.83 For the generation of mice with inducible Il24 loss of function specifically in skin epithelium, we used the sleeping beauty sys- tem and mir-E based shRNA method.84 For TRE-inducible Il24 knockdown in vivo, we designed Il24 shRNA with the algorithm from splashRNA,85 and cloned the shRNA with the optimal antisense sequences (TAGAATTTCTGCATCCAGGTCA) into the mir-E backbone86 placed at the 3’UTR of a nucleus-localized H2B-GFP reporter driven by a TRE promoter. After validation of efficient knockdown in keratinocytes in vitro, the TRE-H2B-GFP-shIl24 cassette was cloned into a sleeping beauty transposon (Addgene Plasmid #108352) for injection into the zygotes of K14rtTA mice.87 The transposon plasmid was then mixed with a plasmid encoding trans- posase (pCMV-SB100; Addgene Plasmid #34879) in injection buffer (2.5 ng/μl transposon plasmid; 1.25 ng/μl SB100 transposase plasmid; 5 mM Tris-cl pH 7.4, and 0.1mM EDTA), and injected into the pronucleus of fertilized single-cell mouse embryos of K14rtTA, and embryos were implanted into the oviducts of pseudo-pregnant C57BL/6 female mice. Once the sleeping beauty mice were born, female mice and control littermates were subjected to wounding experiments, while male mice with high transduc- tion efficiency were used as founder mice to back-cross with Krt14-rtTA C57BL/6 female mice to generate F1 offspring mice. Animals were assigned randomly to experimental groups and studies were not blinded. However, age- and sex-matched, and whenever possible, littermates were used for each experiment. For the full-thickness wound healing time course and wound imaging experiments, female mice in the telogen phase of the hair cycle (P50-P65) were used, as males tend to fight and introduce wounds that might preclude accurate analyses. Mice were maintained in the Association for Assessment and Accreditation of Laboratory Animal Care-accredited animal facility of The Rockefeller University (RU), and procedures were performed with Institutional Animal Care and Use Committee (IACUC)-approved protocols. Mice of all strains were housed in an environment with controlled temperature and humidity under specific-pathogen-free conditions, on 12 hour-light:dark cycles, and fed with regular rodent’s chow or doxycycline as described. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 15 Cell lines—293TN HEK cells for lentiviral production were cultured in DMEM medium with 10% FCS (Gibco) and 1 mM sodium pyruvate, 2 mM glutamine, 100 units/mL streptomycin, and 100 mg/mL penicillin. Primary cell cultures—Primary epidermal stem cells (EpdSCs) were maintained at 37°C in a humidified atmosphere containing 7.5% CO2. Cells were cultured in E-low calcium (50 μM Ca 2+) medium made in-house from DMEM/F12 (3:1 ratio) medium supplemented with 15% chelated FBS, 5 μg/mL insulin, 5 μg/mL transferrin, 2 nM triiodothyroxine, 40 μg/mL hydrocortisone, 10 nM cholera toxin and Pen-Strep. 88 C57BL/6 mouse primary dermal microvascular endothelial cells were purchased from Cellbiologics (C57–6064) and pure CD31+ blood endothelial cells were FACS-purified based on markers endomucin, CD31, PDPN and LYVE1.89 The purified blood endothelial cells were then cultured in commercially available endothelial media from Cellbiologics (M1168) containing 5% FBS. Primary fibro- blasts were cultured in DMEM:F12 (3:1) containing Pen-Strep and 10% FBS. METHOD DETAILS Cell culture experiments—For in vitro hypoxia experiments, primary EpdSCs with GFP or IL24-receptor reconstitution were cultured under 21% oxygen (normoxia) or 1% oxygen (hypoxia) in DMEM/F12 (3:1 ratio) medium supplemented with 15% chelated FBS, 5 μg/mL insulin, 5 μg/mL transferrin, 2 nM triiodothyroxine, 40 μg/mL hydrocortisone, 10 nM cholera toxin and Pen-Strep.88 For the generation of each nutrient-deprived condition, amino acid/glucose/glutamine deficient DMEM/F12 (complete deficient media) was made in-house by the MSKCC media core (dialyzed chelated FBS was used), and reconstituted with each nutrient, and the complete medium re- supplemented with all missing nutrients served as a control. Cells were also cultured on the plates coated with poly-L-lysine, fibrin, or collagen as indicated, according to the manufacturer’s instructions. For IL17A stimulation, IL24- receptor reconstituted keratinocytes cells were cultured either under 21% oxygen (normoxia) or 1% oxygen (hypoxia) conditions, with 10ng/ml or 100ng/ml recombinant IL17A for 4 days to mimic chronic hypoxic conditions in the later wound edge. For IL24 stimulation, both endothelial cells and fibroblasts were cultured in low serum condition (1%) for 6 hours before stimulation, followed by 100 ng/ml IL24 treatment for 40 minutes. EdU was added to the culture 15 minutes before harvest. Metabolic analysis in vitro—For measuring glucose uptake and lactate production, GFP control and IL24-receptor reconstituted keratinocytes were plated in triplicates in 12-well plates at 50,000 cells/well and were allowed to attach overnight in E-low calcium medium. The next day, following same media change, cells were placed in normoxic or hypoxic conditions overnight. Media glucose consumption and lactate production were then measured using the YSI 2900 analyzer and normalized by cell number. IL24-receptor reconstitution and Hif1α KO cells—For IL24-receptor reconstitution, either a GFP control or a mouse cDNA encoding IL20RB was cloned into pTY-EF1A- puroR-2a lentiviral vector, and either a GFP control or a mouse cDNA encoding IL22RA1 were cloned into pTY-EF1A-HygromycinR-2a lentiviral vector. Lentivirus was packaged in 293TN cells and then used to infect wild-type or Krt14CreER+;Hif1αfl/ fl keratinocytes, Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 16 which were selected by puromycin 1 μg/ml and hygromycin 50 μg/ml for a week. For floxing out Hif1α exon2 in Krt14CreER+; Hif1αfl/ fl cells to generate Hif1α loss of function cells, 3 μM 4-Hydroxytamoxifen (4OH-Tam) was added to the culture for 4 days. Alternatively, guide RNA targeting Hif1α 90 was cloned into pLentiCRISPRv2-blasticidin construct (Addgene Plasmid #98293). Lentivirus was packaged in 293TN cells and then used to infect GFP control or IL24-receptor reconstituted keratinocytes, which were selected by blasticidin (3 μg/ml, InvivoGen) for 4 days prior to the experiments. Full-thickness wounding—Punch biopsies were performed on anesthetized mice in the telogen phase of the hair cycle (P50-P65).91 For wounding the back skin, dorsal hairs were shaved with clippers and skin was swabbed with ethanol prior to wounding. 4mm or 6 mm biopsy punches (Miltex) were used to make full-thickness wounds. After wounding, tissues were collected at 1, 3, 5 or 7 days after wounding as indicated. Immunofluorescence microscopy—Mouse back skin was dissected, fixed with 4% paraformaldehyde diluted in PBS for 1–2 hours at 4°C, washed with PBS three times, incubated with 30% sucrose at 4°C overnight, and then embedded in OCT (Tissue Tek). Frozen tissue blocks were sectioned at 14 μm on a cryostat (Leica) and mounted on SuperFrost Plus slides (Fisher). The tissue sections were blocked for 1 hour at room temperature with the blocking solution (5% normal donkey serum, 0.5% bovine serum albumin, 2.5% fish gelatin, and 0.3% Triton X-100 in PBS). Sections were then incubated with the indicated primary antibodies diluted in the blocking solution at 4°C overnight. For staining the tissues with an anti-p-STAT3 or an anti-HIF1α antibody, the sections were pretreated with ice-cold 100% methanol prior to the blocking step. The sections were then washed three times with 0.3% Triton X-100 in PBS and incubated with secondary antibodies diluted in the blocking solution at room temperature for 1 hour. Finally, the sections were washed three times with 0.3% Triton X-100 in PBS, three times with PBS containing DAPI at a 1:3,000 dilution, and then mounted with ProLong Dimond Antifade Mountant (Thermo Fisher Scientific). EdU click-it reaction was performed according to the manufacturer’s instructions (Life Technologies) after the secondary antibody incubation and was followed by washing with PBS containing DAPI, as needed. The samples were visualized with an AxioOberver.Z1 epifluorescence microscope equipped with a Hamamatsu ORCA-ER camera and an ApoTome.2 (Carl Zeiss) slider. Tiled and stitched images of sagittal sections were collected using a 20X objective, controlled by Zen software (Carl Zeiss). Alternatively, whole wound images were captured using a BioTek Cytation 5 using a 4x air objective. In order to present a larger wounded area, most of the immunofluorescence images presented (except for Figure 4E, 7D and S3G) were tiled images taken by either AxioOberver.Z1 or Biotek Cytation 5 automatically, and were then stitched into bigger images by respective software Zen (Zeiss) or Gen5 (BioTek). Please note some of the images such as Figure S6B may still show a straight line in between two stitched single images due to imperfect shading correction after processed by Zen. BioTek images did not show such shading correction problem. ImageJ software was used to project Z-stacks and process images. The size of the images was adjusted and assembled in Adobe Illustrator. Scale bars were indicated in the figures and legends. Antibodies against following mouse pro- teins were used for immunofluorescence staining in the study: p-STAT3 (rabbit, Cell Signaling), HIF1α (rabbit, Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 17 Cell Signaling), GLUT1 (rabbit, Abcam), CD31 (rat, Biolegend), Endomucin (Rat, Santa Cruz), GFP (chicken, Abcam), PDGFRa/CD140a (Rat, Biolegend), Intergrin-α5/CD49e (Rat, Biolegend), Krt14 (Chicken, Biolegend), Collagen-I (Rabbit, Abcam), CD31 (Hamster, Millipore), Ki67(Rabbit, Cell Signaling), ARG1(Goat, Novusbio), MHCII(Rat, Biolegend), VEGFA(Goat, R&D Systems). Whole-mount immunostaining for wounded skin—For adult skin wounds, the entire wound bed and 1 mm of skin surrounding the wound were dissected from the back skin and placed on Whatman paper. The tissue was then soaked in PBS for half an hour, and the scab was gently removed if needed, and excess fat tissue was gently removed from the dermis side using sharp forceps. The wounded tissues were then fixed in 4% PFA in PBS for one hour at room temperature, followed by extensive washing in PBS. Tissues were then permeabilized for at least 5 hours (and up to overnight) in 0.3% Triton X-100 in PBS, followed by blocking buffer (2.5% fish gelatin, 5% normal donkey serum, 3% BSA, 0.3% Triton) for additional 2 hours. For immunolabeling, primary antibodies (Krt14, 1:500; Endomucin, 1:300) were incubated at room temperature for two days, followed by extensive washing with 0.3% Triton X-100 in PBS. Samples were then incubated for additional two days at room temperature with secondary antibodies conjugated with Alexa 488, RRX, or 647 (1:500 Life), and DAPI (0.2 μg/ml; 1:500). Samples were washed with 0.3% Triton X-100 and DAPI (1:500) in PBS for 4 hours at room temperature and proceeded to tissue clearing. Tissue clearing—Tissue clearing was performed as previously described with some modifications.44 Stained back skin tissues were transferred through increasing concentrations of ethanol diluted in molecular grade water and adjusted to pH 9.0: 30%, 50%, and 70% for 2 hours each, all at room temperature under gentle shaking. Dehydrated tissues were then incubated for two rounds of 100% ethanol for 2 hours each, at room temperature under gentle shaking, before transferring into 1 ml ethyl cinnamate in Eppendorf tubes (polypropylene) for clearing. Cleared skin was mounted with ethyl cinnamate drops between 2 cover glass sizes 22×40 mm, #0 (Electron Microscopy Science), and placed in the microscope slide holder to acquire images. Images could be acquired within 30 min of tissue clearing or up to 3 months of staining and clearing. Proximity ligation in situ hybridization—Proximity ligation in situ hybridization technology (PLISH) is performed as previously described33 with slight modifications. Mouse skin samples were fixed with 4% paraformaldehyde in DEPC-treated PBS at 4°C for 1 hour, rinsed three times with DEPC-treated PBS, incubated with DEPC-treated 30% Sucrose/PBS solution for a few hours, and embedded in OCT. 10 μm tissue sections were prepared from frozen OCT blocks, pretreated with 25 μg/ml pepsin in 0.1 M HCl at 37°C for 5 minutes, and rinsed with DEPC-treated PBS. After drying at room temperature for approximately 5 minutes, tissue sections on the microscope slides are sealed with adhesive chambers (Grace Bio-Labs, GBL622514), rinsed with Hybridization Buffer (1 M NaTCA, 5 mM EDTA, 50 mM Tris pH 7.4, 0.2 mg/ mL Heparin, and 0.1% LDS in DEPC-treated water), and incubated with a mixture of hybridization probes (sequences listed below, 100 nM final concentration each) in Hybridization buffer at 37°C. After a 2 hour-incubation in Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 18 a humid hybridization oven, the tissue sections were rinsed four times with Hybridization Buffer, incubated with High Salt Buffer (0.25 M NaCl, 50 mM Tris, 2 mM EDTA, and 0.1% LDS in DEPC-treated water) at 37°C for 10 minutes, rinsed once with Circle Hybridization Buffer (2x SSC/20% Formamide, 0.2 mg/mL Heparin, and 0.1% LDS in DEPC treated water), and incubated with 116 nM phosphorylated Common Connector Circle (CCC) oligo and phosphorylated Variable Bridge (VB) oligo (sequences listed below) in Circle Hybridization Buffer at 37°C in a humid hybridization oven. After a 1 hour-incubation, the tissue sections were rinsed twice in Circle Hybridization Buffer, once with 1x T4 DNA ligase buffer (NEB, B0202S) in nuclease-free water (Invitrogen, AM9937), and incubated with a ligation reaction mixture (10 unit/μL T4 DNA ligase (NEB, M0202M), 1x T4 DNA ligase buffer, 0.4 μg/μL BSA, 0.4 unit/μL RNaseOUT (Invitrogen, 10-777-019), 250 mM NaCl, 0.005% Tween-20 in nuclease-free water) at 37°C in a humid hybridization oven. After a 2 hour-incubation, tissue sections were rinsed twice with Circle Hybridization Buffer, rinsed once with 1x phi29 polymerase buffer (Lucigen, NxGen kit 30221) in nuclease-free water, and incubated with a rolling-circle amplification (RCA) reaction mixture (1 unit/μL phi29 polymerase (Lucigen, NxGen kit 30221), 1x phi29 polymerase buffer, 5% Glycerol, 0.25 mM each dNTP, 0.4 μg/μL BSA, 0.4 unit/μL RNaseOUT in nuclease- free water) at 37°C in a humid hybridization oven. After overnight (~16 hours) RCA reaction, the tissue sections were rinsed twice with Label Probe Hybridization Buffer (2x SSC/20% Formamide, 0.2 mg/mL Heparin in nuclease-free water) and incubated with 50 nM Label Probe (sequence listed below) in Label Probe Hybridization Buffer at 37°C in a humid hybridization oven for 2 hours. The labeled samples were washed twice with 0.05% Tween-20 in DEPC-treated PBS, stained with 1 μg/ml DAPI in DEPC-treated PBS, rinsed with DEPC-treated PBS, and imaged on the MIDAS microscope. The DNA oligos used for PLISH were purchased from Eurofins Genomics. The sequences (from 5’ to 3’ end) are listed below: -CCC (5’ phosphorylated, HPLC purification): ATTCCTGACCTAACAAACATGCGTCTATAGTGGAGCCACATAATTAAACCTGGCTA T -VB (5’ phosphorylated, HPLC purification): ACTACTCGACCTATAACCATAACGACGTAAGT -Label Probe (5’ conjugated with Alexa Fluor 647, HPLC purification): ACTATACTACTCGACCTATA -Design of H probes: Il24-H1L: AGGTCAGGAATACTTACGTCGTTATGGAGGGTCCTAAAGTGAAGCCG Il24-H1R: AAAGGGCCAGTGCTCCTGCTTTATAGGTCGAGTAGTATAGCCAGGTT Il24-H2L: AGGTCAGGAATACTTACGTCGTTATGGAGGCTCAGGCAGGGGAGAAT Il24-H2R: GGTTCCAAAGAAGAAGGATTTTATAGGTCGAGTAGTATAGCCAGGTT Il24-H3L: AGGTCAGGAATACTTACGTCGTTATGGTCACTAATGGGAAGCATGGA Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 19 Il24-H3R: AAAACCGCTGGTGTGCACTCTTATAGGTCGAGTAGTATAGCCAGGTT Krt14-H1L: AGGTCAGGAATACTTACGTCGTTATGGTGGCGGTTGGTGGAGGTCAC Krt14-H1R: CCATGACCTTGGTGCGGATCTTATAGGTCGAGTAGTATAGCCAGGTT Krt14-H2L: AGGTCAGGAATACTTACGTCGTTATGGAAAGAGTGAAGCCTATAGGG Krt14-H2R: AGGAAGGACAAGGGTCAAGTTTATAGGTCGAGTAGTATAGCCAGGTT Evolutionary analysis of cytokines/receptors—We retrieved the protein family containing IL24 from Pfam and ECOD databases.92,93 Pfam classifies proteins using sequences while ECOD takes similarity in protein structure into consideration. IL24 belongs to the Pfam family IL10 (PF00726), which is a member of the Pfam clan 4H cytokine (CL0053). 4H cytokine clan is equivalent to the 4-helical cytokine homologous group of ECOD, and we included all the 29 Pfam families from this clan in our study. We identified Pfam domains in each human protein from Uniprot using HMMER (e-value < 0.00001).94,95 A total of 59 human proteins contained Pfam domains from the 4H cytokine clan, and we extracted the sequences of these domains and aligned them using PROMALS3D96 (Table S3). The multiple sequence alignment (MSA) of these Pfam domains were used for phylogenetic analysis by RAxML (-m PROTGAMMAAUTO).97 After initial alignment, we picked representative cytokine from each clade that highlighted in yellow from Table S3, and used the same method to generate a smaller phylogenic tree for presentation. We identified the receptors for all human cytokines based on literature (Table S3). We identified Pfam domains in these cytokine receptors using HMMER and found that majority (35 out of 40) of them contain >=2 tandem immunoglobulin-like (Ig-like) domains in their extracellular regions. We built MSA for two Ig-like domains from these receptors using the following approach. First, we focused on receptors containing two Ig-like domains and obtained the MSA of the tandem Ig-like domains in these receptors. Second, for each cytokine receptor with >= 3 Ig-like domains, we iterated all combinations of two Ig-like domains from it and identified the combination showing maximal sequence similarity, measured by BLOSUM55 matrix to the MSA we built in the first stage. We extracted regions for the best combination for each receptor and concatenated the sequences for the two Ig-like domains to represent this receptor. Finally, we aligned the sequences of two representative Ig-like domains from all the receptors with >= 2 such domains using PROMALS3D, and the resulting MSA was used to reconstruct the phylogeny of these receptors through RAxML. Germ-free mice wounding—Germ-free (GF) C57BL/6 wild-type (WT) mice were kept in germ-free flexible film isolators (Class Biologically Clean Ltd) at Rockefeller University. For wounding experiments, GF C57BL/6 mice were exported to isocages bioexclusion system (Tecniplast, PA, USA) and housed in isocages for the duration of the experiment. Wounding of GF mice was performed in a sterile hood using sterile autoclaved instruments. Wounding of specific-pathogen-free (SPF) C57BL/6 WT mice was performed in the same hood after GF mice were transferred into the isocages. Both GF and SPF mice were then housed in the isocages under the same conditions for 1 or 5 days as described before Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 20 harvesting skin wounds. Mice housed in the isocages were provided with autoclaved food and water. Flow Cytometry Analysis and Cell Sorting—In order to isolate and stain EpdSCs from the homeostatic mouse back skin, subcutaneous fat was removed from the skin with a scalpel, and the skin was placed dermis side down on 0.25% trypsin (Gibco) and 0.1 mg/ml DNase at 37 °C for 45 minutes while shaking gently. For isolating Day-1 wound edge EpdSCs, skin wounds were first excised at about 1–2 mm from the wound edge. Subcutaneous fat was then removed, and the skin was placed on a Whatman filter paper, faced down to be soaked entirely in trypsin, and incubated for 15–18 minutes while shaking gently. For Day-5 or 7 wounds, wounds were excised at 1 mm from the wound edge, placed on a Whatman filter paper, faced down to be soaked entirely in 50 mM EDTA in PBS, and incubated at 37°C for 1 hour while shaking gently. After the incubation, the scabs were firstly removed, the wound edge epidermis including the migrating tongue was then carefully dissected and isolated from the dermis under a dissection microscope. The isolated epidermis was then incubated in trypsin for about 12 minutes while shaking gently. Single-cell suspensions were obtained by scraping the skin to remove the epidermis and hair follicles from the dermis of homeostatic skin or Day-1 wounds. Single-cell suspensions for Day-5 or 6 wounds were obtained by pipetting the suspension to release single cells. Cell suspensions were then filtered through 70 mm, followed by 40 mm strainers. Cell suspensions were incubated with the indicated antibodies for 30 minutes on ice. The following anti-mouse an- tibodies were used for FACS: α6-integrin-PE or BV650 (BD Pharmingen, 1:1,000), CD34-efluor660 or BV421 (eBiosciences, 1:100), Sca-1- PerCP-Cy5.5 (Biolegend, 1:1,000), CD45-APC-Cy7 (Biolegend,1:200), CD31-PE-Cy7 (Biolegend,1:300), biotin-CD117 (Bio- legend, 1:200), CD140a-APC (Biolegend, 1:100), Streptavidin- PE-Cy7 (eBioscience, 1:500), CD90-BV421 (Biolegend, 1:200). For biotin- conjugated primary antibodies, after washing with FACS buffer, cells were incubated with Streptavidin PE-Cy7 (1:500). DAPI was used to exclude dead cells. Cell isolations were performed on BD FACSAriaII SORP running BD FACSDiva software (BD Biosciences). Flow Cytometry Analyses (data acquisitions) were performed using BD LSRFortessa and BD LSRII analyzers running BD FACSDiva software, and the data were analyzed with FlowJo software (BD Biosciences). For the analysis of dermal cells at the wound site, wound tissue was isolated from the back skin, keeping margins as close as 1 mm. The whole wounds were first excised and placed on a Whatman filter paper, faced down to be entirely soaked in PBS for half an hour, softened scabs were then carefully removed to expose live tissue underneath. Tissue was minced in media (RPMI with L-glutamine, β-mercaptoethanol, sodium pyruvate, acid-free HEPES, penicillin and streptomycin), added with Liberase TL (Roche; 250 μg/ml) and 0.1 mg/ml DNase, and digested for 60–90 minutes at 37°C while shaking gently. The digest reaction was stopped by adding 20 μl of 0.5 M EDTA. Single-cell suspensions were then obtained by pipetting the suspension to release single cells. Cells were filtered through a 70 μm strainer, and then a 40 μm strainer. For 10x single cell RNA-seq, the cell suspensions were additionally incubated with ACK lysing buffer (Thermofisher) to remove red blood cells, and then live, single cells were sorted after adding DAPI. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 21 Cell suspensions for other analysis and sorting experiments were then stained with the following antibodies from Biolegend: α6-integrin-PE (1:1,000), CD45-APC-Cy7 (1:200), CD31- PE-Cy7 (1:300), CD11b-BV421 (1:1,500), MHCII-AF700 (1:1,000), CD45-APC- Cy7 (1:200), CD140a-APC (1:100), ITGA5-Ax488 or APC (1:100), Ly6G-PE or APC (1:500). In particular, for the wound bed innate immune cell panel analysis, we used the following combination: CD45-APC-Cy7 (Biolegend, 1:200), CD117-PerCP-Cy5.5 (Biolegend 1:200), Ly6C-FITC, (Biolegend, 1:200), Ly6G-PE (Biolegend, 1:200), Siglec F-APC (Biolegend 1:200), FceRIa-PE-Cy7 (eBioscience, 1:200), CD64-BV605 (Biolegend 1:200), CD11b-BV421 (Biolegend 1:200), MHCII-(I-A/I-E) AF700, (Biolegend 1:200). Dead cells were excluded using a LIVE/DEAD Fixable Blue Dead Cell Stain Kit (Molecular Probes) or DAPI. Flow Cytometry Analyses (data acquisitions) were performed using BD LSRFortessa and BD LSRII analyzers running BD FACSDiva software, and the data were analyzed with FlowJo software (BD Biosciences). Bulk RNA-seq and quantitative RT-PCR—Total RNA from sorted EpdSCs, endothelial cells, dermal fibroblasts, and innate immune cells was purified using the Direct-zol RNA MiniPrep kit (Zymo Research) per the manufacturer’s instructions. DNase treatment was performed to remove genomic DNA (RNase-Free DNase Set, Qiagen). The quality of RNA samples was determined using an Agilent 2100 Bioanalyzer, and all samples for sequencing had RNA integrity (RIN) numbers >8. cDNA library construction using the Illumina TrueSeq mRNA sample preparation kit was performed by the Weill Cornell Medical College Genomic Core facility (New York, NY), and cDNA libraries were sequenced on an Illumina HiSeq 2000 or Illumina Novaseq 6000 instruments. The bulk RNA-seq data analysis was mainly processed in R (version 4.0) environment. The reference genome sequence was fetched from BSGenome.Mmusculus.UCSC.mm10 package (https://bioconductor.org/packages/release/ data/annotation/html/BSgenome.Mmusculus.UCSC.mm10.html ); the GTF file was fetched from TxDb.Mmusculus.UCSC.mm10.knownGene package (https://bioconductor.org/ packages/release/data/annotation/html/TxDb.Mmusculus.UCSC.mm10.knownGene.html). The fastq files were aligned to reference genome by Salmon (version 1.4.0, https:// salmon.readthedocs.io/en/latest/salmon.html), and the counts for each feature were calculated by Salmon. The counting results were imported into DESeq2 object by tximport (https://bioconductor.org/packages/release/bioc/html/tximport.html ). For real-time PCR, equivalent amounts of RNA from FACS-purified cells were reverse-transcribed using the SuperScript™ VILO™ cDNA Synthesis Kit (ThermoFisher Scientific). All cDNAs were normal- ized to equal amounts using housekeeping genes Eef1a1 and Ppib. If not specified in the figure legends, data normalized to Eef1a1 are presented and similar expression trends were also confirmed with Ppib. cDNAs were mixed with indicated gene-specific primers and SYBR green PCR Master Mix (Sigma), and qRT-PCR was performed on an Applied Biosystems 7900HT Fast Real-Time PCR system. 10x single-cell RNA-seq analysis—The raw fastq files of 10X data were mapped to mouse genome (mm10), and the gene expression of each gene in each cell was estimated by the count function of Cell Ranger (v 3.0.2). The counting matrices of the two samples were Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 22 then merged by the aggr function of Cell Ranger. The “.cloupe” file was applied for data visualization with Loupe Browser (v 3.0.0). More customized analyses were processed by Seurat (v 3.0.0) which was developed on R language (version 3.5.2). The following steps were derived from Seurat vignette. First, the filtered counting matrices of the samples were loaded into Seurat object. The features detected in less than five cells were removed. The proportion of mitochondrial genes oriented UMI counts (percent.mt) was also estimated. Then, the Seurat object was subjected to log normalization (Seurat::NormalizeData) and variable features identification (Seurat::FindVariableFeatures). After this step, amount 2000 variable features were identified by vst method. To merge the Seurat objects for all samples, the CCA-based workflow was applied. After merging all samples, the cells with the following criteria were removed: (i) too few genes detected (nFeature_RNA < 200); (ii) potential doublets (nCount_RNA > 99% quantile of UMI counts); potential cell debris (percent.mt > 10%). After removing low quality cells, a principal component analysis was performed (Seurat::RunPCA). The PCs used was determined by an Elbow plot (Seurat::ElbowPlot). In this case, we decided to use the first 15 PCs for the following steps, including identify neighbors (Seurat::FindNeighbors), made UMAP projection (Seurat::RunUMAP). Finally, the clusters were identified by using Louvain clustering with resolution as 0.5 (Seurat::FindClusters). The UMAP projection and clustering information were extracted and imported into Loupe Browser for more customized visualization. EdU and pimonidazole injections—In order to label mitotic cells with EdU, mice were injected intraperitoneally with thymidine analogue 5-Ethynyl-2′-deoxyuridine (EdU, 50 μg/g) (Sigma-Aldrich) 3 hours before sample collection. For labeling tissue hypoxia, pimonidazole (Hypoxyprobe) was prepared as 100 mg/ml in 0.9% saline, and was injected intraperitoneally (60 mg/kg) 1.5 to 2 hours before sample collection. Tamoxifen treatment on mice—Mice expressing Krt14-CreER, as well as their wild- type controls, were treated with the topical application of 0.1% 4-Hydroxytamoxifen (4OH-Tam) diluted in 100% ethanol for 4 days, to manipulate the gene expression in the epidermis. After three days of resting period, the experiments were performed on the back skin of mice as indicated. Doxycycline treatment on mice—Second telogen mice expressing Krt14-rtTA, as well as their control littermates, were put on a high-dose doxycycline (Dox, 2 mg/kg) food chow starting 2 days before the first punch biopsy. The mice were also injected intraperitoneally with 25 μg of Dox per gram of body weight at the time of first punch biopsy. For neonatal mice experiments, pregnant females were put on the high-dose Dox chow one day before they gave birth. Neonatal mice skins were harvested 48, 72, 96 hours after the start of doxy chow. In utero lentiviral transduction—Concentrated lentiviral solutions were produced, and ultrasound-guided in utero injection of concentrated lentivirus was performed in the Comparative Biology Center at The Rockefeller University. Specifically, female mice were anesthetized with isoflurane at embryonic day 9.5, and 500 nL to 1 μL of lentivirus was Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 23 injected into the amniotic sacs of the animal to selectively transduce individual progenitors within the surface ectoderm that will give rise to the skin epithelium. Histology—Mouse back skin was dissected, and fixed with 4% paraformaldehyde diluted in PBS at 4°C overnight. After extensive washing with PBS, the tissues were incubated with 35% Ethanol for 1 hour and then 70% Ethanol for 1 hour. Samples in 70% Ethanol were then sent to Histowiz for processing as well as H&E and Trichrome staining. Toluidine blue staining and TEM—Skin samples were fixed in 2% glutaraldehyde, 4% paraformaldehyde, and 2 mM CaCl2 in 0.1 M sodium cacodylate buffer (pH 7.2) for >1 hour at room temperature, post-fixed in 1% osmium tetroxide, and processed for Epon embedding; ultrathin sections (60–65 nm) were counterstained with uranyl acetate and lead citrate. Images were acquired with a transmission electron microscope (TEM, Tecnai G2–12; FEI, Hillsboro, OR) equipped with a digital camera (AMT BioSprint29). Semithin sections (800 nm) were stained with toluidine blue and photographed with a Zeiss Axio Scope equipped with a Nikon Digital Sight camera. Immunoblot analysis—Cells were lysed in chilled 1x RIPA buffer (10x stock, EMD Millipore) diluted in PBS containing 1 tablet of cOmplete EDTA free pro- tease inhibitor and PhosSTOP phosphatase inhibitor for 30 minutes on ice. Protein was quantified using a Pierce BCA protein quantification kit. 20 μg of total protein lysates were loaded and separated on NuPAGE 4–12% Bis-Tris gels (Thermo Scientific). Proteins were transferred to nitrocellulose membranes, blocked for 1 hour with 5% milk in TBS-T, and incubated with the indicated primary antibodies diluted in TBS-T at 4°C overnight. Membranes were washed in TBS-T and incubated in HRP-coupled secondary anti- bodies at room temperature. Proteins were detected by chemiluminescence using ECL (Thermo Scientific) in a Bio-Rad ChemiDoc Imager. The following primary antibodies and dilutions were used: vinculin (Sigma, V9131 1:2000), HIF‐1α (Cayman Chemical, 10006421, 1:1000), STAT3 (124H6, Cell Signaling, 1:1000), p-STAT3 (D3A7, Cell Signaling 1:1000), LDHA (21799– 1-AP, Proteintech Group, 1:5000) and GLUT1 (ab115730 Abcam 1:1000). Western blot images were processed using Adobe Photoshop CS5. ATAC-Seq library preparation and sequencing—ATAC-seq was performed on 70,000 FACS-purified cells from control and Day-1 wounded samples and processed as previously described.58 Briefly, cells were lysed in ATAC lysis buffer for 5 minutes and then transposed with TN5 transposase (Illumina) for 30 minutes at 37°C. Samples were uniquely barcoded, and the sequencing library was prepared according to manufacturer guidelines (Illumina). Libraries were sequenced on Illumina NextSeq 500. 40-bp paired-end ATAC-seq FASTQs were aligned to the mm10 genome from the Bsgenome.Mmusculus.UCSC.mm10 Bioconductor package (version 1.4.0) using Rsubread’s align method in paired-end mode with fragments between 1 to 5000 base-pairs considered properly paired.98 Normalized, fragment signal bigWigs were created.99 Peak calls for each replicate were made with MACS2 software in BAMPE mode.76,100 Cut and Run-Seq analysis—Cultured EpdSCs from GFPctrl_21%O2 (24hr) and IL22RA/IL20RB_1%O2 (24 hr) were trypsinized into single cell suspensions, and Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 24 CUT&RUN was performed as previously described with minor modifications indicated below.59 Briefly, 650,000 cells were resuspended in crosslinking buffer (10 mM HEPES- NaOH pH 7.5, 100 mM NaCl, 1 mM EGTA, 1 mM EDTA, 1% formaldehyde) and rotated at room temperature for 10 minutes. Crosslinked cells were quenched with glycine at a final concentration of 0.125 M for 5 minutes at room temperature. Cells were washed with cold PBS and resuspended in NE1 buffer (20 mM HEPES-KOH pH7.9, 10 mM KCl, 1mM MgCl2, 1 mM DTT, 0.1% triton X-100 supplemented with Roche complete protease inhibitor EDTA-free) and rotated for 10 minutes at 4°C. Nuclei were washed twice with CUT&RUN wash buffer (20 mM HEPES pH7.5, 150 mM NaCl, 0.5% BSA, 0.5 mM spermidine supplemented with protease inhibitor) and incubated with concanavalin-A (ConA) beads washed with CUT&RUN binding buffer (20 mM HEPES-KOH pH 7.9, 10 mM KCl, 1 mM CaCl2, 1 mM MnCl2) for 10 minutes at 4°C. ConA-bead-bound nuclei were incubated CUT&RUN antibody buffer (CUT&RUN wash buffer supplemented with 0.1% triton X-100 and 2 mM EDTA) and antibody at 4°C overnight. After antibody incubation, ConA-bead-bound nuclei were washed once with CUT&RUN triton wash buffer (CUT&RUN wash buffer supplemented with 0.1% triton X-100) then resuspended and incubated at 4°C for 1 hour in CUT&RUN antibody buffer and 2.5 μL pAG-MNase (EpiCypher). ConA-bound-nuclei were then washed twice with CUT&RUN triton wash buffer, resuspended in 100μL of triton wash buffer, and incubated on ice for 5 minutes. Each 100 μl ConA-bound-nuclei was added with 2 μL 100 mM CaCl incubated on ice for 30 minutes. After adding 100 μL 2x stop buffer (30 mM EGTA), the reaction was incubated at 37°C for 10 minutes. After incubation, ConA-bound-nuclei were captured using a magnet, and the supernatant containing CUT&RUN DNA fragments was collected. The supernatant was incubated at 70°C for 2 hours with 2 μL 10% SDS and 2.5 μL 20mg/mL proteinase K. DNA was purified using PCI and overnight ethanol precipitation with glycogen at −20°C, and was resuspended in 15 μL of buffer EB. CUT&RUN sequencing libraries were generated using NEBNext Ultra II DNA Library Prep Kit for Illumina and NEBNext Multiplex Oligos for Illumina (Index Primer Set 1 and 2). PCR-amplified libraries were purified using 1.2x ratio of AMPure XP beads and eluted in 15 μL 0.1x TE buffer. All CUT&RUN libraries were sequenced on Illumina NextSeq using 40-bp paired-end reads. Reads were aligned to reference genome (mm10) using Bowtie2 (version 2.2.9) and deduplicated with Java (version 2.3.0) Picard tools (http:// broadinstitute.github.io/picard). Reads were flittered to reads smaller or equal to 120 bp using samtools (version 1.3.1). BAM files for each replicate were combined using samtools. Bigwigs were generated using deeptools (version 3.1.2) with RPKM normalization and presented by Integrative Genomics Viewer (IGV) software. Peaks were called using SEACR using a stringent setting and a numeric threshold of 0.01. Peaks were further filtered to have peaks scores greater than 600 for a set of high confident peaks per antibody and condition. The motif analysis was performed with HOMER (version 4.10). 2, mixed gently, and QUANTIFICATION AND STATISTICAL ANALYSIS Group sizes were determined on the basis of the results of the preliminary experiment and mice were assigned at random to groups. The number of animals shown in each figure is indicated in the legends as n = x mice per group and in times, and data are presented with multiple measurements per animal. Experiments were not performed in a blinded fashion. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 25 Statistical analysis was calculated using Prism software (GraphPad). All error bars are mean ± SEM. Experiments were independently replicated, and representative data are shown. Unpaired two-tailed Student’s t-tests were used to ascertain statistical significance between two groups, and one-way ANOVA was used to assess statistical significance between three or more groups with one experimental parameter; Two-way ANOVA was used to assess statistical significance between two or more groups with two experimental parameters. , p < 0.05; , p< 0.01; , p < 0.001; **. p < 0.0001; ns, not significant. See figure legends for more information on statistical tests. Supplementary Material Refer to Web version on PubMed Central for supplementary material. ACKNOWLEDGMENTS We thank J. Racelis, E. Wong, L. Polak, M. Nikolova, and L. Hidalgo for technical support; I. Matos, Y. Miao, L. Xi, and T. Feinberg for experimental contributions; S. Ellis, R. Niec, Y. Miao, H. Yang, M. Schernthanner, A. Gola, C.P. Lu, C. Ng, R. Yang, Y. Yu, J.-L. Casanova, C.M. Rice, and A. Rudensky for discussions. FACS was conducted by RU’s Flow Cytometry Resource Center (S. Mazel, director); ATAC-seq, Cut&Run-seq, and 10x scRNA-seq were conducted by RU’s Genomics Core (C. Zhao, director); RNA-seq was conducted by Weill Cornell Genomics Core Facility (J. Xiang, director). All mouse work was per- formed in RU’s Center for Comparative Biology, under ALAAC accreditation and according to guidelines for animal care set by the National Institutes of Health. E.F. and D.M. are investigators of the Howard Hughes Medical Institute. The following received postdoctoral fellowships: S.L. (RU Women & Science, Jane Coffin Childs); Y.H.H. (AACR-Incyte immuno-oncology research); X.C. (NIH K99 Pathway to Independence award); C.X. (C.H.Li Memorial, Na tional Cancer Center, Charles Revson); K.A.U.G. (Cancer Research Institute Carson Family, Human Frontier Science Program); C.J.C. (NIH F99 Transition award); S.M.P. (Cancer Research Institute Carson Family); and B.H. (NIH F30 award, Tri-institutional Medical Scientist Training Program). This study was supported by grants from the NIH (R01-AR050542 and R01-AR27833, E.F.; K99 AR072780, S.L.), Starr Foundation (E.F.), Robertson Therapeutic Development Funds (S.L. and E.F.), NCI (P30 CA008748, C.B.T.), and NIH (R01 DK093674, D.M.). INCLUSION AND DIVERSITY We worked to ensure diversity in experimental samples through the selection of the cell lines. One or more of the authors of this paper self-identifies as a member of the LGBTQIA+ community. While citing references scientifically relevant for this work, we also actively worked to promote gender balance in our reference list. We avoided “helicopter science” practices by including the participating local contributors from the region where we conducted the research as authors on the paper. REFERENCES 1. Medzhitov R. (2008). Origin and physiological roles of inflammation. Nature 454, 428–435. 10.1038/nature07201. [PubMed: 18650913] 2. Akira S, Uematsu S, and Takeuchi O. (2006). Pathogen recognition and innate immunity. Cell 124, 783–801. 10.1016/j.cell.2006.02.015. [PubMed: 16497588] 3. Constan DA., Nic TJ., and Rauc I. (2021). Innate immune sensing by epithelial barriers. Curr. Opin. Immunol. 73, 1–8. 10.1016/j.coi.2021.07.014. 4. Schneider WM, Chevillotte MD, and Rice CM (2014). Interferon-stimulated genes: a complex web of host defenses. Annu. Rev. Immunol 32, 513–545. 10.1146/annurev-immunol-032713-120231. [PubMed: 24555472] 5. Singer AJ, and Clark RA (1999). Cutaneous wound healing. N. Engl. J. Med 341, 738–746. 10.1056/NEJM199909023411006. [PubMed: 10471461] Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 26 6. Naik S, Larsen SB, Gomez NC, Alaverdyan K, Sendoel A, Yuan S, Polak L, Kulukian A, Chai S, and Fuchs E. (2017). Inflammatory memory sensitizes skin epithelial stem cells to tissue damage. Nature 550, 475–480. 10.1038/nature24271. [PubMed: 29045388] 7. Heller E, Kumar KV, Grill SW, and Fuchs E. (2014). Forces generated by cell intercalation tow epidermal sheets in mammalian tissue morphogenesis. Dev. Cell 28, 617–632. 10.1016/ j.devcel.2014.02.011. [PubMed: 24697897] 8. Park S, Gonzalez DG, Guirao B, Boucher JD, Cockburn K, Marsh ED, Mesa KR, Brown S, Rompolas P, Haberman AM, et al. (2017). Tissue-scale coordination of cellular behaviour promotes epidermal wound repair in live mice. Nat. Cell Biol 19, 155–163. 10.1038/ncb3472. [PubMed: 28248302] 9. Aragona M, Dekoninck S, Rulands S, Lenglez S, Mascré , G., Simons, B.D., and Blanpain, C. (2017). Defining stem cell dynamics and migration during wound healing in mouse skin epidermis. Nat. Commun 8, 14684. 10.1038/ncomms14684. [PubMed: 28248284] 10. Gonzales KAU, Polak L, Matos I, Tierney MT, Gola A, Wong E, Infarinato NR, Nikolova M, Luo S, Liu S, et al. (2021). Stem cells expand potency and alter tissue fitness by accumulating diverse epigenetic memories. Science 374, eabh2444. 10.1126/sci-ence.abh2444. [PubMed: 34822296] 11. Konieczny P, Xing Y, Sidhu I, Subudhi I, Mansfield KP, Hsieh B, Biancur DE, Larsen SB, Cammer M, Li D, et al. (2022). Interleukin-17 governs hypoxic adaptation of injured epithelium. Science 377, eabg9302. 10.1126/science.abg9302. [PubMed: 35709248] 12. Martin P. (1997). Wound healing–aiming for perfect skin regeneration. Science 276, 75–81. [PubMed: 9082989] 13. Keyes BE, Liu S, Asare A, Naik S, Levorse J, Polak L, Lu CP, Nikolova M, Pasolli HA, and Fuchs E. (2016). Impaired epidermal to dendritic T cell signaling slows wound repair in aged skin. Cell 167. 1323–1338.e14. 10.1016/j.cell.2016.10.052. [PubMed: 27863246] 14. Enyedi B, and Niethammer P. (2015). Mechanisms of epithelial wound detection. Trends Cell Biol. 25, 398–407. 10.1016/j.tcb.2015.02.007. [PubMed: 25813429] 15. Shook BA, Wasko RR, Rivera-Gonzalez GC, Salazar-Gatzimas E, López-Giráldez F, Dash BC, Muñoz-Rojas AR, Aultman KD, Zwick RK, Lei V, et al. (2018). Myofibroblast proliferation and heterogeneity are supported by macrophages during skin repair. Science 362. 10.1126/ science.aar2971. 16. Shook BA, Wasko RR, Mano O, Rutenberg-Schoenberg M, Rudolph MC, Zirak B, Rivera- Gonzalez GC, López-Giráldez F, Zarini S, Rezza A, et al. (2020). Dermal adipocyte lipolysis and myofibroblast conversion are required for efficient skin repair. Cell Stem Cell 26. 880–895.e6. 10.1016/j.stem.2020.03.013. [PubMed: 32302523] 17. Niethammer P, Grabher C, Look AT, and Mitchison TJ (2009). A tissue-scale gradient of hydrogen peroxide mediates rapid wound detection in zebrafish. Nature 459, 996–999. 10.1038/nature08119. [PubMed: 19494811] 18. Katikaneni A, Jelcic M, Gerlach GF, Ma Y, Overholtzer M, and Niethammer P. (2020). Lipid peroxidation regulates long-range wound detection through 5-lipoxygenase in zebrafish. Nat. Cell Biol 22, 1049–1055. 10.1038/s41556-020-0564-2. [PubMed: 32868902] 19. Uderhardt S, Martins AJ, Tsang JS, Lämmermann T, and Germain RN (2019). Resident macrophages cloak tissue microlesions to prevent neutrophil-driven inflammatory damage. Cell 177. 541–555.e17. 10.1016/j.cell.2019.02.028. [PubMed: 30955887] 20. Razzell W, Evans IR, Martin P, and Wood W. (2013). Calcium flashes orchestrate the wound inflammatory response through DUOX activation and hydrogen peroxide release. Curr. Biol 23, 424–429. 10.1016/j.cub.2013.01.058. [PubMed: 23394834] 21. Bosanquet DC, Harding KG, Ruge F, Sanders AJ, and Jiang WG (2012). Expression of IL-24 and IL-24 receptors in human wound tissues and the biological implications of IL-24 on keratinocytes. Wound Repair Regen. 20, 896–903. 10.1111/j.1524-475X.2012.00840.x. [PubMed: 23110359] 22. Soo C, Shaw WW, Freymiller E, Longaker MT, Bertolami CN, Chiu R, Tieu A, and Ting K. (1999). Cutaneous rat wounds express c49a, a novel gene with homology to the human melanoma differentiation associated gene, mda-7. J. Cell. Biochem 74, 1–10. [PubMed: 10381256] 23. Kolumam G, Wu X, Lee WP, Hackney JA, Zavala-Solorio J, Gandham V, Danilenko DM, Arora P, Wang X, and Ouyang W. (2017). IL-22R ligands IL-20, IL-22, and IL-24 promote Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 27 wound healing in diabetic db/db mice. PLoS One 12, e0170639. 10.1371/journal.pone.0170639. [PubMed: 28125663] 24. Poindexter NJ, Williams RR, Powis G, Jen E, Caudle AS, Chada S, and Grimm EA (2010). IL-24 is expressed during wound repair and inhibits TGFalpha-induced migration and proliferation of keratinocytes. Exp. Dermatol 19, 714–722. 10.1111/j.1600-0625.2010.01077.x. [PubMed: 20545760] 25. Sano S, Itami S, Takeda K, Tarutani M, Yamaguchi Y, Miura H, Yoshikawa K, Akira S, and Takeda J. (1999). Keratinocyte-specific ablation of Stat3 exhibits impaired skin remodeling, but does not affect skin morphogenesis. EMBO J. 18, 4657–4668. 10.1093/emboj/18.17.4657. [PubMed: 10469645] 26. Meraz MA, White JM, Sheehan KC, Bach EA, Rodig SJ, Dighe AS, Kaplan DH, Riley JK, Greenlund AC, Campbell D, et al. (1996). Targeted disruption of the Stat1 gene in mice reveals unexpected physiologic specificity in the JAK-STAT signaling pathway. Cell 84, 431–442. 10.1016/s0092-8674(00)81288-x. [PubMed: 8608597] 27. Schindler C, and Darnell JE Jr. (1995). Transcriptional responses to polypeptide ligands: the JAK-STAT pathway. Annu. Rev. Biochem 64, 621–651. 10.1146/annurev.bi.64.070195.003201. [PubMed: 7574495] 28. Leaman DW, Leung S, Li X, and Stark GR (1996). Regulation of STAT-dependent pathways by growth factors and cytokines. FASEB J. 10, 1578–1588. [PubMed: 9002549] 29. Ouyang W, Rutz S, Crellin NK, Valdez PA, and Hymowitz SG (2011). Regulation and functions of the IL-10 family of cytokines in inflammation and disease. Annu. Rev. Immunol 29, 71–109. 10.1146/annurev-immunol-031210-101312. [PubMed: 21166540] 30. Renauld JC (2003). Class II cytokine receptors and their ligands: key antiviral and inflammatory modulators. Nat. Rev. Immunol 3, 667–676. 10.1038/nri1153. [PubMed: 12974481] 31. Jiang H, Lin JJ, Su ZZ, Goldstein NI, and Fisher PB (1995). Subtraction hybridization identifies a novel melanoma differentiation associated gene, mda-7, modulated during human melanoma differentiation, growth and progression. Oncogene 11, 2477–2486. [PubMed: 8545104] 32. Haensel D, Jin S, Sun P, Cinco R, Dragan M, Nguyen Q, Cang Z, Gong Y, Vu R, MacLean AL, et al. (2020). Defining epidermal basal cell states during skin homeostasis and wound healing using single-cell transcriptomics. Cell Rep. 30, 3932–3947.e6. 10.1016/j.celrep.2020.02.091. [PubMed: 32187560] 33. Nagendran M, Riordan DP, Harbury PB, and Desai TJ (2018). Automated cell-type classification in intact tissues by single-cell molecular profiling. eLife 7. 10.7554/eLife.30510. 34. Ouyang W, and O’Garra A. (2019). IL-10 family cytokines IL-10 and IL-22: from basic science to clinical translation. Immunity 50, 871–891. 10.1016/j.immuni.2019.03.020. [PubMed: 30995504] 35. Foste SL., and Medzhito R. (2009). Gene-specific control of the TLR-induced inflammatory response. Clin. Immunol. 130, 7–15. 10.1016/j.clim.2008.08.015. 36. Ivashkiv LB, and Donlin LT (2014). Regulation of type I interferon responses. Nat. Rev. Immunol 14, 36–49. 10.1038/nri3581. [PubMed: 24362405] 37. Truong C, Guo W, Woodside L, Gang A, Savage P, Infarinato N, Stewart K, Polak L, Levorse J, Pasolli A, et al. (2021). Skin stem cells orchestrate de novo generation of extrathymic regulatory T cells to establish a temporary protective niche during wound healing. Preprint at bioRxiv. 10.1101/2021.08.16.456570. 38. Shinkai Y, Rathbun G, Lam KP, Oltz EM, Stewart V, Mendelsohn M, Charron J, Datta M, Young F, and Stall AM (1992). RAG-2-deficient mice lack mature lymphocytes owing to inability to initiate V(D)J rearrangement. Cell 68, 855–867. 10.1016/0092-8674(92)90029-c. [PubMed: 1547487] 39. Cao X, Shores EW, Hu-Li J, Anver MR, Kelsall BL, Russell SM, Drago J, Noguchi M, Grinberg A, and Bloom ET (1995). Defective lymphoid development in mice lacking expression of the common cytokine receptor gamma chain. Immunity 2, 223–238. 10.1016/1074-7613(95)90047–0. [PubMed: 7697543] 40. Johnson BZ, Stevenson AW, Preˆ le CM, Fear MW, and Wood FM (2020). The role of IL-6 in skin fibrosis and cutaneous wound healing. Biomedicines 8. 10.3390/biomedicines8050101. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 28 41. McGee HM, Schmidt BA, Booth CJ, Yancopoulos GD, Valenzuela DM, Murphy AJ, Stevens S, Flavell RA, and Horsley V. (2013). IL-22 promotes fibroblast-mediated wound repair in the skin. J. Invest. Dermatol 133, 1321–1329. 10.1038/jid.2012.463. [PubMed: 23223145] 42. Short WD, Rae M, Lu T, Padon B, Prajapati TJ, Faruk F, Olutoye OO, 2nd, Yu, L., Bollyky, P., Keswani, S.G., and Balaji, S. (2023). Endogenous interleukin-10 contributes to wound healing and regulates tissue repair. J. Surg. Res 285, 26–34. 10.1016/j.jss.2022.12.004. [PubMed: 36640607] 43. Rutz S, Wang X, and Ouyang W. (2014). The IL-20 subfamily of cytokines–from host defence to tissue homeostasis. Nat. Rev. Immunol 14, 783–795. 10.1038/nri3766. [PubMed: 25421700] 44. Gur-Cohen S, Yang H, Baksh SC, Miao Y, Levorse J, Kataru RP, Liu X, de la Cruz-Racelis J, Mehrara BJ, and Fuchs E. (2019). Stem cell-driven lymphatic remodeling coordinates tissue regeneration. Science 366, 1218–1225. 10.1126/science.aay4509. [PubMed: 31672914] 45. Zhou X, Franklin RA, Adler M, Jacox JB, Bailis W, Shyer JA, Flavell RA, Mayo A, Alon U, and Medzhitov R. (2018). Circuit design features of a stable two-cell system. Cell 172. 744–757.e17. 10.1016/j.cell.2018.01.015. [PubMed: 29398113] 46. Dabitao D, Hedrich CM, Wang F, Vacharathit V, and Bream JH (2018). Cell-specific requirements for STAT proteins and type I IFN receptor signaling discretely regulate IL-24 and IL-10 expression in NK cells and macrophages. J. Immunol 200, 2154–2164. 10.4049/jimmunol.1701340. [PubMed: 29436412] 47. Chong WP, Mattapallil MJ, Raychaudhuri K, Bing SJ, Wu S, Zhong Y, Wang W, Chen Z, Silver PB, Jittayasothorn Y, et al. (2020). The cytokine IL-17A limits Th17 pathogenicity via a negative feedback loop driven by autocrine induction of IL-24. Immunity 53. 384–397.e5. 10.1016/j.immuni.2020.06.022. [PubMed: 32673565] 48. Liu G, Jia J, Zhong J, Yang Y, Bao Y, and Zhu Q. (2022). TCDD-induced IL-24 secretion in human chorionic stromal cells inhibits placental trophoblast cell migration and invasion. Reprod. Toxicol 108, 10–17. 10.1016/j.reprotox.2022.01.001. [PubMed: 34995713] 49. Smillie CS, Biton M, Ordovas-Montanes J, Sullivan KM, Burgin G, Graham DB, Herbst RH, Rogel N, Slyper M, Waldman J, et al. (2019). Intra- and inter-cellular rewiring of the human colon during ulcer- ative colitis. Cell 178, 714–730.e22. 10.1016/j.cell.2019.06.029. [PubMed: 31348891] 50. Kawada S, Nagasawa Y, Kawabe M, Ohyama H, Kida A, Kato-Kogoe N, Nanami M, Hasuike Y, Kuragano T, Kishimoto H, et al. (2018). Iron-induced calcification in human aortic vascular smooth muscle cells through interleukin-24 (IL-24), with/without TNF-alpha. Sci. Rep 8, 658. 10.1038/s41598-017-19092-1. [PubMed: 29330517] 51. He M, and Liang P. (2010). IL-24 transgenic mice: in vivo evidence of overlapping functions for IL-20, IL-22, and IL-24 in the epidermis. J. Immunol 184, 1793–1798. 10.4049/ jimmunol.0901829. [PubMed: 20061404] 52. Beronja S, Livshits G, Williams S, and Fuchs E. (2010). Rapid functional dissection of genetic networks via tissue-specific transduction and RNAi in mouse embryos. Nat. Med 16, 821–827. 10.1038/nm.2167. [PubMed: 20526348] 53. Kumari S, Bonnet MC, Ulvmar MH, Wolk K, Karagianni N, Witte E, Uthoff-Hachenberg C, Renauld JC, Kollias G, Toftgard R, et al. (2013). Tumor necrosis factor receptor signaling in keratinocytes triggers interleukin-24-dependent psoriasis-like skin inflammation in mice. Immu- nity 39, 899–911. 10.1016/j.immuni.2013.10.009. 54. Shweiki D, Itin A, Soffer D, and Keshet E. (1992). Vascular endothelial growth factor induced by hypoxia may mediate hypoxia-initiated angiogenesis. Nature 359, 843–845. 10.1038/359843a0. [PubMed: 1279431] 55. Hong WX, Hu MS, Esquivel M, Liang GY, Rennert RC, McArdle A, Paik KJ, Duscher D, Gurtner GC, Lorenz HP, and Longaker MT (2014). The role of hypoxia-inducible factor in wound healing. Adv. Wound Care (New Rochelle) 3, 390–399. 10.1089/wound.2013.0520 . 56. Li F, Lee KE, and Simon MC (2018). Detection of hypoxia and HIF in paraffin-embedded tumor tissues. Methods Mol. Biol 1742, 277–282. 10.1007/978-1-4939-7665-2_24. 57. Schofield CJ, and Ratcliffe PJ (2004). Oxygen sensing by HIF hydroxylases. Nat. Rev. Mol. Cell Biol 5, 343–354. 10.1038/nrm1366. [PubMed: 15122348] Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 29 58. Buenrostro JD, Wu B, Chang HY, and Greenleaf WJ (2015). ATAC-seq: a method for assaying chromatin accessibility genome-wide. Curr. Protoc. Mol. Biol 109, 21.29.1–21.29.9. 10.1002/0471142727.mb2129s109. 59. Meers MP, Bryson TD, Henikoff JG, and Henikoff S. (2019). Improved CUT&RUN chromatin profiling tools. eLife 8. 10.7554/eLife.46314. 60. Colegio OR, Chu NQ, Szabo AL, Chu T, Rhebergen AM, Jairam V, Cyrus N, Brokowski CE, Eisenbarth SC, Phillips GM, et al. (2014). Functional polarization of tumour-associated macrophages by tumour-derived lactic acid. Nature 513, 559–563. 10.1038/nature13490. [PubMed: 25043024] 61. Carmona-Fontaine C, Deforet M, Akkari L, Thompson CB, Joyce JA, and Xavier JB (2017). Metabolic origins of spatial organization in the tumor microenvironment. Proc. Natl. Acad. Sci. USA 114, 2934–2939. 10.1073/pnas.1700600114. [PubMed: 28246332] 62. Liu S, Cai X, Wu J, Cong Q, Chen X, Li T, Du F, Ren J, Wu YT, Grishin NV, and Chen ZJ (2015). Phosphorylation of innate immune adaptor proteins MAVS, STING, and TRIF induces IRF3 activation. Science 347, aaa2630. 10.1126/science.aaa2630. [PubMed: 25636800] 63. Zhang Z, Zi Z, Lee EE, Zhao J, Contreras DC, South AP, Abel ED, Chong BF, Vandergriff T, Hosler GA, et al. (2018). Differential glucose requirement in skin homeostasis and injury identifies a therapeutic target for psoriasis. Nat. Med 24, 617–627. 10.1038/s41591-018-0003-0. [PubMed: 29662201] 64. Medzhitov R, Schneider DS, and Soares MP (2012). Disease tolerance as a defense strategy. Science 335, 936–941. 10.1126/science.1214935. [PubMed: 22363001] 65. McCarville JL, and Ayres JS (2018). Disease tolerance: concept and mechanisms. Curr. Opin. Immunol 50, 88–93. 10.1016/j.coi.2017.12.003. [PubMed: 29253642] 66. Filbin MR, Mehta A, Schneider AM, Kays KR, Guess JR, Gentili M, Fenyves BG, Charland NC, Gonye ALK, Gushterova I, et al. (2021). Longitudinal proteomic analysis of severe COVID-19 reveals survival-associated signatures, tissue-specific cell death, and cell-cell interactions. Cell Rep. Med 2, 100287. 10.1016/j.xcrm.2021.100287. [PubMed: 33969320] 67. Stringer BW, Day BW, D’Souza RCJ, Jamieson PR, Ensbey KS, Bruce ZC, Lim YC, Goasdoué , K., Offenhäuser, C., Akgül, S., et al. (2019). A reference collection of patient-derived cell line and xenograft models of proneural, classical and mesenchymal glioblastoma. Sci. Rep 9, 4902. 10.1038/s41598-019-41277-z. [PubMed: 30894629] 68. Wang Y, Pryputniewicz-Dobrinska D, Nagy EÉ ., Kaufman, C.D., Singh, M., Yant, S., Wang, J., Dalda, A., Kay, M.A., Ivics, Z., and Izsvá k, Z. (2017). Regulated complex assembly safeguards the fidelity of Sleeping Beauty transposition. Nucleic Acids Res. 45, 311–326. 10.1093/nar/ gkw1164. [PubMed: 27913727] 69. Mátés L, Chuah MK, Belay E, Jerchow B, Manoj N, Acosta-Sanchez A, Grzela DP, Schmitt A, Becker K, Matrai J, et al. (2009). Molecular evolution of a novel hyperactive sleeping beauty transposase enables robust stable gene transfer in vertebrates. Nat. Genet 41, 753–761. 10.1038/ ng.343. [PubMed: 19412179] 70. Schneider CA, Rasband WS, and Eliceiri KW (2012). NIH Image to ImageJ: 25 years of image analysis. Nat. Methods 9, 671–675. 10.1038/nmeth.2089. [PubMed: 22930834] 71. R Development Core Team (2013). R: a language and environment for statistical computing (R Foundation for Statistical Computing). 72. Team BC; Maintainer BP (2019). TxDb.Mmusculus.UCSC.mm10.known Gene: annotation package for TxDb object(s). R package version 3.4.7. https://bioconductor.org/packages/release/ data/annotation/html/TxDb.Mmusculus.UCSC.mm10.knownGene.html. 73. Patro R, Duggal G, Love MI, Irizarry RA, and Kingsford C. (2017). Salmon provides fast and bias- aware quantification of transcript expression. Nat. Methods 14, 417–419. 10.1038/nmeth.4197. [PubMed: 28263959] 74. Soneson C, Love MI, and Robinson MD (2015). Differential analyses for RNA-seq: transcript- level estimates improve gene-level inferences. F1000Res 4, 1521. 10.12688/f1000research.7563.2. [PubMed: 26925227] 75. Team TBD (2021). BSgenome.Mmusculus.UCSC.mm10: full genome sequences for Mus musculus (UCSC version mm10, based on GRCm38.p6). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 30 R package version 1.4.3. https://bioconductor.org/packages/release/data/annotation/html/ BSgenome.Mmusculus.UCSC.mm10.html. 76. Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, and Liu XS (2008). Model-based analysis of ChIP-Seq (MACS). Genome Biol. 9, R137. 10.1186/gb-2008-9-9-r137. [PubMed: 18798982] 77. Langmead B, and Salzberg SL (2012). Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359. 10.1038/nmeth.1923. [PubMed: 22388286] 78. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, and Durbin R; 1000 Genome Project Data Processing Subgroup (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079. 10.1093/bioin-formatics/btp352. [PubMed: 19505943] 79. Ramıŕez F, Ryan DP, Grüning B, Bhardwaj V, Kilpert F, Richter AS, Heyne S, Dündar F, and Manke T. (2016). deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res. 44, W160–W165. 10.1093/nar/gkw257. [PubMed: 27079975] 80. Robinson JT, Thorvaldsdó ttir, H., Winckler, W., Guttman, M., Lander, E.S., Getz, G., and Mesirov, J.P. (2011). Integrative genomics viewer. Nat. Biotechnol 29, 24–26. 10.1038/nbt.1754. [PubMed: 21221095] 81. Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre C, Singh H, and Glass CK (2010). Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589. 10.1016/j.molcel.2010.05.004. [PubMed: 20513432] 82. Zheng Y, Valdez PA, Danilenko DM, Hu Y, Sa SM, Gong Q, Abbas AR, Modrusan Z, Ghilardi N, de Sauvage FJ, and Ouyang W. (2008). Interleukin-22 mediates early host defense against attaching and effacing bacterial pathogens. Nat. Med 14, 282–289. 10.1038/nm1720. [PubMed: 18264109] 83. Vasioukhin V, Degenstein L, Wise B, and Fuchs E. (1999). The magical touch: genome targeting in epidermal stem cells induced by tamoxifen application to mouse skin. Proc. Natl. Acad. Sci. USA 96, 8551–8556. 10.1073/pnas.96.15.8551. [PubMed: 10411913] 84. Garrels W, Talluri TR, Ziegler M, Most I, Forcato DO, Schmeer M, Schleef M, Ivics Z, and Kues WA (2016). Cytoplasmic injection of murine zygotes with Sleeping Beauty transposon plasmids and minicircles results in the efficient generation of germline transgenic mice. Biotechnol. J 11, 178–184. 10.1002/biot.201500218. [PubMed: 26470758] 85. Pelossof R, Fairchild L, Huang CH, Widmer C, Sreedharan VT, Sinha N, Lai DY, Guan Y, Premsrirut PK, Tschaharganeh DF, et al. (2017). Prediction of potent shRNAs with a sequential classification algorithm. Nat. Biotechnol 35, 350–353. 10.1038/nbt.3807. [PubMed: 28263295] 86. Fellmann C, Hoffmann T, Sridhar V, Hopfgartner B, Muhar M, Roth M, Lai DY, Barbosa IA, Kwon JS, Guan Y, et al. (2013). An optimized microRNA backbone for effective single-copy RNAi. Cell Rep. 5, 1704–1713. 10.1016/j.celrep.2013.11.020. [PubMed: 24332856] 87. Garrels W, Talluri TR, Apfelbaum R, Carratalá , Y.P., Bosch, P., Pözsch, K., Grueso, E., Ivics, Z., and Kues, W.A. (2016). One-step multiplex transgenesis via sleeping beauty transposition in cattle. Sci. Rep 6, 21953. 10.1038/srep21953. [PubMed: 26905416] 88. Nowak JA, and Fuchs E. (2009). Isolation and culture of epithelial stem cells. Methods Mol. Biol 482, 215–232. 10.1007/978-1-59745-060-7_14. [PubMed: 19089359] 89. Niec RE, Chu T, Schernthanner M, Gur-Cohen S, Hidalgo L, Pasolli HA, Luckett KA, Wang Z, Bhalla SR, Cambuli F, et al. (2022). Lymphatics act as a signaling hub to regulate intestinal stem cell activity. Cell Stem Cell 29. 1067–1082.e18. 10.1016/j.stem.2022.05.007. [PubMed: 35728595] 90. Schwörer S, Berisa M, Violante S, Qin W, Zhu J, Hendrickson RC, Cross JR, and Thompson CB (2020). Proline biosynthesis is a vent for TGFbeta-induced mitochondrial redox stress. EMBO J. 39, e103334. 10.15252/embj.2019103334. [PubMed: 32134147] 91. Hsu YC, Li L, and Fuchs E. (2014). Transit-amplifying cells orchestrate stem cell activity and tissue regeneration. Cell 157, 935–949. 10.1016/j.cell.2014.02.057. [PubMed: 24813615] Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 31 92. El-Gebali S, Mistry J, Bateman A, Eddy SR, Luciani A, Potter SC, Qureshi M, Richardson LJ, Salazar GA, Smart A, et al. (2019). The Pfam protein families database in 2019. Nucleic Acids Res. 47, D427–D432. 10.1093/nar/gky995. [PubMed: 30357350] 93. Cheng H, Schaeffer RD, Liao Y, Kinch LN, Pei J, Shi S, Kim BH, and Grishin NV (2014). ECOD: an evolutionary classification of protein domains. PLoS Comput. Biol 10, e1003926. 10.1371/journal.pcbi.1003926. [PubMed: 25474468] 94. Consortium UniProt (2021). UniProt: the universal protein knowledge-base in 2021. Nucleic Acids Res. 49, D480–D489. 10.1093/nar/gkaa1100. [PubMed: 33237286] 95. Eddy SR (2009). A new generation of homology search tools based on probabilistic inference. Genome Inform. 23, 205–211. [PubMed: 20180275] 96. Pei J, and Grishin NV (2014). PROMALS3D: multiple protein sequence alignment enhanced with evolutionary and three-dimensional structural information. Methods Mol. Biol 1079, 263–271. 10.1007/978-1-62703-646-7_17. [PubMed: 24170408] 97. Stamatakis A. (2014). RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics 30, 1312–1313. 10.1093/bioinformatics/btu033. [PubMed: 24451623] 98. Lia Y., Smyt GK., and Sh W. (2019). The R package Rsubread is easier, faster, cheaper and better for alignment and quantification of RNA sequencing reads. Nucleic Acids Res. 47, e47. 10.1093/nar/gkz114. [PubMed: 30783653] 99. Lawrence M, Gentleman R, and Carey V. (2009). rtracklayer: an R package for interfacing with genome browsers. Bioinformatics 25, 1841–1842. 10.1093/bioinformatics/btp328. [PubMed: 19468054] 100. Feng J, Liu T, Qin B, Zhang Y, and Liu XS (2012). Identifying ChIP-seq enrichment using MACS. Nat. Protoc 7, 1728–1740. 10.1038/nprot.2012.101. [PubMed: 22936215] 101. Wang M, Tan Z, Zhang R, Kotenko SV, and Liang P. (2002). Interleukin 24 (MDA-7/MOB-5) signals through two heterodimeric receptors, IL-22R1/IL-20R2 and IL-20R1/IL-20R2. J. Biol. Chem 277, 7341–7347. 10.1074/jbc.M106043200. [PubMed: 11706020] Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 32 Highlights • • • • Upon injury, IL-24 is induced specifically in epithelial stem cells at wound edges Il24 is regulated by hypoxia and STAT3, independent of microbes, B cells, or T cells IL-24 acts in autocrine and paracrine signaling to regulate proliferation and metabolism Epithelial stem cells sense tissue damage and orchestrate organ-level repair Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 33 Figure 1. IL-24 is specifically produced by epithelial stem cells near the wound site (A) Schematic of the wound repair process in mouse skin. (B) Sagittal sections of homeostatic skin, and wounds (days indicated) immunolabeled for p-STAT3 at Tyr705 (n = 5 mice). (C) qRT-PCR for putative STAT3-targeting cytokines in homeostatic skin and day-1 wound. Il1β served as a positive control.6 Values were normalized to Ppib (n = 3 mice). (D) qRT-PCR of Il24 mRNA in FACS-purified cell populations isolated from homeostatic and wounded skin (n = 3 mice). (E) IL-10 cytokine family expression from RNA-seq performed on FACS-purified EpdSCs from homeostatic and wounded skin. TPM, transcripts per kilobase million (n = 3 mice). (F) PLISH (proximity-ligation-based in situ hybridization) images of sagittal sections of homeostatic and wounded skin, probed for Il24 and Krt14 mRNA. Serial skin sections of Il24 PLISH and immunolabeling of integrin-α5 in day-3 wounds. The red-boxed region was magnified and shown at the right to highlight the Il24 PLISH signal in the re-epithelializing (migrating) epidermis. Asterisk () denotes autofluorescence of hair shaft and stratum corneum (n = 3 mice). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 34 Experiments were performed R ≥3×. White dotted lines, epidermal-dermal border; wound site, red dotted line; epidermal migration direction, red arrow. DAPI, nuclei; scale bars, 100 μm. Data in (D) and (E) are presented as mean ± SEM. N.D., not detected. See also Figure S1 and Tables S1 and S2. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 35 Figure 2. Injury-induced IL-24 signaling resembles infection-induced interferon signaling (A) qRT-PCR of Il24 mRNA in FACS-purified EpdSCs from homeostatic and wounded skin from specific-pathogen-free (SPF) vs. germ-free (GF) C57BL/6J WT mice (SPF, n = 5–6, GF, n = 5–9 mice). (B) qRT-PCR of Il24 mRNA in epidermis microdissected from homeostatic and wounded skin from WT vs. Myd88−/−Trif−/− mice (n = 3 mice per genotype; representative of 3 independent experiments). (C) qRT-PCR analysis of Il24 mRNA in EpdSCs FACS-purified from homeostatic and wounded skin from WT vs. Rag2/IL2rg DKO mice. Note that Rag2/IL2rg DKO mice lack all functional lymphocytes (n = 5–7 mice per genotype). (D) Diagram depicting our central hypothesis that parallel but distinct signaling pathways are used for responding to and resolving pathogen infection and tissue injury. Steps tackled in current study are highlighted by question marks. Data in (A)–(C) are presented as mean ± SEM. Statistical significance was determined using two-tailed unpaired Student’s t tests; ns; not significant; N.D.; not detected. Dots in the graphs indicate data from individual mice. See also Table S3. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 36 Figure 3. Epithelial-expressed IL-24 coordinates dermal repair and re-epithelialization (A) Schematic of two C57BL/6J Il24−/− mouse strains generated by CRISPR-Cas9-mediated frameshift deletions within Il24 exon 2. Impairments of wound repair were indistinguishable between two loss-of-Il24-function strains, used interchangeably for experiments. (B) Sagittal sections of day-3 wounds from wild-type (WT) vs. Il24 null mice immunolabeled for p-STAT3. Note that p-STAT3 is still seen in Il24 null wounded epidermis (asterisk). Graphs show quantifications of the percentage of EpdSCs expressing p-STAT3 (upper), and the thickness of keratin 14 (KRT14+) progenitor layers (lower) (n = 5 mice per genotype). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 37 (C) Il20rb RNA-seq of FACS-purified cell populations from homeostatic skin and day-5 wounds (note: immune cells were only from day-5 wounds). TPM, transcripts per kilobase million (n = 5 mice). (D) Sagittal sections of day-5 wounds immunolabeled for KRT14 (epidermis), CD31 (endothelial cells), and labeled with 5-ethynyl-2′-deoxyuridine (EdU) (proliferation). Boxed regions are magnified in insets to better visualize EdU incorporation of S-phase cells (scale bars, 10 μm). Graphs show quantifications of percentage of EdU + cells in epidermis and dermis. For epidermis, quantifications were performed separately for the cells in the migrating zone (to the right of the wound site) and behind the migrating zone (to the left of the wound site) (n = 5 mice per genotype). (E) Left: quantifications of the percentages of migrating epidermis displaying adjacent CD31+ endothelial cells (top) and the percentages of the wound beds at day-5 and −7 post wounding that were repopulated with sprouting blood vessels (CD31+ cells) (middle and bottom). Mouse genotypes are as indicated (see STAR Methods). Top and middle: WT: n = 5, Il24 Het: n = 6, Il24−/−: n = 9 mice, one-way ANOVA, Tukey’s multiple comparisons test; bottom, WT: n = 5, Il20rb−/−:n =6 mice, two-tailed unpaired t test; dots in the graphs indicate data from individual mice. Right: Images of whole-mount immunofluorescence microscopy and 3D image reconstruction performed on day-5 wounds from WT vs. Il24 null mice (scale bars, 50 μm. Immunolabeling was for KRT14 [epidermis] and endomucin [blood vessels]) (n = 3 mice per genotype). (F) Sagittal sections of day-5 wounds immunolabeled for CD31 and PDGFRα (left), or for PDGFRα, collagen-I, and KRT14 (right). Asterisk () denotes a paucity of fibroblasts (PDGFRα+) and their deposition of collagen-I ECM in the dermis of Il24−/− skin. The boxed region magnified in the color-coded insets shows additional Ki67 immunolabeling (Scale bars, 20 μm). Yellow arrows denote Ki67 + proliferating fibroblasts (Ki67+PDGFR+). Quantifications are of fibroblast amount (PDGFRα intensity, upper) and collagen deposition (lower) (n = 5 per genotype). (G) Sagittal sections of day-5 wounds immunolabeled for p-STAT3 and KRT14. Percentage and number/area of p-STAT3+ dermal cells beneath the wound bed are quantified (n = 3 mice per genotype). (H) Left: sleeping beauty system used to generate epidermal-specific Il24 mRNA knockdown mice. Middle top: qRT-PCR of Il24 mRNA in FACS-purified EpdSCs from homeostatic and day-1 wounded skins from control (Ctrl) vs. shIl24 mice (n = 5–6 mice for each genotype). Right: sagittal sections of day-5 wounds from control (Ctrl) vs. shIl24 mice immunolabeled for CD31, KRT14 and labeled with EdU. Percentage of migrating epidermis adjacent to CD31+ capillaries is quantified in middle bottom panel (n = 6 mice per genotype). White dotted lines: epidermal-dermal border; wound site, red dotted line; epidermal migration direction, red arrow. DAPI, nuclei; scale bars except for boxed regions and whole mount: 100 μm. Data in (B)–(H) are presented as mean ± SEM. Dots in the graphs (E) and (H) indicate data from individual mice. Statistical significance was determined using two-tailed unpaired Student’s t tests in (D), (E; bottom panel), (F), (G), and (H); and using one-way ANOVA, Tukey’s multiple comparisons test in (B) and (E; top two panels); * p < 0.0001; * p < 0.001; ** p < 0.01; * p < 0.05; and ns, not significant. See also Figures S2–S5. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 38 Figure 4. Ectopic IL-24 induction in homeostatic skin epithelium elicits a wound-like response without injury (A) Schematic of the generation of TRE-IL-24 mice. Selective targeting to skin EpdSCs was achieved by packaging the transgene in a lentivirus and in utero injection into the amniotic sacs of E9.5 mouse embryos genetic for the Krt14-rtTA doxycycline inducible transcriptional activator. The lentivirus also contained a constitutively expressed Pgk- H2BGFP to monitor integration efficiency. Skins were harvested after mice were fed Dox food for 2, 3, or 4 days. (B) Left: images of mice at postnatal days 1 and 4. Note flaky skin phenotype, evident by day-4. Right: Images of hematoxylin and eosin (H&E) staining and trichrome staining performed on sagittal sections of homeostatic skins from Dox-fed WT and Tre-Il24 mice. Quantifications are of epidermal thickness and intensity of trichrome staining to evaluate dermal collagen deposition (n = 3 mice per genotype). (C) Sagittal sections of homeostatic skins from WT and Tre-Il24 mice immunolabeled for Ki67, GFP and CD31. Quantifications are of percentages of proliferating (Ki67+) EpdSCs (top), and underlying endothelial cells (Ki67+CD31+) (middle) and non-endothelial dermal cells (Ki67+CD31−) (bottom) (n = 3 mice per genotype). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 39 (D) Sagittal sections of homeostatic skins from WT and Tre-Il24 mice immunolabeled for GFP and CD31. Quantifications are of percentage of interfollicular epidermis close to CD31+ endothelial cells (top), and the distance (mm) between epidermis and CD31+ vasculature (bottom) (n = 3 mice per genotype per time point). (E) Sagittal sections of homeostatic skins from WT and Tre-Il24 mice were immunolabeled for GFP and p-STAT3. Prior to collecting skins, mice were given Dox food for 2 days. Quantifications are of percentage of p-STAT3+ epidermal, endothelial, and fibroblast cells. Quantifications of dermal cell types were made by performing similar immunofluorescence as for epidermis, but using antibodies against CD31 and PDGFα, respectively (n = 3 mice per genotype). White dotted lines: epidermal-dermal border. DAPI, nuclei; scale bars, 100 μm. Data in (B)–(E) are presented as mean ± SEM. Experiments were performed R ≥3×. Statistical significance was determined using two-tailed unpaired Student’s t tests; ** p < 0.0001; * p < 0.001; ** p < 0.01; and * p < 0.05. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 40 Figure 5. Tissue-damage-associated hypoxia and HIF1α in wounds are important for robust Il24 expression (A) Sagittal section of day-3 wound harvested just after pimonidazole injection to label tissue hypoxia (n = 5 mice). (B) Sagittal section of day-3 wound immunolabeled for CD31 and HIF1α. The distance (μm) from HIF1αLow vs. HIF1αHigh EpdSCs to the nearest CD31+ blood vessels is quantified (n = 5 mice). (C) Sagittal sections of day-5 wounds from WT and Il24 null mice immunolabeled for CD31 and HIF1α. Boxed regions of the migrating epidermal tongue are magnified at right (scale bars, 20 μm). b, basal EpdSCs; sb, suprabasal epidermal cells (n = 5 mice per genotype). (D) Schematic of the experiment and qRT-PCR of Il24 and Vegfa mRNA in YFP− (Hif1α WT) or YFP+ (Hif1αΔ exon2) FACS-purified EpdSCs from homeostatic skin and from day-1 wounds of Krt14CreER; Hif1αfl/ fl; RosaYFP+/fl mice treated with topical 4OH-Tam (n = 5 mice). White dotted lines: epidermal-dermal border; wound site, red dotted line; epidermal migration direction, red arrow. DAPI, nuclei; scale bars except for the boxed regions: 100 μm. Data in (B) and (D) are presented as mean ± SEM. Experiments were performed R ≥ 3×. Statistical significance was determined using two-tailed unpaired Student’s t tests; **** p < 0.0001; * p < 0.05. See also Figures S6. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 41 Figure 6. Critical roles for both hypoxia/HIF1α and STAT3 in governing robust Il24 expression (A) qRT-PCR of Il24 mRNA in keratinocytes with GFP or IL-24-receptor reconstitution cultured under different oxygen, nutrient, substrate, glycolytic product, and oxidative stress conditions for 48 h. AA, amino acid; Leu, leucine. Note that the native IL-24-receptor, robustly expressed by EpdSCs in their native niche in vivo, is silenced under the culture conditions in vitro. (B) EpdSCs were isolated from skins of Krt14CreER; Hif1αfl/ fl mice, reconstituted with either GFP or IL-24-receptor, and cultured in normoxic (21% O2) or hypoxic (1% O2) conditions. 4OH-Tam was used to replace the endogenous HIF1α with HIF1α lacking the bHLH DNA binding domain (Hif1αΔ exon2). Cells were then immunoblotted for HIF1α, LDHA (lactate dehydrogenase A; encoded by a classical hypoxia-sensitive gene), p-STAT3, STAT3, and vinculin as the loading control. (C) qRT-PCR of Il24 mRNA in the cells described in (B). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 42 (D) Il24 expression from RNA-seq data performed on FACS-purified EpdSCs from homeostatic skin and day-1 wounds from WT and Krt14Cre; Stat3fl/fl (Stat3 cKO)mice treated with 4OH-Tam. TPM, transcripts per kilobase million (n = 3 mice for each genotype). (E) Normalized peaks of RNA-seq, assay for transposase-accessible chromatin sequencing (ATAC-seq), and Cut&Run-seq (with IgG control or antibodies against at HIF1α or STAT3) at the Il24 locus. Red boxes indicate the 5 chromatin regions at the Il24 locus that opened upon wounding (ATAC) and have both HIF1α and STAT3 binding peaks (Cut&Run). Peaks from the same experiments are indicated on the same scale. Data in (A), (C), and (D) are presented as mean ± SEM. Sequencing experiments were in duplicates; others were performed ≥3×. Statistical significance was determined using two-tailed unpaired Student’s t tests; * p < 0.001; and p < 0.01. See also Figure S7. Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 43 Figure 7. IL-24 signaling promotes epithelial glucose uptake and influences dermal repair (A) GLUT1 expression is dependent upon both hypoxia and IL-24-receptor-signaling. qRT- PCR and immunoblot analyses showing that both events are essential for optimal GLUT1 expression. (B) Glucose transport family expression from RNA-seq performed on EpdSCs that were FACS-purified from homeostatic skin (unwd_Epi) and day-5 wound (5d_migrating Epi). TPM, transcripts per kilobase million. (C) Sagittal sections of day-3 wounds from WT vs. Il20rb null skins immunolabeled for GLUT1. Graphs show quantifications of the thickness of GLUT1-expressing epidermis (n = 3 mice per genotype). Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 44 (D) Glut1 expression depends upon STAT3. Left: Slc2a1 mRNA TPM value from RNA- seq of FACS-purified EpdSCs from homeostatic and wounded skin in WT vs. Krt14-Cre; Stat3flfl (Stat3 cKO) mice. Right: sagittal sections of day-3 wounds from WT vs. Krt14- Cre; Stat3flfl;Yfp+/fl (Stat3 cKO) immunolabeled with GLUT1 and YFP (n = 3 mice per genotype). (E) Graphs show relative rates of glucose consumption (left) and lactate production (right) by keratinocytes with GFP or IL-24-receptor reconstitution under normoxic vs. hypoxic conditions. Note that under conditions of hypoxia and IL-24-receptor reconstitution, both measurements are the most elevated. (F) Sagittal sections of day-3 wounds from WT vs. Krt14Cre; Glut1fl/fl mice treated with topical 4OH-Tam. Sections were immunolabeled for GLUT1 and CD31 (left), or for GLUT1 and PDGFRα (right). Asterisk () in the right images denotes a paucity of fibroblasts (PDGFRα+) in the dermis of Glut1 cKO skin. Quantifications at right (n = 6 mice per genotype). (G) Model depicting the similarities between evolutionarily conserved pathogen-induced IFN signaling for defense and injury-induced IL-24 signaling for repair. In contrast to pathogens, which lead to induction of IFN and p-STAT1/2, tissue damage causes hypoxia, leading to HIF1α, IL-24, and p-STAT3. Specifically, EpdSCssense wound hypoxia caused by severed blood vessels, and induce IL-24 and receptor signaling, which subsequently activates STAT3 and further fuels Il24 expression to promote a coordinated dermal repair and re-epithelialization. The autocrine and paracrine mechanisms underlying wound-induced IL-24-signaling in tissue repair are parallel and functionally analogous to pathogen-induced IFN signaling in pathogen defense, and the two pathways share multiple levels of homology. White dotted lines, epidermal-dermal border; wound site, red dotted line; epidermal migration direction, red arrow. DAPI, nuclei; scale bars, 100 μm. Data in and (C)–(F) are presented as mean ± SEM. Experiments were performed ≥3×. Statistical significance was determined using two-way ANOVA and Tukey’s multiple comparisons tests in (A) and (E), using one-way ANOVA and Tukey’s multiple comparisons tests in figure (C), and using two-tailed unpaired Student’s t tests in (F); * p < 0.0001; * p < 0.001; ** p < 0.01; * p < 0.05; and ns, not significant. Cell. Author manuscript; available in PMC 2023 July 05. Liu et al. Page 45 KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Rabbit monoclonal anti-Phospho-STAT3 (Y705) antibody Cell Signaling Technology Cat# 9145S; RRID:AB_2491009 Rabbit monoclonal anti-HIF‐1α antibody Rabbit monoclonal anti-Glucose transporter GLUT1 antibody Rat monoclonal anti-CD31 antibody Cell Signaling Technology Cat# 36169; RRID:AB_2799095 Abcam Biolegend Cat# ab115730; RRID:AB_10903230 Cat# 102502; RRID:AB_312909 Rat monoclonal anti-Endomucin antibody Santa Cruz Biotechnology Cat# sc-65495; RRID:AB_2100037 Chicken polyclonal anti-GFP antibody Rat monoclonal anti-CD140a antibody Rat monoclonal anti-CD49e (Integrin-α5) antibody Chicken polyclonal anti-Keratin 14 antibody Rabbit polyclonal anti-Collagen I antibody Abcam Biolegend Biolegend Biolegend Abcam Cat# ab13970; RRID:AB_300798 Cat# 135909; RRID:AB_2043973 Cat#103801; RRID:AB_31305 Cat# 906004; RRID:AB_2616962 Cat# 21286; RRID:AB_446161 Armenian Hamster monoclonal anti-PECAM-1 antibody Millipore Sigma Cat# MAB1398Z; RRID:AB_94207 Rabbit monoclonal anti-Ki-67 antibody Cell Signaling Technology Cat# 12202; RRID:AB_2620142 Goat polyclonal anti-Arginase 1 antibody Novus Biologicals Cat# NB100-59740; RRID:AB_892299 Polyclonal Goat anti-VEGFA antibody Mouse monoclonal anti-Vinculin antibody Rabbit polyclonal anti-HIF‐1α antibody Mouse monoclonal anti-STAT3 antibody R&D systems Millipore Sigma Cat# AF-493-NA; RRID:AB_354506 Cat# V9131; RRID:AB_477629 Cayman Chemical Cat# 10006421; RRID:AB_409037 Cell Signaling Technology Cat# 9139; RRID:AB_331757 Rabbit polyclonal anti-LDHA antibody Proteintech Group Cat# 21799-1-AP; RRID:AB_10858925 Rat monoclonal PE anti-Integrin α6 antibody Rat monoclonal anti-CD49f (Integrin α6) antibody Rat monoclonal eFluor660 anti-CD34 antibody Rat monoclonal BV421 anti-CD34 antibody Rat monoclonal PerCP/Cy5.5 anti-Sca-1 antibody Rat monoclonal APC/Cy7 anti-CD45 antibody Rat monoclonal PE/Cy7 anti-CD31 antibody Rat monoclonal Biotin anti-CD117 antibody Rat monoclonal BV421 anti-CD140a antibody Rat monoclonal APC anti-CD140a antibody eBioscience BD Pharmingen eBioscience BD Biosciences Biolgened Biolegend Biolegend Biolegend Biolegend Biolegend Cat# 12-0495-82; RRID:AB_891474 Cat# 555734; RRID:AB_2296273 Cat# 50-0341-82; RRID:AB_10596826 Cat# 562608; AB_11154576 Cat# 108124; RRID:AB_893615 Cat# 103116; RRID:AB_312981 Cat# 102524; RRID:AB_2572182 Cat# 105804; RRID:AB_313213 Cat# 135923; RRID:AB_2814036 Cat# 135907; RRID:AB_2043969 Streptavidin PE-Cy7 conjugate eBioscience Cat# 25-4317-82; RRID:AB_10116480 Rat monoclonal BV421 anti-CD90.2 antibody Rat monoclonal BV421 anti-CD11b antibody Rat monoclonal BV421 anti-I-A/I-E (MHCII) antibody Rat monoclonal AF488 anti-CD49e (Integrin α5) antibody Rat monoclonal APC anti-CD49e (Integrin α5) antibody Rat monoclonal PE anti-Ly-6G antibody Rat monoclonal APC anti-Ly-6G antibody Rat monoclonal PE/Cy7 anti-CD117 (c-Kit) antibody Rat monoclonal FITC anti-Ly-6C antibody Biolegend Biolegend Biolegend Biolegend Biolegend Biolegend Biolegend Biolegend Biolegend Cat# 140327; RRID:AB_2686992 Cat# 101235; RRID:AB_10897942 Cat# 107621; RRID:AB_493726 Cat# 103810; RRID:AB_528839 Cat# 103813; RRID:AB_2750076 Cat# 127607; RRID:AB_1186104 Cat# 127613; RRID:AB_1877163 Cat# 105813; RRID:AB_313222 Cat# 128006; RRID:AB_1186135 Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 46 REAGENT or RESOURCE Rat monoclonal APC anti-Siglec F antibody Armenian Hamster monoclonal PE/Cy7 anti-FcεRIα antibody Mouse monoclonal BV605 anti-CD64 antibody Rat monoclonal anti-CD16/CD32 antibody Donkey polyclonal AF488 anti-Rabbit IgG antibody Donkey polyclonal AF488 anti-Chicken IgG antibody Donkey polyclonal AF488 anti-Rat IgG antibody Goat polyclonal AF488 anti-Armenian hamster IgG antibody Donkey polyclonal AF546 anti-Rabbit IgG antibody Donkey polyclonal RRX anti-Rat IgG antibody Donkey polyclonal AF647 anti-Rabbit IgG antibody Donkey polyclonal AF647 anti-Rat IgG antibody Donkey polyclonal HRP anti-Rabbit IgG antibody Donkey polyclonal HRP anti-Mouse IgG antibody Bacterial and virus strains SOURCE Biolegend eBioscience Biolegend eBioscience Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories IDENTIFIER Cat# 155508; RRID:AB_2750237 Cat# 25-5898-82; RRID:AB_2573493 Cat# 139323; RRID:AB_2629778 Cat# 14-0161-85; RRID:AB_467134 Cat# 711-545-152; RRID:AB_2313584 Cat# 703-545-155; RRID:AB_2340375 Cat# 712-545-150; RRID:AB_2340683 Cat# 127-545-099; RRID:AB_2338996 Cat# 711-165-152; RRID:AB_2307443 Cat# 712-295-150; RRID:AB_2340675 Cat# 711-605-152; RRID:AB_2492288 Cat# 712-605-150; RRID:AB_2340693 Cat# 711-035-152; RRID:AB_10015282 Cat# 715-035-150; RRID:AB_2340770 NEB® Stable Competent E. coli (High Efficiency) New England Biolabs Cat# C3040H Chemicals, peptides, and recombinant proteins Normal Donkey Serum Normal Goat serum Jackson ImmunoResearch Laboratories Jackson ImmunoResearch Laboratories Cat# 017-000-121; RRID:AB_2337258 Cat# 005-000-121; RRID:AB_2336990 ProLong™ Diamond Antifade Mountant with DAPI Thermo Fisher Scientific Cat# P36962 Trypsin-EDTA (0.25%), phenol red Liberase TL Research Grade ACK lysing buffer T4 DNA ligase reaction buffer Nuclease-free water RNaseOUT NxGen phi29 DNA Polymerase Gibco Sigma-Aldrich Cat# 25200056 Cat# 5401020001 Thermo Fisher Scientific Cat# A1049201 New England Biolabs Invitrogen Invitrogen Lucigen Cat# B0202S Cat# AM9937 Cat# 10-777-019 Cat# 30221-1-LU Power SYBR™ Green PCR Master Mix Thermo Fisher Scientific Cat# 4367659 5-Ethynyl-2’-deoxyuridine (EdU) RIPA Lysis and Extraction Buffer cOmplete™ Protease Inhibitor Cocktail PhosSTOP™ Millipore Sigma Cat# 900584 Thermo Fisher Scientific Cat# 89901 Millipore Sigma Millipore Sigma Cat# 11836145001 Cat# 4906837001 Cat# 20034197 Illumina Tagment DNA Enzyme and Buffer Small Kit Illumina NuPAGE™ LDS Sample Buffer (4X) Thermo Fisher Scientific Cat# NP0008 4’6’-diamidino-2-phenylindole (DAPI) Millipore Sigma Cat# 28718-90-3 Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 47 REAGENT or RESOURCE Blasticidin Alt-R® CRISPR-Cas9 tracrRNA, 20 nmol Alt-R™ S.p. Cas9 Nuclease V3, 100 μg TRI Reagent Complete mouse endothelial cell medium kit Hypoxyprobe Kit (100 mg pimonidazole HCl plus 1.0 ml of 4.3.11.3 mouse MAb) SOURCE InvivoGen IDT IDT Millipore Sigma Cell Biologics Hypoxyprobe IDENTIFIER Cat# ant-bl-05 Cat# 1072533 Cat# 1081058 Cat# T3934 Cat# M1168 Cat# HP1-100Kit Precision Plus Protein™ Dual Color Standards Biorad Cat# 1610374EDU NuPAGE™ 4 to 12%, Bis-Tris, 1.0-1.5 mm Thermo Fisher Scientific Cat# NPG321BOX NuPAGE™ MOPS SDS Running Buffer (20X) Thermo Fisher Scientific Cat# NPGGG1 NuPAGE™ Transfer Buffer (20X) 1x Tris Buffered Saline (TBS) Thermo Fisher Scientific Cat# NPGGG61 Biorad Cat# 161G782 Pierce™ ECL Plus Western Blotting Substrate Thermo Fisher Scientific Cat# 32132 DNase 1 from bovine pancreas Human Plasma Fibronectin Purified Protein Corning Collagen I, Rat Tail Poly-L-lysine Millipore Sigma Millipore Sigma Corning Millipore Sigma Cat# D4263 Cat# FCG1G Cat# 354236 Cat# P47G7 DreamTaq Green PCR Master Mix (2X) Thermo Fisher Scientific Cat# K1G82 R&D Systems Cat# 78G7-ML-G1G/CF Recombinant Mouse IL-24 (NS0-expressed) Protein (Carrier- free) Recombinant Mouse IL-17A Protein (Carrier Free) Doxycycline hydrochloride (Z)-4-Hydroxytamoxifen Critical commercial assays RNase-Free DNase Set TruSeq RNA Library Preparation Kit R&D Systems Millipore Sigma Millipore Sigma Qiagen Illumina NEBNext Ultra II DNA Library Prep kit for Illumina New England BioLabs NEBNext® Multiplex Oligos for Illumina® (Index Primers Set 1) New England BioLabs Cat# 421-ML-G1G/CF Cat# D3447 Cat# H79G4 Cat# 79254 Cat# RS-122-2GG1 Cat# E7645L Cat# E7335S NEBNext® Multiplex Oligos for Illumina® (Index Primers Set 2) New England BioLabs Cat# E75GGS Agencourt AMPure XP beads Direct-zol RNA Microprep Direct-zol RNA Miniprep SuperScript™ VILO™ cDNA Synthesis Kit Power SYBR™ Green PCR Master Mix Pierce™ BCA Protein Assay Kit Beckman Coulter Zymo Research Zymo Research ThermoFisher ThermoFisher ThermoFisher A6388G Cat# R2G62 Cat# R2G5G Cat# 11754G5G Cat# 4368577 Cat# 23225 LIVE/DEAD Fixable Blue Dead Cell Stain Kit Thermo Fisher Scientific Cat# L231G5 Click-iT™ EdU Cell Proliferation Kit for Imaging, Alexa Fluor™ 647 dye Click-iT™ EdU Cell Proliferation Kit for Imaging, Alexa Fluor™ 594 dye Deposited data Thermo Fisher Scientific Cat#C1G34G Thermo Fisher Scientific Cat# C1G339 Raw bulk RNA-sequencing data This paper GEO: PRJNA7313G4 Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 48 REAGENT or RESOURCE Raw 10x single cell RNA-sequencing data Raw ATAC-sequencing and Cut-and-Run sequencing data Experimental models: Cell lines SOURCE This paper This paper IDENTIFIER GEO: PRJNA885G18 GEO: PRJNA731164 C57BL/6 mouse primary dermal microvascular endothelial cells Cell Biologics Cat# C57-6G64 293TN Producer Cell Line Primary mouse fibroblasts Primary mouse keratinocytes J2 fibroblast feeder cells Experimental models: Organisms/strains Mouse: C57BL/6J Mouse: Rosa26-stop-lox-stop YFP: B6.129X1- Gt(ROSA)26Sortm1(EYFP)Cos/J Mouse: Il24−/− Mouse: Il2Grb−/− Mouse: Hif1αfl/fl: B6.129-Hif1atm3Rsjo/J Mouse: K14CreER: K14-CreER-Rosa26-YFP Mouse: Glut1fl/fl: Slc2a1tm1.1Stma/AbelJ Mouse: Stat3fl/fl: B6.129S1-Stat3tm1Xyfu/J System Biosciences Cat# LV9GGA-1 Fuchs Lab Fuchs Lab Fuchs Lab N/A N/A N/A The Jackson Laboratory The Jackson Laboratory Cat# GGG664; RRID:IMSR_JAX:GGG664 Cat# GG6148; RRID:IMSR_JAX:GG6148 Fuchs Lab Genentech N/A N/A The Jackson Laboratory Cat# 7561; RRID:IMSR_JAX:GG7561 Fuchs Lab N/A The Jackson Laboratory Cat# 31871; RRID:IMSR_JAX:031871 The Jackson Laboratory Cat# 16923; RRID:IMSR_JAX:016923 Mouse: Myd88−/−: B6.129P2(SJL)-Myd88tm1.1Defr/J The Jackson Laboratory Cat# 9088; RRID:IMSR_JAX:009088 Mouse: Trif−/−: C57BL/6J-Ticam1Lps2/J The Jackson Laboratory Cat# 005037; RRID:IMSR_JAX:005037 Mouse: Rag2−/−Il2rg−/−: C57BL/6NTac.Cg-Rag2tm1Fwa Il2rgtm1Wjl Mouse: C57BL/6NTac Mouse: TNFR1/TNFR2 DKO: B6.129S-Tnfrsf1btm1Imx Tnfrsf1atm1Imx/J Mouse: Krt14-rtTA Mouse: Krt14-rtTA; sleeping beauty shIl24 Oligonucleotides Taconic Taconic Cat# 4111-F Cat# B6-F The Jackson Laboratory Cat# 003243; RRID:IMSR_JAX:003243 Fuchs Lab Fuchs Lab N/A N/A N/A Quantitative real-time PCR primers (see Table S4) Eurofins Genomics Recombinant DNA pTY-EF1A-puroR-2a pTY-EF1A-HygromycinR-2a lentiCRISPRv2 blast pT4/HB pCMV(CAT)T7-SB100 pMD2.G psPAX2 Software and algorithms Liu et al.62 Liu et al.62 Stringer et al.67 Wang et al.68 Má té s et al.69 A gift from Didier Trono A gift from Didier Trono A gift from Zhijian Chen A gift from Zhijian Chen Addgene plasmid #98293; RRID:Addgene_98293 Addgene plasmid #108352; RRID:Addgene_108352 Addgene plasmid #34879; RRID:Addgene_34879 Addgene plasmid #12259; RRID:Addgene_12259 Addgene plasmid #12260; RRID:Addgene_12260 Cell. Author manuscript; available in PMC 2023 July 05. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Liu et al. Page 49 REAGENT or RESOURCE SOURCE IDENTIFIER Prism ImageJ FlowJo Adobe Photoshop Adobe Illustrator CS5 R https://www.graphpad.com/ scientific-software/prism/ N/A Schneider et al.70 https://imagej.nih.gov/ij/ https://www.flowjo.com Adobe.com Adobe.com N/A N/A N/A R Development Core Team71 http://www.r-project.org/ TxDb.Mmusculus.UCSC.mm10.knownGene (R package) Team BC and Maintainer BP72 Salmon (version 1.4.0) tximport (R package) Bsgenome.Mmusculus.UCSC.mm10 (version 1.4.0) (R package) Patro et al.73 Soneson et al.74 Team TBD75 https://bioconductor.org/packages/ release/data/annotation/html/ TxDb.Mmusculus.UCSC.mm10.known Gene.html https://github.com/COMBINE-lab/ salmon/releases https://bioconductor.org/packages/ release/bioc/html/tximport.html https://bioconductor.org/packages/ release/data/annotation/html/ BSgenome.Mmusculus.UCSC.mm10.ht ml MACS2 software in BAMPE mode Zhang et al.76 https://pypi.org/project/MACS2/ Langmead and Salzberg77 https://sourceforge.net/projects/bowtie- bio/files/bowtie2/2.2.9/ http://www.java.com N/A Bowtie2 (version 2.2.9) Java (version 2.3.0) SAM tools (version 1.3.1) deeptools (version 3.1.2) Integrative Genomics Viewer (IGV) software HOMER (version 4.10) BD FACSDiva software Zen software Other Li et al.78 Ramıŕez et al.79 Robinson et al.80 Heinz et al.81 BD Biosciences Carl Zeiss Axio Observer Z1 epifluorescence microscope BioTek Cytation 5 Cell Imaging Multimode Reader 2100 Bioanalyzer Instrument Carl Zeiss Agilent Agilent 7900HT Fast Real-Time PCR system Applied Biosystems BD FACSAria Cell Sorter BD LSRII Analyzer BD LSRFortessa Analyzer ChemiDoc Imager 2900 Biochemistry Analyzer Sterile 4 mm biopsy punch Sterile 6 mm biopsy punch BD Bioscience BD Bioscience BD Bioscience Bio-Rad YSI Integra Miltex Integra Miltex Cell. Author manuscript; available in PMC 2023 July 05. https://sourceforge.net/projects/ samtools/files/samtools/1.3.1/ https://pypi.org/project/deepTools/ https://software.broadinstitute.org/ software/igv/ http://homer.ucsd.edu/homer/ N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A Cat# 33-34 SH Cat# 33-36 SH A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1073_pnas.2221415120	RESEARCH ARTICLE NEUROSCIENCE Reward expectations direct learning and drive operant matching in Drosophila Adithya E. Rajagopalana,b ID , Ran Darshana,c, Karen L. Hibbarda, James E. Fitzgeralda, and Glenn C. Turnera,1 ID Edited by Leslie C. Griﬀith, Brandeis University, Waltham, MA; received January 3, 2023; accepted August 11, 2023 by Editorial Board Member Michael Rosbash Foraging animals must use decision-making strategies that dynamically adapt to the changing availability of rewards in the environment. A wide diversity of animals do this by distributing their choices in proportion to the rewards received from each option, Herrnstein’s operant matching law. Theoretical work suggests an elegant mechanistic explanation for this ubiquitous behavior, as operant matching follows automatically from simple synaptic plasticity rules acting within behaviorally relevant neural circuits. However, no past work has mapped operant matching onto plasticity mechanisms in the brain, leaving the biological relevance of the theory unclear. Here, we discovered operant matching in Drosophila and showed that it requires synaptic plasticity that acts in the mushroom body and incorporates the expectation of reward. We began by developing a dynamic foraging paradigm to measure choices from individual ﬂies as they learn to associate odor cues with probabilistic rewards. We then built a model of the ﬂy mushroom body to explain each ﬂy’s sequential choice behavior using a family of biologically realistic synaptic plasticity rules. As predicted by past theoretical work, we found that synaptic plasticity rules could explain ﬂy matching behavior by incorporating stimulus expectations, reward expectations, or both. However, by optogenetically bypassing the representation of reward expectation, we abolished matching behavior and showed that the plasticity rule must speciﬁcally incorporate reward expectations. Altogether, these results reveal the ﬁrst synapse-level mechanisms of operant matching and provide compelling evidence for the role of reward expectation signals in the ﬂy brain. dopamine \| learning-rules \| decision-making \| mushroom body \| foraging An animal’s survival depends on its ability to adaptively forage between multiple potentially rewarding options (1, 2). To guide these foraging decisions appropriately, animals learn associations between options and rewards (3–5). Learning these associations in natural environments is complicated by the uncertainty of rewards. Both vertebrates and invertebrates employ decision-making strategies that account for this uncertainty (6–14). A commonly observed strategy across the animal kingdom is to divide choices between options in proportion to the rewards received from each (7–15). It has been hypothesized that animals that use this operant matching strategy make use of the expectation of reward—the recency-weighted rolling average over past rewards—to learn option–reward associations (14, 15). Many studies further posit that this learning involves synaptic plasticity (16–18), and theoretical work has identiﬁed a characteristic relationship between operant matching and a speciﬁc form of expectation-based plasticity rule that incorporates the covariance between reward and neural activity (19). Despite this strong link between plasticity rules and the matching strategy, there has been no mapping of these rules onto particular synapses or plasticity mechanisms in any animal. As a result, deeply investigating these theories by manipulating and testing the nature of plasticity rules underlying operant matching has been intractable. The fruit ﬂy, Drosophila melanogaster, offers a promising system within which to address these challenges. Over the last half century, researchers have shown that ﬂies can learn a wide variety of Pavlovian associations between cues and rewards (20–26). With the help of advances in functional and anatomical tools (27–32), they have identiﬁed the mushroom body (MB) as the neural substrate for these learning processes, including the assignment of value to sensory cues, and the underlying plasticity mechanisms have been extensively characterized (33–42). Recent theoretical work has also attempted to formalize the fea- tures of the learning rule that is mediated by these plasticity mechanisms (43–47). Despite this progress, evidence has been mixed as to whether this learning rule makes use of reward expectations (48–52), and there is a dearth of understanding about how ﬂies learn in natural environments (but see ref. 53). Studying foraging behaviors would allow us Signiﬁcance Unraveling how humans and other animals learn to make adaptive decisions is a unifying aim of neuroscience, economics, and psychology. In 1961, Richard Herrnstein formulated a long-standing empirical law that quantitatively describes many decision-making paradigms across these fields. Herrnstein’s matching law states that choices between options are divided in proportion to the rewards received, a strategy that equalizes the return on investment across options. Identifying mechanistic principles that could explain this universal behavior is of great theoretical interest. Here, we show that Drosophila obey Herrnstein’s matching law, and we pinpoint a plasticity rule involving the computation of reward expectations that could mechanistically explain the behavior. Our study thus provides a powerful example of how fundamental biological mechanisms can drive sophisticated economic decisions. J.E.F., and G.C.T. Author contributions: A.E.R., R.D., designed research; A.E.R. performed the experiments and simulations; K.L.H. contributed new reagents/ analytic tools; A.E.R., R.D., J.E.F., and G.C.T. analyzed data and interpreted models; and A.E.R., R.D., K.L.H., J.E.F., and G.C.T. wrote the paper. The authors declare no competing interest. This article is a PNAS Direct Submission. L.C.G. is a guest editor invited by the Editorial Board. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2221415120/-/DCSupplemental. Published September 21, 2023. PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 1 of 12 OPENACCESSto not only clarify these gaps in the understanding of ﬂy learning but could also provide an insightful framework for testing the neural computations underlying decision-making strategies such as matching. Leveraging this foraging framework in ﬂies requires us to address several open questions. First, animals in real foraging scenarios have to be able to form associations between multiple different options and rewards, yet evidence in ﬂies suggests that some associations are labile and easily overwritten (24). Second, choice behavior has rarely been investigated at the individual ﬂy level (53–56), and never in the context of ﬂies making repeated choices between two probabilistically rewarding options. It is therefore unclear whether ﬂies can learn associations between options and probabilistic rewards. Finally, it is unknown whether they can integrate probabilistic reward events over multiple past experiences to form analog expectations. Even if such analog expectations can be formed, it is unclear whether they lead to matching behavior through covariance-based plasticity in the ﬂy brain. To answer these questions, we designed an olfactory two- alternative forced choice (2AFC) task for individual Drosophila, inspired by earlier behavior assays for ﬂies (57–61) and foraging related 2AFC tasks in vertebrates (10–12, 14). The assay allows us to measure hundreds of sequential choices from individual ﬂies as we vary the probability of reward associated with different odor cues. Such measurement of behavior over time allows us to distinguish between different foraging strategies, such as the matching law versus a simpler win-stay, lose-switch strategy. Importantly, our assay breaks the hard dichotomy between Pavlovian and operant conditioning. Unlike purely Pavlovian tasks (20, 24), ﬂies in our task do not passively experience olfactory cues and rewards. Rather, the choices made by the ﬂy dictate the odors and rewards experienced, a hallmark of operant- learning tasks. However, unlike purely operant tasks, where animals learn that speciﬁc actions lead to rewards or punishment (62, 63), ﬂies in our task have to learn to perform stimulus- dependent actions. This relationship between stimulus, action, and reward is very similar to the dynamic foraging tasks where operant matching has been observed in other species (11–14). The dynamic foraging task structure thereby allows us to readily translate past theoretical work into the context of the Drosophila brain to seek a mechanistic understanding of decision-making behaviors that could apply across animals. Results Flies Learn Multiple Probabilistic Cue–Reward Associations. In our Y-arena, a single ﬂy begins a trial in an arm ﬁlled with clean air and can choose between two odor cues that are randomly assigned to the other two arms (Fig. 1A, Materials and Methods and SI Appendix, Information 1). The ﬂy can freely move between arms, with a choice deﬁned as the ﬂy crossing into the reward zone at the end of the arm (Fig. 1A). Once a choice is made, we provide reward by optogenetically activating sugar-sensing neurons using a Gr64f driver (64, 65). The Y-arena then resets, with the arm chosen by the ﬂy ﬁlled with clean air and the other two arms randomly ﬁlled with the two odors. This task design permits us to precisely control reward delivery without satiating the ﬂy and enables us to monitor the choices of a single ﬂy over hundreds of trials. We ﬁrst established that ﬂies learn effectively in this apparatus by reliably rewarding ﬂies only when they chose one of the odors—what we term a 100:0 protocol. Each ﬂy ﬁrst experienced the two odors (3-octanol; OCT and 4-methylcyclohexanol; MCH) unrewarded for a block of 60 trials, and then reward Fig. 1. Flies learn multiple probabilistic cue–reward associations. (A) Sc- hematic of Y-arena (Top). Air flows from tips of each arm to an outlet in the center. Reward zones are demarcated by lines. A choice is registered when a fly crosses into the reward zone of an odorized arm, triggering Gr64f sugar sensory neuron optogenetic activation with a 500-ms pulse of red light. The next trial commences as the chosen arm switches to air and the two odors (green/orange) are randomly reassigned to the other two arms (Bottom). (B) Cumulative choices made toward each option are shown (n = 9 flies, mean & individual flies). No rewards were available for the first 60 trials (Naive—black) and became available for the green option from the 61st trial onward (Training—red). Inset: Percentage rewarded choices in naive and training blocks. Flies prefer the rewarded option in the training block compared to naive (Wilcoxon signed-rank test: P = 0.0039, n = 9). (C) Example trajectory of a fly in the Y before (Left) and after (Right) green odor is paired with reward. Distance in air arm is represented as negative values (black), while distances in odorized arms are represented as positive values (green/orange). Choices are represented by colored rasters. At choice points the arena resets and that arm switches to air, so the fly’s position jumps to the tip of the air arm. (D) The probability of accept decisions are plotted as a function of time in the 100:0 protocol (n = 9 flies, mean—solid line, SE—shaded area). Flies show a high probability of accepting the rewarded odor (Left). The probability of accepting the unrewarded odor drops over time (Right). (E) Controls (Left) show higher percentage of choice made toward the rewarded option than DopR1 K.O. (Right) flies in one 100:0 block of 60 trials (mean ± SE − red point & line; individual fly scores—black; Mann–Whitney rank-sum test: P = 0.0022, control: n = 7, DopR1−/−: n = 6). (F ) Schematic describes the reward structure of the task (Top). Cumulative rewarded and unrewarded choices plotted against each other, for three diﬀerent protocols 100:0, 80:0, 40:0 (Bottom). Slope of all curves indicates that flies show a preference for the rewarded odor in all cases compared to a naive preference indicated by the black line (Mann–Whitney rank-sum test: 100:0, P = 4.4500 × 10−8, n = 18; 80:0, P = 5.8927 × 10−5, n = 10; 40:0, P = 0.0014, n = 10). (G) Schematic of the protocol for training flies with two simultaneous probabilistic cue–reward contingencies (Top). Two diﬀerent odor choices are alternated throughout an unrewarded naive block and a reward block where options were rewarded with baiting probability of 0.4 or 0.8. Performance (percentage of choices in which the potentially rewarding option was chosen) on the low and high reward choices (Bottom) indicates that flies learn both associations. An increased preference for the rewarded odors over unrewarded is observed (compared to naive preference) (Mann–Whitney rank-sum test: P = 2.3059 × 10−4 for high rewarding odor; n = 10, P = 0.01 for low rewarding odor, n = 10). 2 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org Trial 2ExhaustOdor 2Odor 1AirRewardZoneOptogenetic Sugar RewardTrial 1AC Air Arm Naive0 30Time (min)Time (min)TrainingG/O Arms 0 5 -5 Displacement (cm)D0 300t0t00.51.0 P(Accept)Rewarded odor Unrewarded odor Odor Experience NumberOdor Experience Number% Rewarded Choice1000Naive Training B5cmELowReward HighReward% Rewarded Choice0000GNaive% Rewarded Choice100500Control DopR1-/- 400800Training**020406080100120020406080100120Cumulative Rewarded ChoicesCumulative Unrewarded ChoicesF0 20 400 40 X = 100X = 80X = 4020 0 XCumulative Rewarded ChoicesCumulative Unrewarded Choices100500delivery was activated for the following block of 60 trials. As observed previously, although individual ﬂies exhibited different odor biases in this naive phase (54, 55, 66), those biases averaged out over the population (Fig. 1 B, inset). In this phase, ﬂies spent a lot of time in the air arm and made variable choices, with little preference for either odor (Fig. 1 C, Left, example ﬂy). Once reward was made available, ﬂies rapidly shifted to choosing the rewarded odor (Fig. 1B). This was accompanied by a faster interval between choices (SI Appendix, Fig. S1B) and a decrease in meandering trajectories (Fig. 1C). To analyze this choice behavior at a more elemental level, we adopted the common framework of considering foraging choices as a series of accept–reject decisions, where the animal decides whether or not to pursue an option (67). We deﬁned reject decisions as when a ﬂy enters an odorized arm but turns around and exits the arm before reaching the reward zone, while accept decisions reﬂect cases where the ﬂy reaches the reward zone (Materials and Methods). Associating options with rewards changed the probability of accept decisions gradually over the course of a block. Acceptance probability increased for the rewarded odor and decreased for the unrewarded odor (Fig. 1D and SI Appendix, Fig. S1E). On average, ﬂies were around four times more likely to reject the unrewarded odor and seven times more likely to accept the rewarded odor (SI Appendix, Fig. S1D). Interestingly, ﬂies tended to reject odors quite close to the tip of the arm (SI Appendix, Fig. S1F ), suggesting that ﬂies might accumulate evidence over time to make and commit to their decision—an aspect of ﬂy behavior that has previously been studied (68). These results indicate that ﬂy choice behavior in this task can be thought of as a series of accept–reject decisions. We found that the odor–reward associations learnt by ﬂies in our assay were MB dependent. Learning-related plasticity in the MB circuit requires the activity of dopaminergic neurons (DANs) (24, 34, 37–41). Dopamine is sensed by odor-representing Kenyon cells (KCs) and induces synaptic plasticity between these KCs and downstream mushroom body output neurons (MBONs) (37, 39). To interfere with this plasticity, we used a tissue-speciﬁc CRISPR knock-out strategy (69) to knock out DopR1 receptors selectively in the KCs (Materials and Methods), which are necessary for ﬂies to associate odors with rewards in other paradigms (41). These ﬂies showed no detectable learning in the 100:0 protocol, compared to control animals (Fig. 1E). These ﬁndings establish that odor–reward associations in our behavioral assay are mediated by MB plasticity. is thought We then asked whether ﬂies could link odor cues with probabilistic rewards and distinguish between different reward probabilities, a key aspect of natural foraging. Importantly, we incorporated reward baiting into our probabilistic reward tasks (12, 14). This means that rewards probabilistically become available and then persist until the reward is collected (Materials and Methods). Baiting is commonly used in mammalian 2AFC tasks, as it to mimic the natural processes of resource depletion and replenishment over time. We began with experiments in which a single odor was rewarded with a range of baiting probabilities: 1 (100:0 task), 0.8 (80:0 task), or 0.4 (40:0 task). Flies showed a preference toward the rewarded odor in all cases compared to a naive lack of preference indicated by the black line (Fig. 1F ). The extent of the preference varied with the probability of reward—a higher probability of reward led to a stronger preference. Interestingly, ﬂies made faster choices when rewards were more probable (SI Appendix, Fig. S1C). These results show that ﬂies can learn from probabilistic rewards but do not determine whether they can store two associations simultaneously—another necessity for foraging. To test this, we designed a paradigm with a third odor, pentyl acetate (PA), included. This served as the unrewarded cue while we tested memory formation with the other two odors (Fig. 1 G, Top). Flies ﬁrst made 80 unrewarded choices consisting of 40 choices between OCT and PA and 40 choices between MCH and PA. In the next 80 (Training) trials, one of OCT or MCH was assigned a high reward baiting probability (0.8) and the other a low probability (0.4). We alternated the training trials for the two different odors, to ensure that both relationships would be learnt simultaneously (Materials and Methods). After pairing, ﬂies preferred both rewarded odors over PA compared to their naive preference (Fig. 1 G, Bottom). This choice preference was also reﬂected in their accept/reject behavior, with ﬂies exhibiting a clear preference for accepting the high-rewarding odor (SI Appendix, Fig. S1 G, Right). Interestingly, in trials with the low-reward cue presented, there was an increased probability of rejecting both rewarded and unrewarded odors, as compared to naive trials (SI Appendix, Fig. S1 G, Left). This suggests the possibility that ﬂies keep track of all the odor options potentially available in the environment and actually increase their rejection rate in the absence of the high-reward odor. Overall, these experiments establish the ﬂy as a capable animal model for studying foraging behaviors. Individual ﬂies in the Y-arena can learn multiple odor-reward associations and can do so in the face of probabilistic reward. Importantly, these relationships are mediated by synaptic plasticity at the KC- MBON synapses in the MB. This establishes a foundation to test how these animals perform in dynamic foraging tasks and assess how they respond to reward baiting probabilities that change over time. Flies Follow Herrnstein’s Operant Matching Law. Foraging tasks are cognitively complex, involving two cues paired with different baiting probabilities that change with time. This requires animals to keep track of choice and reward history and form expectations to make adaptive choices. We designed our own dynamic foraging protocol to investigate how ﬂies behave in such a scenario. The protocol consisted of three consecutive blocks of 80 trials each. Flies made choices between two odors (OCT & MCH) that were paired with different baiting probabilities (Materials and Methods). These probabilities remained ﬁxed within a block and changed across blocks (Fig. 2A, example). We found that ﬂies exhibit operant matching behavior, similar to observations in monkeys, mice, and honeybees (8, 11, 12, 14). Individual ﬂies exhibited a strong correlation between choice ratio (deﬁned as the ratio between the number of choices made toward option A and option B) and reward ratio (deﬁned as the ratio between the number of rewards received upon choosing option A and option B), either calculated over entire blocks or over a short (ten-trial) window to capture short-term dynamics (Fig. 2 A and B—example ﬂy, SI Appendix, Fig. S2—all 18 ﬂies, and Materials and Methods). This holds true across ﬂies, as seen in the relationship between block-averaged reward ratios and their choice ratios (Fig. 2C). In such a plot, the matching law predicts that all points will fall along a line with slope equal to one (the unity line). Flies appear to approximately follow the matching law with a slight amount of undermatching, signiﬁed by a slope less than one. Undermatching is commonly observed across species (11–15), and several reasons have been suggested for this tendency (13, 19) (Discussion). Past work has suggested that animals form expectations of reward and use this to guide behavior in such dynamic foraging tasks (13–15, 19). When rewards are delivered probabilistically, PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 3 of 12 Fig. 2. Flies follow Herrnstein’s operant matching law (A) Matching of instantaneous choice ratio (blue) and reward ratio (black) in an example fly. Schematics indicate the reward baiting probabili- ties for each odor in the three 80-trial blocks (Top). Individual odor choices are denoted by rasters, tall rasters—rewarded choices, short rasters— unrewarded choices. Curves show 10-trial aver- aged choice and reward ratio, and horizontal lines the corresponding averages over the 80- trial blocks. A description of how rewards are determined on any given trial in this task can be found in Materials and Methods. (B) Cumulative choices of the same fly. The slope of the black lines indicates the block-averaged reward ratio in each block; the blue line indicates the cumulative choices with slope representing choice ratio. The parallel slopes of the two lines indicate matching. (C) Block-averaged choice ratio is approximately equal to reward ratio, following the matching law with some undermatching (n = 54 blocks from n = 18 flies). (D) A “win–stay; lose–switch” model does not accurately capture the trial-by- trial staying and switching probabilities of flies. A 2 × 2 probability table indicating the joint probability of the action predicted by the model and the action made by the fly (n = 18 flies and Materials and Methods). (E) Change in instantaneous choice ratio around block changes (n = 16 transitions with large changes in baiting probabilities between blocks). (F ) Analysis of choices following particular histories of experience. Choices made by flies over three consecutive past trials are represented by boxes of diﬀerent colors representing odors chosen. Filled boxed indicate rewarded choices. Probabilities of choosing the green and orange odor on the current trial (0) conditional on this history are illustrated with associated values. (G) Coeﬀicients from logistic regression performed on fly choice behavior to determine the influence of 15 past rewards (Top) and choices (Bottom) on a fly’s present choice (blue). These coeﬀi- cients were compared to coeﬀicients predicted on shuﬀled data (gray) (Wilcoxon signed-rank test: P < 0.001, P < 0.01, P < 0.05, n = 18 flies). (H) Model fit quality (percentage deviance explained) for 15-trial, 7-trial, and 1-trial logistic regression models. Null model used to calculate the quality metric is a logistic regression with 0-trial history and only bias (Wilcoxon signed- rank test comparing the null model prediction with each test model prediction; shown here as test model prediction subtracted by null model: *P < 0.001, n = 18 flies). (I) 15-trial logistic regression fit (purple) on behavior (blue) from the example fly from panel A, plotted from the 15th trial onward to avoid edge eﬀects. (J) Exponential timescales for each fly shown in SI Appendix, Fig. S2, estimated from fitting the leaky integrator model (Materials and Methods). animals can only derive an expectation of reward by tallying information over multiple trials. However, such tallying could reﬂect a computation beyond the capabilities of ﬂies. We wanted to explicitly address the alternative hypothesis that ﬂies follow a simple win–stay/lose–switch strategy (Materials and Methods), which would suggest that their behavior is dictated by only the most recent reward/omission experience. Simulating choice sequences using this learning rule produced output that somewhat resembled that of the ﬂy (example in SI Appendix, Fig. S3A). However, it poorly captured the stay/switch probabilities actually observed in ﬂy behavior data (Fig. 2D). In particular, switching occurred much more frequently than predicted. As further evidence that multiple past outcomes affected behavior, choices of an individual ﬂy at block transitions showed a lag between the choice ratio curve and the updated reward ratio at transition points (Fig. 2B), suggesting that the ﬂy takes a few trials to adjust its behavior. Quantifying this across multiple transitions for all ﬂies in the task showed ﬂies require 15 to 20 trials to reach a new steady state choice behavior following block switches (Fig. 2E). It is possible that this lag could arise from averaging across multiple ﬂies that switch at different trials after the transition. This could occur even if ﬂies use just one past trial’s worth of information to learn about the change in reward, consistent with observations in larvae (56). To qualitatively illustrate that ﬂies learn using multiple trials worth of past information, we ﬁrst looked at the decisions made by ﬂies following example triplets of choices and outcomes (Fig. 2F ), inspired by recent work in mice (6). For example, following three unrewarded choices of one par- ticular odor, ﬂies’ next choice was roughly random (Fig. 2 F, Top Left). However, when an odor was rewarded on the most recent trial or more distant trials, choices were biased toward that option (Fig. 2 F, Middle and Bottom Left). In another comparison, ﬂies’ tendency to switch back to an earlier choice (i.e., choose the green 4 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org 100:050:500:100Choice Ratio11891189Choice RatioReward Ration = 18 flies100:050:500:10050:50100:0% Deviance Explained30015-trial Model0 80 160 240Trial Number8020100:050:500:100oitaReciohCFly Behavior0 80 160 240Trial Number60120070140Cumulative G ChoicesCumulative O Choices-150154575Trials Since Block SwitchChoice Ratio100:050:500:100n = 16 transitionsABCEHI01515 trial7 trial1 trialF4654316938623235654357683232#Trials InThe PastDStaySwitchStaySwitch0.20.30.4Fly BehaviorModel Prediction1010************************************00.1051015PastTrialRewardsCoefficient ValueChoices G00.08***051015ExponentialTimescaleJRewardedUnrewardedChoice RatioReward RatioytilibaborPodor after an unrewarded choice of the orange odor) increased if that odor was rewarded in the recent past (Fig. 2 F, Right). To measure the relationship between current choice and past outcomes more systematically, we used logistic regression to determine how a ﬂy’s decisions depended on choice and reward history. Like other animals (6, 12), ﬂies showed a small amount of habitualness by choosing options that had been recently chosen more often; regression coefﬁcients for a short history of recent choices were signiﬁcantly positive compared to coefﬁcients ﬁt to shufﬂed data (Fig. 2 G, Bottom). This approach also showed that the reward history was important for predicting choice, with many recent rewards weighted signiﬁcantly (Fig. 2 G, Top). We compared regression models that predicted behavior based on different lengths of outcome histories (15, 7, and 1 trial) and found that the percentage of deviance explained over a null model with a 0-trial history was greater for models that used longer outcome histories (Fig. 2H and Materials and Methods). An example ﬁt from a regression model with a 15-trial history is shown in Fig. 2I. In alignment with the results of the regression model (Fig. 2G), we found that when ﬁtting a leaky integrator model(14), which assigns value to options using exponentially weighted reward histories (SI Appendix, Fig. S3 B–E), to the behavior of individual ﬂies, an exponential timescale of 7 trials on average best-predicted behavior (Fig. 2J ). Together, these results show that ﬂies’ choices follow operant matching, with choices depending on the history of many past choices and outcomes. Covariance-Based Learning Is Required for Matching Behavior in a Model of the MB. Theoretical work has placed strong, testable constraints on the nature of learning rules that could underlie operant matching. An elegant theory put forward by Loewenstein and Seung (19) proves that operant matching is the inevitable outcome of plasticity rules that modify synaptic weights according to the covariation of neural activity signaling reward and sensory input (Materials and Methods). Mathematically, covariance is the averaged product of two variables with at least one being subtracted by its mean. The mean is simply the average reward and/or sensory input the animal experiences—an average that can also be expressed as the animal’s expectation. Comparing the current value to its expectation ensures that weights can be adjusted up or down. Importantly, only an animal that follows operant matching would receive rewards at a rate equal to the reward expectation for both options, which leads weights to stabilize. Loewenstein and Seung mathematically formalized this intuitive link between expectation and matching and showed that when synaptic plasticity is the basis for operant matching, a covariance-based plasticity rule can account for matching. They used a simple neural circuit model to illustrate their theory, with two different sensory inputs controlling different motor outputs and a decision determined by a winner-take- all interaction between those outputs (SI Appendix, Fig. S4A). Interestingly, the structure of this model maps nicely onto the circuitry of the ﬂy MB (Fig. 3 A, Left). Sensory inputs are represented by the KCs, each odor activating a sparse subset of the KC population (70–72). KCs synapse onto MBONs, which guide motor output by signaling the valence of an odor, i.e., its attractive/repulsive quality, rather than a speciﬁc action (22, 38, 42). KC-MBON synapses are modiﬁed by a plasticity rule that depends on the coincident activity of odor- representing KCs and release of dopamine by reward-signaling DANs (24, 34, 37–41, 43) (Fig. 3 A, Center, box). Current evidence indicates that postsynaptic activity of the MBON does not play a role in the plasticity (73), so only the sensory and reward activities need to be considered. Either or both of these terms could incorporate an expectation resulting in a covariance-based rule (Fig. 3 A, Center, box). DANs could incorporate reward expectation [(R - E(R))] by subtracting a running average of reward activity (E(R)) from the current reward-related activity (R). Similarly, KCs could incorporate sensory expectation [(Si - E(Si))] by calculating an average sensory experience, possibly by a mechanism that involves metaplasticity and synaptic eligibility traces. To fully adapt the theoretical framework of Loewenstein and Seung to the biological network in the MB, we had to make a few changes (Fig. 3 A, Right and Materials and Methods). Fig. 3. Covariance-based learning is required for matching behavior in a model of the MB (A) Left: Schematic representing the MB with all relevant neurons shown in diﬀerent colors (key). Center: Box containing candidate reward-dependent synaptic plasticity rules at the KC-MBON synapse. Right: Schematic of our MB model developed by adapting Loewenstein and Seung’s model to more closely resemble the MB and the features of our olfactory task. In the modified task, agents only experience one odor at a time. Reward information is provided to this circuit via DAN activity which either represents simply reward (R) or reward minus reward expectation (R-E(R)). Weights between inputs and MBON are modified according to plasticity rules shown in Center, where 𝜂 < 0 to match the fly’s depression-based learning rule. MBON output determines probability of rejecting an odor and is passed through a sigmoidal nonlinearity to determine action. (B) Left: Block-averaged choice ratio produced by the [Si-E(Si)] · [R-E(R)] covariance-based rule (box) plotted against reward ratio. The model exhibits matching behavior (slope is 1). Right: An example simulation showing the performance in a 3-block task of a model incorporating a covariance-based rule [Si-E(Si)] · [R-E(R)]. Task reward contingencies are the same as shown for the example fly in Fig. 2A. (C) Same as (B), but simulated with a noncovariance learning rule. Left: The model produces behavior that does not show matching (slope < 1). Right: Performance in a 3 block task does not show matching of choice and reward ratio. PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 5 of 12 DANMBONW2W1Odor 1Mushroom Body ModelProbabilityof rejectBC100:00:100100:0Choice RatioReward Ratio100:00:100100:0Choice RatioReward Ratio100:050:500:100Choice Ratio0 80 160 240118911898020Mushroom Body CircuitAMB Model:∆Wi = η∙ [Si-E(Si)] ∙ [R-E(R)]100:050:500:100Choice Ratio0 80 160 240MB Model:∆Wi = η ∙ [Si] ∙ [R]RewardedUnrewardedChoice RatioReward RatioTrial Number118911898020Behavioral Output ∆Wi = η∙ KCi∙ DANKCi = Si or [Si-E(Si)]DAN = R or [R-Ε(R)]Candidate Fly Plasticity RulesOdor Input - Kenyon Cell (KC)Dopamineric Neuron (DAN)MB Output Neuron (MBON)Reward Input -Gr64f NeuronTrial Number50:5050:5050:5050:50KCsActionRewardGr64fFirst, odors are represented by noisy populations of KCs (70– 72). We thus parameterized input representations in the model to incorporate noise and overlap of KC subsets between options. Second, in our task, ﬂies only experience one odor at a time, so only one set of KCs is active during reward delivery. Although Loewenstein and Seung’s original theory does not account for this possibility in its proof, we extended it to this case (Materials and Methods). Third, plasticity between MBONs and KCs is modiﬁed by a synaptic depression rule (37, 38). We thus ﬂipped the sign of the synaptic weight update rule. Finally, MBON activity determines whether ﬂies accept or reject an odor rather than a winner-take-all decision mechanism (22, 38) (Fig. 3A). We incorporated this into our model by having MBON activity encode the probability of rejecting an odor, with higher activity representing a greater probability to reject. This MBON activity was then passed through a sigmoidal nonlinearity to determine behavioral output. We then evaluated whether these changes affect the rela- tionship between covariance-based rules and matching. We used this MB-aligned model to simulate behavior arising from covariance rules that incorporated stimulus–expectation, reward– expectation, or both (Fig. 3 and SI Appendix, Fig. S5). Consistent with the theory, all three covariance-based rules gave rise to a choice-ratio versus reward-ratio relationship that followed the matching law (Fig. 3 B, Left and SI Appendix, Fig. S5 A–C, Left). In contrast, a rule that did not incorporate either reward or stimulus expectation did not follow the matching law and instead yielded a ﬂat slope (Fig. 3 C, Left and SI Appendix, Fig. S5 D, Left). For comparison, we also examined the behavior produced by the original model in a distinctly different task and observed similar results (SI Appendix, Fig. S4 A–E). Note that in the Loewenstein and Seung task, both options are always present when reward is delivered, which leads to a slope in between ﬂat and unity when a noncovariance rule is used (SI Appendix, Fig. S4 E, Left). However, if only one option is present when an animal is rewarded, as in the ﬂy task, synapses saturate and a noncovariance rule leads to a ﬂat choice-ratio versus reward-ratio relationship (Fig. 3 C, Left). To get a more reﬁned view of model performance, we examined the trial-by-trial behavior each plasticity rule generates. Models that incorporate covariance-based plasticity rules nicely replicate the trial-by-trial behavior of ﬂies, tracking changes in the reward contingencies across blocks, with the resulting instantaneous choice ratio biased toward the more rewarded option in each block (Fig. 3 B, Right and SI Appendix, Figs. S4 B–D, Right and 5 A–C, Right). On the other hand, both the MB-inspired model and Loewenstein and Seung’s model do not capture trial- by-trial behavior well when a noncovariance rule is incorporated, with choices made roughly equally to both options throughout (Fig. 3 C, Right and SI Appendix, Figs. S4 E, Right and S5 D, Right). This reﬂects the fact that when value updates only depend on sensory input and reward, plasticity is unidirectional. Consequently, synapses representing the two options will both be driven to low levels, although at slightly different rates, so that ultimately both options are chosen roughly equally. Overall, these results show that a model constrained by the network architecture of the MB more closely reproduces ﬂy behavior when it operates according to a covariance-based plasticity rule. Identifying Learning Rules Underlying Dynamic Foraging in the Mushroom Body. To test whether our theoretical prediction of a covariance-based rule is supported by the observed behavior, we developed an approach that estimated the form of the plasticity rule being used in the ﬂy MB. Our goal was to break the plasticity rule into components that span a space of possible rules and use optimization approaches to predict trial-by-trial behavior of each individual ﬂy to assign coefﬁcients to each of these components. In this way, we would identify the form of the plasticity rule that best explained observed behavior and be able to conclude whether this rule was a covariance-based rule. We used the structure of the MB-inspired generative model (Fig. 3A) to build a predictive model and test how it ﬁts the accept/rejection decisions made by the ﬂy on each odor encounter. However, rather than utilizing a predeﬁned plasticity rule, the predictive model used a rule composed of four terms that were candidate components of the MB learning rule (Fig. 4A). We then used logistic regression to assess which of these terms contributed the most when ﬁtting ﬂy behavioral data (Materials and Methods). The four terms were a constant term, a KC term reﬂecting sensory input, a DAN term representing reward, and ﬁnally, the product of KC and DAN activity. By deﬁnition, this product term becomes a covariance calculation when either of its elements are subtracted by their mean values, i.e., when either reward and/or sensory expectation are incorporated (Fig. 3 A, Center box). We considered four model variants, a noncovariance one that lacked any expectation term and three different covariance-based rules where either KC or DAN or both were subtracted by their expectation. At every iteration of the logistic regression, the model prediction was compared to experimentally observed ﬂy behavior, and regression coefﬁcients were updated. Once the ﬁt was optimized, we evaluated which term contributed the most to the ﬁt by examining the weights of each coefﬁcient. Before applying this approach to ﬂy data, we validated it by determining whether it correctly identiﬁed the relevant term when tested with choice sequences that were simulated using a covariance-based learning rule that only incorporated reward expectation. Indeed, the ﬁt quality was clearly better with a model that incorporated reward expectation (SI Appendix, Fig. S6 A and B). Moreover, the largest weights were correctly assigned to the KC-DAN product term, the term that calculates the covariance between these two elements (SI Appendix, Fig. S6C). Additionally, our simulations suggested that the extent of matching and the accuracy of learning rule ﬁts were largely unaffected by either the degree of overlap in KC activity patterns or the timescale over which rewards were integrated (SI Appendix, Fig. S6 D–I ). Consequently, for simplicity, we then used overlap of zero and an exponential timescale of 3.5 trials in all future analyses. We then applied our approach directly to the behavioral data from individual ﬂies performing the dynamic foraging task. A representative example showing ﬂy behavior and model predictions can be seen in Fig. 4B. This example suggests that models with covariance-based rules may better resemble the ﬂies’ behavior. To quantitatively compare ﬁt quality of the different models, we calculated the percentage deviance explained for every individual ﬂy. This metric showed that regressions that utilized rules with sensory expectation, reward expectation, or both were objectively better ﬁts for ﬂy behavior (Fig. 4C and SI Appendix, Fig. S7A). Overall, we found that learning rules that incorporated either sensory or reward expectation both yielded better ﬁts to ﬂy behavior than noncovariance rules. To distinguish between these different expectation-based learning rules, we examined which regression coefﬁcients had the biggest weights. When we ﬁt a rule with only reward expectation, the regression assigned the 6 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org Identifying learning rules underlying dy- Fig. 4. (A) namic foraging in the mushroom body. Schematic detailing the logic of the MB-inspired regression model. This model was used to pre- dict the behavior of and learning rules used by each individual fly that experienced the task described in Fig. 2. (B) Example fly data (blue) showing the probability of accepting odor 1 (Top) and odor 2 (Bottom) calculated over a 6- trial window as a function of the number of times the fly experienced the given odor. These data were fit using an MB-inspired regression model (A) that incorporates either a covariance- based rule with sensory and reward expectations (brown), just reward expectation (gray), or a noncovariance rule (red). (C) Change in percentage deviance explained, computed by subtracting the percent- age deviance explained of the noncovariance- based model from a covariance-based rule that incorporates reward expectation (n = 18 flies). On average, fly behavior was better predicted by the covariance-based model (Wilcoxon signed- rank test: P = 0.0018). Individual flies that were better fit by the covariance-based model have a positive value on this plot (gray region), while flies better fit by the noncovariance-based model have a negative value (red region). (D) Regres- sion coeﬀicients assigned to each term of the plasticity rule when the MB-inspired regression model using a covariance-based rule with reward expectation was fit to the flies’ behavior. As in (C), the model was fit to each fly resulting in 18 diﬀerent values for the coeﬀicients. The largest coeﬀicients were observed to have been assigned to the product term. (E) Change in percentage deviance explained (shown in C), plotted against a measure of undermatching (mean square error between instantaneous choice ratio and reward ratio lines) for each fly (n = 18). The best fit line of the scatter, calculated by a linear regression is shown in orange. (F ) Coeﬀicient value assigned to the product term (shown in D), plotted against a measure of undermatching for each fly (n = 18). The best fit line of the scatter, calculated by a linear regression is shown in orange. just sensory expectations (black), KC-DAN product, i.e., the covariance term, with the largest weight (Fig. 4D and SI Appendix, Fig. S7B). On the other hand, ﬁtting using either of the two covariance rules that incorporated sensory expectations yielded large coefﬁcients for the noncovariance terms containing either KC or DAN activity alone (SI Appendix, Fig. S7B). We observed a similar result when we ﬁt simulated data from an agent using a reward expectation– based learning rule (SI Appendix, Fig. S7C). Nevertheless, when the behavior was simulated using the same sensory expectation rule, the covariance term was given the most weight (SI Appendix, Fig. S7E). These results suggest that ﬂies use a covariance rule based on reward expectations to guide their behavior. in some ﬂies, Interestingly, we found that the simple expectation-free noncovariance rule was a better ﬁt. One possible explanation for this result is that these ﬂies showed operant matching to a lesser extent. We thus quantiﬁed matching by calculating the mean squared error between instantaneous choice and reward ratios and found that different strengths of matching across ﬂies were correlated with how well an expectation-free plasticity rule ﬁt the behavior data (Materials and Methods). Flies that were better ﬁt by the expectation-free rule tended to show more undermatching, in line with our predictions (Fig. 4E). Consistent with this, the weight of the covariance term coefﬁcient was greater in ﬂies that exhibited stronger matching behavior (Fig. 4F ). To examine whether some ﬂies were better ﬁt by a noncovariance rule because our approach might inaccurately assign weights to a combination of correlated terms in the learning rule, we examined the correlations between pairs of coefﬁcients. However, we found no consistent statistical relationship (SI Appendix, Fig. S7D and Materials and Methods). Overall, this general approach allowed us to estimate the learning rule the ﬂy uses directly from behavioral data, providing clear evidence that a reward-expectation-based covariance rule is important in the MB. Behavioral Evidence of Reward Expectation in DANs. We next wanted to experimentally verify that a reward-expectation-based covariance rule in particular guided learning and choice behavior in the ﬂy MB. The mathematical differences between the three different covariance rules suggested a way forward (Materials and Methods). In particular, the rules differ in which terms— sensory input or reward—incorporate an expectation. Thus, to distinguish between the possible different covariance-based rules in the MB, we designed an experiment to manipulate the calculation of reward expectation using genetic tools that override the natural activity of the DANs. Speciﬁcally, we provided reward via optogenetic activation of the reward- related protocerebral anterior medial (PAM) DANs. This would bypass any upstream computation of reward expectation and simply provide a consistent reward signal on every trial. Such a manipulation would change the learning rule from a covariance- based rule to a noncovariance rule if the following conditions were met: i) the animal’s learning rule depended on the product of DAN and KC activities; ii) DAN activity incorporated reward PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 7 of 12 BP(Accept)EA036∆ % Deviance ExplainedCcov-basednoncov-based-301200-3036∆ % DevUndermatching IndexF0120000.30.6‘d’ CoefficientUndermatching IndexPredicted Behavior Error∆Wi = -a - b ∙ KCi - c ∙ DAN - d ∙ KCi ∙ DAN010242# Odor 1 Experience 010265# Odor 2 Experience Fly DataModel Fit: [Si-E(Si)] ∙ R Model Fit: [Si-E(Si)] ∙ [R-E(R)]Model Fit: Si∙ RModel Fit: Si∙ [R-E(R)]-0.8-0.400.40.8BiasabcdDCoefficient ValueDANMBONW2W1KCsObserved Behavior Fly ModelUpdate Coefficients ofModel Plasticity RuleP(Accept)expectation; and iii) KC activity did not incorporate sensory expectation. This would in turn result in modiﬁed behavior. For this test, we initially focused on a task consisting of two blocks (naive and training) of 60 trials each, with a reward ratio of either 100:0 (one odor has a baiting probability of 100% and the other is never rewarded) or 80:20 (one odor has a baiting probability of 80% and the other 20%) in the second block (Materials and Methods). We ﬁrst predicted how behavior in these protocols would differ between covariance-based and noncovariance rules using simulations. As expected, covariance-based models learnt to choose the more rewarded option more often, with choice ratios reﬂecting reward ratios (Fig. 5A and SI Appendix, Fig. S8 A and B). The behavior of the model with any covariance-based rule was similar to the ﬂy behavior when it was rewarded using the sugar neurons (Fig. 5 B and C). On the other hand, noncovariance rules led to preferences saturated around 75% in 100:0 and 50% with the 80:20 reward ratio (Fig. 5D). These theoretically predicted preferences very closely match our observations of ﬂy behavior in the DAN activation experiments (Fig. 5 E and F ). We observed low plateau performance in both tasks (Fig. 5 E and F ), with values strikingly similar to that predicted by the noncovariance rule (Fig. 5D). One potential concern with these experiments is that differ- ences in the efﬁcacy of optogenetic activation of the DANs deep in the central brain versus the peripherally located Gr64f neurons could contribute to these behavioral differences. However, when ﬂies were instead made to choose between reward-associated or unrewarded odors in a circular arena previously used to assess learning in ﬂies (24), we found that both PAM DAN and Gr64f sugar neuron activation drove similar learning (SI Appendix, Fig. S9 A–D). Since the LED intensity in the circular arena (2.3 mW/cm2) was closely matched to that in the Y-arena (1.9 mW/cm2), differences in optogenetic efﬁcacy cannot explain the range of behavioral patterns seen in the circular and Y arenas. All data are consistent with the interpretation that PAM activation bypasses the computation of reward expectation and converts a covariance rule into a noncovariance rule. In particular, learning via the noncovariance plasticity rule only modiﬁes weights from Kenyon cells that respond to the rewarded odor, which increases the acceptance probability of the rewarded odor without changing behavior to the unrewarded odor. According to this model, performance saturates in the Y-arena because the ﬂy repeatedly encounters the unrewarded odor by chance, and their initial tendencies for accepting the odor option never change; performance does not saturate in the circular arena because a ﬂy Fig. 5. Behavioral evidence of reward expec- tation in DANs. (A) Instantaneous choice ratio over trial number, for a simulated agent using a covariance-based rule with reward expectation in 80:20 (orange) and 100:0 (red) tasks. (B) As (A), except it shows fly behavior when provid- ing sugar sensory optogenetic reward instead of simulation (100:0, n = 8 flies; 80:20, n = 6). (C) Average choice ratios of individual flies from (B) showing significant learning in both 100:0 and 80:20 protocols (Wilcoxon signed-rank test: 100:0, P = 0.0039; 80:20, P = 0.0312). (D) As (A), except for an agent using a noncovariance rule. (E) As (B), except reward provided via the PAM DANs using R58E02-Gal4 to drive UAS- CSChrimson (n = 8 flies in both 80:20 and 100:0). Dashed line in (D) and (E) indicates the max- imum possible performance of agent in D in the 100:0 protocol. (F ) Average choice ratios of individual flies from (E). Flies showed a significant preference toward the rewarded odor in 100:0 but not 80:20 (100:0, P = 0.0391; 80:20, P = 0.1875). (G) The instantaneous choice ratio of an example fly receiving DAN optogenetic reward performing the dynamic foraging protocol plot- ted against trial number as in Fig. 2A. (H) Block- averaged choice ratios against reward ratios for flies with DAN reward (n = 26 flies, 3 blocks each). Best fit lines : red—DAN reward, blue—Gr64f sugar sensory reward (Fig. 3C). (I) Block-averaged choice ratios against reward ratios (n = 50) from data simulated using a noncovariance- based rule. Best fit lines : black—simulated data, red—DAN reward. (J) Instantaneous choice ratio around block changes. Flies trained with Gr64f (K ) activation in blue, DAN activation in red. As (J) but with simulated agents using either a covariance-based rule in blue or noncovari- ance rule in red. (L) Example fly data showing probability of accepting odors against experience number (blue) with DANs activated as reward. Fit using a model (Fig. 4) that incorporates ei- ther a covariance-based rule (gray) or a non- (M) Change in percent- covariance rule (red). age deviance explained, computed by subtract- ing the percentage deviance explained of the noncovariance-based model from a covariance- based rule, plotted for each fly (n = 26). On average, fly behavior was better predicted by the noncovariance-based model (Wilcoxon signed-rank test: P = 0.0164). 8 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org -4-202P(Accept)01Odor Experience Number350∆ % Deviance ExplainedFly η∙ Si∙ Rη∙ Si∙ [R-E(R)]Trial Number060060060060ACLM Choice Ratio100:080:20Trial NumberBChoice RatioReward Ratio100:050:500:10050:50100:0HG100:050:500:100Choice RatioTrial Number0 80 160 240336733678020F100:080:20Choice RatioReward Ratio100:050:500:10050:50100:0INaiveNaiveTrainingTraining***100 : 080 : 20DEcov-basednoncov-basedPAM DANGr64fNon-cov model0100:050:500:100Choice RatioFly Behavior:Gr64f activationMB Model:∆Wi = η∙ Si∙ [R-E(R)]Fly Behavior:PAM DAN activationMB Model:∆Wi = η∙ Si∙ R100:050:500:100Choice Ratio100:050:500:100Choice Ratio100:050:500:100Choice Ratio100:050:500:100 Choice Ratio100:050:500:100MB Model:∆Wi = η∙ Si∙ R-150154575Trials Since Block SwitchChoice Ratio100:050:500:100J-150154575Trials Since Block SwitchChoice Ratio100:050:500:100KFly Behavior:PAM DAN activationNaiveNaiveTrainingTraining30303030n.sFly Behavior:PAM DAN activationPAM DANGr64fR ∙ S[R-E(R)] ∙ SPAM DANthat has learned to accept the rewarded odor will stop exploring and cease to encounter the unrewarded option. We next examined how bypassing reward expectation affects matching behavior. When tested with the same three-block matching design as earlier, but now providing a consistent reward signal via direct DAN stimulation, ﬂies exhibited strongly diminished matching behavior (Fig. 5 G and H ). The slope of the choice-ratio versus reward-ratio plot was lower than that observed with Gr64f-driven reward and approached the ﬂat line predicted by simulations of behavior with a noncovariance based learning rule (Fig. 5I ). The instantaneous choice ratio and reward ratio of an example ﬂy (Fig. 5G) suggested that this ﬂattening arises because choice ratios are never strongly biased to either odor. This is again explained by the unidirectional noncovariance rule. In agreement with this, changes in choice ratio at block transitions were much ﬂatter with DAN reward than with Gr64f, as expected by the differences between the covariance-based and noncovari- ance models (Fig. 5 J and K ). To quantitatively evaluate whether providing reward with DAN activation changed the learning rule from covariance-based to a noncovariance rule, we ﬁt our MB- inspired regression models (Fig. 4A) to ﬂy data produced with DAN reward. We found that the noncovariance rule was the better ﬁt (Fig. 5 L and M ). We ﬁnd through these experiments that bypassing the computation of reward expectation changes ﬂy choices from resembling behavior produced by a covariance- based learning rule to behavior expected from a noncovariance rule. In particular, the results suggest that this covariance-based rule is located in the ﬂy MB and incorporates reward expectation but not sensory expectation. Altogether, our results support the theory that covariance- based learning rules that incorporate reward expectation underlie operant matching in ﬂies. It suggests that a reward expectation signal is calculated in the DANs of the ﬂy MB and provides the ﬁrst mapping of learning rules underlying operant matching onto plasticity mechanisms at speciﬁc synapses. Discussion The foraging strategies used by animals play a key role in their survival. Operant matching is one simple and ubiqui- tous behavioral strategy, utilized in dynamically changing and probabilistic environments. Despite the ubiquity of this strategy and its strong theoretical foundation, little is known about the underlying biological mechanisms. We leveraged the growing body of knowledge regarding learning in the fruit ﬂy, and the plethora of available anatomical tools, to identify these mechanisms. We developed a foraging task that allowed us to monitor choices of individual fruit ﬂies and showed, for the ﬁrst time, that ﬂies follow Herrnstein’s operant matching law. Combining experimental results with computational modeling, we found that this behavior requires synaptic plasticity and uses a rule that incorporates expectation of reward via the rewarding PAM DANs. Our results provide the ﬁrst mapping of the learning rule underlying operant matching onto the plasticity of speciﬁc synapses—the KC-MBON synapses in the MB. Does the Ubiquity of Operant Matching Imply a Common Mechanistic Framework? When choosing between options that predict reward with different probabilities, mammals, birds, and insects all follow Herrnstein’s matching law (8, 9, 11–15). This is clear at the global, trial-averaged level, where choice ratios are roughly equal to reward ratios, but is also true at the trial-by-trial level (Fig. 2A). In fact, we found that individual choices made by ﬂies depended on choice and reward information received over multiple past trials (Fig. 2 G and H ). This is in agreement with what has been observed in mice and monkeys (12, 15) and suggests that these animals all make use of similar kinds of information to guide their behavior. Flies also show an increase in speed of choice when rewarded, another common signature of learnt behavior in mice and monkeys (11, 12) (SI Appendix, Fig. S1 B and C). It is unclear whether these behavioral similarities result from underlying mechanisms that are shared across species. At its surface, mechanistic similarities seem likely. For example, neural signals that subtract reward expectation from reward—a key com- ponent of the plasticity rules underlying matching shown here— can be found in the form of a reward prediction error in many different animals (74, 75). Nevertheless, such a signal on its own is not sufﬁcient to produce matching; it needs to be incorporated into a covariance-based plasticity rule in a behaviorally relevant circuit. On the other hand, while learning values of options via synaptic plasticity is the traditional mechanistic framework thought to underlie such foraging decisions (16, 17), recent work has found signatures of graded neural responses proportional to value during inter-trial-intervals, suggesting a persistent-activity- based mechanism for foraging decisions that may not require synaptic plasticity (12, 76). Associated modeling efforts suggest matching can arise from models that don’t incorporate synaptic learning (19, 77). While both synaptic plasticity and nonplasticity mechanisms can explain the observed behaviors, each makes different testable assumptions about the underlying neural architecture (18) and the effect of changing environmental conditions on the behavior. For example, if one eliminated reward baiting in our experiment, a circuit using a covariance-based plasticity rule would still give rise to behavior that follows Herrnstein’s matching law. In this case, following such a law would lead the animal to always choose the option with higher reward probability. On the other hand, if matching behavior was produced using a different mechanism, the lack of reward-baiting might give rise to different strategies, such as the probability matching strategy commonly observed in mice under these conditions (6). Experiments to identify which mechanisms are used by different brains, and theoretical work to understand why, would therefore provide important insight into circuit function and the neural basis of operant matching. Beyond Covariance-Based Synaptic Plasticity. Our behavioral evidence suggests that synaptic plasticity in the mushroom body depends on reward expectations through a simple covariance- based plasticity rule. We identiﬁed this plasticity rule by the process of elimination. First, we narrowed our focus to the three minimal covariance-based plasticity rules inspired by the architecture of the MB. Importantly, Loewenstein and Seung showed that these rules produce matching. We then showed that only one of the three rules also explains the results of the DAN- activation experiment. It’s important to recognize that more complex plasticity rules may be consistent with our data and necessary to explain future mechanistic and behavioral data. For instance, the plasticity rule could be augmented by adding any term that averages to zero in the matching task. The plasticity rule could also be changed to involve a nonlinear function of the current synaptic weight, presynaptic KC activity, and postsynaptic MBON activity. The fundamental requirement of Loewenstein and Seung’s theory is merely that the plasticity rule ultimately drives the covariance between neural activity and reward to zero. Loewenstein and Seung’s theory provides an impressively general link between operant matching and covariance-based PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 9 of 12 plasticity, but it does make several assumptions that may be violated in the ﬂy. For instance, the theory assumes that plasticity only occurs when the animal makes a choice, with weights ﬁxed between decisions (Materials and Methods). In our current paradigm, this means that no plasticity occurs when the ﬂy rejects an odor or otherwise explores and navigates through its environment. In contrast, DANs encode a variety of motor variables and are not locked to choice or reward (39, 78). These motor-related DAN signals would presumably modify synaptic connections in the MB, and such off-task plasticity could generate important variability in synaptic weights and choice behavior. Interestingly, recent work has also found that these same DANs do not have a consistent effect on action- reward learning in a purely operant task (63). This suggests that motor-related DAN signals are not the substrate for operant learning, and MB plasticity may speciﬁcally act to link sensory cues to rewarding actions. Further, the theory assumes that neural activity and reward are conditionally independent given choice. The MB represents reward via DAN activity, so this assumption could be violated if KC and DAN activity have correlated variability across trials that is not related to choice. Such correlations are feasible given indirect connections from KCs to DANs and the complexity of DAN activity (31, 32, 78). Finally, the theory pertains to tasks where the animal decides between two options. Some animals have also been found to exhibit operant matching behavior when choosing between three or more options (8, 79). In this setting, operant matching still implies that the covariance between neural activity and reward vanishes, so there is hope that covariance-based plasticity rules would generate matching. However, other behavioral strategies can also lead to vanishing covariance (SI Appendix). It would be interesting to investigate whether modiﬁed learning rules can more reliably produce matching in naturalistic foraging scenarios or multioption choice tasks. Plasticity in Multiple MB Compartments Could Explain Devi- ations from Matching. One complication to the framework of expectation-based learning rules and matching is that ﬂies, like several other animals, don’t perfectly follow the matching law; rather they undermatch. Two hypotheses have been proposed to account for this deviation. The ﬁrst proposes that animals that undermatch make use of a learning rule that deviates from a strictly covariance-based rule (19). Synaptic saturation and representation of motor variables in DAN response, as discussed in the previous section, offer particularly simple possibilities. Another important possibility for how this could occur is to have plasticity at multiple sites contributing to the overall learning, with different plasticity rules at each site. Indeed, the MB is divided into multiple compartments that contribute to behavior but exhibit important differences in learning (22, 24). It is possible that some compartments make use of reward expectation in a covariance-based learning rule, while others do not. Alternatively, undermatching can also result if reward expectations are estimated over long timescales (13), even if all compartments made use of a covariance-based rule. This idea suggests that in a dynamic environment where baiting probabilities change quickly, the memory of past experiences acts as a bias that prevents the animal from correctly estimating the present cue–reward relationships. This is possible in the MB, as different compartments form and decay over different time scales (24). Whether either or both of these hypotheses explain undermatching in ﬂies can be studied in future experiments by manipulating different compartments of the MB circuitry and analyzing the effect of such a manipulation on undermatching. Relatedly, it would be interesting to check whether animals could adapt the timescales used to estimate reward expectations to the dynamics of the behavior task. An Approach for Inferring Learning Rules from Behavior. Here, we introduced a statistical method that uses logistic regression to infer learning rules from behavioral data. While we speciﬁcally applied our approach to infer learning rules for the ﬂy mushroom body, the inference of learning rules is of importance to many areas of neuroscience (80–82). In fact, this method could be similarly applied to model other learnt behaviors in the ﬂy and other animals. In the current work, we considered learning rules that only depended on the current sensory stimulus (KC response) and reward (DAN response), but our methodology would also generalize to the inference of learning rules that incorporated a longer time-scale history of sensory input and reward. For example, the framework would be able to estimate rules that incorporated the weighted average of recent sensory experience. However, it’s important to realize that the logistic regression formalization would break down entirely for learning rules that depend on the magnitudes of synaptic weights or postsynaptic activity. Such terms would induce different nonlinear dependen- cies between the choice sequence and learning rule parameters, preventing us from converting these choice and reward histories into regression inputs related to each component of the learning rule (Materials and Methods). Our approach was appropriate here because the plasticity rule in the mushroom body is not thought to involve these terms. However, many biological learning rules do depend on postsynaptic activity and current synaptic weights, and future work should explore more ﬂexible methodologies from modern machine learning to develop generally applicable approaches (82). Circuit Mechanisms for Matching and Reward Expectation in Drosophila. We have shown that operant matching is mediated by synaptic plasticity in the ﬂy mushroom body and involves the calculation of a reward expectation. However, the mechanisms underlying this calculation remain unclear. The proposed mechanism underlying the calculation of reward prediction error (RPE) in mammals provides a hint at one option (74). Here, dopaminergic neurons implicitly represent expectation by calculating the difference between the received reward and the reward expectation. This has been found to involve the summation of positive “reward” inputs and negative GABA-ergic “expected reward” inputs to the dopaminergic neurons (83). MB DANs could represent reward expectation in a similar way. In fact, the recently released hemibrain connectome (32) has found many direct and indirect feedback connections from MBONs to DANs that theoretical work has shown could support such a computation (43, 47). In the MB circuit, MBON activity is related to the expectation of reward associated with a given odor (22, 38, 42). An inhibitory feedback loop, via GABA- ergic interneuron(s) for example, could potentially carry reward expectation–related information from MBONs to DANs. The negative expected reward signal from this interneuron could be combined in the DANs with a positive reward signal from sensory neurons, allowing DAN activity to represent the type of reward expectation signal needed by a covariance-based rule. It is important to note that such a mechanism would have a major difference from mammalian RPEs. Since MBON activity is linked to the presence of odor, the reward expectation signal would vary across stimuli and only be present when the stimulus was too. Thus, this signal would not have the temporal features 10 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org of mammalian RPEs. This difference in temporal structure of the reward expectation signal could explain the mixed observations from past studies aimed at identifying reward expectation in ﬂies. For instance, a study that used temporally distinct cues and reinforcements suggested that DANs do not incorporate reward expectation (49), while studies that used temporally overlapping cues and reinforcements did ﬁnd signatures of reward expectation (48, 52), albeit with different temporal properties than the typical mammalian RPE. It’s also possible that reward expectations are incorporated into mushroom body plasticity by adjusting the levels of reward and punishment needed to achieve a given dopamine signal. In this scheme, reward-related dopamine neurons could represent how much a reward exceeds expectations, and punishment- related dopamine neurons could respond when expectations are not met. This is reminiscent of the idea from Felsen- berg et al. that interactions between reward and punishment- related compartments in the MB guide bidirectional learning (26, 45, 46, 51). However, here we extend the idea by proposing that reward would not only modify KC-MBON synapses but also modulate the baseline dopamine release or ﬁring threshold of reward-related dopaminergic neurons. Similarly, upon missing an expected reward, learning would do the same for MBONs and DANs in punishment-related compartments. The resulting behavior would depend on the balance between the activity of both reward and punishment compartments; if the reward and punishment baselines were updated correctly, such a mechanism could produce a covariance-based rule and support operant matching. This mechanism would also tie into the notion that phasic dopamine release (i.e., the difference of dopamine from its baseline level) mediates the RPE signal in mammals. Future experiments can distinguish between these hypotheses. For instance, neural recordings can probe how DAN activity changes over the course of the task, and connectomics can identify other neurons in the system that may be important for the computing of reward expectation. These types of experiments are easily doable in the D. melanogaster model. Paired with further modeling efforts and the foraging framework we developed, the ﬂy MB promises to be a system in which we can understand decision-making at a level of detail that is presently unparalleled in systems neuroscience. Materials and Methods Thefollowingisabriefdescriptionofthepaper’smethods.Acompletedescription can be found in SI Appendix. Both brief and supplemental methods consist of the same section headings. Fly Strains and Rearing. D. melanogaster were raised on standard cornmeal food supplemented with 0.2 mM all-trans-retinal at 25 ◦C (for Gr64f lines) or 21 ◦C (for other lines) with 60% relative humidity and kept in dark throughout. using https://ﬂycrispr.org/target-ﬁnder (69). The gRNA were then cloned into pCFD5_5 (85). Y-arena. A detailed schematic of the apparatus is provided in SI Appendix and Information 1. A description of the custom MATLAB code (MATLAB 2018b, Mathworks) used to control the Y-arena can be found in SI Appendix. Circular Olfactory Arena. Group learning experiments (SI Appendix, Fig. S9) were performed in a previously described circular arena (22). Behavioral Experiments. For all experiments in the paper, two or three of the odorants, 3-octanol (OCT) [Sigma-Aldrich 218405], 4-methylcyclohexanol (MCH) [Sigma-Aldrich 153095], and pentyl acetate (PA) [Sigma-Aldrich 109584] were used. Details regarding the instantiation of probabilistic rewards etc. can be found in SI Appendix. Quantitative Analysis and Behavioral Modeling. All analysis and modeling were performed using MATLAB 2020b (Mathworks). Details are described in SI Appendix. Neural Circuit Model of Dynamic Foraging. We designed two versions of a neural circuit model, inspired by work from Loewenstein and Seung (19), that were used to simulate behavior. The ﬁrst version aimed to directly replicate the model used by Loewenstein and Seung (SI Appendix and Fig. S4A). The second version incorporated modiﬁcations that made it more appropriate to our task and the mushroom body (Fig. 3 A, Right and SI Appendix). Plasticity Requirements of Operant Matching in the Mushroom Body Model. An expansion of the mathematical proof provided by Loewenstein and Seung’s to incorporate the structure of our task and architecture of the MB can be found in the eponymous section of SI Appendix. Logistic Regression Model for Estimating Learning Rules. To determine the learning rules that best predict ﬂy behavior, we designed a logistic regression model that made use of the known relationship between MBON activity and behavior (Fig. 4A). The mathematical working of this model can be found in the eponymous section of SI Appendix. Data, Materials, and Software Availability. Matlab code and data have been deposited in Zenodo repositories (86, 87). ACKNOWLEDGMENTS. This work was supported by HHMI. We thank Igor Ne- grashov, Tobias Goulet, Peter Polidoro, Steven Sawtelle, and Jon Arnold for help designing and fabricating the Y-arena and the Janelia Fly Facility for ﬂy rearing support. We also thank all the members of the Turner and Fitzgerald groups for insightful discussions, and Brad Hulse, Eyal Gruntman, Sandro Romani, Mehrab Modi, Yichun Shuai, Laura Grima, Luke Coddington, and Yoshi Aso for feedback on the manuscript. A.E.R. would like to thank The Solomon H. Snyder Department of Neuroscience’s Graduate Training Program and thesis committee members Vivek Jayaraman, Ann Hermundstad, Jeremiah Cohen, Christopher Potter, Yoshi Aso, and Erik Snapp for their guidance. Cloning. The Gr64f promoter was ampliﬁed using Q5 High-Fidelity 2X- Master Mix (New England Biolabs) from the Gr64f-GAL4 plasmid (84) and cloned into the FseI/EcoRI digested backbone of pBPLexAp65 (27) using NEBuilder HiFi DNA Assembly(NewEnglandBiolabs).FourgRNAforthegeneDop1R1weredesigned Author aﬀiliations: aJanelia Research Campus, HHMI, Ashburn, VA 20147; bSolomon H. Snyder Department of Neuroscience, Johns Hopkins School of Medicine, Baltimore, MD 21205; and cDepartment of Physiology and Pharmacology, Sackler Faculty of Medicine, Sagol School of Neuroscience, The School of Physics and Astronomy, Tel Aviv University, Tel Aviv 6997801, Israel D. W. Stephens, J. R. Krebs, Foraging Theory (Princeton University Press, 1986). B. Y. Hayden, M. E. Walton, Neuroscience of foraging. Front. Neurosci. 8, 81 (2014). 1. 2. 3. W. Schultz, Neural coding of basic reward terms of animal learning theory, game theory, microeconomics and behavioural ecology. Curr. Opin. Neurobiol. 14, 139–147 (2004). J. M. Pearce, Animal Learning and Cognition: An Introduction (Psychology Press, ed. 3, 2008). P. W. Glimcher, E. Fehr, Neuroeconomics: Decision Making and the Brain (Academic Press, 2013). 4. 5. 6. C. C. Beron, S. Q. Neufeld, S. W. Linderman, B. L. Sabatini, Mice exhibit stochastic and efﬁcient action switching during probabilistic decision making. Proc. Natl. Acad. Sci. U.S.A. 119, e2113961119 (2022). 7. W. D. Pierce, W. F. Epling, Choice, matching, and human behavior: A review of the literature. 8. Behav. Anal. 6, 57–76 (1983). U. Greggers, R. Menzel, Memory dynamics and foraging strategies of honeybees. Behav. Ecol. Sociobiol. 32, 17–29 (1993). PNAS 2023 Vol. 120 No. 39 e2221415120 https://doi.org/10.1073/pnas.2221415120 11 of 12 9. R. J. Herrnstein, The Matching Law: Papers in Psychology and Economics (Harvard University Press, 1997). 50. J. Felsenberg et al., Re-evaluation of learned information in Drosophila. Nature 544, 240–244 (2017). 10. B. Lau, P. W. Glimcher, Value representations in the primate striatum during matching behavior. 51. J. Felsenberg et al., Integration of parallel opposing memories underlies memory extinction. Cell Neuron 58, 451–463 (2008). 175, 709–722.e15 (2018). 11. K. I. Tsutsui, F. Grabenhorst, S. Kobayashi, W. Schultz, A dynamic code for economic object 52. C. Eschbach et al., Recurrent architecture for adaptive regulation of learning in the insect brain. Nat. valuation in prefrontal cortex neurons. Nat. Commun. 7, 12554 (2016). Neurosci. 23, 544–555 (2020). 12. B. A. Bari et al., Stable representations of decision variables for ﬂexible behavior. Neuron 103, 922–933 (2019). 13. K. Iigaya et al., Deviation from the matching law reﬂects an optimal strategy involving learning 14. over multiple timescales. Nat. Commun. 10, 1466 (2019). L. P. Sugrue, G. S. Corrado, W. T. Newsome, Matching behavior and the representation of value in the parietal cortex. Science 304, 1782–1787 (2004). 15. B. Lau, P. W. Glimcher, Dynamic response-by-response models of matching behavior in rhesus monkeys. J. Exp. Anal. Behav. 84, 555–579 (2005). 53. S. E. Seidenbecher, J. I. Sanders, A. C. von Philipsborn, D. Kvitsiani, Reward foraging task and model-based analysis reveal how fruit ﬂies learn value of available options. PLoS One 15, e0239616 (2020). 54. A. Claridge-Chang et al., Writing memories with light-addressable reinforcement circuitry. Cell 139, 405–415 (2009). 55. K. S. Honegger, M. A. Y. Smith, M. A. Churgin, G. C. Turner, B. L. de Bivort, Idiosyncratic neural coding and neuromodulation of olfactory individuality in Drosophila. Proc. Natl. Acad. Sci. U.S.A. 117, 23292–23297 (2020). 16. J. R. Wickens, J. N. J. Reynolds, B. I. Hyland, Neural mechanisms of reward-related motor learning. 56. A. Lesar, J. Tahir, J. Wolk, M. Gershow, Switch-like and persistent memory formation in individual Curr. Opin. Neurobiol. 13, 685–690 (2003). Drosophila larvae. eLife 10, e70317 (2021). 17. A. Soltani, X. J. Wang, A biophysically based neural model of matching law behavior: Melioration 57. M. M. Simonnet, M. Berthelot-Grosjean, Y. Grosjean, Testing Drosophila olfaction with a Y-maze by stochastic synapses. J. Neurosci. 26, 3731–3744 (2006). assay. J. Vis. Exp., (2014). 18. U. Pereira-Obilinovic, H. Hou, K. Svoboda, X. J. Wang, Brain mechanism of foraging: Reward- 58. R. Mohandasan, F. M. Iqbal, M. Thakare, M. Sridharan, G. Das, Enhanced olfactory memory dependent synaptic plasticity or neural integration of values? bioRxiv [Preprint] (2022). https://doi. org/10.1101/2022.09.25.509030 (Accessed 13 October 2022). performance in trap-design Y-mazes allows the study of novel memory phenotypes in Drosophila (2021). 19. Y. Loewenstein, H. S. Seung, Operant matching is a generic outcome of synaptic plasticity based on the covariance between reward and neural activity. Proc. Natl. Acad. Sci. U.S.A. 103, 15224–15229 (2006). 59. S. A. Lewis, D. C. Negelspach, S. Kaladchibachi, S. L. Cowen, F. Fernandez, Spontaneous alternation: A potential gateway to spatial working memory in Drosophila. Neurobiol. Learn. Mem. 142, 230–235 (2017). 20. W. G. Quinn, W. A. Harris, S. Benzer, Conditioned behavior in Drosophila melanogaster. Proc. Natl. 60. S. M. Buchanan, J. S. Kain, B. L. de Bivort, Neuronal control of locomotor handedness in Acad. Sci. U.S.A. 71, 708–712 (1974). 21. Y. Shuai, Y. Hu, H. Qin, R. A. A. Campbell, Y. Zhong, Distinct molecular underpinnings of Drosophila olfactory trace conditioning. Proc. Natl. Acad. Sci. U.S.A. 108, 20201–20206 (2011). 22. Y. Aso et al., Mushroom body output neurons encode valence and guide memory-based action 23. selection in Drosophila. eLife 3, 1–42 (2014). T. Ichinose et al., Reward signal in a recurrent circuit drives appetitive long-term memory formation. eLife 4, e10719 (2015). 24. Y. Aso, G. M. Rubin, Dopaminergic neurons write and update memories with cell-type-speciﬁc rules. eLife 5, 1–15 (2016). 25. S. Sayin et al., A neural circuit arbitrates between persistence and withdrawal in hungry Drosophila. Neuron, (2019). 61. Drosophila. Proc. Natl. Acad. Sci. U.S.A. 112, 6700–6705 (2015). T. D. Wiggin, Y. Hsiao, J. B. Liu, R. Huber, L. C. Grifﬁth, Rest is required to learn an Appetitively- Reinforced operant task in Drosophila. Front. Behav. Neurosci. 15, 681593 (2021). 62. B. Brembs, M. Heisenberg, The operant and the classical in conditioned orientation of Drosophila melanogaster at the ﬂight simulator. Learn. Mem. 7, 104–115 (2000). 63. C. Rohrsen et al., Pain is so close to pleasure: The same dopamine neurons can mediate approach and avoidance in Drosophila. bioRxiv [Preprint] (2021). https://doi.org/10.1101/2021.10.04. 463010 (Accessed 3 May 2023). 64. Y. Jiao, S. J. Moon, X. Wang, Q. Ren, C. Montell, Gr64f is required in combination with other gustatory receptors for sugar detection in Drosophila. Curr. Biol. 18, 1797–1801 (2008). 26. C. König, A. Khalili, T. Niewalda, S. Gao, B. Gerber, An optogenetic analogue of second-order 65. H. Haberkern et al., Visually guided behavior and optogenetically induced learning in Head-Fixed reinforcement in Drosophila. Biol. Lett. 15, 20190084 (2019). ﬂies exploring a virtual landscape. Curr. Biol. 29, 1647–1659.e8 (2019). 27. B. D. Pfeiffer et al., Reﬁnement of tools for targeted gene expression in Drosophila. Genetics 186, 66. M. A. Y. Smith, K. S. Honegger, G. Turner, B. de Bivort, Idiosyncratic learning performance in ﬂies. 735–755 (2010). Biol. Lett. 18, 20210424 (2022). 28. A. Jenett et al., A GAL4-Driver line resource for Drosophila neurobiology. Cell Rep. 2, 991–1001 67. B. Y. Hayden, Economic choice: The foraging perspective. Curr. Opin. Behav. Sci. 24, 1–6 (2012). 29. N. C. Klapoetke et al., Independent optical excitation of distinct neural populations. Nat. Methods 68. 11, 338–346 (2014). (2018). L. N. Groschner, L. C. W. Hak, R. Bogacz, S. DasGupta, G. Miesenböck, Dendritic integration of sensory evidence in perceptual Decision-Making. Cell 173, 894–905.e13 (2018). 30. O. Riabinina et al., Improved and expanded Q-system reagents for genetic manipulations. Nat. 69. S. J. Gratz et al., Highly speciﬁc and efﬁcient CRISPR/Cas9-catalyzed homology-directed repair in Methods 12, 219–222 (2015). Drosophila. Genetics 196, 961–971 (2014). 31. Z. Zheng et al., A complete electron microscopy volume of the brain of adult Drosophila 70. G. C. Turner, M. Bazhenov, G. Laurent, Olfactory representations by Drosophila mushroom body melanogaster. Cell 174, 730–743.e22 (2018). neurons. J. Neurophysiol. 99, 734–746 (2008). 32. F. Li et al., The connectome of the adult Drosophila mushroom body provides insights into function. 71. R. A. A. Campbell et al., Imaging a population code for odor identity in the Drosophila mushroom eLife 9, 1–86 (2020). body. J. Neurosci. 33, 10568–10581 (2013). 33. M. Heisenberg, A. Borst, S. Wagner, D. Byers, Drosophila mushroom body mutants are deﬁcient in 72. S. J. C. Caron, V. Ruta, L. F. Abbott, R. Axel, Random convergence of olfactory inputs in the olfactory learning. J. Neurogenet. 2, 1–30 (1985). Drosophila mushroom body. Nature 497, 113–117 (2013). 34. J. Séjourné et al., Mushroom body efferent neurons responsible for aversive olfactory memory 73. M. N. Modi, Y. Shuai, G. C. Turner, The Drosophila mushroom body: From architecture to algorithm retrieval in Drosophila. Nat. Neurosci. 14, 903–910 (2011). in a learning circuit. Annu. Rev. Neurosci. 43, 465–484 (2020). 35. C. Liu et al., A subset of dopamine neurons signals reward for odour memory in Drosophila. Nature 74. W. Schultz, P. Dayan, P. R. Montague, A neural substrate of prediction and reward. Science 275, 488, 512–516 (2012). 1593–1599 (1997). 36. N. Yamagata et al., Distinct dopamine neurons mediate reward signals for short- and long-term 75. G. S. Berns, S. M. McClure, G. Pagnoni, P. R. Montague, Predictability modulates human brain 37. memories. Proc. Natl. Acad. Sci. U.S.A. 112, 578–583 (2014). T. Hige, Y. Aso, M. N. Modi, G. M. Rubin, G. C. Turner, Heterosynaptic plasticity underlies aversive olfactory learning in Drosophila article. Neuron 88, 985–998 (2015). response to reward. J. Neurosci. 21, 2793–2798 (2001). 76. R. Hattori, B. Danskin, Z. Babic, N. Mlynaryk, T. Komiyama, Area-Speciﬁcity and plasticity of History- Dependent value coding during learning. Cell 177, 1858–1872.e15 (2019). 38. D. Owald et al., Activity of deﬁned mushroom body output neurons underlies learned olfactory 77. J. X. Wang et al., Prefrontal cortex as a meta-reinforcement learning system. Nat. Neurosci. 21, behavior in Drosophila. Neuron 86, 417–427 (2015). 860–868 (2018). 39. R. Cohn, I. Morantte, V. Ruta, Coordinated and compartmentalized neuromodulation shapes 78. A. Zolin et al., Context-dependent representations of movement in Drosophila dopaminergic sensory processing in Drosophila. Cell 163, 1742–1755 (2015). reinforcement pathways. Nat. Neurosci. 24, 1555–1566 (2021). 40. J. A. Berry, A. Phan, R. L. Davis, Dopamine neurons mediate learning and forgetting through 79. J. A. Harris, J. S. Carpenter, Response rate and reinforcement rate in Pavlovian conditioning. J. Exp. bidirectional modulation of a memory trace. Cell Rep. 25, 651–662.e5 (2018). Psychol. Anim. Behav. Process. 37, 375–384 (2011). 41. A. Handler et al., Distinct dopamine receptor pathways underlie the temporal sensitivity of 80. S. Lim et al., Inferring learning rules from distributions of ﬁring rates in cortical neurons. Nat. associative learning. Cell 178, 60–75.e19 (2019). Neurosci. 18, 1804–1810 (2015). 42. M. E. Villar et al., Differential coding of absolute and relative aversive value in the Drosophila brain. 81. Z. Ashwood, N. A. Roy, J. H. Bak, J. W. Pillow, “Inferring learning rules from animal decision- 43. Curr. Biol., (2022). L. Jiang, A. Litwin-Kumar, Models of heterogeneous dopamine signaling in an insect learning and memory center. PLoS Comput. Biol. 17, e1009205 (2021). 44. M. Springer, M. P. Nawrot, A mechanistic model for reward prediction and extinction learning in the fruit ﬂy. eNeuro 8 (2021). making” in Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. F. Balcan, H. Lin, Eds. (Curran Associates, Inc., 2020), vol. 33, pp. 3442–3453. 82. B. Confavreux, F. Zenke, E. Agnes, T. Lillicrap, T. Vogels, A meta-learning approach to (re) discover plasticity rules that carve a desired function into a neural network. Adv. Neural Inf. Process. Syst. 33, 16398–16408 (2020). 45. M. Adel, L. C. Grifﬁth, The role of dopamine in associative learning in Drosophila: An updated 83. R. Keiﬂin, P. H. Janak, Dopamine prediction errors in reward learning and addiction: From theory uniﬁed model. Neurosci. Bull. 37, 831–852 (2021). to neural circuitry. Neuron 88, 247–263 (2015). 46. E. Gkanias, L. Y. McCurdy, M. N. Nitabach, B. Webb, An incentive circuit for memory dynamics in 84. A. Dahanukar, Y. T. Lei, J. Y. Kwon, J. R. Carlson, Two Gr genes underlie sugar reception in the mushroom body of Drosophila melanogaster. eLife 11 (2022). Drosophila. Neuron 56, 503–516 (2007). 47. J. E. M. Bennett, A. Philippides, T. Nowotny, Learning with reinforcement prediction errors in a 85. F. Port, S. L. Bullock, Augmenting CRISPR applications in Drosophila with tRNA-ﬂanked sgRNAs. 48. model of the Drosophila mushroom body. Nat. Commun. 12, 2569 (2021). T. Riemensperger, T. Völler, P. Stock, E. Buchner, A. Fiala, Punishment prediction by dopaminergic neurons in Drosophila. Curr. Biol. 15, 1953–1960 (2005). 49. K. V. Dylla, G. Raiser, C. G. Galizia, P. Szyszka, Trace conditioning in Drosophila induces associative plasticity in mushroom body Kenyon cells and dopaminergic neurons. Front. Neural Circuits 11, 42 (2017). Nat. Methods 13, 852–854 (2016). 86. A. E. Rajagopalan et al., DATA: Reward expectations direct learning and drive operant matching in Drosophila (v1.0.0) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7449214. Deposited 16 December 2022. 87. A. E. Rajagopalan et al., CODE: Reward expectations direct learning and drive operant matching in Drosophila (v1.1.0). Zenodo. https://doi.org/10.5281/zenodo.7986372. Deposited 30 May 2023. 12 of 12 https://doi.org/10.1073/pnas.2221415120 pnas.org
10.1073_pnas.2217096120	RESEARCH ARTICLE \| BIOPHYSICS AND COMPUTATIONAL BIOLOGY OPEN ACCESS Structure of WNT inhibitor adenomatosis polyposis coli down-regulated 1 (APCDD1), a cell-surface lipid-binding protein Fu-Lien Hsieha,b,1, Tao-Hsin Changa,b,1,2 , and Jeremy Nathansa,b,f,g,2 , Sandra B. Gabellic,d,e,3 Contributed by Jeremy Nathans; received October 6, 2022; accepted April 13, 2023; reviewed by Samuel Bouyain and Bart O. Williams Diverse extracellular proteins negatively regulate WNT signaling. One such regulator is adenomatosis polyposis coli down-regulated 1 (APCDD1), a conserved single-span transmembrane protein. In response to WNT signaling in a variety of tissues, APCDD1 transcripts are highly up-regulated. We have determined the three-dimensional structure of the extracellular domain of APCDD1, and this structure reveals an unusual archi- tecture consisting of two closely apposed β-barrel domains (ABD1 and ABD2). ABD2, but not ABD1, has a large hydrophobic pocket that accommodates a bound lipid. The APCDD1 ECD can also bind to WNT7A, presumably via its covalently bound palmi- toleate, a modification that is common to all WNTs and is essential for signaling. This work suggests that APCDD1 functions as a negative feedback regulator by titrating WNT ligands at the surface of responding cells. Wnt signaling \| negative feedback regulation \| extracellular domain \| lipid-binding protein \| X-ray structure WNT signaling plays a central role in embryonic development and tissue homeostasis. WNT signaling is negatively regulated by diverse extracellular and intracellular pathways (1). These include competitive inhibition of WNT–FRIZZLED binding [sFRP (2) and WIF1 (3)], blockade and/or increased turnover of LRP5/LRP6 [DKK (4), KREMEN (5), and SOST (6)], deacylation and cleavage of WNTs [NOTUM (7, 8) and TIKI1 (9), respectively], ubiquitination and degradation of FRIZZLED [ZNRF3 (10) and RNF43 (11)], and increased β-CATENIN phosphorylation and degradation [AXIN2 (12)]. Transcripts for several of these negative regulators, including AXIN2 and NOTUM, are up-regulated by WNT signaling, implying that the encoded proteins function as part of an adjustable autocrine or paracrine feedback loop. Among the genes most consistently and most highly induced by WNT signaling is Apcdd1 (13), which codes for a conserved single-pass transmembrane protein of ~55 kDa with a large glycosylated extracellular domain (ECD), a small cytoplasmic domain, and no discernable homology to any protein of known function. In the context of human disease, elevated WNT signaling is most prominently associated with cancer, and elevated expression of APCDD1 has been reported in colon cancer and Ewing sarcoma cells (13–16). In transfected cells, APCDD1 inhibits WNT signaling (17–19), and, in humans, an APCDD1 mutation (Leu9Arg in the signal peptide) is associated with hereditary hypo- trichosis simplex, a defect in hair follicle development (17, 20). In the brain and retina, WNT signaling is required for angiogenesis and vascular barrier formation, i.e., the blood–brain barrier and the blood–retina barrier. In mice, loss of Apcdd1 causes a transient hyperplasia of the retinal vasculature, enhanced expression of Lama2 in pericytes, and precocious development of tight junctions, an essential component of the blood–retina barrier (21, 22). Conversely, Apcdd1 overexpression in vascular endothelial cells leads to retarded vascular growth and defective tight junctions (21). These data are consistent with a model in which APCDD1 acts as a negative regulator of WNT signaling. APCDD1 has also been implicated in central nervous system (CNS) myelination, astro- cyte migration, and adipocyte differentiation. In mice, APCDD1 promotes oligodendrocyte precursor cell (OPC) differentiation ex vivo and it enhances regenerative myelination after white matter injury (18), consistent with experiments showing that WNT signaling inhibits OPC differentiation and myelination (23). In the context of demyelinating disease, APCDD1 is increased in endothelial cells in mice with experimental autoimmune encephalitis, and APCDD1 protein and APCDD1 mRNA are increased in human multiple sclerosis (MS) lesions (18, 24). In the embryonic chicken spinal cord, overproduction of APCDD1 pro- motes migration of astrocyte precursors (25). In cultured adipocytes, APCDD1 accumulates with differentiation, and siRNA-based reduction in APCDD1 inhibits adipocyte differen- tiation (26). Significance Adenomatosis polyposis coli down-regulated 1 (APCDD1)—a conserved single-span transmembrane protein containing a large extracellular domain—negatively regulates WNT signaling and plays important roles in hair follicle development, CNS vascular development, and glial differentiation. We report here the three-dimensional structure of the ECD of APCDD1, revealing an unusual architecture. The APCDD1 ECD consists of two closely apposed β-barrel domains (ABD1 and ABD2). ABD2 contains a large hydrophobic pocket that accommodates a bound lipid. In an in vitro assay, the ECD of APCDD1 bound to WNT7A, which contains a covalently linked palmitoleate. Collectively, the results of this study suggest that APCDD1 serves as a negative feedback regulator of WNT signaling by neutralizing WNT ligands. The authors declare no competing interest. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1F.-L.H. and T.-H.C. contributed equally to this work. 2To whom correspondence may be addressed. Email: [email protected] or [email protected]. 3Present address: Merck & Co., Inc., West Point, PA 19486. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2217096120/-/DCSupplemental. Published May 8, 2023. PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 1 of 12 At present, the mechanism of APCDD1 action is uncertain, as APCDD1 has been variously reported to associate with LRP5, WNT3A, and β-CATENIN (17–19). To gain a mechanistic under- standing of APCDD1 function, we have determined the three-dimensional structure of the ECD of APCDD1 by X-ray crystallography. The structure reveals an architecture containing two closely related β-barrel domains (ABD1 and ABD2). Structural and functional analyses show that APCDD1 has a large hydrophobic pocket in ABD2 and that APCDD1 can bind to WNT7A, pre- sumably via its covalently bound palmitoleate, a modification that is common to all WNTs and is essential for signaling. Taken together, the results reported here suggest that APCDD1 acts as a negative feedback regulator by lowering the concentration of avail- able WNT ligands at the surface of responsive cells. Results Protein Production and Structure Determination of the APCDD1 ECD. To determine the three-dimensional structure of the ECD of mouse APCDD1 (referred to hereafter simply as “APCDD1”), we produced this domain as a glycosylated secreted protein using human embryonic kidney (HEK293) cells, crystallized it, and collected X-ray diffraction data to 1.95 Å and 2.15 Å resolution from two crystal forms (SI Appendix, Fig. S1A and Table S1). Attempts to use molecular replacement for structure determination failed because of a lack of discernable homology to known structures and the low quality of predicted structures. Therefore, we produced, purified, and crystallized a fusion protein between engineered maltose-binding protein (eMBP) (27) and APCDD1 (SI Appendix, Fig. S1B), collected X-ray diffraction data to 2.3 Å resolution, and used molecular replacement with eMBP to obtain phase information (SI Appendix, Table S1). Although the resulting electron density map provided a good fit to the eMBP half of the fusion protein, the map of APCDD1 was of insufficient quality for model building (Fig. 1A). Recognizing that the N- and C-terminal halves of APCDD1 have ~25% amino acid identity and similar predicted secondary structure patterns, and, therefore, that APCDD1 very likely con- sists of two domains with conserved three-dimensional structures, we used this constraint, together with the RoseTTAFold predic- tion algorithm (28), to generate a three-dimensional model of APCDD1. Although the resulting model was a relatively poor match to the final structure (Fig. 1B), it correctly captured the paired β-barrel domain structure of APCDD1 (Fig. 1C; described below). Fitting this model to the electron density map of eMBP-APCDD1 provided the starting point for iteratively solving the structure of the APCDD1 half of the fusion protein (Fig. 1 D–F and SI Appendix, Table S1). Molecular replacement with this structure was then used to solve the structure of APCDD1 in the two crystal forms described above (Fig. 1 G and H). Architecture of the APCDD1 ECD. The structure of APCDD1 reveals an unusual architecture consisting of two β-barrel domains (ABD1 and ABD2) that differ in orientation by ~90° (Fig. 2 and SI Appendix, Fig. S2). ABD1 and ABD2 have, respectively, three and two intradomain disulfide bonds, a large area of interfacial contact, and one interdomain disulfide bond (Cys52–Cys466). This structure is observed in all three crystal forms (eight unique APCDD1 structures), with a global rmsd of 0.7 to 1.1 Å among structures (SI Appendix, Fig. S2E and Table S2). ABD1 and ABD2 present highly similar folds, consisting of two curved antiparallel β-sheets connected by an α-helix and loop domain (AHLD) (Fig. 2 B and C). In all three crystal forms, AHLD1 and AHLD2 differ by ~30° in orientation relative to their linked β-barrel domains (Fig. 2C and SI Appendix, Fig. S2 E and F). For both ABD1 and ABD2, the interior surface of the β-barrel is lined by hydrophobic amino acids. The barrels are open to the solvent on the side adjacent to AHLD, whereas the other side is closed by tight packing between side chains from β1 and β2 for ABD1 and from α3 and β12 for ABD2 (Fig. 3 A and B). Hydrophobic Pockets and Lipid Binding. Intriguingly, we observed an electron density consistent with the acyl chain of a lipid molecule in the hydrophobic pocket of ABD2 in two of the four eMBP-APCDD1 structures (chains A and C; Figs. 3 C and D and 4A). We suspected that these lipid molecules became bound following APCDD1 secretion from HEK293 cells, which were grown in the presence of 2% bovine serum, a plentiful source of lipids. To define the identities of the bound molecules, we extracted hydrophobic molecules from the purified APCDD1 sample used for crystallization and analyzed the extracted species by high-performance liquid chromatography (HPLC) coupled with mass spectrometry (MS). This analysis revealed stearic acid, a C18 fatty acid (i.e., with 18 carbon atoms), as the most prevalent species (SI Appendix, Fig. S3 A and B). Therefore, we modeled a molecule of stearic acid into the electron density in the ABD2 pocket; the subsequent structural refinement revealed a good fit between stearic acid and the electron density (Fig. 3C and SI Appendix, Fig. S3 C and D). Since APCDD1 inhibits WNT signaling and WNTs have a covalently linked palmitoleate (PAM; C16 fatty acid; ref. 1) that contributes to FRIZZLED binding and signaling (31–33), we also fitted a PAM molecule into the electron density in the ABD2 pocket (Figs. 3D and 4 A–C). The subsequent refinement showed a good fit between PAM and the electron density. Interestingly, this structural analysis also revealed different lipid conformations in the two APCDD1 structures (Fig. 4B), both of which exhibit multiple contacts between the lipid and hydrophobic side chains lining the ABD2 pocket (Fig. 4 D and E). A comparison between the bound and unbound APCDD1 structures shows no major conformational change upon lipid binding (SI Appendix, Fig. S2F). Despite the nearly identical backbone configurations of the β-barrel residues in ABD1 and ABD2, the interior volumes of their hydrophobic pockets are dramatically different: ~35 Å3 for ABD1 vs. ~350 Å3 for ABD2 (Fig. 3 A and B and SI Appendix, Table S3), as calculated with CASTp (29). To define the molecular basis for this difference, we superimposed the structures of ABD1 and ABD2 (Figs. 2, 5, and 6). Two loops, which we refer to as Gate-I and Gate-II, differ markedly in both primary sequence and configura- tion between ABD1 and ABD2 (Figs. 3 A and B, 5, and 6). In ABD1, Gate-I and Gate-II are positioned closer to the center of the β-barrel, whereas, in ABD2, Gate-I and Gate-II are positioned away from the center of the β-barrel (Figs. 3 A and B and 6). The differ- ences between ABD1 and ABD2 can be further appreciated by comparing the identities and spatial locations of the individual amino acid side chains that line their hydrophobic pockets (Fig. 5 B–D). In particular, Gate-1 side chains show a concerted inward shift within the ABD1 pocket and, together with multiple other side chains in ABD1, largely occupy the interior of the superim- posed ABD2 pocket (Fig. 5 E–H). These distinguishing features of the ABD1 and ABD2 pockets are observed in all eight APCDD1 structures (Fig. 5C and SI Appendix, Table S3). With respect to the function of the individual ABD1 and ABD2 domains, our attempts to address this question by producing the individual domains have thus far failed, most likely due to the instability of the individual domains. We note that the two domains have a large area of interdomain contact and an interdo- main disulfide bond, which likely stabilizes them. 2 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org Fig. 1. Electron density map and chain tracing of eMBP-APCDD1. (A) The initial electron density map (purple meshes) of eMBP-APCDD1 contoured at the 1.8 σ level after molecular replacement in PHASER using eMBP structures as the search model. Three eMBP copies (yellow, green, and magenta) fit into the electron density, whereas the electron density for APCDD1 is not interpretable. (B) Superposition of predicted APCDD1 model which was generated using RoseTTAFold and X-ray structure of APCDD1 (with rmsd of 3.79 Å over 358 Cα atoms) revealed high structural difference. (C) Superposition of predicted models of the β-barrel regions of ABD1 and ABD2, generated using RoseTTAFold, referred to as the core domain for molecular replacement. (D) The electron density (pink meshes) contoured at the 1.8 σ level after molecular replacement in PHASER with three eMBP copies fixed and using predicted models of core domains as search models. (E) The electron density modified map (red meshes) from PARROT contoured at the 1.8 σ level. (F) The sigmaA-weighted 2\|FO\|-\|FC\| electron density (blue meshes) after refinement in PHENIX contoured at the 1.3 σ level. The structure of the eMBP-APCDD1 fusion protein is shown as a ribbon representation (magenta). The linker between eMBP and APCDD1 is indicated by a green arrow. (G) The sigmaA-weighted 2\|FO\|-\|FC\| electron density (blue meshes) contoured at the 1.0 σ level. A close-up view of APCDD1 structure (chain A of crystal-form I) is shown as sticks. (H) The sigmaA-weighted 2\|FO\|-\|FC\| electron density (blue meshes) contoured at the 1.0 σ level. A close-up view of the APCDD1 structure (chain A of crystal-form II) is shown as sticks. PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 3 of 12 Fig. 2. Structure of the APCDD1 ECD in the apo- form. (A) Schematic diagram of APCDD1 (SP, signal peptide; TM, transmembrane domain). ABD1 and ABD2 are colored in blue and cyan, respectively. Six disulfide bonds and three N-linked glycosylation sites are denoted as orange lines and green hexagons, respectively. (B) Ribbon representation of the ECD of APCDD1 (ABD1, blue; ABD2, cyan) in two views. AHLD1 and AHLD2 are marked with purple dotted circles. Disulfide bonds and N-linked glycans are shown as sticks. The N- and C-termini are labeled. The top-left Inset shows a cartoon of APCDD1 on the cell surface. The ABD1 and ABD2 β-barrels differ in orientation by approximately 90° and have an interdomain disulfide bond (SS-1). (C) Superposition of ABD1 and ABD2 (with rmsd of 2.11 Å over 160 Cα atoms) reveals high structural similarity between the β-barrels and an orientation difference between AHLD1 and AHLD2, as shown by the purple arrow. Comparisons between APCDD1 and Other Lipid-Binding Proteins. Binding to the PAM that is covalently linked to WNT ligands is a recurrent theme among protein that transduce or modulate WNT signals. These proteins include FRIZZLED (33), NOTUM (7, 8), WNTLESS (WLS) (34), and DALLY-LIKE (DLP) (35). A comparison of the structures of these and other proteins in complex with hydrophobic ligands shows great diversity in the volumes of their lipid binding cavities, ranging from ~55 Å3 to ~400 Å3 (Fig. 7 A–G and SI Appendix, Fig. S4 and Table S3). At ~350 Å3, the ABD2 pocket is at the high end of this distribution and is predicted to Fig. 3. Structural analyses of ABD1, ABD2, and ABD2 in complex with a lipid. (A) Ribbon diagram of ABD1 (blue). The interior volume of the ABD1 pocket was rendered with CASTp (29) using a 1.4 Å probe and is colored green. The regions involved in determining pocket size are colored orange (Gate-I) and yellow (Gate-II). N-linked glycans are shown as sticks. (B) The structure of ABD2 (cyan) reveals a large hydrophobic pocket for ligand binding, rendered as described for (A) and is shown with a green interior volume. (C) Structure of ABD2 (eMBP-APCDD1 chain A; charcoal gray) was refined with stearic acid (green sticks) bound. The 2\|FO\|-\|FC\| electron density for stearic acid (blue meshes) is contoured at 0.9 σ. (D) Structure of ABD2 (eMBP-APCDD1 chain A; gray) was refined with PAM (magenta sticks) bound. The 2\|FO\|-\|FC\| electron density for PAM (green meshes) is contoured at 0.9 σ. 4 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org Fig. 4. Lipid-binding pocket of ABD2. (A) Structural superposition of APCDD1 with bound PAM (magenta sticks) from eMBP-APCDD1 chain A and APCDD1 with bound PAM (yellow sticks) from eMBP-APCDD1 chain C. No major conformational changes are observed. The PAM-binding pocket of ABD2 is marked with a red dotted circle. (B) A close-up view of the PAM molecules as shown in (A). (C) The 2\|FO\|-\|FC\| electron density (blue meshes) of PAM (yellow sticks) from eMBP- APCDD1 chain C contoured at the 0.8 σ level. (D and E) Diagrams of ABD2 and PAM interactions generated with LigPlot+ (30). Panel (D), eMBP-APCDD1 chain A. Panel (E), eMBP-APCDD1 chain C. Atoms are colored as follows: nitrogen, blue; oxygen, red; carbon, black. Hydrogen bonds are displayed as green dashed lines. Red eyelashes denote hydrophobic interactions. accommodate hydrophobic ligands in the 230 to 440 Da range (Fig. 7H and SI Appendix, Table S3). The shallow ABD1 pocket might accommodate a small ligand or part of a larger ligand. To explore the evolutionary origin of APCDD1, we performed a 3D structure search of the Protein Data Bank (PDB) and AlphaFold (36) databases using the DALI server (37). This search returned weak homologs, including the lipocalin and peripheral myelin protein 2 (P2) families, with primary sequence identities in the 5 to 10% range. As seen for ABD1 and ABD2, the members of these protein families have a hydrophobic binding pocket formed by a β-barrel domain (SI Appendix, Fig. S5). Detailed structural comparisons between APCDD1 and members of the lipocalin and P2 families showed marked differences in the shapes of their β-barrels and the orientations of β-sheets, giving an average rmsd of ~2.4 to 3.7Å (SI Appendix, Fig. S5). Proteins in the lipocalin and P2 families bind a wide variety of hydrophobic molecules via the pocket within their β-barrel domains (38). Across both families, the mean volume of the hydrophobic pocket is ~267 Å3, as calculated with CASTp (29). Evolution of APCDD1 and APCDD1L. Nearly all vertebrate genomes code for two APCDD1 homologous sequences: APCDD1 and APCDD1-like (APCDD1L), which exhibit ~50% amino acid identity throughout their length (SI Appendix, Fig. S6). The APCDD1L sequences retain the principal features of the APCDD1 sequence, including the 12 conserved and disulfide-bonded cysteines (SI Appendix, Fig. S6). Interestingly, APCDD1L amino acid sequences show a several-fold greater rate of evolutionary change compared to APCDD1 sequences, as seen in the dendrogram in SI Appendix, Fig. S7A. Threading the rat APCDD1L sequence into the mouse APCDD1 structure suggests that APCDD1L adopts very nearly the same structure, with an ABD2 pocket that can also accommodate a lipid (SI Appendix, Fig. S7). At present, the function of APCDD1L is unknown. APCDD1 Binds to WNT7A. Based on the structure of APCDD1, the most attractive hypothesis for its action as a WNT inhibitor is that it binds directly to WNTs via their covalently linked PAM. To test this idea, we chose WNT7A from among the 19 mammalian WNTs, because WNT7A, like APCDD1, regulates CNS vascular development and barrier maturation (39). More specifically, WNT7A produced by glia and neurons activates FRIZZLED receptors on CNS endothelial cells, which respond with high-level WNT signaling, including high-level Apcdd1 expression. Most WNTs, including WNT7A, are not well behaved biochem- ically and they typically form inactive aggregates following secretion into conditioned medium (40). However, in some instances, coex- pression and coassembly with a binding partner facilitates secretion of a native WNT complex (33, 40, 41). Therefore, we decided to coexpress in HEK293T cells ALFA epitope-tagged WNT7A with either a C-terminal biotinylated APCDD1-mVenus fusion or, as a positive control, a C-terminal biotinylated FZD4 cysteine-rich domain (CRD)-mVenus fusion. We then applied immobilized metal affinity chromatography (IMAC) to concentrate the PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 5 of 12 Fig. 5. Structural analysis and comparison of hydrophobic pockets of ABD1 and ABD2. (A) Structure-based sequence alignment of ABD1 with ABD2. Secondary structure elements are represented. Purple and green rectangles highlight residues lining the hydrophobic pockets of ABD1 and ABD2, respectively. The regions of Gate-I and Gate-II are marked with horizontal colored lines. (B) Residues lining the hydrophobic pocket of ABD2 are shown as sticks and labeled. (C) The interior volume of the ABD2 pocket (gray), rendered with CASTp ((29)) together with residues that line it, are shown as green sticks, for the eight APCDD1 structures determined in this study. (D) Stick representation of ABD1 residues corresponding to the residues that line the ABD2 hydrophobic pocket. (E–G) ABD1 and ABD2 were superimposed (Fig. 3A and B) and subsets of the corresponding pairs of ABD1 and ABD2 residues were visualized as sticks (magenta for ABD1 and green for ABD2): (E) pairs with high spatial similarity; (F) pairs with moderate spatial similarity; (G) pairs with low spatial similarity (connected by arrows). (H) ABD2 ribbon diagram (cyan; with Gate-I and Gate-II colored light and dark gray, respectively) and the hydrophobic pocket rendered as a green interior volume. Select ABD1 residues that occupy or partially occupy the ABD2 hydrophobic pocket are visualized as magenta sticks with their positions and orientations determined by the superposition of ABD1 and ABD2. preformed protein complexes from the conditioned medium, using the histidine tags on the epitope-tagged WNT7A and the mVenus fusions of APCDD1 and FZD4 CRD. The partially purified com- plexes were captured on streptavidin-coated wells, and the epitope-tagged WNT7A protein was detected with an anti-ALFA nanobody-alkaline phosphatase (AP) fusion protein and a colori- metric AP assay (Fig. 8A). The FZD4 CRD showed a strong WNT7A binding signal and APCDD1 showed a weaker WNT7A binding signal. Omitting either the biotinylated bait or the WNT7A ligand eliminated the binding signal. 6 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org Fig. 6. Structural comparison of ABD1 with ABD2. (A) Comparison between the structures of ABD1 (blue) and ABD2 (cyan). Gate-I (orange for ABD1 and gray for ABD2) and Gate-II (yellow for ABD1 and black for ABD2) are the two most important regions for determining pocket size. Red arrows indicate the change in these gates in comparing the ABD2 (large pocket) to ABD1 (small pocket) configurations. (B) Residues for the ABD1 pocket highlighted in Fig. 5A are shown as sticks (atom coloring: purple, carbon; blue, nitrogen; red, oxygen; yellow, sulfur). The red dotted circle highlights the region of the ABD1 pocket. (C) Residues for the ABD2 pocket highlighted in Fig. 5A are shown as sticks (atom coloring: green, carbon; blue, nitrogen; red, oxygen; yellow, sulfur). To address the hypothesis that the PAM linked to WNT7A binds to the hydrophobic pocket in APCDD1, we compared the WNT7A capture efficiency in the presence vs. the absence of neutral detergents [dodecyl maltoside (DDM) and Triton X-114], which would be expected to both compete for occupancy of the lipid-binding pocket and stabilize the unbound PAM. Both deter- gents reduced the WNT7A-APCDD1 binding signal, with 1% DDM showing a greater potency than 0.1% DDM (Fig. 8A). In contrast, the same detergent treatment had little effect on the interaction of WNT7A and FZD4 CRD (Fig. 8A), most likely because this interaction is stabilized by both protein–lipid and protein–protein contacts and because the PAM-binding pocket in the FZD CRD consists of a narrow groove that makes a tight fit to the PAM (33, 41). Although a definitive analysis of the WNT–APCDD1 interaction will require a high-resolution struc- ture of the complex, these biochemical data are consistent with a model in which the PAM moiety of WNT binds to the hydro- phobic pocket in APCDD1. To determine whether APCDD1 directly interacts with the extracellular four tandem β-propeller-epidermal growth factor-like domain pairs (PE1-4) of LRP5/6 coreceptors, we produced PE1-4 of LRP5 and LRP6 with a C-terminal biotin tag, captured them on streptavidin-coated wells, and probed the wells with APCDD1-AP and AP-APCDD1 fusion proteins or, as a positive control, with DKK1-AP and AP-DKK1 fusion proteins (Fig. 8B). We observed no detectable binding between the APCDD1 and the PE1-4 domains of LRP5 or LRP6, consistent with a model in which APCDD1 acts via binding to WNT ligands rather than to WNT coreceptors. The lower WNT7A-APCDD1 binding signal compared to the WNT7A-FZD4 CRD binding signal presumably reflects a lower affinity for the former interaction. If this differential binding also applies in vivo, it may be partially offset by the greater abundance of transcripts coding for APCDD1 relative to transcripts coding for WNT receptors and coreceptors. For example, in mouse brain vascular endothelial cells, in which a high level of canonical WNT signaling maintains the blood–brain barrier, Apcdd1 transcripts are 5 to 100 fold more abundant than transcripts coding for WNT receptor (FZD4), coreceptor (LRP5 and LRP6), and coactivator (GPR124 and RECK) proteins (SI Appendix, Fig. S8). Although the corresponding protein abundances in vivo are not presently known, the relative transcript abundances suggest that in brain vascular endothelial cells APCDD1 may be substantially more abundant than WNT receptors, coreceptors, and coactivators. In sum, the experiments described here reveal APCDD1 to be a transmembrane lipid-binding protein, and they suggest that APCDD1 exerts its inhibitory effect on WNT signaling by bind- ing to lipidated WNTs (Fig. 8C). PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 7 of 12 Fig. 7. Structural analyses of hydrophobic ligand-binding pockets of cell surface and secreted proteins. Hydrophobic ligand volumes (gray) and hydrophobic ligands (magenta sticks), rendered with CASTp, are shown for the following lipid-binding pockets. (A) FZD8 CRD in complex with the PAM of WNT8A (PDB ID 4F0A). (B) FZD4 CRD dimer interface in complex with PAM (PDB ID 5UWG). (C) NOTUM in complex with the PAM of a WNT7A peptide (PDB ID 4UZQ). (D) WLS in complex with the PAM of WNT8A (PDB ID 7KC4). (E) DLP in complex with the PAM of a WNT7A peptide (PDB ID 6XTZ). (F) SMOOTHENED (SMO) in complex with cholesterol (PDB ID 5L7D). (G) Retinol binding protein 4 (RBP4) in complex with retinol (PDB ID 1RBP). (H) Plot of the interior volume of the hydrophobic ligand-binding pocket in the indicated proteins vs. the molecular weight of their cognate hydrophobic ligands. The range of values for the ABD2 hydrophobic pocket volume (horizontal red bar) reflects differences among the eight APCDD1 structures and the vertical bar estimates the range of potential ligand molecular weights predicted from the volume. Quantifications are in SI Appendix, Table S3. Discussion The present study shows that the ECD of APCDD1 consists of two homologous β-barrel domains (ABD1 and ABD2), linked by one interdomain disulfide bond. Protein sequence searches and protein fold comparisons show that APCDD1 exhibits an unusual architecture. Interestingly, structural analyses show that ABD2, but not ABD1, has a hydrophobic pocket that is larger than the hydrophobic binding pockets in FRIZZLED (33), NOTUM (7, 8), and WLS (34), each of which binds the PAM that is covalently linked to WNT ligands. Evidence that ABD2 could plausibly bind the WNT-linked PAM comes from our observations of a) electron density consistent with a C16 or C18 lipid in the hydrophobic pocket of ABD2 in two of four eMBP-APCDD1 structures, and b) copurification of stearic acid with APCDD1. In support of this model, in vitro binding assays show that APCDD1 binds to WNT7A. Building on earlier reports that APCDD1 inhibits WNT signaling in transfected cells (17–19), the present study suggests that APCDD1 serves as a negative feedback regulator by titrating WNT ligands at the cell surface to reduce WNT signaling. At a technical level, this work also demonstrates the synergy that is possible between artificial intelligence-based structure prediction algorithms and traditional molecular replacement using fusion-partner-derived phase information for protein structure determination. The PAM that is covalently joined to WNT proteins plays a central role in WNT-CRD binding (31, 33, 41–44), secretion dependent upon WNT association with WLS (34), the formation of a WNT morphogen gradient via interaction with DLP (35), and WNT signaling inhibition by NOTUM, a deacylase that 8 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org APCDD1L homologues evolved as WNT regulators and that they maintain this function in present day Metazoa. In addition to the mouse Apcdd1 knockout and overexpression phenotypes described in the Introduction, Xenopus embryo experiments with Apcdd1 morpholino oligonucleotide knock- down reveal expansion in the expression domain of the ventral marker Sizzled and a reduction in the expression of the dorsal (neural tube) marker Sox2 (46). Apcdd1l morpholino oligonu- cleotide knockdown and TALEN-mediated germ-line elimina- tion of Apcdd1l in zebrafish show that a) homozygous loss of Apcdd1l is dispensable for viability and fertility, and b) progeny embryos from Apcdd1l homozygous null parents, which lack both maternal and zygotic Apcdd1l function, show expansion of the Spemann organizer region (visualized by the expression of gsc), a phenotype that appears to be of little consequence for subsequent development (46). The Apcdd1 and Apcdd1l loss-of-function phenotypes are consistent with enhanced WNT signaling, and they further suggest that APCDD1 may addition- ally inhibit Bone Morphogenetic Protein (BMP) signaling (46). It will be interesting to further explore the BMP inhibitor hypothesis and to determine whether combined elimination of both Apcdd1 and Apcdd1l in mice, frogs, or fish produces a distinctive and/or more severe developmental phenotype than elimination of either gene alone. The structure of APCDD1 presents a distinctive protein archi- tecture with ABD1 and ABD2 closely packed in an orientation that differs by ~90°, with one interdomain disulfide bond (Fig. 2). In comparisons to the lipocalin and P2 families, the ABD1 and ABD2 β-barrel domains show 5 to 10% amino acid sequence iden- tity and relatively weak structural homology. The lipocalin/P2 superfamily predates APCDD1, with superfamily members present in genome sequences in all domains of life except for Archaea. The members of the lipocalin/P2 superfamily are highly divergent at the primary sequence level and are well known for their roles in binding and transporting hydrophobic molecules (e.g., lipids, pher- omones, steroid hormones, and retinoids) using the hydrophobic pocket within the β-barrel domain (38). As a distant relative of this family, APCDD1 is unusual in having a membrane anchor and two β-barrels in the same polypeptide. Interestingly, a secreted lipocalin family member in Drosophila, SWIM, has been reported to interact with WINGLESS (one of seven Drosophila WNTs) in a PAM-dependent manner to maintain WINGLESS solubility (47). Therefore, SWIM, like APCDD1, appears to use its hydro- phobic pocket within the β-barrel domain for WNT binding via the WNT-linked PAM. It will be interesting to determine whether APCDD1 or SWIM exhibits any binding selectivity for specific WNTs. A diverse collection of extracellular and intracellular proteins have been found to negatively regulate WNT signaling. sFRP and WIF1 prevent WNT and FRIZZLED interaction by capturing WNT ligands (2, 3); DKK, SOST, and KREMEN block the formation of the WNT–FRIZZLED-LRP complex and/or reduce the cell surface concentration of LRP5/LRP6 (4–6); NOTUM a WNT-specific deacylase, inhibits WNT signaling by removing the PAM from WNT ligands (7, 8); TIKI1, a transmembrane protease, suppresses WNT signaling by degrading WNT ligands (9); ZNRF3 and RNF43, transmembrane E3 ubiquitin ligases, ubiquitylate FRIZZLED to promote its degradation (10, 11); and AXIN2 intracellularly promotes β-CATENIN phosphorylation and degradation (12). APCDD1 represents a distinctive type of negative feedback regulator of WNT signaling. The APCDD1 structure presented here, together with the WNT-binding data, imply a mode of action in which APCDD1 titrates WNTs at the surface of responding cells to reduce WNT-FRIZZLED binding. Fig. 8. Functional characterization of APCDD1. (A) APCDD1 binds to WNT7A. Left, diagram of the protein–protein interaction assay used in this study. The secreted complex between the APCDD1-mVenus-His-tag fusion protein with biotinylated Avi-tag at its C-terminus and WNT7A-His tag with an ALFA- tag at its C terminus (magenta hexagon) was concentrated by IMAC affinity chromatography and then immobilized on streptavidin-coated wells. Captured WNT7A was detected using an anti-ALFA nanobody (Nb)-AP fusion protein and a colorimetric AP reaction. Right, WNT7A binding to APCDD1 and the FZD4 CRD. The binding and washing were conducted in the following buffer conditions: 1, standard buffer; 2, in the presence of 1% DDM; 3, in the presence of 0.1% DDM; 4, in the presence of 2.25% Triton X-114 (Materials and Methods). The FZD4 CRD bait serves as a positive control. (B) APCDD1 does not detectably bind to the extracellular PE1-4 domains of LRP5 or LRP6 as determined by probing with APCDD1-AP and AP-APCDD1. The DKK1-AP and AP-DKK1 probes serve as positive controls. (C) Model showing how Apcdd1 expression and APCDD1 function provide negative regulatory feedback in canonical WNT signaling. recognizes palmitoylated WNT and releases the PAM moiety (7, 8). As shown in Fig. 7 and SI Appendix, Fig. S4, the hydropho- bic PAM-binding pockets in these WNT-interacting proteins exhibit substantial shape and size diversity. In the FRIZZLED CRD monomer and dimer-binding modes, and in WLS and DLP, the PAM-binding pockets are extended and narrow, whereas NOTUM contains a compact and globular hydrophobic PAM-binding pocket. By contrast, the structures of APCDD1 determined here reveal that the ABD2 pocket has a large, open, and globular shape that is predicted to accommodate hydrophobic ligands with molecular masses in the 230 to 440 Da range, includ- ing PAM (254 Da). Determining whether the APCDD1–WNT interaction also involves protein–protein interactions will likely require a high-resolution structure of the complex. Genome sequences predict the existence of APCDD1 homo- logues in nearly all vertebrates, as well as in a wide variety of invertebrates, including many evolutionarily distant Metazoa. Among the 100 most similar invertebrate APCDD1 homologues, amino acid alignments with mammalian APCDD1 show 25 to 40% amino acid identity spread across the entire ECD. The dis- tinction between APCDD1 and APCDD1L is clear in comparing vertebrate homologues (SI Appendix, Fig. S7A), but many inver- tebrate genomes carry one or more APCDD1/APCDD1L sequences with approximately the same degree of similarity to APCDD1 and APCDD1L. Since WNT signaling arose early in metazoan evolution (45), it is possible that the earliest APCDD1/ PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 9 of 12 Whether APCDD1 binding leads to internalization and/or deg- radation of WNTs and/or WNT-associated surface proteins remains to be determined. The upregulation of Apcdd1 transcripts in response to WNT signaling in both normal and pathologic contexts implies that APCDD1 feedback effects are likely to be of broad biological relevance. Materials and Methods Sequence-Based and Phylogenetic Analyses. The sequence-based homology search for the mouse APCDD1 sequence (UniProt code: Q3U128; ECD, residues 27 to 492) was carried out over the PDB database using the HHpred server (48). For the prediction of protein secondary structure and disordered regions, Ali2D (49–51) and Quick2D (52) were used. For phylogenetic analysis, the amino acid sequences of APCDD1 and APCDD1L were aligned using ClustalΩ (53), and the phylogenetic tree was constructed with the neighbor-joining method using MEGA7 (54, 55). The optimal tree with the sum of branch lengths = 2.41744267 is shown. The phylogenetic tree is drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the tree. The evolutionary distances were computed using the Poisson correction method and are in the units of the number of amino acid substitutions per site (55). Plasmid Design and Construction. For mouse APCDD1 constructs, a series of coding segments of the APCDD1 ECD were cloned into pHLsec-mVenus-12H (27, 56, 57) for expression in human embryonic kidney (HEK293) cells. For crys- tallization, the APCDD1 ECD (residues 27 to 481) was cloned into pHLsec-8H (58) with a C-terminal 8xHis tag. For the eMBP-APCDD1 construct, the APCDD1 ECD (residues 47 to 481) was cloned into pHLsec-eMBP-8H with an N-terminal eMBP fragment [PCR amplified from pET-11d-eMBP-8H (27)] and a C-terminal 8xHis tag. For protein–protein interaction experiments, coding segments of the APCDD1 ECD and the human FZ4 CRD (56) were cloned into pHLsec-3C-mVenus-Avi-8H (57) with a C-terminal Human Rhinovirus-3C protease cleavage site followed by a monomeric (m)Venus fusion protein, an Avi tag and finally an 8xHis tag; the resulting plasmids were named APCDD1-mV-Avi and FZ4-mV-Avi, respectively. The mouse LRP5 PE1-4 construct (LRP6-Avi) and the human LRP6 PE1-4 construct (LRP6-Avi) were generated in pHL-Avitag3 with a C-terminal Avi tag as described previously (59). For AP fusion protein constructs, the APCDD1 ECD (residues 27 to 486) and human DKK1 (UniProt code: O94907; residues 32 to 266) were cloned into pHL-N-AP-Myc-8H (with an N-terminal human AP followed by a Myc tag and an 8xHis tag) and pHLsec-C-Myc-AP-8H (with a Myc tag followed by human AP and an 8xHis tag) (57); the resulting plasmids were named AP-APCDD1, APCDD1-AP, DKK1-AP, and AP-DKK1. Human WNT7A (UniProt code: O00755; residues 32 to 349) was cloned into pHLsec-ALFA-8H with an N-terminal ALFA sequence (SRLEEELRRRLTE) (60) and a C-terminal 8xHis tag. The nanobody against ALFA (NbALFA) (60) was cloned into pET-11d-C-Myc-eAP-8H (with a C-terminal Myc tag followed by Escherichia coli AP and finally an 8xHis tag, generating Nb-AP. All constructs were confirmed by DNA sequencing. Protein Expression and Purification. HEK293T (ATCC CRL-11268) cells were maintained in a humidified 37 °C incubator with 5% CO2 in Dulbecco’s modified eagle medium (MilliporeSigma) supplemented with 2 mM L-Glutamine (L-Glu, Gibco), 0.1 mM nonessential amino acids (Gibco) and 10% [v/v] Fetal Bovine Serum (FBS, Gibco). The FBS concentration was lowered to 2% [v/v] after trans- fection with the DNA using polyethylenimine (PEI; MilliporeSigma) as described previously (61). For crystallization experiments, APCDD1 and eMBP-APCDD1 were expressed in HEK293T cells grown in HYPERFlask® Cell Culture Vessels (Corning) and cultured in the presence of 5 μM of the class I α-mannosidase inhibitor, kifunensine (62) and 4 mM valproic acid (56) after transfection. Conditioned media were collected 5 d posttransfection and supplemented with 10 mM HEPES, pH 7.5, and 5 mM imidazole. The His-tagged sample was purified by IMAC using Ni Sepharose Excel resin (GE Healthcare Life Sciences). The IMAC eluted sample was subjected to size-exclusion chromatography using HiLoad Superdex 200 pg (GE Healthcare Life Sciences) in 10 mM Bis-Tris, pH 6.5, 0.3 M NaCl (for APCDD1) and 10 mM HEPES, pH 7.5, 0.15 M NaCl (for eMBP-APCDD1). For E. coli expression and purification of Nb-AP, the plasmid DNA was trans- formed into E. coli BL21 Star™ (DE3) cells (ThermoFisher) and induced with 0.2 mM isopropyl β-thiogalactopyranoside in Luria broth containing 100 μg/mL ampicillin (MilliporeSigma) at room temperature (~25°C) overnight. The cell pellets were harvested by centrifugation and resuspended in B-PER bacterial protein extract reagent (ThermoFisher) supplemented with 50 mM HEPES, pH 7.5, 0.3 M NaCl, 30 mM imidazole, 1 mM MgCl2, 500 U benzonase (MilliporeSigma), 0.2 mg/mL lysozyme, and cOmplete Protease Inhibitor Cocktail (MilliporeSigma). The cell lysate was clarified by centrifugation, and the supernatant was filtered using a 0.45-μm Steritop filter (MilliporeSigma). Proteins were purified by IMAC using Ni Sepharose 6 Fast Flow resin (GE Healthcare Life Sciences). The eluted sample was dialyzed against 10 mM HEPES, pH 7.5, 0.15 M NaCl. Crystallization and Data Collection. Purified APCDD1 protein was concen- trated to 12.5 mg/mL in the presence of 0.5% Flavobacterium meningosepticum endoglycosidase-F1 (Endo-F1) prepared as described previously (56) for in situ deglycosylation (58). Purified eMBP-APCDD1 protein was concentrated to 11 mg/ mL in the presence of 0.18 mM zinc acetate, 20 mM maltose, 0.5% Endo-F1, and 0.5% carboxypeptidase A/B (MilliporeSigma) for in situ deglycosylation and pro- teolysis (58). Using a Mosquito LCP crystallization robot (TTP Labtech), the protein samples were then subjected to sitting drop vapor diffusion crystallization trials in 96-well MRC 2 Well UVXPO plates (Hampton Research) by mixing 100 nL protein solution with 100 nL reservoir. APCDD1 crystal-form I crystallized in 0.7 M magne- sium formate, 0.1 M Bis-Tris propane, pH 7.0. APCDD1 crystal-form II crystallized in 0.1 M magnesium acetate, 0.1 M sodium citrate, pH 5.8, 14% polyethylene glycol (PEG) 5K MME. Crystals of eMBP-APCDD1 were grown and optimized in 0.2 M ammonium citrate, pH 7.0, 20% PEG 3350, 4% NDSB-256, 5% glycerol. For cryoprotection, crystals were transferred into a reservoir solution supple- mented with 70% Tacsimate™, pH 7.0 for crystal-form I, with 20% glycerol for crystal-form II, and with a 5% increase gradually to 15% glycerol for eMBP-AP- CDD1, and subsequently cryocooled in liquid nitrogen. X-ray diffraction data were collected at 100°K at the 17-ID-2 FMX beamline (63) using a beam size of 1 × 1.5 µm and an EIGER 16M detector (DECTRIS) at the National Synchrotron Light Source II (NSLS II), Brookhaven National Laboratory. Diffraction data from APCDD1 crystal-form I were indexed, integrated, and scaled using the autoPROC toolbox (64), coupled with XDS (65), POINTLESS (66), and AIMLESS (67). Diffraction data from APCDD1 crystal-form II and eMBP-APCDD1 were indexed, integrated, and scaled using the XIA2 system (68), coupled with DIALS (69, 70) and POINTLESS (66). Diffraction anisotropy was further corrected using STARANISO (71). A randomly selected subset of 5% of the diffraction data was used as a cross-validation dataset to calculate Rfree. Structure Determination and Refinement. The structure of eMBP-APCDD1 was determined by molecular replacement (MR) using PHASER (72) using the diffraction data scaled at the 30.0 to 2.5 Å resolution range without the anisotropy correction and using the MBP structure (PDB ID 3SET) as a template to obtain the initial phases. The resulting map contained three MBP copies that fit well into the electron density, whereas the electron density for APCDD1 was not interpretable. RoseTTAFold (28) was used to predict models of APCDD1, ABD1, and ABD2. All predicted models were superimposed to assess conserved core domains of ABD1 and ABD2 using the SSM algorithm of SUPERPOSE (73) in the CCP4 suite (74). Prior to MR, the values of Angstroms error estimates of the core domains were set to a constant value (=30 Å2) for the B-factors for all atoms. The second round of MR using PHASER (72) was conducted by fixing the position of the three MBP copies and using the core domains of ABD1 and ABD2 as templates to obtain the improved phases. The electron density was further improved after density modification with PARROT (75) and subsequently fed into BUCCANEER in the CCP4 suite (30, 74) for initial model building. The model of eMBP-APCDD1 was completed by manual building in COOT (76) and refinement was performed using REFMAC5 (77) and PHENIX Refine (78) with translation-libration-screw (TLS) parameterization. The resulting APCDD1 model from the eMBP-APCDD1 structure was used to determine the structures of APCDD1 crystal-form I and II by MR using PHASER (72). The subsequent model building and refinement were conducted using COOT (71) and PHENIX Refine (78) with TLS parameterization. Finally, the APCDD1 structures were built for crystal-form I (residues 50 to 473 for chain A and residues 50 to 468 for chain B), crystal-form II (residues 50 to 467 for chain A and residues 50 to 466 for chain B), and eMBP-APCDD1 (residues 47 to 468 for chain A, residues 47 to 469 for chain B, residues 47 to 472 for chain C, and residues 47 to 469 for chain D), except the regions where the electron densities were not interpretable for model building: crystal-from I (residues 389 to 391 and 406 to 408 for chain A; residues 193 to 195, 387 to 393, and 407 to 410 for 10 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org chain B), crystal-form II (residues 175 to 178 and 391 to 394 for chain B), and eMBP-APCDD1 (residues 387 to 393 and 409 to 411 for chain A, 387 to 393 and 406 to 410 for chain C, and 389 to 390 and 409 to 414 for chain D). MOLPROBITY (79) was used to validate the models. The crystallographic statistics are listed in SI Appendix, Table S1. Structure Analysis. Structure-based multiple sequence alignment was per- formed using Clustal Omega (53) and ESPript (80). Structure superposition was performed using the SSM algorithm of SUPERPOSE (73) in the CCP4 suite (74). The interior volume of the pocket was calculated using CASTp (29) with a 1.4 Å radius probe. Schematic 2D representations of protein and ligand interac- tions were generated using LigPlot+ (81). Searches for structure-based similar- ities to APCDD1, ABD1, and ABD2 were performed against the databases of the PDB and AlphaFold (36) using the DALI server (37). The computational model of APCDD1L based on APCDD1 was generated with Modeller (82). High-quality images of the molecular structures were generated with the PyMOL Molecular Graphic System (Version 2.5, Schrödinger, LLC). Schematic figures and other illustrations were prepared using GraphPad Prism (GraphPad Software, LLC) and Corel Draw (Corel Corporation). Structural biology applications used in this work were compiled and configured by SBGrid (83). Mass Spectrometry Analysis. To identify the bound ligand in APCDD1 structures, HPLC coupled with mass spectrometry (MS) was used. Briefly, purified APCDD1 protein solution or conditioned media without APCDD1 (serving as a control for the compounds initially present) was mixed with the extraction solution (methyl tert-bu- tyl ether/methanol/water in a ratio of 10:3:2.5) and the phases were separated. The organic phase was dried in a SpeedVac concentrator. The dried pellet was reconsti- tuted in a solution of n-butanol and methanol in a 1:1 ratio. The sample was then analyzed by HPLC/MS on an Ultimate 3000 UPLC (using an Accucore C30 column) and Q-Exactive Plus Orbitrap. Ligand identification was performed by comparison of the mass spectrum of the analyte with the library database at Cayman Chemical. AP-Based Binding Assay. For the AP fusion proteins (AP-APCDD1, APCDD1-AP, DKK1-AP, and AP-DKK1), HEK293T cells were grown in six-well plates and transfected with plasmids and PEI as described previously (27, 57). Conditioned media were collected 2 d posttransfection. For the biotinylated bait preparations, a 3:1:1 ratio of either LRP5-Avi or LRP6-Avi, MESD pHLsec (59), and pHLsec-BirA-ER (56) plasmids was transfected into HEK293T cells in the presence of 0.1 mM biotin (MilliporeSigma) and 4 mM valproic acid. Similarly, a 2.5:1.5:1 ratio of WNT7A, APCDD1-mV-Avi (or FZD4-mV-Avi, or empty vector, pLEXm), and pHLsec-BirA-ER (56) plasmids was trans- fected into HEK293T cells in the presence of 0.1 mM biotin and 4 mM valproic acid. The biotinylated baits were purified by the IMAC method (57) from 20 mL conditioned media collected 2 or 3 d posttransfection and eluted in 400 μL. IMAC-concentrated complexes (100 μL each) were immobilized on 96-well streptavidin-coated plates (Thermo Fisher Scientific) at 4 °C overnight. For the detergent addition assays, 1% DDM (Anatrace), 0.1% DDM (Anatrace), or 2.25% Triton X-114 (MilliporeSigma) were included in all steps from the 96-well plate immobilization to the washes. The wells were then washed three times with wash buffer [10 mM HEPES, pH 7.5, 0.15 M NaCl, 0.05% (w/v) Tween- 20] supplemented or not with the indicated detergents and incubated with a 10-fold dilution of bovine serum albumin (BSA) blocker buffer (Thermo Fisher Scientific 37525) in wash buffer for 1 h at 25 °C. The wells were washed with wash buffer with the indicated detergents and incubated with conditioned media containing AP probes (APCDD1-AP, AP-APCDD1, DKK-AP, or AP-DKK) or recombinant Nb-AP proteins at 4 °C overnight. The wells were subsequently washed three times with wash buffer with the indicated detergents and incu- bated with BluePhos phosphatase substrate solution (Kirkegaard and Perry Laboratories 50-88-00) to visualize the bound AP probes. The binding assays were performed twice. Biotinylated baits were immobilized on 96-well streptavidin-coated plates (Thermo Fisher Scientific) at 4 °C overnight. The wells were then washed three times with wash buffer [10 mM HEPES, pH 7.5, 0.15 M NaCl, 0.05% (w/v) Tween- 20] and incubated with a 10-fold dilution of BSA blocker buffer (Thermo Fisher Scientific 37525) in wash buffer for 1 h at 25 °C. The wells were washed with wash buffer and incubated with conditioned media containing AP probes (APCDD1-AP, AP-APCDD1, DKK-AP, or AP-DKK) or recombinant Nb-AP proteins at 4 °C overnight. The wells were subsequently washed three times with wash buffer and incubated with BluePhos phosphatase substrate solution (Kirkegaard and Perry Laboratories 50-88-00) to visualize the bound AP probes. Data, Materials, and Software Availability. X-ray structure data have been deposited in Protein Data Bank (8E0P, 8E0R, and 8E0W) for eMBP-APCDD1, APCDD1 crystal-form I and form II, respectively (84–86). ACKNOWLEDGMENTS. We thank Amir Rattner for comments on the manu- script, Randy J. Read (University of Cambridge) for discussion of molecular replacement, and Cayman Chemical (Ann Arbor, MI) for performing the liquid chromatography-mass spectrometry (LC-MS) analysis. This work was supported by the Howard Hughes Medical Institute and the National Eye Institute (NIH) (R01EY018637). S.B.G. was supported by the National Cancer Institute (NIH) (R01CA204345). T.-H.C. was supported by the Human Science Frontier Program Organization Fellowship (LT000130/2017-L) and the Howard Hughes Medical Institute. Work at the Highly Automated Macromolecular Crystallography (AMX) (17-ID-1) and Frontier Microfocusing Macromolecular Crystallography (FMX) (17-ID-2) beamlines was supported by the NIH, the National Institute of General Medical Sciences (P41GM111244), the U.S. Department of Energy (DOE) Office of Biological and Environmental Research (KP1605010), and the National Synchrotron Light Source II at Brookhaven National Laboratory, which is supported by the DOE Office of Basic Energy Sciences under contract DE-SC0012704 (KC0401040). Author affiliations: aDepartment of Molecular Biology and Genetics, Johns Hopkins University School of Medicine, Baltimore, MD 21205; bHHMI, Johns Hopkins University School of Medicine, Baltimore, MD 21205; cDepartment of Biophysics and Biophysical Chemistry, Johns Hopkins University School of Medicine, Baltimore, MD 21205; dDepartment of Medicine, Johns Hopkins University School of Medicine, Baltimore, MD 21205; eDepartment of Oncology, Johns Hopkins University School of Medicine, Baltimore, MD 21205; fDepartment of Neuroscience, Johns Hopkins University School of Medicine, Baltimore, MD 21205; and gWilmer Eye Institute, Johns Hopkins University School of Medicine, Baltimore, MD 21205 Author contributions: F.-L.H., T.-H.C., S.B.G., and J.N. designed research; F.-L.H. and T.-H.C. performed research; F.-L.H., T.-H.C., and J.N. analyzed data; and F.-L.H., T.-H.C., S.B.G., and J.N. wrote the paper. Reviewers: S.B., University of Missouri-Kansas City; and B.O.W., Van Andel Research Institute. 1. 2. 3. 4. 5. E. Y. Rim, H. Clevers, R. Nusse, The Wnt pathway: From signaling mechanisms to synthetic modulators. Annu. Rev. Biochem. 91, 571–598 (2022). A. Rattner et al., A family of secreted proteins contains homology to the cysteine-rich ligand-binding domain of frizzled receptors. Proc. Natl. Acad. Sci. U.S.A. 94, 2859–2863 (1997). J. C. Hsieh et al., A new secreted protein that binds to Wnt proteins and inhibits their activities. Nature 398, 431–436 (1999). A. Glinka et al., Dickkopf-1 is a member of a new family of secreted proteins and functions in head induction. Nature 391, 357–362 (1998). B. Mao et al., Kremen proteins are Dickkopf receptors that regulate Wnt/beta-catenin signalling. Nature 417, 664–667 (2002). 10. H. X. Hao et al., ZNRF3 promotes Wnt receptor turnover in an R-spondin-sensitive manner. Nature 485, 195–200 (2012). 11. B. K. Koo et al., Tumour suppressor RNF43 is a stem-cell E3 ligase that induces endocytosis of Wnt receptors. Nature 488, 665–669 (2012). 12. D. Yan et al., Elevated expression of axin2 and hnkd mRNA provides evidence that Wnt/ beta -catenin signaling is activated in human colon tumors. Proc. Natl. Acad. Sci. U.S.A. 98, 14973–14978 (2001). 13. M. Takahashi et al., Isolation of a novel human gene, APCDD1, as a direct target of the beta- Catenin/T-cell factor 4 complex with probable involvement in colorectal carcinogenesis. Cancer Res. 62, 5651–5656 (2002). 6. M. Semenov, K. Tamai, X. He, SOST is a ligand for LRP5/LRP6 and a Wnt signaling inhibitor. J. Biol. 14. M. van de Wetering et al., Prospective derivation of a living organoid biobank of colorectal cancer 7. 8. 9. Chem. 280, 26770–26775 (2005). S. Kakugawa et al., Notum deacylates Wnt proteins to suppress signalling activity. Nature 519, 187–192 (2015). X. Zhang et al., Notum is required for neural and head induction via Wnt deacylation, oxidation, and inactivation. Dev. Cell 32, 719–730 (2015). X. Zhang et al., Tiki1 is required for head formation via Wnt cleavage-oxidation and inactivation. Cell 149, 1565–1577 (2012). patients. Cell 161, 933–945 (2015). 15. L. Lin et al., Super-enhancer-associated MEIS1 promotes transcriptional dysregulation in Ewing sarcoma in co-operation with EWS-FLI1. Nucleic Acids Res. 47, 1255–1267 (2019). 16. D. Shiokawa et al., Slow-cycling cancer stem cells regulate progression and chemoresistance in colon cancer. Cancer Res. 80, 4451–4464 (2020). 17. Y. Shimomura et al., APCDD1 is a novel Wnt inhibitor mutated in hereditary hypotrichosis simplex. Nature 464, 1043–1047 (2010). PNAS 2023 Vol. 120 No. 20 e2217096120 https://doi.org/10.1073/pnas.2217096120 11 of 12 18. H. K. Lee et al., Apcdd1 stimulates oligodendrocyte differentiation after white matter injury. Glia 63, 1840–1849 (2015). 53. F. Sievers, D. G. Higgins, Clustal omega. Curr. Protoc. Bioinform. 48, 3 13 11-16 (2014). 54. N. Saitou, M. Nei, The neighbor-joining method: A new method for reconstructing phylogenetic 19. S.-G. Cho, APC downregulated 1 inhibits breast cancer cell invasion by inhibiting the canonical WNT trees. Mol. Biol. Evol. 4, 406–425 (1987). signaling pathway. Oncol. Lett. 2017, 4845–4852 (2017). 55. S. Kumar, G. Stecher, K. Tamura, MEGA7: Molecular evolutionary genetics analysis Version 7.0 for 20. M. Li, R. Cheng, Y. Zhuang, Z. Yao, A recurrent mutation in the APCDD1 gene responsible for bigger datasets. Mol. Biol. Evol. 33, 1870–1874 (2016). hereditary hypotrichosis simplex in a large Chinese family. Br. J. Dermatol. 167, 952–954 (2012). 56. T. H. Chang et al., Structure and functional properties of Norrin mimic Wnt for signalling with 21. J. Mazzoni et al., The Wnt inhibitor Apcdd1 coordinates vascular remodeling and barrier maturation Frizzled4, Lrp5/6, and proteoglycan. Elife 4, e06554 (2015). of retinal blood vessels. Neuron 96, 1055–1069.e1056 (2017). 22. S. Biswas et al., Mural Wnt/β-catenin signaling regulates Lama2 expression to promote neurovascular unit maturation. Development 149, dev200610 (2022). 57. F. L. Hsieh, T. H. Chang, Antibody display of cell surface receptor Tetraspanin12 and SARS-CoV-2 spike protein. bioRxiv (2021), 10.1101/2021.05.29.446300. 58. F. L. Hsieh et al., The structural basis for CD36 binding by the malaria parasite. Nat. Commun. 7, 23. S. P. Fancy et al., Dysregulation of the Wnt pathway inhibits timely myelination and remyelination in 12837 (2016). the mammalian CNS. Genes Dev. 23, 1571–1585 (2009). 59. S. Chen et al., Structural and functional studies of LRP6 ectodomain reveal a platform for Wnt 24. J. E. Lengfeld et al., Endothelial Wnt/beta-catenin signaling reduces immune cell infiltration in signaling. Dev. Cell 21, 848–861 (2011). multiple sclerosis. Proc. Natl. Acad. Sci. U.S.A. 114, E1168–E1177 (2017). 60. H. Gotzke et al., The ALFA-tag is a highly versatile tool for nanobody-based bioscience applications. 25. P. Kang et al., Sox9 and NFIA coordinate a transcriptional regulatory cascade during the initiation of Nat. Commun. 10, 4403 (2019). gliogenesis. Neuron 74, 79–94 (2012). 61. A. R. Aricescu, W. Lu, E. Y. Jones, A time- and cost-efficient system for high-level protein production in 26. N. K. H. Yiew et al., A novel role for the Wnt inhibitor APCDD1 in adipocyte differentiation: mammalian cells. Acta Crystallogr. D. Biol. Crystallogr. 62, 1243–1250 (2006). Implications for diet-induced obesity. J. Biol. Chem. 292, 6312–6324 (2017). 62. V. T. Chang et al., Glycoprotein structural genomics: Solving the glycosylation problem. Structure 15, 27. T. H. Chang, F. L. Hsieh, P. M. Smallwood, S. B. Gabelli, J. Nathans, Structure of the RECK CC domain, 267–273 (2007). an evolutionary anomaly. Proc. Natl. Acad. Sci. U.S.A. 117, 15104–15111 (2020). 63. D. K. Schneider et al., FMX - the frontier microfocusing macromolecular crystallography 28. M. Baek et al., Accurate prediction of protein structures and interactions using a three-track neural network. Science 373, 871–876 (2021). beamline at the National Synchrotron Light Source II. J. Synchrotron Radiat. 28, 650–665 (2021). 29. W. Tian, C. Chen, X. Lei, J. Zhao, J. Liang, CASTp 3.0: Computed atlas of surface topography of 64. C. Vonrhein et al., Data processing and analysis with the autoPROC toolbox. Acta Crystallogr. D. Biol. proteins. Nucleic Acids Res. 46, W363–W367 (2018). Crystallogr. 67, 293–302 (2011). 30. K. Cowtan, The Buccaneer software for automated model building. 1. Tracing protein chains. Acta Crystallogr. D. Biol. Crystallogr. 62, 1002–1011 (2006). 65. W. Kabsch, Xds. Acta Crystallogr. D. Biol. Crystallogr. 66, 125–132 (2010). 66. P. Evans, Scaling and assessment of data quality. Acta Crystallogr. D. Biol. Crystallogr. 62, 72–82 31. K. Willert et al., Wnt proteins are lipid-modified and can act as stem cell growth factors. Nature 423, (2006). 448–452 (2003). 67. P. R. Evans, G. N. Murshudov, How good are my data and what is the resolution? Acta Crystallogr. 32. R. Takada et al., Monounsaturated fatty acid modification of Wnt protein: its role in Wnt secretion. D. Biol. Crystallogr. 69, 1204–1214 (2013). Dev. Cell 11, 791–801 (2006). 68. G. Winter, xia2: An expert system for macromolecular crystallography data reduction. J. Appl. 33. C. Y. Janda, D. Waghray, A. M. Levin, C. Thomas, K. C. Garcia, Structural basis of Wnt recognition by Crystallogr. 43, 186–190 (2010). Frizzled. Science 337, 59–64 (2012). 34. R. Nygaard et al., Structural basis of WLS/Evi-mediated Wnt transport and secretion. Cell 184, 35. 194–206.e114 (2021). I. J. McGough et al., Glypicans shield the Wnt lipid moiety to enable signalling at a distance. Nature 585, 85–90 (2020). 36. J. Jumper et al., Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). 37. L. Holm, Using dali for protein structure comparison. Methods Mol. Biol. 2112, 29–42 (2020). 38. D. R. Flower, The lipocalin protein family: Structure and function. Biochem. J. 318, 1–14 (1996). 39. A. Rattner, Y. Wang, J. Nathans, Signaling pathways in neurovascular development. Annu. Rev. Neurosci. 45, 87–108 (2022). 69. J. Beilsten-Edmands et al., Scaling diffraction data in the DIALS software package: Algorithms and new approaches for multi-crystal scaling. Acta Crystallogr. D. Struct. Biol. 76, 385–399 (2020). 70. G. Winter et al., DIALS: Implementation and evaluation of a new integration package. Acta Crystallogr. D. Struct. Biol. 74, 85–97 (2018). I. J. F. Tickle et al., STARANISO (Global Phasing Ltd., Cambridge, United Kingdom, 2021). 71. 72. A. J. McCoy et al., Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007). 73. E. Krissinel, K. Henrick, Secondary-structure matching (SSM), a new tool for fast protein structure alignment in three dimensions. Acta Crystallogr. D. Biol. Crystallogr. 60, 2256–2268 (2004). 74. M. D. Winn et al., Overview of the CCP4 suite and current developments. Acta Crystallogr. D. Biol. 40. M. Vallon et al., A RECK-WNT7 receptor-ligand interaction enables isoform-specific regulation of Wnt Crystallogr. 67, 235–242 (2011). bioavailability. Cell Rep. 25, 339–349.e339 (2018). 75. K. Cowtan, Recent developments in classical density modification. Acta Crystallogr. D. Biol. 41. H. Hirai, K. Matoba, E. Mihara, T. Arimori, J. Takagi, Crystal structure of a mammalian Wnt-frizzled Crystallogr. 66, 470–478 (2010). complex. Nat. Struct. Mol. Biol. 26, 372–379 (2019). 76. P. Emsley, B. Lohkamp, W. G. Scott, K. Cowtan, Features and development of Coot. Acta Crystallogr. D. 42. Z. J. DeBruine et al., Wnt5a promotes Frizzled-4 signalosome assembly by stabilizing cysteine-rich Biol. Crystallogr. 66, 486–501 (2010). domain dimerization. Genes Dev. 31, 916–926 (2017). 77. G. N. Murshudov et al., REFMAC5 for the refinement of macromolecular crystal structures. Acta 43. A. H. Nile, S. Mukund, K. Stanger, W. Wang, R. N. Hannoush, Unsaturated fatty acyl recognition by Crystallogr. D. Biol. Crystallogr. 67, 355–367 (2011). Frizzled receptors mediates dimerization upon Wnt ligand binding. Proc. Natl. Acad. Sci. U.S.A. 114, 4147–4152 (2017). 78. D. Liebschner et al., Macromolecular structure determination using X-rays, neutrons and electrons: recent developments in Phenix. Acta Crystallogr. D. Struct. Biol. 75, 861–877 (2019). 44. T. H. Chang, F. L. Hsieh, K. Harlos, E. Y. Jones, Structural insights into Frizzled assembly by acylated 79. C. J. Williams et al., MolProbity: More and better reference data for improved all-atom structure Wnt and frizzled connector domain. bioRxiv (2022), 10.1101/2022.09.29.510206. validation. Protein Sci. 27, 293–315 (2018). 45. G. S. Richards, B. M. Degnan, The dawn of developmental signaling in the metazoa. Cold Spring 80. X. Robert, P. Gouet, Deciphering key features in protein structures with the new ENDscript server. Harb. Symp. Quant. Biol. 74, 81–90 (2009). Nucleic Acids Res. 42, W320–W324 (2014). 46. A. Vonica et al., Apcdd1 is a dual BMP/Wnt inhibitor in the developing nervous system and skin. 81. R. A. Laskowski, M. B. Swindells, LigPlot+: Multiple ligand-protein interaction diagrams for drug Dev. Biol. 464, 71–87 (2020). discovery. J. Chem. Inf. Model 51, 2778–2786 (2011). 47. K. A. Mulligan et al., Secreted wingless-interacting molecule (Swim) promotes long-range signaling 82. N. Eswar et al., Comparative protein structure modeling using Modeller. Curr. Protoc. Bioinform. by maintaining wingless solubility. Proc. Natl. Acad. Sci. U.S.A. 109, 370–377 (2012). (2006), 10.1002/0471250953.bi0506s15. 48. H. Hasegawa, L. Holm, Advances and pitfalls of protein structural alignment. Curr. Opin. Struct. Biol. 19, 341–348 (2009). 49. F. Gabler et al., Protein sequence analysis using the MPI bioinformatics toolkit. Curr. Protoc. Bioinform. 72, e108 (2020). 83. A. Morin et al., Collaboration gets the most out of software. Elife 2, e01456 (2013). 84. F. L. Hsieh, T. H. Chang, S. B. Gabelli, J. Nathans, Crystal structure of mouse APCDD1 in fusion with engineered MBP. Worldwide Protein Data Bank (wwPDB). https://doi.org/10.2210/pdb8E0P/pdb. Deposited 9 August 2022. 50. D. T. Jones, W. R. Taylor, J. M. Thornton, A model recognition approach to the prediction of all-helical 85. F. L. Hsieh, T. H. Chang, S. B. Gabelli, J. Nathans, Crystal structure of mouse APCDD1 in P21 space membrane protein structure and topology. Biochemistry 33, 3038–3049 (1994). 51. L. J. McGuffin, K. Bryson, D. T. Jones, The PSIPRED protein structure prediction server. Bioinformatics 16, 404–405 (2000). 52. L. Zimmermann et al., A completely reimplemented MPI bioinformatics toolkit with a new HHpred server at its core. J. Mol. Biol. 430, 2237–2243 (2018). group. Worldwide Protein Data Bank (wwPDB). https://doi.org/10.2210/pdb8E0R/pdb. Deposited 9 August 2022. 86. F. L. Hsieh, T. H. Chang, S. B. Gabelli, J. Nathans, Crystal structure of mouse APCDD1 in P1 space group. Worldwide Protein Data Bank (wwPDB). https://doi.org/10.2210/pdb8E0W/pdb. Deposited 9 August 2022. 12 of 12 https://doi.org/10.1073/pnas.2217096120 pnas.org
10.1016_j.isci.2022.105093	iScience ll OPEN ACCESS Article NOX-like ROS production by glutathione reductase Julia M. Diaz, Xinying Shi [email protected] Highlights Glutathione reductase expressed from a microbial phototroph is found to produce ROS This promiscuous reaction shows similar kinetics and inhibition as NOX-derived ROS The GR pathway of ROS production has cellular beneﬁts under physiological stress GR may function similarly to NOX in other taxa, providing metabolic versatility Diaz & Shi, iScience 25, 105093 October 21, 2022 ª 2022 The Authors. https://doi.org/10.1016/ j.isci.2022.105093 iScience ll OPEN ACCESS Article NOX-like ROS production by glutathione reductase Julia M. Diaz1,3,* and Xinying Shi1,2 SUMMARY In organisms from bacteria to mammals, NADPH oxidase (NOX) catalyzes the production of beneﬁcial reactive oxygen species (ROS) such as superoxide (cid:1)). However, our previous research implicated glutathione reductase (GR), a (O2 canonical antioxidant enzyme, as a source of extracellular superoxide in the ma- rine diatom Thalassiosira oceanica. Here, we expressed and characterized the two GR isoforms of T. oceanica. Both coupled the oxidation of NADPH, the native electron donor, to oxygen reduction, giving rise to superoxide in the absence of glutathione disulﬁde, the native electron acceptor. Superoxide production by ToGR1 exhibited similar kinetics as representative NOX enzymes, and inhibition assays agreed with prior organismal studies, supporting a physiological role. ToGR is similar to GR from human, yeast, and bacteria, suggesting that NOX- like ROS production by GR could be widespread. Yet unlike NOX, GR-mediated ROS production is independent of iron, which may provide an advantageous way of making ROS under micronutrient stress. INTRODUCTION Superoxide is a reactive oxygen species (ROS) formed by the one-electron reduction of oxygen. All aerobic forms of life generate superoxide and other ROS, which can accumulate to toxic levels under adverse con- ditions. Yet physiological levels of ROS serve a broad diversity of beneﬁcial signaling roles, as well. In or- ganisms from bacteria to humans, the beneﬁcial roles of ROS include growth regulation (Hansel et al., 2019; Huang et al., 2019; Patel et al., 2018; Rossi et al., 2017), innate immunity (Bardaweel et al., 2018; Wein- berger, 2007), and defense (Armoza-Zvuloni et al., 2016; Diaz and Plummer, 2018; Minibayeva et al., 2009). In the ocean, the production of extracellular superoxide by microbes is widespread (Diaz et al., 2013; Diaz and Plummer, 2018; Hansel and Diaz, 2021; Sutherland et al., 2019), but the mechanisms and physiological functions of this microbial ROS production are not completely understood. Addressing this knowledge gap will improve our basic understanding of microbial ecosystem services and ocean health under continued global change. Phytoplankton are photosynthetic microbes in the ocean that form the base of marine food webs, contribute to global climate by absorbing the greenhouse gas carbon dioxide, and produce at least half of the world’s oxygen supply. In a previous study, we investigated the mechanism and role of extracellular superoxide production by the phytoplankton species Thalassiosira oceanica, which represents the diatom lineage of marine phytoplankton (Malviya et al., 2016), a dominant group responsible for approximately In 20% of global photosynthesis (Falkowski et al., 1998; Field et al., 1998; Nelson et al., 1995). T. oceanica, we found that NADPH-dependent extracellular superoxide production is vital to photophysi- ology (Diaz et al., 2019). An expected source of this ROS production would be the enzyme NADPH oxidase (NOX), which couples the oxidation of NADPH to the generation of superoxide. Indeed, since its discovery in human phagocytes, NOX has been found in bacteria (Hajjar et al., 2017; Magnani et al., 2017) and every major eukaryotic lineage (Bedard et al., 2007) and has been implicated in extracellular superoxide produc- tion by several phytoplankton taxa (Hansel and Diaz, 2021). Yet surprisingly, our previous results revealed an unexpected enzyme as a source of extracellular superoxide production by T. oceanica: glutathione reductase (GR) (Diaz et al., 2019). GR is a highly conserved enzyme across the tree of life. It typically promotes a reducing environment by coupling the oxidation of NADPH to the reduction of glutathione disulﬁde (GSSG), yielding reduced gluta- thione (GSH). For example, GSH directly eliminates some ROS, with the exception of superoxide (Winter- bourn, 2016), acts as a thiol buffer maintaining proteins in their reduced state, and serves as a substrate of glutathione peroxidase, which degrades the ROS hydrogen peroxide (H2O2). GR can donate electrons to a 1Geosciences Research Division, Scripps Institution of Oceanography, University of California San Diego, La Jolla, CA 92093, USA 2Present address: Skaggs School of Pharmacy and Pharmaceutical Science, University of California San Diego, La Jolla, CA 92093 USA 3Lead contact Correspondence: [email protected] https://doi.org/10.1016/j.isci. 2022.105093 iScience 25, 105093, October 21, 2022 ª 2022 The Authors. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 1 ll OPEN ACCESS iScience Article wide variety of acceptors besides the native substrate GSSG (Carlberg and Mannervik, 1986; Ce´ nas et al., 1989; Nordman et al., 2003; Paulı´kova´ and Berczeliova´ , 2005; Pinto et al., 1985), including oxygen (Angiulli et al., 2015; Korge et al., 2015). ROS production by GR has been reported previously in plant (Asada, 1999), yeast (Massey et al., 1969), bovine (Liochev and Fridovich, 1992), and human (Coppo et al., 2022; Korge et al., 2015), but GR is not widely recognized as a physiological ROS source. Here, we characterized the capacity of T. oceanica GR (ToGR) to produce superoxide in vitro in order to evaluate the potential physiological relevance of this promiscuous activity. Two putative GR isoforms exist in the T. oceanica genome: ToGR1 and ToGR2 (Diaz et al., 2019; Lommer et al., 2012). Both ToGR proteins possess unique (cid:3)130 amino acid N-terminal putative localization domains that appear absent in other GRs, yet the func- tional ToGR proteins show high sequence similarity to each other and to other GRs from human, yeast, and bac- teria, including several highly conserved sites of known catalytic or structural importance (Diaz et al., 2019). In addition, GRs share several broad features with NOX enzymes, including the ability to bind and oxidize NADPH and to transfer electrons using a ﬂavin adenine dinucleotide (FAD) cofactor in the active site. Overall, results revealed that both ToGRs produce superoxide, and that ROS production by ToGR1 likely has a physio- logical role, with a catalytic performance that is similar to representative NOX enzymes. RESULTS AND DISCUSSION ToGR1 and ToGR2 are glutathione reductases We overexpressed recombinant ToGR1 and ToGR2 in Escherichia coli with the N-terminal localization do- mains removed (see STAR Methods), which yielded proteins enriched in the (cid:3)55 kDa size fraction under denaturing conditions (Figure 1A; Figure S1). This result is consistent with the size of previously character- ized GR proteins, which are homodimeric enzymes of (cid:3)110 kDa (Deponte, 2013) that would disassociate into equivalent protein subunits no more than (cid:3)55 kDa each. Furthermore, absorption spectra of both pro- teins showed peaks at 370–375 and 460 nm in oxidized form or 345 nm and 430–435 nm in reduced form (Figures 1B and 1C), which is characteristic of GR and other ﬂavoproteins (Ji et al., 2015). Consistent with the predicted native function of ToGR1 and ToGR2, both puriﬁed proteins coupled the oxidation of NADPH to the reduction of GSSG (Figures 1D, 1E and S2A). Kinetic analysis of the native ac- tivity revealed that ToGR1 and ToGR2 are similar to previously characterized GRs. For example, the turn- (cid:1)1 over rate (kcat) of ToGR1 ((cid:3)12000 min (Can et al., 2010; Ji et al., 2015; Mapson and Isherwood, 1963; Massey and Williams, 1965; Savvides and Karplus, 1996; Worthington and Rosemeyer, 1976). ToGR2 has a slower turnover rate than ToGR1 (cid:1)1), but it is in line with previously reported values of 8295 (Lo´ pez-Barea and Lee, 1979) ((cid:3)7500 min and (cid:3)3000 min (cid:1)1) is similar to most GRs, which ranges from 12500 to 16,000 min (cid:1)1 (Hakam and Simon, 2000). The Km values for ToGR native activity are also similar to previously characterized GRs, which typically range from 4 to 15 mM NADPH (Can et al., 2010; Hakam and Simon, 2000; Ji et al., 2015; Lo´ pez-Barea and Lee, 1979; Mapson and Isherwood, 1963; Massey and Williams, 1965; Worthington and Rosemeyer, 1976) (Ta- ble 1). Compared to ToGR1, ToGR2 was more readily saturated with NADPH by nearly a factor of three, (cid:1)1) as ToGR1 (Table 1). ToGR1 and ToGR2 speciﬁcity con- making ToGR2 almost twice as speciﬁc (kcat Km (cid:1)1) (Can stants for the native activity (Table 1) are also in agreement with typical GRs (1.73107–6.63107 M et al., 2010; Ji et al., 2015; Lo´ pez-Barea and Lee, 1979; Mapson and Isherwood, 1963; Massey and Williams, 1965; Worthington and Rosemeyer, 1976). (cid:1)1s ToGR1 and ToGR2 catalyze NADPH-dependent superoxide production In the absence of GSSG, the native electron acceptor, both ToGR1 and ToGR2 still oxidized NADPH, as observed by the decline in NADPH absorbance at 340 nm over time (Figures 2A and 2B). To test whether this promiscuous NADPH oxidation was coupled to superoxide production, we used the probe nitroblue tetrazolium (NBT), which is reduced to the chromogenic product monoformazan (MF+) in the presence of superoxide. Consistent with superoxide production, MF+ accumulated over time, as indicated by the in- crease in absorbance at 530 nm (Figures 2C and 2D). Superoxide production is proportional to the amount of MF+ production inhibited by superoxide dismut- ase (SOD), which selectively degrades superoxide (Figures 2C–2F; Figure S2B). At most, we found that 50% and 80% of MF+ production by ToGR1 and ToGR2, respectively, were due to reaction of NBT with super- oxide. Increasing the concentration of SOD did not alter this ratio (Figure S3), suggesting that the 2 iScience 25, 105093, October 21, 2022 iScience Article ll OPEN ACCESS A D B C E Figure 1. Protein properties and native activity (A) SDS-PAGE analysis of protein fractions obtained during the puriﬁcation of recombinant ToGR1 (see STAR Methods). Arrow indicates (cid:3)55 kDa. (B and C) Absorption spectra of ToGR1 (B) and ToGR2 (C) in 100 mM phosphate buffer (1 mM EDTA, pH 8.0). Proteins (10 mM) were reduced with DTT (0.1 mM) or oxidized with GSSG (1 mM). (D and E) Michaelis-Menten curves of NADPH oxidation (D) coupled to glutathione disulﬁde (GSSG) reduction to glutathione (GSH) (E) by ToGR1 and ToGR2 (0.045 nM enzyme concentrations) in 100 mM phosphate buffer (1 mM EDTA, pH 8.0). Rate data represent the avg. G std. dev. of triplicate measurements, where error bars smaller than the data symbol are not visible. remainder of MF+ production was due to direct enzymatic reduction of NBT. A similar amount of superox- ide-mediated reduction of the probe cytochrome c was reported in a study of bacterial NOX (Hajjar et al., 2017). Moreover, no MF+ formation occurred in control reactions lacking ToGR or NADPH, indicating that superoxide production was driven by NADPH-dependent enzyme activity (Figures 2C and 2D). ToGR1 is more specialized for superoxide production than ToGR2 Based on the concentrations obtained from Figures 2A to 2D, we calculated an MF+:NADPH stoichiometry of 0.70–0.98 for ToGR1 (95% conﬁdence interval, n = 43), which approaches the ideal value of 1 (Bielski et al., 1980). However, ToGR2 only produced 0.04–0.10 molecules of MF+ for every molecule of NADPH oxidized (95% conﬁdence interval, n = 18), which was signiﬁcantly less than ToGR1 (p < 0.0001, Student’s t test). The remainder of NADPH oxidation could potentially be accounted for by the production of other reduced products, including hydrogen peroxide (Korge et al., 2015) and water (Angiulli et al., 2015), which do not reduce NBT, but can arise from NADPH-dependent oxygen reduction by GR. This result suggests that ToGR1 is more specialized for superoxide production than ToGR2. Promiscuous NADPH oxidation and superoxide production followed Michaelis-Menten kinetics (Figures 2G and 2H). Overall, kinetic parameters revealed that the native GR function was superior to the promiscuous activity in both enzymes (Figure 3; Table 1). Yet promiscuous superoxide production (cid:1)1), (NADPH oxidation) by ToGR1 was (cid:3)20 ((cid:3)8) times faster (kcat), (cid:3)70 ((cid:3)30) times more speciﬁc (kcat Km iScience 25, 105093, October 21, 2022 3 ll OPEN ACCESS iScience Article Table 1. Kinetic parameters of glutathione reductase activity and superoxide production by ToGR1 and ToGR2 in 100 mM phosphate buffer (pH 8.0) Enzyme Activity ToGR1 Native Substrate or product NADPH GSH Promiscuous NADPH ToGR2 Native (cid:1) O2 NADPH GSH Promiscuous NADPH (cid:1) O2 kcat (cid:1)1 (min(cid:1)1) Km (mM) kcat Km, (cid:1)1 (M(cid:1)1 s(cid:1)1) a kcat/kuncat 11,707 G 671 11,765 G 676 12.9 G 1.0 10.0 G 0.8 7692 G 312 7273 G 243 1.66 G 0.13 0.42 G 0.11 11.7 G 1.6 13.0 G 1.3 23.3 G 2.8 47.5 G 4.1 4.4 G 0.5 5.0 G 0.4 82.7 G 7.3 210 G 98 (1.7 G 0.1)3107 (1.5 G 0.1)3107 9626 G 979 3109 G 289 (3.0 G 0.3)3107 (2.5 G 0.1)3107 341 G 47 42 G 11 107.7 107.0 104.7 104.1 107.5 106.8 103.9 102.7 Data represent the avg. G std. err. of n = 3 (native activities), n = 8 (ToGR1 promiscuous) or n = 5 observations (ToGR2 pro- miscuous). All Km values are for NADPH. aValues for kuncat are listed in Table S2. and exhibited Km values that were (cid:3)80% (70%) lower than ToGR2 (Figure 3; Table 1). Furthermore, both enzymes enhanced the uncatalyzed rates of NADPH oxidation and superoxide production (kuncat), but (cid:1)1) the effect of ToGR1 was more pronounced than ToGR2 by about an order of magnitude (kcat kuncat (Table 1). The differences in activity between ToGR1 and ToGR2 likely come down to three amino acid substitutions that distinguish the two functional protein sequences (Diaz et al., 2019). Two of these substitutions occur within known regions of structural or catalytic importance: the C-terminal interface-binding domain that forms the functional homodimer, and a ﬂavin adenine dinucleotide (FAD)-binding domain (Diaz et al., 2019). However, the potential inﬂuence of these substitutions on enzyme activity remains unclear. Superoxide production by ToGR1 is physiologically relevant ToGR1 may be almost as effective at catalyzing the NADPH-dependent production of ROS as some NOX en- (cid:1)1) is comparable to zymes (Figure 3). For example, the rate (kcat) of superoxide production by ToGR1 (10 min (cid:1)1) (Wu et al., 2021). rates of H2O2 production by the NOX family protein dual oxidase, or DUOX (6–40 min (cid:1)1) is only 1.3 (cid:1)1) by ToGR1 (3109 M Furthermore, the speciﬁcity constant of superoxide production (kcat Km (cid:1)1). Similarly, to 19 times lower than the speciﬁcity constant of H2O2 production by DUOX (4100–58000 M the Km of NADPH oxidation by ToGR1 (23.3 mM) is (cid:3)2.5-fold enhanced relative to cyanobacterial NOX5 (cid:1)1) is only (cid:3)4 (58.6 mM), and the speciﬁcity constant of promiscuous NADPH oxidation by ToGR1 (9626 M (cid:1)1) (Magnani et al., 2017). These results are based on the activities of pu- times lower than NOX5 (37,000 M riﬁed enzymes in vitro, which are likely to be different under physiological conditions. Nonetheless, these com- parisons illustrate that as a source of ROS production, ToGR1 has the potential to perform similarly to represen- tative NOX enzymes, which suggests that this promiscuous activity is physiologically relevant. (cid:1)1s (cid:1)1s (cid:1)1s (cid:1)1s In addition to kinetic analyses, we also tested the effect of several chemical compounds on ToGR-mediated su- peroxide production. Some compounds eliminate in vivo extracellular superoxide production by T. oceanica and other phytoplankton, including GSSG (Diaz et al., 2019) and diphenyleneiodonium (DPI) (Anderson et al., 2016; Diaz et al., 2019; Kustka et al., 2005; Laohavisit et al., 2015; Park et al., 2009; Saragosti et al., 2010), whereas other compounds like formaldehyde do not (Schneider et al., 2016). We found that these compounds had iden- tical effects on superoxide production by both ToGRs, but especially ToGR1, consistent with a physiological role. For example, DPI is an inhibitor of NOX enzymes and is commonly used to evaluate the potential involvement of NOX in ROS production, yet DPI can target ﬂavoenzymes like GR, as well (Reis et al., 2020). Indeed, the appli- cation of DPI resulted in a concentration-dependent inhibition of superoxide production (Figure 4C) and NADPH oxidation (Figure S4) by ToGR1 and ToGR2. ToGR1 was more sensitive to DPI inhibition than ToGR2, with IC50 values that were (cid:3)two orders of magnitude lower (Table S1). These results conﬁrm that DPI cannot be used to distinguish between NOX and GR as potential sources of ROS production in vivo. A previous study suggested that extracellular superoxide production by Thalassiosira spp. occurs by a passive or non-enzymatic photochemical process based on the ﬁnding that extracellular superoxide 4 iScience 25, 105093, October 21, 2022 ll OPEN ACCESS iScience Article A C E B D F G H Figure 2. NADPH-dependent superoxide production (A–F) Absorbance of NADPH (60 mM initial concentration) at 340 nm, (C,D) simultaneous absorbance of monoformazan (MF+) at 530 nm, and (E,F) rates of MF+ production with or without superoxide dismutase (SOD). Reactions were with ToGR1 (A,C,E), ToGR2 (B,D,F), or no protein controls. (G and H) Michaelis-Menten curves of NADPH oxidation (G) coupled to the production of superoxide (H). Data reﬂect single representative trials (A–D) or the avg. G std. dev. of triplicate measurements (E–H), where error bars smaller than the data symbol are not visible. All reactions were in 100 mM phosphate buffer (1 mM EDTA, pH 8.0) in the absence of GSSG. ToGR1 (40 nM), ToGR2 (200 nM). production could not be completely quenched by using formaldehyde to kill the cells (Schneider et al., 2016). However, our results show that formaldehyde failed to eliminate superoxide production by either ToGR1 or ToGR2 (Figure 4D). Therefore, we suggest that residual enzyme activity in dead cells, or en- zymes interacting directly with the formaldehyde, could create the effect observed in the study by Schneider et al. In fact, formaldehyde decreased superoxide production by ToGR1 without completely abolishing it, similar to previous results from T. oceanica cells (Schneider et al., 2016). This result is consistent with ToGR1 as an in vivo source of superoxide in T. oceanica. On the other hand, formaldehyde stimulated superoxide production by ToGR2, which was not seen in vivo (Schneider et al., 2016). This disagreement suggests that ToGR2 may not be a major source of superoxide in T. oceanica in vivo. (2016). Based on the observation that GSSG inhibits extracellular superoxide production by T. oceanica, we hypothesized previously that GR reduces oxygen and gives rise to superoxide at the GSSG binding site of the enzyme (Diaz et al., 2019). In agreement, our ﬁndings suggest that GSSG is a competitive inhibitor of superoxide production by ToGR1 (Figures 4A and 4B). Indeed, the rate constant of super- (cid:1)1, n = 8) or presence (12.2 G oxide production (kcat) was similar in the absence (10.0 G 2.4 min (cid:1)1, n = 2) of GSSG, while the Km of superoxide production (47.5 G 11.4 mM, n = 8) increased 0.1 min by approximately 3-fold (140.7 G 33.3 mM, n = 2) in the presence of GSSG (avg. G std. dev.), consis- tent with competitive inhibition. Rates of superoxide production by ToGR2 were too low to allow a similar analysis. A major factor controlling the ability of GR to produce superoxide under physiological conditions is GSSG. GR-mediated production of ROS is favored at low concentrations of GSSG. This process has been called ROS ‘‘spillover,’’ which may be triggered by reductive stress (high GSH:GSSG and NADPH:NADP+) and has a proposed role in redox signaling and oxidative injury in mitochondria (Korge iScience 25, 105093, October 21, 2022 5 ll OPEN ACCESS A B iScience Article Figure 3. Kinetic parameters (A and B). ToGR1 and ToGR2 catalyze the oxidation of NADPH (A) and the production of glutathione or superoxide (B) in the presence (diamonds) or absence (stars) of glutathione disulﬁde, respectively. NOX data (circles) for NADPH oxidation (Magnani et al., 2017) and ROS production (Wu et al., 2021) are provided for reference. et al., 2015). Furthermore, GR-mediated superoxide production has been implicated in multiple human disease traits (Coppo et al., 2022). While these examples highlight the harmful potential of ROS, we pre- viously found that the production of extracellular superoxide by ToGR may beneﬁt the photophysiolog- ical health of T. oceanica by alleviating reductive stress that occurs with excessive light exposure (Diaz et al., 2019). In phototrophs, extracellular superoxide production may help alleviate reductive stress that occurs under other conditions as well, including carbon dioxide limitation (Yuasa et al., 2020b) or nutrient deﬁciency (Yuasa et al., 2020a). Conclusions In a previous study, we identiﬁed GR as a potential source of extracellular superoxide production by the marine diatom T. oceanica (Diaz et al., 2019). Here, we expressed, puriﬁed, and characterized the two ToGR isoforms to assess their physiological relevance. Our results conﬁrmed that both enzymes are GRs capable of superoxide production, yet results from kinetic analyses and inhibitor assays point to ToGR1 as the likelier source of GR- mediated extracellular superoxide production in vivo. These ﬁndings support the view that ToGR1 may be a multifunctional ROS-generating enzyme whose activity is predominantly controlled by the presence or absence of GSSG, the native electron acceptor. A similar characterization has also been proposed for human GR (Korge et al., 2015). Indeed, given the reports of ROS production by multiple GRs (Asada, 1999; Coppo et al., 2022; Korge et al., 2015; Liochev and Fridovich, 1992; Massey et al., 1969) and the high sequence similarity of several prototypical GRs to ToGR (Diaz et al., 2019), NOX-like ROS production by GR should be considered in organisms that are taxonomically widespread beyond T. oceanica. GR-mediated ROS production would not necessarily rule out a role for NOX. For example, the T. oceanica genome encodes several putative NOX homologs (Diaz et al., 2019), suggesting the coexistence of at least two broad mechanisms for extracellular superoxide production in the same microorganism. This metabolic versatility may provide an adaptive advantage. For instance, NOX proteins are heme-dependent enzymes with a high iron requirement, whereas GR does not depend on iron or any other metal. Iron scarcity con- strains the growth and productivity of marine phytoplankton across wide areas of the ocean (Moore et al., 2013), and T. oceanica shows a remarkable degree of metabolic ﬂexibility in dealing with chronically low levels of iron (Lommer et al., 2012). We speculate that the substitution of NOX by GR as a source of ROS production may furnish a low-iron beneﬁt to T. oceanica, which may also be relevant to other organisms due to the broad role of iron in diverse biological processes. This metal-dependent regulation could potentially help explain why GR would switch from its canonical antioxidant function to a paradoxically pro-oxidant, ROS-generating role. 6 iScience 25, 105093, October 21, 2022 iScience Article A C B D ll OPEN ACCESS Figure 4. Inhibition of superoxide production (A–D). Superoxide production in the presence of (A,B) glutathione disulﬁde (GSSG), (C) diphenylene iodonium (DPI), or (D) after treatment of ToGR with formaldehyde. In panels c and d, data represent the avg. G std. dev. of triplicate observations. Stars depict signiﬁcant difference from the control condition (* = p < 0.001; *** = p < 0.0001, Student’s t test). Limitations of the study This study presents enzyme activity results from pure enzymes in vitro, which provide insights into the po- tential activity of these enzymes in vivo, as discussed above. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d KEY RESOURCES TABLE d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d METHOD DETAILS B Cloning, protein expression, and puriﬁcation B Protein spectroscopy B Enzyme activity assays B Formaldehyde treatment of ToGR B Numerical modeling d QUANTIFICATION AND STATISTICAL ANALYSIS SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.isci.2022.105093. ACKNOWLEDGMENTS The authors thank the Simpson Joseph lab for help with protein prep. This work was supported by grants to J.M.D. from the Simons Foundation (678537) and the Alfred P. Sloan Foundation (FG-2019-12550). AUTHOR CONTRIBUTIONS J.M.D. conceived and oversaw the study. X.S. designed experiments and performed analyses. J.M.D. wrote the paper with X.S. iScience 25, 105093, October 21, 2022 7 ll OPEN ACCESS iScience Article DECLARATION OF INTERESTS The authors declare that they have no conﬂicts of interest with the contents of this article. INCLUSION AND DIVERSITY We support inclusive, diverse, and equitable conduct of research. Received: June 8, 2022 Revised: July 20, 2022 Accepted: September 2, 2022 Published: October 21, 2022 REFERENCES Anderson, A., Laohavisit, A., Blaby, I.K., Bombelli, P., Howe, C.J., Merchant, S.S., Davies, J.M., and Smith, A.G. (2016). Exploiting algal NADPH oxidase for biophotovoltaic energy. Plant Biotechnol. J. 14, 22–28. Angiulli, G., Lantella, A., Forte, E., Angelucci, F., Colotti, G., Ilari, A., and Malatesta, F. (2015). Leishmania infantum trypanothione reductase is a promiscuous enzyme carrying an NADPH:O2 oxidoreductase activity shared by glutathione reductase. Biochim. Biophys. Acta 1850, 1891– 1897. Armoza-Zvuloni, R., Schneider, A., Sher, D., and Shaked, Y. (2016). Rapid hydrogen peroxide release from the coral Stylophora pistillata during feeding and in response to chemical and physical stimuli. Sci. Rep. 6, 21000. Asada, K. (1999). The water-water cycle in chloroplasts: Scavenging of active oxygens and dissipation of excess photons. Annu. Rev. Plant Physiol. Plant Mol. Biol. 50, 601–639. Bardaweel, S.K., Gul, M., Alzweiri, M., Ishaqat, A., ALSalamat, H.A., and Bashatwah, R.M. (2018). Reactive oxygen species: the dual role in physiological and pathological conditions of the human body. Eurasian J. Med. 50, 193–201. Bedard, K., Lardy, B., and Krause, K.H. (2007). NOX family NADPH oxidases: not just in mammals. Biochimie 89, 1107–1112. Bielski, B.H.J., Shiue, G.G., and Bajuk, S. (1980). - and Reduction of nitro blue tetrazolium by CO2 - radicals. J. Phys. Chem. 84, 830–833. O2 Can, B., Kulaksiz Erkmen, G., Dalmizrak, O., Ogus, I.H., and Ozer, N. (2010). Puriﬁcation and characterisation of rat kidney glutathione reductase. Protein J. 29, 250–256. Carlberg, I., and Mannervik, B. (1986). Reduction of 2, 4, 6-trinitrobenzenesulfonate by glutathione reductase and the effect of NADP+ on the electron transfer. J. Biol. Chem. 261, 1629–1635. Ce´ nas, N.K., Rakauskiene´ , G.A., and Kulys, J.J. (1989). One- and two-electron reduction of quinones by glutathione reductase. Biochim. Biophys. Acta 973, 399–404. Coppo, L., Mishra, P., Siefert, N., Holmgren, A., Pa¨ a¨ bo, S., and Zeberg, H. (2022). A substitution in the glutathione reductase lowers electron leakage and inﬂammation in modern humans. Sci. Adv. 8, eabm1148. 8 iScience 25, 105093, October 21, 2022 Deponte, M. (2013). Glutathione catalysis and the reaction mechanisms of glutathione-dependent enzymes. Biochim. Biophys. Acta 1830, 3217– 3266. Diaz, J.M., Hansel, C.M., Voelker, B.M., Mendes, C.M., Andeer, P.F., and Zhang, T. (2013). Widespread production of extracellular superoxide by heterotrophic bacteria. Science 340, 1223–1226. Diaz, J.M., and Plummer, S. (2018). Production of extracellular reactive oxygen species by phytoplankton: Past and future directions. J. Plankton Res. 40, 655–666. Diaz, J.M., Plummer, S., Hansel, C.M., Andeer, P.F., Saito, M.A., and McIlvin, M.R. (2019). NADPH-dependent extracellular superoxide production is vital to photophysiology in the marine diatom Thalassiosira oceanica. Proc. Natl. Acad. Sci. USA 116, 16448–16453. Eyer, P., Worek, F., Kiderlen, D., Sinko, G., Stuglin, A., Simeon-Rudolf, V., and Reiner, E. (2003). Molar absorption coefﬁcients for the reduced Ellman reagent: reassessment. Anal. Biochem. 312, 224–227. Falkowski, P.G., Barber, R.T., and Smetacek, V. (1998). Biogeochemical controls and feedbacks on ocean primary production. Science 281, 200–207. Field, C.B., Behrenfeld, M.J., Randerson, J.T., and Falkowski, P. (1998). Primary production of the biosphere: integrating terrestrial and oceanic components. Science 281, 237–240. Hajjar, C., Cherrier, M.V., Dias Mirandela, G., Petit-Hartlein, I., Stasia, M.J., Fontecilla-Camps, J.C., Fieschi, F., and Dupuy, J. (2017). The NOX family of proteins is also present in bacteria. mBio 8. e01487-17-01417. Hakam, N., and Simon, J.-P. (2000). Molecular forms and thermal and kinetic properties of puriﬁed glutathione reductase from two populations of barnyard grass (Echinochloa crus- galli (L.) Beauv.: Poaceae) from contrasting climatic regions in North America. Can. J. Bot. 78, 969–980. Hansel, C.M., and Diaz, J.M. (2021). Production of extracellular reactive oxygen species by marine Biota. Ann. Rev. Mar. Sci. 13, 177–200. Hansel, C.M., Diaz, J.M., and Plummer, S. (2019). Tight regulation of extracellular superoxide points to its vital role in the physiology of the globally relevant Roseobacter clade. mBio 10. e02668-18-02618. Huang, H., Ullah, F., Zhou, D.-X., Yi, M., and Zhao, Y. (2019). Mechanisms of ROS regulation of plant development and stress responses. Front. Plant Sci. 10, 800. Ji, M., Barnwell, C.V., and Grunden, A.M. (2015). Characterization of recombinant glutathione reductase from the psychrophilic Antarctic bacterium Colwellia psychrerythraea. Extremophiles 19, 863–874. Korge, P., Calmettes, G., and Weiss, J.N. (2015). Increased reactive oxygen species production during reductive stress: the roles of mitochondrial glutathione and thioredoxin reductases. Biochim. Biophys. Acta 1847, 514–525. Kustka, A.B., Shaked, Y., Milligan, A.J., King, D.W., and Morel, F.M.M. (2005). Extracellular production of superoxide by marine diatoms: contrasting effects on iron redox chemistry and bioavailability. Limnol. Oceanogr. 50, 1172–1180. Laohavisit, A., Anderson, A., Bombelli, P., Jacobs, M., Howe, C.J., Davies, J.M., and Smith, A.G. (2015). Enhancing plasma membrane NADPH oxidase activity increases current output by diatoms in biophotovoltaic devices. Algal Res. 12, 91–98. Liochev, S.I., and Fridovich, I. (1992). Superoxide generated by glutathione reductase initiates a vandante-dependent free radical chain oxidation of NADH. Arch. Biochem. Biophys. 294, 403–406. Lommer, M., Specht, M., Roy, A.S., Kraemer, L., Andreson, R., Gutowska, M.A., Wolf, J., Bergner, S.V., Schilhabel, M.B., Klostermeier, U.C., et al. (2012). Genome and low-iron response of an oceanic diatom adapted to chronic iron limitation. Genome Biol. 13, R66. Lo´ pez-Barea, J., and Lee, C.-Y. (1979). Mouse- liver glutathione reductase. Eur. J. Biochem. 98, 487–499. Magnani, F., Nenci, S., Millana Fananas, E., Ceccon, M., Romero, E., Fraaije, M.W., and Mattevi, A. (2017). Crystal structures and atomic model of NADPH oxidase. Proc. Natl. Acad. Sci. USA 114, 6764–6769. Malviya, S., Scalco, E., Audic, S., Vincent, F., Veluchamy, A., Poulain, J., Wincker, P., Iudicone, D., Vargas, C.d., Bittner, L., et al. (2016). Insights into global diatom distribution and diversity in iScience Article the world’s ocean. Proc. Nat. Acad. Sci. USA 113, E1516–E1525. Mapson, L.W., and Isherwood, F.A. (1963). Glutathione reductase from germinated peas. Biochem. J. 86, 173–191. Massey, V., Strickland, S., Mayhew, S.G., Howell, L.G., Engel, P.C., Matthews, R.G., Schuman, M., and Sullivan, P.A. (1969). Production of superoxide anion radicals in reaction of reduced ﬂavins and ﬂavoproteins with molecular oxygen. Biochem. Biophys. Res. Commun. 36, 891–897. Massey, V., and Williams, C.H., Jr. (1965). On the reaction mechanism of yeast glutathione reductase. J. Biol. Chem. 240, 4470–4480. Minibayeva, F., Kolesnikov, O., Chasov, A., Beckett, R.P., Lu¨ thje, S., Vylegzhanina, N., Buck, F., and Bo¨ ttger, M. (2009). Wound-induced apoplastic peroxidase activities: their roles in the production and detoxiﬁcation of reactive oxygen species. Plant Cell Environ. 32, 497–508. Moore, C.M., Mills, M.M., Arrigo, K.R., Berman- Frank, I., Bopp, L., Boyd, P.W., Galbraith, E.D., Geider, R.J., Guieu, C., Jaccard, S.L., et al. (2013). Processes and patterns of oceanic nutrient limitation. Nat. Geosci. 6, 701–710. Nelson, D.M., Tre´ guer, P., Brzezinski, M.A., Leynaert, A., and Que´ guiner, B. (1995). Production and dissolution of biogenic silica in the ocean: revised global estimates, comparison with regional data and relationship to biogenic sedimentation. Global Biogeochem. Cycles 9, 359–372. Nordman, T., Xia, L., Bjo¨ rkhem-Bergman, L., Damdimopoulos, A., Nalvarte, I., Arne´ r, E.S.J., Spyrou, G., Eriksson, L.C., Bjo¨ rnstedt, M., and Olsson, J.M. (2003). Regeneration of the antioxidant ubiquinol by lipoamide dehydrogenase, thioredoxin reductase and glutathione reductase. Biofactors 18, 45–50. Park, S.Y., Choi, E.S., Hwang, J., Kim, D., Ryu, T.K., and Lee, T.-K. (2009). Physiological and biochemical responses of Prorocentrum minimum to high light stress. Ocean Sci. J. 44, 199–204. Patel, R., Rinker, L., Peng, J., and Chilian, W. (2018). Reactive oxygen species: the good and the bad. In Reactive oxygen species (ROS) in living cells, F. Cristiana and A. Elena, eds. (IntechOpen), pp. 7–20. Paulı´kova´ , H., and Berczeliova´ , E. (2005). The effect of quercetin and galangin on glutathione reductase. Biomed. Pap. Med. Fac. Univ. Palacky. Olomouc. Czech Repub. 149, 497–500. Pinto, M.C., Mata, A.M., and Lo´ pez-Barea, J. (1985). The redox interconversion mechanism of Saccharomyces cerevisiae glutathione reductase. Eur. J. Biochem. 151, 275–281. Reis, J., Massari, M., Marchese, S., Ceccon, M., Aalbers, F.S., Corana, F., Valente, S., Mai, A., Magnani, F., and Mattevi, A. (2020). A closer look into NADPH oxidase inhibitors: Validation and insight into their mechanism of action. Redox Biol. 32, 101466. Rossi, D.C.P., Gleason, J.E., Sanchez, H., Schatzman, S.S., Culbertson, E.M., Johnson, C.J., McNees, C.A., Coelho, C., Nett, J.E., Andes, D.R., et al. (2017). Candida albicans FRE8 encodes a member of the NADPH oxidase family that produces a burst of ROS during fungal morphogenesis. PLoS Pathog. 13, e1006763. Saragosti, E., Tchernov, D., Katsir, A., and Shaked, Y. (2010). Extracellular production and degradation of superoxide in the coral Stylophora pistillata and cultured Symbiodinium. PLoS One 5, e12508. Savvides, S.N., and Karplus, P.A. (1996). Kinetics and Crystallographic analysis of human glutathione reductase in complex with a Xanthene inhibitor (*). J. Biol. Chem. 271, 8101– 8107. ll OPEN ACCESS Schneider, R.J., Roe, K.L., Hansel, C.M., and Voelker, B.M. (2016). Species-level variability in extracellular production rates of reactive oxygen species by diatoms. Front. Chem. 4, 5. Smith, I.K., Vierheller, T.L., and Thorne, C.A. (1988). Assay of glutathione reductase in crude tissue homogenates using 5, 50-dithiobis(2- nitrobenzoic acid). Anal. Biochem. 175, 408–413. Sutherland, K.M., Coe, A., Gast, R.J., Plummer, S., Suffridge, C.P., Diaz, J.M., Bowman, J.S., Wankel, S.D., and Hansel, C.M. (2019). Extracellular superoxide production by key microbes in the global ocean. Limnol. Oceanogr. 64, 2679–2693. https://doi.org/10.1002/lno.11247. Weinberger, F. (2007). Pathogen-induced defense and innate immunity in macroalgae. Biol. Bull. 213, 290–302. Winterbourn, C.C. (2016). Revisiting the reactions of superoxide with glutathione and other thiols. Arch. Biochem. Biophys. 595, 68–71. Worthington, D.J., and Rosemeyer, M.A. (1976). Glutathione reductase from human erythrocytes. Catalytic properties and aggregation. Eur. J. Biochem. 67, 231–238. Wu, J.-X., Liu, R., Song, K., and Chen, L. (2021). Structures of human dual oxidase 1 complex in low-calcium and high-calcium states. Nat. Commun. 12, 155. Yuasa, K., Shikata, T., Ichikawa, T., Tamura, Y., and Nishiyama, Y. (2020a). Nutrient deﬁciency stimulates the production of superoxide in the noxious red-tide-forming raphidophyte Chattonella antiqua. Harmful Algae 99, 101938. Yuasa, K., Shikata, T., Kitatsuji, S., Yamasaki, Y., and Nishiyama, Y. (2020b). Extracellular secretion of superoxide is regulated by photosynthetic electron transport in the noxious red-tide- forming raphidophyte Chattonella antiqua. J. Photochem. Photobiol. B 205, 111839. iScience 25, 105093, October 21, 2022 9 ll OPEN ACCESS STAR+METHODS KEY RESOURCES TABLE REAGENT or RESOURCE Bacterial and virus strains E. coli BL21 (DE3) Chemicals, peptides, and recombinant proteins IPTG Imidazole Glutathione disulﬁde NADPH DTNB Nitroblue tetrazolium Superoxide dismutase Diphenyleneiodonium T. oceanica glutathione reductase (ToGR1) T. oceanica glutathione reductase (ToGR2) Critical commercial assays Q5 site-directed mutagenesis kit Coomassie (Bradford) kit Recombinant DNA ToGR2 gene Software and algorithms Origin Pro (9.9) JMP Pro (16.0) Other pET21 vector HisPur Ni-NTA resin Polyacrylamide gels Amicon Ultra centrifugal ﬁlter RESOURCE AVAILABILITY Lead contact iScience Article IDENTIFIER CMC0014-4X40UL I3325 792527-500G G4376-500MG 10107824001 D8130-500MG BP1081 S5395-30KU D2926-10mg N/A N/A E0552 23200 SOURCE Sigma Teknova Sigma Sigma Roche Sigma Fisher Sigma Sigma This study This study New England Biolabs ThermoFisher Twist Bioscience GenBank AGNL01048094.1 OriginLab SAS Twist Bioscience ThermoFisher BioRad Millipore Sigma N/A N/A n/a 88221 4561036 UFC801024 Further information and requests for resources and reagents should be directed to and will be fulﬁlled by the lead contact, Julia Diaz ([email protected]). Materials availability This study did not generate new unique reagents. Data and code availability d All data reported in this paper will be shared by the lead contact upon request. d This paper does not report original code. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. 10 iScience 25, 105093, October 21, 2022 iScience Article ll OPEN ACCESS METHOD DETAILS Cloning, protein expression, and puriﬁcation The ToGR2 gene (GenBank AGNL01048094.1) was synthesized by Twist Bioscience after intron removal and codon optimization for expression in E. coli. The gene was subcloned into pET21 vector with C-terminal His-tag. The ribosome binding site was inserted upstream of the gene. To remove the putative N-terminal localization domain of ToGR2, N-terminal truncated ToGR2 (NT-ToGR2) was created by deleting 130 amino acids at the N-terminus using the Q5 site-directed mutagenesis kit (New England BioLabs). NT-ToGR1 was created from NT-ToGR2 by replacing three amino acids, Asn248Asp, Lys325Ile, and Asp480Glu, consistent with previous sequence analysis (Diaz et al., 2019). E. coli BL21 (DE3) cells were used to overexpress NT-ToGR1 and NT-ToGR2 (hereafter ToGR1 and ToGR2). Brieﬂy, cells were grown at 37(cid:4)C to an optical density (at 600 nm) of 0.5–0.8, and chilled to 25(cid:4)C. Protein overexpression was induced with 0.1 mM IPTG, and the cells were grown for another 3 h. To purify ToGR1 and ToGR2, the cell pellet was resuspended in buffer A (50 mM Tris-HCl pH 8, 200 mM NaCl, 10% glycerol) with 10 mM imidazole and lysed using a French press at 18000 psi (lysate). The lysate was centrifuged (15,0003g, 50 min, 4(cid:4)C) to isolate the soluble proteins (supernatant), which were then puriﬁed by Ni-NTA afﬁnity chromatography (ThermoFisher) following the manufacturer’s instructions. Brieﬂy, the sample ﬂow-through was discarded, the column was washed with buffer A containing 20 mM imidazole, and the target proteins were eluted by 300 mM imidazole in buffer A. Protein fractions were analyzed with SDS-PAGE (180 V, 30 min) using 10% pre-cast polyacrylamide gels (Bio-Rad) and stained with Coomas- sie blue. Puriﬁed ToGR1 and ToGR2 proteins were stored at (cid:1)80(cid:4)C in buffer containing 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM DTT, and 10% glycerol. Protein concentrations were measured using the Bradford method (ThermoFisher). Protein spectroscopy Protein absorption spectra were recorded on a SpectraMax M3 plate reader (Molecular Devices) at 25(cid:4)C in phosphate buffer (100 mM potassium phosphate and 1 mM EDTA, pH 8.0) or Tris buffer (100 mM Tris-HCl and 1 mM EDTA at pH 8.0) using 10 mM of puriﬁed ToGR1 or ToGR2. Proteins were fully reduced with 0.1 mM DTT or fully oxidized with 1 mM of GSSG. Additional spectra were collected under conditions of superoxide production in the presence of NADPH (120 mM) and in the absence of GSSG. All protein spectra were blank-corrected using enzyme-free controls. Peak positions were determined using the Quick Peaks Gadget in Origin Pro (9.9) with default settings. Enzyme activity assays All reactions were carried out in 200 mL reaction volume in a clear, ﬂat-bottom 96-well plate at 25(cid:4)C on a SpectraMax M3 plate reader (Molecular Devices) in the presence or absence of ToGR1 or ToGR2. Enzyme-free controls were used to determine uncatalyzed reaction rates. All concentrations indicated are ﬁnal. Glutathione reductase native activity ToGR-mediated GSSG reduction and NADPH oxidation were measured using the DTNB assay. In this assay, GR reduces GSSG to GSH, which reacts with 5,50-Dithio-bis-(2-nitrobenzoic acid) (DTNB) to pro- duce the yellow product 2-nitro-5-thiobenzoic acid (TNB). Kinetic analyses were performed in phosphate buffer (100 mM potassium phosphate, 1 mM EDTA, pH 8.0) by varying the concentration of NADPH (0- 80 mM) at ﬁxed concentrations of GSSG (2 mM) and DTNB (0.1 mM). The reaction was started by the addition of ToGR1 or ToGR2 (0.045 nM, ﬁnal enzyme concentration). Absorbance measurements were taken at 412 and 340 nm every 1.5 min for 1 h to monitor TNB production, and NADPH oxidation, respectively. To determine linear reaction rates (R2 > 0.98), each absorbance value was ﬁrst blank-corrected for con- (cid:1)1) were converted to trols containing all components except NADPH. Next, linear reaction rates (abs min (cid:1)1) molar units (nmol L (cid:1)1). Finally, these rates were corrected by subtracting (Eyer et al., 2003) or NADPH (6.22 mM enzyme-free controls. GSH production rates were calculated assuming 1:1 stoichiometry of TNB:GSH (Smith et al., 1988). (cid:1)1) by applying the molar extinction coefﬁcient for TNB (14.15 mM (cid:1)1 min (cid:1)1 cm (cid:1)1 cm iScience 25, 105093, October 21, 2022 11 iScience Article ll OPEN ACCESS Superoxide production ToGR-mediated superoxide production and NADPH oxidation were measured using the NBT assay. In this assay, superoxide reacts with nitroblue tetrazolium (NBT) to produce the purple product monoformazan (MF+). The amount of superoxide produced by GR is proportional to the amount of MF+ production that is inhibited by the enzyme superoxide dismutase (SOD). Kinetic analyses were performed at pH 8.0 in phos- phate buffer (100 mM potassium phosphate, 1 mM EDTA) or Tris buffer (100 mM Tris-HCl, 1 mM EDTA) in (cid:1)1) by varying the concentration the presence of 0.1 mM NBT with or without the addition of SOD (200 U mL of NADPH from 0-300 mM (ToGR1) or 0-600 mM (ToGR2). Reactions were assumed to be in equilibrium with atmospheric oxygen levels. Unless otherwise stated, reactions were started by the addition of ToGR1 (40 nM) or ToGR2 (200 nM). Absorbance measurements were taken at 530 and 340 nm every 1.5 min for 1 h to monitor MF+ production, and NADPH oxidation, respectively. Several types of inhibition experiments were carried out with the following modiﬁcations to the NBT assay. To test the effect of DPI, ToGR1 (40 nM) or ToGR2 (400 nM) was incubated with DPI (0-50 mM in 0.5% DMSO) for 10 min at room temperature, and reactions were started in the presence of DPI by the addition of 200 mM NADPH (ﬁnal concentrations). To test the effect of formaldehyde, ToGR1 and ToGR2 were pre-treated with formaldehyde (see below), which was removed before beginning the NBT assay containing 200 mM NADPH and 40 nM ToGR1 or 400 nM ToGR2 (ﬁnal concentrations). In select inhibition experiments, GSSG was added to 20 mM. To determine rates of MF+ and superoxide production, the absorbance of MF+ was ﬁrst blank-corrected for controls containing all components except NADPH. To quantify superoxide production rates, these mea- (cid:1)1) were surements were further corrected by subtracting SOD controls. Rates of MF+ production (abs min (cid:1)1) by applying the determined over the linear range (R2 > 0.98) and converted to molar units (nmol L (cid:1)1) (Bielski et al., 1980). These rates were converted to su- molar extinction coefﬁcient of MF+ (20 mM (cid:1) :MF+ reaction stoichiometry of 2:1(Bielski et al., 1980). Finally, reaction peroxide production using the O2 rates were corrected by subtracting enzyme-free controls. Rates of NADPH oxidation were determined as above for the DTNB assay. Reaction stoichiometries were determined from multiple observations of NADPH, MF+, and superoxide concentrations across the reaction time course. (cid:1)1 min (cid:1)1 cm Formaldehyde treatment of ToGR ToGR1 and ToGR2 were treated with formaldehyde as described previously (Diaz et al., 2019). Brieﬂy, proteins were incubated at 4(cid:4)C for 2 h in Tris buffer (100 mM Tris-HCl, 1 mM EDTA, pH 8.0) amended with 0% (control), 1.5%, or 4% formaldehyde (ﬁnal concentrations). Formaldehyde was removed by changing into Tris buffer using a 10 kDa Amicon Ultra-0.5 centrifugal ﬁlter device (Millipore Sigma), according to the manufacturer’s instructions. Formaldehyde-treated proteins were incubated for an hour at 4(cid:4)C prior to analysis with the NBT assay. Numerical modeling Catalyzed rates (R) of NADPH oxidation, GSH production, and superoxide production were ﬁt to the following Michaelis-Menten equation by minimizing the sum of squared residuals using the Solver tool in Microsoft Excel: R = kcat½E(cid:5)t½S(cid:5) Km + ½S(cid:5) (Equation 1) where [S] is the concentration of NADPH, [E]t is the concentration of enzyme, and the ﬁtted parameters are the enzyme turnover number (kcat) and half saturation constant (Km). Uncatalyzed reaction rates were deter- mined as the log-linear (R2 > 0.9) pseudo ﬁrst-order rate constant of the enzyme-free controls (Table S2). To quantify DPI inhibition, IC50 values were calculated from the linear regression of log-transformed DPI concentrations versus the percent inhibition of NADPH oxidation or superoxide production (R2 >0.94). IC50 values were calculated according to the formula: IC50 = (0.5-b)/a, where b is the y-intercept and a is the slope of the linear regression. QUANTIFICATION AND STATISTICAL ANALYSIS Statistical analyses were performed in JMP Pro (16.0). Reaction rates and stoichiometries were compared using Student’s t test or Tukey’s Honest Signiﬁcant Difference (HSD) test, and p values <0.05 were consid- ered signiﬁcantly different. 12 iScience 25, 105093, October 21, 2022
10.1016_j.aim.2022.108778	Burness, T., & Guralnick, R. (2022). Fixed point ratios for finite primitive groups and applications. Advances in Mathematics, 411(A), Article 108778. https://doi.org/10.1016/j.aim.2022.108778 Publisher's PDF, also known as Version of record License (if available): CC BY Link to published version (if available): 10.1016/j.aim.2022.108778 Link to publication record on the Bristol Research Portal PDF-document This is the final published version of the article (version of record). It first appeared online via Elsevier at https://doi.org/10.1016/j.aim.2022.108778.Please refer to any applicable terms of use of the publisher. University of Bristol – Bristol Research Portal General rights This document is made available in accordance with publisher policies. Please cite only the published version using the reference above. Full terms of use are available: http://www.bristol.ac.uk/red/research-policy/pure/user-guides/brp-terms/ Advances in Mathematics 411 (2022) 108778 Contents lists available at ScienceDirect Advances in Mathematics journal homepage: www.elsevier.com/locate/aim Fixed point ratios for ﬁnite primitive groups and applications Timothy C. Burness a,∗, Robert M. Guralnick b a School of Mathematics, University of Bristol, Bristol BS8 1UG, UK b Department of Mathematics, University of Southern California, Los Angeles, CA 90089-2532, USA a r t i c l e i n f o a b s t r a c t Article history: Received 16 May 2022 Received in revised form 4 November 2022 Accepted 8 November 2022 Available online 18 November 2022 Communicated by A. Kleshchev Keywords: Fixed point ratios Primitive groups Simple groups Minimal degree Minimal index Let G be a ﬁnite primitive permutation group on a set Ω and recall that the ﬁxed point ratio of an element x ∈ G, denoted fpr(x), is the proportion of points in Ω ﬁxed by x. Fixed point ratios in this setting have been studied for many decades, ﬁnding a wide range of applications. In this paper, we are interested in comparing fpr(x) with the order of x. Our main theorem provides a classiﬁcation of the triples (G, Ω, x) as above with the property that x has prime order r and fpr(x) > 1/(r + 1). There are several applications. Firstly, we extend earlier work of Guralnick and Magaard by determining the primitive permutation groups of degree m with minimal degree at most 2m/3. Secondly, our main result plays a key role in recent joint work with Moretó and Navarro on the commuting probability of p-elements in ﬁnite groups. Finally, we use our main theorem to investigate the minimal index of a primitive permutation group, which allows us to answer a question of Bhargava. © 2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). * Corresponding author. E-mail addresses: [email protected] (T.C. Burness), [email protected] (R.M. Guralnick). https://doi.org/10.1016/j.aim.2022.108778 0001-8708/© 2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). 2 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Contents 1. 2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Aﬃne groups, diagonal groups and twisted wreath products . . . . . . . . . . . . . . . . . . . . . . 10 3. Almost simple groups with non-classical socle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 4. Almost simple classical groups: non-subspace actions . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 5. Almost simple classical groups: subspace actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Product type groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 6. 7. Minimal index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 1. Introduction Let G (cid:2) Sym(Ω) be a ﬁnite transitive permutation group with point stabilizer H. For x ∈ G, we write fpr(x) = \|CΩ(x)\| \|Ω\| = \|xG ∩ H\| \|xG\| for the ﬁxed point ratio of x, where CΩ(x) = {α ∈ Ω : αx = α} is the set of ﬁxed points of x and xG denotes the conjugacy class of x in G. Sometimes we will write fpr(x, Ω) if we wish to highlight the permutation domain Ω. Fixed point ratios have been extensively studied for many decades, ﬁnding a wide range of applications. In one direction, we can view fpr(x) as the probability that a random element in Ω is ﬁxed by x and this explains why ﬁxed point ratios often arise naturally in a probabilistic setting. For example, upper bounds on ﬁxed point ratios are a key ingredient in a powerful probabilistic approach for bounding the base size of a ﬁnite permutation group. This method was originally introduced by Liebeck and Shalev in [37] and it has played a major role in recent proofs of inﬂuential base size conjectures of Cameron, Kantor and Pyber. Fixed point ratios have also turned out to be very useful for studying the generation and random generation properties of ﬁnite groups. For example, bounds on ﬁxed point ratios are applied extensively in [17], which provides the ﬁnal step in the proof of a conjecture of Breuer, Guralnick and Kantor [9] on 3 2 -generated ﬁnite groups. In a diﬀerent direction, ﬁxed point ratios have also been used to study the structure of monodromy groups of coverings of the Riemann sphere, playing a prominent role in the proof of the Guralnick-Thompson genus conjecture [24]. We refer the reader to the survey article [11] for a more detailed discussion of these applications. In this paper, we study ﬁxed point ratios in the setting where G (cid:2) Sym(Ω) is a primi- tive permutation group. Recall that a transitive group G is primitive if Ω has no nontrivial G-invariant partition (equivalently, the point stabilizer H is a maximal subgroup of G). The structure and action of a primitive group is described by the Aschbacher-O’Nan- Scott theorem, which divides the ﬁnite primitive groups into several families. The almost T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 3 simple primitive groups form one of these families and there is an extensive literature on the corresponding ﬁxed point ratios, stretching back several decades. First recall that G is almost simple if there exists a nonabelian ﬁnite simple group G0 (the socle of G) such that G0 (cid:2) G (cid:2) Aut(G0). The possibilities for G0 are determined by the classiﬁcation of ﬁnite simple groups; G0 is either an alternating group, a sporadic group or a group of Lie type (classical or exceptional). When studying ﬁxed point ratios in this setting, it is natural to partition the primitive almost simple classical groups into two collections (the subspace and non-subspace actions), according to the action of H ∩ G0 on the natural module V for G0. Roughly speaking, the action of G on Ω is a subspace action if H ∩ G0 acts reducibly on V , which allows us to identify Ω with a set of subspaces (or pairs of subspaces) of V , otherwise the action is non-subspace. We refer the reader to Deﬁnition 4.1 for the formal deﬁnition of a subspace action that we will work with in this paper. Before stating our main results, let us brieﬂy highlight some of the earlier work on ﬁxed point ratios for almost simple primitive groups of Lie type. So let G (cid:2) Sym(Ω) be such a group, where G0 is a group of Lie type over the ﬁnite ﬁeld Fq of order q. One of the main results in this setting is due to Liebeck and Saxl [34], which states that fpr(x) (cid:2) 4 3q for all nontrivial elements x ∈ G, with a small list of known exceptions, mainly involving groups with socle G0 = L2(q). This bound is essentially best possible. For example, if G = Ln(q), x is a transvection and Ω is the set of 1-dimensional subspaces of V , then it is easy to show that fpr(x) = (qn−1 − 1)/(qn − 1), which is roughly 1/q. However, stronger bounds can be obtained by imposing some additional conditions on G. For instance, an in-depth analysis of ﬁxed point ratios for primitive actions of exceptional groups of Lie type is presented in [33]. For classical groups, we refer the reader to [25,27] for a more detailed treatment of ﬁxed point ratios for subspace actions. For non-subspace actions of classical groups, a key theorem is due to Liebeck and Shalev [37], which states that there exists a universal constant (cid:3) > 0 (independent of G) such that fpr(x) < \|xG\|−(cid:2) for all x ∈ G of prime order. An eﬀective version of this result is established in the series of papers [12–15], which shows that a constant (cid:3) ∼ 1/2 is essentially best possible (see Theorem 4.4). The proofs of these results rely heavily on Aschbacher’s celebrated subgroup structure theorem [1] for ﬁnite classical groups, which divides the possibilities for the point stabilizer into several subgroup collections. With a view towards new applications, in this paper we seek an upper bound on fpr(x) that is given in terms of the order of x. Moreover, we want a bound that applies to all 4 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 ﬁnite primitive groups. With this aim in mind, we present the following result, which is the main theorem of this paper. (Note that Table 6 is presented in Section 5.1.) Theorem 1. Let G (cid:2) Sym(Ω) be a ﬁnite primitive permutation group with point stabilizer H and let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) 1 r + 1 (1) or one of the following holds (up to permutation isomorphism): (i) G is almost simple and one of the following holds: (a) G = Sn or An acting on (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. (b) G = Sn, H = Sn/2 (cid:5) S2, x is a transposition and fpr(x) = 1 3 + n − 4 6(n − 1) . (c) G = M22:2, H = L3(4).22, x ∈ 2B and fpr(x) = 4/11. (d) G is classical in a subspace action and (G, H, x, fpr(x)) is listed in Table 6. (ii) G = V :H is an aﬃne group with socle V = (Cp)d and point stabilizer H (cid:2) GLd(p), r = p, x is conjugate to a transvection in H and fpr(x) = 1/r. (iii) G (cid:2) L (cid:5) Sk is a product type primitive group with its product action on Ω = Γk, where k (cid:3) 2 and L (cid:2) Sym(Γ) is one of the almost simple groups in (i). Remark 1. Some remarks on the statement of Theorem 1 are in order. (a) In part (i)(a), it is plain to see that there are many exceptions to the bound in (1). For example, if G = Sn and (cid:4) = 1, then fpr(x) = 1 − 2/n when x is a transposition. More generally, it is straightforward to show that fpr(x) is maximal when x is an r-cycle (or a double transposition if r = 2 and G = An) and it is easy to compute fpr(x) in this case (see Proposition 3.4 and Remark 3.5). (b) In part (i)(c), we use the standard Atlas [21] notation. As noted above, Table 6 in (i)(d) is presented in Section 5.1 and we refer the reader to Remark 5.4 for information on the notation adopted in this table. It is worth noting that most of the special cases in Table 6 correspond to the action of G on a set of 1-dimensional subspaces (or hyperplanes) of the natural module V and the relevant elements x ∈ G with fpr(x) > (r + 1)−1 typically have an eigenspace on V of codimension 1. (c) Let x = (x1, . . . , xk)π ∈ G be an element of prime order r, where G (cid:2) L (cid:5) Sk is a product type group as in part (iii). Let J be a point stabilizer in the action of L on Γ. In Section 6 we will show that fpr(x) > (r + 1)−1 only if π = 1, in which case fpr(x) = i fpr(xi, Γ). Moreover, Proposition 6.2 states that either L is permutation isomorphic to Sn or An acting on (cid:4)-element subsets of {1, . . . , n}, or x is conjugate (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 5 to (x1, 1, . . . , 1) and (L, J, x1) is one of the special cases arising in part (b), (c) or (d) of Theorem 1(i). (d) The special cases arising in Theorem 1 are described up to permutation isomorphism in order to avoid unnecessary repetition. For instance, if G = A8 and H = AGL3(2), then either fpr(x) (cid:2) (r + 1)−1, or x is an involution with cycle-shape (24) and fpr(x) = 7/15. But here G is permutation isomorphic to L4(2) acting on the set of 1-dimensional subspaces of the natural module (with x corresponding to a transvec- tion), so this case is included in part (i)(d). Similarly, consider the case where G = Sp4(2) and H = O(cid:2) 4(2) is a subspace subgroup. If (cid:3) = + then G is permu- tation isomorphic to S6 acting on the set of partitions of {1, . . . , 6} into two subsets of size 3 (as in part (i)(b) of Theorem 1), and it is permutation isomorphic to S6 in its natural action on {1, . . . , 6} when (cid:3) = − (and therefore included in part (i)(a)). The following result is an immediate corollary of Theorem 1. Corollary 2. Let G be a ﬁnite primitive permutation group and let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) 1√ r + 1 , or (An)k (cid:2) G (cid:2) Sn (cid:5) Sk with k (cid:3) 1, where the action of Sn is on (cid:4)-element subsets of {1, . . . , n} and the wreath product has the product action of degree (cid:4) k . (cid:3) n (cid:3) Remark 2. Consider the special case G (cid:2) Sn (cid:5) Sk arising in the statement of Corollary 2. Let Γ be the set of (cid:4)-element subsets of {1, . . . , n} and note that we may assume 1 (cid:2) (cid:4) < n/2. By combining Proposition 3.4 with the proof of Proposition 6.2, we deduce that fpr(x) (cid:2) (r + 1)−1 if r > n − (cid:4), so we may assume r (cid:2) n − (cid:4). Then fpr(x) is maximal when x is conjugate to an element in (Sn)k of the form (y, 1, . . . , 1) with y ∈ Sn an r-cycle, in which case fpr(x) = fpr(y, Γ). An expression for fpr(y, Γ) is given in part (ii) of Proposition 3.4 and we deduce that fpr(x) (cid:2) 1 − r/n (see Remark 3.5). We also obtain the following result on almost simple primitive groups. Corollary 3. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) 1 r or one of the following holds (up to permutation isomorphism): (i) G = Sn or An acting on (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. (ii) G is a classical group in a subspace action and (G, H, x, fpr(x)) is listed in Table 1. 6 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Table 1 The subspace actions in part (ii) of Corollary 3. G0 H P1 L2(q) P2 U4(2) Spn(2) O− n (2) Ω− Ω+ n (2) n (2) P1 N1 x (ω, ω−1) τ (J2, J n−2 (Λ, In−2) (J2, J n−2 (J2, J n−2 1 1 1 r q − 1 2 2 3 2 2 ) ) ) fpr(x) q−1 + q−3 q2−1 1 5 9 1 2 + 5 14 1 2 + 1 2 + 1 2(2n/2−1) 1 2(2n/2+1) 1 2(2n/2−1) Conditions q (cid:2) 8 G = U4(2).2 n (cid:2) 6 n = 6 G = O− G = O+ n (2) n (2) Remark 3. Let us brieﬂy comment on the notation used in Table 1. Firstly, G0 denotes the socle of G. In the ﬁrst row, r = q − 1 (cid:3) 7 is a Mersenne prime, H = P1 is a Borel subgroup and x is any element of order r. In the second row, H = P2 is the stabilizer of a 2-dimensional totally singular subspace of the natural module and x is an involutory graph automorphism with CG0(x) = Sp4(2). Next, in the third row x is a transvection (and similarly in rows 5 and 6), while x is an element of order 3 with an (n − 2)-dimensional 1-eigenspace on the natural module in the fourth row. In the ﬁnal row, H is the stabilizer of a nonsingular 1-space. We now turn to some of the applications of Theorem 1. One of our motivations for seeking a bound as in Theorem 1 stems from a widely applied theorem of Guralnick and Magaard [28] on the minimal degree of a ﬁnite primitive permutation group G (cid:2) Sym(Ω). Recall that the minimal degree of G, denoted μ(G), is the minimal number of points moved by a nonidentity element of G. This is a classical invariant in permutation group theory, which has been investigated by many authors for more than a century (for example, see Babai [3], Bochert [5], Jordan [29] and Manning [40]). The main theorem of [28] determines the primitive groups G of degree m with μ(G) (cid:2) m/2, extending an earlier result of Liebeck and Saxl [34], which describes the groups with μ(G) (cid:2) m/3. In order to do this, Guralnick and Magaard determine the ﬁnite primitive groups G with the property that fpr(x) > 1 2 for some nonidentity element x ∈ G. We can use Theorem 1 to determine all the primitive groups that contain a nonidentity element x with fpr(x) > 1/3, which allows us to establish the following result on the minimal degree of a ﬁnite primitive permutation group. This can be viewed as a natural extension of the earlier work of Liebeck and Saxl [34] and Guralnick and Magaard [28]. Theorem 4. Let G (cid:2) Sym(Ω) be a ﬁnite primitive permutation group of degree m with point stabilizer H and minimal degree μ(G). Then either μ(G) (cid:3) 2m/3, or one of the following holds (up to permutation isomorphism): T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 7 Table 2 The subspace actions in part (iv) of Theorem 4. G0 Ln(2) Ln(3) U4(q) Spn(2) Ωn(3) PΩ(cid:2) n(q) H P1 P1 P2 P1 O(cid:2) P1 − N 1 P1 n(2) N1 m 2n − 1 2 (3n − 1) 1 (q3 + 1)(q + 1) 2n − 1 2n/2−1(2n/2 + (cid:2)) 2 (3n−1 − 1) 1 2 3(n−1)/2(3(n−1)/2 − 1) 1 (2n/2 − (cid:2))(2n/2−1 + (cid:2)) 2 (3n/2 + 1)(3n/2−1 − 1) 1 2n/2−1(2n/2 − (cid:2)) 2 3n/2−1(3n/2 − 1) 1 2 3n/2−1(3n/2 + 1) 1 μ(G) 2n−1 3n−1 − 1 q2(q2 − 1) 2n−1 2n/2−1(2n/2−1 + (cid:2)) 3(n−3)/2(3(n−1)/2 − 1) 3n−2 − 2.3(n−3)/2 − 1 2n/2−1(2n/2−1 + (cid:2)) 3n/2−1(3n/2−1 − 1) 2n/2−1(2n/2−1 − (cid:2)) 3n/2−1(3n/2−1 − 1) 3n−2 − 1 Conditions n (cid:2) 3 n (cid:2) 3, r ∈ G q ∈ {2, 3}, τ ∈ G n (cid:2) 6 n (cid:2) 6 r+ ∈ G r− ∈ G G = O(cid:2) (q, (cid:2)) = (3, −), r ∈ G G = O(cid:2) (q, (cid:2)) = (3, +), r(cid:2) ∈ G (q, (cid:2)) = (3, −), r(cid:3) ∈ G n(2) n(2) (i) G = Sn or An acting on (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. (ii) G = Sn, H = Sn/2 (cid:5) S2 and μ(G) = (cid:6) (cid:5) 1 4 1 + 1 n − 1 n! (n/2)!2 . (iii) G = M22:2, H = L3(4).22, m = 22 and μ(G) = 14. (iv) G is an almost simple classical group in a subspace action and (G, H, m, μ(G)) is listed in Table 2, where G0 is the socle of G. (v) G = V :H is an aﬃne group with socle V = (C2)d, H (cid:2) GLd(2) contains a transvec- tion and μ(G) = 2d−1 = m/2. (vi) G (cid:2) L (cid:5) Sk is a product type primitive group with its product action on Ω = Γk, where k (cid:3) 2 and L (cid:2) Sym(Γ) is one of the almost simple groups in (i)-(iv). Remark 4. In Table 2, we adopt the standard Pm notation for maximal parabolic sub- groups (in which case, we can identify Ω with the set of totally singular m-dimensional subspaces of the natural module for G0). In the second row, r ∈ PGLn(3) is the image of a reﬂection (−In−1, I1) and we note that r ∈ G0 if and only if n is odd. Similarly, in the third row, τ is an involutory graph automorphism with CG0(τ ) = PSp4(q). For G0 = Ωn(3), we write N − 1 for the stabilizer of a nondegenerate 1-space U of the natural module such that U ⊥ is a minus-type orthogonal space. We also write r(cid:2) with (cid:3) = ± for a reﬂection (−In−1, I1) with an (cid:3)-type (−1)-eigenspace (we note that r(cid:2) ∈ G if and only if G = SOn(3) or n ≡ (cid:3) (mod 4)). Similarly, if G0 = PΩ(cid:2) n(3) with n even, then n(3) is the image of any reﬂection of the form (−In−1, I1), while we write rδ r ∈ PGO(cid:2) with δ ∈ {(cid:4), (cid:5)} if we need to specify the discriminant of the 1-dimensional 1-eigenspace of r (which is either a square or nonsquare). In addition, N1 denotes the stabilizer of a nonsingular 1-space (respectively, a nondegenerate 1-space with square discriminant) when q = 2 (respectively, q = 3). 8 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 In order to describe our next application, let G be a ﬁnite group and recall that the commuting probability of G is the probability that two random elements of G commute. In [18], Moretó, Navarro and the authors introduce a natural analogue of this widely studied notion, which is deﬁned in terms of a prime r. Let Prr(G) be the probability that two random r-elements in G commute. Then [18, Theorem A] is the following. Theorem 5. Let G be a ﬁnite group and let r be a prime. Then Prr(G) > r2 + r − 1 r3 if and only if G has a normal and abelian Sylow r-subgroup. The proof of Theorem 5 relies on bounding the ratio \|CG(x)r\|/\|Gr\| for every nontrivial r-element x ∈ G, where Kr denotes the set of r-elements in the subgroup K of G. By embedding CG(x) in a maximal subgroup of G, we can bring bounds on ﬁxed point ratios for primitive groups into play and Theorem 1 turns out to be a key ingredient in the proof. We refer the reader to [18] for further details. Our ﬁnal application concerns the minimal index of a permutation group. Let G (cid:2) Sym(Ω) be a primitive permutation group of degree m. For x ∈ G we deﬁne ind(x) = m − orb(x) = m ⎛ ⎝1 − 1 \|x\| (cid:9) y∈(cid:4)x(cid:5) ⎞ fpr(y) ⎠ to be the index of x, where orb(x) is the number of orbits of x on Ω. Note that ind(x) is also the minimal number t such that x is a product of t transpositions in the symmetric group Sm. Let us also observe that if x has order r, then ind(x) (cid:2) m(1 − 1/r). This quantity arises naturally in various number theoretic estimates, including the Riemann-Hurwitz formula for the genus of a branched covering of a smooth projective curve. Similarly, Ind(G) = min{ind(x) : 1 (cid:8)= x ∈ G}, which we call the minimal index of G, also appears in various number theoretic settings (see the work of Malle [38,39], for example). In particular, it plays a crucial role in a recent beautiful paper of Bhargava [4], where he estimates the number of number ﬁelds of given discriminant with a given Galois group. In response to a question from Manjul Bhargava, we can use Theorem 1 to investigate Ind(G) for an arbitrary primitive permutation group G. Our main results are Theorems 6 and 7 below, which will be proved in Section 7 as an application of Theorem 1. Theorem 6. Let G (cid:2) Sym(Ω) be a primitive permutation group of degree m with point stabilizer H and assume \|G\| is even. Let x ∈ G be an element with Ind(G) = ind(x). Then the following hold: T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 9 Table 3 The cases arising in part (i) of Theorem 7. L J \|Γ\| Ind(L, Γ) Conditions U4(2).2 Spn(2) O− n (2) O+ n (2) n (2) P2 O− P1 N1 27 2n/2−1(2n/2 − 1) (2n/2−1 − 1)(2n/2 + 1) 2n/2−1(2n/2 − 1) 6 2n/2−2(2n/2−1 − 1) 2n/2−2(2n/2−1 − 1) 2n/2−2(2n/2−1 − 1) n (cid:2) 6 n (cid:2) 8 n (cid:2) 8 (i) \|x\| ∈ {2, 3} and there exists an involution x ∈ G with ind(x) = Ind(G). (ii) If \|x\| = 3 then one of the following holds (up to permutation isomorphism): (a) Ind(G) = m/2 and \|H\| is odd. (b) Ind(G) = 4m/9, G = V :H is an aﬃne group with socle (C3)d, x ∈ H (cid:2) GLd(3) is a transvection and H does not contain an involution of the form (−I1, Id−1). (c) Ind(G) = 4m/9, G = L (cid:5) P with its product action on Ω = Γk, where k (cid:3) 1, P (cid:2) Sk is transitive, L = L2(8):3 in its standard action of degree 9 and x ∈ Lk is conjugate to (x1, 1, . . . , 1) with x1 a ﬁeld automorphism of L2(8) of order 3. (d) Ind(G) = 2m/n, (An)k (cid:2) G (cid:2) Sn (cid:5)Sk, n (cid:3) 5, k (cid:3) 1, Sn has its natural action on {1, . . . , n}, the wreath product has the product action of degree nk, x is conjugate to (x1, 1, . . . , 1) ∈ (An)k with x1 a 3-cycle and G ∩(Sn)k does not contain elements of the form (y1, 1, . . . , 1), (y1, y2, 1, . . . , 1) or (y1, y2, y3, 1, . . . , 1) (n = 5 only), up to conjugacy, where each yi is a transposition. Theorem 7. Let G (cid:2) Sym(Ω) be a primitive permutation group of degree m with point stabilizer H and assume \|G\| is even. Then either m 4 (cid:2) Ind(G) (cid:2) m 2 , or one of the following holds (up to permutation isomorphism): (i) G = L (cid:5) P with its product action on Ω = Γk, where k (cid:3) 1, P (cid:2) Sk is transitive, L (cid:2) Sym(Γ) is an almost simple primitive classical group in a subspace action with point stabilizer J and 3m 14 (cid:2) Ind(G) = \|Γ\|k−1Ind(L, Γ) < m 4 , where (L, J, \|Γ\|, Ind(L, Γ)) is one of the cases in Table 3. (ii) G is a subgroup of Sn (cid:5) Sk containing (An)k with k (cid:3) 1, where the action of Sn is on (cid:4)-element subsets of {1, . . . , n} and the wreath product has the product action of degree (cid:4) k . (cid:3) n (cid:3) If G is a primitive group of odd order, then G is solvable by the Feit-Thompson theorem and thus G is an aﬃne group of degree m = pd for some odd prime p. In Theorem 7.4 we will show that 10 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Table 4 The ﬁnite primitive permutation groups. Type I II III(a)(i) III(a)(ii) III(b)(i) III(b)(ii) III(c) Description Aﬃne: G = V :H (cid:3) AGL(V ), H (cid:3) GL(V ) irreducible Almost simple: T (cid:3) G (cid:3) Aut(T ) Diagonal type: T k (cid:3) G (cid:3) T k.(Out(T ) × P ), P (cid:3) Sk primitive Diagonal type: T 2 (cid:3) G (cid:3) T 2.Out(T ) Product type: G (cid:3) L (cid:6) P , L primitive of type II, P (cid:3) Sk transitive Product type: G (cid:3) L (cid:6) P , L primitive of type III(a), P (cid:3) Sk transitive Twisted wreath product (cid:12) min m (cid:5) 1 − 3 2r + 1 (cid:6) (cid:5) 1 − 1 p , m (cid:13) (cid:6) 2 (cid:2) Ind(G) (cid:2) m (cid:5) 1 − 1 r (cid:6) , where r is the smallest prime divisor of \|G\|. A framework for our proof of Theorem 1 is provided by the Aschbacher-O’Nan-Scott theorem, which divides the ﬁnite primitive permutation groups into several families (see Table 4 for a rough description). We proceed by considering each family in turn. As one might expect, most of the work involves the almost simple groups, with a long and delicate analysis required for the subspace actions of classical groups (this is carried out in Section 5). Our proof for almost simple groups relies heavily on some of the earlier results on ﬁxed point ratios referred to above (in particular, the main theorem of [34], combined with [12–15,27] for classical groups and [33] for exceptional groups of Lie type). 2. Aﬃne groups, diagonal groups and twisted wreath products In this section we prove Theorem 1 when G is either a primitive group of aﬃne type, diagonal type or a twisted wreath product. 2.1. Aﬃne groups Proposition 2.1. Let G (cid:2) Sym(Ω) be a ﬁnite primitive permutation group of aﬃne type with socle (Cp)d and point stabilizer H (cid:2) GLd(p), where p is a prime. If x ∈ G has prime order r, then either (i) fpr(x) (cid:2) (r + 1)−1; or (ii) r = p, x is conjugate to a transvection in H and fpr(x) = r−1. Proof. Write G = V :H, where V = (Fp)d and H (cid:2) GL(V ) is irreducible. By replacing x by a suitable conjugate, we may assume that x ∈ H (otherwise fpr(x) = 0). Set e = dim CV (x) and note that fpr(x) = pe−d. If r = p then either e (cid:2) d − 2 and fpr(x) (cid:2) p−2 < (r + 1)−1, or e = d − 1, x is a transvection and fpr(x) = r−1. Now assume r (cid:8)= p. Here r divides \|GLd−e(p)\|, so r (cid:2) pd−e − 1 and thus fpr(x) (cid:2) (r + 1)−1 as required. (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 11 2.2. Diagonal groups Next we turn to the primitive groups of diagonal type. We will need the following lemma on ﬁnite simple groups. Lemma 2.2. Let T be a nonabelian ﬁnite simple group. Then the following hold: (i) \|Out(T )\|3 < \|T \|. (ii) \|{t ∈ T : tα = t−1}\| (cid:2) 4\|T \|/15 for all α ∈ Aut(T ). (iii) \|T \| (cid:3) (\|α\| + 1)\|CInn(T )(α)\| for all α ∈ Aut(T ) of prime order. Proof. Part (i) is [22, Lemma 4.8] and part (ii) follows from [43, Theorem 3.1]. Now consider part (iii) and let α ∈ Aut(T ) be an automorphism of prime order r. First assume α ∈ Inn(T ), in which case it suﬃces to show that \|αInn(T )\| (cid:3) r + 1. If \|αInn(T )\| (cid:2) r then \|αInn(T )\| (cid:2) r − 1 since no simple group has a nontrivial conjugacy class of prime-power length (this is a classical result of Burnside), whence Inn(T ) is isomorphic to a subgroup of Sr−1 and this contradicts the fact that Inn(T ) contains an element of order r. Now assume α ∈ Aut(T ) \ Inn(T ). Since T is simple, the index of CInn(T )(α) in Inn(T ) is at least 5 and so we may assume \|α\| = r (cid:3) 5. This implies that T is a group of Lie type over Fq and we write Inndiag(T ) for the subgroup of Aut(T ) generated by the inner and diagonal automorphisms of T . Then either (a) T = L(cid:2) (b) q = qr n(q), α ∈ Inndiag(T ) \ Inn(T ) and r divides (n, q − (cid:3)); or 0 for some prime power q0 and α is a ﬁeld automorphism. If (a) holds, then \|T \| \|CInn(T )(α)\| (cid:3) \|GL(cid:2) n(q)\| n−1(q)\|\|GL(cid:2) 1(q)\| = \|GL(cid:2) qn−1(qn − (cid:3)) q − (cid:3) (cid:3) r + 1 as required. Similarly, if (b) holds then \|CInn(T )(α)\| (cid:2) \|Inndiag(S)\|, where S is a group of the same type as T , but deﬁned over the subﬁeld Fq0. Once again, it is easy to verify the desired bound. (cid:2) Proposition 2.3. Let G (cid:2) Sym(Ω) be a ﬁnite primitive permutation group of diagonal type and let x ∈ G be an element of prime order r. Then fpr(x) (cid:2) (r + 1)−1. Proof. Let T k be the socle of G, where T is a nonabelian simple group and k (cid:3) 2. If α ∈ Aut(T ) then let ¯α denote the coset αInn(T ) in Out(T ). Adopting Fawcett’s notation for diagonal groups presented in [22], we may assume that G = A(k, T ):Sk (cid:2) Aut(T ) (cid:5) Sk and H = D(k, T ), where A(k, T ) = {(α1, . . . , αk) ∈ Aut(T )k : ¯α1 = ¯αi for all i} 12 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 D(k, T ) = {(α, . . . , α)π : α ∈ Aut(T ), π ∈ Sk}. Let R(G) be a set of representatives of the G-classes of elements of prime order in H. Following [22, Section 4], we partition R(G) into three collections (here we write [k] for {1, . . . , k}): R1(G) = {(α, . . . , α)π ∈ R(G) : π is ﬁxed-point-free on [k]} R2(G) = {(α, . . . , α)π ∈ R(G) : π = 1} R3(G) = {(α, . . . , α)π ∈ R(G) : π (cid:8)= 1 and iπ = i for some i ∈ [k]}. First assume x = (α, . . . , α)π ∈ R1(G) has order r. Then r divides k and [22, Lemma 4.6] gives \|CG(x)\| (cid:2) \|CSk (π)\|\|Out(T )\|\|T \|k/r. Since \|xG ∩ H\| (cid:2) \|Aut(T )\|\|πSk \|, it follows that fpr(x) (cid:2) \|Out(T )\| \|T \|k−k/r−1 and it suﬃces to show that (r + 1)\|Out(T )\| (cid:2) \|T \|k−k/r−1. (2) Suppose r < k, so k (cid:3) 2r. By setting k = 2r and T = A5, it is easy to verify the bound in (2). Similarly, the desired result follows if r = k (cid:3) 3. Finally, suppose r = k = 2. Here we claim that \|CΩ(x)\| = \|{t ∈ T : tα = t−1}\|. To see this, we identify Ω with the set of cosets {D(1, t) : t ∈ T } of D = {(s, s) : s ∈ T } in T 2. Then D(1, t)x = D(1, tα)π = D(tα, 1) and thus D(1, t)x = D(1, t) if and only if (s, st) = (tα, 1) for some s ∈ T . The latter equality holds if and only if s = tα and tα = t−1, which justiﬁes the claim. By applying Lemma 2.2(ii) we deduce that \|CΩ(x)\| (cid:2) 4\|T \|/15 and thus fpr(x) = \|CΩ(x)\| \|T \| (cid:2) 4 15 < 1 3 = 1 r + 1 as required. Next let us assume x = (α, . . . , α) ∈ R2(G), so \|α\| = r. By applying [22, Lemmas 4.5, 4.6] we see that \|CΩ(x)\| = \|CInn(T )(α)\|k−1 and thus Lemma 2.2(iii) yields fpr(x) = (cid:5) \|CInn(T )(α)\| \|T \| (cid:6)k−1 (cid:5) (cid:2) (cid:6) k−1 . 1 r + 1 The result follows. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 13 Finally let us assume x = (α, . . . , α)π ∈ R3(G), so k (cid:3) 3 and π is a nontrivial permutation of order r with f ﬁxed points on [k], where 1 (cid:2) f (cid:2) k − r. Note that \|α\| = r. Using [22, Lemmas 4.5, 4.6] we deduce that fpr(x) = \|Out(T )\|\|CInn(T )(α)\|f −1 \|T \|k−1−(k−f )/r = \|Out(T )\| \|T \|(k−f )(1−1/r) · (cid:5) \|CInn(T )(α)\| \|T \| (cid:6)f −1 (cid:2) \|Out(T )\| \|T \|r−1 and the result follows by applying the bound \|Out(T )\| < \|T \|1/3 in Lemma 2.2(i). (cid:2) 2.3. Twisted wreath groups Proposition 2.4. Let G (cid:2) Sym(Ω) be a ﬁnite primitive permutation group of twisted wreath type and let x ∈ G be an element of prime order r. Then fpr(x) (cid:2) (r + 1)−1. Proof. Write G = T k:H, where T is a nonabelian ﬁnite simple group and the point stabilizer H is a transitive subgroup of Sk. Let x ∈ H be an element of prime order r. Then by applying [23, Lemmas 5.3, 5.4] we deduce that fpr(x) (cid:2) \|T \|(cid:3)−k, where (cid:4) is the number of r-cycles in the cycle-shape of x (with respect to the action on {1, . . . , k}). The result now follows since (cid:4) (cid:2) k/r. (cid:2) In order to complete the proof of Theorem 1, we may assume G is either an almost simple group or a product type group. The latter groups will be handled in Section 6 and we will see that the desired result is easily obtained by combining Proposition 2.3 with our main result for almost simple groups. So the almost simple groups will be our main focus for the remainder of the paper and we divide the analysis into three cases: (a) Non-classical groups (Section 3); (b) Classical groups in non-subspace actions (Section 4); and (c) Classical groups in subspace actions (Section 5). 3. Almost simple groups with non-classical socle In this section we assume G (cid:2) Sym(Ω) is a ﬁnite primitive almost simple group with socle G0. We postpone the analysis of classical groups to Sections 4 (non-subspace actions) and 5 (subspace actions), so here we take G0 to be a sporadic, alternating or exceptional group of Lie type. In part (ii) of the following result, we adopt the standard labeling of conjugacy classes from the Atlas [21]. Proposition 3.1. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0, a sporadic simple group. Let x ∈ G be an element of prime order r. Then either (i) fpr(x) (cid:2) (r + 1)−1; or 14 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (ii) G = M22:2, H = L3(4):22, x is an involution in the class 2B and fpr(x) = 4/11. Proof. For G (cid:8)= B, M we can verify the bound using GAP [26]. Indeed, in each case the character tables of G and H are available in the GAP Character Table Library [8] (we use the Maxes function to access the character table of H), together with the fusion map from H-classes to G-classes. This allows us to compute fpr(x) precisely for all x ∈ G and it is routine to check the desired result. A very similar approach also applies when G = B is the Baby Monster. As before, the character tables of G and H are available in [8] and we can also access the stored fusion map if H (cid:8)= (22 × F4(2)).2. This quickly reduces the problem to the case H = (22 × F4(2)).2. Here we use the function PossibleClassFusions to determine a set of candidate fusion maps (there are 64 such maps in total) and for each possibility one checks that fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. To complete the proof, we may assume G = M is the Monster group. As discussed in [44], G has 44 known conjugacy classes of maximal subgroups and any additional maximal subgroup is almost simple with socle one of L2(8), L2(13), L2(16) or U3(4). In addition, we note that r (cid:2) 71 and by inspecting the character table of G we can compute ar = min{\|xG\| : x ∈ G, \|x\| = r}, which yields the trivial bound fpr(x) (cid:2) \|H\|/ar. In this way, we immediately deduce that fpr(x) (cid:2) (r+1)−1 if \|H\| < 1020 and so by inspecting the list of known maximal subgroups of G, we have reduced the problem to the cases where H is one of the following: 2.B, 21+24.Co1, 3.Fi24, 22.2E6(2).S3, 210+16.Ω+ 10(2), 22+11+22.(M24 × S3). In the ﬁrst four cases, we can use the function NamesOfFusionSources to access the character table of H in GAP and as above we can check the bound fpr(x) (cid:2) (r + 1)−1 by working with the stored fusion map from H-classes to G-classes. Now let us turn to the ﬁnal two cases, where H is a 2-local subgroup. If r (cid:3) 3 then ar r + 1 (cid:3) 53644422509007885434880000000 > \|H\| and the result follows. Now assume r = 2 and note that G has two conjugacy classes of involutions, labeled 2A and 2B, where \|2A\| = 97239461142009186000, \|2B\| = 5791748068511982636944259375. If x ∈ 2B then \|xG\| > 3\|H\| and the result follows. On the other hand, if x ∈ 2A then \|xG ∩ H\| is given in [20, Proposition 3.9]; this allows us to compute fpr(x) precisely and it is easy to check that fpr(x) (cid:2) 1/3 as required. (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 15 Table 5 Some special cases with G0 = A6. G A6 S6 A6.22 H A5 (prim) S5 (prim) S2 (cid:6) S3 (S3 (cid:6) S2).2 x (32) (23) (32) (2, 14) (2, 14) r 3 2 3 2 2 fpr(x) 1/2 2/3 1/2 7/15 2/5 Next we consider the groups with socle an alternating group. We refer the reader to Proposition 3.4 for further information on the groups arising in part (i) of the following result. In Table 5, we write (rh, 16−rh) to denote any element in S6 that is a product of h disjoint r-cycles. We also write “prim” if the given subgroup H acts primitively on {1, . . . , 6}. Proposition 3.2. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0 = An. Let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds (up to permutation isomorphism): (i) G = Sn or An acting on (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. (ii) G = Sn, H = Sn/2 (cid:5) S2, x is a transposition and fpr(x) = 1 3 + n − 4 6(n − 1) . (iii) G = A8, H = AGL3(2), x is an involution with cycle shape (24) and fpr(x) = 7/15. (iv) n = 6 and (G, H, x, r, fpr(x)) is one of the cases in Table 5. Proof. Let x ∈ G be an element of prime order r. Note that r (cid:2) n and recall that we may as well assume x ∈ H. For n = 6 we can use Magma [6] to check that fpr(x) (cid:2) (r + 1)−1 unless (i) or (ii) holds, or (G, H, x) is one of the cases recorded in Table 5. For the remainder, we may assume that G = Sn or An, with n (cid:8)= 6. We partition the analysis into two cases according to the action of H on [n] = {1, . . . , n}, noting that (i) holds if H is intransitive. Case 1. H is primitive. Suppose H acts primitively on [n]. For n (cid:2) 12, a straightforward Magma computation shows that fpr(x) (cid:2) (r + 1)−1 unless G = A8, H = AGL3(2) and x has cycle-shape (24), in which case r = 2 and fpr(x) = 7/15. For the remainder, we will assume n (cid:3) 13. Our aim is to establish the bound fpr(x) (cid:2) (r + 1)−1. Let μ(H) = min{\|supp(x)\| : 1 (cid:8)= x ∈ H} be the minimal degree of H, which is deﬁned to be the minimal number of points moved by a nontrivial element of H. Suppose μ(H) (cid:3) n/2 and observe that this forces 16 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 \|xG\| (cid:3) n! 2n/4(cid:9)n/4(cid:10)!(cid:9)n/2(cid:10)! (3) for all x ∈ H of prime order (minimal if n ≡ 0 (mod 4) and x is an involution with cycle-shape (2n/4, 1n/2)). If n (cid:3) 25 then [41, Corollary 1.2] gives \|H\| < 2n and thus fpr(x) < 25n/4(cid:9)n/4(cid:10)!(cid:9)n/2(cid:10)! n! . One can check that this upper bound is at most 1/(n +1) for all n (cid:3) 25, which establishes the desired bound (recall that r (cid:2) n). For 13 (cid:2) n (cid:2) 24 we work with the bound fpr(x) (cid:2) a/br, where a is the maximal order of a core-free primitive subgroup of G (set a = 0 if no such subgroup exists) and br = min (cid:14) n! rhh!(n − hr)!s : (cid:9)n/2r(cid:10) (cid:2) h (cid:2) (cid:11)n/r(cid:12) (cid:15) is the minimal size of a conjugacy class in G containing elements of order r with at most n/2 ﬁxed points on [n] (here s = 2 if G = An and r = n, otherwise s = 1). Using Magma, it is easy to calculate a and br for every prime r (cid:2) n and one checks that a/br (cid:2) (r + 1)−1 in all cases. To complete the analysis of Case 1, we may assume n (cid:3) 13 and μ(H) < n/2. Here [28, Theorem 1] describes the possibilities for H (note that in our setting, H is a maximal subgroup of G, which simpliﬁes the list of cases we need to consider): (a) H is an almost simple classical group with socle H0 = Ω(cid:2) m(2), where m (cid:3) 6 and H is acting on a set of hyperplanes of the natural module for H0; (b) H = Sm or Am acting on the set of (cid:4)-element subsets of [m], where we have m (cid:3) 5, 2 (cid:2) (cid:4) < m/2 and n = (cid:3) (cid:4) m (cid:3) ; (c) H = (S(cid:3) (cid:5) Sk) ∩ G where (cid:4) (cid:3) 5, k (cid:3) 2 and H is acting with its product action on n = (cid:4)k points. First consider case (a). As explained in [28], either (a(cid:7)) H0 = Ω2(cid:3)+1(2) or Ω+ (a(cid:7)(cid:7)) H0 = Ω− 2(cid:3)(2), n = 2(cid:3)−1(2(cid:3) − 1) and μ(H) = n/2 − 2(cid:3)−2; or 2(cid:3)(2), n = (2(cid:3) + 1)(2(cid:3)−1 − 1) and μ(H) = n/2 − (2(cid:3)−1 − 1)/2. In both cases we have n (cid:3) 25 (so \|H\| < 2n as before) and we can proceed as above, working with a slightly modiﬁed version of the lower bound on \|xG\| in (3) to account for the fact that μ(H) < n/2. We omit the details. Next let us assume we are in case (b) above. Here n = (see [28, p.130], where it is noted that a transposition has the largest number of ﬁxed points). Suppose (cid:4) = 2. Here n = m(m − 1)/2 and μ(H) (cid:3) 2m − 4, which implies that and μ(H) (cid:3) 2 (cid:3) (cid:4) m−2 (cid:3)−1 (cid:3) (cid:4) m (cid:3) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 17 \|xG\| (cid:3) n! 2m−2(m − 2)!(n − 2m + 4)! for all x ∈ H of prime order (minimal if x is an involution with n − 2m + 4 ﬁxed points). Since \|H\| (cid:2) m!, it follows that fpr(x) (cid:2) 2m−2(m − 2)!(n − 2m + 4)!m! n! and it is straightforward to check that this upper bound is less than (m + 1)−1 for all m (cid:3) 5. The result now follows since r (cid:2) m. The reader can check that a very similar argument applies if (cid:4) > 2. Finally, let us turn to case (c). Here n = (cid:4)k and μ(H) = 2(cid:4)k−1 (see [28, p.130]). Suppose k = 2, so n = (cid:4)2, \|H\| (cid:2) 2((cid:4)!)2 and r (cid:2) (cid:4). Given the lower bound on μ(H), it follows that and thus \|xG\| (cid:3) n! 2(cid:3)(cid:4)!(n − 2(cid:4))! fpr(x) (cid:2) 2(cid:3)+1((cid:4)!)3(n − 2(cid:4))! n! , which is easy to check is at most ((cid:4) + 1)−1 for all (cid:4) (cid:3) 5. This gives the desired result for k = 2 and the cases with k > 2 can be handled in the same way. Case 2. H is transitive and imprimitive. To complete the proof of the proposition, we may assume H acts transitively and imprimitively on [n]. The groups with n (cid:2) 13 can be checked using Magma, noting that the only exceptions (excluding n = 6) are the cases where G = Sn, H = Sn/2 (cid:5) S2 and x is a transposition. Here \|xG ∩ H\| = 2 and thus (cid:4) (cid:3) , \|xG\| = n/2 2 (cid:4) n 2 (cid:3) fpr(x) = 2 (cid:4) (cid:3) n/2 (cid:4) = (cid:3) 2 n 2 1 3 + n − 4 6(n − 1) . In particular, this gives an inﬁnite family of exceptions to the bound fpr(x) (cid:2) (r + 1)−1 and this special case is recorded in part (i)(b) of Theorem 1. Now assume n (cid:3) 14. Fix a divisor (cid:4) of n with 1 < (cid:4) < n and identify Ω with Π(cid:3), the set of partitions of [n] into (cid:4) parts of equal size. Note that H = (Sn/(cid:3) (cid:5)S(cid:3)) ∩G. If 3 (cid:2) r (cid:2) (cid:4) then [19, Lemma 4.5] implies that fpr(x) < (cid:4)−2 and the result follows. Similarly, if r = 2 then the bound in [19, Lemma 4.6] is suﬃcient unless x is a transposition and (cid:4) = 2, which is a genuine exception, as noted above. So to complete the analysis of this case, we may assume r > (cid:4). Note that 3 (cid:2) r (cid:2) n/(cid:4). Let us also note that x ﬁxes a partition in Π(cid:3) if and only if it ﬁxes each part of the 18 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 partition setwise. It follows that fpr(x) (cid:2) fpr(y), where y ∈ G is an r-cycle, and so it (cid:4) (cid:3) suﬃces to show that fpr(y) (cid:2) (r + 1)−1. Clearly, we have \|yG ∩ H\| = (r − 1)! n/(cid:3) (cid:4) and (cid:3) r \|yG\| = (r − 1)! , whence (cid:4) n r fpr(y) = (n/(cid:4))!(n − r)!(cid:4) n!(n/(cid:4) − r)! (cid:2) (r + 1)−1 if and only if f (r) (cid:3) 1, where f (r) := n!(n/(cid:4) − r)! (n/(cid:4))!(n − r)!(cid:4)(r + 1) . It is routine to check that f is increasing as a function of r and the result follows since f (3) = (n − 1)(n − 2) 4(n/(cid:4) − 1)(n/(cid:4) − 2) (cid:3) (n − 1)(n − 2) 4(n/2 − 1)(n/2 − 2) > 1. In conclusion, if n (cid:8)= 6 and H is transitive and imprimitive, then fpr(x) (cid:2) (r + 1)−1 unless G = Sn, H = Sn/2 (cid:5) S2 and x is a transposition. This completes the proof of the proposition. (cid:2) Remark 3.3. Let us consider the special cases arising in parts (iii) and (iv) of Propo- sition 3.2. First observe that the example in (iii) does not appear in the statement of Theorem 1 since G is permutation isomorphic to the classical group L4(2) acting on the set of 1-dimensional subspaces of its natural module. Similarly, if G = S6 or A6 and H = S5 or A5 is primitive, then there is a permutation isomorphism to the natural action of G on {1, . . . , 6}, which is included in part (i). If G = S6 and H = S2 (cid:5) S3 then G is permutation isomorphic to Sp4(2) acting on the set of 1-dimensional subspaces of the natural module. And if G = A6.22 and H = (S3 (cid:5) S2).2, then G is permutation isomorphic to the natural action of PΓL2(9) on 1-spaces. The following result provides more information on the groups appearing in part (i) of Proposition 3.2. Proposition 3.4. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with socle G0 = An, where Ω is the set of (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. Assume G = Sn or An and let x ∈ G be an element of prime order r. (i) We have fpr(x) (cid:2) fpr(y), where y is an r-cycle. (ii) If r > n − (cid:4) then fpr(y) = 0, otherwise fpr(y) = (cid:5) r−1(cid:16) i=0 1 − (cid:4) n − i (cid:6) (cid:5) r−1(cid:16) + α i=0 (cid:6) , 1 − n − (cid:4) n − i where α = 1 if r (cid:2) (cid:4), otherwise α = 0. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 19 Proof. Let Λ ∈ Ω be an (cid:4)-set and observe that x ﬁxes Λ if and only if the support of each r-cycle comprising x is contained in Λ or {1, . . . , n} \ Λ. In particular, the number of (cid:4)-sets ﬁxed by x is at most the number ﬁxed by a single r-cycle, which gives the bound in (i). For the expression in (ii), we may as well assume G = Sn, in which case H = S(cid:3)×Sn−(cid:3). Clearly, if r > n − (cid:4) then yG ∩ H is empty and thus fpr(y) = 0. Now assume r (cid:2) n − (cid:4). Then \|yG ∩ H\| = a(cid:3) + an−(cid:3) and \|yG\| = an, where am is the number of r-cycles in Sm. The result now follows since am = m!/(m − r)!r if r (cid:2) m, otherwise am = 0. (cid:2) Remark 3.5. Consider the expression for fpr(y) in part (ii) of Proposition 3.4 and assume r (cid:2) n − (cid:4). If we ﬁx n and r, then it is straightforward to check that fpr(y) is decreasing as a function of (cid:4), which implies that fpr(y) (cid:2) 1 − r/n, with equality if and only if (cid:4) = 1. Similarly, for r = 2 we deduce that fpr(y) (cid:3) 1/2 − 1/2n, with equality if and only if n is odd and (cid:4) = (n − 1)/2. In particular, if r = 2 then fpr(y) > 1/3 for all n and (cid:4). Notice that if r and (cid:4) are ﬁxed, then fpr(y) tends to 1 as n tends to inﬁnity. For the remainder of this section, we will assume G0 is a simple exceptional group of Lie type (recall that the classical groups will be handled in the next two sections). In Proposition 3.7 below we assume G0 (cid:8)= G2(2)(cid:7), 2G2(3)(cid:7) since these groups are isomorphic to U3(3) and L2(8), respectively. Remark 3.6. For the record, in the two excluded cases we get fpr(x) (cid:2) (r + 1)−1 unless G0 = 2G2(3)(cid:7), H ∩ G0 = 23:7 and either r = 7 (fpr(x) = 2/9) or x ∈ G \ G0 has order 3 (fpr(x) = 1/3). Note that the special cases arising here correspond (up to permutation isomorphism) to the natural actions of L2(8) or PΓL2(8) on the set of 1-dimensional subspaces of the natural module for L2(8). In particular, they are included in part (i)(d) of Theorem 1. Proposition 3.7. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0, an exceptional group of Lie type over Fq. Then fpr(x) (cid:2) (r + 1)−1 for every element x ∈ G of prime order r. Proof. Write q = pf with p a prime and let x ∈ G be an element of prime order r. Recall that the possibilities for x are as follows, where Inndiag(G0) denotes the subgroup of Aut(G0) generated by the inner and diagonal automorphisms of G0: (a) x ∈ Inndiag(G0) is either semisimple (r (cid:8)= p) or unipotent (r = p); (b) x is a graph automorphism (r (cid:2) 3 only); (c) x is a ﬁeld automorphism (q = qr 0 only); (d) x is a graph-ﬁeld automorphism (r (cid:2) 3 and q = qr 0 only). In [33], Lawther, Liebeck and Seitz conduct an extensive study of ﬁxed point ratios for exceptional groups of Lie type. In particular, [33, Theorem 1] gives an explicit upper 20 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 bound on fpr(x), which is presented as a function of q and is valid for all nontrivial x ∈ G. It is easy to check that this bound immediately reduces the problem to case (a) above with r (cid:8)= p and r (cid:3) 5. For example, if G0 = G2(q) and q (cid:8)= 2, 4 then [33, Theorem 1] gives fpr(x) (cid:2) (q2 − q + 1)−1. This is less than (q + 1)−1, which in turn is at most (r + 1)−1 in cases (b), (c) and (d), as well as case (a) with r = p. For the remainder, let x ∈ G0 be a semisimple element of prime order r (cid:3) 5 and note that x is contained in a maximal torus of G0. Here we appeal to the more reﬁned bounds presented in [33, Theorem 2], which make a distinction between three diﬀerent possibilities for the maximal subgroup H: (I) H is a maximal parabolic subgroup; (II) H is a non-parabolic maximal rank subgroup in the sense of [36]; (III) The remaining subgroups. We proceed by considering the various possibilities for G0 in turn. We ﬁrst determine an upper bound on r and we then apply the bound on fpr(x) in [33, Theorem 2], con- sidering cases (I), (II) and (III) separately if needed. This approach is eﬀective, with the exception of a handful of cases where we need to produce a slightly sharper ﬁxed point ratio estimate. For some low rank groups deﬁned over small ﬁelds, we can also use Magma [6] to verify the result. Set H0 = H ∩ G0. Case 1. G0 = E8(q) or E7(q). First assume G0 = E8(q). By expressing \|G0\|p(cid:2) as a product of cyclotomic polynomials, we note that r (cid:2) Φ30(q) = q8 + q7 − q5 − q4 − q3 + q + 1 and the result follows since [33, Theorem 1] gives fpr(x) (cid:2) q−8(q4 − 1)−1. Similarly, if G0 = E7(q) then r (cid:2) Φ7(q) = (q7 − 1)/(q − 1) and [33, Theorem 2] implies that fpr(x) (cid:2) 2q−7(q4 − 1)−1, which is suﬃcient. Case 2. G0 = E(cid:2) 6(q). Next assume G0 = E(cid:2) 6(q). Once again, by considering the order of \|G0\| we deduce that r (cid:2) q6 + (cid:3)q3 + 1. We proceed by considering cases (I)-(III) in turn, working with the appropriate upper bound on fpr(x) in [33, Theorem 2]. First suppose H is a parabolic subgroup, which is labeled in the usual way. If (cid:3) = −, or if (cid:3) = + and H is not of type P1, P6 or P1,6, then fpr(x) (cid:2) 1 q3(q3 − 2)(q2 − 1) and the result follows. Now assume (cid:3) = + and H is of type P1, P6 or P1,6. Here fpr(x) (cid:2) 1 q(q3 − 1)(q2 − 1) , T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 21 which is suﬃcient unless r + 1 > q(q3 − 1)(q2 − 1). Since r divides \|G0\|, it is easy to check that the latter inequality holds if and only if r = q6 + q3 + 1, in which case x is regular and we have \|xG ∩ H\| (cid:2) \|H0\| < q62, \|xG\| (cid:3) \|Inndiag(G0)\| q6 + q3 + 1 > 1 2 q72. This gives fpr(x) < 2q−10 < (r + 1)−1 as required. Next assume H is a maximal rank subgroup as in (II). If q (cid:3) 3 then [33, Theorem 2] gives fpr(x) (cid:2) 2q−12 and the result follows. On the other hand, if q = 2 then fpr(x) (cid:2) 2−5 and the problem is reduced to the case where (cid:3) = + and r = 73 (note that the character table of G0 is available in [8]). Here the maximal subgroups of G are determined (up to conjugacy) in [32] and by inspection we deduce that H0 = L3(8):3 is the only possibility with fpr(x) > 0. In this case, \|xG\| = \|G0\|/73 and the trivial bound \|xG ∩ H\| (cid:2) \|H0\| is suﬃcient. Finally, let us assume H is of type (III). Here we observe that [33, Theorem 2] gives fpr(x) (cid:2) q−6(q6 − q3 + 1)−1 and the result follows. Case 3. G0 = F4(q) or G2(q). First assume G0 = F4(q). Here r (cid:2) q4 + 1 and we proceed as above, working with the upper bounds on fpr(x) in [33, Theorem 2]. If H is a parabolic subgroup, then fpr(x) (cid:2) q−2(q3 − 2)−1 for H (cid:8)= P1, which is suﬃcient. On the other hand, for H = P1 we have fpr(x) (cid:2) (q4 − q2 + 1)−1 and so we may assume r (cid:3) q4 − q2 + 1. By considering the prime divisors of \|G0\|, we deduce that r = q4 − q2 + 1 or q4 + 1 are the only options. Then \|xG ∩ H\| (cid:2) \|H0\| < q37, \|xG\| (cid:3) \|G0\| q4 + 1 > 1 2 q48 and thus fpr(x) < 2q−11 < (r + 1)−1. For cases (II) and (III), it is easy to check that the bounds in [33, Theorem 2] are suﬃcient. Next suppose G0 = G2(q)(cid:7) with q (cid:3) 2. The groups with q (cid:2) 5 can be han- dled using Magma, working with the functions AutomorphismGroupSimpleGroup and MaximalSubgroups to construct G and H, and the ConjugacyClasses function to com- pute \|xG∩H\| and \|xG\|, which yields fpr(x). Now assume q (cid:3) 7 and note that r (cid:2) q2+q+1. To begin with, let us assume H is a maximal parabolic subgroup of G. If H = P1 (or P1,2) then [33, Theorem 2] gives fpr(x) (cid:2) 2(q3 + 1)−1 and the result follows. Otherwise, if H = P2 then fpr(x) (cid:2) (q2 − q + 1)−1 and so we may assume r > q2 − q, which forces r = q2 ± q + 1. Then \|xG ∩ H\| (cid:2) \|H0\| = q5(q − 1)\|SL2(q)\|, \|xG\| (cid:3) \|G2(q)\| q2 + q + 1 , which gives 22 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 fpr(x) (cid:2) q − 1 q(q + 1)(q2 − q + 1) (cid:2) 1 r + 1 as required. Next suppose H is a maximal rank subgroup of type (II). Here [33, Theorem 2] gives fpr(x) (cid:2) (q2 − q + 1)−1 and so as above we reduce to the case where r = q2 ± q + 1. By inspecting the maximal subgroups of G (for example, see [7, Tables 8.30, 8.41, 8.42]), we deduce that \|H0\| (cid:2) 2\|SU3(q)\| and by arguing as above we obtain fpr(x) (cid:2) 2q−3(q − 1)−1, which is suﬃcient. Similarly, if H is of type (III) then by [33, Theorem 2] we have fpr(x) (cid:2) q−2 and so we may assume r = q2 + q + 1. Since x generates a maximal torus of G0, it follows that every maximal subgroup of G containing x is of type (I) or (II), so this situation does not arise. Case 4. G0 = 3D4(q). Now assume G0 = 3D4(q). The case q = 2 can be checked using Magma, so we will assume q (cid:3) 3. By expressing \|G0\|p(cid:2) as a product of cyclotomic polynomials, we deduce that r (cid:2) q4 − q2 + 1. If H is a maximal parabolic subgroup, then [33, Theorem 2] gives fpr(x) (cid:2) q−2(q3−2)−1 and this bound is suﬃcient. Now suppose H is of type (II). If q = 3 then fpr(x) (cid:2) 3−4 and the result follows. Now assume q (cid:3) 4. Here fpr(x) (cid:2) (q4 −q2 +1)−1 and so we may assume r = q4 − q2 + 1. By inspecting [7, Table 8.51] we deduce that H = NG((cid:13)x(cid:14)) is the only option, whence \|xG ∩ H\| (cid:2) \|H0\| = 4(q4 − q2 + 1), \|xG\| (cid:3) \|3D4(q)\| q4 − q2 + 1 and the desired result follows. Finally, let us assume H is of type (III). Here we have fpr(x) (cid:2) (q4 − q2 + 1)−1 and so as above we may assume r = q4 − q2 + 1. But then x generates a maximal torus, so it is not contained in a subgroup of type (III). This completes the argument for the groups with socle G0 = 3D4(q). Case 5. G0 = 2F4(q)(cid:7), 2G2(q) or 2B2(q). First assume G0 = 2F4(q)(cid:7), so q = 22m+1 with m (cid:3) 0. The case q = 2 can be handled using Magma, so we may assume q (cid:3) 8. By considering \|G0\|, we observe that (cid:17) (cid:17) r (cid:2) q2 + 2q3 + q + 2q + 1. It is now straightforward to check that the bounds on fpr(x) presented in [33, Theorem 2] are suﬃcient in every case. Next suppose G0 = 2G2(q), where q = 32m+1 and m (cid:3) 1 (recall that we exclude the 3q + 1 and case G0 = 2G2(3)(cid:7) since this group is isomorphic to L2(8)). Here r (cid:2) q + once again the bounds in [33, Theorem 2] are good enough. √ Finally, suppose G0 = 2B2(q) with q = 22m+1 and m (cid:3) 1. The cases q ∈ {8, 32} can be veriﬁed using Magma, so let us assume q (cid:3) 27. If H is a Borel subgroup (that is, a subgroup of type (I)) then [33, Theorem 2] gives fpr(x) (cid:2) (q2 +1)−1 and the result follows T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 23 √ since r (cid:2) q+ and once again the result follows. (cid:2) 2q+1. Similarly, if H is of type (II) or (III) then fpr(x) (cid:2) (q2/3+1)/(q2+1) 4. Almost simple classical groups: non-subspace actions In this section we establish Theorem 1 in the case where G is an almost simple classical group in a so-called non-subspace action (see Deﬁnition 4.1 below). So let G (cid:2) Sym(Ω) be a ﬁnite primitive almost simple group with socle G0, which is a classical group over Fq with natural module V . Write q = pf with p a prime and set n = dim V . Let H be a point stabilizer in G and note that H is a maximal subgroup and G = HG0 since G0 is transitive. The main theorem on the subgroup structure of ﬁnite almost simple classical groups is due to Aschbacher [1]. Nine collections of subgroups of G are deﬁned, typically labeled C1, . . . , C8 and S, and Aschbacher proves that every maximal subgroup of G is contained in one of these collections (some adjustments are needed when G0 = PSp4(q) (with q even) or PΩ+ 8 (q), noting that a complete result in the latter case was established in later work by Kleidman [30]). Here the so-called geometric subgroups that comprise the Ci collections are deﬁned in terms of the geometry of the underlying vector space V . For example, they include the stabilizers of subspaces and appropriate direct sum and tensor product decompositions of V (see [31, Table 1.2.A] for a brief description of each geometric collection). The non-geometric subgroups in S are almost simple and irreducibly embedded in G. In [31], Kleidman and Liebeck present a great deal of information on the maximal subgroups of classical groups, including a complete description of the structure and conjugacy of the maximal geometric subgroups when n (cid:3) 13. This is complemented by work of Bray, Holt and Roney-Dougal [7], which gives complete information on the low-dimensional groups with n (cid:2) 12. It is important to note that we adopt the pre- cise deﬁnition of the Ci collection given in [31], which diﬀers slightly from Aschbacher’s original description in [1]. There is also an extensive literature on conjugacy classes in classical groups; in this regard, [16, Chapter 3] will be a convenient reference for our purposes. Deﬁnition 4.1. Let G (cid:2) Sym(Ω) be a ﬁnite primitive almost simple classical group over Fq with socle G0 and point stabilizer H. We say that the action of G on Ω is a subspace action if one of the following holds: (i) H is in the collection C1; or (ii) G0 = Spn(q), q is even and H ∩ G0 = O± n (q). In this situation, we will sometimes refer to the subgroup H as a subspace subgroup of G. Non-subspace actions and subgroups are deﬁned accordingly. 24 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Note that the subspace actions appearing in part (ii) correspond to certain subgroups in the C8 collection. If we view G0 as an orthogonal group On+1(q) with natural module W , then we may identify Ω with a set of nondegenerate hyperplanes in W , which explains why it is reasonable to view the action of G on Ω in this situation as a subspace action. In this section, our aim is to prove Theorem 1 when G is a classical group in a non- subspace action, postponing the analysis of subspace actions to Section 5. Our main result is the following. Theorem 4.2. Let G (cid:2) Sym(Ω) be an almost simple ﬁnite primitive permutation group with socle G0 and point stabilizer H. Let x ∈ G be an element of prime order r and assume G0 is a classical group and H is a non-subspace subgroup. Then either fpr(x) (cid:2) 1 r + 1 or G is permutation isomorphic to one of the groups recorded in part (i) of Theorem 1. Remark 4.3. There are only a handful of exceptions to the main bound in Theorem 4.2. For example, if G = L2(7) and H = S4 is a C6-subgroup of type 21+2 2 (2), then fpr(x) = 3/7 if x is an involution. But here G is permutation isomorphic to L3(2) acting on 1-dimensional subspaces of the natural module, so this case is included in part (i)(d) of Theorem 1. Similarly, if G = Ω+ 8 (2) and H = Sp6(2) is an irreducibly embedded subgroup in S, then fpr(x) = 3/10 when x = (Λ4) is an element of order 3 with a trivial 1-eigenspace. Here the action is permutation isomorphic to the action on nonsingular 1-spaces, so once again this corresponds to a case in (i)(d) of Theorem 1. We also refer the reader to the statement of Lemma 4.12 for further examples with G0 = L(cid:2) − .O− 4(q). Our main reason for making the distinction between subspace and non-subspace ac- tions is encapsulated in the following result, which only applies in the non-subspace setting. This is [12, Theorem 1], which is proved in the sequence of papers [13–15]. It will be our main tool in this section. Theorem 4.4. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple classical primitive permutation group with point stabilizer H and socle G0. Assume H is a non-subspace subgroup. Then fpr(x) < \|xG\|− 1 2 + 1 n +ι for all x ∈ G of prime order, where n is the dimension of the natural module for G0 and either ι = 0 or ι is listed in [12, Table 1]. For non-subspace actions of classical groups, the bound in Theorem 4.4 can be viewed as a signiﬁcant strengthening of the following more general result, which is due to Liebeck and Saxl (see [34, Theorem 1]). It is worth noting that almost all of the special cases appearing in [34, Table 1] involve groups with socle L2(q). T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 25 Theorem 4.5. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0, which is a simple group of Lie type over Fq. Then either fpr(x) (cid:2) 4 3q for all nontrivial x ∈ G, or (G, H, x) is one of the cases appearing in [34, Table 1]. Finally, we need one more deﬁnition before we are ready to begin the proof of Theo- rem 4.2. Recall that V is the natural module for G0, which is an n-dimensional vector space over Fqu, where u = 2 if G0 = Un(q) and u = 1 in all other cases. Deﬁnition 4.6. Let K be the algebraic closure of Fqu. For any element x ∈ G ∩ PGL(V ), write x = ˆxZ with ˆx ∈ GLn(qu) and Z = Z(GLn(qu)), and let ν(x) be the codimension of the largest eigenspace of ˆx as an element of GLn(K). Note that this is independent of the choice of ˆx. Let x ∈ G be an element of prime order r. If x ∈ G ∩ PGL(V ), then x is either unipotent (if r = p) or semisimple (if r (cid:8)= p), and we refer the reader to [13, Section 3] for bounds on \|xG\| in terms of n, q and ν(x). In addition, we will use the notation from [16, Chapter 3] to describe representatives of the conjugacy classes of unipotent and semisimple elements of prime order. In particular, if G0 is a symplectic or orthogonal group in even characteristic then we adopt the notation from [2] for the conjugacy classes of unipotent involutions. Now suppose x /∈ PGL(V ). Here x is either a ﬁeld, graph or graph-ﬁeld automorphism of G0 and once again we refer the reader to [13, Section 3] for bounds on \|xG\|. Let us also observe that if x is a ﬁeld or graph-ﬁeld automorphism, then q = qr 0 and unless we are in one of the handful of special cases recorded in [34, Table 1], we deduce that fpr(x) (cid:2) 4 3q (cid:2) 4 3 · 2r (cid:2) 1 r + 1 (4) via Theorem 4.5. We are now ready to begin the proof of Theorem 4.2. First we handle the linear and n(q) and n (cid:3) 5. For small values of n, the bound unitary groups with socle G0 = L(cid:2) in Theorem 4.4 is less eﬀective and we will consider the groups with n ∈ {2, 3, 4} in Lemmas 4.10, 4.11 and 4.12 below. Lemma 4.7. Suppose G0 = L(cid:2) fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. n(q) with n (cid:3) 5. If H is a non-subspace subgroup, then Proof. Let x ∈ G be an element of prime order r and note that ι (cid:2) 1/n in Theorem 4.4 (see [12, Table 1]), whence 26 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 fpr(x) < \|xG\|− 1 2 + 2 n . (5) First assume x /∈ PGL(cid:2) n(q), so either r = 2 or q = qr 0. By [13, Corollary 3.49] we have \|xG\| > (cid:6) (cid:5) 1 2 q q + 1 1 2 (n2−n−4) q and by combining this bound with (5) we obtain fpr(x) (cid:2) (q +1)−1 (cid:2) (r +1)−1 for n (cid:3) 6. And for n = 5 we have ι = 0 (see [12, Table 1]) and we deduce that fpr(x) (cid:2) (q + 1)−1 via the bound in Theorem 4.4. For the remainder, we may assume x ∈ PGL(cid:2) n(q). First observe that \|xG\| > (cid:6) (cid:5) 1 2 q q + 1 q2n−2 by [13, Corollary 3.38] and by combining this bound with (5) we deduce that if n (cid:3) 6 then fpr(x) (cid:2) (q + 2)−1 unless (n, q) = (6, 2). Similarly, if n = 5 then ι = 0 and we get fpr(x) (cid:2) (q + 2)−1 unless q = 2. The groups with n ∈ {5, 6} and q = 2 can be handled using Magma, so it remains to consider semisimple elements of odd prime order. Let i (cid:3) 1 be minimal such that r divides qi − 1. Notice that if i (cid:2) 2 then r + 1 (cid:2) q + 2 and the result follows as above, so we may assume i (cid:3) 3. In particular, this forces ν(x) (cid:3) 3 and thus [13, Corollary 3.38] implies that \|xG\| > (cid:6) (cid:5) 1 2 q q + 1 q6n−18. By considering \|G0\| we see that r (cid:2) (qn − 1)/(q − 1) and one can check that this lower bound on \|xG\| (combined with (5)) is suﬃcient if n (cid:3) 9. In fact, the same bound is also eﬀective if n = 8 and q (cid:3) 4. It is easy to check that if n = 8 and q = 3 then r (cid:2) 1093, whereas r (cid:2) 127 if q = 2; the previous argument now goes through. Similarly, if n = 7 then ι = 0 in Theorem 4.4 and once again the result follows as above. To complete the proof, we may assume n ∈ {5, 6} (with i (cid:3) 3 as above). If n = 5 then ι = 0, \|xG\| > 1 2 q18 (minimal if (cid:3) = + and i = 3) and we conclude by applying the bound in Theorem 4.4. Now assume n = 6 and recall that we may assume q (cid:3) 3 since we have already handled the case q = 2 using Magma. If ι = 0 then the previous argument applies, so let us assume ι > 0, in which case ι = 1/6 and H is of type Sp6(q) (see [12, Table 1]). Here we have i ∈ {3, 4, 6} and it is easy to check that \|xG\| > 1 2 q24 and r (cid:2) q2 + q + 1. By applying the bound in (5) we deduce that fpr(x) (cid:2) (r + 1)−1 and the argument is complete. (cid:2) In the next lemma we assume G0 = PSpn(q) with n (cid:3) 6, noting that the case n = 4 is handled separately in Lemma 4.13 below. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 27 Lemma 4.8. Suppose G0 = PSpn(q) with n (cid:3) 6. If H is a non-subspace subgroup, then fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. Proof. Suppose x ∈ G has prime order r and note that ι (cid:2) 1/n, so (5) holds once again. Now \|xG\| (cid:3) (qn − 1)/2 (minimal if G = G0, q is odd and x is a transvection) and one can check that (5) yields fpr(x) (cid:2) (q + 2)−1 unless n = 6 or (n, q) = (8, 2). The latter case, together with (n, q) = (6, 2), can be handled using Magma. For n = 6 with q (cid:3) 3 we have ι = 0 and the upper bound in Theorem 4.4 is suﬃcient. For the remainder, we may assume x is semisimple and r (cid:3) 5. Let i (cid:3) 1 be minimal such that r divides qi − 1. By arguing as above, we may assume that r > q + 1, so i (cid:3) 3 and thus ν(x) (cid:3) 4. Notice that r (cid:2) qn/2 + 1 and [13, Proposition 3.36] gives \|xG\| > 1 2 q4n−16. One can now check that the bound in (5) is suﬃcient for n (cid:3) 10, so the problem is reduced to the groups where n ∈ {6, 8} and q (cid:3) 3. Suppose n = 6. Here ι = 0, i ∈ {3, 4, 6} and 2 q16 (minimal if i = 4). In addition, r (cid:2) q2 + q + 1 and the result follows via \|xG\| > 1 the bound in Theorem 4.4. Similarly, if n = 8 then r (cid:2) q4 + 1, \|xG\| > 1 2 q24 (once again, minimal if i = 4) and the bound in (5) is suﬃcient. (cid:2) Lemma 4.9. Suppose G0 = PΩ(cid:2) fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. n(q) with n (cid:3) 7. If H is a non-subspace subgroup, then Proof. This is very similar to the proof of the previous lemma. First assume n is odd (so q is also odd) and write n = 2m + 1. The case (n, q) = (7, 3) can be handled using Magma, so we may assume (n, q) (cid:8)= (7, 3). By inspecting [12, Table 1] we see that ι = 0 if n (cid:3) 9, otherwise ι (cid:2) 0.108. Let us also note that \|xG\| (cid:3) \|SOn(q)\| − n−1(q)\| 2\|SO = 1 2 qm(qm − 1) with equality if x ∈ SOn(q) is an involution with a minus-type eigenspace on V of dimension n − 1. By applying Theorem 4.4 we deduce that fpr(x) (cid:2) (q + 2)−1. To complete the argument for n odd, we may assume x is semisimple, r (cid:3) 5 and i (cid:3) 3, where i is the smallest positive integer such that r divides qi − 1. In particular, we have ν(x) (cid:3) 4 and we quickly deduce that \|xG\| (cid:3) \|SOn(q)\| \|SOn−4(q)\|\|GU1(q2)\| > 1 2 q4n−12. In addition, we note that r (cid:2) 1 now follows via Theorem 4.4. 2 (qm + 1) and it is routine to check that the desired bound For the remainder, we may assume n = 2m (cid:3) 8 is even. The groups with (n, q) = (8, 2) or (8, 3) can be handled using Magma. Let us highlight the special case G = Ω+ 8 (2) with H = Sp6(2) acting irreducibly on V : if x ∈ G0 has order 3 and CV (x) = 0, then fpr(x) = 3/10 > 1/4. Here the action of G on Ω is permutation isomorphic to the action 28 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 of G on the set of 1-dimensional nonsingular subspaces of the natural module, so this special case is included in part (i)(d) of Theorem 1. For the remainder we may assume (n, q) (cid:8)= (8, 2), (8, 3). We will postpone the analysis of the special case where (n, (cid:3)) = (8, +) and H is an irreducible subgroup with socle Sp6(q) (if p = 2) or Ω7(q) (if p is odd) to the end of the proof. By excluding this special case, we observe that ι (cid:2) 1/(n − 2) in Theorem 4.4 and it is easy to check that \|xG\| (cid:3) \|O(cid:2) n(q)\| 2d\|Spn−2(q)\| = 1 d qm−1(qm − (cid:3)) with d = (2, q − 1). By combining this lower bound with Theorem 4.4, setting ι = 1/(n − 2), we deduce that fpr(x) (cid:2) (q + 2)−1. Therefore, to complete the analysis we may assume x is semisimple, r (cid:3) 5 and i (cid:3) 3, so ν(x) (cid:3) 4 and we have \|xG\| (cid:3) \|SO+ n (q)\| − n−4(q)\|\|GU1(q2)\| \|SO > 1 2 q4n−12. (6) Now r (cid:2) qm + 1 and by applying Theorem 4.4 we deduce that fpr(x) (cid:2) (r + 1)−1 as required. Finally, to complete the proof we may assume (n, (cid:3)) = (8, +), q (cid:3) 4 and H is ir- reducible with socle Sp6(q) (if p = 2) or Ω7(q) (if p is odd). Here [12, Table 1] gives ι = 0.219. If r = 2 then the bound in Theorem 4.5 is suﬃcient, so we may assume r is odd and thus \|xG\| (cid:3) 8 (q)\| \|SO+ 4 (q)\|\|Sp2(q)\| q9\|SO+ = (q2 + 1)2(q6 − 1) (minimal if x is unipotent with Jordan form (J 2 1 )). Then as above, by applying the lower bound in Theorem 4.4 with ι = 0.219, we deduce that fpr(x) (cid:2) (q + 2)−1. Finally, we may assume x is semisimple, r (cid:3) 5 and i (cid:3) 3, so i ∈ {3, 4, 6} and r (cid:2) q2 + q + 1. We can now proceed as before, using the lower bound on \|xG\| in (6). (cid:2) 2 , J 4 In order to complete the proof of Theorem 1 for classical groups in non-subspace actions, we may assume G0 is one of the following groups: L2(q), L(cid:2) 3(q), L(cid:2) 4(q), PSp4(q). First assume G0 = L2(q). We refer the reader to [7, Tables 8.1, 8.2] for a convenient list of the maximal subgroups of G up to conjugacy. We will assume q (cid:3) 7 and q (cid:8)= 9, noting that G0 is isomorphic to an alternating group when q = 4, 5 or 9. Note that in the special case arising in the following lemma, G is permutation isomorphic to L3(2) acting on the set of 1-dimensional subspaces of its natural module. In particular, this case is included in part (i)(d) of Theorem 1. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 29 Lemma 4.10. Suppose G0 = L2(q) and H is a non-subspace subgroup, where q (cid:3) 7 and q (cid:8)= 9. If x ∈ G has prime order r, then either fpr(x) (cid:2) (r + 1)−1, or G = L2(7), H = S4, x is an involution and fpr(x) = 3/7. Proof. We need to consider the various possibilities for H arising in [7, Tables 8.1, 8.2]. We begin by handling the subﬁeld subgroups. Case 1. H is a subﬁeld subgroup. Suppose H is a subﬁeld subgroup of type GL2(q0), where q = qk 0 with k a prime and q0 (cid:3) 3 (see [7, Table 8.1]). Let x ∈ G be an element of prime order r and note that H ∩ PGL(V ) (cid:2) PGL2(q0). Let im(X) be the number of elements of order m in the ﬁnite group X. If r = p and x is unipotent, then \|xG ∩ H\| (cid:2) q2 0 − 1 and \|xG\| (cid:3) (q2 − 1)/2, whence fpr(x) (cid:2) 2(q2 − 1) 0 q2 − 1 (cid:2) 2 q2 0 + 1 (cid:2) 1 q0 + 1 and the result follows. Similarly, if x is a semisimple involution, then we observe that \|xG ∩ H\| (cid:2) i2(PGL2(q0)) = q2 0 and \|xG\| (cid:3) q(q − 1)/2, which implies that fpr(x) (cid:2) 1/3 as required. Next assume x is semisimple and r is odd. As usual, we may assume r divides \|H0\|, − 1. If r divides q0 − 1 then which implies that r divides q2 0 fpr(x) = \|xG0 ∩ H0\| \|xG0\| = q0(q0 + 1) q(q + 1) (cid:2) q0 + 1 q0(q2 0 + 1) (cid:2) 1 q0 and the result follows since r (cid:2) q0 − 1. Similarly, if r divides q0 + 1 then fpr(x) (cid:2) q0 − 1 0 + 1) (cid:2) 1 q0 + 2 q0(q2 and once again this bound is suﬃcient. Finally, suppose q = qr 1 and x is a ﬁeld automorphism. First assume k is odd. If r = 2 then \|xG ∩ H\| (cid:2) \|PGL2(q0)\| \|PGL2(q1/2 )\| 0 = q1/2 0 (q0 + 1), \|xG\| (cid:3) 1 2 q1/2(q + 1) and we quickly deduce that fpr(x) (cid:2) 1/3. Similarly, if r (cid:3) 3 then which implies that \|xG ∩ H\| (cid:2) \|PGL2(q0)\| < q, \|xG\| > 1− 1 r (cid:4) , (cid:3) q3 1 2 fpr(x) (cid:2) 2q3−2r 1 (cid:2) 24−2r (cid:2) (r + 1)−1. 30 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Now assume k = 2. If r = 2 then \|xG ∩ H\| (cid:2) i2(PGL2(q0)) + 1 (cid:2) q + 1, \|xG\| (cid:3) 1 2 q1/2(q + 1) and we obtain fpr(x) (cid:2) 2q−1 0 . This is at most 1/3 if q0 (cid:3) 7 and the remaining cases q0 ∈ {4, 5} can be checked directly (recall that q (cid:8)= 4, 9). Finally, if r (cid:3) 3 then q1 = q2 2 and we have \|xG ∩ H\| (cid:2) \|PGL2(q0)\| < q3/2, \|xG\| = \|PGL2(q)\| \|PGL2(q1/r)\| > (q + 1)2. Therefore, fpr(x) < q−1 0 (cid:2) 2−r and the result follows. Case 2. H is of type GL1(q) (cid:5) S2 or GL1(q2). Here H is the normalizer of a maximal torus of G0 and we have H∩PGL(V ) (cid:2) D2(q−(cid:2)), where (cid:3) = 1 if H is of type GL1(q) (cid:5) S2, otherwise (cid:3) = −1. Let x ∈ H be an element of prime order r. First assume x is semisimple or unipotent. If r = 2 then \|xG ∩H\| (cid:2) i2(D2(q−(cid:2))) (cid:2) q +2 and \|xG\| (cid:3) q(q−1)/2. These bounds yield fpr(x) (cid:2) 1/3 for q (cid:3) 11 and the cases q ∈ {7, 8} can be checked directly. On the other hand, if r is odd then r divides q − (cid:3) and the result follows since \|xG0 ∩ H\| = 2 and \|xG\| (cid:3) q(q − 1). Finally, suppose q = qr 0 and x is a ﬁeld automorphism. If r (cid:3) 3 then \|xG ∩H\| (cid:2) 2(q+1) and \|xG\| > (q + 1)2, whence fpr(x) (cid:2) 2(q + 1)−1 < 21−r and the result follows. For r = 2 we have fpr(x) = 0 if (cid:3) = −1, whereas \|xG ∩ H\| (cid:2) 2q1/2 and \|xG\| (cid:3) q1/2(q + 1)/2 if (cid:3) = 1. From the latter bounds we obtain fpr(x) (cid:2) 1/3 since q (cid:3) 16. Case 3. The remaining possibilities for H. − .O− First assume H is of type 21+2 2 (2), so H = A4 or S4, q = p (cid:3) 7 and we may assume r ∈ {2, 3}. Here \|xG ∩ H\| (cid:2) ir(H) (cid:2) 9 and \|xG\| (cid:3) q(q − 1)/2; these bounds are suﬃcient unless r = 2 and q = 7. In the latter case, G = L2(7), H = S4, \|xG ∩ H\| = 9 and \|xG\| = 21, which gives fpr(x) = 3/7 > 1/3. This is the special case recorded in the statement of the lemma. Finally, suppose H = S5 or A5, q ∈ {p, p2} and p ≡ ±1, ±3 (mod 10), so r ∈ {2, 3, 5}. First assume r = 2, so \|xG\| (cid:3) q1/2(q + 1)/2 (minimal if x is an involutory ﬁeld automor- phism) and we note that i2(H) (cid:2) 25. The subsequent bound on fpr(x) is less than 1/3 if q (cid:3) 29 and the cases with q < 29 can be checked very easily with the aid of Magma. Similarly, if r ∈ {3, 5} then the bounds \|xG ∩ H\| (cid:2) 24 and \|xG\| (cid:3) q(q − 1) are suﬃcient for q (cid:3) 13 and once again we can use Magma when q < 13. (cid:2) Lemma 4.11. Suppose G0 = L(cid:2) (r + 1)−1 for all x ∈ G of prime order r. 3(q) and H is a non-subspace subgroup. Then fpr(x) (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 31 Proof. Since U3(2) is solvable and L3(2) ∼ = L2(7), we may assume q (cid:3) 3. The groups with 3 (cid:2) q (cid:2) 13 can be checked using Magma, so for the remainder, we may assume q (cid:3) 16. Let x ∈ G be an element of prime order r and set H0 = H ∩ G0. By Theorem 4.5, we have fpr(x) (cid:2) 4/3q (there are no exceptions in [34, Table 1] with 3(q) and q (cid:3) 16). As a consequence, we may assume r (cid:3) 13 divides \|H0\| and G0 = L(cid:2) x ∈ G0 is either semisimple or unipotent (see (4)). Therefore, \|xG\| (cid:3) \|GU3(q)\| q3\|GU1(q)\|2 = (q3 + 1)(q − 1), (7) with equality if (cid:3) = −, r = p and x has Jordan form (J2, J1). Let us also note that r (cid:2) q2 + q + 1. If r (cid:8)= p then let i (cid:3) 1 be minimal such that r divides qi − 1. By inspecting [7, Tables 8.3-8.6], recalling that we may assume r divides \|H0\|, we can reduce to the cases where H is a geometric subgroup in one of the collections C2, C3, C5 or C8 (recall that we follow [31] in deﬁning the various subgroup collections arising in Aschbacher’s theorem [1], which is consistent with [7]). Suppose H is a C2-subgroup of type GL(cid:2) 1(q) (cid:5) S3. Since we may assume r divides \|H0\|, it follows that r (cid:2) q + 1 and the trivial bound \|xG ∩ H\| (cid:2) (q + 1)2 with (7) is suﬃcient. Next assume H is a C3-subgroup of type GL(cid:2) 1(q3). Here r (cid:8)= p and ((cid:3), i) = (+, 3) or (−, 6) since r > 3. Moreover, \|xG0 ∩ H\| = 3 and the result follows via (7). Now assume H is a subﬁeld subgroup of type GL(cid:2) 0 and k is an odd prime. Here \|H0\| < q8/3 and r is at most q2 0 +q0 +1, which implies that r (cid:2) q2/3 +q1/3 +1. It is easy to check that the trivial bound \|xG ∩ H\| < q8/3 combined with the lower bound on \|xG\| in (7) is suﬃcient. Similarly, if H is of type O3(q), then q is odd, r (cid:2) (q + 1)/2 and \|H0\| = q(q2 − 1). Once again, the bound in (7) is eﬀective. 3(q0), where q = qk To complete the proof of the lemma, we may assume (cid:3) = +, q = q2 0 and H is a subgroup of type GL(cid:2)(cid:2) 3 (q0). Note that r (cid:2) q + q1/2 + 1. If ν(x) = 2 then \|xG\| (cid:3) q(q2 − 1)(q3 − 1)/3 (minimal if r = p and x has Jordan form (J3)) and the result follows since \|H0\| < q4. On the other hand, if ν(x) = 1 then either r = p and x has Jordan form (J2, J1), or r (cid:2) q0 + 1 and x = (I2, ω) (up to conjugacy), where ω is a primitive r-th root of unity. 0 + 1) and \|xG\| = (q + 1)(q3 − 1). And in the In the former case, \|xG ∩ H\| (cid:2) (q0 − 1)(q3 latter, we have \|xG0 ∩ H\| (cid:2) q2 0 + q0 + 1) and \|xG\| (cid:3) q2(q2 + q + 1). In both cases, it is routine to check that the given bounds are suﬃcient. (cid:2) 0(q2 Note that in the next lemma we can assume G0 (cid:8)= L4(2) since L4(2) ∼ = A8. Let us also observe that each special case appearing in the statement is permutation isomorphic to a subspace action of an isomorphic classical group. For example, in (i), the action is permutation isomorphic to the action of an almost simple group with socle PSp4(3) on the set of 2-dimensional totally singular subspaces of the natural module. Similarly, in (ii) with the action of O− 6 (3).2 (extended by an involutory graph automorphism) on a set of nondegenerate 1-spaces in (iii). In addition, let us say that an involutory graph automorphism x of G0 = L(cid:2) 4(q) is of symplectic-type if CG0(x) has socle PSp4(q). 6 (2) on the set of nonsingular 1-spaces, and PΩ+ 32 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 4(q) and H is a non-subspace subgroup. Assume G0 (cid:8)= L4(2) Lemma 4.12. Suppose G0 = L(cid:2) and let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) (r + 1)−1, or r = 2 and one of the following holds: (i) G0 = U4(2), H is of type GU1(2) (cid:5) S4, x = (J2, J 2 (ii) G = U4(2).2, H is of type Sp4(2), x is a symplectic-type graph automorphism and 1 ) and fpr(x) = 2/5; fpr(x) = 4/9; or (iii) G = L4(3).22, H is of type Sp4(3), x is a symplectic-type graph automorphism and fpr(x) = 5/13. Proof. The result for q (cid:2) 7 can be checked using Magma, so we will assume q (cid:3) 8. Let x ∈ G be an element of prime order r. In view of Theorem 4.5, we may assume r (cid:3) 7 and x ∈ G0 is semisimple or unipotent. Note that r (cid:2) q2 + q + 1. Let us also observe that \|xG\| (cid:3) \|GL4(q)\| \|GL3(q)\|\|GL1(q)\| = q3(q2 + 1)(q − 1), (8) minimal if (cid:3) = + and x is semisimple with ν(x) = 1. By inspecting [7, Tables 8.8-8.11], noting that we may assume r divides \|H0\|, we deduce that H is either contained in one of the geometric subgroup collections labeled C(cid:3) with (cid:4) ∈ {2, 3, 5, 8}, or H ∈ S is a non-geometric subgroup with socle L2(7) or A7. For the non-geometric subgroups we have r = 7, q = p (cid:3) 11 and the bound in Theorem 4.5 is suﬃcient. We now consider the remaining possibilities for H in turn. As usual, if x is semisimple then we deﬁne i to be the smallest positive integer such that r divides qi − 1. It will be useful to note that the constant ι in Theorem 4.4 is zero, unless H is a C8-subgroup of type Sp4(q), in which case ι = 1/4. First assume H is a C2-subgroup of type GL(cid:2) 2(q) (cid:5) S2. In the former case, we have r (cid:2) q + 1 and the trivial bound \|xG ∩ H\| (cid:2) (q + 1)3 combined with (8) is 2(q) (cid:5) S2 and note that r (cid:2) q + 1 once again. If suﬃcient. Now assume H is of type GL(cid:2) ν(x) = 1 then 1(q) (cid:5) S4 or GL(cid:2) \|xG0 ∩ H\| (cid:2) 2 (cid:6) (cid:5) \|GL2(q)\| \|GL1(q)\|2 = 2q(q + 1) (maximal if (cid:3) = + and x is semisimple) and the bound in (8) is suﬃcient. Now assume ν(x) (cid:3) 2. Here \|xG\| (cid:3) \|GL4(q)\| 2q4\|GL2(q)\| = 1 2 q(q3 − 1)(q4 − 1) (9) (minimal if (cid:3) = +, G = G0 and x is unipotent with Jordan form (J 2 2 )). By applying the bound in Theorem 4.4, noting that ι = 0, we deduce that fpr(x) (cid:2) (q + 2)−1 and the result follows. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 33 Next let us assume H is of type GL2(q2). Here r (cid:2) q2 + 1 and ν(x) (cid:3) 2 for all x ∈ H0. In particular, if r (cid:2) q + 1 then the previous argument applies (using Theorem 4.4 with ι = 0). Now assume r > q + 1, so i = 4 and x is a regular semisimple element. Here \|xG\| (cid:3) \|GL4(q)\| \|GL1(q4)\| = q6(q − 1)(q2 − 1)(q3 − 1) (10) and once again the desired result follows by applying Theorem 4.4. Now suppose H is of type GL(cid:2)(cid:2) 4 (q0), where q = qk 0 and k is a prime. Here we have r (cid:2) q2 0 + q0 + 1 (cid:2) q + q1/2 + 1 and the result follows via Theorem 4.4, using the lower bound on \|xG\| in (8). In order to complete the proof of the lemma, we may assume H is a C8-subgroup of type Sp4(q) or O(cid:2)(cid:2) First assume H is of type O(cid:2)(cid:2) 4 (q), in which case q is odd. If r (cid:2) q + 1 then the usual argument (using Theorem 4.4 and (8)) is suﬃcient. On the other hand, if r > q + 1 then (cid:3)(cid:7) = −, x is semisimple, i = 4, r (cid:2) (q2 + 1)/2 and the bound in (10) is satisﬁed. We now conclude by applying Theorem 4.4. 4 (q). Finally, suppose H is of type Sp4(q). Here ι = 1/4, so the bound in Theorem 4.4 is not useful and we need to consider the various possibilities for x in turn. Fortunately, the embedding of H in G is transparent and it is easy to determine good bounds on fpr(x). First assume r = p. If x has Jordan form (J2, J 2 1 ), then \|xG ∩ H\| (cid:2) \|Sp4(q)\| q3\|Sp2(q)\| = q4 − 1 and the bound \|xG\| (cid:3) (q2 − q + 1)(q4 − 1) is suﬃcient. Similarly, if x = (J 2 2 ) then \|xG ∩ H\| (cid:2) \|Sp4(q)\| 2 (q)\| q3\|O+ + \|Sp4(q)\| q3\|O− 2 (q)\| = q2(q4 − 1) and the result follows since (9) holds. And for x = (J4), the bounds \|xG ∩ H\| < q8 and \|xG\| > 1 2 q12 are suﬃcient (here we are using the fact that PGSp4(q) contains precisely q8 unipotent elements). Now suppose r (cid:8)= p, so i ∈ {1, 2, 4} and ν(x) ∈ {2, 3}. If ν(x) = 3 then x is regular, r (cid:2) q2 + 1, \|xG0 ∩ H\| < 2q8 and the bound \|xG\| > 1 2 q12 is suﬃcient. Now assume ν(x) = 2, so i ∈ {1, 2} and r (cid:2) q + 1. Here it is straightforward to verify the bounds 2 q8, whence fpr(x) < 4q−2 (cid:2) (q + 2)−1 and the result \|xG0 ∩ H\| < 2q6 and \|xG\| > 1 follows. (cid:2) Finally, we handle the almost simple symplectic groups with socle PSp4(q). Note that we may assume q (cid:3) 4 since PSp4(2)(cid:7) ∼ = A6 and PSp4(3) ∼ = U4(2). Lemma 4.13. Suppose G0 = PSp4(q) and H is a non-subspace subgroup, where q (cid:3) 4. Then fpr(x) (cid:2) (r + 1)−1 for all elements x ∈ G of prime order r. 34 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Proof. Let x ∈ G be an element of prime order r and recall that we may assume r divides \|H0\|, where H0 = H ∩ G0. The groups with q (cid:2) 16 can be checked using Magma, so we may assume q (cid:3) 17. In addition, by applying Theorem 4.5, we may assume that r (cid:3) 13 and x ∈ G0 is either semisimple or unipotent. Note that either r = p and \|xG\| (cid:3) \|Sp4(q)\| dq3\|Sp2(q)\| = 1 d (q4 − 1), where d = (2, q − 1), or r (cid:8)= p and \|xG\| (cid:3) \|Sp4(q)\| \|GU2(q)\| = q3(q − 1)(q2 + 1). (11) (12) We now partition the argument into two cases, according to the parity of q. Case 1. q odd. To begin with, we will assume q is odd. By inspecting [7, Tables 8.12, 8.13], we may assume H is either a geometric subgroup in one of the collections C2, C3 or C5, or H is a non-geometric subgroup with socle L2(q). We consider each possibility in turn. First assume H is a C2-subgroup of type Sp2(q) (cid:5) S2. Suppose x is unipotent, so r = p and x has Jordan form (J2, J 2 1 ) or (J 2 2 ) since we may assume x ∈ H. In the ﬁrst case, (cid:5) (cid:6) \|xG ∩ H\| (cid:2) 2 = 2(q2 − 1) \|Sp2(q)\| q and the bound in (11) is suﬃcient. Similarly, if x = (J 2 the result follows since 2 ) then \|xG ∩ H\| (cid:2) (q2 − 1)2 and \|xG\| (cid:3) \|Sp4(q)\| q3\|O− 2 (q)\| = 1 2 q(q − 1)(q4 − 1). (13) Now assume r (cid:8)= p and note that r (cid:2) q + 1. Here \|xG0 ∩ H\| (cid:2) (cid:6) 2 (cid:5) \|Sp2(q)\| \|GL1(q)\| = q2(q + 1)2 and the bound in (12) is suﬃcient. Next suppose H is of type GL(cid:2) 2 ) is the only option and it is easy to check that the bounds \|xG ∩ H\| (cid:2) q2 − 1 and (13) are eﬀective. Similarly, if x is semisimple then 2(q), so r (cid:2) q + 1. If r = p then x = (J 2 \|xG0 ∩ H\| (cid:2) 2 (cid:6) (cid:5) \|GL2(q)\| \|GL1(q)\|2 = 2q(q + 1) and the result follows via (12). T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 35 A similar argument applies when H is of type Sp2(q2). Indeed, if r = p then x = (J 2 2 ), \|xG ∩ H\| (cid:2) q4 − 1 and we conclude by applying the bound in (13). For r (cid:8)= p we have r (cid:2) q2 + 1 and the bounds \|xG0 ∩ H\| (cid:2) q2(q2 + 1) and (12) are suﬃcient. To complete the argument for q odd, we may assume that H is either a subﬁeld subgroup of type Sp4(q0), where q = qk 0 with k a prime, or H is a non-geometric subgroup with socle L2(q) and p (cid:3) 5. The latter case is easy to handle. Indeed, we have \|H0\| < q3, r (cid:2) q +1 and it is easy to check that the nontrivial unipotent elements in H0 have Jordan form (J4) on the natural module V for G0 (this follows from the fact that V = S3(W ), where W is the natural 2-dimensional module for H0). Therefore, the bound in (12) holds for all x ∈ H of prime order and the result follows. Now assume H is a subﬁeld subgroup of type Sp4(q0), where q = qk 1 ) then \|xG ∩H\| (cid:2) q4 0 0 . First assume −1 (cid:2) q4/3 −1 k (cid:3) 3, in which case r (cid:2) q2 and the bound in (11) is suﬃcient. For the remaining elements, the bound in (13) is satisﬁed and the trivial bound \|xG ∩ H\| (cid:2) \|H0\| < q10 0 0 +1 (cid:2) q2/3 +1. If x = (J2, J 2 (cid:2) q10/3 is good enough. Finally, suppose k = 2. If r = p, then either x = (J2, J 2 1 ), \|xG ∩ H\| (cid:2) q2 − 1 and (11) holds, or \|xG ∩ H\| < q8 0 = q4 (this upper bound is the total number of unipotent elements in H0) and we have the bound on \|xG\| in (13). In both cases, the given bounds are suﬃcient. Now assume r (cid:8)= p and note that r (cid:2) q + 1. If ν(x) = 3 then \|xG\| > 1 2 q8 and the trivial bound \|xG ∩ H\| < \|H0\| < q5 is good enough. Similarly, if ν(x) = 2 then (12) holds and the result follows since \|xG0 ∩ H\| (cid:2) \|Sp4(q0)\| \|GL2(q0)\| = q3/2(q1/2 + 1)(q + 1). Case 2. q even. To complete the proof of the lemma, we may assume q (cid:3) 32 is even. In view of Theorem 4.5, we may also assume that r (cid:3) 23. In particular, x is semisimple and (12) holds. We now work through the various possibilities for H arising in [7, Table 8.14]. First assume H is a Borel subgroup, so H0 = [q4]:C 2 q−1 and thus r (cid:2) q − 1 (note that 2 q8 and \|xG0 ∩ H\| = 8q4 as H is maximal when G (cid:8)(cid:2) ΓSp4(q)). If ν(x) = 3, then \|xG\| > 1 explained in the proof of [10, Lemma 5.8]. Similarly, if ν(x) = 2 then \|xG0 ∩ H\| = 4q3 and we have the bound on \|xG\| in (12). In both cases, the given bounds are suﬃcient. The argument when H is of type Sp2(q) (cid:5) S2, Sp2(q2) or Sp4(q0) is entirely similar to the one given above in the case where q is odd. For this reason, we omit the details. Next suppose H is a non-geometric subgroup with socle 2B2(q), in which case log2 q is odd and we note that \|H0\| < q5 and r (cid:2) q + 2 q8 and the trivial bound \|xG ∩ H\| < q5 is suﬃcient. Similarly, if ν(x) = 2 then r divides q − 1, 2q + 1. If ν(x) = 3 then \|xG\| > 1 √ \|xG\| (cid:3) \|Sp4(q)\| \|GL2(q)\| = q3(q + 1)(q2 + 1) and once again the trivial bound \|xG ∩ H\| < q5 is good enough. 36 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Finally, let us assume H = NG(T ) is the normalizer of a maximal torus (recall that the C8-subgroups of type O(cid:2) 4(q) are subspace subgroups, so they are excluded here; see Remark 4.14). Here H0 < M < G0 for some maximal non-subspace subgroup M of G0 (indeed, H is maximal only if G (cid:8)(cid:2) ΓSp4(q)) and so the desired bound on fpr(x) automatically holds by our earlier work in this proof. (cid:2) Remark 4.14. Using the same approach as in the proof of Lemma 4.13, it is straightfor- ward to show that if G0 = PSp4(q), q (cid:3) 4 is even and H is a subspace subgroup of type 4(q), then fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. See Lemma 5.20 for the O(cid:2) details. 5. Almost simple classical groups: subspace actions In this section we handle the subspace actions of classical groups, which will complete the proof of Theorem 1 for almost simple groups. Recall from Deﬁnition 4.1 that the subspace actions correspond to the groups where a point stabilizer H is either contained in Aschbacher’s C1 collection of maximal subgroups, or G0 = Spn(q) is a symplectic group with q even and H ∩ G0 = O(cid:2) n(q) is a naturally embedded orthogonal group (the stabilizer of a suitable nondegenerate quadratic form). So in almost all cases we may identify Ω with a set of subspaces (or pairs of subspaces) of the natural module, which makes subspace actions more amenable to direct computation since we have a concrete description of the action. 5.1. Main result and notation Our main theorem is the following. Theorem 5.1. Let G (cid:2) Sym(Ω) be an almost simple ﬁnite primitive permutation group with socle G0 and point stabilizer H. Let x ∈ G be an element of prime order r and assume G0 is a classical group over Fq and H is a subspace subgroup. Then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) G is permutation isomorphic to a group recorded in part (a) or (b) in Theorem 1(i); (ii) (G, H, x, fpr(x)) is one of the cases listed in Table 6. Remark 5.2. Write q = pf where p is a prime. In the proof of Theorem 5.1, we will often establish stronger upper bounds. For example, if x is unipotent, then r = p and we usually aim to show that fpr(x) (cid:2) (q + 1)−1. Similarly, if x is semisimple and r is odd then we typically establish a bound on fpr(x) in terms of q and the order i of q mod r (that is, in terms of the smallest positive integer i such that r divides qi − 1). More precisely, if i is even then we will often show that fpr(x) (cid:2) (qi/2 + 2)−1, which establishes T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 37 Table 6 The exceptional subspace actions of classical groups. G0 Ln(q) n (cid:2) 2 Un(q) n (cid:2) 3 PSpn(q) n (cid:2) 4 Ωn(q) n (cid:2) 7 n(q) PΩ(cid:2) n (cid:2) 8 (cid:2) = ± Type of H x P1 P1 P2 N1 P1 O(cid:2) O− n(q) n (q) P1 − N 1 P1 P2 N1 ) 1 (J2, J n−2 (ω, In−1) ϕ (ω, In−1) τ (ωI2, I2) (ω, In−1) 1 (J2, J n−2 (J2, J n−2 (Λ, In−2) 1 ) ) (−In−1, I1)+ (−In−1, I1)− (J2, J n−2 ) 1 (−In−1, I1) (Λ, In−2) (Λ, I6) (−In−1, I1)(cid:2) (−In−1, I1)(cid:3) (J2, J n−2 (Λ, In−2) ) 1 r q q − 1 3 3 2 3 3 q 2 3 2 2 2 2 3 5 2 2 2 3 1 fpr(x) q+1 + q(qn−2−1) q + (q−1)2 q(qn−1) (q+1)(qn−1) 1 1 3 1 3 1 1 3 4 + 4(2n+1) 9 (q = 2); 5 5 14 (q = 3) 1 1 2n−1(2n−1) (q+1)(qn−1) 2 3(3(n−1)/2+1) 3(2n/2+(cid:2)) 3 4(2n/2−1) 4 + 3(2n−3+1) q+1 + q(qn−2−1) 3 + 2n/2−1−(cid:2) 1 4 + 1 3 + 2(3(n−3)/2+1) 1 3 + 3(n−1)/2(3(n−1)/2−1) 3 + 2n−2−(cid:2)2n/2−1−2 1 3 + 1 4 + 1 5 1 3 + 3 + 2(3n/2−2+1) 1 3 + 2n/2−1+(cid:2) 1 4 + 2 3(3n/2+1) 3 4(2n/2+1) 3(2n/2−(cid:2)) 3 4(2n/2−1) 4 3(3n/2−1) 3n/2−1(3n/2+1) 1 1 3(2n/2−1+(cid:2))(2n/2−(cid:2)) Conditions n (cid:2) 3 (n, q) = (2, 8) n (cid:2) 5 odd, q = 2 n = 4, q ∈ {2, 3} (n, q) = (4, 2) n even, q = 2 n (cid:2) 6, q = 2 n (cid:2) 6, q = 2 q = 3 q = 3 q = 2 (q, (cid:2)) = (3, −) (q, (cid:2)) = (2, −) (n, q, (cid:2)) = (8, 4, +) (q, (cid:2)) = (3, +) (q, (cid:2)) = (3, −) q = 2 (q, (cid:2)) = (2, +) the desired bound since r divides qi/2 + 1. Similarly, if i is odd then we typically aim to show that fpr(x) (cid:2) q−i. Remark 5.3. Explicit bounds on ﬁxed point ratios for subspace actions of classical groups are presented in [25,27]. In particular, we will repeatedly apply the results of Guralnick and Kantor given in [27, Section 3]. Before we embark on the proof of Theorem 5.1, we ﬁrst need to deﬁne our notation for subspace actions, which we will use throughout Section 5. The additional notation appearing in Table 6 is discussed in Remark 5.4 below. The various possibilities for G and H that we need to consider are recorded in Table 7, which provides a framework for the proof of Theorem 5.1. In the second column, we present the type of H, which is designed to describe the type of subspace (or pair of subspaces) stabilized by H. For parabolic subgroups, our notation is consistent with [31]. In particular, we write Pm to denote the stabilizer of a totally singular m-space (for G0 = Ln(q), we adopt the standard convention that every subspace of V is totally singular). Similarly, if G0 = Ln(q) then Pm,n−m and GLm(q) ⊕ GLn−m(q) denote the stabilizers of a pair of subspaces (U, W ) with dim U = m and dim W = n − m, where we have U ⊂ W in the ﬁrst case and V = U ⊕ W in the second. As indicated in the 38 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Table 7 The subspace actions. G0 Ln(q) Un(q) PSpn(q) PΩ(cid:2) n(q) Type of H Pm Pm,n−m GLm(q) ⊕ GLn−m(q) Pm Nm Pm P1,2 Nm O(cid:2) Pm N η m P1,3,4 n(q) Conditions 1 (cid:3) m (cid:3) n/2 1 (cid:3) m < n/2, G (cid:8)(cid:3) PΓLn(q) 1 (cid:3) m < n/2, G (cid:8)(cid:3) PΓLn(q) 1 (cid:3) m (cid:3) n/2 1 (cid:3) m < n/2 1 (cid:3) m (cid:3) n/2 n = 4, p = 2, G (cid:8)(cid:3) PΓSp4(q) 2 (cid:3) m < n/2, m even p = 2 1 (cid:3) m (cid:3) n/2 1 (cid:3) m (cid:3) n/2, ((cid:2), η) = (−, +) if m = n/2 (n, (cid:2)) = (8, +), G (cid:8)(cid:3) PΓO+ 8 (q) ﬁnal column, these two subgroups are maximal only if G contains graph or graph-ﬁeld automorphisms. Similarly, if G0 = PSp4(q) and q is even then we write P1,2 to denote a Borel subgroup of G. There is also a parabolic subgroup labeled P1,3,4, which arises when G0 = PΩ+ 8 (q) and G contains triality automorphisms. For G0 = Un(q) or PSpn(q), we write Nm for the stabilizer of a nondegenerate m-space and this notation extends (with suitable modiﬁcations) to orthogonal groups. Suppose G0 is an orthogonal group and let Q be the deﬁning quadratic form on V . First recall that if U is a nondegenerate m-dimensional subspace of V with m even, then U is a plus-type space if it contains a totally singular subspace of dimension m/2, otherwise it is a minus-type space (and every maximal totally singular subspace has dimension m/2 − 1). Now, if G0 = Ωn(q) with nq odd then we write N η m with η = ± to denote the stabilizer of a nondegenerate m-space U with the property that either m is even and U has type η, or m is odd and the orthogonal complement U ⊥ is a nondegenerate (n − m)-space of type η. Similarly, if G0 = PΩ(cid:2) m is the stabilizer of a nondegenerate m-space of type η when m is even. On the other hand, if m is odd then q is odd and we adopt the convention that Nm is the stabilizer of a nondegenerate m-space U with square discriminant (that is, the discriminant of the restriction of Q to U is a square in F × q ). We will use the term square m-space to describe such a subspace, noting that the actions on the set of square and nonsquare m-spaces are permutation isomorphic (so there is no need to consider nonsquare spaces). Finally, if n and q are even, then we also write N1 to denote the stabilizer of a nonsingular 1-space. n(q) with n even, then N η Various conditions are recorded in the ﬁnal column of Table 7, which are designed to eliminate an unnecessary repetition of cases. For example, in the ﬁrst row we may assume m (cid:2) n/2 because the action of G on the set of m-spaces is permutation isomorphic to the action on (n − m)-spaces. We refer the reader to [7,31] for the precise conditions needed to ensure that the given subgroup H is maximal in G. Remark 5.4. Let us comment on the conditions and notation in Table 6. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 39 (a) In the ﬁrst column, we record the socle G0 of G, noting that the given conditions on n are justiﬁed in view of the well known isomorphisms among the low-dimensional classical groups (see [31, Proposition 2.9.1]). (b) In the second column, we describe the type of H using the same notation as in Table 7. The relevant elements x are described in the third column. If x is unipotent, then the Jordan form of x on the natural module V is presented, where Ji denotes a standard unipotent Jordan block of size i. Similarly, if x is semisimple, then we describe the eigenvalues of x on V , up to scalars. Here ω ∈ F × qu is a primitive r-th root of unity and we write Λ to denote an arbitrary irreducible element in GL2(q) of order r (in this case, r divides q + 1). (c) In the two rows with G0 = Ωn(q), we write (−In−1, I1)(cid:2) to denote a semisimple in- volution whose (−1)-eigenspace on V is nondegenerate of type (cid:3) ∈ {+, −}. Similarly, n(q) with n even, then (−In−1, I1)δ is an involution whose 1-eigenspace if G0 = PΩ(cid:2) (cid:13)v(cid:14) has discriminant δ ∈ {(cid:4), (cid:5)}. That is, if Q is the deﬁning quadratic form, then Q(v) ∈ F × q is a square or nonsquare according to δ. (d) Finally, the element ϕ in the third row is a ﬁeld automorphism of order 3, while τ in row 5 is an involutory graph automorphism with CG0(τ ) = PSp4(q). We continue to adopt the notation in [16, Chapter 3] for unipotent and semisimple elements. In particular, we use the notation of Aschbacher and Seitz [2] for unipotent involutions in symplectic and orthogonal groups when p = 2. 5.2. Linear groups In this section, we begin the proof of Theorem 5.1 by handling the linear groups with socle G0 = Ln(q). First we consider the groups with n = 2, in which case H = P1 is a Borel subgroup. As before, we may assume q (cid:3) 7 and q (cid:8)= 9. Note that in part (i) of the following result, q is even and r is a Mersenne prime. Proposition 5.5. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H = P1 and socle G0 = L2(q) with q (cid:3) 7 and q (cid:8)= 9. Let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) x ∈ G0 has order r = q − 1 and fpr(x) = 1 q + q − 1 q(q + 1) . (ii) G = L2(8):3, x is a ﬁeld automorphism of order 3 and fpr(x) = 1/3. Proof. Here \|Ω\| = q + 1 and we may identify Ω with the set of 1-dimensional subspaces of V . First assume x ∈ H ∩ PGL2(q). Then either r = p and \|CΩ(x)\| = 1, in which 40 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 case the bound fpr(x) (cid:2) (r + 1)−1 clearly holds, or r divides q − 1 and \|CΩ(x)\| = 2. In the latter case, we have fpr(x) = 2(q + 1)−1, which is at most (r + 1)−1 if and only if r (cid:2) (q − 1)/2. Notice that if r > (q − 1)/2 then r = q − 1, so q is even, r is a Mersenne prime and fpr(x) = 2(r + 2)−1, which is the special case appearing in part (i). Finally, if 0 and x is a ﬁeld automorphism, then \|CΩ(x)\| = q0 + 1 and we deduce that either q = qr fpr(x) (cid:2) (r + 1)−1, or q0 = 2, r = 3 and fpr(x) = 1/3 as in part (ii). (cid:2) For the remainder of Section 5.2, we will assume G0 = Ln(q) with n (cid:3) 3. Our main = L2(7) result is the following (in the statement, we assume G0 (cid:8)= L3(2), L4(2) since L3(2) ∼ and L4(2) ∼ = A8). Proposition 5.6. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0 = Ln(q) with n (cid:3) 3 and (n, q) (cid:8)= (3, 2), (4, 2). Assume H is a subspace subgroup and x ∈ G has prime order r. Then either fpr(x) (cid:2) (r + 1)−1, or H = P1 and one of the following holds: (i) r = p = q, x = (J2, J n−2 1 ) and fpr(x) = 1 q + 1 + q(qn−2 − 1) (q + 1)(qn − 1) . (ii) r = q − 1, x = (ω, In−1) and fpr(x) = 1 q + (q − 1)2 q(qn − 1) . We will prove Proposition 5.6 in a sequence of lemmas, where we consider each possi- bility for H arising in Table 7. Before launching into the details, we present the following elementary lemma on the ﬁxed points of semisimple elements in a subspace action. In (cid:18) part (ii), the Gaussian binomial coeﬃcient q denotes the number of m-dimensional subspaces in an n-dimensional space over Fq. That is, n m (cid:19) (cid:20) n m (cid:21) m−1(cid:16) = q (cid:3)=0 qn−(cid:3) − 1 qm−(cid:3) − 1 . Lemma 5.7. Consider the natural action of G = GLn(q) on the set Ω of m-dimensional subspaces of the natural module V . Let x ∈ G be a semisimple element of odd prime order r and let i (cid:3) 1 be minimal such that r divides qi − 1, so x is conjugate to (Λa1 1 , . . . , Λat t , Ie), where Λ1, . . . , Λt represent the distinct conjugacy classes in GLi(q) of elements of order r and the aj are non-negative integers such that i (cid:22) j aj (cid:2) n. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 41 (i) We have \|CΩ(x)\| (cid:2) \|CΩ(y)\|, where y = (Λa (ii) If m = i + k with 0 (cid:2) k < i, then 1, Ie) and a = (cid:22) j aj. \|CΩ(y)\| = (cid:20) e m (cid:21) (cid:20) + q e k (cid:21) (cid:5) q (cid:6) . qia − 1 qi − 1 Proof. Part (i) is clear because we can choose y so that there is a natural containment CΩ(x) ⊆ CΩ(y) of the ﬁxed point sets. Now consider (ii) and observe that y preserves a decomposition V = W1 ⊕ · · · ⊕ Wa ⊕ CV (y) where the Wj are isomorphic i-dimensional irreducible Fq(cid:13)y(cid:14)-modules. Write \|CΩ(y)\| = α + β, where α is the number of m-spaces on which y acts trivially (this is simply the number of m-spaces in CV (y)). Next let U be an m-space in CΩ(y) on which y acts nontrivially. Since m = i + k with 0 (cid:2) k < i, it follows that y preserves a decomposition U = U1 ⊕ U2, acting irreducibly on U1 and trivially on the k-space U2. Therefore β = β1β2, where β1 is the number of k-spaces in CV (y) and β2 is the number of m-spaces in W = W1 ⊕ · · · ⊕ Wa ﬁxed by z = (Λa) ∈ GLai(q) = GL(W ). To compute β2 we may view z as a scalar matrix in an extension ﬁeld subgroup GLa(qi) < GLai(q). Then each m-space in W ﬁxed by z corresponds to a 1-dimensional subspace of the natural module for GLa(qi). Since z is a scalar in GLa(qi), it ﬁxes every 1-space and thus β2 is the total number of 1-spaces in (Fqi)a. Therefore (cid:20) e m (cid:21) q α = , β1 = (cid:20) e k (cid:21) q , β2 = (qi)a − 1 qi − 1 and this completes the proof of part (ii). (cid:2) Lemma 5.8. The conclusion to Proposition 5.6 holds if H = P1. Proof. First observe that \|Ω\| = (qn − 1)/(q − 1) and the maximality of H implies that G (cid:2) PΓLn(q). Let x ∈ G be an element of prime order r. In view of (4), we may assume x ∈ PGLn(q) is semisimple or unipotent with ν(x) = s (see Deﬁnition 4.6). If r = p then CΩ(x) is the set of 1-dimensional subspaces in the 1-eigenspace CV (x), so \|CΩ(x)\| = qn−s − 1 q − 1 , fpr(x) = qn−s − 1 qn − 1 = 1 qs − qs − 1 qs(qn − 1) . 42 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 From here, it is straightforward to check that fpr(x) > (r + 1)−1 if and only if s = 1 and q = p, which is the case recorded in part (i) of Proposition 5.6. Now suppose r (cid:8)= p. Let i (cid:3) 1 be minimal such that r divides qi − 1 and note that r + 1 (cid:2) qi. First assume i = 1. If s = 1 then x is of the form (ω, In−1) (up to scalars) and \|CΩ(x)\| = 1 + qn−1 − 1 q − 1 , fpr(x) = 1 q + (q − 1)2 q(qn − 1) . This is greater than (r + 1)−1 if and only if r = q − 1 (so either r = 2, or r is a Mersenne prime) and this special case appears in part (ii) of Proposition 5.6. Now assume i = 1 and s (cid:3) 2. If n = 3 then x is regular, \|CΩ(x)\| = 3 and the result follows. For n (cid:3) 4 we observe that \|CΩ(x)\| is maximal when x is of the form (ωI2, In−2), whence \|CΩ(x)\| (cid:2) q2 − 1 q − 1 + qn−2 − 1 q − 1 and we deduce that fpr(x) (cid:2) (r + 1)−1. Finally, suppose r (cid:8)= p and i (cid:3) 2. Here \|CΩ(x)\| is equal to the number of 1-dimensional subspaces in CV (x), whence \|CΩ(x)\| is maximal when x = (Λ, In−i) in the notation of [16, Proposition 3.2.1]. Here the notation indicates that x preserves a decomposition V = U ⊕ W , acting irreducibly on the i-space U and trivially on W . This implies that \|CΩ(x)\| is at most (qn−i − 1)/(q − 1) and the result follows since fpr(x) (cid:2) q−i. (cid:2) Lemma 5.9. The conclusion to Proposition 5.6 holds if H = Pm with 2 (cid:2) m (cid:2) n/2. Proof. Here we identify Ω with the set of m-dimensional subspaces of V and we note that \|Ω\| = \|GLn(q)\| qm(n−m)\|GLm(q)\|\|GLn−m(q)\| = (cid:20) n m (cid:21) m−1(cid:16) = q (cid:3)=0 qn−(cid:3) − 1 qm−(cid:3) − 1 . By applying [25, Lemma 2.1] we obtain qm(n−m) < \|Ω\| < 2 (cid:6) (cid:5) q q − 1 qm(n−m). (14) For m (cid:8)= n/2, notice that the maximality of H implies that G (cid:2) PΓLn(q). Let x ∈ G be an element of prime order r. By arguing as in the proof of the previous lemma, we may assume that either (a) x ∈ PGLn(q) is semisimple or unipotent; or (b) n is even, m = n/2 and x is an involutory graph automorphism. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 43 By combining [27, Proposition 3.1] with [27, Lemma 3.11(d)] we observe that fpr(x) < 2q−m. (15) In particular, if (b) holds then fpr(x) (cid:2) 1/3 unless m = q = 2. But in the latter case we have G0 = L4(2), which is excluded in Proposition 5.6 (for the record, fpr(x) = 3/7 > 1/3 if x is a symplectic-type graph automorphism, otherwise fpr(x) = 3/35). For the remainder, we may assume x ∈ PGLn(q) is semisimple or unipotent. If r = p then (15) is suﬃcient unless m = q = 2. Here n (cid:3) 5 and we claim that \|CΩ(x)\| is maximal when x = (J2, J n−2 ) with 1 (cid:2) (cid:4) (cid:2) n/2. ). To see this, suppose x = (J (cid:3) If U is a 2-space ﬁxed by x, then either U ⊆ CV (x) or U = (cid:13)u, xu(cid:14) with u ∈ V \ CV (x). Therefore, 2, J n−2(cid:3) 1 1 \|CΩ(x)\| = 1 3 (2n−(cid:3) − 1)(2n−(cid:3)−1 − 1) + (cid:4) (cid:3) (2n − 1) − (2n−(cid:3) − 1) 1 2 and it is easy to check that this is maximal when (cid:4) = 1. This justiﬁes the claim and we quickly deduce that fpr(x) (cid:2) 1/3 for all n (cid:3) 5. So to complete the proof, we may assume r (cid:8)= p. As usual, let i (cid:3) 1 be minimal such that r divides qi − 1. The bound in (15) implies that fpr(x) < q−1, so we may assume i (cid:3) 2. Note that r divides t = (qi − 1)/(q − 1). There are two cases to consider. First assume i > m. Here \|CΩ(x)\| is the number of m-spaces in CV (x), whence x = (Λ, In−i) has the most ﬁxed points and thus \|CΩ(x)\| (cid:2) (cid:21) (cid:20) n − i m q (cid:5) < 2 q q − 1 (cid:6) qm(n−i−m). By applying the lower bound on \|Ω\| in (14) we get fpr(x) < 2 (cid:6) (cid:5) q q − 1 q−mi (cid:2) 2 (cid:6) (cid:5) q q − 1 q−2i (cid:2) (t + 1)−1. Now suppose 2 (cid:2) i (cid:2) m. Here the bound in (15) is suﬃcient unless m = i and q ∈ {2, 3}, or m = i + 1 and q = 2. The groups with n (cid:2) 6 can be checked directly with the aid of Magma, so we may assume n (cid:3) 7. First assume m = i and q ∈ {2, 3}. Now dim CV (x) = n − (cid:4)i for some 1 (cid:2) (cid:4) (cid:2) (cid:11)n/i(cid:12) and by applying Lemma 5.7(i) we see that \|CΩ(x)\| is maximal when x = (Λ(cid:3), In−(cid:3)i). Then by part (ii) of Lemma 5.7 we get \|CΩ(x)\| = (cid:20) n − (cid:4)i i (cid:21) + q qi(cid:3) − 1 qi − 1 . For (cid:4) = 1, this implies that 44 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 fpr(x) < 2 (cid:6) (cid:5) q q − 1 q−i2 (cid:2) (t + 1)−1 and the result follows. Now assume (cid:4) (cid:3) 2, in which case we compute fpr(x) < 4q−i2(cid:3) + 2q−i(n+1−(cid:3)−i) = q−i (cid:23) 4q−i(i(cid:3)−1) + 2q−i(n−(cid:3)−i) (cid:24) . (16) Since n (cid:3) 7, i (cid:3) 2 and 2 (cid:2) (cid:4) (cid:2) (cid:11)n/i(cid:12) we deduce that 4q−i(i(cid:3)−1) + 2q−i(n−(cid:3)−i) (cid:2) 4q−6 + 2q4−n (cid:2) 1 and thus (16) implies that fpr(x) < q−i. A very similar argument applies when m = i + 1 and q = 2. Once again we can reduce to the case where x = (Λ(cid:3), In−(cid:3)i) and Lemma 5.7 gives Therefore \|CΩ(x)\| = (cid:20) n − (cid:4)i i + 1 (cid:21) + 2 (2n−(cid:3)i − 1)(2i(cid:3) − 1) 2i − 1 . fpr(x) < 2−i(i+1)(cid:3)+2 + 2−i(n−i−1)+2 and by setting (cid:4) = 1 we deduce that fpr(x) < 2−i (cid:2) (r + 1)−1 as required. (cid:2) In order to complete the proof of Proposition 5.6, we may assume G (cid:8)(cid:2) PΓLn(q) and H is of type Pm,n−m or GLm(q) ⊕ GLn−m(q) (in both cases we have m < n/2). Lemma 5.10. The conclusion to Proposition 5.6 holds if G (cid:8)(cid:2) PΓLn(q) and H is of type Pm,n−m or GLm(q) ⊕ GLn−m(q). Proof. In both cases, we may identify Ω with a set of pairs (U, W ) of subspaces of V , where dim U = m and dim W = n − m. For H = Pm,n−m, each pair (U, W ) in Ω satisﬁes the condition U ⊂ W , whereas V = U ⊕ W when H is of type GLm(q) ⊕ GLn−m(q). Let x ∈ G be an element of prime order r and note that (15) holds (see [27, Lemma 3.12(a)]). By the usual argument, we may assume x is not a ﬁeld or graph-ﬁeld automorphism. Next assume x is an involutory graph automorphism. Here (15) is suﬃcient unless m = 1, or m = q = 2. In both cases, it is clear that \|CΩ(x)\| is at most the total number of m- spaces in V . Therefore, if m = q = 2 then \|CΩ(x)\| (cid:2) (cid:20) n 2 (cid:21) = 2 1 3 (2n − 1)(2n−1 − 1) and we quickly deduce that fpr(x) (cid:2) 1/3. For m = 1 we have \|CΩ(x)\| (cid:2) (qn − 1)/(q − 1) and the same conclusion holds. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 45 To complete the proof, we may assume x ∈ PGLn(q) is semisimple or unipotent. Since x ﬁxes a pair (U, W ) ∈ Ω only if it ﬁxes the m-space U , by applying Lemmas 5.8 and 5.9 we can immediately reduce to the case where m = 1 and x is one of the elements arising in parts (i) and (ii) in the statement of Proposition 5.6. Since \|CΩ(x)\| is at most the number of 1-spaces ﬁxed by x, we deduce that \|CΩ(x)\| (cid:2) 1 + (qn−1 − 1)/(q − 1) and one can check that this bound is suﬃcient since r ∈ {q − 1, q}. (cid:2) This completes the proof of Proposition 5.6. 5.3. Unitary groups Here is our main result for subspace actions of unitary groups. Note that in part (ii), an involutory graph automorphism x of G0 = U4(q) is said to be of symplectic-type if CG0(x) has socle PSp4(q). Proposition 5.11. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0 = Un(q) with n (cid:3) 3. Assume H is a subspace subgroup of G. If x ∈ G has prime order r, then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) H = P1, r = 3, n is odd, q = 2, x = (ω, In−1) and fpr(x) = 1 4 + 3 4(2n + 1) . (ii) H = P2, r = 2, n = 4, q ∈ {2, 3}, x is a symplectic-type graph automorphism and fpr(x) = 5/9 or 5/14 for q = 2 or 3, respectively. (iii) H = P2, r = 3, n = 4, q = 2, x = (ωI2, I2) and fpr(x) = 1/3. (iv) H = N1, r = 3, n is even, q = 2, x = (ω, In−1) and fpr(x) = 1 4 + 3(2n−3 + 1) 2n−1(2n − 1) . Lemma 5.12. The conclusion to Proposition 5.11 holds if H = P1. Proof. First observe that we may identify Ω with the set of 1-dimensional totally singular subspaces of V and note that \|Ω\| = \|GUn(q)\| q2n−3\|GUn−2(q)\|\|GL1(q2)\| = (qn − (−1)n)(qn−1 − (−1)n−1) q2 − 1 . Let x ∈ G be an element of prime order r. If x is a ﬁeld automorphism, then r is odd, q = qr 0 and the usual argument via (4) applies. Next suppose x is an involutory graph automorphism. As explained in the 46 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 proof of [27, Lemma 3.14] (see Case D), \|CΩ(x)\| is at most the number of 1-dimensional subspaces of (Fq)n, so \|CΩ(x)\| (cid:2) (qn − 1)/(q − 1) and we deduce that fpr(x) (cid:2) (qn − 1)(q + 1) (qn − (−1)n)(qn−1 − (−1)n−1) . One can check that this expression is at most 1/3 unless (n, q) = (3, 3), in which case a straightforward Magma calculation gives fpr(x) = 1/7. For the remainder of the proof, we may assume x ∈ PGUn(q) is semisimple or unipotent. First assume r = p and note that \|CΩ(x)\| coincides with the number of totally singular ) then dim CV (x) (cid:2) 1-spaces in CV (x). We claim that fpr(x) (cid:2) (r +1)−1. If x (cid:8)= (J2, J n−2 n − 2 and one checks that the bound 1 \|CΩ(x)\| (cid:2) (cid:21) (cid:20) n − 2 1 q2 = q2n−4 − 1 q2 − 1 yields fpr(x) (cid:2) (q + 1)−1. Now assume x = (J2, J n−2 ). Here x preserves an orthogonal decomposition V = U ⊥ W , where x has Jordan form (J2) on the nondegenerate 2-space U . Setting CU (x) = (cid:13)u(cid:14), which is totally singular, we have 1 CΩ(x) = {(cid:13)u(cid:14), (cid:13)λu + w(cid:14) : λ ∈ Fq2 and (cid:13)w(cid:14) ⊆ W is totally singular}. Therefore, \|CΩ(x)\| = αq2 + 1, where α = (qn−2 − (−1)n−2)(qn−3 − (−1)n−3) q2 − 1 is the number of totally singular 1-spaces in W (this can also be computed via [25, Lemma 2.13(2)], noting that (cid:13)u(cid:14) is the radical of CV (x)). It is now a routine exercise to check that fpr(x) (cid:2) (q + 1)−1 for all n and q. Finally, suppose r (cid:8)= p and let i (cid:3) 1 be minimal such that r divides qi − 1. We will adopt the notation for semisimple elements given in [16, Proposition 3.3.2]. First assume i = 2, so r is odd, r divides q + 1 and \|CΩ(x)\| is equal to the total number of totally singular 1-spaces in each eigenspace of ˆx on V (where x is the image of ˆx ∈ GUn(q) modulo scalars). Since every eigenspace of ˆx is nondegenerate, it follows that \|CΩ(x)\| is maximal when x is of the form (ω, In−1), in which case fpr(x) = qn−2 − (−1)n qn − (−1)n . (17) If n is even, then it is straightforward to check that fpr(x) (cid:2) (q + 2)−1 and the result follows. On the other hand, if n is odd then the same conclusion holds if and only if q (cid:3) 3. Indeed, if q = 2 then r = 3 and T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 47 fpr(x) = 1 4 + 3 4(2n + 1) , which corresponds to the special case recorded in part (i) of Proposition 5.11. If n is odd, q = 2 and x is any other element of order 3, then \|CΩ(x)\| is maximal when x = (ωI2, In−2) and thus \|CΩ(x)\| (cid:2) (22 − 1)(2 + 1) 22 − 1 + (2n−2 + 1)(2n−3 − 1) 22 − 1 . It is easy to check that fpr(x) (cid:2) 1/4. Next assume i ≡ 2 (mod 4) and i (cid:3) 6. Note that r divides t = (qi/2 + 1)/(q + 1) and \|CΩ(x)\| coincides with the number of totally singular 1-spaces in CV (x). It follows that \|CΩ(x)\| is maximal when x = (Λ, In−i/2) and it is easy to check that fpr(x) (cid:2) (t + 1)−1 as required. Now suppose i ≡ 0 (mod 4), so r divides qi/2 + 1. Once again, \|CΩ(x)\| is the number of totally singular 1-spaces in CV (x) and thus elements of the form (Λ, Λ−q, In−i) ﬁx the most points. Given this observation, it is straightforward to verify the bound fpr(x) (cid:2) (qi/2 + 2)−1. Finally, suppose i is odd. First assume r = 2, so q is odd and i = 1. Here \|CΩ(x)\| is maximal when x is of the form (−I1, In−1), in which case (17) holds and it is easy to check that fpr(x) (cid:2) 1/3. Next assume r is odd and i = 1, so \|CΩ(x)\| is maximal when x is of the form ((Λ, Λ−q)(cid:3), In−2(cid:3)) for some (cid:4) (cid:3) 1. Here the 1-eigenspace is nondegenerate, while the other two eigenspaces are totally singular, whence \|CΩ(x)\| = 2 (cid:6) (cid:5) q2(cid:3) − 1 q2 − 1 + (qn−2(cid:3) − (−1)n)(qn−2(cid:3)−1 − (−1)n−1) q2 − 1 . If n = 4 then \|CΩ(x)\| is maximal when (cid:4) = 2, in which case \|CΩ(x)\| = 2(q2 + 1) and fpr(x) = 2(q3 + 1)−1. For n (cid:3) 5 one can check that \|CΩ(x)\| is maximal when (cid:4) = 1 and it is plain to see that the same conclusion holds when i (cid:3) 3 (since in this case, \|CΩ(x)\| is just the number of totally singular 1-spaces in CV (x)). Here x = (Λ, Λ−q, In−2i) preserves a decomposition V = (U1 ⊕ U2) ⊥ W , where U1 and U2 are totally singular i-spaces and W = CV (x) is nondegenerate (or trivial). Moreover, \|CΩ(x)\| = α + β, where α is the number of totally singular 1-spaces in W and we set β = 2 if i = 1 (since x also ﬁxes the totally singular 1-spaces U1 and U2), otherwise β = 0. It is now straightforward to verify the bound fpr(x) (cid:2) q−i and the result follows. (cid:2) Lemma 5.13. The conclusion to Proposition 5.11 holds if H = Pm with 2 (cid:2) m (cid:2) n/2. Proof. Identify Ω with the set of totally singular m-dimensional subspaces of V and note that \|Ω\| = \|GUn(q)\| qm(2n−3m)\|GUn−2m(q)\|\|GLm(q2)\| . 48 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 By applying [13, Proposition 3.9], we get qm(2n−3m) < \|Ω\| < 2 1 2 (cid:6) (cid:5) q + 1 q qm(2n−3m). Suppose x ∈ G has prime order r. If n (cid:3) 6 then [27, Proposition 3.15] gives fpr(x) < q−(n−1)/2 + 2q−(n−2) + q−2m. (18) As usual, the desired bound holds if x is a ﬁeld automorphism. Next suppose x is an involutory graph automorphism, so r = 2 and we may assume q (cid:2) 3 in view of Theorem 4.5. The groups with n (cid:2) 5 can be checked using Magma, noting the two special cases that arise when n = 4, H = P2 and x is a symplectic-type graph automorphism (see part (ii) in the statement of Proposition 5.11). On the other hand, if n (cid:3) 6 then the bound in (18) is suﬃcient unless (n, q) = (6, 2), which we can handle using Magma (we get fpr(x) (cid:2) 5/33, with equality if m = 3 and CG0(x) = Sp6(2)). For the remainder, we may assume x ∈ PGUn(q) is semisimple or unipotent. First assume r = p. If n (cid:3) 6 then the bound in (18) gives fpr(x) (cid:2) (q + 1)−1 unless (n, q) = (6, 2), which we can check using Magma. Now assume n ∈ {4, 5} and m = 2. If p = 2 and q (cid:3) 4 then Theorem 4.5 yields fpr(x) (cid:2) 1/3 as required, while the case q = 2 can be handled using Magma. Therefore, we may assume q is odd. For n = 4, we claim that \|CΩ(x)\| (cid:2) q2 + q + 1, which immediately implies that − fpr(x) (cid:2) (q + 1)−1. To see this, ﬁrst observe that G0 6 (q) and the action of G0 − 6 (q) on the set Γ of totally singular on Ω is permutation isomorphic to the action of PΩ 1-dimensional subspaces of the 6-dimensional orthogonal module W . The eﬀect of this isomorphism on unipotent elements is as follows: ∼ = PΩ (J2, J 2 1 ) (cid:19)→ (J 2 2 , J 2 1 ), (J 2 2 ) (cid:19)→ (J3, J 3 1 ), (J3, J1) (cid:19)→ (J 2 3 ), (J4) (cid:19)→ (J5, J1). − In particular, if x ∈ G0 is sent to y ∈ PΩ 6 (q), then \|CΩ(x)\| = \|CΓ(y)\| is equal to the number of totally singular 1-spaces in CW (y). If dim CW (y) (cid:2) 2, then we deduce that \|CΩ(x)\| (cid:2) q + 1. In the remaining two cases, we can appeal to the analysis of unipotent elements in the proof of Lemma 5.27, which allows us to conclude that \|CΩ(x)\| = q + 1 if x = (J2, J 2 1 ) and \|CΩ(x)\| = q2 + q + 1 when x = (J 2 2 ). This justiﬁes the claim. Next assume n = 5, so \|Ω\| = (q5 + 1)(q3 + 1). First observe that \|CΩ(x)\| = 1 when x = (J3, J2), (J4, J1) or (J5). Next let x = (J2, J 3 1 ). Here x preserves an orthogonal decomposition V = V1 ⊥ V2, where dim V1 = 2 and CV1(x) = (cid:13)u(cid:14) is totally singular. Then every space in CΩ(x) is of the form (cid:13)u, w(cid:14), where (cid:13)w(cid:14) is a totally singular 1-space in the nondegenerate 3-space V2, and thus \|CΩ(x)\| = q3 + 1. Suppose x = (J3, J 2 1 ). Here we claim that \|CΩ(x)\| (cid:2) (q2 − 1)2 + q + 1 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 49 which implies that fpr(x) (cid:2) (q + 1)−1. To see this, write V = V1 ⊥ V2 as an orthogonal decomposition into nondegenerate spaces, where x has Jordan form (J3, J1) on V1 and V2 = (cid:13)v(cid:14) is centralized by x. If ( , ) is the deﬁning unitary form on V , then we may assume (v, v) = 1. Let U be a totally singular 2-space ﬁxed by x. By our above analysis of the case n = 4, we see that there are at most q + 1 such spaces contained in V1. Now assume U ∩ V1 = (cid:13)u(cid:14) is 1-dimensional, so U = (cid:13)u, w + v(cid:14) for some w ∈ (cid:13)u(cid:14)⊥ ∩ V1 with (w, w) = −1. Here (cid:13)u(cid:14) has to be the radical of the 2-space CV1(x) and we calculate that there are (q2 − 1)2 nondegenerate 1-spaces in the 3-space (cid:13)u(cid:14)⊥ ∩ V1. This justiﬁes the claim. Finally, suppose x = (J 2 2 , J1). Here the claim is \|CΩ(x)\| (cid:2) (q4 − 1)(q2 − 1) + q2 + q + 1, which once again is suﬃcient. Write V = V1 ⊥ V2, where x has Jordan form (J 2 2 ) on V1 and x centralizes V2 = (cid:13)v(cid:14). From our earlier work in the case n = 4, we see that x ﬁxes q2 + q + 1 totally singular 2-spaces in V1. Now suppose U = (cid:13)u, w + v(cid:14) is a totally singular 2-space ﬁxed by x. Here (cid:13)u(cid:14) is contained in the totally singular 2-space CV1(x), so there are q2 + 1 choices for (cid:13)u(cid:14). As before, there are (q2 − 1)2 nondegenerate 1-spaces in (cid:13)u(cid:14)⊥ ∩ V1, whence there are at most (q2 + 1)(q2 − 1)2 totally singular 2-spaces ﬁxed by x that are not contained in V1. The result follows. Now assume r (cid:8)= p and let i (cid:3) 1 be minimal such that r divides qi − 1. First assume r = 2. For q (cid:3) 5, the bound in Theorem 4.5 is clearly suﬃcient and so we may assume q = 3. If n (cid:3) 6 then the bound in (18) is eﬀective, while the remaining cases with n ∈ {4, 5} and m = 2 can be checked using Magma. Now assume r is odd. There are several cases to consider. Case 1. i ≡ 2 (mod 4). First assume i = 2, so r divides q + 1. If n (cid:3) 6 then the bound in (18) is suﬃcient unless (n, q) = (6, 2), which we can handle directly using Magma. Now assume n ∈ {4, 5} and m = 2. For n = 4, it is easy to check that \|CΩ(x)\| is maximal when x = (ωI2, I2), whereas x = (ω, I4) has the most ﬁxed points when n = 5. Therefore, if n = 4 we have \|CΩ(x)\| (cid:2) (q + 1)2 since every nondegenerate 2-space contains q + 1 totally singular 1-spaces. Similarly, \|CΩ(x)\| is at most (q3 + 1)(q + 1) when n = 5, which is the number of totally singular 2-spaces in a nondegenerate 4-space. One can check that these bounds yield fpr(x) (cid:2) (q + 2)−1 unless (n, q) = (4, 2). Here r = 3 and fpr(x) = 1/3; this special case is recorded in part (iii) of Proposition 5.11. Now assume i (cid:3) 6. Note that r divides t = (qi/2 + 1)/(q + 1), so it suﬃces to show that fpr(x) (cid:2) (t + 1)−1. If m (cid:3) i/2 then n (cid:3) i and the bound in (18) implies that fpr(x) < q−(i−1)/2 + 2q−(i−2) + q−i. One can check that this yields fpr(x) (cid:2) (t + 1)−1 unless (i, q) = (6, 2). But 26 − 1 does not have a primitive prime divisor, so the case (i, q) = (6, 2) does not arise. Now assume 50 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 m < i/2. Here \|CΩ(x)\| is the number of totally singular m-spaces in the nondegenerate 1-eigenspace CV (x), which is maximal when x = (Λ, In−i/2). Therefore, \|CΩ(x)\| < 2 (cid:6) (cid:5) q + 1 q qm(2(n−i/2)−3m) = 2 (cid:6) (cid:5) q + 1 q qm(2n−3m) · q−mi and the result follows since fpr(x) < 4 (cid:6) (cid:5) q + 1 q q−mi (cid:2) (t + 1)−1. Case 2. i ≡ 0 (mod 4). Now suppose i ≡ 0 (mod 4), so n (cid:3) i and r divides qi/2 +1. First assume n ∈ {i, i +1}, in which case every element of order r is of the form (Λ, Λ−q, In−i) and we see that \|CΩ(x)\| = 2 if m = i/2, otherwise \|CΩ(x)\| = 0. Therefore, fpr(x) < 4q−n2/4, which in turn implies that fpr(x) < (qi/2 + 2)−1 unless (n, q) = (4, 2). In this special case, we have (m, r) = (2, 5) and we compute fpr(x) = 2/27. For the remainder, we may assume n (cid:3) i + 2. If m (cid:3) i/2 then the bound in (18) is suﬃcient unless q = 2 and (n, i) = (6, 4), (7, 4), (8, 4) or (10, 8). Each of these cases can be handled directly. For example, suppose (n, q, i) = (10, 2, 8), so r = 17, m ∈ {4, 5} and x = (Λ, Λ−2, I2) preserves an orthogonal decomposition V = (U1 ⊕ U2) ⊥ W , where U1 and U2 are totally singular 4-spaces and W = CV (x). If m = 4 then CΩ(x) = {U1, U2} and the result follows. Similarly, if m = 5 then each space in CΩ(x) is of the form Uj ⊕ (cid:13)w(cid:14), where (cid:13)w(cid:14) ⊆ W is totally singular. Since W contains q + 1 = 3 totally singular 1-spaces, it follows that \|CΩ(x)\| = 6 and once again the desired bound holds. The other cases can be handled in a similar fashion (either by hand or via Magma). Now assume m < i/2. Here \|CΩ(x)\| is the number of totally singular m-spaces in CV (x), whence x = (Λ, Λ−q, In−i) has the most ﬁxed points. Working with this element, we compute fpr(x) < 4 (cid:6) (cid:5) q + 1 q q−2mi (cid:2) (qi/2 + 2)−1 and the result follows. Case 3. i odd. Finally, let us assume i is odd. Suppose i = 1, so q (cid:3) 4 (since r is odd) and r + 1 (cid:2) q. If n (cid:3) 6 then the bound in (18) is suﬃcient, so we may assume n ∈ {4, 5} and m = 2. We claim that (cid:12) \|CΩ(x)\| (cid:2) 2(q3 + 1) q2 + 3 if n = 5 if n = 4. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 51 To see this, let us ﬁrst observe that \|CΩ(x)\| = 4 if x is regular, which means that we may assume x = (Λ, Λ−q, In−2) or ((Λ, Λ−q)2, In−4). Now x = (Λ, Λ−q, In−2) preserves a decomposition V = (U1 ⊕ U2) ⊥ W with W = CV (x) and each totally singular 2- space ﬁxed by x is of the form Uj ⊕ (cid:13)w(cid:14), where (cid:13)w(cid:14) is a totally singular 1-space in W = CV (x). Therefore, \|CΩ(x)\| is 2(q3 + 1) if n = 5 and 2(q + 1) if n = 4. Similarly, x = ((Λ, Λ−q)2, In−4) preserves a decomposition V = (U1 ⊕ U2) ⊥ W , where W = CV (x) and x acts as a scalar on the totally singular 2-spaces U1 and U2. Therefore, the totally singular 2-spaces ﬁxed by x are U1, U2 and (cid:13)u(cid:14) ⊕ (cid:13)u(cid:7)(cid:14), where (cid:13)u(cid:14) ⊆ U1 is an arbitrary 1- space and (cid:13)u(cid:7)(cid:14) = U2 ∩(cid:13)u(cid:14)⊥. This implies that \|CΩ(x)\| = q2 +3, which is the total number of subspaces of U1. This justiﬁes the claim and it is easy to check that fpr(x) (cid:2) q−1. Now assume i (cid:3) 3, so n (cid:3) 2i (cid:3) 6 and r divides t = (qi − 1)/(q − 1). If n ∈ {2i, 2i + 1} then x = (Λ, Λ−q, In−2i) is the only possibility and it is easy to check that fpr(x) (cid:2) (t + 1)−1 since \|CΩ(x)\| = 2 if m = i, otherwise \|CΩ(x)\| = 0. Now assume n (cid:3) 2i + 2. If m (cid:3) i then (18) is suﬃcient unless (n, q, i) = (8, 2, 3). Here m ∈ {3, 4}, r = 7 and the result follows since \|CΩ(x)\| (cid:2) 6. Finally, suppose m < i. In this case, \|CΩ(x)\| is the number of totally singular m-spaces in CV (x) and we deduce that fpr(x) < 2 (cid:6) (cid:5) q + 1 q q−4mi (cid:2) (t + 1)−1 as required. (cid:2) Lemma 5.14. The conclusion to Proposition 5.11 holds if H = N1. Proof. We may identify Ω with the set of nondegenerate 1-dimensional subspaces of V and we note that \|Ω\| = \|GUn(q)\| \|GUn−1(q)\|\|GU1(q)\| = qn−1(qn − (−1)n) q + 1 . Let x ∈ G be an element of prime order r. If x is a ﬁeld automorphism, then the usual argument applies. Next suppose x is an involutory graph automorphism. By embedding GUn(q) in GLn(q2), we may view x as an involutory ﬁeld automorphism of GLn(q2) and thus \|CΩ(x)\| is at most the number of 1-dimensional subspaces of V that are deﬁned over Fq. In other words, \|CΩ(x)\| (cid:2) (qn −1)/(q −1) and the result follows unless (n, q) = (4, 2). In the latter case, we have \|Ω\| = 40 and using Magma we calculate that fpr(x) (cid:2) 1/5. To complete the argument, we may assume x ∈ PGUn(q) is semisimple or unipotent. If r = p then \|CΩ(x)\| is the number of nondegenerate 1-spaces in CV (x). In particular, if dim CV (x) (cid:2) n − 2 then \|CΩ(x)\| is at most (q2n−4 − 1)/(q2 − 1), which is the total number of 1-spaces in an (n − 2)-dimensional vector space over Fq2, and we immediately deduce that fpr(x) (cid:2) (q + 1)−1. Now assume dim CV (x) = n − 1, so x = (J2, J n−2 ) preserves a decomposition V = U ⊥ W into nondegenerate spaces, where dim U = 2 and CV (x) = (cid:13)u(cid:14) ⊕ W with (cid:13)u(cid:14) totally singular. Then every nondegenerate 1-space in 1 52 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 CV (x) is of the form (cid:13)λu + w(cid:14), where λ ∈ Fq2 and (cid:13)w(cid:14) is a nondegenerate 1-space in W . Therefore, \|CΩ(x)\| = q2 (cid:5) qn−3(qn−2 − (−1)n−2) q + 1 (cid:6) and once again it is straightforward to check that fpr(x) (cid:2) (q + 1)−1. For the remainder, let us assume r (cid:8)= p and x is semisimple. Let i (cid:3) 1 be minimal such that r divides qi − 1 and set ⎧ ⎪⎨ ⎪⎩ i/2 i 2i j = if i ≡ 2 (mod 4) if i ≡ 0 (mod 4) if i is odd. (19) First assume i = 2, so r divides q+1. Here the eigenspaces of ˆx on V are nondegenerate (where x is the image of ˆx ∈ GUn(q) modulo scalars) and \|CΩ(x)\| is the total number of nondegenerate 1-spaces in each eigenspace. As a consequence, we quickly deduce that \|CΩ(x)\| is maximal when x = (ω, In−1), in which case fpr(x) = qn−2(qn−1 − (−1)n−1) + q + 1 qn−1(qn − (−1)n) (20) and one can check that this is at most (q + 2)−1 unless n is even and q = 2. In this special case we have r = 3 and fpr(x) = 1 4 + 3(2n−3 + 1) 2n−1(2n − 1) , so this is a genuine exception and it is recorded in part (iv) of Proposition 5.11. Let us also observe that if n is even, q = 2, r = 3 and ν(x) (cid:3) 2, then \|CΩ(x)\| (cid:2) q(q2 − 1) + qn−3(qn−2 − 1) q + 1 (maximal if x = (ωI2, In−2)) and it is easy to verify the bound fpr(x) (cid:2) 1/4. Now assume i (cid:8)= 2. If r = 2 then \|CΩ(x)\| is maximal when x = (−I1, In−1), in which case (20) holds and we deduce that fpr(x) (cid:2) 1/3. Now assume r is odd. Here \|CΩ(x)\| is the number of nondegenerate 1-spaces in the 1-eigenspace CV (x) and we deduce that \|CΩ(x)\| (cid:2) qn−j−1(qn−j − (−1)n−j) q + 1 . (21) For example, suppose i ≡ 2 (mod 4). Here j = i/2 (cid:3) 3, r divides t = (qj + 1)/(q + 1) and \|CΩ(x)\| is maximal when x = (Λ, In−j), which gives the upper bound in (21). Moreover, T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 53 it is straightforward to show that fpr(x) (cid:2) (t + 1)−1. A very similar argument applies when i (cid:8)≡ 2 (mod 4) and we omit the details. (cid:2) Finally, we complete the proof of Proposition 5.11 by handling the case where H is the stabilizer of a nondegenerate m-space with m (cid:3) 2. Lemma 5.15. The conclusion to Proposition 5.11 holds if H = Nm with 2 (cid:2) m < n/2. Proof. Let us identify Ω with the set of nondegenerate m-dimensional subspaces of V and observe that n (cid:3) 5 and (cid:5) (cid:6) q − 1 q q2m(n−m) < \|Ω\| = \|GUn(q)\| \|GUm(q)\|\|GUn−m(q)\| < q2m(n−m) (22) (see Section 2 in [25], for example). Let x ∈ G be an element of prime order r. If n (cid:3) 6 then [27, Proposition 3.16] gives fpr(x) < 2q−(n−4) + q−(n−1) + q−2(cid:10)m/2(cid:11) + q−2(n−m). (23) For integers a (cid:3) b, it will be convenient to write f (a, b) for the number of nondegenerate b-spaces in an a-dimensional unitary space over Fq2. In particular, f (a, b) (cid:2) q2b(a−b) and \|Ω\| = f (n, m). (cid:18) (cid:19) q and by combining n m For an involutory graph automorphism x we have \|CΩ(x)\| (cid:2) the relevant bounds in (14) and (22) we deduce that fpr(x) < 2 (cid:6) 2 (cid:5) q q − 1 q−m(n−m) (cid:2) 1 3 . Field automorphisms can be handled in the usual way, so for the remainder we may assume x ∈ PGUn(q) is semisimple or unipotent. First assume r = p. If n (cid:3) 6 then the bound in (23) is suﬃcient unless q = 2 and n ∈ {6, 7, 8}, or (n, q) = (6, 3). All of these cases can be checked using Magma. Now assume n = 5, so m = 2. Here \|CΩ(x)\| is maximal when x = (J2, J 3 1 ) and we observe that \|xG ∩ H\| = α(2, q) + α(3, q) and \|xG\| = α(5, q), where α(d, q) = \|GUd(q)\| q2d−3\|GUd−2(q)\|\|GU1(q)\| is the number of transvections in GUd(q). Since fpr(x) = \|xG ∩ H\|/\|xG\|, it is straight- forward to check that fpr(x) (cid:2) (q + 1)−1 as required. For the remainder, let us assume x is semisimple. As before, let i (cid:3) 1 be minimal such that r divides qi − 1 and deﬁne j as in (19). Suppose i ≡ 2 (mod 4). We may write x = (x1, x2) in terms of an orthogonal de- composition V = V1 ⊥ V2 into nondegenerate spaces, where (cid:13)x(cid:14) acts homogeneously 54 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (and nontrivially) on V1 with the additional property that V1 and V2 have no com- mon (cid:13)x(cid:14)-irreducible constituent. If U is a nondegenerate m-space ﬁxed by x, then U = U1 ⊥ U2 and Ui ⊆ Vi is ﬁxed by xi. In particular, U is ﬁxed by y = (x1, 1) and thus CΩ(x) ⊆ CΩ(y). Therefore, in the notation of [16, Proposition 3.3.2], we may assume that x = (Λ(cid:3), In−j(cid:3)) for some (cid:4) (cid:3) 1. Similar reasoning shows that we may assume x = ((Λ, Λ−q)(cid:3), In−j(cid:3)) when i (cid:8)≡ 2 (mod 4). With this observation in hand, let us begin the main analysis by considering the case i = 2. If m (cid:3) 4 then the upper bound in (23) is suﬃcient, so we may assume m ∈ {2, 3} and x = (ωI(cid:3), In−(cid:3)). In terms of the f (a, b) notation introduced above, we have \|CΩ(x)\| = m(cid:9) k=0 f ((cid:4), m − k) · f (n − (cid:4), k) and it is easy to check that fpr(x) (cid:2) (q + 2)−1. For example, if m = 2 then we get \|CΩ(x)\| < q4(cid:3)−8 + q2n−4 + q4n−4(cid:3)−8, which is suﬃcient when combined with the lower bound on \|Ω\| in (22). Next assume i ≡ 2 (mod 4) and i (cid:3) 6, in which case r divides t = (qi/2 + 1)/(q + 1). Suppose m (cid:3) i/2. Here (23) is suﬃcient unless i = 6, m = 3 and either n = 7 or (n, q) = (8, 3) (note that q (cid:3) 3 if i = 6). These cases can be handled directly. For example, if (n, m, i) = (7, 3, 6) then we may assume x = (Λ, I4) or (Λ2, I1). In the latter case, we compute \|xG0 ∩ H\| = \|GU3(q)\| \|GU1(q3)\| · \|GU4(q)\| \|GU1(q3)\|\|GU1(q)\| < 2q18 and thus fpr(x) < 4q−18 since \|xG\| > 1 2 q36. Finally, if m < i/2 then \|CΩ(x)\| is equal to the number of nondegenerate m-spaces in CV (x), which implies that x = (Λ, In−i/2) has the most ﬁxed points. Therefore \|CΩ(x)\| (cid:2) f (n − i/2, m) and we quickly deduce that fpr(x) (cid:2) (t + 1)−1 as required. Now suppose i ≡ 0 (mod 4), so r divides qi/2 + 1. If m (cid:3) i then the bound in (23) is suﬃcient. On the other hand, if m < i then we may assume x = (Λ, Λ−q, In−i), in which case \|CΩ(x)\| = f (n − i, m) and it is easy to check that fpr(x) (cid:2) (qi/2 + 2)−1. Finally, suppose i is odd. If i = 1 then q (cid:3) 3 and for n (cid:3) 6 one can check that the bound in (23) is suﬃcient unless (n, m, q) = (6, 2, 3). In the latter case, r = 2, \|CΩ(x)\| is maximal when x = (−I1, I5) and we compute fpr(x) (cid:2) 1/81. Similarly, if i = 1 and n = 5 then m = 2 and \|CΩ(x)\| is maximal when x = (−I1, I4), in which case \|CΩ(x)\| = f (4, 1) + f (4, 2) and we obtain fpr(x) (cid:2) q−1. Finally, suppose i (cid:3) 3 and note that r divides t = (qi − 1)/(q − 1). If m (cid:3) 2i then it is easy to check that (23) is suﬃcient. For m < 2i we observe that \|CΩ(x)\| is maximal when x = (Λ, Λ−q, In−2i). Here \|CΩ(x)\| = f (n −2i, m) and it is straightforward to verify the bound fpr(x) (cid:2) (t +1)−1. (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 55 This completes the proof of Proposition 5.11. 5.4. Symplectic groups Next we turn to the subspace actions of almost simple symplectic groups. Through- out this section, we can assume G0 (cid:8)= PSp4(2)(cid:7), PSp4(3) in view of the isomorphisms PSp4(2)(cid:7) ∼ = A6 and PSp4(3) ∼ = U4(2). Proposition 5.16. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0 = PSpn(q) with n (cid:3) 4 and (n, q) (cid:8)= (4, 2), (4, 3). Assume H is a subspace subgroup of G and let x ∈ G be an element of prime order r. Then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) H = P1, r = q = p, x = (J2, J n−2 1 ) and fpr(x) = 1 q + 1 + q(qn−2 − 1) (q + 1)(qn − 1) . (ii) H is of type O(cid:2) n(q), r = q = 2, x = (J2, J n−2 1 ) and fpr(x) = 1 3 + 2n/2−1 − (cid:3) 3(2n/2 + (cid:3)) . (iii) H is of type O− n (q), r = 3, q = 2, x = (ω, ω−1, In−2) and fpr(x) = 1 4 + 3 4(2n/2 − 1) . Note that in view of the proof of Lemma 4.13, we are free to assume that G (cid:2) PΓSp4(q) when n = 4. Lemma 5.17. The conclusion to Proposition 5.16 holds if H = P1. Proof. First identify Ω with the set of 1-dimensional subspaces of V (note that every 1-dimensional subspace is totally singular) and observe that \|Ω\| = (qn − 1)/(q − 1). Let x ∈ G be an element of prime order r. The usual argument applies if x is a ﬁeld automorphism, so we may assume x ∈ PGSpn(q) is semisimple or unipotent. First assume r = p and recall that each block Ji in the Jordan form of x on V has even multiplicity if i is odd. Here \|CΩ(x)\| is equal to the number of 1-dimensional subspaces of CV (x), so \|CΩ(x)\| is maximal when x = (J2, J n−2 ). Working with this element, we get \|CΩ(x)\| = (qn−1 − 1)/(q − 1) and thus fpr(x) = (qn−1 − 1)/(qn − 1). If q (cid:8)= p, then it is easy to check that this gives fpr(x) (cid:2) (r + 1)−1 as required. However, if q = p then fpr(x) > (q + 1)−1 and this case is recorded in part (i) of Proposition 5.16. Finally, if 1 56 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 q = p and ν(x) (cid:3) 2, then \|CΩ(x)\| is maximal when x has Jordan form (J 2 1 which case \|CΩ(x)\| = (qn−2 − 1)/(q − 1) and we deduce that fpr(x) (cid:2) (q + 1)−1. 2 , J n−4 ), in For the remainder, let us assume r (cid:8)= p. If r = 2 then \|CΩ(x)\| is maximal when x = (−I2, In−2) and we compute \|CΩ(x)\| = q2 − 1 q − 1 + qn−2 − 1 q − 1 . This implies that fpr(x) (cid:2) 1/3. Now assume r is odd and let i (cid:3) 1 be minimal such that r divides qi − 1. Set j = 2i if i is odd, otherwise j = i. Note that if i (cid:3) 2 then \|CΩ(x)\| coincides with the number of 1-dimensional subspaces in CV (x). In particular, if i is even, then \|CΩ(x)\| is maximal when x = (Λ, In−i), so \|CΩ(x)\| (cid:2) (qn−j − 1)/(q − 1) and it is easy to check that fpr(x) (cid:2) (qi/2 + 2)−1 as required. The same upper bound on \|CΩ(x)\| holds if i (cid:3) 3 is odd (with equality if x = (Λ, Λ−1, In−j)) and we deduce that fpr(x) (cid:2) q−i. Finally, if i = 1 then \|CΩ(x)\| (cid:2) 2 + (qn−2 − 1)/(q − 1) and once again the result follows. (cid:2) Lemma 5.18. The conclusion to Proposition 5.16 holds if H = Pm with 2 (cid:2) m (cid:2) n/2. Proof. Here we identify Ω with the set of totally singular m-spaces in V , so qm(2n−3m+1)/2 < \|Ω\| = 1 2 \|Spn(q)\| qm(2n−3m+1)/2\|Spn−2m(q)\|\|GLm(q)\| (cid:5) (cid:6) < 2 q q − 1 qm(2n−3m+1)/2. We may assume x ∈ PGSpn(q) has prime order r. For n (cid:3) 6, [27, Proposition 3.15] gives fpr(x) < 2q−(n/2−1) + q−n/2 + q−m. (24) First assume r = p. If n (cid:3) 6 then one can check that the upper bound in (24) is suﬃcient unless q = 2 and n ∈ {6, 8, 10}, or (n, q) = (6, 3). All of these special cases can be handled using Magma. For example, suppose (n, q) = (6, 2). If m = 2, then \|Ω\| = 315 and \|CΩ(x)\| (cid:2) 75 (maximal if x = (J2, J 4 1 )), and for m = 3 we have \|Ω\| = 135 and \|CΩ(x)\| (cid:2) 39 (maximal if x is an a2-type involution in the notation of [2]). In both cases, fpr(x) (cid:2) 1/3 as required. Now assume r = p and n = 4, so m = 2. If p = 2, then H is Aut(G0)-conjugate to P1 and so the result in this case follows from the proof of the previous lemma. For q ∼ odd, we can use the fact that G0 = Ω5(q) and the action of G0 on Ω is permutation isomorphic to the action of Ω5(q) on the set Γ of 1-dimensional totally singular subspaces of the 5-dimensional orthogonal module. In terms of the respective Jordan forms, this isomorphism induces the following correspondence: (J2, J 2 1 ) (cid:19)→ (J 2 2 , J1), (J 2 2 ) (cid:19)→ (J3, J 2 1 ), (J4) (cid:19)→ (J5). T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 57 If x = (J4) then it is easy to see that \|CΩ(x)\| = 1. In the remaining cases, we can appeal to the proof of Lemma 5.22, which shows that \|CΓ(y)\| = q + 1 if y ∈ Ω5(q) has Jordan 1 ). In particular, \|CΩ(x)\| (cid:2) 2q + 1 when 2 , J1), and \|CΓ(y)\| = 2q + 1 if y = (J3, J 2 form (J 2 q is odd and once again we deduce that fpr(x) (cid:2) (q + 1)−1. Next suppose r = 2 and p is odd, so x is a semisimple involution. If q (cid:3) 5 then the bound in Theorem 4.5 is suﬃcient, so we may assume q = 3 (and thus n (cid:3) 6 since we are excluding the case (n, q) = (4, 3)). Here the bound in (24) is suﬃcient unless (n, m) = (6, 2), in which case a Magma computation yields fpr(x) (cid:2) 5/91. For the remainder, let us assume r (cid:8)= p and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. First assume i = 1, so q (cid:3) 4. If n (cid:3) 6 then the bound in (24) is eﬀective. Similarly, if i = 2 and n (cid:3) 6, then the same bound is suﬃcient unless m = q = 2 or (n, m, q) = (8, 4, 2). In the latter case we have r = 3 and a Magma calculation shows that fpr(x) (cid:2) 1/51. Now suppose m = i = q = 2, so n (cid:3) 6, r = 3 and x = (Λ(cid:3), In−2(cid:3)) for some (cid:4) (cid:3) 1. Let α be the number of totally singular 2-spaces in CV (x). Then by arguing as in the proof of Lemma 5.7(ii) we deduce that \|CΩ(x)\| (cid:2) α + (22(cid:3) − 1)/3 and it is straightforward to check that fpr(x) (cid:2) 1/4. Now assume i ∈ {1, 2} and n = 4, in which case m = 2. If i = 1 then \|CΩ(x)\| (cid:2) 2(q+1) (maximal if x = (Λ, Λ−1, I2)) and thus fpr(x) (cid:2) q−1 as required. Similarly, if i = 2 then fpr(x) > 0 if and only if x = (Λ2), in which case \|CΩ(x)\| = q + 1 and we conclude that fpr(x) (cid:2) (q + 2)−1. Next suppose i (cid:3) 4 is even. First assume m (cid:3) i, so n (cid:3) 2i. Here one can check that the bound in (24) yields fpr(x) (cid:2) (qi/2 + 2)−1 unless i = 4 and (n, q) = (8, 2), (8, 3) or (10, 2). Each of these cases can be handled using Magma. Now assume m < i. In this case, \|CΩ(x)\| coincides with the number of totally singular m-spaces in CV (x), so by working with the element x = (Λ, In−i) we deduce that fpr(x) < 4 (cid:6) (cid:5) q q − 1 q−mi (cid:2) (qi/2 + 2)−1 and the result follows. Finally, let us assume i (cid:3) 3 is odd. Note that r divides t = (qi − 1)/(q − 1). If m < i then it is clear that \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i) and thus (cid:5) (cid:6) fpr(x) < 4 q−2mi (cid:2) (t + 1)−1. q q − 1 Now suppose m (cid:3) i. If n < 4i then x = (Λ, Λ−1, In−2i) is the only possibility and we have \|CΩ(x)\| = β + 2γ, where β (respectively, γ) is the number of totally singular m-spaces (respectively (m − i)-spaces) in CV (x). Therefore, (cid:5) (cid:6) qm(2n−3m+1)/2 (cid:23) q−2im + 2q−i(2n−2m+1−i)/2 (cid:24) \|CΩ(x)\| < 2 and thus q q − 1 58 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 fpr(x) < 4 (cid:6) (cid:23) (cid:5) q q − 1 q−2im + 2q−i(2n−2m+1−i)/2 (cid:24) . One can check that this gives fpr(x) (cid:2) (t + 1)−1 unless (n, m, q) = (6, 3, 2) and i = 3. Here r = 7, \|CΩ(x)\| = 2 and the result follows. To complete the proof, we may assume i (cid:3) 3 is odd with m (cid:3) i and n (cid:3) 4i. If m (cid:3) i +1 then one can check that the bound in (24) is suﬃcient unless (n, m, q, i) = (12, 4, 2, 3). Here r = 7 and either x = ((Λ, Λ−1)2) and \|CΩ(x)\| = 0, or x = (Λ, Λ−1, I6) and \|CΩ(x)\| = β + 2γ as above. The reader can check that fpr(x) (cid:2) 1/8. Finally, suppose m = i. In this case, we ﬁnd that (24) is suﬃcient unless q = 2 and r = 2i − 1 is a Mersenne prime. Here dim CV (x) = n − 2m(cid:4) with (cid:4) (cid:3) 1 and it is straightforward to see that \|CΩ(x)\| is maximal when x = ((Λ, Λ−1)(cid:3), In−2m(cid:3)). Then by arguing as in the proof of Lemma 5.7(ii), viewing z = ((Λ, Λ−1)(cid:3)) as an element of the ﬁeld extension subgroup Sp2(cid:3)(2m) < Sp2(cid:3)m(2), we deduce that \|CΩ(x)\| = δ + 2 (cid:5) (cid:6) , 2m(cid:3) − 1 2m − 1 where δ is the number of totally singular m-spaces in CV (x). From here, the desired bound fpr(x) (cid:2) 2−m quickly follows. (cid:2) Lemma 5.19. The conclusion to Proposition 5.16 holds if H = Nm with 2 (cid:2) m < n/2 even. Proof. Here n (cid:3) 6 and we identify Ω with the set of nondegenerate m-dimensional subspaces of V . Note that qm(n−m) < \|Ω\| = \|Spn(q)\| \|Spm(q)\|\|Spn−m(q)\| < 2qm(n−m). Let x ∈ G be an element of prime order r. By [27, Proposition 3.16] we have fpr(x) < 2q−(n/2−d) + q−n/2 + q−m/2 + q−(n−m), (25) where d = (2, q − 1). Our aim is to establish the bound fpr(x) (cid:2) (r + 1)−1 with no exceptions. By the usual argument, we may assume x ∈ PGSpn(q). Suppose x is unipotent, so r = p. If m (cid:3) 4 then it is easy to check that the bound in (25) is suﬃcient unless (n, m, q) = (10, 4, 2). In the latter case, an easy Magma computation shows that fpr(x) (cid:2) 26/341 (maximal if x is a b1 involution). Now assume m = 2. Here \|CΩ(x)\| is maximal when x = (J2, J n−2 ) and we compute the bounds 1 \|xG ∩ H\| (cid:2) (q2 − 1) + (qn−2 − 1), \|xG\| (cid:3) 1 d (qn − 1), whence T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 59 fpr(x) (cid:2) d(qn−2 + q2 − 2) qn − 1 (cid:2) (q + 1)−1 and the result follows. Now assume r (cid:8)= p and x is semisimple. Suppose r = 2. By inspecting the bounds in Theorem 4.5 and (25), we may assume m = 2 and q = 3. The case n = 6 can be handled using Magma, so let us assume n (cid:3) 8. Here one can check that \|CΩ(x)\| is maximal when x = (−I2, In−2), which preserves an orthogonal decomposition V = U ⊥ W into nondegenerate spaces with dim U = 2. In particular, CΩ(x) comprises U and every nondegenerate 2-space in W , whence \|CΩ(x)\| = 1 + \|Spn−2(3)\| \|Sp2(3)\|\|Spn−4(3)\| < 2 · 32(n−4) and thus fpr(x) < 2/81. To complete the proof, we may assume x is semisimple and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. First assume i is even, so r divides qi/2 + 1. By arguing as in the proof of Lemma 5.15, we observe that \|CΩ(x)\| is maximal when x is of the form (Λ(cid:3), In−(cid:3)i) for some (cid:4) (cid:3) 1. If m < i then \|CΩ(x)\| coincides with the number of nondegenerate m-spaces in CV (x), so \|CΩ(x)\| is maximal when x = (Λ, In−i) and we deduce that fpr(x) < 2q−mi. This yields fpr(x) (cid:2) (qi/2 + 2)−1 and the result follows. Next assume m (cid:3) i + 2. Here one can check that the upper bound in (25) is suﬃcient unless (n, m, i, q) = (14, 6, 4, 2) or (m, i, q) = (4, 2, 2). In the former case, r = 5 and x = (Λ(cid:3), I14−4(cid:3)) with 1 (cid:2) (cid:4) (cid:2) 3, and it is straightforward to verify the bound fpr(x) (cid:2) 1/6 by computing \|xG ∩ H\| and \|xG\|. For example, if (cid:4) = 1 then \|xG ∩ H\| = \|Sp6(2)\| \|GU1(4)\|\|Sp2(2)\| + \|Sp8(2)\| \|GU1(4)\|\|Sp4(2)\| < 216(28 + 1) and \|xG\| > 247. Similarly, if (m, i, q) = (4, 2, 2) then r = 3 and we have x = (Λ(cid:3), In−2(cid:3)) with 1 (cid:2) (cid:4) (cid:2) n/2. Here \|xG\| > 1 2 22n(cid:3)−3(cid:3)2+(cid:3) and for (cid:4) (cid:3) 2 we compute \|xG ∩ H\| < \|Sp4(2)\| \|GU2(2)\| · \|Spn−4(2)\| \|GU(cid:3)−2(2)\|\|Spn−2(cid:3)(2)\| \|Spn−4(2)\| \|GU(cid:3)−1(2)\|\|Spn−2(cid:3)−2(2)\| + + · \|Sp4(2)\| \|GU1(2)\|\|Sp2(2)\| \|Spn−4(2)\| \|GU(cid:3)(2)\|\|Spn−2(cid:3)−4(2)\| < 22n(cid:3)−3(cid:3)2+(cid:3) (cid:3) 2−4n+4(cid:3)+8 + 2−2n−2(cid:3)+10 + 2−8(cid:3) (cid:4) . It is easy to check that this yields fpr(x) (cid:2) 1/4 and a very similar argument shows that the same conclusion holds when (cid:4) = 1. 60 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 To complete the argument when i is even, it remains to handle the case m = i. As noted above, we may assume x = (Λ(cid:3), In−m(cid:3)) and we compute \|xG\| > 1 2 qm(cid:3)(2n−(cid:3)−m(cid:3)+1)/2 \|xG ∩ H\| < \|Spm(q)\| \|GU1(qm/2)\| · < qm(cid:3)(2n−(cid:3)−m(cid:3)+1)/2 \|Spn−m(q)\| \|GU(cid:3)−1(qm/2)\|\|Spn−m(cid:3)(q)\| + (cid:24) (cid:23) qm(cid:3)+m2−m−mn + q−m2(cid:3) \|Spn−m(q)\| \|GU(cid:3)(qm/2)\|\|Spn−m−m(cid:3)(q)\| . It is straightforward to check that these bounds imply that fpr(x) (cid:2) (qm/2 + 2)−1. Finally, suppose i is odd. If m < 2i then \|CΩ(x)\| is the number of nondegenerate m-spaces in CV (x), which is maximal when x = (Λ, Λ−1, In−2i). The reader can check that fpr(x) < 2q−2mi (cid:2) q−i and the result follows. For m (cid:3) 2i + 2 it is easy to show that the upper bound in (25) is suﬃcient, so we may assume m = 2i. As noted above, we may also assume that x = ((Λ, Λ−1)(cid:3), In−m(cid:3)) and it is straightforward to show that fpr(x) < 8 (cid:23) qm(cid:3)+m2−m−mn + q−m2(cid:3) (cid:24) , which yields fpr(x) (cid:2) q−i. (cid:2) Lemma 5.20. The conclusion to Proposition 5.16 holds if q is even and H is of type O(cid:2) n(q). Proof. Here H ∩ G0 = O(cid:2) prime order r. As usual, we may assume x ∈ PGSpn(q). n(q) and \|Ω\| = qn/2(qn/2 + (cid:3))/2. Let x ∈ G be an element of First assume r = 2. If x = (J2, J n−2 1 ) is a b1-type involution then n(q)\| \|xG ∩ H\| = \|O(cid:2) 2\|Spn−2(q)\| = qn/2−1(qn/2 − (cid:3)), and thus fpr(x) = qn/2−1/(qn/2 + (cid:3)). This is at most 1/3 if and only if q (cid:3) 4, so the case q = 2 is an exception and it is recorded in part (ii) of Proposition 5.16. Now assume x ∈ G is an involution with ν(x) = s (cid:3) 2. By applying the bounds in the proof of [13, Proposition 3.22], we deduce that fpr(x) < 4q−s, which is suﬃcient unless q = 2 and s ∈ {2, 3}. If x = a2 then \|xG\| = qn − 1 \|xG ∩ H\| = and n(2)\| \|O(cid:2) n−4(2)\|\|Sp2(2)\| = 22n−7\|O(cid:2) (2n−2 − 1)(2n/2−2 + (cid:3))(2n/2 − (cid:3)) 1 3 \|xG\| = \|Spn(2)\| 22n−5\|Spn−4(2)\|\|Sp2(2)\| = 1 3 (2n−2 − 1)(2n − 1) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 61 which yields fpr(x) = (2n/2−2 + (cid:3))/(2n/2 + (cid:3)) (cid:2) 1/3 (with equality if (n, (cid:3)) = (6, +)). A similar calculation shows that the same conclusion holds when x = c2. And for x = b3 we get fpr(x) = 2n/2−3 2n/2 + (cid:3) (cid:2) 1 7 and the result follows. For the remainder, let us assume r is odd. As usual, let i (cid:3) 1 be minimal such that r divides qi − 1. First assume i is even and write x = (Ie, Λa1 t ), where e = dim CV (x). If e = 0 then by computing \|xG0 ∩ H\| and \|xG0\| we deduce that \|CΩ(x)\| = 1 and fpr(x) (cid:2) (qi/2 + 2)−1 as required. Now assume e > 0, so we may view CV (x) as a nondegenerate orthogonal space of type (cid:3)(cid:7) and we get (cid:5) 1 , . . . , Λat (cid:6) fpr(x) = qe/2(qe/2 + (cid:3)(cid:7)) qn/2(qn/2 + (cid:3)) (cid:2) q−i/2 q(n−i)/2 + 1 qn/2 − 1 since e (cid:2) n − i. One can now check that this bound gives fpr(x) (cid:2) (qi/2 + 2)−1 unless i = q = 2. Here r = 3 and we quickly reduce to the case where x = (Λ, In−2). If (cid:3) = + then (cid:3)(cid:7) = − and we obtain fpr(x) (cid:2) 1/4. However, if (cid:3) = − then (cid:3)(cid:7) = + and we compute fpr(x) = 1 4 + 3 4(2n/2 − 1) . The latter case is recorded in part (iii) of Proposition 5.16. Finally, suppose i is odd. As above, if e = 0 then \|CΩ(x)\| = 1 and we deduce that fpr(x) (cid:2) q−i. Similarly, for e > 0 we get fpr(x) = qe/2(qe/2 + (cid:3)) qn/2(qn/2 + (cid:3)) (cid:2) q−i (cid:5) qn/2−i + 1 qn/2 + 1 (cid:6) (cid:2) q−i and the result follows. (cid:2) 5.5. Odd dimensional orthogonal groups We begin our analysis of subspace actions of orthogonal groups by handling the groups with socle G0 = Ωn(q), where nq is odd. Note that we may assume n (cid:3) 7 since Ω3(q) ∼ = L2(q) and Ω5(q) ∼ = PSp4(q). Our main result is the following. Note that in part (i), x ∈ SOn(q) is an involution of type (−In−1, I1) with a plus-type (−1)-eigenspace. Whereas in part (ii), the (−1)- eigenspace of x is a minus-type space. Proposition 5.21. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group with point stabilizer H and socle G0 = Ωn(q), where n (cid:3) 7 and q is odd. Assume H is a 62 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 subspace subgroup of G. If x ∈ G has prime order r, then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) H = P1, r = 2, q = 3, x = (−In−1, I1)+ and fpr(x) = 1 3 + 2 3(3(n−1)/2 + 1) . (ii) H = N − 1 , r = 2, q = 3, x = (−In−1, I1)− and fpr(x) = 1 3 + 2(3(n−3)/2 + 1) 3(n−1)/2(3(n−1)/2 − 1) . Lemma 5.22. The conclusion to Proposition 5.21 holds if H = P1. Proof. As usual, we identify Ω with the set of 1-dimensional totally singular subspaces of V , noting that \|Ω\| = \|SOn(q)\| qn−2\|SOn−2(q)\|\|GL1(q)\| = qn−1 − 1 q − 1 . Let x ∈ G be an element of prime order r. By arguing in the usual fashion, we may assume x ∈ SOn(q) is semisimple or unipotent. First assume r = p, in which case \|CΩ(x)\| is the number of totally singular 1-spaces in CV (x). Let us also recall that each Ji block in the Jordan form of x on V has even 2 , J n−4 ) then CV (x) = U ⊕ W , where U is a totally multiplicity if i is even. If x = (J 2 1 singular 2-space and W is a nondegenerate (n − 4)-space, and we compute \|CΩ(x)\| = q2 − 1 q − 1 + q2 (cid:5) (cid:6) qn−5 − 1 q − 1 = qn−3 − 1 q − 1 . Similarly, if x = (J3, J n−3 ) then CV (x) = (cid:13)u(cid:14) ⊕ W , where u is totally singular and W is nondegenerate of dimension n − 3. Here CΩ(x) comprises (cid:13)u(cid:14) and every 1-space of the form (cid:13)λu + w(cid:14), where (cid:13)w(cid:14) ⊆ W is totally singular and λ ∈ Fq. Therefore, \|CΩ(x)\| is maximal when W is a plus-type space and in this situation we get 1 \|CΩ(x)\| = 1 + q (cid:5) (q(n−5)/2 + 1)(q(n−3)/2 − 1) q − 1 (cid:6) . In both cases, it is easy to check that fpr(x) (cid:2) (q + 1)−1. If x is any other unipotent element of order p, then dim CV (x) (cid:2) n − 4 and the bound \|CΩ(x)\| (cid:2) (qn−4 − 1)/(q − 1) is suﬃcient. Now assume r (cid:8)= p. If r = 2 then \|CΩ(x)\| is maximal when x ∈ SOn(q) is an involution of the form (−In−1, I1) with a plus-type (−1)-eigenspace. Here T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 63 and thus \|CΩ(x)\| = (q(n−3)/2 + 1)(q(n−1)/2 − 1) q − 1 fpr(x) = (q(n−3)/2 + 1)(q(n−1)/2 − 1) qn−1 − 1 . For q (cid:3) 5 it is easy to check that this is at most 1/3 and the result follows. On the other hand, if q = 3 then fpr(x) = 1 3 + 2 3(3(n−1)/2 + 1) and this special case is recorded in part (i) of Proposition 5.21. Note that if q = 3 and x is any other involution, then \|CΩ(x)\| is maximal when x = (−In−1, I1) has a minus-type (−1)-eigenspace, in which case \|CΩ(x)\| = 1 2 (3(n−3)/2 − 1)(3(n−1)/2 + 1) and we conclude that fpr(x) (cid:2) 1/3. To complete the proof, we may assume x is semisimple and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. Note that if i (cid:3) 2 then CΩ(x) comprises the totally singular 1-spaces in the nondegenerate space CV (x). First assume i is even. By the previous observation, it follows that \|CΩ(x)\| is maximal when x = (Λ, In−i), in which case \|CΩ(x)\| = (qn−i−1 − 1)/(q − 1) and it is easy to check that fpr(x) (cid:2) (qi/2 + 2)−1. Similarly, if i (cid:3) 3 is odd then \|CΩ(x)\| (cid:2) (qn−2i−1 −1)/(q −1) and we obtain fpr(x) (cid:2) q−i. Finally, if i = 1 then \|CΩ(x)\| (cid:2) 2 +(qn−3 −1)/(q −1) (maximal when x = (Λ, Λ−1, In−2)) and once again the desired bound holds. (cid:2) Lemma 5.23. The conclusion to Proposition 5.21 holds if H = Pm with 2 (cid:2) m < n/2. Proof. Here Ω is the set of totally singular m-spaces in V and we have qm(2n−3m−1)/2 < \|Ω\| = 1 2 qm(2n−3m−1)/2\|SOn−2m(q)\|\|GLm(q)\| < 2qm(2n−3m−1)/2. \|SOn(q)\| Let x ∈ G be an element of prime order r. As usual, we may assume x ∈ SOn(q) is semisimple or unipotent. By [27, Proposition 3.15] we have fpr(x) < 2q−(n−3)/2 + q−(n−1)/2 + q−m. (26) If r = p then the bound in (26) is suﬃcient unless (n, q) = (7, 3), in which case a Magma computation yields fpr(x) (cid:2) 37/280. Similarly, if r = 2 and p is odd, then the 64 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 bound in (26) is suﬃcient unless (n, m, q) = (7, 2, 3), where we obtain fpr(x) (cid:2) 1/7 by a straightforward Magma calculation. Finally, suppose r (cid:8)= p and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. If i (cid:2) 2 then it is easy to check that the bound in (26) is always suﬃcient, so we may assume i (cid:3) 3. Suppose i (cid:3) 4 is even. If m (cid:3) i then n (cid:3) 2i + 1 and (26) implies that fpr(x) < 2q−i + 2q1−i, which in turn is less than (qi/2 + 1)−1 unless (n, q, i) = (9, 3, 4). Here r = 5 and one checks that the previous bound yields fpr(x) (cid:2) 1/6. On the other hand, if m < i then \|CΩ(x)\| is the number of totally singular m-spaces in CV (x), whence \|CΩ(x)\| < 2qm(2n−3m−1)/2 · q−mi (maximal when x = (Λ, In−i)) and thus fpr(x) < 4q−mi (cid:2) (qi/2 + 2)−1. Similar reasoning applies when i (cid:3) 3 is odd. Here let us ﬁrst observe that r divides t = (qi − 1)/(q − 1). Now, if m < i then \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i), in which case \|CΩ(x)\| < 2qm(2n−3m−1)/2 · q−2mi and we get fpr(x) < 4q−2mi (cid:2) (t +1)−1 as required. Now assume m (cid:3) i. If n = 2i +1 then m = i, x = (Λ, Λ−1, I1) and \|CΩ(x)\| = 2. Similarly, if n = 2i + 3 then x = (Λ, Λ−1, I3) and either m = i and \|CΩ(x)\| = 2, or m = i + 1 and \|CΩ(x)\| = 2(q + 1). In each of these cases, the desired bound holds. Finally, if n (cid:3) 2i + 5 then the bound in (26) is suﬃcient. (cid:2) Lemma 5.24. The conclusion to Proposition 5.21 holds if H = N (cid:2) 1. Proof. First identify Ω with the set of nondegenerate 1-spaces U such that U ⊥ has type (cid:3) (in this situation, we will refer to U itself as an (cid:3)-type 1-space). Note that \|Ω\| = \|SOn(q)\| 2\|SO(cid:2) n−1(q)\| = q(n−1)/2(q(n−1)/2 + (cid:3)). 1 2 Let x ∈ G be an element of prime order r. We may assume x ∈ SOn(q) is semisimple or unipotent. First assume r = p and note that \|CΩ(x)\| coincides with the total number of (cid:3)- type 1-spaces in CV (x). In particular, if dim CV (x) (cid:2) n − 3 then \|CΩ(x)\| is at most (qn−3 − 1)/(q − 1) and it is easy to check that fpr(x) (cid:2) (q + 1)−1. Therefore, we may assume x = (J 2 ). In the former case, we compute ) or (J3, J n−3 2 , J n−4 1 1 \|xG ∩ H\| = n−1(q)\| \|SO(cid:2) q2n−9\|Sp2(q)\|\|SO(cid:2) n−5(q)\| , \|xG\| = \|SOn(q)\| q2n−7\|Sp2(q)\|\|SOn−4(q)\| and we obtain fpr(x) (cid:2) (q + 1)−1. A very similar calculation establishes the same bound when x = (J3, J n−3 ). 1 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 65 Now assume r (cid:8)= p. First we handle the case r = 2. Let x = (−In−1, I1) be an involution with a minus-type (−1)-eigenspace. If (cid:3) = + then \|xG ∩ H\| = \|SO+ n−1(q)\| 2\|SOn−2(q)\| , \|xG\| = \|SOn(q)\| − n−1(q)\| 2\|SO and we compute fpr(x) = q−1 (cid:2) 1/3. Similarly, if (cid:3) = − then fpr(x) = 2 + q(n−3)/2(q(n−1)/2 + 1) q(n−1)/2(q(n−1)/2 − 1) , which is at most 1/3 if q (cid:3) 5. However, if q = 3 then one checks that fpr(x) > 1/3 and so this case is recorded in part (ii) of Proposition 5.21. Now if x is any other involution, then \|CΩ(x)\| is maximal when x = (−In−1, I1) has a plus-type (−1)-eigenspace and it is straightforward to check that fpr(x) (cid:2) 1/3. Finally, let us assume x is semisimple and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1 and note that \|CΩ(x)\| is maximal when x = (Λ, In−i) (for i even) or x = (Λ, Λ−1, In−2i) (for i odd). If we ﬁx such an element x, then \|CΩ(x)\| is the number of (cid:3)(cid:7)-type 1-spaces in CV (x), where (cid:3)(cid:7) = (cid:3) if i is odd, otherwise (cid:3)(cid:7) = −(cid:3). For example, if i is even and (cid:3) = +, then x preserves an orthogonal decomposition V = U ⊥ W , with x acting irreducibly on the minus-type i-space U . It follows that \|CΩ(x)\| is the number of nondegenerate 1-spaces in W = CV (x) with a minus-type orthogonal complement in W , whence (cid:3)(cid:7) = −. By setting j = i if i is even and j = 2i if i is odd, we deduce that \|CΩ(x)\| (cid:2) 1 2 q(n−j−1)/2(q(n−j−1)/2 + (cid:3)(cid:7)). For i even, it is easy to check that this bound implies that fpr(x) (cid:2) (qi/2 +2)−1. Similarly, if i is odd then we obtain fpr(x) (cid:2) q−i. (cid:2) Lemma 5.25. The conclusion to Proposition 5.21 holds if H = N η m with 2 (cid:2) m < n/2. Proof. Recall that if m is even, then we may identify Ω with the set of nondegenerate m-spaces of type η ∈ {+, −}. Similarly, if m is odd then we take Ω to be the set of nondegenerate m-spaces U such that U ⊥ has type η. If m is even, then (cid:6) (cid:5) 1 4 q q + 1 qm(n−m) < \|Ω\| = \|SOn(q)\| m(q)\|\|SOn−m(q)\| < qm(n−m) 2\|SOη and it is easy to check that the same upper and lower bounds on \|Ω\| are also valid when m is odd. Let x ∈ G be an element of prime order r. As usual, we may assume x ∈ SOn(q) is semisimple or unipotent. By [27, Proposition 3.16] we note that fpr(x) < 2q−(n−3)/2 + q−(n−1)/2 + q−m + q−(n−m−α)/2, (27) 66 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 where α = 1 − η if m is odd, otherwise α = 1. First assume r = p. If n (cid:3) 9 then the bound in (27) is suﬃcient, so let us assume n = 7. If m = 2, or if m = 3 and η = +, then one checks that (27) is eﬀective if q (cid:3) 7, while the cases q ∈ {3, 5} can be checked using Magma. Now assume m = 3 and η = −. Here we ﬁnd that \|CΩ(x)\| is maximal when x = (J3, J 4 1 ) and we compute \|xG ∩ H\| = \|xG\| = \|SO + − 4 (q)\| 2q2 \|SO7(q)\| 2q5\|SO − 4 (q)\| \|SO3(q)\| q = 1 2 (q2 − 1)(q2 + 3) = 1 2 q2(q2 − 1)(q6 − 1), which immediately implies that fpr(x) (cid:2) (q + 1)−1. If r = 2 then the bound in (27) is suﬃcient unless (n, q) = (7, 3); in the latter case, we can use Magma to show that fpr(x) (cid:2) 1/3. Finally, suppose r (cid:8)= p and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. First assume i is even. As in previous cases, we are free to assume that x = (Λ(cid:3), In−(cid:3)i) for some (cid:4) (cid:3) 1. If m < i then \|CΩ(x)\| is the number of η-type m-spaces in CV (x), which is maximal when x = (Λ, In−i) and we deduce that fpr(x) < 4 (cid:6) (cid:5) q + 1 q q−mi (cid:2) (qi/2 + 2)−1. For m (cid:3) i one checks that the bound in (27) is suﬃcient unless n = 2m + 1 and m ∈ {i, i + 1}. Suppose n = 2m + 1 and m = i + 1. Here m is odd and there are three cases to consider. If x = (Λ, Im+2) then \|xG0 ∩ H\| = \|SOm(q)\| \|GU1(qi/2)\| + \|SOη m+1(q)\| −η 2 (q)\|\|GU1(qi/2)\| \|SO 1 2 m2 < q and by estimating \|xG\| we deduce that (cid:6) (cid:5) fpr(x) < 2 q−m2+m+1/2 (cid:2) (qi/2 + 2)−1. q + 1 q Similarly, one can check that the same conclusion holds if x = (Λ2, I3). Finally, if n = 7 and x = (Λ3, I1) then \|xG0 ∩ H\| (cid:2) \|SO3(q)\| \|GU1(q)\| · 4 (q)\| \|O+ \|GU2(q)\| < 2q4 and the result follows since \|xG\| > 1 2 (q + 1)−1q13. Now assume n = 2m + 1 and m = i, in which case x = (Λ, Im+1) or (Λ2, I1). If x = (Λ2, I1) then T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 67 \|xG0 ∩ H\| (cid:2) \|O− m(q)\| \|GU1(qm/2)\| · \|SOm+1(q)\| \|GU1(qm/2)\| < 2qm(m−1) and we deduce that fpr(x) < 4 (cid:6) (cid:5) q + 1 q q−m2 (cid:2) (qi/2 + 2)−1. One can check that the same conclusion holds when x = (Λ, Im+1). A very similar argument applies when i is odd. If m < 2i then \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i) and we deduce that (cid:6) (cid:5) fpr(x) < 4 q−2mi (cid:2) q−i. q + 1 q For m (cid:3) 2i, one can check that the bound in (27) is suﬃcient unless n = 2m + 1 and m ∈ {2i, 2i + 1}. The analysis of these remaining cases is essentially identical to the argument given above in the analogous cases with i even. We leave the reader to check the details. (cid:2) 5.6. Even dimensional orthogonal groups To complete the proof of Theorem 5.1, it remains to handle the subspace actions of the even dimensional orthogonal groups. Our main result is Proposition 5.26 below. Notice that in part (v), we use the notation x = (−In−1, I1)δ to denote a semisimple involution such that CV (x) = (cid:13)v(cid:14) has discrimi- nant δ ∈ {(cid:4), (cid:5)} (that is, if Q is the deﬁning quadratic form on V , then Q(v) ∈ F × is a q square if δ = (cid:4) and a nonsquare if δ = (cid:5)). Also note that in part (ii), both (−In−1, I1)(cid:4) and (−In−1, I1)(cid:5) have the same number of ﬁxed points on Ω, so there is no need to specify a label. Proposition 5.26. Let G (cid:2) Sym(Ω) be a ﬁnite almost simple primitive permutation group n(q) with n (cid:3) 8 even. Assume H is a subspace with point stabilizer H and socle G0 = PΩ(cid:2) subgroup of G. If x ∈ G has prime order r, then either fpr(x) (cid:2) (r + 1)−1, or one of the following holds: (i) H = P1, r = q = 2, x = (J2, J n−2 1 ) and fpr(x) = 1 3 + 2n−2 − (cid:3)2n/2−1 − 2 3(2n/2−1 + (cid:3))(2n/2 − (cid:3)) . (ii) H = P1, (cid:3) = −, r = 2, q = 3, x = (−In−1, I1) and fpr(x) = 1 3 + 2 3(3n/2 + 1) . 68 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (iii) H = P1, (cid:3) = −, r = 3, q = 2, x = (ω, ω−1, In−2) and fpr(x) = 1 4 + 3 4(2n/2 + 1) . (iv) H = P2, ((cid:3), n, q) = (+, 8, 4), r = 5, x = (ω, ω−1, I6) or (ωI4, ω−1I4), and we have fpr(x) = 1/5. (v) H = N1, r = 2, q = 3, x = (−In−1, I1)δ and (cid:12) fpr(x) = 4 3(3n/2−1) 1 3 + 3 + 2(3n/2−2+1) 1 3n/2−1(3n/2+1) if (cid:3) = + if (cid:3) = −, where δ = (cid:5) if (cid:3) = +, otherwise δ = (cid:4). ) and (vi) H = N1, r = q = 2, x = (J2, J n−2 1 fpr(x) = 1 3 + 2n/2−1 + (cid:3) 3(2n/2 − (cid:3)) . (vii) H = N1, (cid:3) = +, r = 3, q = 2, x = (ω, ω−1, In−2) and fpr(x) = 1 4 + 3 4(2n/2 − 1) . The proof of Proposition 5.26 will be given in the sequence of lemmas presented below. In the proofs, we will sometimes write (cid:4)η to denote a nondegenerate (cid:4)-space of type η. For example, if (cid:3) = − then V = U ⊥ W = 2+ ⊥ (n − 2)− denotes an orthogonal decomposition of V into nondegenerate spaces, where U is a plus-type 2-space and W is a minus-type space of dimension n − 2. Lemma 5.27. The conclusion to Proposition 5.26 holds if H = P1. Proof. Identify Ω with the set of totally singular 1-dimensional subspaces of V and note that \|Ω\| = n(q)\| \|SO(cid:2) n−2(q)\|\|GL1(q)\| = qn−2\|SO(cid:2) (qn/2−1 + (cid:3))(qn/2 − (cid:3)) q − 1 . Let x ∈ G be an element of prime order r. If G0 = PΩ+ 8 (q) then the maximality of H implies that G does not contain any triality automorphisms (see [30] or [7, Table 8.50], for example). Since ﬁeld and graph-ﬁeld automorphisms can be handled in the usual manner, we may assume x ∈ PGO(cid:2) n(q) is semisimple or unipotent. First assume r = p, in which case \|CΩ(x)\| is equal to the number of totally singular 1- spaces in CV (x). In particular, if dim CV (x) (cid:2) n −3 then \|CΩ(x)\| (cid:2) (qn−3−1)/(q−1) and T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 69 we quickly deduce that fpr(x) (cid:2) (q + 1)−1. Therefore, we may assume dim CV (x) (cid:3) n − 2 and we will consider the cases p = 2 and p odd separately. Suppose p = 2 and note that the condition dim CV (x) (cid:3) n −2 implies that x is of type b1, a2 or c2 in the notation of [2]. First assume x = (J2, J n−2 ) is a b1-type involution, which ﬁxes an orthogonal decomposition U ⊥ W = 2+ ⊥ (n − 2)(cid:2) of the natural module. Here CU (x) = (cid:13)u(cid:14) is nonsingular and we may assume Q(u) = 1. Then CΩ(x) comprises the set of totally singular 1-spaces in W , together with the spaces (cid:13)u +w(cid:14), where (cid:13)w(cid:14) ⊆ W is nonsingular and Q(w) = 1. It follows that \|CΩ(x)\| coincides with the total number of 1-dimensional subspaces of W and thus 1 fpr(x) = qn/2−1 − (cid:3) qn/2 − (cid:3) . (28) If q (cid:3) 4 then it is easy to check that fpr(x) (cid:2) 1/3. However, it q = 2 then fpr(x) > 1/3 and this case is recorded in part (i) of Proposition 5.26. 2 , J n−4 1 Now assume x = (J 2 ) is of type a2 or c2. We claim that fpr(x) (cid:2) 1/3. If x = a2 then x ﬁxes an orthogonal decomposition U ⊥ W = 4+ ⊥ (n − 4)(cid:2) with CU (x) = (cid:13)u1, u2(cid:14) totally singular. It follows that CΩ(x) comprises the set of 1-dimensional subspaces of CU (x), together with every 1-space of the form (cid:13)u + w(cid:14), where u ∈ CU (x) and (cid:13)w(cid:14) ⊆ W is totally singular. Therefore, \|CΩ(x)\| = q2 (cid:5) (qn/2−3 + (cid:3))(qn/2−2 − (cid:3)) q − 1 (cid:6) + q + 1. Similarly, if x = c2 then x ﬁxes a decomposition U1 ⊥ U2 ⊥ W = 2+ ⊥ 2− ⊥ (n − 4)−(cid:2) with CUi (x) = (cid:13)ui(cid:14) and Q(ui) = 1. Here CΩ(x) comprises the spaces (cid:13)λ(u1 +u2) +w(cid:14) and (cid:13)u + w(cid:7)(cid:14), where λ ∈ Fq and w, w(cid:7) ∈ W , u ∈ (cid:13)u1, u2(cid:14) are vectors such that Q(w) = 0 and Q(w(cid:7)) = Q(u) = 1. Let α be the number of totally singular 1-spaces in W . Then there are 1 + qα spaces of the form (cid:13)λ(u1 + u2) + w(cid:14) and there are q((qn−4 − 1)/(q − 1) − α) spaces of the form (cid:13)u + w(cid:7)(cid:14), whence \|CΩ(x)\| = q (cid:6) (cid:5) qn−4 − 1 q − 1 + 1. In both cases, it is easy to check that fpr(x) (cid:2) 1/3. 1 1 2 , J n−4 Next assume r = p > 2. Since we are free to assume dim CV (x) (cid:3) n − 2, it follows that ) or (J3, J n−3 x = (J 2 ) because every even size Jordan block must occur with an even multiplicity. First assume x = (J 2 ), which ﬁxes an orthogonal decomposition U ⊥ W = 4+ ⊥ (n − 4)(cid:2) with CU (x) = (cid:13)u1, u2(cid:14) totally singular. Then CΩ(x) comprises every 1-dimensional subspace of (cid:13)u1, u2(cid:14), together with the 1-spaces (cid:13)u + w(cid:14), where u ∈ (cid:13)u1, u2(cid:14) and (cid:13)w(cid:14) ⊆ W is totally singular. Therefore, if α denotes the number of totally singular 1-spaces in W , then \|CΩ(x)\| = αq2 + q + 1 and it is easy to check that fpr(x) (cid:2) (q + 1)−1. Now assume x = (J3, J n−3 ), ﬁxing a decomposition U ⊥ W = 3 ⊥ 2 , J n−4 1 1 70 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (n − 3) with CU (x) = (cid:13)u(cid:14). Here CΩ(x) comprises (cid:13)u(cid:14) and (cid:13)λu + w(cid:14), where λ ∈ Fq and (cid:13)w(cid:14) is a totally singular 1-space in W . Therefore, \|CΩ(x)\| = q(qn−4 − 1)/(q − 1) + 1 and once again the desired bound holds. For the remainder, we may assume r (cid:8)= p. Suppose r = 2 and note that the bound in Theorem 4.5 is suﬃcient if q (cid:3) 5, so we may assume q = 3. Here \|CΩ(x)\| is maximal when x is of the form (−In−1, I1); there are two such conjugacy classes and elements in both classes have the same number of ﬁxed points. We get \|CΩ(x)\| = (qn−2 − 1)/(q − 1) and thus (28) holds. One can now check that fpr(x) (cid:2) 1/3 unless ((cid:3), q) = (−, 3), which is the case recorded in part (ii) of Proposition 5.26. Now, if ((cid:3), q) = (−, 3) and x is some other involution, then \|CΩ(x)\| is maximal when x = (−In−2, I2) with a plus-type (−1)-eigenspace and we compute \|CΩ(x)\| = 1 2 (3n/2−2 + 1)(3n/2−1 − 1), which coincides with the number of totally singular 1-spaces in the (−1)-eigenspace of x. Once again, it is easy to check that fpr(x) (cid:2) 1/3. To complete the proof, suppose r (cid:8)= p and r is odd. As usual, let i (cid:3) 1 be minimal such that r divides qi − 1. Note that if i (cid:3) 2 then \|CΩ(x)\| is the number of totally singular 1-spaces in CV (x). First assume i is even. By the previous observation, we see that \|CΩ(x)\| is maximal when x = (Λ, In−i), in which case \|CΩ(x)\| (cid:2) (q(n−i)/2−1 − (cid:3))(q(n−i)/2 + (cid:3)) q − 1 and one can check that this bound yields fpr(x) (cid:2) (qi/2+2)−1 unless (cid:3) = − and q = i = 2. Here r = 3 and we get fpr(x) = 1 4 + 3 4(2n/2 + 1) . This special case is recorded in part (iii) of Proposition 5.26. If (cid:3) = −, q = i = 2 and x is any other element of order 3, then \|CΩ(x)\| is maximal when x = (Λ2, In−4). Here \|CΩ(x)\| = (2n/2−3 − 1)(2n/2−2 + 1) is the number of totally singular 1-spaces in a nondegenerate minus-type space of dimension n − 4 and we quickly deduce that fpr(x) (cid:2) 1/4. Now assume i is odd. Here \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i), in which case \|CΩ(x)\| = α + β, where α = (q(n−2i)/2−1 + (cid:3))(q(n−2i)/2 − (cid:3)) q − 1 is the number of totally singular 1-spaces in CV (x) and we set β = 2 if i = 1, otherwise β = 0. It is straightforward to check that fpr(x) (cid:2) q−i and the result follows. (cid:2) T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 71 Lemma 5.28. The conclusion to Proposition 5.26 holds if H = Pm with 2 (cid:2) m (cid:2) n/2. Proof. For m < n/2 we identify Ω with the set of totally singular m-spaces in V . On the other hand, if m = n/2 then (cid:3) = + and we identify Ω with one of the two G0-orbits on the set of totally singular m-spaces (the actions of G on each orbit are permutation isomorphic, so it does not matter which orbit we choose). Note that \|Ω\| = \|SO(cid:2) qm(2n−3m−1)/2\|SO(cid:2) n(q)\| n−2m(q)\|\|GLm(q)\| if m < n/2, whereas \|Ω\| = \|O+ n (q)\| 2qn(n−2)/8\|GLn/2(q)\| if m = n/2. In particular, we have (cid:6) (cid:5) 1 2 q q + 1 qm(2n−3m−1)/2 < \|Ω\| < 4qm(2n−3m−1)/2 (29) 8 (q) then we may assume m ∈ {2, 3} since the action of for all m. Note that if G0 = PΩ+ G on each orbit of totally singular 4-spaces is permutation isomorphic to its action on the set of totally singular 1-spaces, which we handled in the previous lemma. Let x ∈ G be an element of prime order r. As usual, we may assume x is not a ﬁeld or graph-ﬁeld automorphism. If (n, q) ∈ {(8, 2), (8, 3), (8, 4), (10, 2), (12, 2)} (30) then the desired result can be checked directly using Magma, so for the remainder we will exclude these cases from the analysis. By [27, Proposition 3.15] we have fpr(x) < 2q−(n−2−2α)/2 + q−(n−2α)/2 + q−m, (31) where α = 1 if (cid:3) = +, otherwise α = 0. This bound immediately implies that fpr(x) (cid:2) 1/3 and so we may assume r is odd. In addition, we note that fpr(x) (cid:2) 1/4 if n = 8 and q (cid:3) 4, which handles the special case where G0 = PΩ+ 8 (q) and x is a triality graph automorphism. In particular, for the remainder we may assume x ∈ PGO(cid:2) n(q) is semisimple or unipotent. If r = p > 2 then the bound in (31) implies that fpr(x) (cid:2) (q + 1)−1, so we may assume r (cid:8)= p and r is odd. Let i (cid:3) 1 be minimal such that r divides qi − 1. There are two cases to consider, according to the parity of i. First assume i is even. If m < i then \|CΩ(x)\| is maximal when x = (Λ, In−i), in which case \|CΩ(x)\| is the number of totally singular m-spaces in the (−(cid:3))-type space CV (x). By applying the bounds in (29), we deduce that 72 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 fpr(x) < 8 (cid:6) (cid:5) q + 1 q q−mi (cid:2) (qi/2 + 2)−1 and the result follows. If m (cid:3) i + 1 then one can check that the bound in (31) is suﬃcient (recall that we may exclude the cases in (30)). Therefore, we may assume m = i. If n (cid:3) 2m + 4 then (31) is suﬃcient unless m = q = 2, in which case r = 3 and x = (Λ(cid:3), In−2(cid:3)) for some (cid:4) (cid:3) 1. In addition, we may assume n (cid:3) 14 (see (30)). Let β be the number of totally singular 2-spaces in CV (x). If (cid:4) = 1 then \|CΩ(x)\| = β and we compute fpr(x) (cid:2) (2n/2−3 − (cid:3))(2n/2−2 − (cid:3)) (2n/2−1 − (cid:3))(2n/2 − (cid:3)) (cid:2) 1 4 as required. Now assume (cid:4) (cid:3) 2. By arguing as in the proof of Lemma 5.7(ii) we deduce that \|CΩ(x)\| (cid:2) β + (22(cid:3) − 1)/3 and once again we conclude that fpr(x) (cid:2) 1/4. To complete the analysis for i even, we may assume m = i (cid:3) 4 and n ∈ {2m, 2m + 2}. Suppose n = 2m, in which case x is of the form (Λ, Im), (Λ2) or (Λ1, Λ2) (the latter two possibilities only occur when (cid:3) = +). Here x has ﬁxed points on Ω if and only if x = (Λ2), so let us assume x is of this form. We may embed ˆx as a scalar in a subgroup GU2(qm/2) < O+ 2m(q) and we note that there is a bijective correspondence between the set of totally singular m-spaces in V ﬁxed by x and the set of totally singular 1-spaces in the natural module for GU2(qm/2). This implies that \|CΩ(x)\| = qm/2 + 1 and the bound fpr(x) (cid:2) (qm/2 + 2)−1 quickly follows. Similarly, if n = 2m + 2 then we may assume x = (Λ2, I2) and the result follows since \|CΩ(x)\| = qm/2 + 1. Now assume i is odd. If m < i then it is plain to see that \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i), in which case CΩ(x) is the set of totally singular m-spaces in CV (x) and by applying the bounds in (29) we deduce that fpr(x) < 8 (cid:6) (cid:5) q + 1 q q−2mi (cid:2) q−i and the result follows. Now assume m (cid:3) i. If i = 1 then q (cid:3) 4 and one checks that the bound in (31) implies that fpr(x) (cid:2) q−1. Now assume i (cid:3) 3 and note that r divides t = (qi − 1)/(q − 1). If n < 4i then x = (Λ, Λ−1, In−2i) is the only possibility and we get \|CΩ(x)\| = 2β, where β is the number of totally singular (m − i)-spaces in CV (x), which is a nondegenerate (n − 2i)-space of type (cid:3). Therefore, fpr(x) < 8 (cid:6) (cid:5) q + 1 q q−i(2n−2m−i−1)/2 and we deduce that fpr(x) (cid:2) (t + 1)−1. To complete the proof, suppose i (cid:3) 3 is odd, m (cid:3) i and n (cid:3) 4i. If m (cid:3) i + 1 then the bound in (31) is suﬃcient unless (n, m, q, i) = (14, 4, 2, 3), in which case r = 7 and T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 73 x = (Λ, Λ−1, I8) or ((Λ, Λ−1)2, I2). With the aid of Magma, it is easy to check that fpr(x) (cid:2) 1/8 in both cases. Finally, suppose m = i. In this case, one checks that (31) is suﬃcient unless q = 2 and r = 2i − 1 is a Mersenne prime. Here we proceed as in the ﬁnal paragraph in the proof of Lemma 5.18. In particular, we ﬁrst observe that \|CΩ(x)\| is maximal when x is of the form ((Λ, Λ−1)(cid:3), In−2m(cid:3)) for some (cid:4) (cid:3) 1 and we compute (cid:6) (cid:5) \|CΩ(x)\| (cid:2) γ + 2 2m(cid:3) − 1 2m − 1 , where γ is the number of totally singular m-spaces in CV (x). It is now routine to check that this yields fpr(x) (cid:2) 2−m as required. (cid:2) If G0 = PΩ+ 8 (q) and G contains triality graph or graph-ﬁeld automorphisms, then G has a maximal parabolic subgroup H of type P1,3,4. This special case is handled in the following lemma. Lemma 5.29. The conclusion to Proposition 5.26 holds if G0 = PΩ+ 8 (q) and H = P1,3,4. Proof. Let x ∈ G be an element of prime order r and note that \|Ω\| = \|O+ 8 (q)\| 2q11\|GL2(q)\|\|GL1(q)\|2 = (q4 + q2 + 1)(q2 + 1)2(q + 1)3. The usual argument applies if x is a ﬁeld or graph-ﬁeld automorphism. Next assume x is a triality graph automorphism, so r = 3. If q (cid:3) 7 then the bound in Theorem 4.5 is suﬃcient, so we may assume q (cid:2) 5. In each of these cases we can use Magma to show that \|CΩ(x)\| (cid:2) (q6 − 1)/(q − 1), which yields fpr(x) (cid:2) 1/225, with equality if q = 2 and CG0(x) = G2(2). (For q = 5 we can construct H by observing that H = NG(P ), where P is a suitable index-ﬁve subgroup of a Sylow 5-subgroup of G0.) For the remainder, we may assume x ∈ PGO+ 8 (q) is either semisimple or unipotent. Since fpr(x) = \|xG ∩ H\|/\|xG\|, it follows that fpr(x) is at most the corresponding ﬁxed point ratio for the action of x on totally singular 1-spaces. Therefore, our earlier analysis of the case H = P1 (see Lemma 5.27) immediately implies that fpr(x) (cid:2) (r + 1)−1 unless q = 2 and x = b1 is a transvection. But here an easy Magma computation yields fpr(x) = 1/15 and the result follows. (cid:2) Lemma 5.30. The conclusion to Proposition 5.26 holds if H = N1 and q is odd. Proof. Here we may assume Ω is the set of nondegenerate 1-dimensional subspaces of V with square discriminant (recall that we refer to such a subspace as a square 1-space). Note that \|Ω\| = n(q)\| \|SO(cid:2) 2\|SOn−1(q)\| = 1 2 qn/2−1(qn/2 − (cid:3)). 74 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 We may assume x ∈ PGO(cid:2) implies that G does not contain any triality automorphisms when (n, (cid:3)) = (8, +). n(q) has prime order r, noting that the maximality of H First assume r = p and note that \|CΩ(x)\| is the number of square 1-spaces in CV (x). In particular, if dim CV (x) (cid:2) n − 3 then \|CΩ(x)\| (cid:2) (qn−3 − 1)/(q − 1) and it is easy to check that fpr(x) (cid:2) (q + 1)−1. Therefore, we may assume dim CV (x) = n − 2, in which case x has Jordan form (J 2 ). If x = (J 2 ) or (J3, J n−3 ) then 2 , J n−4 2 , J n−4 1 1 1 \|xG ∩ H\| = \|SOn−1(q)\| q2n−9\|Sp2(q)\|\|SOn−5(q)\| , \|xG\| = n(q)\| \|SO(cid:2) q2n−7\|Sp2(q)\|\|SO(cid:2) n−4(q)\| and we deduce that fpr(x) = q(n−4)/2 − (cid:3) qn/2 − (cid:3) . Similarly, if x = (J3, J n−3 1 ) then \|xG ∩ H\| (cid:2) \|SOn−1(q)\| qn−3\|SO+ n−4(q)\| , \|xG\| = and we get \|SO(cid:2) n(q)\| 2qn−2\|SOn−3(q)\| fpr(x) (cid:2) 2(q(n−4)/2 + 1) qn/2 − (cid:3) . In both cases, one can check that fpr(x) (cid:2) (q + 1)−1 and the result follows. Next assume r = 2. Here \|CΩ(x)\| is equal to the number of square 1-spaces in the eigenspaces of x on V . As a consequence, we observe that \|CΩ(x)\| is maximal when x = (−In−1, I1) ﬁxes an orthogonal decomposition V = U ⊥ W with W = CV (x). There are two cases to consider, according to the discriminant of the deﬁning quadratic form Q restricted to W . First assume this discriminant is a square, so x = (−In−1, I1)(cid:4), W is contained in Ω and we have \|CΩ(x)\| = 1 + α, where α is the number of square 1-spaces in U . If Y ⊆ U is such a subspace, then the orthogonal complement of Y in U is a nondegenerate (n − 2)-space of type η and we get α = 1 2 q(n−2)/2(q(n−2)/2 + η). If η = −, then it is easy to check that fpr(x) (cid:2) 1/3. The same conclusion holds if η = + and q (cid:3) 5. However, if η = + and q = 3 then fpr(x) > 1/3 and this case is recorded in part (v) of Proposition 5.26. Moreover, one can check that η = −(cid:3), so we only get fpr(x) > 1/3 when (cid:3) = −. Similarly, if x = (−In−1, I1)(cid:5) then \|CΩ(x)\| = α as above with η = (cid:3) and we deduce that fpr(x) (cid:2) 1/3 unless (q, (cid:3)) = (3, +). Again, this special case is recorded in Proposition 5.26. To summarize, fpr(x) > 1/3 if and only if q = 3 and x = (−In−1, I1)δ, where δ = (cid:5) if (cid:3) = +, and δ = (cid:4) if (cid:3) = −. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 75 Finally, let us assume r (cid:8)= p and r is odd. As usual, let i (cid:3) 1 be minimal such that r divides qi − 1 and note that \|CΩ(x)\| is equal to the number of square 1-spaces in CV (x). In particular, if i is even then \|CΩ(x)\| is maximal when x = (Λ, In−i) and we compute \|CΩ(x)\| (cid:2) 1 2 q(n−i)/2−1(q(n−i)/2 + (cid:3)), which implies that fpr(x) (cid:2) (qi/2 + 2)−1. Similarly, if i is odd then \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i) and we get \|CΩ(x)\| (cid:2) 1 2 q(n−2i)/2−1(q(n−2i)/2 − (cid:3)). This gives fpr(x) (cid:2) q−i and the result follows. (cid:2) Lemma 5.31. The conclusion to Proposition 5.26 holds if H = N1 and q is even. Proof. Here we identify Ω with the set of nonsingular 1-dimensional subspaces of V , whence \|Ω\| = n(q)\| \|Ω(cid:2) \|Spn−2(q)\| = qn/2−1(qn/2 − (cid:3)). Note that if G0 = Ω+ any triality automorphisms. Also note that if G = O(cid:2) centralizer of a b1-type involution. 8 (q) then the maximality of H implies that G does not contain n(q), then H = 2 × Spn−2(q) is the Let x ∈ G be an element of prime order r. As usual, we may assume x ∈ O(cid:2) n(q) is semisimple or unipotent. First assume r = 2. By arguing as in the proof of the previous lemma, we may assume dim CV (x) (cid:3) n −2 and thus x is an involution of type b1, a2 or c2 in the notation of [2]. In addition, we may assume q = 2 since the bound in Theorem 4.5 is suﬃcient if q (cid:3) 4. If x = b1 then \|xG ∩ H\| = 1 + \|Spn−2(2)\| 2n−3\|Spn−4(2)\| = 2n−2, \|xG\| = \|Ω\| and we deduce that fpr(x) = 1 3 + 2n/2−1 + (cid:3) 3(2n/2 − (cid:3)) . This special case is recorded in part (vi) of Proposition 5.26. Similarly, if x = a2 then we compute \|xG ∩ H\| = \|Spn−2(2)\| 22n−9\|Sp2(2)\|\|Spn−6(2)\| , \|xG\| = n(2)\| \|O(cid:2) 22n−7\|Sp2(2)\|\|O(cid:2) n−4(2)\| 76 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 and this gives fpr(x) = 2n/2−2 − (cid:3) 2n/2 − (cid:3) (cid:2) 1 3 . And for x = c2 we obtain fpr(x) = 2n/2−2/(2n/2 − (cid:3)) and once again fpr(x) (cid:2) 1/3. To complete the proof, we may assume r is odd, so x is semisimple. Let i (cid:3) 1 be minimal such that r divides qi − 1 and observe that CΩ(x) is the set of nonsingular 1-spaces in CV (x). Suppose i is even. Then \|CΩ(x)\| is maximal when x = (Λ, In−i), in which case \|CΩ(x)\| = q(n−i)/2−1(q(n−i)/2 + (cid:3)) and we deduce that fpr(x) (cid:2) (qi/2 + 2)−1 unless (cid:3) = + and q = i = 2. Here r = 3 and we get fpr(x) = 1 4 + 3 4(2n/2 − 1) as recorded in part (vii) of Proposition 5.26. Note that if q = 2 and x is any other element of order r = 3, then \|CΩ(x)\| is maximal when x = (Λ2, In−4) and it is easy to check that fpr(x) (cid:2) 1/4. Finally, if i is odd then \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i), so \|CΩ(x)\| (cid:2) qn/2−i−1(qn/2−i − (cid:3)) and it is straightforward to check that fpr(x) (cid:2) q−i. (cid:2) Lemma 5.32. The conclusion to Proposition 5.26 holds if H = N η m with 2 (cid:2) m (cid:2) n/2. Proof. Here we identify Ω with an orbit of nondegenerate m-dimensional subspaces of V of type η. More precisely, if m is even then Ω is the complete set of subspaces of the given type and either m < n/2, or ((cid:3), η) = (−, +) and m = n/2. On the other hand, if m is odd then m < n/2, q is odd and we may assume that Ω is the set of nondegenerate m-spaces with square discriminant. In all cases, let us observe that (cid:6) (cid:5) 1 4 q q + 1 qm(n−m) < \|Ω\| = n(q)\| \|O(cid:2) m(q)\|\|Oη(cid:2) \|Oη n−m(q)\| < 2qm(n−m), where (cid:3) = ηη(cid:7). Let x ∈ G be an element of prime order r. As usual, the desired bound quickly follows if x is a ﬁeld or graph-ﬁeld automorphism, while the maximality of H implies that G does not contain triality automorphisms when (n, (cid:3)) = (8, +). Therefore, for the remainder we may assume x ∈ PGO(cid:2) n(q) is unipotent or semisimple. By inspecting [27, Proposition 3.16] we deduce that T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 77 fpr(x) < 2q−(n−4)/2 + q−(n−2)/2 + q−m + q−(n−m−α)/2, (32) where α = 2 if m is even, otherwise α = 1. The groups with (n, q) as in (30) can be handled directly using Magma, so we will exclude these cases for the remainder of the proof. First assume r = p = 2. If q (cid:3) 4 then the bound in Theorem 4.5 is suﬃcient, so we may assume q = 2. As noted above, we may also assume that n (cid:3) 14 and one can check that the bound in (32) is eﬀective unless (n, m) = (14, 2). Here H = N − 2 (as noted in [31], N + 1 ) is a b1-type involution, in which case 2 is non-maximal when q (cid:2) 3) and \|CΩ(x)\| is maximal when x = (J2, J 12 \|xG ∩ H\| (cid:2) 1 2 \|O− 2 (2)\| + 12(2)\| \|O− 2\|Sp10(2)\| = 2083 and we deduce that fpr(x) (cid:2) 1/3 since \|xG\| (cid:3) 8128. Next assume r = p > 2. Here the bound in (32) is suﬃcient for n (cid:3) 10, so we may assume n = 8 and q (cid:3) 5. If m (cid:2) 3 then (32) is good enough unless (m, q) = (2, 5). In fact, by carefully inspecting [27, Proposition 3.16], we may assume that ((cid:3), η) = (+, −) and with the aid of Magma one checks that fpr(x) (cid:2) 1/620 (maximal if x has Jordan form 1 )). Finally, suppose (n, m) = (8, 4). Here (cid:3) = − and we are free to assume that (J3, J 5 4 , which means that [27, Proposition 3.16] yields fpr(x) (cid:2) 3q−2+2q−4 (cid:2) (q+1)−1 H = N + for q (cid:3) 5. To complete the proof, we may assume r (cid:8)= p. If r = 2 then one checks that the bound in (32) is suﬃcient (recall that we may assume q (cid:3) 5 when n = 8). Now assume r is odd and let i (cid:3) 1 be minimal such that r divides qi − 1. As before, we note that \|CΩ(x)\| is maximal when x is of the form (Λ(cid:3), In−(cid:3)i) (if i is even) or ((Λ, Λ−1)(cid:3), In−2(cid:3)i) (if i is odd) for some (cid:4) (cid:3) 1. Suppose i is even. If m < i then CΩ(x) is the set of η-type m-spaces in CV (x), so \|CΩ(x)\| is maximal when x = (Λ, In−i) and we deduce that fpr(x) < 8 (cid:6) (cid:5) q + 1 q q−mi (cid:2) (qi/2 + 2)−1 as required. Now assume m (cid:3) i (with i even). If i = 2 then one can check that (32) is suﬃcient unless m = q = 2 and n (cid:3) 14. Here η = − (since H is a maximal subgroup of G), r = 3 and we compute (cid:29) \|xG ∩ H\| (cid:2) 2 (cid:30) − n−2(q)\| \|SO n−2(cid:3)(q)\| \|GU(cid:3)−1(q)\|\|SO+ (cid:3) 2q−2n+2(cid:3)+2 + q−4(cid:3) + < 2q2n(cid:3)−3(cid:3)2−(cid:3) − n−2(q)\| \|SO \|GU(cid:3)(q)\|\|SO+ (cid:4) n−2(cid:3)−2(q)\| and 78 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 \|xG\| > (cid:6) (cid:5) 1 2 q q + 1 q2n(cid:3)−3(cid:3)2−(cid:3). The resulting upper bound on fpr(x) is suﬃcient unless (cid:4) = 1. For (cid:4) = 1, we verify the bound fpr(x) (cid:2) 1/4 by working with the precise values for \|xG ∩ H\| and \|xG\|. Now assume m (cid:3) i (cid:3) 4 (we are continuing to assume that i is even). If m < n/2 then (32) is suﬃcient unless (n, m, q, i) = (10, 4, 3, 4). Here r = 5 and x = (Λ, I6) or (Λ2, I2). For x = (Λ, I6) we get \|xG ∩ H\| (cid:2) \|O− 4 (3)\| \|GU1(9)\| + − 6 (3)\| \|SO \|GU1(9)\|\|SO+ 2 (3)\| < 2 · 312 and the bound \|xG\| > 1 8 329 is clearly suﬃcient. The case x = (Λ2, I2) is similar. Finally, suppose n = 2m, so (cid:3) = − and we can slightly strengthen the bound in (32) by replacing the term q−(n−2)/2 by q−n/2 (see [27, Proposition 3.16]). Working with this modiﬁed bound, we can now reduce the problem to the cases where (m, q, i) = (6, 3, 6) or (5, 3, 4), or m = i = 4. In the latter case, r divides q2 + 1 and x = (Λ, I4) since (cid:3) = −. In particular, we calculate fpr(x) = 2 q8(q4 + q2 + 1)(q4 + 1) (cid:2) (q2 + 2)−1 and the result follows. The two other special cases that we need to consider can be handled in a similar fashion. Finally, let us assume i is odd. If m < 2i then we ﬁnd that \|CΩ(x)\| is maximal when x = (Λ, Λ−1, In−2i), which implies that (cid:5) fpr(x) < 8 (cid:6) q + 1 q q−2mi (cid:2) q−i. On the other hand, if m (cid:3) 2i then one can check that the bound in (32) is suﬃcient, noting that r divides (qi − 1)/(q − 1) if i (cid:3) 3. (cid:2) This completes the proof of Proposition 5.26, which in turn completes the proof of Theorem 5.1 and our analysis of subspace actions of classical groups. 6. Product type groups In this section, we complete the proof of Theorem 1 by handling the product-type primitive groups. Here we have G (cid:2) L (cid:5) Sk in its product action on Ω = Γk, where k (cid:3) 2 and L (cid:2) Sym(Γ) is a primitive group of diagonal or almost simple type. In addition, if T denotes the socle of L, then T k is the socle of G and the subgroup of Sk induced by the conjugation action of G on the k factors of T k is transitive. There are two cases to consider. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 79 Proposition 6.1. If L is a diagonal type group, then fpr(x) (cid:2) (r + 1)−1 for all x ∈ G of prime order r. Proof. Write x = (x1, . . . , xk)π ∈ G, where xi ∈ L and π ∈ Sk. If π = 1 then xr all i and at least one xi is nontrivial, whence i = 1 for fpr(x) = (cid:16) i fpr(xi, Γ) (cid:2) (r + 1)−1 by Proposition 2.3. Now assume π (cid:8)= 1, in which case π has cycle-shape (rh, 1k−hr) for some h (cid:3) 1. In this situation, a straightforward computation with the product action shows that \|CΩ(x)\| (cid:2) \|Γ\|k−h(r−1) and thus fpr(x) (cid:2) \|Γ\|−h(r−1) (cid:2) \|Γ\|1−r (cid:2) (r + 1)−1 since \|Γ\| (cid:3) 60. (cid:2) Proposition 6.2. Suppose L (cid:2) Sym(Γ) is almost simple with point stabilizer J and let x = (x1, . . . , xk)π ∈ G be an element of prime order r. Then fpr(x) > (r + 1)−1 only if π = 1 and one of the following holds (up to permutation isomorphism): (i) L = Sn or An acting on (cid:4)-element subsets of {1, . . . , n} with 1 (cid:2) (cid:4) < n/2. (ii) x is conjugate to (x1, 1, . . . , 1) and (L, J, x1) is one of the cases in parts (b)-(d) of Theorem 1(i). Proof. Suppose fpr(x) > (r+1)−1. By arguing as in the proof of the previous proposition, noting that \|Γ\| (cid:3) 5, we see that π = 1 and thus fpr(x) = i fpr(xi, Γ). In addition, we may assume L is neither Sn nor An acting on (cid:4)-sets. If x1 is nontrivial, then either fpr(x1, Γ) (cid:2) (r + 1)−1 or (L, J, x1) is one of the special cases arising in parts (b)-(d) of Theorem 1(i). By Corollary 2, which is an easy consequence of Theorem 1 for almost simple groups (see below), we see that (cid:2) fpr(x1, Γ) · fpr(x2, Γ) (cid:2) (r + 1)−1 if x1 and x2 are both nontrivial. Therefore, we conclude that x is conjugate to (x1, 1, . . . , 1) and the proof is complete. (cid:2) This completes the proof of Theorem 1. Finally, let us prove Corollary 2. Proof of Corollary 2. Let G (cid:2) Sym(Ω) be a ﬁnite primitive group and let x ∈ G be an element of prime order r. By Theorem 1, we may assume that G (cid:2) L (cid:5) Sk acts on 80 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Ω = Γk with the product action, where L (cid:2) Sym(Γ) is almost simple and k (cid:3) 1 (note that the desired conclusion clearly holds if G is an aﬃne group since r + 1 < r2). In addition, we can exclude the special case where L is permutation isomorphic to Sn or An acting on (cid:4)-element subsets of {1, . . . , n}. If k = 1 then G is almost simple and the desired conclusion follows by inspecting the special cases that arise in parts (b)-(d) of Theorem 1(i). Finally, by combining this observation with Proposition 6.2, we conclude that fpr(x) (cid:2) (r + 1)−1/2 when k (cid:3) 2. (cid:2) 7. Minimal index Let G (cid:2) Sym(Ω) be a primitive permutation group of degree m with point stabilizer H. Recall that Ind(G) = min{ind(x) : 1 (cid:8)= x ∈ G} is the minimal index of G, where ind(x) = m − orb(x) = m − 1 \|x\| (cid:9) y∈(cid:4)x(cid:5) \|CΩ(y)\| is the index of x and orb(x) is the number of orbits of x on Ω. If we wish to specify the set Ω, we will write Ind(G, Ω) and ind(x, Ω). Observe that if x ∈ G has order r, then orb(x) (cid:3) m/r and thus ind(x) (cid:2) m(1 − 1/r). In particular, if x is an involution then ind(x) (cid:2) m/2, with equality if and only if x acts ﬁxed point freely on Ω. Consequently, if \|G\| is even then Ind(G) (cid:2) m/2, with equality only if \|H\| is odd (of course, if \|H\| is even then G contains involutions with ﬁxed points). In this ﬁnal section we prove Theorems 6 and 7. We will also establish Theorem 7.4 in the special case where \|G\| is odd. We begin with the following easy lemma. Lemma 7.1. If Ind(G) = ind(x) then x has prime order. Proof. Suppose otherwise, say \|x\| = pq with p a prime and q > 1. Then z = xq has order p and each (cid:13)x(cid:14)-orbit is a union of (cid:13)z(cid:14)-orbits. Since Ind(G) = ind(x), it follows that every (cid:13)x(cid:14)-orbit is a (cid:13)z(cid:14)-orbit and vice versa. But if {α, αz, . . . , αzp−1 } is an orbit of (cid:13)z(cid:14) of length p, then (cid:13)xp(cid:14) is the stabilizer of α in (cid:13)x(cid:14) and we deduce that orb(xp) > orb(x), which is a contradiction. (cid:2) Next we handle two important special cases. Proposition 7.2. Let G = Sn or An acting on (cid:4)-element subsets of {1, . . . , n}, where n (cid:3) 5 and 1 (cid:2) (cid:4) < n/2. Then Ind(G) = (cid:12) (cid:3) (cid:4) n−2 (cid:23)(cid:3) (cid:3)−1 n (cid:3) 1 2 (cid:4) − (cid:4) (cid:3) n−4 (cid:3) (cid:3) (cid:4) n−4 (cid:3)−2 (cid:3) − − 2 (cid:4)(cid:24) n−4 (cid:3)−4 if G = Sn if G = An. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 81 Moreover, Ind(G) = ind(x) if and only if (i) G = Sn and x is a transposition; (ii) G = An and x is a double transposition; or (iii) G = An, (cid:4) = 1 and x is a 3-cycle. Proof. If 1 (cid:8)= y ∈ G has odd order, then ind(y) is minimal when y is a 3-cycle, in which case ind(y) = (cid:5)(cid:5) (cid:5) (cid:6) n (cid:4) − n − 3 (cid:4) (cid:6) (cid:5) − 2 3 (cid:6)(cid:6) n − 3 (cid:4) − 3 . Now let x ∈ G be an involution and observe that ind(x) is minimal when x is a trans- position (for G = Sn) or a double transposition (for G = An). If x is a transposition, then ind(x) = (cid:6) (cid:5) n − 2 (cid:4) − 1 < ind(y). Similarly, if x is a double transposition then ind(x) = (cid:5)(cid:5) (cid:6) (cid:6) (cid:5) − n (cid:4) n − 4 (cid:4) 1 2 (cid:5) n − 4 (cid:4) − 2 (cid:6) (cid:5) − n − 4 (cid:4) − 4 − 2 (cid:6)(cid:6) and this is strictly less than ind(y) when (cid:4) (cid:3) 2. However, if (cid:4) = 1 then we calculate that ind(x) = ind(y) = 2. The result follows. (cid:2) Proposition 7.3. Let G (cid:2) Sym(Ω) be an almost simple primitive classical group of degree m in a subspace action with socle G0 and point stabilizer H, where (G, H) is one of the cases appearing in Table 6. Then the following hold: (i) Ind(G) = ind(x) only if \|x\| ∈ {2, 3}. (ii) Ind(G) = ind(x) for some involution x ∈ G. (iii) Ind(G) = ind(x) for some element x ∈ G of order 3 if and only if G = L2(8):3 and H = P1, in which case m = 9 and Ind(G) = 4. (iv) Ind(G) < m/4 if and only if (G, H, m, Ind(G)) is one of the cases in Table 8. In each of these cases, Ind(G) (cid:3) 3m/14. Proof. First observe that \|H\| is even (for example, see [35, Theorem 2]) and thus Ind(G) < m/2 as noted above. Let x ∈ G be an element such that Ind(G) = ind(x), so x has prime order r by Lemma 7.1. Seeking a contradiction, suppose r (cid:3) 5. If fpr(y) (cid:2) r−1 for every element y ∈ G of order r, then ind(x) (cid:3) (cid:6) 2 (cid:5) 1 − 1 r m > m 2 > Ind(G) 82 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 Table 8 The subspace actions in part (iv) of Proposition 7.3. G H m Ind(G) Conditions U4(2).2 Spn(2) O− n (2) O+ n (2) n (2) P2 O− P1 N1 27 2n/2−1(2n/2 − 1) (2n/2−1 − 1)(2n/2 + 1) 2n/2−1(2n/2 − 1) 6 2n/2−2(2n/2−1 − 1) 2n/2−2(2n/2−1 − 1) 2n/2−2(2n/2−1 − 1) n (cid:2) 6 n (cid:2) 8 n (cid:2) 8 and we have reached a contradiction. Therefore, we may assume fpr(y) > r−1 for some y ∈ G of order r. Then Corollary 3 implies that G0 = L2(q), H = P1 and r = q − 1 (cid:3) 7 is a Mersenne prime. Here m = q + 1 and \|CΩ(x)\| = 2, so ind(x) = q − 2 > m/2 and once again this is a contradiction. This proves part (i) and so for the remainder we may assume r ∈ {2, 3}. For r ∈ {2, 3}, set fr = max{fpr(x) : \|x\| = r}, mr = min{ind(x) : \|x\| = r} and note that Ind(G) = min{m2, m3}. If f3 (cid:2) 1/4 then ind(x) (cid:3) m/2 > Ind(G) for every x ∈ G of order 3, in which case Ind(G) = m2 (so (ii) holds) and (G, H) does not arise as a special case in part (iii). In addition, if f2 (cid:2) 1/3 then ind(x) (cid:3) m/3 for every involution x and thus (G, H) does not appear in part (iv). Therefore, to complete the proof of the proposition we may assume f2 > 1/3 or f3 > 1/4, in which case (G, H, x) is one of the special cases in Table 6 with \|x\| = 2 or 3. We now inspect each of these cases in turn, computing m2, and also m3 if f3 > 1/4. Case 1. G0 = Ln(q). Here H = P1, m = (qn − 1)/(q − 1) and we may assume q ∈ {2, 3, 4} or (n, q) = (2, 8) since we are only interested in the cases where f2 > 1/3 or f3 > 1/4. If (n, q) = (2, 8) then G = L2(8):3, m = 9 and Ind(G) = 4, with ind(x) = 4 if and only if x is an involution or a ﬁeld automorphism of order 3 (in particular, this is the special case recorded in part (iii)). If q = 2 then f3 (cid:2) 1/4 and Ind(G) = m2 = ind(x) with x = (J2, J n−2 ). Here fpr(x) is recorded in Table 6 and we compute Ind(G) = 2n−2 > m/4. Similarly, if q = 3 then m3 = 2.3n−2 and 1 (cid:12) m2 = 2 (3n−1 − 1) 1 2(3n−2 − 1) if (−In−1, I1) ∈ G otherwise, whence Ind(G) = m2 > m/4. Finally, suppose q = 4. Here f3 > 1/4 if and only if G contains an element of order 3 of the form x = (ω, In−1) (modulo scalars), in which case m3 = ind(x) = 2(4n−1 − 1)/3. If x ∈ G is an involution, then ind(x) is minimal when x = (J2, J n−2 ) is a transvection and we compute m2 = 22n−3. We conclude that Ind(G) = m2 > m/4. Case 2. G0 = Un(q), n (cid:3) 3. 1 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 83 First assume H = P1, so n (cid:3) 5 is odd, q = 2 and m = (2n + 1)(2n−1 − 1)/3. Here f3 > 1/4 if and only if G contains an element x = (ω, In−1) of order 3, in which case m3 = ind(x) = 2n−1(2n−1 − 1)/3. If x ∈ G is an involution, then ind(x) is minimal when x = (J2, J n−2 ). Then by inspecting the proof of Lemma 5.12, we calculate that \|CΩ(x)\| = 4(2n−2 + 1)(2n−3 − 1)/3 + 1 and thus m2 = ind(x) = 22n−4. In particular, Ind(G) = m2 > m/4. 1 Now suppose H = P2, so n = 4 and q ∈ {2, 3}. Here it is straightforward to check that Ind(G) = m2 < m3. More precisely, if q = 2 then m = 27 and either G = G0 and Ind(G) = 10 > m/4, or G = G0.2 and Ind(G) = 6 < m/4 (so the latter case is recorded in part (iv)). Similarly, if q = 3 then m = 112 and Ind(G) (cid:3) 36 > m/4. Finally suppose H = N1, so n (cid:3) 4 is even, q = 2 and m = 2n−1(2n−1)/3. If r = 3 then ind(x) is minimal when x = (ω, In−1) and we compute ind(x) = 2(22n−3 − 2n−2 − 1)/3. Similarly, if r = 2 then ind(x) is minimal when x = (J2, J n−2 ) and by inspecting the proof of Lemma 5.14 we deduce that ind(x) = 22n−4. Therefore, Ind(G) = 22n−4 > m/4 and Ind(G) = ind(x) if and only if x = (J2, J n−2 Case 3. G0 = PSpn(q), n (cid:3) 4. ). 1 1 1 First assume H = P1, so m = (qn − 1)/(q − 1) and we may assume q ∈ {2, 3}. If q = 2 then ind(x) is minimal when x = (J2, J n−2 ) and we compute ind(x) = 2n−2. Therefore, Ind(G) = 2n−2 > m/4, with Ind(G) = ind(x) if and only if x = (J2, J n−2 ). Now assume q = 3. If r = 3, then ind(x) is minimal when x = (J2, J n−2 ) and we calculate ind(x) = 2.3n−2. For r = 2, we ﬁnd that ind(x) is minimal when x = (−I2, In−2). Here the proof of Lemma 5.17 gives \|CΩ(x)\| = (3n−2 − 1)/2 + 4 and we deduce that ind(x) = 2(3n−2 − 1). Therefore, Ind(G) = 2(3n−2 − 1) > m/4 and Ind(G) = ind(x) only if x is an involution. 1 1 Next suppose n (cid:3) 6, q = 2 and H = O(cid:2) n(2), so m = 2n/2−1(2n/2 + (cid:3)). Here f3 > 1/4 if and only if (cid:3) = −, in which case m3 = 2n/2−1(2n/2−1 − 1) (with ind(x) minimal if x = (Λ, In−2)). If r = 2 then ind(x) is minimal when x = (J2, J n−2 ) and we compute ind(x) = 2n/2−2(2n/2−1 + (cid:3)). Therefore, Ind(G) = 2n/2−2(2n/2−1 + (cid:3)), which is less than m/4 if and only if (cid:3) = − (note that in this situation we have Ind(G) (cid:3) 3m/14, with equality if n = 6). We also deduce that Ind(G) = ind(x) only if x is an involution. 1 Case 4. G0 = Ωn(q), n (cid:3) 7 odd, q odd. Here q = 3, H = P1 or N − 1 , and f3 (cid:2) 1/4, so we may assume r = 2. First assume H = P1, so m = (3n−1 − 1)/2. If G = SOn(3) then ind(x) is minimal when x = (−In−1, I1)+, in which case ind(x) = 3(n−3)/2(3(n−1)/2 − 1)/2. Similarly, if G = Ωn(3) and (−In−1, I1)+ /∈ G, then ind(x) is minimal when x = (−In−1, I1)− and we compute ind(x) = 3(n−3)/2(3(n−1)/2 + 1)/2. In conclusion, if H = P1 then Ind(G) = 3(n−3)/2(3(n−1)/2 − δ)/2, where δ = 1 if (−In−1, I1)+ ∈ G, otherwise δ = −1. In particular, Ind(G) > m/4 and Ind(G) = ind(x) only if x is an involution. Now assume H = N − 1 , so m = 3(n−1)/2(3(n−1)/2 − 1). As in the previous case, we only need to consider involutions of type (−In−1, I1)(cid:2). If G = SOn(3), then ind(x) is minimal 84 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 when x = (−In−1, I1)− and we get ind(x) = (3n−2 − 2.3(n−3)/2 − 1)/2. And if G = Ωn(3) does not contain (−In−1, I1)−, then ind(x) is minimal when x = (−In−1, I1)+ and we calculate that fpr(x) = 1/3, which yields ind(x) = 3(n−3)/2(3(n−1)/2 − 1)/2. Therefore, (cid:12) Ind(G) = (3n−2 − 2.3(n−3)/2 − 1)/2 if (−In−1, I1)− ∈ G 3(n−3)/2(3(n−1)/2 − 1)/2 otherwise. Once again we conclude that Ind(G) > m/4 and Ind(G) = ind(x) only if x is an involu- tion. Case 5. G0 = PΩ(cid:2) n(q), n (cid:3) 8 even. First assume H = P1, so m = (qn/2−1 + (cid:3))(qn/2 − (cid:3))/(q − 1) and q ∈ {2, 3}. Suppose q = 3, in which case (cid:3) = −, f3 (cid:2) 1/4 and we may assume G contains a reﬂection x = (−In−1, I1) (otherwise f2 (cid:2) 1/3). Then we compute m2 = ind(x) = 1 2 3n/2−1(3n/2−1 − 1) > m 4 . n(2) then ind(x) is minimal when x = (J2, J n−2 Now assume q = 2. If r = 3, then we may assume (cid:3) = −, in which case ind(x) is minimal when x = (Λ, In−2). Here we compute ind(x) = 2n/2−1(2n/2−1 − 1). Now assume r = 2. If G = O(cid:2) ) is a b1-type involution and we calculate ind(x) = 2n/2−2(2n/2−1 + (cid:3)). Therefore, if G = O(cid:2) n(2) then Ind(G) = 2n/2−2(2n/2−1 +(cid:3)), which is less than m/4 if and only if (cid:3) = − (note that if (cid:3) = − then it is easy to check that Ind(G) > 3m/14). Now suppose G = Ω(cid:2) n(2). Here ind(x) is minimal when x = a2 or c2. If x = a2 then \|CΩ(x)\| is given in the proof of Lemma 5.27 and we compute ind(x) = 3.2n−4. Similarly, if x = c2 then ind(x) = 2n/2−2(3.2n/2−2 +(cid:3)). Since 3.2n−4 < 2n/2−1(2n/2−1 − 1), we conclude that if G = Ω(cid:2) 1 n(2) then (cid:12) Ind(G) = 3.2n−4 2n/2−2(3.2n/2−2 − 1) if (cid:3) = + if (cid:3) = − is greater than m/4 and Ind(G) = ind(x) only if x is an involution. To complete the analysis of orthogonal groups, we may assume H = N1 and q ∈ {2, 3}. Note that m = qn/2−1(qn/2 − (cid:3))/d, where d = (2, q − 1). First assume q = 2. If r = 3 then we may assume (cid:3) = +, x = (Λ, In−2) and we compute ind(x) = 2n/2−1(2n/2−1 − 1). Now suppose r = 2. If G = O(cid:2) n(2) then ind(x) is minimal when x = b1, in which case ind(x) = 2n/2−2(2n/2−1 − (cid:3)). So in this situation, we have Ind(G) = 2n/2−2(2n/2−1 −(cid:3)), which is less than m/4 if and only if (cid:3) = + (here one checks that Ind(G) (cid:3) 3m/14). Now assume G = Ω(cid:2) n(2). Here ind(x) is minimal when x = a2 or c2. We calculated fpr(x) in the proof of Lemma 5.31 and we deduce that ind(x) = 3.2n−4 if x = a2 and ind(x) = 2n/2−2(3.2n/2−2 − (cid:3)) if x = c2. Now 3.2n−4 < 2n/2−1(2n/2−1 − 1) and thus T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 85 (cid:12) Ind(G) = 3.2n−4 2n/2−2(3.2n/2−2 − 1) if (cid:3) = − if (cid:3) = + and Ind(G) = ind(x) only if x is an involution. In addition, Ind(G) > m/4. Finally, suppose q = 3. We may assume r = 2 since f3 (cid:2) 1/4. First assume (cid:3) = + and note that we may assume x = (−In−1, I1)(cid:5) ∈ G (otherwise f2 (cid:2) 1/3). Then Ind(G) = ind(x) = 3n/2−1(3n/2−1 − 1)/2. Similarly, if (cid:3) = − then we may assume G contains x = (−In−1, I1)(cid:4) and we deduce that Ind(G) = (3n−2 − 1)/2. In both cases, Ind(G) > m/4 and the result follows. (cid:2) With these two special cases in hand, we are now ready to prove Theorem 6. Proof of Theorem 6. Assume \|G\| is even and recall that Ind(G) (cid:2) m/2, with equality only if \|H\| is odd. Let x ∈ G be an element of order r such that Ind(G) = ind(x). By Lemma 7.1, r is a prime. Seeking a contradiction, suppose r (cid:3) 5. By arguing as in the proof of Proposition 7.3, it follows that fpr(y) > r−1 for some y ∈ G of order r, so by applying Theorem 1 we deduce that (a) G is almost simple; or (b) G (cid:2) L (cid:5) Sk is a product type primitive group with its product action on Ω = Γk, where k (cid:3) 2 and L (cid:2) Sym(Γ) is almost simple. If G is almost simple then the possibilities for (G, H) are described in Corollary 3 and the result follows via Propositions 7.2 and 7.3. Now assume (b) holds and let y = (y1, . . . , yk)π ∈ G be an element of prime order r (cid:3) 5 with fpr(y) > r−1. Then Theorem 1 (also see Remark 1(c)) implies that π = 1 and either L = Sn or An acting on (cid:4)-element subsets of {1, . . . , n}, or y = (y1, 1, . . . , 1) up to conjugacy. In the latter case, we have fpr(y) = fpr(y1, Γ), ind(y) = \|Γ\|k−1 · ind(y1, Γ) and Ind(L, Γ) < ind(y1, Γ) by the result for almost simple groups handled in case (a). Similarly, if L = Sn or An acting on (cid:4)-sets, then Proposition 7.2 implies that Ind(G) = ind(x) only if x has order 2 or 3. We have now proved that Ind(G) = ind(x) only if \|x\| ∈ {2, 3}. To complete the proof of Theorem 6, let us assume we have equality for some element x of order 3. If fpr(x) (cid:2) 1/4 then ind(x) (cid:3) m/2, so we must have fpr(x) = 1/4, Ind(G) = m/2 and every involution in G acts ﬁxed point freely on Ω (otherwise there would be an involution y ∈ G with ind(y) < m/2). In particular, this forces \|H\| to be odd and we have Ind(G) = ind(y) for every involution y ∈ G. So to complete the argument, we may assume fpr(x) > 1/4, which implies that (G, H, x) is one of the cases arising in Theorem 1. 86 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 If G is almost simple then by applying Propositions 7.2 and 7.3 we deduce that one of the following holds (up to permutation isomorphism): (i) G = Am in its natural action of degree m and Ind(G) = ind(x) = 2 with x a 3-cycle. (ii) G = L2(8):3, H = P1, m = 9 and Ind(G) = ind(x) = 4 with x a ﬁeld automorphism of order 3. Next assume G = V :H is an aﬃne group with socle (Cp)d and point stabilizer H (cid:2) GLd(3) as in part (ii) of Theorem 1. Here p = 3, x is conjugate to a transvection in H and fpr(x) = 1/3, so ind(x) = 4m/9. Recall that we are assuming \|G\| is even, so H contains involutions. We claim that H contains an involution y with dim CV (y) (cid:3) d − 2. In particular, if H contains a reﬂection y = (−I1, Id−1) then fpr(y) = 1/3 and ind(y) = m/3 < ind(x), whence Ind(G) = m/3 and G does not contain an element of order 3 with Ind(G) = ind(x). On the other hand, if H does not contain such an element, then ind(y) is minimal when y ∈ H is an involution with dim CV (y) = d − 2, in which case ind(y) = Ind(G) = 4m/9. This is the case recorded in part (ii)(b) of Theorem 6. Therefore, it remains to justify the claim. To do this, we apply a theorem of McLaugh- lin [42]. Recall that we are assuming H contains a transvection, so we can consider the normal subgroup H0 generated by the transvections in H. Since H acts irreducibly on V = (F3)d, it follows that H0 is semisimple, preserving a direct sum decomposition V = V1 ⊕ · · · ⊕ Vt. Moreover, H0 acts on V as a direct product H1 × · · · × Ht, where each Hi (cid:2) GL(Vi) is either SL(Vi) or Sp(Vi). Therefore, H0 contains involutions of the form (−I2, Id−2) and the claim follows. Finally, let us turn to the product type groups in part (iii) of Theorem 1. Here G (cid:2) L (cid:5) Sk acts on Ω = Γk with its product action, where k (cid:3) 2 and L (cid:2) Sym(Γ) is one of the almost simple primitive groups in part (i) of Theorem 1. Recall that we are assuming there exists an element x ∈ G of order 3 with fpr(x) > 1/4 and we note that x is of the form (x1, . . . , xk) ∈ Lk (see Remark 1(c)). Therefore, up to permutation isomorphism, L is either Sn or An acting on (cid:4)-element subsets of {1, . . . , n}, or L is a classical group in a subspace action as in Table 6. First assume L = Sn or An acting on (cid:4)-sets, where 1 (cid:2) (cid:4) < n/2. If (cid:4) (cid:3) 2 then Proposition 7.2 implies that Ind(G) = ind(x) only if x is an involution, so we may assume (cid:4) = 1. Now, if x ∈ G has order 3, then ind(x) is minimal when x = (x1, 1, . . . , 1) ∈ (An)k and x1 ∈ An is a 3-cycle, in which case we compute ind(x) = 2m/n. If y ∈ G ∩ (Sn)k is of the form (y1, . . . , yt, 1, . . . , 1) and each yi is a transposition, then ind(y) = (cid:29) (cid:5) 1 − m 2 1 − 2 n (cid:30) (cid:6) t and we deduce that ind(y) < 2m/n if and only if t ∈ {1, 2}, or if t = 3 and n = 5. This gives the conclusion recorded in part (d) of Theorem 6(ii). T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 87 Finally, suppose L is a classical group in a subspace action. By applying part (iii) of Proposition 7.3 we deduce that G = L (cid:5) P is the only possibility, where L = L2(8):3, Γ is the set of 1-dimensional subspaces of the natural module for L2(8) and P (cid:2) Sk is transitive. Here m = 9k, Ind(G) = 4.9k−1 and Ind(G) = ind(x) if and only if x is conjugate to (x1, 1, . . . , 1), where x1 ∈ L is either an involution or a ﬁeld automorphism of L2(8) of order 3. (cid:2) Next we prove Theorem 7. Proof of Theorem 7. As before, \|G\| is even and thus Ind(G) (cid:2) m/2, with equality only if \|H\| is odd. By Theorem 6, there exists an involution x ∈ G with Ind(G) = ind(x). If fpr(x) (cid:2) 1/2, then ind(x) (cid:3) m/4, so we may assume fpr(x) > 1/2 and thus (G, H, x) is one of the special cases appearing in the statement of Theorem 1. If G is almost simple, then by applying Corollary 3 we reduce to the case where G is a classical group in a subspace action and the result follows via Proposition 7.3(iv). Finally, we may assume G (cid:2) L (cid:5) Sk is a product type primitive group with its product action on Ω = Γk, where k (cid:3) 2 and L (cid:2) Sym(Γ) is an almost simple primitive group with point stabilizer J. In addition, we may assume L is not Sn or An acting on (cid:4)-element subsets of {1, . . . , n} (since this is covered by case (ii) in Theorem 7). Therefore, our involution x with fpr(x) > 1/2 must be conjugate to (x1, 1, . . . , 1) in Lk and thus It follows that Ind(G) = ind(x) = \|Γ\|k−1 · ind(x1, Γ). Ind(L, Γ) = ind(x1, Γ) < \|Γ\| 1 4 and thus (L, J, \|Γ\|, Ind(L, Γ)) is one of the cases in Table 8. Let T denote the socle of L. If L = Spn(2) with n (cid:3) 6 then L = T and thus G = L (cid:5) P for some transitive group P (cid:2) Sk. In the three remaining cases in Table 8, we have \|L : T \| = 2, but in each case every involution y ∈ L with Ind(L, Γ) = ind(y, Γ) is contained in L \ T (for T = U4(2), y is an involutory graph automorphism with CT (y) = Sp4(2), while y is a b1-type involution for the cases with T = Ω(cid:2) n(2)). Therefore, G must contain Lk and thus G = L (cid:5) P for some transitive group P (cid:2) Sk. (cid:2) Finally, let us consider the minimal index of a primitive group G (cid:2) Sym(Ω) of odd order. Here G is solvable by the Feit-Thompson theorem, so G = V :H is an aﬃne group with socle V = (Cp)d and point stabilizer H (cid:2) GLd(p) for some odd prime p and positive integer d. Theorem 7.4. Let G = V :H be a primitive permutation group of odd order with socle V = (Cp)d and point stabilizer H. Then 88 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (cid:12) min m (cid:5) 1 − 3 2r + 1 (cid:6) (cid:5) , m 1 − 1 p (cid:13) (cid:6) 2 (cid:2) Ind(G) (cid:2) m (cid:6) (cid:5) 1 − 1 r where r is the smallest prime divisor of \|G\|. Proof. As previously noted, if x ∈ G has order r, then orb(x) (cid:3) m/r and thus Ind(G) (cid:2) ind(x) (cid:2) m (cid:5) 1 − 1 r (cid:6) . Next let x ∈ G be an element of prime order s. We may assume fpr(x) > 0 (otherwise ind(x) = m(1 − 1/s) (cid:3) m(1 − 1/r)), so by replacing x by a suitable conjugate we may assume x ∈ H. Then fpr(x) = pe−d, where e = dim CV (x). Suppose s = p. If e = d − 1 then x is a transvection, fpr(x) = p−1 and we compute ind(x) = m (cid:5) (cid:6) 2 . 1 − 1 p On the other hand, if e < d − 1 then fpr(x) (cid:2) p−2 and thus ind(x) (cid:3) m (cid:5) 1 − p2 + p − 1 p3 (cid:6) (cid:5) 1 − 1 p (cid:6) 2 . > m Now suppose s (cid:8)= p. Then s (cid:2) (pd−e − 1)/2 since p is odd and thus fpr(x) (cid:2) 1/(2s + 1). In turn, this implies that ind(x) (cid:3) m (cid:5) (cid:5) 1 − 1 s 1 + s − 1 2s + 1 (cid:6)(cid:6) (cid:5) 1 − 3 2s + 1 (cid:6) (cid:5) (cid:3) m 1 − 3 2r + 1 (cid:6) = m and the result follows. (cid:2) Acknowledgments We thank an anonymous referee for their careful reading of the paper and helpful suggestions. Burness thanks the Department of Mathematics at the University of Padua for their generous hospitality during a research visit in autumn 2021. Guralnick was partially supported by the NSF grant DMS-1901595 and a Simons Foundation Fellowship 609771. References [1] M. Aschbacher, On the maximal subgroups of the ﬁnite classical groups, Invent. Math. 76 (1984) 469–514. [2] M. Aschbacher, G.M. Seitz, Involutions in Chevalley groups over ﬁelds of even order, Nagoya Math. J. 63 (1976) 1–91. T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 89 [3] L. Babai, On the order of uniprimitive permutation groups, Ann. Math. 113 (1981) 553–568. [4] M. Bhargava, Galois groups of random integer polynomials and van der Waerden’s Conjecture, preprint, arXiv :2111 .06507, 2021. [5] A. Bochert, Über die Zahl der verschiedenen Werthe, die eine Function gegebener Buchstaben durch Vertauschung derselben erlangen kann, Math. Ann. 33 (1889) 584–590. [6] W. Bosma, J. Cannon, C. Playoust, The Magma algebra system I: the user language, J. Symb. Comput. 24 (1997) 235–265. [7] J.N. Bray, D.F. Holt, C.M. Roney-Dougal, The Maximal Subgroups of the Low-Dimensional Finite Classical Groups, London Math. Soc. Lecture Notes Series, vol. 407, Cambridge University Press, 2013. [8] T. Breuer, The GAP character table library, version 1.3.3, GAP package, http://www .math .rwth - aachen .de /~Thomas .Breuer /ctbllib, 2022. [9] T. Breuer, R.M. Guralnick, W.M. Kantor, Probabilistic generation of ﬁnite simple groups, II, J. Algebra 320 (2008) 443–494. (2021) 1755–1807. [10] T.C. Burness, Base sizes for primitive groups with soluble stabilisers, Algebra Number Theory 15 [11] T.C. Burness, Simple groups, ﬁxed point ratios and applications, in: Local Representation Theory and Simple Groups, in: EMS Ser. Lect. Math., Eur. Math. Soc., Zürich, 2018, pp. 267–322. [12] T.C. Burness, Fixed point ratios in actions of ﬁnite classical groups I, J. Algebra 309 (2007) 69–79. [13] T.C. Burness, Fixed point ratios in actions in ﬁnite classical groups II, J. Algebra 309 (2007) 80–138. [14] T.C. Burness, Fixed point ratios in actions of ﬁnite classical groups III, J. Algebra 314 (2007) 693–748. [15] T.C. Burness, Fixed point ratios in actions of ﬁnite classical groups IV, J. Algebra 314 (2007) 749–788. [16] T.C. Burness, M. Giudici, Classical Groups, Derangements and Primes, Australian Mathematical Society Lecture Series, vol. 25, Cambridge University Press, Cambridge, 2016. [17] T.C. Burness, R.M. Guralnick, S. Harper, The spread of a ﬁnite group, Ann. Math. 193 (2021) 619–687. [18] T.C. Burness, R.M. Guralnick, A. Moretó, G. Navarro, On the commuting probability of p-elements in a ﬁnite group, Algebra Number Theory (2022), in press. [19] T.C. Burness, S. Harper, Finite groups, 2-generation and the uniform domination number, Isr. J. [20] T.C. Burness, E.A. O’Brien, R.A. Wilson, Base sizes for sporadic simple groups, Isr. J. Math. 177 [21] J.H. Conway, R.T. Curtis, S.P. Norton, R.A. Parker, R.A. Wilson, Atlas of Finite Groups, Oxford Math. 239 (2020) 271–367. (2010) 307–333. University Press, 1985. [22] J.B. Fawcett, The base size of a primitive diagonal group, J. Algebra 375 (2013) 302–321. [23] J.B. Fawcett, Bases of twisted wreath products, J. Algebra 607 (2022) 247–271, part A. [24] D. Frohardt, K. Magaard, Composition factors of monodromy groups, Ann. Math. 154 (2001) [25] D. Frohardt, K. Magaard, Grassmannian ﬁxed point ratios, Geom. Dedic. 82 (2000) 21–104. [26] The GAP Group, GAP – groups, algorithms, and programming, version 4.11.0, http://www .gap - [27] R.M. Guralnick, W.M. Kantor, Probabilistic generation of ﬁnite simple groups, J. Algebra 234 [28] R.M. Guralnick, K. Magaard, On the minimal degree of a primitive permutation group, J. Algebra [29] C. Jordan, Théorèmes sur les groupes primitifs, Math. Pures Appl. 16 (1871) 383–408. [30] P.B. Kleidman, The maximal subgroups of the ﬁnite 8-dimensional orthogonal groups PΩ+ 8 (q) and of their automorphism groups, J. Algebra 110 (1987) 173–242. [31] P.B. Kleidman, M.W. Liebeck, The Subgroup Structure of the Finite Classical Groups, London Math. Soc. Lecture Note Series, vol. 129, Cambridge University Press, 1990. [32] P.B. Kleidman, R.A. Wilson, The maximal subgroups of E6(2) and Aut(E6(2)), Proc. Lond. Math. Soc. 60 (1990) 266–294. [33] R. Lawther, M.W. Liebeck, G.M. Seitz, Fixed point ratios in actions of ﬁnite exceptional groups of Lie type, Pac. J. Math. 205 (2002) 393–464. [34] M.W. Liebeck, J. Saxl, Minimal degrees of primitive permutation groups, with an application to monodromy groups of covers of Riemann surfaces, Proc. Lond. Math. Soc. 63 (1991) 266–314. 327–345. system .org, 2020. (2000) 743–792. 207 (1998) 127–145. 90 T.C. Burness, R.M. Guralnick / Advances in Mathematics 411 (2022) 108778 (1991) 2777–2786. 12 (1999) 497–520. 585–599. [35] M.W. Liebeck, J. Saxl, On point stabilizers in primitive permutation groups, Commun. Algebra 19 [36] M.W. Liebeck, J. Saxl, G.M. Seitz, Subgroups of maximal rank in ﬁnite exceptional groups of Lie type, Proc. Lond. Math. Soc. 65 (1992) 297–325. [37] M.W. Liebeck, A. Shalev, Simple groups, permutation groups, and probability, J. Am. Math. Soc. [38] G. Malle, On the distribution of Galois groups, J. Number Theory 92 (2002) 315–329. [39] G. Malle, On the distribution of Galois groups, II, Exp. Math. 13 (2004) 129–135. [40] W.A. Manning, The degree and class of multiply transitive groups, Trans. Am. Math. Soc. 35 (1933) [41] A. Maróti, On the orders of primitive groups, J. Algebra 258 (2002) 631–640. [42] J. McLaughlin, Some groups generated by transvections, Arch. Math. 18 (1967) 364–368. [43] W.M. Potter, Nonsolvable groups with an automorphism inverting many elements, Arch. Math. (Basel) 50 (1988) 292–299. [44] R.A. Wilson, Maximal subgroups of sporadic groups, in: Finite Simple Groups: Thirty Years of the Atlas and Beyond, in: Contemp. Math., vol. 694, Amer. Math. Soc., Providence, RI, 2017, pp. 57–72.
10.1080_15476286.2023.2231280	RNA BIOLOGY 2023, VOL. 20, NO. 1, 444–456 https://doi.org/10.1080/15476286.2023.2231280 RESEARCH PAPER SARS-CoV-2 Nsp1 mediated mRNA degradation requires mRNA interaction with the ribosome Soraya I. Shehata a and Roy Parker b aDepartment of Molecular, Cellular, and Developmental Biology, University of Colorado Boulder, Boulder, CO, USA; Medical Scientist Training Program, University of Colorado Anschutz Medical Campus, Aurora, CO, USA; bDepartment of Biochemistry, University of Colorado Boulder, Boulder, CO, USA; Howard Hughes Medical Institute, University of Colorado Boulder, Boulder, CO, USA ABSTRACT Nsp1 is a SARS-CoV-2 host shutoff factor that both represses cellular translation and promotes host RNA decay. However, it is unclear how these two activities are connected and interact with normal translation processes. Here, we performed mutational analyses of Nsp1, and these revealed that both the N and C terminal domains of Nsp1 are important for translational repression. Furthermore, we demonstrate that specific residues in the N terminal domain are required for cellular RNA degradation but not bulk translation shutoff of host mRNAs, thereby separating RNA degradation from translation repression. We also present evidence that Nsp1 mediated RNA degradation requires engagement of the ribosome with mRNA. First, we observe that cytosolic lncRNAs, which are not translated, escape Nsp1 mediated degradation. Second, inhibition of translation elongation with emetine does not prevent Nsp1 mediated degradation, while blocking translation initiation before 48S ribosome loading reduces mRNA degrada- tion. Taken together, we suggest that Nsp1 represses translation and promotes mRNA degradation only after ribosome engagement with the mRNA. This raises the possibility that Nsp1 may trigger RNA degradation through pathways that recognize stalled ribosomes. ARTICLE HISTORY Revised 13 June 2023 Accepted 19 June 2023 KEYWORDS SARS-CoV-2; Nsp1; RNA decay; translation; host shutoff; stress granule Introduction SARS-CoV-2 (SARS2) is the origin of the COVID-19 pan- demic, which has killed nearly 7 million people globally to date [1]. SARS-CoV-2 is the third β-coronavirus (βCoV) in two decades to cause a respiratory disease epidemic and the first to progress to a global pandemic, and it is unlikely to be the last novel pathogenic βCoV to pose a significant threat to public health [2]. COVID-19 is also expected to progress to endemicity [3]. Therefore, it is critical to under- stand common virulence factors of coronaviruses with the goal to developing therapies that could be used not only against SARS2 infection, but potentially against other cor- onavirus infections as well. Many βCoVs restrict host cell gene expression (called host shutoff) in order to promote viral gene expression, free up interferon cellular translation machinery, and slow the response [4–7]. In both SARS and SARS2, host shutoff is largely mediated by nonstructural protein 1 (Nsp1). Nsp1 is a 20kD protein translated as part of the ORF1a/b polyprotein and is one of the first proteins to be expressed after viral entry into the cell [8,9]. It has three domains: an alpha helical C-terminal domain (CTD), a flexible linker, and a globular N-terminal domain (NTD). Nsp1 is highly conserved between SARS and SARS2 and much of our knowledge of Nsp1 derives from work studying SARS. Nsp1 restricts host gene expression in two ways: by arrest- ing host translation and by causing the degradation of host RNAs [4,10–14]. The CTD of Nsp1 binds the 40S ribosome and blocks the mRNA entry channel, which is thought to prevent the proper interaction of the 40S ribosome with the mRNA [15–17]. Mutations in the CTD disrupt Nsp1-40S binding and restore cellular translation [13,16]. The NTD has been implicated in mediating viral mRNA escape from translation repression through interactions with the first stem loop in the viral 5’ UTR [18–21]. How Nsp1 expression confers rapid mRNA degradation is still unknown. It has not been shown to have ribonuclease activity in vitro and bears no similarity to any known RNases [22,23]. Knockdown of the 5’-3’ exonuclease Xrn1 only par- tially blocks Nsp1-induced mRNA degradation for SARS, which suggests that it is not the primary nuclease responsible for mRNA decay [24]. The link between translation repression and mRNA decay is also not well understood. mRNA decay appears to be a separate function of Nsp1 that requires trans- lation repression, but the Nsp1-triggered nuclease that degrades the mRNAs has not been identified [11,12,19,25]. Here, we developed a system for single-cell analyses of Nsp1-mediated translation repression and endogenous RNA decay. In contrast to analyses using transfected reporter mRNAs, this approach allows the examination of host mRNAs in individual cells expressing Nsp1. We used this CONTACT Roy Parker University of Colorado Boulder, Boulder, CO, USA [email protected] Department of Biochemistry, University of Colorado Boulder, CO, USA; Howard Hughes Medical Institute, Supplemental data for this article can be accessed online at https://doi.org/10.1080/15476286.2023.2231280 © 2023 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent. repression and that RNA degradation approach to study the relative contributions of the N- and C- terminal domains on translation repression and RNA decay. We found that both the NTD and CTD are necessary for translation is a distinct function of Nsp1, which can be separated from translation repression by specific mutations. Moreover, we present several observations that Nsp1 mediated RNA degra- dation requires mRNAs to be engaged with a ribosome. Specifically, we observe that cytosolic lncRNAs, which are not translated, escape Nsp1 mediated degradation. In addi- tion, compounds blocking ribosome attachment to mRNAs limit Nsp1 mediated degradation, while blocking ribosomes during translation elongation still allows mRNA degradation by Nsp1. These findings suggest that Nsp1 represses transla- tion and promotes mRNA degradation only after ribosome engagement with the mRNA. This raises the possibility that Nsp1 may trigger RNA degradation pathways that recognize stalled ribosomes. Results Nsp1 degrades mRNAs and represses bulk translation Previous studies showed that Nsp1 encoded by SARS-CoV-1 (SARS) or SARS-CoV-2 (SARS2) inhibits translation and pro- motes mRNA degradation [10,12,14,16,17]. However, these findings largely relied on reporter mRNAs to assess the extent of translation repression and RNA degradation. Since these analyses typically examine a pool of transfected cells and show a partial effect on translation repression and mRNA degradation, we hypothesized that the effects might be larger in individual cells based on our observations that Nsp1 is sufficient to essen- tially completely degrade the host GAPDH mRNA [4]. Given this, we analysed Nsp1-mediated effects on translation and host mRNA stability on endogenous host mRNAs in cells by exam- ining individual cells transfected with Nsp1 constructs. For these experiments, we transfected U-2 OS cells with a GFP-Nsp1 expression vector (Figure 1A). We then measured bulk translation of individual cells by pulse labelling cells for 4 hours with L-azidohomoalanine (AHA), a methionine analog, and measured AHA incorporation in proteins via fluorescence (Figure 1B) [26]. Cells expressing Nsp1 (Nsp1+) displayed a 75% reduction in AHA signal compared to untransfected controls, indicating that Nsp1 represses bulk translation of cellular mRNAs. The magnitude of translation repression was similar in cells treated with 1μM emetine, a translation inhibi- tor (Figure 1B). This demonstrates that Nsp1 substantially reduces the translation of essentially all host mRNAs. We also assessed mRNA degradation via smFISH by exam- ining mRNAs of different length and abundance. Specifically, we examined GAPDH mRNA (1kb, ~4k copies/cell) and AHNAK mRNA (18kb; ~100 copies/cell) [27]. Consistent with our previous studies, we observed an 88% reduction in GAPDH mRNA levels in Nsp1 expressing cells, confirming that Nsp1 reduces bulk cellular mRNA levels (Figure 1C) [4]. We also observed that Nsp1 similarly reduces cytoplasmic AHNAK mRNA levels (Figure 1D). We conclude that Nsp1 causes the degradation of cytoplasmic mRNAs in general, regardless of their abundance or length. RNA BIOLOGY 445 Interestingly, neither AHNAK nor GAPDH mRNAs nota- bly accumulated in the nucleus, despite the mRNA export block that can arise following bulk cytosolic mRNA degrada- tion (Fig S1B) [28–30]. We did observe that Nsp1 expressing cells had increased nuclear poly-A binding protein (PABP) signal and more nuclear oligo-dT smFISH signal (Figs S1A, S4). This suggests that many, but not all poly-A RNAs, are trapped in the nucleus and that the nuclear export block applies to specific transcripts. Montelukast does not alter GAPDH mRNA degradation A recent report suggested that montelukast, a common asthma medication, binds Nsp1 and can rescue translation repression of luciferase in transient transfection experiments [31]. An inhibitor of Nsp1 would be a useful tool for SARS2 research and potentially in the clinic as well, so we examined if montelukast altered mRNA degradation. For this experi- ment, we treated cells both before and after Nsp1 expression, then assayed GAPDH mRNA levels. We observed no signifi- cant difference in GAPDH degradation in either experiment: mRNA was reduced by the same levels as in Nsp1 controls (Figure 2). Since mRNA degradation appears downstream of translation repression (see below), this suggests that montelu- kast does not globally inhibit the function of Nsp1. Indeed, we observed that montelukast did not alter Nsp1’s ability to repress translation (Fig S2). We cannot rule out the possibility that montelukast can inhibit the action of Nsp1 on specific luciferase reporter mRNAs. Nsp1 does not degrade non-coding RNAs We examined whether Nsp1 could promote the degradation of untranslated lncRNAs. We examined this possibility given earlier work suggesting that Nsp1 can bind the 40S ribosome to block the mRNA entry channel and prevent ribosome engagement with the mRNA [14–17]. This has suggested two possible models by which Nsp1 represses translation and promotes mRNA degradation. In one model, Nsp1 could prevent the 40S ribosome from interacting with mRNAs and thereby destabilize those mRNAs. This first model predicts that Nsp1 could degrade both translating and untranslating mRNAs. Alternatively, Nsp1 could interact with 40S ribosomes engaged with the mRNA and thereby alter translation in a manner that promotes mRNA degradation. This alternative model predicts that Nsp1 would only be able to degrade translating mRNAs. To test these possibilities, we performed smFISH for the NORAD lncRNA (5.3kb; predomi- lncRNA nantly cytoplasmic, ~ 20 copies/cell), the GAS5 (725bp; cytoplasmic, ~200 copies/cell), and MALAT1, a nuclear localized lncRNA (~8kb; predominantly in nuclear speckles, ~90 copies/cell). We observed that Nsp1 does not reduce any of the untran- slating lncRNAs tested (Figure 3). Specifically, Nsp1 expres- sing cells did not display a difference in the diffuse localization or abundance of NORAD or GAS5 in the cyto- plasm in comparison to controls (Figure 3A). Similarly, Nsp1 expressing cells did not display alterations to MALAT1 RNA abundance or localization (Figure 3B). These data suggest that 446 S. I. SHEHATA AND R. PARKER Figure 1. Nsp1 represses endogenous protein translation and causes the degradation of cytosolic mRnas. A) Schematic of GFP-Nsp1 expression construct. B) IF for GFP and AHA-AF647 labelling of Nsp1+, untransfected cells, and cells treated with 1 μM emetine for 90 minutes. Quantified below. C) if for GFP and smFISH for GAPDH, quantified below. D) if for GFP and smFISH for AHNAK, quantified to the right. Ordinary one-way ANOVA compared to eGFP transfection control. **p < 0.0001; ns = not significant. RNA BIOLOGY 447 Figure 2. Montelukast treatment does not block GAPDH degradation in Nsp1+ cells. smFISH for GAPDH in cells treated with DMSO or 10 μM montelukast, either before (pre-treated) or after Nsp1 expression, quantified on the right. Ordinary one-way ANOVA compared to wtNsp1 + DMSO.; *p < 0.0001; ns = not significant. Nsp1 does not target either cytoplasmic- or nuclear-localized lncRNAs for degradation. We interpret these observations to argue that Nsp1 only degrades translating RNAs in the cyto- plasm, which implies that an mRNA needs to associate with a ribosome in a 48S initiation complex, or some alternative configuration, for it to be degraded by Nsp1. mRNA degradation and translation repression are two separate functions of Nsp1 Nsp1‘s sequence and structure are highly conserved through- out SARS1 and SARS2 and have remained very conserved throughout the SARS2 pandemic and evolution of variants [32]. Variants that do express mutated Nsp1 reduce the viral load in infected cells [33]. This suggests that Nsp1 is well evolved as a host shutoff factor and is sensitive to mutations. To test this possibility and to examine the importance of the different domains, we made a series of alanine-scanning mutations in conserved residues. We targeted alleles that are conserved across βCoVs, on the surface of the protein, and charged. We generated 12 mutants across the N-terminal and C-terminal domains of Nsp1 (Figure 4A) and screened each mutant for its ability to repress translation and/or degrade mRNAs by AHA-labelling and GAPDH smFISH. We used the K164A/H165A double mutant as a positive control, which is not able to bind the 40S ribosome and therefore fails to inhibit translation or degrade mRNAs [13,16,17]. This analysis revealed three broad categories of Nsp1 mutants: i) mutations that disrupt both translation repression and RNA degradation, ii) mutations that disrupt only mRNA decay, and iii) mutations that have no effect on Nsp1 func- tion. Many of the residues we tested fall in the latter category and have no effect on Nsp1 as a host shutoff factor (Fig S4). We suggest the mutations that do not affect Nsp1’s role in translation repression and/or mRNA degradation are likely to be conserved for other functions of Nsp1. Consistent with earlier results showing that the CTD of Nsp1 interacts with the ribosome, several mutations in the CTD of Nsp1 abolished both translation repression and mRNA degrada- tion [11,16,17,19]. As previously described, the K164A/H165A double mutant neither repressed translation nor degraded GAPDH mRNA (Figure 4). These residues are essential for Nsp1 binding the 40S, and when that interaction is disrupted, Nsp1 expression has no effect on cellular translation or mRNA stability. Additionally, the ribosome-interacting residues at posi- tions Y154, F157, R171, and R175 are also necessary for Nsp1’s ability to both repress translation and trigger RNA degradation [16,17]. Specifically, both Y154A/F157A and R171E/R175E dou- ble mutants restored AHA-labelled proteins to levels observed in control cells, and neither showed an ability to degrade GAPDH (Figure 4). All six of these mutations are in the distal region of the CTD, suggesting that a 20 amino acid stretch of the CTD must be intact for Nsp1 to suppress translation. Supporting this, the D152A point mutant also reduced the ability of Nsp1 to repress translation and degrade mRNA, although less completely than the more downstream residues (Figure 4B). This suggests that these residues help stabilize the Nsp1-40S interaction but are less critical for binding than the other ribosome-interacting residues, which bind directly to the 40S [17]. A surprising result was that the NTD residue L16 is also essential for Nsp1 translation repression and mRNA decay. We observed that the L16S mutant prevented translation repression by Nsp1 and was not able to degrade RNA (Figure 5). An NTD point mutation that abrogates transla- tion repression has not been observed before. Previously, an R99A mutant was identified and shown to have a moderate 448 S. I. SHEHATA AND R. PARKER Figure 3. Nsp1 does not degrade lncRNAs. A) smFISH for GAPDH and NORAD, quantified on the right. B) smiFISH for GAS5. C) smFISH for MALAT1. Ordinary one-way ANOVA compared to eGFP transfection control. ns = not significant. effect on restoring translation and mRNA stability [19], but L16S restores both GAPDH expression and protein synth- esis to normal levels in cells (Figure 5). L16S is expressed similarly to wtNsp1 in cells, suggesting that this point mutation isn’t significantly altering protein expression or folding (Fig S3B). We also tested translation inhibition of nano-luciferase by recombinant L16S in rabbit reticulocyte lysate, and we found that the L16S mutation partially reduces the ability of Nsp1 to repress translation compared to WT protein (Fig S3A). Thus, both in cells and in vitro, the L16S mutation compromises the ability of Nsp1 to repress translation. We identified two residues, E36 and P109, that are required for Nsp1 mediated RNA decay but not translation shutoff. The Nsp1 mutants E36A and P109A reduced transla- tion to the same extent as wtNsp1 but still had abundant levels of GAPDH mRNA. Consistent with these results, the E36A mutant was identified previously as having a mild RNA BIOLOGY 449 Figure 4. CTD of Nsp1 is critical for translation repression, which is upstream of mRNA decay. A) Amino acid sequence of Nsp1, with mutated residues highlighted in red. The three rows delineate the three domains of Nsp1: the NTD, linker, and CTD, respectively. B) AHA-AF647 labelling of bulk protein synthesis and smFISH for GAPDH in cells expressing mutant Nsp1 constructs. C) and D) quantification of AHA labelling and GAPDH smFISH. Ordinary one-way ANOVA compared to wtNsp1. p < 0.05; **p < 0.0001. mRNA degradation defect when combined with a E37A mutation [19]. This demonstrates that Nsp1-mediated trans- lational repression can occur without mRNA decay and that is more than a consequence of mRNA destabilization translation shutoff. We observed two additional residues in the NTD, P19 and D25, that slightly alter mRNA decay but have no effect on translation repression (Fig S4). This is similar to the CTD, where some residues are necessary for 450 S. I. SHEHATA AND R. PARKER Figure 5. NTD is involved in translation repression and mediates mRNA decay. L16S mutant neither blocks translation repression nor degrades RNA. E36A and P109A mutants abolish GAPDH degradation without restoring translation. A) AHA-AF647 labelling of bulk protein synthesis and smFISH for GAPDH in cells expressing mutant Nsp1 constructs. B) and C) quantification of AHA labelling and GAPDH smFISH. Ordinary one-way ANOVA compared to wtNsp1. p < 0.001; **p < 0.0001; ns = not significant. the Nsp1-40S interaction and some are needed for stabiliza- tion. Therefore, the translation repression and mRNA decay functions of Nsp1 can be decoupled from each other. Stalled translation initiation blocks mRNA decay The above data suggest that Nsp1 has two discrete functions, translation repression and RNA decay. Moreover, since cyto- solic lncRNAs are not sensitive to Nsp1, it suggests that an mRNA may need to interact with a ribosome for Nsp1 mediated degradation. To test this possibility further, we blocked translation at specific stages using translation inhibi- tors to ask how stalling ribosomes in different states affected Nsp1 mediated RNA degradation. In these experiments, we treated Nsp1-expressing cells with a translation inhibitor for five hours to allow for the production of new mRNAs, which would accumulate if Nsp1 mediated degradation was inhibited. In the first experiment, we inhibited translation elongation with emetine, which stalls elongating ribosomes along the mRNA by binding the E site and blocking translocation of the tRNA-mRNA [34]. Strikingly, the GAPDH mRNA was still strongly reduced in Nsp1 expressing cells treated with emetine (Figure 6A). This suggests that Nsp1 mediated mRNA decay can occur in the presence of stalled ribosomes on the mRNAs. In a second experiment, we inhibited translation initiation with pateamine A (PatA), which inhibits eIF4A function and RNA BIOLOGY 451 Figure 6. Nsp1 acts downstream of translation initiation and requires a monosome at the start codon before mRNA degradation can occur. smFISH for GAPDH and IF for G3BP in cells expressing GFP-Nsp1 or eGFP and treated with DMSO or 1 μM emetine (A), 100 nM pateamine A (B), or 2 μg/mL harringtonine (C). Quantification on the right. Yellow arrows point to G3BP+ stress granules. Ordinary one-way ANOVA compared to wtNsp1 + DMSO. *p < 0.0001; ns = not significant. blocks recruitment of the 43S pre-initiation complex (PIC) while the ribosomes along the mRNA run off, causing stress granule formation [35]. After five hours of PatA treatment in Nsp1+ cells, we observed increased GAPDH mRNAs by smFISH (Figure 6B). We interpret this observation to suggest that at least 48S engagement with the mRNA is required for efficient Nsp1 mediated mRNA degradation. This result implies that Nsp1, bound to the 40S, is included in the PIC and that the PIC must be able to interact with the mRNA before that mRNA is degraded. Consistent with PatA blocking Nsp1 mediated degradation, we observed that Nsp1 expressing cells treated with PatA form 452 S. I. SHEHATA AND R. PARKER stress granules based on the punctate formation of G3BP1 foci containing GAPDH mRNAs (Figure 6B). In contrast, untreated Nsp1 expressing cells are unable to make stress granules, presumably because most of the cellular mRNAs have been degraded [36]. The presence of stress granules validates the increased mRNA levels after PatA treatment in Nsp1 expressing cells. Interestingly, Nsp1 is enriched in stress granules formed in the presence of PatA, suggesting it can stably interact with 48S subunits and/or mRNAs present in stress granules. Trapping one monosome at translation initiation site blocks Nsp1-mediated mRNA degradation The above data indicate that Nsp1-mediated RNA decay occurs between early translation initiation and elongation. To test this further, we treated Nsp1 expressing cells with harringtonine (HTN), which binds the peptidyl transferase centre on the 60S ribosome and inhibits peptide bond formation during the first step of elongation. This traps a single 80S ribosome at the initiation codon while other translating ribosomes run off the mRNA [37]. HTN has been tested as a COVID therapeutic in clinical trials and has been shown to slow viral replication [38,39]. If a 43S PIC interacting with the mRNA is sufficient for Nsp1 mediated degradation, or if mRNA decay occurs before subunit joining, we expect to observe no impact of HTN on RNA degradation. Conversely, if elongating ribosomes are important for mRNA degradation, then we would anticipate HTN would reduce Nsp1 mediated RNA degradation. We observed that HTN inhibited the Nsp1 dependent degradation of GAPDH mRNAs (Figure 6C). Moreover, HTN treatment prevents the increased poly(A)+ RNA accu- mulation in the nucleus seen with Nsp1 expression (Fig S5). This suggests that a single 80S ribosome stalled at the AUG is not sufficient for Nsp1 mediated RNA degradation and that subunit the presence of Nsp1. Alternatively, it is possible that HTN blocks subunit joining when Nsp1 is bound to the 40S, and that mRNA decay requires at least a fully assembled monosome at the start codon. This result suggests that HTN treatment traps Nsp1 in an assembly with mRNAs that reduces their degradation, and further experiments are needed to characterize this assembly. joining can occur in Discussion Our results strengthen the previous conclusion that the Nsp1 C terminal domain is required for translation repression, which has been established by showing Nsp1 directly interacts with the ribosome and mutations altering that interaction are defective in translation repression [13,16,17]. Here we expand upon that observation, validating that there are at least four other residues, Y154, F157, R171, R175, and to a lesser effect D152, in the C terminal domain that are also required for the full degree of Nsp1 translation repression (Figure 4) and presumably affect the Nsp1-40S interaction [17]. This is con- sistent with a model wherein additional parts of the Nsp1 CTD are involved in the Nsp1-40S complex [19] and that SARS-CoV-2 variants with CTD deletions replicate less effec- tively [33]. We also observe that the NTD contains residues important for translation repression. Specifically, we identify the L16S point mutation that diminishes the ability of Nsp1 to repress translation in cells (Figure 5). Moreover, this mutant protein is also partially defective at translation repression in vitro (Fig S3). This is consistent with deletion analysis where removal of the entire Nsp1 N terminal domain leads to reduced transla- tion repression [19]. This suggests a possible interaction between the NTD and CTD that could alter the Nsp1-40S interaction, or the N terminal domain could function in translation repression through an additional, yet to be identi- fied interaction. Several observations now suggest that RNA degradation is downstream of translation repression and requires additional functions/interactions of Nsp1 after binding the 40S ribo- some. This has been initially suggested by the observation that mutations in Nsp1 blocking translation repression, all lead to stabilization of the mRNA (Figures 3, 4) [16,17,19]. Moreover, a key observation is that two specific mutations in Nsp1, E36A and P109A, both still strongly repress host trans- lation but do not degrade host mRNAs (Figure 5). This is consistent with previous work showing that R99 and R124/ K125 residues are important for mRNA stability [12,19], revealing that one role of the NTD is to promote mRNA decay. However, the previously described mutations also res- cue translation repression, presumably by altering the Nsp1- 40S complex. It remains unknown how the NTD residues E36 and P109 contribute to mRNA decay, but one potential mechanism is that they recruit or stabilize a host nuclease. Past experiments have proposed that Xrn1 mediates 5’-3’ mRNA degradation after SARS1 Nsp1 expression [24], and an Nsp1 interactome indicated that Nsp1 interacts with Xrn1 as well as DIS3, part of the 3’-5’ exosome [18]. However, these relationships have not been tested directly in cells, and the mechanism of Nsp1 mediated RNA degradation is still unknown. efficiency that cytosolic untranslated Several observations now argue that Nsp1 requires the mRNA to interact with ribosomes to be degraded. First, although Nsp1 RNA decay targets multiple mRNAs, we observed lncRNAs are not degraded by Nsp1 (Figure 3). Second, RNA-seq data has demonstrated a correlation between translation efficiency and Nsp1-dependent mRNA degradation [40]. Increased translation indicates more mRNA-ribosome engagement, which is consistent with the model in which Nsp1 mRNA degradation depends upon at least an mRNA- 40S-Nsp1 complex. Third, we observed that either preventing 40S ribosome interaction with the mRNA by PatA treatment, or inhibiting the transition to elongation by 80S subunits with HTN stabilizes mRNAs from Nsp1-dependent degradation (Figure 6). These results are consistent with a model wherein elongating 80S ribosomes are required for Nsp1 to trigger mRNA degradation. An unresolved issue is the exact nature and diversity of the Nsp1-ribosome-mRNA complex(es). Nsp1 competes with eIF3J for ribosome binding, preventing formation of a functional 48S complex [14]. The same study showed through structural analyses that Nsp1 and the cricket paralysis virus (CrPV) IRES mRNA can bind the 40S ribo- some simultaneously, but that Nsp1 limits the 40S head conformational change that is necessary for scanning [14]. This finding is inconsistent with the single-molecule obser- vations that mRNA and Nsp1 are mutually exclusive in the mRNA entry channel: when one is bound, the other cannot interact [15]. However, previous ribosome toe-printing analysis with SARS1 demonstrated that a 48S complex (mRNA-43S PIC-Nsp1) can form, but that subunit joining is blocked [11]. This presents a contradiction. Solved structures of Nsp1 bound to the ribosome show that Nsp1 can bind to an 80S ribosome, but they also reveal a clear block to the mRNA entry channel suggesting that Nsp1-40S interaction would prevent the 40S ribosome interacting with mRNA [16,17]. Additionally, it is still unclear if Nsp1 permits the forma- tion of a functional 48S, wherein the mRNA is properly accommodated on the 40S ribosome. However, our obser- vations suggest that ribosome engagement with the mRNA is required for Nsp1 to promote mRNA degradation (Figures 3 & 6). Although the basis of this apparent contra- diction remains to be solved, we suggest three possibilities. First, it could be that Nsp1 interacts with a second 40S pre- initiation complex and the interaction of the Nsp1-40S complex with an 80S on the mRNA triggers mRNA degra- dation, which could explain Nsp1-80S interactions [16,17]. Alternatively, Nsp1 may insert into the mRNA entry chan- nel at a later stage of the mRNA decay process, and there- fore the structure does not reveal an initial required interaction. Lastly, Nsp1 could force mRNA to bind a different site on the ribosome through interactions with RBPs and eIFs. Understanding the Nsp1-ribosome-mRNA assembly will require further structural analyses. The requirement for ribosome mRNA engagement for Nsp1 mediated degradation raises the possibility that Nsp1 may alter ribosome dynamics. Interestingly, when ribosome movement is inhibited, the collision of either 80S subunits, or 40S and 80S subunits, can trigger ribosome quality control pathways that lead to mRNA cleavage and degradation [41]. This, and other observations, lead to the possibility that Nsp1 leads to activation of RQC pathways and subsequent mRNA decay. For example, Nsp1 from SARS is known to induce endonucleolytic cleavage of mRNAs in specific contexts [11,12,23]. Moreover, Nsp1 overexpression has been proposed to resolve stalled ribosome complexes [42]. Further experi- ments are needed to determine if Nsp1’s involvement with quality control factors is what causes mRNA cleavage and degradation. Materials and methods Cell culture and transfections RNA BIOLOGY 453 mycoplasma contamination by our core facility. Cells were transfected with X-tremeGENE HP transfection reagent (2 μl per 1 μg DNA) (Sigma 6366244001). Four hours after adding transfection mix, the media was replaced. Cells were fixed for downstream analyses 24 hours after transfection. Cloning and mutagenesis Full-length wtNsp1, GFP, and eGFP were synthesized by Integrated DNA Technologies (IDT) as g-Block gene frag- ments. FLAG-Nsp1 was cloned into the pcDNA3.1+ vector as described in Burke, et al 2021. GFP and eGFP were ligated upstream of Nsp1 using EcoRI and XhoI restriction sites. All plasmids sequences were verified using Sanger sequencing or whole plasmid sequencing. Site-directed mutagenesis was used to generate point mutants and double mutants (primers listed in Table S1). Fluorescence microscopy sample preparation Sequential smFISH and IF on U2OS cells were performed as described previously with the following modifications [27]. Cells were grown on glass coverslips in a 24-well plate, then fixed for 10 minutes with 4% paraformaldehyde. GAPDH and oligo-dT smFISH probes labelled with Quasar-570 were pur- chased from Stellaris. Custom AHNAK smFISH probes [27] were purchased as DNA oligos (Table S2) from IDT and labelled with Atto-633 using 5-Propargylamino-ddUTP- ATTO-633 using transferase (Thermo Fisher Scientific) as described previously [43]. Custom NORAD and GAS5 smiFISH probes were purchased as DNA oligos (Table S2) and labelled with ATTO-647n as described previously [44]. terminal deoxynucleotidyl After smFISH/smiFISH was performed, coverslips were rinsed with phosphate buffered saline (PBS) before being fixed again for 10 minutes in 4% paraformaldehyde. Cells were washed twice with PBS then incubated for one hour at room temperature with primary antibody, washed three more times with PBS then incubated with secondary antibody for 1 hour. After washing 3× with PBS, coverslips were mounted slides with ProLong Glass Antifade Mountant on (ThermoFisher P36982). Antibodies used: polyclonal anti-GFP (rabbit, 1:500, Invitrogen A-11122); monoclonal anti-GFP (mouse, 1:500, Santa Cruz sc-9996); monoclonal anti-G3BP (mouse, 1:500, Abcam ab56574); polyclonal anti-PABP (rabbit, 1:500, Abcam ab21060). Secondary antibodies were all used at 1:1000 dilu- tion. Goat anti-rabbit IgG Alexa Fluor 488 (Invitrogen A11008); goat anti-mouse IgG Alexa Fluor 488 (Invitrogen A1101); Goat Anti-Mouse IgG H&L Alexa Fluor 647 (Abcam ab150115); Goat Anti-Rabbit IgG H&L Alexa Fluor 647 (Abcam ab150079). L-azidohomoalanine (AHA) labelling U2OS cell lines (ATCC) were maintained at 5% CO2 and 37°C in Dulbecco’s modified Eagle’s medium (DMEM) supplemen- ted with 10% v/v foetal bovine serum (FBS) and 1% v/v for penicillin/streptomycin. Cells were routinely tested AHA labelling was performed as described by the manufac- turer (Fisher C10102). Briefly, cells growing on glass cover- slips were incubated in methionine-free DMEM (Gibco 21013024) with 10% v/v FBS for one hour. 50 μM Click-iT 454 S. I. SHEHATA AND R. PARKER AHA (Invitrogen C10102) was added to cells for four hours before fixation with 4% paraformaldehyde. Cells were per- meabilized with 0.1% Triton-X 100 in PBS for 15 minutes, then washed with 3% BSA in PBS before adding the Click-iT Cell Reaction cocktail (Invitrogen C10269). Click reaction with Alexa-Fluor 647 alkyne (Invitrogen A10278) was per- formed as described by the manufacturer (https://assets.ther m o f i s h e r . c o m / T F S - A s s e t s % 2 F L S G % 2 F m a n u a l s % 2Fmp10269.pdf). Drug treatments U2OS cells were transiently transfected with GFP-Nsp1 as described above. After 20 hours, media was replaced with media containing either 1:1000 DMSO, 1 μM emetine (Cayman Chemical Company 316-42-7), 100 nM pateamine A, or 2 μg/mL harringtonine (Abcam ab141941). Cells were incubated at 37°C for 5 hours before fixation. For montelukast treatment, cells were treated for 5 hours with 10 μM monte- lukast (Sigma SML0101) as described above. For the pretreat- ment condition, cells were seeded onto glass coverslips in 10 μM montelukast 24 hours before Nsp1 transfection. Microscopy and image analysis Fixed cell imaging was performed using a 100× oil objective on a Nikon A1R Laser Scanning Confocal microscope with an Andor iXon 897 Ultra detector or a Nikon SoRa Spinning Disc Confocal microscope with an Andor iXon Life 897 EMCCD Camera. We imaged 10–20 z-slices at a distance of 0.2 μM/slice. All images shown are a max projection in Z of the z-stack. We used ImageJ (version 2.9.0/1.53t) to quantify GAPDH smFISH spots. After creating a max intensity projection in Z, we set an intensity threshold for the experiment and analysed the number of pixels above that threshold within manually defined regions of interest (ROIs). Each ROI represents a single cell. To quantify AHNAK, NORAD and GAS5 smFISH spots, we first made a mask of the nuclei using CellProfiler (version 4.2.5) to specifically analyse the cytosol (or nuclei), set an intensity threshold as described above, then used the ImageJ feature Analyse Particles to count the number of smFISH spots above the intensity threshold in each ROI. AHA intensity was quantified using ROIs manually defined in ImageJ. We then measured the average intensity of each ROI. SDS-PAGE and Western blotting Cells were transfected with GFP-Nsp1 constructs as described above in 6 well plates. After 24 hours, cells were trypsinized and pelleted, then resuspended in 200 μl lysis buffer (1.25% SDS and 4% β-mercaptoethanol in water). Cells were frozen at −80°C to complete lysis then boiled at 95°C for 10 minutes prior to running on a NuPAGE 4– 12% Bis-Tris Gel (Fisher Scientific, NP0322BOX) and to a nitrocellulose membrane (ThermoFisher, transfer IB23002). The membrane was blocked with 5% milk in Tris-buffered saline with 0.1% Tween-20 (TBS-T) for 1 hour at room temperature then incubated with primary antibody in 5% milk and TBS-T overnight at 4°C. The membrane was washed three times with TBS-T then incu- bated with secondary antibody for 1 hour at room tem- perature, then washed again with TBS-T before incubation with the chemiluminescence substrate (Biorad 170506) for 5 minutes at room temperature prior to imaging with an iBright FL1500 anti-Nsp1 (GeneTex GTX135612) and GAPDH – HRP antibody (Santa Cruz, sc -47724 HRP), anti-rabbit IgG HRP-linked secondary antibody (Cell Signaling Technology, 7074S) were used at 1:1000. System. Rabbit Imaging Protein purification Recombinant wtNsp1 and mutants were expressed and pur- ified as described previously [19]. The pGEX-Nsp1 expres- sion plasmid (Addgene 175512) was mutated using the Quikchange site directed mutagenesis kit (Agilent 200153), then transformed in BL21 competent cells (Agilent 230,240). Overnight Express TB (EMD Millipore 71,491–4) cultures were inoculated and grown at 37°C to an OD600 of 0.6, then moved to 18°C for 24 hours. Cells were pelleted, washed with PBS and a protease inhibitor cocktail (Sigma Aldrich 11836170001), then pelleted again. Cells were resuspended in lysis buffer (500 mM NaCl, 5 mM MgCl2, 20 mM HEPES, 0.5% Triton-X 100, 5% glycerol, 1 mM TECEP, pH 7.5), then lysed at 4°C for 12 minutes with a macrotip sonicator (3 second pulse and 17 second rest), then centrifuged for 35 minutes at 39,000×g. Glutathione Sepharose (Sigma, GE17075601) beads were washed with 5 column volumes (CV) of lysis buffer on a glass Econo-Column (Biorad 7372512), then incubated with the cleared lysate for 2 hours at 4°C. Flow-through was collected and the beads were washed with 10 CV of lysis buffer, then washed with 20 CV of wash buffer (250 mM NaCl, 5 mM MgCl2, 5% glycerol, 20 mM HEPES, 1 mM TECEP, pH 7.5), and finally resus- pended in 1 CV of wash buffer with HRV 3C Protease (Millipore 71493). Nsp1 was eluted overnight at 4°C on a nutator then collected. The beads were washed with 1 CV of wash buffer. The two elutions were pooled and concen- trated with an Amicon Ultra-4 10K centrifugal filter (Millipore UFC801024). Concentrated protein was then pur- ified by size exclusion column chromatography. Purity and mass of the eluted protein were confirmed by SDS-PAGE gel electrophoresis (Invitrogen LC6060). and SimplyBlue SafeStain In vitro transcription Nanoluciferase (NanoLuc) mRNA was generated with T7 RNA polymerase (ThermoFisher AM1334) as previously described [19]. DNA was PCR amplified from a plasmid with the NanoLuc coding sequence (Addgene 175431), then gel purified before T7 transcription. After transcription, RNA was purified using the MEGAClear Transcription Clean-Up Kit (Thermo Scientific AM1908), then capped and 2’O methylated with the Vaccinia Capping System (NEB M2080S) and the mRNA Cap 2 ´-O-Methyltransferase (NEB M0366). Capped NanoLuc mRNA was isolated using Direct-zol RNA Miniprep kits (Zymo R2050). In vitro translation assay Nuclease treated rabbit reticulocyte lysate (RRL) (Promega L4960) was used as described by the manufacturer for non-radioactive reactions with the following modifications. RRL lysate was incubated with 640 nM of Nsp1 or glu- tathione-S transferase (Sigma G6511) for 10 minutes on ice before adding nanoluciferase mRNA. The reaction pro- ceeded for 90 minutes at 30°C before the addition of the NanoGlo (Promega N1110). substrate Luminescence was visualized with a BMG plate reader. luciferase Acknowledgments The imaging work was performed at the BioFrontiers Institute Advanced Light Microscopy Core (RRID: SCR_018302). Laser scanning confocal microscopy was performed on a Nikon A1R microscope sup- ported by NIST-CU Cooperative Agreement award number 70NANB15H226. This work was supported by funds from the National Institute on Aging (F30AG076323) and the Howard Hughes Medical Institute. Disclosure statement No potential conflict of interest was reported by the author(s). Funding The work was supported by the Howard Hughes Medical Institute and the National Institutes of Health awards 5R01GM045443 and F30AG076323. Data availability statement The data that support the findings of this study can be accessed online at https://doi.org/10.6084/m9.figshare.c.6569116. ORCID Soraya I. Shehata Roy Parker http://orcid.org/0000-0002-2909-064X http://orcid.org/0000-0002-8412-4152 References [1] WHO Coronavirus (COVID-19) dashboard [internet]. [cited 2023 Apr 7]; Available from: https://covid19.who.int [2] Goldstein SA, Brown J, Pedersen BS, et al. Extensive recombination-driven coronavirus diversification expands the pool of potential pandemic pathogens. Genome Biol Evol. 2022;14(12):evac161. doi: 10.1093/gbe/evac161 [3] Lavine JS, Bjornstad ON, Antia R. Immunological characteristics govern the transition of COVID-19 to endemicity. Science. 2021;371(6530):741–745. doi: 10.1126/science.abe6522 [4] Burke JM, Clair LAS, Perera R, et al. SARS-CoV-2 infection triggers widespread host mRNA decay leading to an mRNA export block. RNA. 2021;27(11):1318–1329. doi: 10.1261/rna. 078923.121 [5] de Breyne S, Vindry C, Guillin O, et al. Translational control of coronaviruses. Nucleic Acids Res. 2020;48(22):12502–12522. doi: 10.1093/nar/gkaa1116 RNA BIOLOGY 455 [6] Hartenian E, Nandakumar D, Lari A, et al. The molecular virology of coronaviruses. J Biol Chem. 2020;295(37):12910–12934. doi: 10. 1074/jbc.REV120.013930 [7] Tohya Y, Narayanan K, Kamitani W, et al. Suppression of host gene expression by nsp1 proteins of group 2 bat coronaviruses. J Virol. 2009;83(10):5282–5288. doi: 10.1128/JVI.02485-08 [8] Masters PS. The molecular biology of coronaviruses. Adv Virus Res. 2006;66:193–292. [9] Zhou P, Yang X-L, Wang X-G, et al. A pneumonia outbreak associated with a new coronavirus of probable bat origin. Nature. 2020;579(7798):270–273. doi: 10.1038/s41586-020-2012-7 [10] Kamitani W, Narayanan K, Huang C, et al. Severe acute respira- tory syndrome coronavirus nsp1 protein suppresses host gene expression by promoting host mRNA degradation. PNAS. 2006;103(34):12885–12890. doi: 10.1073/pnas.0603144103 [11] Kamitani W, Huang C, Narayanan K, et al. A two-pronged strat- egy to suppress host protein synthesis by SARS coronavirus Nsp1 protein. Nat Struct Mol Biol. 2009;16(11):1134–1140. doi: 10. 1038/nsmb.1680 [12] Lokugamage KG, Narayanan K, Huang C, et al. Severe acute respiratory syndrome coronavirus protein nsp1 is a novel eukar- yotic translation inhibitor that represses multiple steps of transla- tion Initiation. J Virol. 2012;86(24):13598–13608. doi: 10.1128/ JVI.01958-12 [13] Narayanan K, Huang C, Lokugamage K, et al. Severe acute respiratory syndrome coronavirus nsp1 suppresses host gene expression, including that of type I interferon, in infected cells. J Virol. 2008;82(9):4471–4479. doi: 10.1128/JVI.02472-07 [14] Yuan S, Peng L, Park JJ, et al. Nonstructural protein 1 of SARS-CoV-2 is a potent pathogenicity factor redirecting host protein synthesis machinery toward viral RNA. Mol Cell. 2020;80(6):1055–1066.e6. doi: 10.1016/j.molcel.2020.10.034 [15] Lapointe CP, Grosely R, Johnson AG, et al. Dynamic competition between SARS-CoV-2 NSP1 and mRNA on the human ribosome inhibits translation initiation. Proc Natl Acad Sci, USA. 2021;118 (6):e2017715118. doi: 10.1073/pnas.2017715118 [16] Schubert K, Karousis ED, Jomaa A, et al. SARS-CoV-2 Nsp1 binds the ribosomal mRNA channel to inhibit translation. Nat Struct Mol Biol. 2020;27(10):959–966. doi: 10.1038/s41594-020-0511-8 [17] Thoms M, Buschauer R, Ameismeier M, et al. Structural basis for translational shutdown and immune evasion by the Nsp1 protein of SARS-CoV-2. Science. 2020;369(6508):1249–1255. doi: 10.1126/ science.abc8665 [18] Bujanic L, Shevchuk O, von KN, et al. The key features of SARS-CoV-2 leader and NSP1 required for viral escape of NSP1-mediated repression. RNA. 2022;28(5):766–779. doi: 10. 1261/rna.079086.121 [19] Mendez AS, Ly M, González-Sánchez AM, et al. The N-terminal domain of SARS-CoV-2 nsp1 plays key roles in suppression of cellular gene expression and preservation of viral gene expression. Cell Rep. 2021;37(3):109841. doi: 10.1016/j.celrep.2021.109841 [20] Slobodin B, Sehrawat U, Lev A, et al. Cap-independent translation and a precisely localized RNA sequence enable SARS-CoV-2 to control host translation and escape anti-viral response [Internet]. Mol Biol. 2021; [cited 2021 Aug 20]. Available from. http://bior xiv.org/lookup/doi/10.1101/2021.08.18.456855 [21] Vora SM, Fontana P, Mao T, et al. Targeting stem-loop 1 of the SARS-CoV-2 5′ UTR to suppress viral translation and Nsp1 eva- sion Proc Natl Acad Sci USA. 2022;119(9):e2117198119. doi:10. 1073/pnas.2117198119 [22] Almeida MS, Johnson MA, Herrmann T, et al. Novel β-barrel fold in the nuclear magnetic resonance structure of the replicase non- structural protein 1 from the severe acute respiratory syndrome coronavirus. J Virol. 2007;81(7):3151–3161. doi: 10.1128/JVI. 01939-06 [23] Huang C, Lokugamage KG, Rozovics JM, et al. SARS coronavirus nsp1 protein induces template-dependent endonucleolytic clea- vage of mRnas: viral mRNAs are resistant to nsp1-Induced RNA cleavage. PLOS Pathogens. 2011;7(12):e1002433. doi: 10.1371/jour nal.ppat.1002433 456 S. I. SHEHATA AND R. PARKER [24] Gaglia MM, Covarrubias S, Wong W, et al. A common strategy for host RNA degradation by divergent viruses. J Virol. 2012;86 (17):9527–9530. doi: 10.1128/JVI.01230-12 [25] Shi M, Wang L, Fontana P, et al. SARS-CoV-2 Nsp1 suppresses host but not viral translation through a bipartite mechanism. bioRxiv [preprint]. 2020;2020.09.18.302901. doi:10.1101/2020.09. 18.302901 [26] Groskreutz DJ, Babor EC, Monick MM, et al. Respiratory syncy- tial virus limits α subunit of eukaryotic translation initiation factor 2 (eIf2α) phosphorylation to maintain translation and viral repli- cation. J Biol Chem. 2010;285(31):24023–24031. doi: 10.1074/jbc. M109.077321 [27] Khong A, Matheny T, Jain S, et al. The stress granule transcrip- tome reveals principles of mRNA accumulation in stress granules. Mol Cell. 2017;68(4):808–820.e5. doi: 10.1016/j.molcel.2017.10. 015 [28] Burke JM, Gilchrist AR, Sawyer SL, et al. RNase L limits host and viral protein synthesis via inhibition of mRNA export. Sci Adv. 2021;7(23):eabh2479. doi: 10.1126/sciadv.abh2479 [29] Gilbertson S, Federspiel JD, Hartenian E, et al. Changes in mRNA abundance drive shuttling of RNA binding proteins, linking cyto- plasmic RNA degradation to transcription. Elife. 2018;7:e37663. doi: 10.7554/eLife.37663 [30] Zhang K, Miorin L, Makio T, et al. Nsp1 protein of SARS-CoV-2 disrupts the mRNA export machinery to inhibit host gene expression. Sci Adv. 2021;7(6):eabe7386. doi: 10.1126/sciadv. abe7386 [31] Afsar M, Narayan R, Akhtar MN, et al. Drug targeting shows antiviral activity against Nsp1-ribosomal complex SARS-CoV-2 [internet]. eLife2022; [cited 2023 Jan 10]. Available from: https://elifesciences.org/articles/74877/figures [32] Min Y-Q, Mo Q, Wang J, et al. SARS-CoV-2 nsp1: bioinformatics, potential structural and functional features, and implications for drug/vaccine designs. Front Microbiol [Internet]. 2020[cited 2021 Jun 2]; 11. Available from: 10.3389/fmicb.2020.587317 [33] Lin J, Tang C, Wei H, et al. Genomic monitoring of SARS-CoV-2 uncovers an Nsp1 deletion variant type I interferon response. Cell Host Microbe. 2021;29(3):489–502.e8. doi: 10.1016/j.chom.2021.01.015 that modulates [34] Wong W, Bai X, Brown A, et al. Cryo-EM structure of the Plasmodium the 80S anti-protozoan drug emetine. Elife. 2014;3:e03080. doi: 10.7554/ eLife.03080 falciparum ribosome bound to [35] Bordeleau M-E, Cencic R, Lindqvist L, et al. RNA-Mediated sequestration of the RNA helicase eIF4A by pateamine a inhibits translation initiation. Chem Biol. 2006;13(12):1287–1295. doi: 10. 1016/j.chembiol.2006.10.005 [36] Burke JM, Lester ET, Tauber D, et al. RNase L promotes the formation of unique ribonucleoprotein granules distinct from stress granules. J Biol Chem. 2020;295(6):1426–1438. doi: 10. 1074/jbc.RA119.011638 [37] Fresno M, Jiménez A, Vázquez D. Inhibition of translation in eukaryotic systems by harringtonine. Eur J Biochem. 1977;72 (2):323–330. doi: 10.1111/j.1432-1033.1977.tb11256.x [38] Ma H, Wen H, Qin Y, et al. Homo-harringtonine, highly effective against coronaviruses, treating COVID-19 by is safe nebulization. Sci China Life Sci. 2022;65(6):1263–1266. doi: 10. 1007/s11427-021-2093-2 in [39] Wen H-J, Liu F-L, Huang M-X, et al. A proposal for clinical trials of COVID-19 treatment using homo-harringtonine. Natl Sci Rev. 2021;8(1):nwaa257. doi: 10.1093/nsr/nwaa257 [40] Fisher T, Gluck A, Narayanan K, et al. Parsing the role of NSP1 in SARS-CoV-2 infection. Cell Rep. 2022;39(11):110954. doi: 10. 1016/j.celrep.2022.110954 [41] Kim KQ, Zaher HS. Canary in a coal mine: collided ribosomes as sensors of cellular conditions. Trends Biochem Sci. 2022;47 (1):82–97. doi: 10.1016/j.tibs.2021.09.001 [42] Wang X, Rimal S, Tantray I, et al. Prevention of ribosome colli- sion-induced neuromuscular degeneration by SARS CoV-2– encoded Nsp1. Proc Natl Acad Sci USA. 2022;119(42): e2202322119. doi:10.1073/pnas.2202322119 [43] Gaspar I, Wippich F, Ephrussi A. Enzymatic production of single-molecule FISH and RNA capture probes. RNA. 2017;23 (10):1582–1591. doi: 10.1261/rna.061184.117 [44] Tsanov N, Samacoits A, Chouaib R, et al. smiFISH and FISH- quant – a flexible single RNA detection approach with super- resolution capability. Nucleic Acids Res. 2016;44(22):e165. doi: 10.1093/nar/gkw784
10.1021_acscentsci.2c01385	http://pubs.acs.org/journal/acscii Research Article Enzymatic Fluoromethylation Enabled by the S‑Adenosylmethionine Analog Te- Adenosyl‑L‑(fluoromethyl)homotellurocysteine Syam Sundar Neti, Bo Wang,* David F. Iwig, Elizabeth L. Onderko, and Squire J. Booker* Cite This: ACS Cent. Sci. 2023, 9, 905−914 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: Fluoromethyl, difluoromethyl, and trifluoromethyl groups are present in numerous pharmaceuticals and agrochemicals, where they play critical roles in the efficacy and metabolic stability of these molecules. Strategies for late-stage incorporation of fluorine-containing atoms in molecules have become an important area of organic and medicinal chemistry as well as synthetic biology. Herein, we describe the synthesis and use of Te-adenosyl-L-(fluoromethyl)homotellurocysteine (FMeTeSAM), a novel and biologically relevant fluoromethylating agent. FMeTeSAM is structurally and chemically related to the universal cellular methyl donor S-adenosyl-L-methionine (SAM) and supports the robust transfer of fluoromethyl groups to oxygen, nitrogen, sulfur, and some carbon nucleophiles. FMeTeSAM is also used to fluoromethylate precursors to oxaline and daunorubicin, two complex natural products that exhibit antitumor properties. ■ INTRODUCTION Methylation underpins myriad cellular processes central to life, such as transcription, translation, gene regulation, signaling, and general metabolism.1−7 S-Adenosyl-L-methionine (SAM or AdoMet), often referred to as Nature’s universal methylating the agent, is the overwhelmingly predominant source of appended methyl groups.8,9 Enzymes use SAM to methylate numerous biomolecules, including DNA, RNA, proteins, lipids, carbohydrates, and a wide variety of small molecules.5,10−19 The canonical reaction involves a polar SN2 attack of a nucleophile onto the electrophilic methyl substituent of SAM, affording S- adenosyl-L-homocysteine (SAH) as a coproduct (Figure 1A).20,21 The most widely methylated nucleophiles are N, O, and C (where carbanions can be generated). However, P, S, Se, Te, As, Co (in cobalamin), and Hg also receive methyl groups from SAM.10,11,22−26 SAM-dependent methyltransferases (MTases) are ubiquitous in nature, and a variety of them have been shown in vitro to have broad substrate scopes.27 The introduction of alkyl groups�especially methyl groups�is a well-known strategy in the pharmaceutical and agricultural industries for improving the pharmacological properties of drug candidates and natural products by tuning their binding affinities, solubilities, and metabolic profiles.28 However, conventional synthetic alkylating methods have several disadvantages, including toxic reactants (such as alkyl halide) and a lack of robust regio-, chemo-, or stereoselectivities. By contrast, MTase-catalyzed methylations are highly regio- and stereoselective, making them promising biocatalysts for the late- stage diversification of complex molecules. There has been growing interest in exploiting MTases for synthetic applications in recent years.29 A variety of SAM analogs bearing different alkyl groups in place of the methyl substituent have been synthesized and used as cosubstrates in MTase-catalyzed reactions for various purposes.30 However, the transfer of fluorine-bearing alkyl groups is much more difficult due to the instability of the corresponding SAM analog.31 The introduction of fluorine into pharmaceuticals, agro- chemicals, and other molecules of value is an ongoing and major focus of synthetic chemists because of the unique properties that fluorine atoms confer on molecules.32 In fact, fluorine is found in Received: November 21, 2022 Published: May 8, 2023 © 2023 The Authors. Published by American Chemical Society 905 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 1. MTase-catalyzed methylation using (A) SAM and fluoromethylation using (B) FMeTeSAM. Figure 2. Chemical structures of (A) SAM, (B) SeSAM, and (C) TeSAM 20−30% of all pharmaceuticals and ∼30% of all agro- chemicals.33,34 Due to the electronegativity of fluorine, C−F bonds can tune the strength of proximal bonds, resulting in a substantial impact on the physicochemical properties of an entire molecule.35,36 In addition, 19F, the only natural isotope of fluorine, provides a low-noise and highly sensitive tool to study drug−target interactions and other properties of molecules through nuclear magnetic resonance (NMR) spectroscopy.37 Due to the importance of fluorine-containing small molecules for therapeutics, there has been ongoing interest in developing strategies to incorporate fluorine atoms site-specifically into organic compounds, biochemical metabolites, and natural products.38−48 In this work, we describe the synthesis of Te- adenosyl-L-(fluoromethyl)homotellurocysteine (FMeTeSAM) and its application in the fluoromethylation of biological targets (Figure 1B). ■ DESIGN AND SYNTHESIS OF FMETESAM In our studies of the enzyme cyclopropane fatty acid (CFA) synthase, we synthesized the Se- and Te-containing analogs of SAM (Se-adenosyl-L-selenomethionine and Te-adenosyl-L- telluromethionine (SeSAM and TeSAM, respectively)) (Figure 2) as probes of the enzyme’s reaction mechanism and showed that TeSAM exhibits marked stability over SAM and SeSAM.49 SAM exhibits a half-life on the order of days at pH 7.5−10.0 and 37 °C and degrades in three distinct ways (Figure S1). It undergoes racemization at the sulfur to afford the inactive (R, S) diastereomer. At a pH above 3, it undergoes an intramolecular cyclization to render methylthioadenosine (MTA) and homoserine lactone. At pH values above 7, it undergoes deprotonation at C5′, which results in the elimination of adenine and the production of S-ribosylmethionine.49−54 SeSAM does not undergo racemization and only undergoes deprotonation at C5′ at pH values above 12.0. However, it degrades to homoserine lactone and methylselenoadenosine about 10-fold faster than SAM. By contrast, TeSAM does not undergo any significant degradation at 37 °C in the pH range of 2−12. Moreover, we found that L-telluromethionine appears to be as good as L-methionine in SAM synthetase reactions under saturating conditions, suggesting the possibility of generating FMeTeSAM enzymatically from L-(fluoromethyl)- homotellurocysteine.49 Importantly, both SeSAM and TeSAM were good cosubstrates in methylation reactions catalyzed by CFA synthase and catechol O-methyltransferase (COMT).55 Given the stability of TeSAM and the ease by which it can be synthesized enzymatically, we posited that it might be possible to generate and isolate its fluoromethyl analog. Recently, Seebeck and colleagues reported the in situ generation of fluoromethyl-SAM�using fluoromethyl iodide, SAH, and halide methyltransferase�and its application in transferring fluoromethyl groups to a variety of nucleophiles. Although this strategy was clever, the fluoromethyl-SAM was highly unstable and could not be isolated for kinetic studies.31 The synthesis of FMeTeSAM was completed in six steps starting from L-homoserine, with the final step involving an enzymatic transformation (Scheme S1). NMR spectroscopy and 906 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Table 1. Nucleophile Scopes and Key Kinetic Parameters aUnit of kcat is min−1. bUnit of Km is μM. cUnit of kcat/Km is μM−1min−1. dMonofluoromethylation on aliphatic amines resulted in carbinolamine decomposition followed by hydrolysis, presumably giving fluoride ion and formaldehyde. eNot applicable. fkcat and kcat/Km for NNMT is multiplied by 10−3. high-resolution mass spectrometry (HRMS) verified the final product’s authenticity. In particular, the mass spectrum of FMeTeSAM shows the characteristic isotopic distribution associated with tellurium, with the major isotopes being 130Te (34.08%), 128Te (31.74%), 126Te (18.84%), 125Te (7.07%), 124Te (4.74%), and 122Te (2.55%). 19F-NMR shows a resonance split into a triplet by the two protons on the fluoromethyl group. Moreover, the resonance is highly upfield (−229 ppm), due to the attachment of the fluoromethyl group to the positively charged tellurium atom (see Supporting Information). ■ FLUOROMETHYL TRANSFER TO OXYGEN NUCLEOPHILES Mammalian COMT has served as a paradigm for enzymatic SAM-dependent methyl transfer.21,56−62 Its normal function is to methylate hydroxyl groups on catechols, catecholamines, and other small molecules to prepare biologically active and/or potentially toxic hydroxylated metabolites for elimination. It is in the metabolism of catecholamine especially important neurotransmitters and catechol estrogens.63 Because of its historic role in early studies of the mechanism of SAM- dependent MT, COMT was chosen as an initial model to address the transfer of the fluoromethyl group from FMeTeSAM to dihydroxybenzoic acid (DHBA), one of its known substrates. Using SAM as a methylating agent, COMT methylates DHBA with kcat and kcat/Km values of 1.34 ± 0.08 min−1 and 0.65 ± 0.17 μM−1 min−1, respectively (Table 1). In the presence of TeSAM, kcat is reduced only by a factor of 2, although kcat/Km is reduced by a factor of almost 34. Excitingly, COMT also uses FMeTeSAM to fluoromethylate DHBA, exhibiting kcat and kcat/Km values of 0.281 ± 0.011 min−1 and 0.00819 ± 0.00113 μM−1 min−1, respectively. The kcat value for the reaction is reduced only by a factor of ∼5 from that with SAM, indicating that FMeTeSAM is a robust fluoromethylating agent under saturating conditions. As expected, the substantial increase in the size of tellurium (covalent radius, 135 pm) over sulfur (covalent radius, 103 pm) drives the Km value for TeSAM upward, and the addition of the fluorine atom increases it further, which impacts the kcat/Km values substantially.64 Nevertheless, these Km values are still lower than or on the order of the in vivo concentrations of SAM in most organisms.65−67 ■ FLUOROMETHYL TRANSFER TO NITROGEN NUCLEOPHILES We also assessed whether FMeTeSAM could be used to fluoromethylate nitrogen nucleophiles, among the most abundant targets of SAM-derived methyl groups. Indeed, nitrogen atoms on DNA and RNA bases, phospholipid head groups, lysine, histidine, and arginine side chains on proteins, and various small molecules are methylated by SAM-dependent MTases.5,7,8,14,15,18,19,68 Phenylethanolamine N-methyltransfer- 907 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article ase (PNMT), nicotinamide N-methyltransferase (NNMT), and TrmD were chosen as test cases. PNMT methylates norepinephrine, affording epinephrine, while NNMT methyl- ates nicotinamide, yielding 1-methylnicotinamide.69−76 Nor- epinephrine and epinephrine serve as hormones and neuro- transmitters, while 1-methylnicotinamide is produced predom- inantly in human fat and liver cells, where it has been reported to play a role in obesity and type 2 diabetes.72,77,78 Moreover, NNMT is upregulated in several cancers.79−83 Last, TrmD is involved in tRNA modification.84−86 The kinetic parameters for PNMT using SAM, TeSAM, or FMeTeSAM as the methyl or fluoromethyl donor are displayed in Table 1. When SAM is used as the methyl donor, the enzyme exhibits a kcat of 1.86 ± 0.1 min−1 and a Km of 1.26 ± 0.24 μM. The kcat when using TeSAM as the methyl donor is about 10-fold lower, although Km remains relatively unchanged. Interestingly, no fluoromethyl-containing product is observed in the reaction using FMeTeSAM as the methyl donor. However, a careful analysis of the reaction by mass spectrometry revealed that Te- adenosyl-L-homotellurocysteine (TeHCys) is produced in a time-dependent manner, suggesting that methyl transfer does indeed take place (Figure S3). It is known that the attachment of fluoromethyl groups to aliphatic amines results in hydrolysis with the concomitant release of formaldehyde and fluoride.87 Given the instability of these fluoromethylated products, we also tested molecules containing nitrogen atoms in different electronic environments as substrates for N-fluoromethyla- tion.88 Results of a kinetic analysis of NNMT-catalyzed methylation of 4-dimethylaminopyridine (DMAP) using SAM, TeSAM, or FMeTeSAM as the methyl donor are also displayed in Table 1. In this instance, FMeTeSAM is equivalent to or better than SAM as a methylating agent under saturating conditions. However, the Km for FMeTeSAM is approximately 3.5 times higher than that of SAM, resulting in a kcat/Km that is ∼4-fold lower than that when using SAM as the methyl donor. 4- DMAP was chosen as a substrate for NNMT because the native substrate, nicotinamide (NAM), did not yield detectable levels of the fluoromethylated product (fm-NAM), although the coproduct TeHCys was detected in our LC−MS analysis. The fm-NAM could have been degraded via hydrolysis generating formaldehyde and fluoride. It should be noted that the kcat and kcat/Km of SAM and SAM analogs with NNMT using 4-DMAP as a cosubstrate are approximately ∼1000-fold lower compared to other MTases used in this study. In addition, we studied E. coli TrmD, which methylates N1 of guanine37 (G37) in tRNApro (Figure 3), and found that the enzyme can indeed fluoromethylate G37 using FMeTeSAM (Figure S5). As expected, it also uses TeSAM and SAM to transfer methyl tRNApro. TrmD (15 μM) converts groups to G37 of the substrate (180 μM) to the approximately 54% of fluoromethylated product in 2 min. Figure 3. Methylation/fluoromethylation of tRNA guanine N37 by TrmD using SAM, TeSAM, and FMeTeSAM. ■ FLUOROMETHYL TRANSFER TO SULFUR NUCLEOPHILES Thiopurine methyltransferase (TPMT) was used to investigate fluoromethyl transfer to sulfur nucleophiles. TPMT is involved in the metabolism of thiopurine drugs such as 6-mercaptopur- ine, 6-thioguanine, and azathioprine, which are cytotoxic immunosuppressant compounds used to treat childhood acute lymphoblastic leukemia, inflammatory bowel disease, and rheumatological diseases.89−91 Their methylation by TPMT reduces their cytotoxic effects.92,93 Surprisingly, FMeTeSAM is as good a methylating agent under saturating conditions as SAM, and the Km for FMeTeSAM is nearly the same as that for SAM. TeSAM supports about 58% of the activity exhibited with SAM under saturating conditions (Table 1). ■ FLUOROMETHYL TRANSFER TO CARBON NUCLEOPHILES Several C-methyltransferases were chosen to investigate fluoromethyl transfer to carbon nucleophiles. These C-MTases include SgvM, a phenyl pyruvic acid MTase;94 M. SssI, a cytosine C5 DNA MTase;95,96 and NovO, an MTase involved in the biosynthesis of the antibiotic novobiocin.97−99 Initially, a series of phenyl pyruvic acid analogs bearing various substituents on the phenyl ring were used to study SgvM (Figure S7); these compounds yielded detectable however, none of fluoromethylated products as judged by HRMS. Moreover, some of the compounds did not support methylation by TeSAM or SAM. A similar result was obtained with M. SssI, suggesting that these carbon centers may not be sufficiently nucleophilic to displace the fluoromethyl group from FMeTeSAM (Figure S8). Studies with NovO and CFA synthase were more successful. NovO methylates C8 of the coumarin scaffold during the biosynthesis of novobiocin, an antibiotic that targets bacterial DNA gyrase. NovO has been reported to exhibit a broad substrate scope. Therefore, the commercially available com- pound 1 was used as our test substrate.98 In the absence of NovO, the fluoromethylated product 2 (m/z 285) is not observed (Figure 4B, black trace). By contrast, when NovO is added to the reaction, a product exhibiting m/z 285 is detected at a retention time of 3.8 min (Figure 4B, red trace). The exact mass of the fluoromethylated product 2 (observed m/z 287.0718, calculated m/z 287.0714) was verified by HRMS (Figure S9). Interestingly, a second peak (m/z 283) is observed at a retention time of 3 min (Figure 4B, blue trace), which corresponds to the hydroxymethylated product 3 (Figure 4A) as verified by HRMS (Figure S9). The NovO reaction mechanism involves an active site Arg− His dyad that deprotonates the 7-OH group of the substrate, which activates the substrate for the rate-limiting methyl transfer (Figure S10).99 We believe that the active site-assisted deprotonation combined with the leaving group ability of fluorine in the fluoromethylated product 2 induces the formation of an o-quinone methide (o-QM) intermediate, which can be attacked by a water molecule to give the hydroxymethylated product 3. o-QM has been implicated in the biosynthesis of several families of natural products as well as in the chemical synthesis of bioactive compounds as an important precursor.100 To provide additional evidence for the o-QM intermediate and to further derivatize this species, we tested other types of nucleophiles, as shown in Figure 4C. Reactions with thiol and phosphine nucleophiles proceed smoothly to the corresponding adducts, which were confirmed by HRMS 908 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 4. (A) NovO catalyzes C-fluoromethylation using FMeTeSAM on coumarin 1 to yield an o-QM intermediate, which is trapped by a water molecule. (B) EIC trace (black) of reaction conducted in the absence of NovO, showing no fluoromethylated product 2 (m/z 285); EIC trace (red) of reaction in the presence of NovO and FMeTeSAM, showing a fluoromethylated product 2 (m/z 285); EIC trace (blue) of hydroxymethylated product 3 (m/z 283) from reaction of NovO using FMeTeSAM. (C) Nucleophiles tested in the NovO reaction. Other nucleophiles (triethylamine, ethanolamine, potassium cyanide, sodium iodide) failed to give corresponding adducts. (D) Dienophiles tested with the o-QM intermediate. (E) Fluoromethylation of dOPG 11 catalyzed by CFA synthase using FMeTeSAM as the methyl donor. The reaction gave two products, including a fluorocyclopropane species 12 and a terminal fluoroalkene 13 via rearrangement. (Figure S9, panels 1, 4, and 5). Amino nucleophiles, including methylamine and dimethylamine (Figure S9, panels 2 and 3, respectively), give only low amounts of adducts due to their lower nucleophilicities. The o-QM is known to react with dienophiles in a Diels−Alder fashion.101,102 Therefore, we assessed whether the o-QM species generated via fluoromethy- lation reacts with ethyl vinyl ether and tetrahydropyran, as shown in Figure 4D. The corresponding adducts were detected and verified by HRMS as shown in Figure S9 (panels 6 &7). CFA synthase was employed to ascertain whether a less activated carbon nucleophile could be fluoromethylated using FMeTeSAM, as shown in Figure 4E. CFA synthase catalyzes the SAM-dependent cyclopropanation of isolated fatty acid olefinic bonds in membrane phospholipids, yielding SAH and a proton as secondary products.103,104 As shown in Figures S11−S13, E. coli CFA synthase can use FMeTeSAM to catalyze cyclo- propanation on a 1,2-dioleoyl-sn-glycero-3-phospho-(1′-rac- glycerol) (dOPG) substrate 11, yielding a fluorinated cyclo- propane phospholipid product 12, which was observed by HRMS and confirmed using MS/MS and 19F-NMR. Interest- ingly, two peaks with different retention times (4.60 and 5.00 min) but identical m/z values and mass spectra are observed by electrospray ionization in negative mode (ESI−) (Figure S11), both of which correspond to the addition of 1 CHF (32.0062 Da) moiety to dOPG (m/z 805.5429) substrate 11. However, the peaks display distinct MS/MS spectra (Figure S12). The peak at 4.6 min contains a fragment ion from the loss of HF, observed both in ESI− and in ESI positive mode (ESI+), while the peak at 5.00 min does not. This behavior implies that the two peaks correspond to different compounds. Based on this data, we identified these two compounds as 12 and 13 (Figure 4E). Compound 12 is the expected fluorinated cyclopropane the identity of which was product, while compound 13, confirmed by LC−MS/MS and 19F-NMR (Figures S11−S13), presumably arises from a 1,2-hydride shift after methyl transfer. A 1,2-hydride shift after methyl transfer to an isolated alkene to stabilize the secondary carbocation intermediate has been observed previously in the biosynthesis of tuberculostearic 909 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 5. (A) OxaC and (B) DnrK are applied for the late-stage fluoromethylation. acid.105 Taken together, our studies demonstrate the versatility of the FMeTeSAM as a methyl donor by showing that it can be used to fluoromethylate unactivated carbon nucleophiles. ■ LATE-STAGE FLUOROMETHYLATION ON NATURAL PRODUCTS The complicated chemical structures of natural products often render them difficult to modify in a selective way. However, late- stage derivatization strategies may provide an efficient means for the diversification of natural products. Therefore, the enzymes OxaC and DnrK were used to demonstrate that they can use FMeTeSAM to transfer fluoromethyl groups onto more complex scaffolds. OxaC catalyzes the penultimate step in the biosynthesis of oxaline, while DnrK catalyzes the penultimate step in the biosynthesis of daunorubicin (Figure 5A,B). Oxaline is a fungal alkaloid that exhibits anticancer activity in vitro, arresting the cell cycle in M phase by inhibition of tubulin polymerization.106 By contrast, daunorubicin is used as a chemotherapeutic agent to treat various types of leukemias and Kaposi’s sarcoma.107 End point assay analysis shows that both oxaline and daunorubicin are fluoromethylated by OxaC (Figure 5A) and DnrK (Figure 5B), respectively, with detection and verification by UPLC−MS and HRMS (Figures S12 and S13). Using FMeTeSAM as fluoromethylating reagent, 5 μM OxaC converts approximately 90% of an 0.15 mM substrate to product in 1 min, 910 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article and 50 μM DnrK converts ∼99% of an 0.2 mM substrate to product in 1 min. This work shows that FMeTeSAM can serve as a regioselective fluoromethylating reagent on complex natural products, highlighting its potential use in late-stage derivatiza- tion processes. ■ SUMMARY AND CONCLUSIONS In this work, we reported the design and synthesis of FMeTeSAM, the only known stable and isolable SAM analog that bears a fluoromethyl group. Indeed, we showed that several SAM-dependent methyltransferases can use FMeTeSAM to transfer fluoromethyl groups to biologically relevant nucleo- philes, such as O-, N-, S-, and some nucleophilic carbon atoms. The kinetic properties of the enzymes when using FMeTeSAM were robust; in most instances where kinetic parameters were determined, the kcat of the reaction was within a factor of 2 of in the presence of SAM, although the Km value, that understandably, due to the increased size of the tellurium atom, increased by up to a factor of 10, driving the second-order −1) downward. rate constants for some of the reactions (kcat·Km Although the fluoromethylation of O-, S-, and C- atoms gave stable and isolable products, the fluoromethylation of N- atoms often gave unstable adducts that presumably were hydrolyzed to formaldehyde and fluoride ion. However, some N- atoms as constituents of aromatic heterocycles did indeed support the formation of stable N-fluoromethylated products. OxaC and DnrK, two MTases that are involved in the biosynthesis of oxaline and daunorubicin, were active in the fluoromethylation of natural product scaffolds, demonstrating the potential application for late-stage fluoromethylation on more compli- cated molecules. FMeTeSAM could potentially be utilized in biocatalytic strategies to derivatize natural products or synthetic scaffolds whose structures are similar to those of the parent natural product, thereby generating novel fluorinated leads in a regioselective manner. ■ ASSOCIATED CONTENT sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscentsci.2c01385. Kinetic plots, high-resolution liquid chromatography− mass spectrometry (LC−MS) spectra and chromato- grams, reaction schemes, HPLC chromatograms (Figures S1−S17, Schemes S1−S5, Tables S1−S2), DNA and protein sequences, experimental procedures for cloning, protein expression and purification, and kinetic and enzyme assays, LC−MS quantification conditions, chemical synthetic procedures, and NMR spectra (PDF) ■ AUTHOR INFORMATION Corresponding Authors Squire J. Booker − Department of Chemistry, Department of Biochemistry and Molecular Biology, and Howard Hughes Medical Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, United States; orcid.org/0000-0002-7211-5937; Email: squire@ psu.edu Bo Wang − Department of Chemistry and Howard Hughes Medical Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, United States; orcid.org/0000-0002-0381-3686; Email: bzw10@ psu.edu Authors Syam Sundar Neti − Department of Chemistry and Howard Hughes Medical Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, United States; orcid.org/0000-0003-3343-8642 David F. Iwig − Department of Chemistry and Howard Hughes Medical Institute, The Pennsylvania State University, University Park, Pennsylvania 16802, United States; orcid.org/0000-0001-8394-184X Elizabeth L. Onderko − Department of Chemistry, The Pennsylvania State University, University Park, Pennsylvania orcid.org/0009-0003-5329-3468 16802, United States; Complete contact information is available at: https://pubs.acs.org/10.1021/acscentsci.2c01385 Funding This work was supported by NIH (GM-122595 and AI-160172 to S.J.B.) and the Eberly Family Distinguished Chair in Science (to S.J.B.). S.J.B. is an investigator of the Howard Hughes Medical Institute. Notes The authors declare no competing financial interest. ■ ACKNOWLEDGMENTS We thank the NMR facility at Penn State for help with collecting NMR spectra. We also thank the Thompson group at the University of Massachusetts Medical School, USA, the Mueller group at the University of Freiburg, Germany, the Burley group at the University of Strathclyde, Glasgow, UK, and the Sherman group at the University of Michigan, Ann Arbor, USA for providing expression plasmids for NNMT, SgvM, NovO, and OxaC. This article is subject to HHMI’s Open Access to Publications policy. HHMI lab heads have previously granted a non- exclusive CC BY 4.0 license to the public and a sublicensable license to HHMI in their research articles. Pursuant to those licenses, the author-accepted manuscript of this article can be made freely available under a CC BY 4.0 license immediately upon publication. ■ ABBREVIATIONS COMT, catechol O-methyltransferase; DHBA, dihydroxyben- zoic acid; DMAP, 4-dimethylaminopyridine; FMeTeSAM, Te- adenosyl-L-(fluoromethyl)homotellurocysteine; HRMS, high- resolution mass spectrometry; LC−MS, liquid chromatogra- phy−mass spectrometry; MS, mass spectrometry; MTase, methyltransferase; MTA, methylthioadenosine; NMR, nuclear magnetic resonance; NNMT, nicotinamide N-methyltransfer- ase; PNMT, phenylethanolamine N-methyltransferase; SAH, S- adenosylhomocysteine; SAM, S-adenosylmethionine; SeSAM, Se-adenosylselenomethionine; TeHCys, Te-adenosyl-L-homo- tellurocysteine; TPMT, thiopurine methyltransferase; TeSAM, Te-adenosyltelluromethionine ■ REFERENCES (1) Greenberg, M. V. C.; Bourc’his, D. The diverse roles of DNA methylation in mammalian development and disease. Nat. Rev. Mol. Cell Biol. 2019, 20 (10), 590−607. (2) Lu, S. C.; Mato, J. M. S-Adenosylmethionine in liver health, injury, and cancer. Physiol. Rev. 2012, 92, 1515−1542. 911 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article (3) Cantoni, G. L. Biological methylation: selected aspects. Annu. Rev. Biochem. 1975, 44, 435−51. (4) Ouyang, Y.; Wu, Q.; Li, J.; Sun, S.; Sun, S. S-adenosylmethionine: A metabolite critical to the regulation of autophagy. Cell Prolif 2020, 53 (11), e12891. (5) Di Blasi, R.; Blyuss, O.; Timms, J. F.; Conole, D.; Ceroni, F.; Whitwell, H. J. Non-Histone Protein Methylation: Biological Significance and Bioengineering Potential. ACS Chem. Biol. 2021, 16 (2), 238−250. (6) Delaunay, S.; Frye, M. RNA modifications regulating cell fate in cancer. Nat. Cell Biol. 2019, 21 (5), 552−559. (7) Towns, W. L.; Begley, T. J. Transfer RNA methytransferases and their corresponding modifications in budding yeast and humans: activities, predications, and potential roles in human health. DNA Cell Biol. 2012, 31 (4), 434−54. (8) Takusagawa, F.; Fujioka, M.; Spies, A.; Schowen, R. L. S- adenosylmethionine (AdoMet)-dependent methyltransferases. In Comprehensive Biological Catalysis; Sinnott, M., Ed.; Academic Press: New York, 1998; Vol. 1, pp 1−30. (9) Chiang, P. K.; Gordon, R. K.; Tal, J.; Zeng, G. C.; Doctor, B. P.; Pardhasaradhi, K.; McCann, P. P. S-Adenosylmethionine and methylation. FASEB J. 1996, 10 (4), 471−80. (10) Bauerle, M. R.; Schwalm, E. L.; Booker, S. J. Mechanistic diversity of radical S-adenosylmethionine (SAM)-dependent methylation. J. Biol. Chem. 2015, 290, 3995−4002. (11) Zhang, Q.; van der Donk, W. A.; Liu, W. Radical-mediated enzymatic methylation: A tale of two SAMS. Acc. Chem. Res. 2012, 45, 555−564. (12) Schubert, H. L.; Blumenthal, R. M.; Cheng, X. Many paths to methyltransfer: a chronicle of convergence. Trends Biochem. Sci. 2003, 28 (6), 329−35. (13) Sufrin, J. R.; Finckbeiner, S.; Oliver, C. M. Marine-derived metabolites of S-adenosylmethionine as templates for new anti- infectives. Mar Drugs 2009, 7 (3), 401−34. (14) Clarke, S. G. Protein methylation at the surface and buried deep: thinking outside the histone box. Trends Biochem. Sci. 2013, 38 (5), 243−52. (15) Liscombe, D. K.; Louie, G. V.; Noel, J. P. Architectures, mechanisms and molecular evolution of natural product methyltrans- ferases. Nat. Prod Rep 2012, 29 (10), 1238−50. (16) Luo, M. Chemical and Biochemical Perspectives of Protein Lysine Methylation. Chem. Rev. 2018, 118 (14), 6656−6705. (17) Borchardt, R. T. S-Adenosyl-L-methionine-dependent macro- molecule methyltransferases: potential targets for the design of chemotherapeutic agents. J. Med. Chem. 1980, 23 (4), 347−57. (18) Fuhrmann, J.; Clancy, K. W.; Thompson, P. R. Chemical biology of protein arginine modifications in epigenetic regulation. Chem. Rev. 2015, 115 (11), 5413−61. (19) Hori, H. Methylated nucleosides in tRNA and tRNA methyltransferases. Front Genet 2014, 5, 144. (20) Woodard, R. W.; Tsai, M.-D.; Floss, H. G.; Crooks, P. A.; Coward, J. K. Sterochemical course of the transmethylation catalyzed by catechol O-methyltransferase. J. Biol. Chem. 1980, 255, 9124−9127. (21) Hegazi, M. F.; Borchardt, R. T.; Schowen, R. L. a-Deuterium and carbon-13 isotope effects for methyl transfer catalyzed by catechol-O- methyl-transferase. SN2-like transition state. J. Am. Chem. Soc. 1979, 101, 4359−4365. (22) Parks, J. M.; Johs, A.; Podar, M.; Bridou, R.; Hurt, R. A., Jr; Smith, S. D.; Tomanicek, S. J.; Qian, Y.; Brown, S. D.; Brandt, C. C.; Palumbo, A. V.; Smith, J. C.; Wall, J. D.; Elias, D. A.; Liang, L. The genetic basis for bacterial mercury methylation. Science 2013, 339 (6125), 1332−5. (23) Jarrett, J. T.; Hoover, D. M.; Ludwig, M. L.; Matthews, R. G. The mechanism of adenosylmethionine-dependent activation of methio- nine synthase: a rapid kinetic analysis of intermediates in reductive methylation of Cob(II)alamin enzyme. Biochemistry-Us 1998, 37 (36), 12649−12658. (24) Booker, S. J. Anaerobic functionalization of unactivated C−H bonds. Curr. Opin. Chem. Biol. 2009, 13, 58−73. (25) Chasteen, T. G.; Bentley, R. Biomethylation of selenium and tellurium: microorganisms and plants. Chem. Rev. 2003, 103 (1), 1−25. (26) Bentley, R.; Chasteen, T. G. Microbial methylation of metalloids: arsenic, antimony, and bismuth. Microbiol Mol. Biol. Rev. 2002, 66 (2), 250−71. (27) Bennett, M. R.; Shepherd, S. A.; Cronin, V. A.; Micklefield, J. Recent advances in methyltransferase biocatalysis. Curr. Opin Chem. Biol. 2017, 37, 97−106. (28) Barreiro, E. J.; Kummerle, A. E.; Fraga, C. A. The methylation effect in medicinal chemistry. Chem. Rev. 2011, 111 (9), 5215−46. (29) Struck, A. W.; Thompson, M. L.; Wong, L. S.; Micklefield, J. S- adenosyl-methionine-dependent methyltransferases: highly versatile enzymes in biocatalysis, biosynthesis and other biotechnological applications. Chembiochem 2012, 13 (18), 2642−55. (30) Huber, T. D.; Johnson, B. R.; Zhang, J.; Thorson, J. S. AdoMet analog synthesis and utilization: current state of the art. Curr. Opin Biotechnol 2016, 42, 189−197. (31) Peng, J.; Liao, C.; Bauer, C.; Seebeck, F. P. Fluorinated S- Adenosylmethionine as a Reagent for Enzyme-Catalyzed Fluorome- thylation. Angew. Chem., Int. Ed. Engl. 2021, 60 (52), 27178−27183. (32) Neumann, C. N.; Ritter, T. Late-Stage Fluorination: Fancy Novelty or Useful Tool? Angew. Chem. Int. Edit 2015, 54 (11), 3216− 3221. (33) Muller, K.; Faeh, C.; Diederich, F. Fluorine in pharmaceuticals: looking beyond intuition. Science 2007, 317 (5846), 1881−6. (34) Wang, J.; Sanchez-Rosello, M.; Acena, J. L.; del Pozo, C.; Sorochinsky, A. E.; Fustero, S.; Soloshonok, V. A.; Liu, H. Fluorine in pharmaceutical industry: fluorine-containing drugs introduced to the market in the last decade (2001−2011). Chem. Rev. 2014, 114 (4), 2432−506. (35) Inoue, M.; Sumii, Y.; Shibata, N. Contribution of Organofluorine Compounds to Pharmaceuticals. Acs Omega 2020, 5 (19), 10633− 10640. (36) Gillis, E. P.; Eastman, K. J.; Hill, M. D.; Donnelly, D. J.; Meanwell, N. A. Applications of Fluorine in Medicinal Chemistry. J. Med. Chem. 2015, 58 (21), 8315−8359. (37) Vulpetti, A.; Dalvit, C. Fluorine local environment: from screening to drug design. Drug Discov Today 2012, 17 (15−16), 890−7. (38) Thuronyi, B. W.; Chang, M. C. Y. Synthetic Biology Approaches to Fluorinated Polyketides. Acc. Chem. Res. 2015, 48 (3), 584−592. (39) Walker, M. C.; Thuronyi, B. W.; Charkoudian, L. K.; Lowry, B.; Khosla, C.; Chang, M. C. Expanding the fluorine chemistry of living systems using engineered polyketide synthase pathways. Science 2013, 341 (6150), 1089−94. (40) Walker, M. C.; Chang, M. C. Natural and engineered biosynthesis of fluorinated natural products. Chem. Soc. Rev. 2014, 43 (18), 6527−36. (41) Sirirungruang, S.; Ad, O.; Privalsky, T. M.; Ramesh, S.; Sax, J. L.; Dong, H.; Baidoo, E. E. K.; Amer, B.; Khosla, C.; Chang, M. C. Y. Engineering site-selective incorporation of fluorine into polyketides. Nat. Chem. Biol. 2022, 18 (8), 886−893. (42) Eustaquio, A. S.; O’Hagan, D.; Moore, B. S. Engineering fluorinase expression in Salinispora fluorometabolite production: tropica Yields Fluorosalinosporamide. J. Nat. Prod 2010, 73 (3), 378−82. (43) Braun, M. G.; Doyle, A. G. Palladium-catalyzed allylic C-H fluorination. J. Am. Chem. Soc. 2013, 135 (35), 12990−3. (44) Hull, K. L.; Anani, W. Q.; Sanford, M. S. Palladium-catalyzed fluorination of carbon-hydrogen bonds. J. Am. Chem. Soc. 2006, 128 (22), 7134−5. (45) Miao, J.; Yang, K.; Kurek, M.; Ge, H. Palladium-Catalyzed Site- Selective Fluorination of Unactivated C(sp(3))-H Bonds. Org. Lett. 2015, 17 (15), 3738−41. (46) Cho, E. J.; Senecal, T. D.; Kinzel, T.; Zhang, Y.; Watson, D. A.; Buchwald, S. L. The palladium-catalyzed trifluoromethylation of aryl chlorides. Science 2010, 328 (5986), 1679−81. (47) Roque, J. B.; Kuroda, Y.; Gottemann, L. T.; Sarpong, R. Deconstructive fluorination of cyclic amines by carbon-carbon cleavage. Science 2018, 361 (6398), 171−174. 912 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article (48) Liang, T.; Neumann, C. N.; Ritter, T. Introduction of fluorine and fluorine-containing functional groups. Angew. Chem., Int. Ed. Engl. 2013, 52 (32), 8214−64. (49) Iwig, D. F.; Booker, S. J. Insight into the polar reactivity of the onium chalcogen analogues of S-adenosyl-L-methionine. Biochemistry- Us 2004, 43 (42), 13496−509. (50) Parks, L. W.; Schlenk, F. The stability and hydrolysis of S- adenosylmethionine; isolation of S-ribosylmethionine. J. Biol. Chem. 1958, 230 (1), 295−305. (51) Parks, L. W.; Schlenk, F. Formation of alpha-amino-gamma- butyrolactone from S-adenosylmethionine. Arch. Biochem. Biophys. 1958, 75 (1), 291−2. (52) Borchardt, R. T. Mechanism of alkaline hydrolysis of S-adenosyl- L-methionine and related sulfonium nucleosides. J. Am. Chem. Soc. 1979, 101, 458. (53) Wu, S.-E.; Huskey, W. P.; Borchardt, R. T.; Schowen, R. L. Chiral instability at sulfur of S-adenosylmethionine. Biochemistry-Us 1983, 22, 2828−2832. (54) Hoffman, J. L. Chromatographic analysis of the chiral and covalent instability of S-adenosyl-L-methionine. Biochemistry-Us 1986, 25, 4444−4449. (55) Iwig, D. F.; Grippe, A. T.; McIntyre, T. A.; Booker, S. J. Isotope and elemental effects indicate a rate-limiting methyl transfer as the initial step in the reaction catalyzed by Escherichia coli cyclopropane fatty acid synthase. Biochemistry-Us 2004, 43 (42), 13510−13524. (56) Zhang, J. Y.; Klinman, J. P. Enzymatic Methyl Transfer: Role of an Active Site Residue in Generating Active Site Compaction That Correlates with Catalytic Efficiency. J. Am. Chem. Soc. 2011, 133 (43), 17134−17137. (57) Czarnota, S.; Johannissen, L. O.; Baxter, N. J.; Rummel, F.; Wilson, A. L.; Cliff, M. J.; Levy, C. W.; Scrutton, N. S.; Waltho, J. P.; Hay, S. Equatorial Active Site Compaction and Electrostatic Reorganization in Catechol-O-methyltransferase. ACS Catal. 2019, 9 (5), 4394−4401. (58) Kahn, K.; Bruice, T. C. Transition-state and ground-state structures and their interaction with the active-site residues in catechol O-methyltransferase. J. Am. Chem. Soc. 2000, 122, 46−51. (59) Mannisto, P. T.; Kaakkola, S. Catechol-O-methyltransferase (COMT): biochemistry, molecular biology, pharmacology, and clinical efficacy of the new selective COMT inhibitors. Pharmacol Rev. 1999, 51 (4), 593−628. (60) Rutherford, K.; Le Trong, I.; Stenkamp, R. E.; Parson, W. W. Crystal structures of human 108V and 108M catechol O-methyl- transferase. J. Mol. Biol. 2008, 380 (1), 120−30. (61) Ruggiero, G. D.; Williams, I. H.; Roca, M.; Moliner, V.; Tunon, I. QM/MM determination of kinetic isotope effects for COMT-catalyzed methyl transfer does not support compression hypothesis. J. Am. Chem. Soc. 2004, 126 (28), 8634−5. (62) Coward, J. K.; Slisz, E. P.; Wu, Y.-H. F. Kinetic Studies on Catechol 0-Methyltransferase. Product Inhibition and the Nature of the Catechol Binding Site. Biochemistry 1973, 12, 2291−2297. (63) Qayyum, A.; Zai, C. C.; Hirata, Y.; Tiwari, A. K.; Cheema, S.; Nowrouzi, B.; Beitchman, J. H.; Kennedy, J. L. The Role of the Catechol-o-methyltransferase (COMT) Gene Val158Met in Aggres- sive Behavior, A Review of Genetic Studies. Curr. Neuropharmacol 2015, 13 (6), 802−814. (64) Housecroft, C. E.; Sharpe, A. G. Inorganic Chemistry, 2nd ed.; Pearson Education Limited: Harlow, England, 2005. (65) Bennett, B. D.; Kimball, E. H.; Gao, M.; Osterhout, R.; Van Dien, S. J.; Rabinowitz, J. D. Absolute metabolite concentrations and implied enzyme active site occupancy in Escherichia coli. Nat. Chem. Biol. 2009, 5 (8), 593−9. (66) Goldberg, B.; Rattendi, D.; Lloyd, D.; Yarlett, N.; Bacchi, C. J. Kinetics of methionine transport and metabolism by Trypanosoma brucei brucei and Trypanosoma brucei rhodesiense. Arch. Biochem. Biophys. 2000, 377 (1), 49−57. (67) Chiba, P.; Wallner, C.; Kaiser, E. S-adenosylmethionine metabolism in HL-60 cells: effect of cell cycle and differentiation. Biochim. Biophys. Acta 1988, 971 (1), 38−45. (68) Song, H.; van der Velden, N. S.; Shiran, S. L.; Bleiziffer, P.; Zach, C.; Sieber, R.; Imani, A. S.; Krausbeck, F.; Aebi, M.; Freeman, M. F.; Riniker, S.; Kunzler, M.; Naismith, J. H. A molecular mechanism for the enzymatic methylation of nitrogen atoms within peptide bonds. Sci. Adv. 2018, 4 (8), eaat2720. (69) Georgieva, P.; Wu, Q.; McLeish, M. J.; Himo, F. The reaction mechanism of phenylethanolamine N-methyltransferase: a density functional theory study. Biochim. Biophys. Acta 2009, 1794 (12), 1831− 7. (70) Hou, Q. Q.; Wang, J. H.; Gao, J.; Liu, Y. J.; Liu, C. B. QM/MM studies on the catalytic mechanism of phenylethanolamine N- methyltransferase. Biochim. Biophys. Acta 2012, 1824 (4), 533−41. (71) Wu, Q.; McLeish, M. J. Kinetic and pH studies on human phenylethanolamine N-methyltransferase. Arch. Biochem. Biophys. 2013, 539 (1), 1−8. (72) Ziegler, M. G.; Bao, X.; Kennedy, B. P.; Joyner, A.; Enns, R. Location, development, control, and function of extraadrenal phenyl- ethanolamine N-methyltransferase. Ann. N.Y. Acad. Sci. 2002, 971, 76− 82. (73) Alston, T. A.; Abeles, R. H. Substrate specificity of nicotinamide methyltransferase isolated from porcine liver. Arch. Biochem. Biophys. 1988, 260 (2), 601−8. (74) Pissios, P. Nicotinamide N-Methyltransferase: More Than a Vitamin B3 Clearance Enzyme. Trends Endocrinol Metab 2017, 28 (5), 340−353. (75) Cantoni, G. L. Methylation of nicotinamide with soluble enzyme system from rat liver. J. Biol. Chem. 1951, 189 (1), 203−16. (76) Peng, Y.; Sartini, D.; Pozzi, V.; Wilk, D.; Emanuelli, M.; Yee, V. C. Structural basis of substrate recognition in human nicotinamide N- methyltransferase. Biochemistry-Us 2011, 50 (36), 7800−8. (77) Liu, J. R.; Deng, Z. H.; Zhu, X. J.; Zeng, Y. R.; Guan, X. X.; Li, J. H. Roles of Nicotinamide N-Methyltransferase in Obesity and Type 2 Diabetes. Biomed Res. Int. 2021, 2021, 9924314. (78) Kitahama, K.; Denoroy, L.; Goldstein, M.; Jouvet, M.; Pearson, J. Immunohistochemistry of tyrosine hydroxylase and phenylethanol- amine N-methyltransferase in the human brain stem: description of adrenergic perikarya and characterization of longitudinal catecholami- nergic pathways. Neuroscience 1988, 25 (1), 97−111. (79) Xu, J.; Moatamed, F.; Caldwell, J. S.; Walker, J. R.; Kraiem, Z.; Taki, K.; Brent, G. A.; Hershman, J. M. Enhanced expression of nicotinamide N-methyltransferase in human papillary thyroid carcino- ma cells. J. Clin Endocrinol Metab 2003, 88 (10), 4990−6. (80) Roessler, M.; Rollinger, W.; Palme, S.; Hagmann, M. L.; Berndt, P.; Engel, A. M.; Schneidinger, B.; Pfeffer, M.; Andres, H.; Karl, J.; Bodenmuller, H.; Ruschoff, J.; Henkel, T.; Rohr, G.; Rossol, S.; Rosch, W.; Langen, H.; Zolg, W.; Tacke, M. Identification of nicotinamide N- methyltransferase as a novel serum tumor marker for colorectal cancer. Clin. Cancer Res. 2005, 11 (18), 6550−7. (81) Sartini, D.; Muzzonigro, G.; Milanese, G.; Pierella, F.; Rossi, V.; Emanuelli, M. Identification of nicotinamide N-methyltransferase as a novel tumor marker for renal clear cell carcinoma. J. Urol 2006, 176 (5), 2248−54. (82) Tomida, M.; Mikami, I.; Takeuchi, S.; Nishimura, H.; Akiyama, H. Serum levels of nicotinamide N-methyltransferase in patients with lung cancer. J. Cancer Res. Clin Oncol 2009, 135 (9), 1223−9. (83) Sartini, D.; Santarelli, A.; Rossi, V.; Goteri, G.; Rubini, C.; Ciavarella, D.; Lo Muzio, L.; Emanuelli, M. Nicotinamide N- methyltransferase upregulation inversely correlates with lymph node metastasis in oral squamous cell carcinoma. Mol. Med. 2007, 13 (7−8), 415−21. (84) Hou, Y. M.; Matsubara, R.; Takase, R.; Masuda, I.; Sulkowska, J. I. TrmD: A Methyl Transferase for tRNA Methylation With m(1)G37. Enzymes 2017, 41, 89−115. (85) Gamper, H. B.; Masuda, I.; Frenkel-Morgenstern, M.; Hou, Y. M. Maintenance of protein synthesis reading frame by EF-P and m(1)G37- tRNA. Nat. Commun. 2015, 6, 7226. (86) Bjork, G. R.; Wikstrom, P. M.; Bystrom, A. S. Prevention of translational frameshifting by the modified nucleoside 1-methylguano- sine. Science 1989, 244 (4907), 986−9. 913 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914 ACS Central Science http://pubs.acs.org/journal/acscii Research Article (87) Shen, L.; Song, C. X.; He, C.; Zhang, Y. Mechanism and function of oxidative reversal of DNA and RNA methylation. Annu. Rev. Biochem. 2014, 83, 585−614. (88) Prakash, G. K.; Ledneczki, I.; Chacko, S.; Olah, G. A. Direct electrophilic monofluoromethylation. Org. Lett. 2008, 10 (4), 557−60. (89) Peng, Y.; Feng, Q.; Wilk, D.; Adjei, A. A.; Salavaggione, O. E.; Weinshilboum, R. M.; Yee, V. C. Structural basis of substrate recognition in thiopurine s-methyltransferase. Biochemistry 2008, 47 (23), 6216−25. (90) Scheuermann, T. H.; Lolis, E.; Hodsdon, M. E. Tertiary structure of thiopurine methyltransferase from Pseudomonas syringae, a bacterial orthologue of a polymorphic, durg-metabolizing enzyme. J. Mol. Biol. 2003, 333, 573−585. (91) Wu, H.; Horton, J. R.; Battaile, K.; Allali-Hassani, A.; Martin, F.; Zeng, H.; Loppnau, P.; Vedadi, M.; Bochkarev, A.; Plotnikov, A. N.; Cheng, X. Structural basis of allele variation of human thiopurine-S- methyltransferase. Proteins 2007, 67 (1), 198−208. (92) Lennard, L. The clinical pharmacology of 6-mercaptopurine. Eur. J. Clin Pharmacol 1992, 43 (4), 329−39. (93) Coulthard, S.; Hogarth, L. The thiopurines: an update. Invest New Drugs 2005, 23 (6), 523−32. (94) Sommer-Kamann, C.; Fries, A.; Mordhorst, S.; Andexer, J. N.; Muller, M. Asymmetric C-Alkylation by the S-Adenosylmethionine- Dependent Methyltransferase SgvM. Angew. Chem., Int. Ed. Engl. 2017, 56 (14), 4033−4036. (95) Buryanov, Y.; Shevchuk, T. The use of prokaryotic DNA methyltransferases as experimental and analytical tools in modern biology. Anal. Biochem. 2005, 338 (1), 1−11. (96) Slaska-Kiss, K.; Zsibrita, N.; Koncz, M.; Albert, P.; Csabradi, A.; Szentes, S.; Kiss, A. Lowering DNA binding affinity of SssI DNA methyltransferase does not enhance the specificity of targeted DNA methylation in E. coli. Sci. Rep 2021, 11 (1), 15226. (97) Steffensky, M.; Muhlenweg, A.; Wang, Z. X.; Li, S. M.; Heide, L. Identification of the novobiocin biosynthetic gene cluster of Streptomyces spheroides NCIB 11891. Antimicrob. Agents Chemother. 2000, 44 (5), 1214−22. (98) Pacholec, M.; Tao, J.; Walsh, C. T. CouO and NovO: C- methyltransferases for tailoring the aminocoumarin scaffold in coumermycin and novobiocin antibiotic biosynthesis. Biochemistry-Us 2005, 44 (45), 14969−76. (99) Sadler, J. C.; Chung, C. H.; Mosley, J. E.; Burley, G. A.; Humphreys, L. D. Structural and Functional Basis of C-Methylation of Coumarin Scaffolds by NovO. ACS Chem. Biol. 2017, 12 (2), 374−379. (100) Doyon, T. J.; Perkins, J. C.; Baker Dockrey, S. A.; Romero, E. O.; Skinner, K. C.; Zimmerman, P. M.; Narayan, A. R. H. Chemoenzymatic o-Quinone Methide Formation. J. Am. Chem. Soc. 2019, 141 (51), 20269−20277. (101) Yang, B. C.; Gao, S. H. Recent advances in the application of Diels-Alder reactions involving o-quinodimethanes, aza-o-quinone methides and o-quinone methides in natural product total synthesis. Chem. Soc. Rev. 2018, 47 (21), 7926−7953. (102) Bai, W. J.; David, J. G.; Feng, Z. G.; Weaver, M. G.; Wu, K. L.; Pettus, T. R. The domestication of ortho-quinone methides. Acc. Chem. Res. 2014, 47 (12), 3655−64. (103) Grogan, D. W.; Cronan, J. E., Jr Cyclopropane ring formation in membrane lipids of bacteria. Microbiol Mol. Biol. Rev. 1997, 61 (4), 429−441. (104) Cronan, J. E., Jr.; Nunn, W. D.; Batchelor, J. G. Studies on the biosynthesis of cyclopropane fatty acids in Escherichia coli. Biochim. Biophys. Acta 1974, 348 (1), 63−75. (105) Lederer, E. some problems concerning biological C-alkylation reactions and phytosterol biosynthesis. Q. Rev. Chem. Soc. 1969, 23, 453−481. (106) Koizumi, Y.; Arai, M.; Tomoda, H.; Omura, S. Oxaline, a fungal alkaloid, arrests the cell cycle in M phase by inhibition of tubulin polymerization. Biochim. Biophys. Acta 2004, 1693 (1), 47−55. (107) Jansson, A.; Koskiniemi, H.; Mantsala, P.; Niemi, J.; Schneider, G. Crystal structure of a ternary complex of DnrK, a methyltransferase in daunorubicin biosynthesis, with bound products. J. Biol. Chem. 2004, 279 (39), 41149−56. 914 https://doi.org/10.1021/acscentsci.2c01385 ACS Cent. Sci. 2023, 9, 905−914
10.1016_j.celrep.2023.112408	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Cell Rep. Author manuscript; available in PMC 2023 August 22. Published in final edited form as: Cell Rep. 2023 May 30; 42(5): 112408. doi:10.1016/j.celrep.2023.112408. The nuclear Argonaute HRDE-1 directs target gene re- localization and shuttles to nuage to promote small RNA- mediated inherited silencing Yue-He Ding1, Humberto J. Ochoa1, Takao Ishidate1, Masaki Shirayama1, Craig C. Mello1,2,3,* 1RNA Therapeutics Institute, University of Massachusetts Medical School, Worcester, MA 01605, USA 2Howard Hughes Medical Institute, Worcester, MA 01605, USA 3Lead contact SUMMARY Argonaute/small RNA pathways and heterochromatin work together to propagate transgenerational gene silencing, but the mechanisms behind their interaction are not well understood. Here, we show that induction of heterochromatin silencing in C. elegans by RNAi or by artificially tethering pathway components to target RNA causes co-localization of target alleles in pachytene nuclei. Tethering the nuclear Argonaute WAGO-9/HRDE-1 induces heterochromatin formation and independently induces small RNA amplification. Consistent with this finding, HRDE-1, while predominantly nuclear, also localizes to peri-nuclear nuage domains, where amplification is thought to occur. Tethering a heterochromatin-silencing factor, NRDE-2, induces heterochromatin formation, which subsequently causes de novo synthesis of HRDE-1 guide RNAs. HRDE-1 then acts to further amplify small RNAs that load on downstream Argonautes. These findings suggest that HRDE-1 plays a dual role, acting upstream to initiate heterochromatin silencing and downstream to stimulate a new cycle of small RNA amplification, thus establishing a self-enforcing mechanism that propagates gene silencing to future generations. In brief Ding and colleagues investigate inherited silencing in C. elegans. They demonstrate that the nuclear Argonaute HRDE-1 induces subnuclear-co-localization of target genes in heterochromatin. Heterochromatin formation subsequently triggers de novo HRDE-1 guide RNA loading. Finally, This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Correspondence: [email protected]. AUTHOR CONTRIBUTIONS Conceptualization, Y.-H.D. and C.C.M.; investigation, Y.-H.D. and H.J.O.; methodology, Y.-H.D., H.J.O., T.I., and C.C.M.; data analysis, Y.-H.D.; writing – review & editing, Y.-H.D. and C.C.M.; supervision, C.C.M. SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2023.112408. DECLARATION OF INTERESTS The authors declare no competing interests. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 2 HRDE-1 enters nuage and activates small RNA amplification. Thus, HRDE-1 effects multiple steps of a self-enforcing transgenerational silencing process. Graphical Abstract INTRODUCTION In many animal germlines, small RNA/Argonaute pathways function transgenerationally to install and re-inforce chromatin silencing essential for fertility. For example, in flies, worms, and mammals, members of the PIWI Argonaute family engage genomically encoded small RNAs termed PIWI-interacting RNAs (piRNAs) that silence transposons to maintain genome integrity.1–6 Although the details differ, all transgenerational small RNA silencing pathways studied to date require amplification and engagement of secondary Argonautes.7 Many of the components of the amplification machinery localize prominently in peri-nuclear non-membranous organelles called nuage. However, how the amplification system in nuage communicates with and drives the nuclear events during the initiation and maintenance of transgenerational silencing is not well understood. In C. elegans, transgenerational silencing can be initiated by the PIWI pathway, by the canonical double-stranded RNA (dsRNA)-induced RNAi pathway, or by intronless mRNA.8–11 Inherited silencing is maintained by a family of related downstream worm- specific Argonautes (WAGO Argonautes) guided by small RNAs (22G-RNAs) produced Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 3 by cellular RNA-dependent RNA polymerase. Once established, inherited silencing can be propagated independently of the initiating cues via continuous cycles of WAGO 22G-RNA amplification and transmission of the WAGO Argonautes and their small RNA co-factors to progeny.8,12–14 The nuclear WAGO Argonaute, HRDE-1/WAGO-9, plays a central role in transgenerational silencing in C. elegans.15,16 HRDE-1 is thought to engage nascent transcripts at target loci to induce heterochromatin and transcriptional silencing through the nuclear RNAi pathway.15,17 HRDE-1 promotes the transgenerational silencing of many genes18 and is thought to do so by recruiting chromatin remodeling factors, including the nucleosome remodeling and deacetylase complex (NuRD) and histone methyltransferases (e.g., MET-2, SET-25, SET-32).9,18,19 The nuclear RNAi pathway is also required for the spreading of secondary small RNAs from piRNA target sites.14,20 Transgenerational silencing requires a series of events that are thought to occur in the nuage, nucleus, and cytoplasm. Because all of these events are essential for the cycle of inherited silencing, their order has been difficult to determine. For example, it is not known whether the nuclear Argonaute HRDE-1 directly triggers RdRP recruitment and amplification of small RNAs or whether it must first induce heterochromatin at its targets to elicit small RNA amplification. Here, we use the phage lambda N (λN)-boxB tethering system21–25 to recruit—i.e., tether—HRDE-1 or the nuclear silencing factor NRDE-2 to a reporter mRNA. In principle, tethering enables initiation of silencing in the absence of upstream initiators such as piRNAs or dsRNA and, with appropriate genetic tests, can be used to order events in the pathway. We show that tethering either HRDE-1 or NRDE-2 can induce a complete silencing response, including small RNA amplification and transgenerational silencing that persists even after the λN-fusion protein is crossed from the strain. Tethering NRDE-2 initiates chromatin silencing through nrde-4 and independently of hrde-1 but requires hrde-1 for small RNA amplification. By contrast, tethering HRDE-1 stimulates chromatin silencing through NRDE-2 and NRDE-4 but can elicit small RNA amplification independently of both these chromatin-silencing factors. Mutations that block HRDE-1 from binding small RNA disarm silencing and cause HRDE-1 to become cytoplasmic, but tethering HRDE-1 in these mutants nevertheless initiates a strong silencing response that requires small RNA amplification proximal to the tether site. The small RNA amplification machinery is recruited to the tether site by sequences in the N-terminal half of HRDE-1 (the N-terminal domain [NTD]). Like full-length HRDE-1 protein, HRDE-1 NTD co-localizes with MUT-16 in Mutator foci, subdomains of cytoplasmic nuage where the small RNA amplification machinery resides.26 Our findings suggest that HRDE-1 lies at a nexus in the silencing pathway, shuttling from the nucleus to the nuage and back, to coordinate the nuclear and cytoplasmic events of transgenerational silencing. RESULTS HRDE-1 and NRDE-2 tethering induce transgenerational silencing To order events in inherited silencing, we sought to uncouple initiation and maintenance of silencing. To do this, we used the phage λN-boxB tethering system to recruit nuclear silencing factors HRDE-1 or NRDE-2 to a target reporter that is robustly expressed in the Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 4 germline (Figure 1A). We hypothesized that if artificial recruitment of a silencing factor mimics a physiological event, then it should elicit a silencing response that is independent of upstream factors but depends on known downstream factors. For example, directly tethering a chromatin factor should, in principle, induce silencing without requiring machinery necessary to amplify the small RNAs that would normally guide the chromatin silencing machinery to the appropriate targets. Using CRISPR, we inserted an in-frame λN coding sequence at the 5′ end of the endogenous hrde-1 or nrde-2 loci (see STAR Methods). Both fusion genes were fully functional, based on their ability to mediate piRNA silencing (Figure S1A). Moreover, both strains exhibited wild-type patterns and distributions of endogenous small RNA species (Figure S1B).We then tested whether the λN fusions could induce heritable silencing of a reporter gene whose 3′ UTR contains λN-binding sites (i.e., boxB elements) (Figures 1A–1D). Both λN::HRDE-1 and λN::NRDE-2 induced silencing of the reporter beginning at the initial heterozygous generation (Figures 1C and 1D). Notably, silencing of the reporter persisted in subsequent generations after genetically segregating away the λN-fusion alleles (Figures 1E and S1C; data not shown). As expected, inherited silencing (after segregating the λN-fusion alleles) required known components of the transgenerational RNA silencing pathway, including HRDE-1, the small RNA amplification factors RDE-3/MUT-2 and MUT-16,6,27,28 and the nuclear silencing factors NRDE-2 and NRDE-49,29 (Figures 1E– 1G, S1C, and S1H; data not shown). Moreover, λN::HRDE-1 and λN::NRDE-2 tethering induced trimethylation of histone H3 lysine 9 (H3K9me3; Figures 2A and 2B) and reduced both reporter mRNA and pre-mRNA levels (Figures 2C and 2D), consistent with the role of H3K9me3 in transcriptional silencing.15 Thus, artificially recruiting HRDE-1 or NRDE-2 to a target locus was sufficient to initiate the full cycle of events required for inherited silencing, including small RNA amplification and heterochromatin formation. Having established that tethering induces inherited silencing that depends genetically on known components of the RNA silencing pathway, we asked which factors were required for silencing when the tethered protein was continuously present. For example, because the λN-boxB interaction recruits HRDE-1 and NRDE-2 independently of a guide RNA, we reasoned that the small RNA amplification machinery should be unnecessary when nuclear silencing factors are tethered to the reporter. Consistent with this idea, we found that λN::NRDE-2 silenced the reporter in the absence of rde-3, mut-16, and hrde-1 (Figures 2E, S2A–S2C) but failed to silence it in the absence of nrde-4 (Figures 2E and S2D). These results suggest that NRDE-2 acts downstream of HRDE-1 and upstream of NRDE-4 in nuclear silencing. In wild-type animals without tethering, inherited silencing requires nuclear chromatin silencing factors (e.g., nrde-2 and nrde-4) and nuage-localized factors (e.g., rde-3 and mut-16; Figure 1G), indicating that these pathways function together, possibly sequentially, to propagate inherited silencing. In contrast, when λN::HRDE-1 was tethered to the reporter, we found that leaving either pathway intact was sufficient to maintain silencing, as monitored by GFP epifluorescence. For example, silencing of the reporter GFP was maintained independently of nrde-2, nrde-4, or rde-3 and only partly required mut-16 activity (Figures 2F and S2E–S2H). To completely prevent silencing, it was necessary to Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 5 simultaneously mutate components of both the small RNA amplification machinery (rde-3 or mut-16) and components of the chromatin nuclear silencing machinery (nrde-2 or nrde-4) (Figures 2F and S2E–S2K). HRDE-1 tethering in wild-type worms reduced the unspliced pre-mRNA reporter level by 2-fold and the spliced RNA level by 100-fold, as measured by quantitative PCR (qPCR) (Figure 2C). For unknown reasons, nrde-2 mutants exhibited a 4-fold increase in reporter pre-mRNA both with and without HRDE-1 tethering (Figure 2C) but exhibited discordant effects on spliced reporter RNA levels. Removing nrde-2 activity in animals without tethering had little effect on spliced reporter mRNA levels (a slight 1.2-fold increase) compared with wild type, but removing nrde-2 activity in the context of tethering caused spliced RNA levels to increase (compared with levels in wild-type HRDE-1-tethered animals), reaching levels of approximately 40% of wild-type mRNA levels. It is important to note that the qPCR assay cannot distinguish mRNA from template RNA being silenced, as template RNAs derive from spliced RNAs. Moreover, the high levels of spliced RNA in λN::HRDE-1;nrde-2 worms correlate with a marked accumulation of reporter RNA localized in nuage (via RNA fluorescence in situ hybridization [FISH], shown below). Thus, the accumulated spliced RNA likely reflects template RNA engaged in amplifying the small RNA silencing signal, perhaps to compensate for the loss of heterochromatin silencing. Further study is needed to understand the effects of nrde-2 mutants on pre-mRNA levels, such as whether increased pre-mRNA levels in nrde-2 mutants reflect processing defects.30 Nevertheless, in the nrde-2 background, HRDE-1 tethering reduces mRNA and pre-mRNA levels by 2-to 3-fold, suggesting that tethered HRDE-1 can exert effects on both mRNA and pre-mRNA levels independently of NRDE-2. Taken together, our findings suggest that HRDE-1 functions twice during inherited silencing—upstream of nuclear silencing to recruit NRDE-2 and NRDE-4 and again downstream of these factors to induce small RNA amplification and post-transcriptional clearance of mRNA. While these events likely occur sequentially and thus depend on each other during the normal course of inherited silencing,31 tethering HRDE-1 initiates both modes of silencing independently, either of which is sufficient to prevent reporter GFP expression. HRDE-1 acts downstream of NRDE-2 to promote small RNA amplification The above findings indicate that HRDE-1 can initiate inherited silencing independently of nrde-2 and nrde-4, while NRDE-2 requires both nrde-4 and hrde-1. A likely explanation for these findings is that heterochromatin silencing directed by NRDE-2 and NRDE-4 induces the de novo synthesis of small RNAs that engage HRDE-1 and that HRDE-1 can further amplify these small RNAs to propagate silencing to offspring. Indeed, whereas we detected very few small RNAs targeting the reporter in the absence of tethering (Figure 3A), λN::NRDE-2 induced small RNA accumulation that required nrde-4, rde-3, and hrde-1 (Figures 3B–3E). These findings suggest that NRDE-2 tethering induces silencing and heterochromatin formation through NRDE-4 (Figures 2B and 2D) and that downstream events (e.g., heterochromatin formation itself or other NRDE-4-dependent events) act through RDE-3 and HRDE-1 to induce small RNA amplification. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 6 λN::HRDE-1 tethering induced abundant small RNA accumulation that was independent of nrde-2 and nrde-4 (Figures 3F, 3G, and S3A). However, interestingly, both the distribution of small RNAs and their levels of accumulation along the target mRNA were dramatically altered in the nrde mutants. Small RNA levels were markedly increased adjacent to the boxB sites and were diminished on the gfp coding sequences (Figures 3F, 3G, S3A, and S3B). Small RNAs targeting the reporter were greatly reduced by mutations in rde-3 and mut-16, as expected, (Figures 3H and 3I). Interestingly, however, a low level of small RNAs persisted directly adjacent to the boxB sites when λN::HRDE-1 was tethered in the absence of rde-3 but not in the absence of mut-16 (Figure 3H). This result is consistent with the observation that tethering of λN::HRDE-1 can bypass rde-3 but cannot fully bypass mut-16 (Figure 2F). When outcrossed to a hrde-1(+) background to segregate away λN::HRDE-1, the reporter remained silent for at least 13 generations, with no change in penetrance. Moreover, we observed only a slight reduction in small RNA levels primarily in regions juxtaposed to the boxB hairpins (Figure 3J). In contrast, when outcrossed to a hrde-1 null background, the reporter was fully de-silenced, and small RNAs were absent (Figure 3K). As expected, the maintenance of silencing, and of small RNA levels, also required rde-3(+) and mut-16(+) (Figures 3L and 3M). Taken together, these findings suggest that heterochromatin formation at the target locus induces de novo transcription and loading of small RNAs onto the nuclear Argonaute HRDE-1. HRDE-1, in turn, further promotes small RNA amplification and then functions again, perhaps in the next life cycle, to reinitiate heterochromatin silencing (see discussion). HRDE-1 guide RNA loading is not required for small RNA amplification The finding that λN::HRDE-1 can direct chromatin silencing in rde-3 and mut-16 mutants, which are defective in small RNA amplification, suggests that the unloaded Argonaute can direct chromatin silencing when tethered. To further test this idea, we monitored silencing (1) by λN::HRDE-1 in an hrde-2 mutant, which is defective in HRDE-1 small RNA loading13 and (2) by a λN::HRDE-1(Y669E) mutant, predicted by structural work to be defective in guide RNA binding (Figure S5B).32 In both cases, tethering completely silenced the boxB reporter as monitored by GFP fluorescence (Figure 4C and S4D) and by quantitative reverse transcription PCR (qRT-PCR) of the mRNA (Figure 4D). For unknown reasons, compromising nuclear silencing by hrde1-(Y669E) caused elevated pre-mRNA levels as measured by qRT-PCR (Figure 4D), similar to nrde-2 mutants. As expected, the hrde-1(Y669E) mutant was defective in silencing a piRNA reporter (Figure S4A) and showed a collapse of small RNAs resembling that in hrde-1(null) (Figures S4B and S4C). However, in these mutant contexts, loss of rde-3 alone was sufficient to completely de-silence the reporter (Figures S4E and 4C), suggesting that in the absence of guide RNA loading, HRDE-1 fails to engage the NRDE heterochromatin machinery. Deep sequencing revealed an abundant accumulation of rde-3-dependent small RNAs targeting the boxB reporter in λN::HRDE-1(Y669E) animals (Figures 4E and 4F). Notably, the pattern and levels of small RNA accumulation induced by λN::HRDE-1(Y669E) resembled those observed when wild-type λN::HRDE-1 is tethered in a nrde-2 mutant (compare Figures 4E–3G)—i.e., resulting in increased levels of small RNAs targeting sequences adjacent to Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 7 the boxB sites and reduced levels targeting GFP sequences. Taken together, these results suggest that tethering of unloaded HRDE-1 can induce local small RNA amplification and silencing but that tethered HRDE-1 must be loaded with small RNAs to induce chromatin silencing, which is in turn required for small RNA targeting to spread into the 5′ sequences of the target mRNA. HRDE-1 promotes small RNA amplification through its NTD We next attempted to dissect functional domains of HRDE-1 required for small RNA amplification. We used CRISPR to make a series of λN::hrde-1 truncation mutants (Figure 5A). These studies identified the N-terminal half (herein the NTD) as the minimal fragment of HRDE-1 that could fully silence the reporter. The NTD and the remaining C-terminal domain (CTD) truncations of HRDE-1 are predicted by I-TASSER33 to fold into self-contained globular structures, with subdomains similar to those identified in atomic resolution studies on humanAgo234 (Figures 5B, S5A, and S5B). As expected, in the absence of tethering, hrde-1(NTD) and hrde-1(CTD) alleles failed to silence a piRNA sensor (Figure S4A). Silencing by λN::NTD required rde-3 but not nrde-2 (Figures 5C and S5C), and deep sequencing revealed that λN::NTD induces abundant rde-3-dependent small RNAs targeting the boxB reporter (Figures 5D and 5E). Truncations that failed to silence the reporter did not trigger small RNA generation (Figure S5D). The small RNA pattern induced by λN::NTD resembled the patterns caused by λN::HRDE-1 in nrde-2 mutants or by λN::HRDE-1(Y669E)—i.e., dramatically increased levels of small RNAs proximal to the boxB sites and reduced levels of small RNAs targeting GFP sequences. Interestingly, the magnitude of small RNA accumulation induced by λN::NTD at the boxB sites was ~4-fold greater than that induced by either λN::HRDE-1 in nrde-2 mutants or by λN::HRDE-1(Y669E) (compare Figure 5D with Figures 3G and 4E). These results suggest that the NTD of HRDE-1 robustly recruits the small-RNA amplification machinery to the target and promotes silencing that is independent of the NRDE-2 nuclear silencing pathway. HRDE-1 tethering promotes accumulation of poly-UG-modified target fragments During RNA silencing in worms, truncated target RNAs are converted into templates for small RNA production via the RDE-3-dependent addition of poly-UG tails.27 We therefore used a qPCR assay27 to detect poly-UG additions to reporter RNA in the absence of a λN fusion or in worms expressing λN::HRDE-1, λN::NTD, or λN::HRDE-1(Y669E) (Figures 5G and 5H). Priming from an endogenous UGUG motif in the reporter 3′ UTR serves as a control for the presence of full-length mRNA. This analysis revealed that faster-migrating, poly-UG-modified RNAs accumulated in strains where silencing was active. In wild-type λN::HRDE-1 worms, poly-UG-modified RNAs were most robustly detected at truncations within the GFP sequences (Figures 5G and 5H). As expected, only full-length mRNA was detected in rde-3 mutants, confirming that RDE-3 is absolutely required for poly-UG RNA accumulation. Notably, mutation of nrde-2 or tethering the NTD or Y669E mutants shifted poly-UG addition toward the 3′ end of the reporter, close to the boxB elements (Figures 5G and 5H). These results suggest that HRDE-1 tethering induces RDE-3-dependent poly- UG modification of truncation products that are generated near the tethering sites and Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 8 that nuclear silencing promotes the induction of additional truncations far away from the tethering sites that likely support the 5′ spread of small RNA amplification. To further analyze changes in target RNA caused by tethering, we used qRT-PCR. Surprisingly, whereas tethering wild-type λN::HRDE-1 reduced the reporter pre-mRNA by 50% and mRNA by 99% (Figure 2C), λN::NTD increased the reporter pre-mRNA by ~2.5-fold and reduced the mRNA by ~40% (Figure 5F). This result was surprising given that GFP fluorescence was undetectable in λN::NTD worms (Figures 5C and S5C) and suggested that the accumulating species in λN::NTD animals might reflect the accumulation of nearly full-length pUG RNA. Functional HRDE-1 RNA-induced silencing complex (RISC) is not required parentally for transmission of silencing to offspring We next asked if λN::NTD can initiate inherited silencing. To do this, we first established reporter silencing by tethering λN::NTD in otherwise wild-type worms. We then crossed to a reporter strain homozygous for a hrde-1 null allele to generate animals heterozygous for the tethering construct. Finally, we crossed these λN::NTD/null heterozygotes (either as males or hermaphrodites) to a hrde-1(+) reporter strain, resulting in two types of cross progeny—λN::NTD/+ or null/+ heterozygotes. Remarkably, although the λN::NTD/null parents lacked a functional HRDE-1 RISC, they nevertheless robustly transmitted silencing to the next generation (Figures S7A and S7B). As expected, HRDE-1(+) was required in the inheriting generation for silencing to occur (Buckley et al.15 and Figure 1F). Since the NTD fails to establish heterochromatin upon tethering and cannot directly form a RISC complex, these findings suggest that parentally established heterochromatin and HRDE-1 RISC are not required in gametes for inheritance, a finding consistent with previous work in which hrde-1 homozygous mutant hermaphrodites were shown to transmit silencing to their heterozygous progeny.15 Rather, in the parental generation, the tethered NTD can stimulate amplification of small RNAs that likely engage with other Argonautes to propagate silencing to offspring (see discussion). HRDE-1 localizes to Mutator foci HRDE-1 localization is primarily nuclear15; however, template formation and small RNA amplification are thought to occur in domains of peri-nuclear nuage termed Mutator foci, where several components of the small RNA amplification machinery localize.26–28 To examine whether HRDE-1 localizes in Mutator foci, we expressed GFP::HRDE-1 (without tethering) in worms that also express either mCherry::GLH-1, which localizes broadly within nuage, or MUT-16::mCherry, which localizes prominently in Mutator foci. GFP::HRDE-1 co-localized to a subset of peri-nuclear mCherry::GLH-1 foci, especially in association with late pachytene germ nuclei (Figures 6A and S6A). Moreover, the GFP::HRDE-1 foci only partially overlapped with mCherry::GLH-1 foci, suggesting that the HRDE-1+ foci occupy subdomains of larger GLH-1+ nuage, reminiscent of Mutator foci. Indeed, GFP::HRDE-1 foci coincided almost perfectly with MUT-16::mCherry foci (Figure 6B). Similarly, GFP::HRDE-1(NTD) co-localized with GLH-1::mCherry and mCherry::MUT-16 foci (Figures 6C, 6D, and S6B). Taken together, these findings suggest Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 9 that HRDE-1 localizes via its NTD to Mutator foci, where it functions to promote small RNA amplification. Silencing by dsRNA or tethering causes target genes to co-localize To understand how HRDE-1 and nuclear silencing regulate their target genes and RNAs, we performed RNA and DNA FISH studies to visualize the boxB reporter mRNA and DNA. In the absence of silencing, reporter RNA foci were detected throughout the germline cytoplasm (Figures 6E and S6C). In addition, we observed prominent RNA signals in the majority (~70%) of pachytene nuclei (most nuclei, 57%, exhibited at least two closely paired nuclear dots, while the remainder exhibited a single dot; Figures 6E and 6I). The positions of these nuclear signals adjacent to DAPI-stained chromosomes suggests that they correspond to sites of transcription on the paired sister chromatids within the axial loops of synapsed meiotic homologs. Silencing, induced either by exposure to dsRNA targeting the reporter or by tethering λN::HRDE-1, eliminated cytoplasmic reporter RNA signal and greatly reduced the nuclear signal (Figures 6F, 6L, and S6C). More than 80% of the pachytene nuclei with visible RNA signal exhibited a single nuclear focus (Figures 6F, 6L, 6I, and 6O). The changes in nuclear RNA signal induced by silencing correlated with changes in the reporter DNA FISH signal. In the absence of silencing, we observed a pair of nuclear DNA FISH signals in approximately 50% of pachytene nuclei that have visible DNA signal (Figures 6P and 6T), while in the presence of silencing, we observed a single focus of DNA FISH signal in approximately 90% of pachytene nuclei with visible DNA signal (Figures 6Q, 6J, 6T, and S6E). These results suggest that nuclear silencing mediated by HRDE-1 causes the target alleles to become merged from predominantly paired DNA FISH signals into a single focus containing all 4 silenced alleles. Mutations that disarm nuclear silencing cause target RNA to accumulate in nuage subdomains that resemble Mutator foci We next examined how mutations that disarm only the nuclear silencing pathway impact RNA and DNA localization after RNAi or tethering. To do this, we performed RNA and DNA FISH on λN::NTD worms and on nrde-2 mutants. In these mutants, where nuclear silencing is disarmed, we found that nuclear RNA and DNA FISH signals resembled the nuclear signals observed in wild-type animals in the absence of silencing: predominantly two foci of RNA and DNA FISH signals detected in each background (Figures 6M, 6N, 6I, 6J, 6O, 6T, and S6D). In contrast, however, the cytoplasmic RNA FISH signals were dramatically altered. While RNA signal was absent from the bulk cytoplasm throughout the gonad, consistent with cytoplasmic post-transcriptional silencing, we noticed pronounced accumulation of reporter RNA signals in multiple peri-nuclear foci surrounding pachytene nuclei. Co-staining experiments with GFP::GLH-1 or MUT-16::GFP revealed that these RNA foci coincide with most of the nuage subdomains that express MUT-16::GFP (Figures 6G, 6H, 6M, and 6N). The accumulation of target RNA in the MUT-16 foci required RDE-3(+) activity (Figure S6F), suggesting that these RNA signals may correspond to RdRP templates engaged in small RNA amplification. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 10 MUT-16 promotes the nuclear localization of GFP::HRDE-1 but not its nuage localization MUT-16 is required for the co-localization of small RNA amplification factors within Mutator foci.26,28,35 We therefore wondered if MUT-16 is also required for the co- localization of HRDE-1 in Mutator foci. To answer this question, we introduced a null allele of mut-16 into worms expressing both GFP::HRDE-1 and mCherry::GLH-1. As shown previously,24 we found that MUT-16 activity is required for the nuclear localization of HRDE-1 (Figures 7A and 7B). MUT-16 was not, however, required for the localization of GFP::HRDE-1 to nuage (Figures 7A and 7B). The localization of GFP::HRDE-1 in nuage appeared more obvious in mut-16 mutants, but the levels of GFP::HRDE-1 within nuage and the approximate numbers of foci appeared similar with or without mut-16 activity (Figures 7A and 7B). Finally, the localization of MUT-16 itself to nuage was not disrupted in hrde-1 mutants (data not shown), thus HRDE-1 and MUT-16 localize within a nuage subdomain (or domains) independently of each other. DISCUSSION In many eukaryotes, the installation and maintenance of chromatin silencing is coupled to Argonaute small RNA pathways that promote transmission to offspring. Here, we have explored the role of a nuclear Argonaute HRDE-1 in coordinating transgenerational silencing in the C. elegans germline. In addition to its known role in directing heterochromatin silencing downstream of RNAi13,15 and Piwi Argonaute silencing,8,9,14 our tethering studies have shown that HRDE-1 is also de novo loaded with small RNA, downstream of heterochromatin silencing, enabling it to prime a new round of small RNA amplification within nuage (Figure 7C, model). The nuclear silencing events that depend on HRDE-1 cause the target alleles to co-localize into a single focus of DNA FISH signal (Figures 6P–6S and S6E). Presumably, the heterochromatinized alleles within this focus are transcribed at low levels to produce template RNA that feeds transgenerational silencing; indeed, the continued expression of the target locus after heterochromatin induction is a conserved feature of co-transcriptional small RNA silencing.36 Consistent with this idea, the inactivation of heterochromatin silencing caused target alleles to remain separated and increased the levels of the nuclear- and nuage-localized RNA signals as measured by RNA FISH. The failure to engage nuclear silencing did not de-silence protein expression in the context of our tethering studies nor indeed in previously published studies on nuclear-silencing mutants when an RNAi trigger is present.13,15 Instead, our RNA FISH studies suggest that unabated transcription of the target gene feeds increased levels of target RNA localization in nuage (also noted in a recent study by Ouyang et al.37) and that small RNA levels also increase dramatically to compensate and silence mRNA expression. Taken together, our findings suggest that when the nuclear heterochromatin pathways are inactive, the target mRNA is silenced by a combination of cytoplasmic clearance or trapping in the P granule. In the yeast S. pombe, the RNAi-induced transcriptional silencing complex (RITS), which includes an RdRP and a nuclear Argonaute AGO1p, resides in heterochromatin. A previous study showed that tethering of AGO1p to RNA via a boxB reporter system, similar to the Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 11 one used here, was sufficient to recruit the RITS complex, induce small RNA amplification, and drive reporter silencing25. HRDE-1 associates with NRDE-2 and components of the nucleosome re-modeling and deacetylase NuRD complex to establish heterochromatin silencing.15,18,38 How heterochromatin leads to de novo programming of HRDE-1 is nevertheless unknown. In C. elegans, the RdRP EGO-1 has been shown to associate with germline chromatin,39,40 and several of our findings would be consistent with a cycle of nuclear small RNA transcription and de novo HRDE-1 loading within heterochromatin. Such a mechanism could explain why tethering NRDE-2 in the absence of HRDE-1 initiates heterochromatin silencing but not small RNA amplification (Figures 2E and 3E). Perhaps after a nuclear cycle of HRDE-1 loading, the protein exits the nucleus along with nascent target/template RNA to further amplify small RNA production. Consistent with this idea, we have shown that the N-terminal half of HRDE-1 is sufficient to stimulate small RNA amplification and loading and that both the NTD and full-length HRDE-1 (as well as target RNA) localize within a specialized nuage domain known as Mutator foci. Mutator foci accumulate poly-UG-modified templates derived from target RNA27 and are thought to serve in the amplification of small RNA signals that are propagated to offspring. Thus, our findings suggest that HRDE-1 shuttles out of the nucleus to nuage to promote small RNA amplification. A mutant HRDE-1 protein incapable of binding guide RNA was sufficient (when tethered) to induce silencing that transmits to offspring via either the sperm or the egg (Figures S7A and S7B). Thus, as previously reported,15 a functional HRDE-1 RISC is not required in gametes for transgenerational silencing but is required in offspring to renew silencing for another generation (Buckley et al.15 and Figure 1F). In the parental germline, Mutator foci likely serve as locations where HRDE-1 and other upstream Argonautes trigger the expansion of small RNAs that are loaded onto downstream WAGO Argonautes, including the two prominent nuage-localized Argonautes WAGO-18 and WAGO-4.41 Consistent with this idea, silencing induced by λN::HRDE-1(Y669E) was partially dependent on wago-1 (75% de-silenced, N = 32, and Figure S4G). Taken together, our findings suggest that heterochromatin renews small RNA silencing (and vice versa) during each germline life cycle. For example, small RNAs guide heterochromatin formation in the zygote, and heterochromatin then propagates silencing before feeding back into the de novo synthesis of guide RNAs that load onto HRDE-1. HRDE-1 promotes expansion of small RNAs that are then transmitted to offspring through HRDE-1 and other WAGOs to re-establish heterochromatin. Heterochromatin then, in turn, transcribes RNA that forms templates for RdRP-dependent amplification, renewing the cycle. Consistent with these ideas, neither pathway, small RNA or heterochromatin alone, is sufficient to stably transmit silencing signals for multiple generations8,9,13,15 (Figures S7C–S7F). Given the similarities between the worm and yeast mechanisms—and by extension, the intriguing relationships between long non-coding RNAs and chromatin modifiers in flies and mammals7—feedforward RNA-chromatin circuits that amplify and maintain silencing across cell divisions or generations will likely be a common feature of gene regulation in eukaryotes. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Limitations of the study Page 12 In this study, we use an artificial mechanism to recruit RNA silencing factors to their targets. Recruiting, factors via the λN/boxB system may elicit non-physiological mechanisms that block gene expression. For example, tethering factors to the reporter UTR could prevent proper recruitment of translation-initiation machinery or 3′ end processing factors. Transcripts that are not processed properly (for example, unspliced mRNA11) could trigger default recruitment of the same RNA silencing factors that mediate physiological silencing in response to bona fide Argonaute-guided silencing. To control for such possibilities, we used genetics to dissect the nature of the silencing pathways induced by tethering and found that tethering different factors elicited different genetic dependencies for silencing. For example, λN::NRDE-2 required nrde-4(+) activity for silencing but λN::HRDE-1 tethering did not. We have controlled for possible artifacts by initiating parallel studies on untethered factors and by using a combination of genetics, microscopy, and RNA-expression profiling. Together, these studies give us high confidence that tethering, in these instances, has faithfully replicated actual physiological steps in silencing. STAR★METHODS RESOURCE AVAILABILITY Lead contact—Further information and requests for resources and materials should be directed to and will be fulfilled by the lead contact, Craig Mello ([email protected]). Materials availability—All materials generated in this study are available from the lead contact without restrictions. Data and code availability—Original small-RNA sequencing datasets are publicly available in NCBI under the accession number BioProject: PRJNA874806. This study did not generate any new code, but the scripts used in the study are available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS All the strains used in this study were derived from C. elegans Bristol N2 (CGC) and cultured on nematode growth media (NGM) plates with E. coli OP5043 or E. coli HT115 for RNAi experiments. Strains used in this study were generated by CRISPR-cas9 method or Cross (see Table S1 for details). METHOD DETAILS CRISPR-Cas9 genome editing—The Cas9 ribonucleoprotein (RNP) CRISPR strategy44 were used to edit the genome. Plasmid pRF4 containing rol-6 (su-1006) was used as co-injection marker. For short insertions like λN and deletion mutations, synthesized Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 13 single-strand DNAs were used as the donor; for long insertions like GFP, mCherry, and 5xBoxB, the annealed PCR products were used instead. The gRNA and donor sequences were listed in Table S2. The BoxB reporter strain was constructed based on a single copy insertion of Ppie-1:GFP::his-58:unc-54UTR (WM701). The 5xBoxB sequence amplified from a previously published strain JMC00222 was inserted before the unc-54 UTR. Live worm fluorescent image—Young adult worms were transferred to glass slide in M9 buffer with 0.4mM Tetramisole. Epifluorescence and differential interference contrast (DIC) microscopy were performed on a Zeiss Axio Imager M2 Microscope and images were processed with ZEN Microscopy Software (Zeiss). Confocal images were taken by a Andor Dragonfly Spinning Disk confocal microscope. Confocal images were processed with Imaris Microscopy Image Analysis Software. Quantifying reporter RNA using qPCR—Young adult worms were collected and washed with M9 for three times and ddH2O once. Total RNA was extracted with TRIZOL and treated with DNase I to remove DNA contamination. First strand cDNA was synthesized by Superscript IV with random hexamers. Quantitative PCR was performed on a Quant studio 5 Real-time PCR machine together with Fast SYBR Green Master Mix. Actin was used as internal reference (primer set S5265 and S527). Primer set of oYD826 and oYD827 were used for reporter. All primers used were listed in Table S2. CHIP-qPCR—A traditional worm CHIP method45 was applied to the young adult worm samples. Anti H3K9me3 antibody (Upstate 07523) and CHIP grade IgA/G magnetic beads were used for the immunoprecipitation. During elution, RNase A and Protease K were used to remove RNA and proteins. For qPCR, actin was used as internal reference. All primers used were listed in Table S2. Small RNA cloning and data analysis—Small RNA cloning was conducted as previously reported.6 Synchronized young adult worms were collected and total RNA were purified with Trizol. Two biological repeats were included for each strain. Small RNAs were enriched using a mirVana miRNA isolation kit. Homemade PIR-1 was used to remove the di or triphosphate at the 5′ to generate 5′ monophosphorylated small RNA. Adaptors of 3’ (DA35) and 5’ (DA4) were ligated to the small RNA by T4 RNA ligase 2 (NEB) and T4 ligase 1 (NEB) sequentially. Reverse transcription was performed with SuperScript III and RT primer (DA5). After PCR amplification, productions around 150 bp were separated by 12% SDS-PAGE and equally mixed. Libraries were sequenced on a NextSeq 550 sequencer with the illumina NextSeq 500/550 high output kit in 75bp single-end sequencing mod. Reads were trimmed by cutadapt and mapped using Bowtie2.42 For small RNAs mapped to the reporter, total reads with length longer than 16 nt were used to normalized between samples. Plots were generated by R and R studio. QUANTIFICATION AND STATISTICAL ANALYSIS To determine the genes with increased or decreased antisense small RNAs (Figures S4B and S4C), small RNAs were cloned and sequenced as described above with two biological Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 14 repeats for each strain. DEseq2 package in R was used to find out genes with 2-fold decrease of antisense small RNA (p value ≤ 0.05) in hrde-1(null) or hrde-1(Y669E) compared to WT. Structure prediction—The 3D structure of HRDE-1 was predicted by I-TASSER online server33 with default setting. HRDE-1 structure was aligned with hAgo2 by PyMOL46 and its domains were annotated based on the alignment. pUG RNA analysis—As previously reported,27 total RNAs were extracted with Trizol. SuperScript IV was used to generate the first strand DNA with reverse transcription primer oYD1001. A pair of outer primers (oYD998 and oYD1002) were used for the first round PCR amplification with Taq DNA polymerase. After 100-fold dilution, another round of PCR was performed with a pair of inner primers (oYD256 and oYD1003). PCR products were analyzed by 1.5% agarose gels. DNA bands were purified, cloned with TOPO TA Cloning Kit and sent for sanger sequencing. gsa-1 served as a control for pUG PCR analysis. RNA FISH—Worms at young adult stage were dissected in Happy Buffer (81mM HEPES pH 6.9, 42mM NaCl, 5mM KCl, 2mM MgCl2, 1mM EGTA) (From personal correspondence with James Priess). Dissected gonads were transferred to poly-lysine treated dish with 80 μl of Happy Buffer and fixed by adding equal volume of 5% formaldehyde in PBST (PBS+0.1% Tween 20) for 30 min. After one wash with PBST, gonads were treated with PBST-Triton (PBST+0.1% Triton) for 10 min, washed with PBST again and emerged in 70% ethanol for 30 min to overnight. Before hybridization, samples were washed with fresh wash buffer (2xSSC +10% formamide) for 5 min hybridization was performed at 37°C for 18 h to overnight in hybridization buffer (900 μl Stellaris RNA FISH Hybridization Buffer+ 100ul formamide) with 10 pmol RNA FISH probes. Samples were washed with wash buffer, once quick wash, one wash for 30 min at 37°C and two quick washes. Mounting medium with DAPI was added to preserve the signal. Confocal images were taken with an Andor Dragonfly Spinning Disk confocal microscope and processed with Fusion and Imaris. DNA FISH—Same to RNA FISH, gonads were dissected, fixed and washed with PBST and treated with 70% ethanol. Then, samples were washed with wash buffer three times, one at room temperature for 5 min, one at 95°C for 3 min, and one at 60°C for 20 min. Hybridization was performed in hybridization buffer (700 μl Stellaris RNA FISH Hybridization Buffer +300 μl formamide + primary probes (final 10 pmol) + detection probe (final 10 pmol)) at 95°C for 5 min and then transferred to 37°C for 3 h to overnight. After hybridization, samples were wash with 2xSSC for 20 min at 60°C, and then 2xSSCT (2xSSC +0.3% Triton X-100) for 5 min at 60°C and another 20 min at 60°C. After another wash with 2xSSCT for 5 min at room temperature, samples were preserved in the mounting medium with DAPI. Confocal images were taken with an Andor Dragonfly Spinning Disk confocal microscope and processed with Fusion and Imaris. Primary probes of DNA FISH were picked from the oligo lists generated by OligoMiner.47 RNAi experiments—Synchronous L1 worms of the reporter strain were plated on NGM plates for 48 h. Then the worms were collected and washed with M9. About 100 worms Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 15 were plated on every IPTG plate with the gfp RNAi food. After 24 h, worms were dissected for the FISH experiment. RNA FISH and DNA FISH were performed as described above. Supplementary Material Refer to Web version on PubMed Central for supplementary material. ACKNOWLEDGMENTS We thank members of Mello and Ambros labs for discussions; James Priess (Fred Hutchinson Cancer Center) for sharing the receipt of happy buffer and imaging experiences; Weifeng Gu (University of California, Riverside) for providing the PIR-1 protein for small RNA cloning; Ahmet Ozturk for building the small RNA analysis pipeline; Darryl Conte for critical comments and edits on the manuscript; and the RNA Therapeutics Institute for offering the Nextseq 550 sequencing machine. The work was supported by NIH funding (GM058800 and HD078253) to C.C.M. C.C.M. is a Howard Hughes Medical Institute Investigator. REFERENCES 1. Kasschau KD, Fahlgren N, Chapman EJ, Sullivan CM, Cumbie JS, Givan SA, and Carrington JC (2007). Genome-wide profiling and analysis of Arabidopsis siRNAs. PLoS Biol. 5, e57. [PubMed: 17298187] 2. Czech B, Malone CD, Zhou R, Stark A, Schlingeheyde C, Dus M, Perrimon N, Kellis M, Wohlschlegel JA, Sachidanandam R, et al. (2008). An endogenous small interfering RNA pathway in Drosophila. Nature 453, 798–802. [PubMed: 18463631] 3. Ghildiyal M, Seitz H, Horwich MD, Li C, Du T, Lee S, Xu J, Kittler ELW, Zapp ML, Weng Z, and Zamore PD (2008). Endogenous siRNAs derived from transposons and mRNAs in Drosophila somatic cells. Science 320, 1077–1081. [PubMed: 18403677] 4. Tam OH, Aravin AA, Stein P, Girard A, Murchison EP, Cheloufi S, Hodges E, Anger M, Sachidanandam R, Schultz RM, and Hannon GJ (2008). Pseudogene-derived small interfering RNAs regulate gene expression in mouse oocytes. Nature 453, 534–538. [PubMed: 18404147] 5. Watanabe T, Totoki Y, Toyoda A, Kaneda M, Kuramochi-Miyagawa S, Obata Y, Chiba H, Kohara Y, Kono T, Nakano T, et al. (2008). Endogenous siRNAs from naturally formed dsRNAs regulate transcripts in mouse oocytes. Nature 453, 539–543. [PubMed: 18404146] 6. Gu W, Shirayama M, Conte D Jr., Vasale J, Batista PJ, Claycomb JM, Moresco JJ, Youngman EM, Keys J, Stoltz MJ, et al. (2009). Distinct argonaute-mediated 22G-RNA pathways direct genome surveillance in the C. elegans germline. Mol. Cell 36, 231–244. [PubMed: 19800275] 7. Weick E-M, and Miska EA (2014). piRNAs: from biogenesis to function. Development 141, 3458– 3471. [PubMed: 25183868] 8. Shirayama M, Seth M, Lee H-C, Gu W, Ishidate T, Conte D Jr., and Mello CC (2012). piRNAs initiate an epigenetic memory of nonself RNA in the C. elegans germline. Cell 150, 65–77. [PubMed: 22738726] 9. Ashe A, Sapetschnig A, Weick E-M, Mitchell J, Bagijn MP, Cording AC, Doebley A-L, Goldstein LD, Lehrbach NJ, Le Pen J, et al. (2012). piRNAs can trigger a multigenerational epigenetic memory in the germline of C. elegans. Cell 150, 88–99. [PubMed: 22738725] 10. Fire A, Xu S, Montgomery MK, Kostas SA, Driver SE, and Mello CC (1998). Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. nature 391, 806–811. [PubMed: 9486653] 11. Makeyeva YV, Shirayama M, and Mello CC (2021). Cues from mRNA splicing prevent default Argonaute silencing in C. elegans. Dev. Cell 56, 2636–2648.e4. [PubMed: 34547227] 12. Lee HC, Gu W, Shirayama M, Youngman E, Conte D Jr., and Mello CC (2012). C. elegans piRNAs mediate the genome-wide surveillance of germline transcripts. Cell 150, 78–87. 10.1016/ j.cell.2012.06.016. [PubMed: 22738724] 13. Spracklin G, Fields B, Wan G, Becker D, Wallig A, Shukla A, and Kennedy S (2017). The RNAi inheritance machinery of Caenorhabditis elegans. Genetics 206, 1403–1416. [PubMed: 28533440] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 16 14. Sapetschnig A, Sarkies P, Lehrbach NJ, and Miska EA (2015). Tertiary siRNAs mediate paramutation in C. elegans. PLoS Genet. 11, e1005078. 15. Buckley BA, Burkhart KB, Gu SG, Spracklin G, Kershner A, Fritz H, Kimble J, Fire A, and Kennedy S (2012). A nuclear Argonaute promotes multigenerational epigenetic inheritance and germline immortality. Nature 489, 447–451. [PubMed: 22810588] 16. Rechavi O, i-Ze’evi L, Anava S, Goh WSS, Kerk SY, Hannon GJ, and Hobert O (2014). Starvation-induced transgenerational inheritance of small RNAs in C. elegans. Cell 158, 277–287. [PubMed: 25018105] 17. Almeida MV, Andrade-Navarro MA, and Ketting RF (2019). Function and evolution of nematode RNAi pathways. Noncoding. RNA 5, 8. [PubMed: 30650636] 18. Kim H, Ding Y-H, Zhang G, Yan Y-H, Conte D Jr., Dong M-Q, and Mello CC (2021). HDAC1 SUMOylation promotes Argonaute-directed transcriptional silencing in C. elegans. Elife 10, e63299. 19. Towbin BD, González-Aguilera C, Sack R, Gaidatzis D, Kalck V, Meister P, Askjaer P, and Gasser SM (2012). Step-wise methylation of histone H3K9 positions heterochromatin at the nuclear periphery. Cell 150, 934–947. [PubMed: 22939621] 20. Luteijn MJ, Van Bergeijk P, Kaaij LJT, Almeida MV, Roovers EF, Berezikov E, and Ketting RF (2012). Extremely stable Piwi-induced gene silencing in Caenorhabditis elegans. The EMBO journal 31, 3422–3430. [PubMed: 22850670] 21. Baron-Benhamou J, Gehring NH, Kulozik AE, and Hentze MW (2004). Using the λN peptide to tether proteins to RNAs. In mRNA Processing and Metabolism (Springer), pp. 135–153. 22. Wedeles CJ, Wu MZ, and Claycomb JM (2013). Protection of germline gene expression by the C. elegans Argonaute CSR-1. Dev. Cell 27, 664–671. [PubMed: 24360783] 23. Aoki ST, Lynch TR, Crittenden SL, Bingman CA, Wickens M, and Kimble J (2021). C. elegans germ granules require both assembly and localized regulators for mRNA repression. Nat. Commun. 12, 996–1014. [PubMed: 33579952] 24. Cornes E, Bourdon L, Singh M, Mueller F, Quarato P, Wernersson E, Bienko M, Li B, and Cecere G (2022). piRNAs initiate transcriptional silencing of spermatogenic genes during C. elegans germline development. Dev. Cell 57, 180–196.e7. [PubMed: 34921763] 25. Bühler M, Verdel A, and Moazed D (2006). Tethering RITS to a nascent transcript initiates RNAi-and heterochromatin-dependent gene silencing. Cell 125, 873–886. [PubMed: 16751098] 26. Phillips CM, Montgomery TA, Breen PC, and Ruvkun G (2012). MUT-16 promotes formation of perinuclear mutator foci required for RNA silencing in the C. elegans germline. Genes Dev. 26, 1433–1444. [PubMed: 22713602] 27. Shukla A, Yan J, Pagano DJ, Dodson AE, Fei Y, Gorham J, Seidman JG, Wickens M, and Kennedy S (2020). Poly (UG)-tailed RNAs in genome protection and epigenetic inheritance. Nature 582, 283–288. [PubMed: 32499657] 28. Zhang C, Montgomery TA, Gabel HW, Fischer SEJ, Phillips CM, Fahlgren N, Sullivan CM, Carrington JC, and Ruvkun G (2011). mut-16 and other mutator class genes modulate 22G and 26G siRNA pathways in Caenorhabditis elegans. Proc. Natl. Acad. Sci. USA 108, 1201–1208. [PubMed: 21245313] 29. Guang S, Bochner AF, Burkhart KB, Burton N, Pavelec DM, and Kennedy S (2010). Small regulatory RNAs inhibit RNA polymerase II during the elongation phase of transcription. Nature 465, 1097–1101. [PubMed: 20543824] 30. Jiao AL, Perales R, Umbreit NT, Haswell JR, Piper ME, Adams BD, Pellman D, Kennedy S, and Slack FJ (2019). Human nuclear RNAi-defective 2 (NRDE2) is an essential RNA splicing factor. RNA 25, 352–363. [PubMed: 30538148] 31. Billi AC, Fischer SE, and Kim JK Endogenous RNAi pathways in C. elegans (May 7, 2014). WormBook, ed. The C. elegans Research Community, WormBook, 10.1895/wormbook.1.170.1, [http://www.wormbook.org]. 32. Rüdel S, Wang Y, Lenobel R, Körner R, Hsiao H-H, Urlaub H, Patel D, and Meister G (2011). Phosphorylation of human Argonaute proteins affects small RNA binding. Nucleic Acids Res. 39, 2330–2343. [PubMed: 21071408] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 17 33. Yang J, and Zhang Y (2015). I-TASSER server: new development for protein structure and function predictions. Nucleic Acids Res. 43, W174–W181. [PubMed: 25883148] 34. Schirle NT, Sheu-Gruttadauria J, and MacRae IJ (2014). Structural basis for microRNA targeting. Science 346, 608–613. [PubMed: 25359968] 35. Uebel CJ, Anderson DC, Mandarino LM, Manage KI, Aynaszyan S, and Phillips CM (2018). Distinct regions of the intrinsically disordered protein MUT-16 mediate assembly of a small RNA amplification complex and promote phase separation of Mutator foci. PLoS Genet. 14, e1007542. 36. Holoch D, and Moazed D (2015). RNA-mediated epigenetic regulation of gene expression. Nat. Rev. Genet. 16, 71–84. [PubMed: 25554358] 37. Ouyang JPT, Zhang WL, and Seydoux G (2022). The conserved helicase ZNFX-1 memorializes silenced RNAs in perinuclear condensates. Nat. Cell Biol. 24, 1129–1140. [PubMed: 35739318] 38. Wan G, Yan J, Fei Y, Pagano DJ, and Kennedy S (2020). A conserved NRDE-2/MTR-4 complex mediates nuclear RNAi in Caenorhabditis elegans. Genetics 216, 1071–1085. [PubMed: 33055090] 39. Maine EM, Hauth J, Ratliff T, Vought VE, She X, and Kelly WG (2005). EGO-1, a putative RNA-dependent RNA polymerase, is required for heterochromatin assembly on unpaired DNA during C. elegans meiosis. Curr. Biol. 15, 1972–1978. [PubMed: 16271877] 40. Claycomb JM, Batista PJ, Pang KM, Gu W, Vasale JJ, van Wolf-swinkel JC, Chaves DA, Shirayama M, Mitani S, Ketting RF, et al. (2009). The Argonaute CSR-1 and its 22G-RNA cofactors are required for holocentric chromosome segregation. Cell 139, 123–134. [PubMed: 19804758] 41. Xu F, Feng X, Chen X, Weng C, Yan Q, Xu T, Hong M, and Guang S (2018). A cytoplasmic Argonaute protein promotes the inheritance of RNAi. Cell Rep. 23, 2482–2494. [PubMed: 29791857] 42. Langmead B, and Salzberg SL (2012). Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–359. [PubMed: 22388286] 43. Brenner S (1974). The genetics of Caenorhabditis elegans. Genetics 77, 71–94. [PubMed: 4366476] 44. Dokshin GA, Ghanta KS, Piscopo KM, and Mello CC (2018). Robust genome editing with short single-stranded and long, partially single-stranded DNA donors in Caenorhabditis elegans. Genetics 210, 781–787. [PubMed: 30213854] 45. Askjaer P, Ercan S, and Meister P (2014). Modern techniques for the analysis of chromatin and nuclear organization in C. elegans. WormBook, 1–35. 10.1895/wormbook.1.169.1. 46. Schrodinger L (2010). The PyMOL Molecular Graphics System, Version 2.4.0 (Schrodinger, L.). 47. Beliveau BJ, Kishi JY, Nir G, Sasaki HM, Saka SK, Nguyen SC, Wu C. t., and Yin P (2018). OligoMiner provides a rapid, flexible environment for the design of genome-scale oligonucleotide in situ hybridization probes. Proc. Natl. Acad. Sci. USA 115, E2183–E2192. [PubMed: 29463736] Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 18 Highlights • • • • Nuclear Argonaute HRDE-1 separately induces heterochromatin and small RNA production HRDE-1 induces target alleles to merge into a single focus of heterochromatin Transcription within heterochromatin feeds de novo loading of HRDE-1 with small RNAs HRDE-1 shuttles to nuage and promotes small RNA production via its N- terminal domain Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 19 Figure 1. HRDE-1 tethering caused reporter silencing and generated the silencing memory (A) Scheme of λN-BoxB tethering system. A sequence encoding five BoxB hairpins (5xBoxB) was inserted immediately after the coding region of the GFP::his-58(H2B) transgene and before the unc-54 3′ UTR. The reporter is driven by the pie-1 promoter (Ppie-1). The BoxB sites recruit λN::HRDE-1 or λN::NRDE-2 fusion proteins, thereby tethering HRDE-1 or NRDE-2 to the reporter RNA. (B) Representative fluorescence image of a syncytial germline (outlined by dashed lines) in the absence of tethering. The image represents 100% of worms scored, N > 30. (C) Representative fluorescence image in the presence of HRDE-1 tethering. The image represents 100% of worms scored, N > 30. (D) Representative fluorescence image in the presence of NRDE-2 tethering. The image represents 100% of worms scored, N > 30. (E and F) Analysis of inherited silencing triggered by λN::HRDE-1 tethering. After outcross to hrde-1 wild type (E) or hrde-1 null (F), reporter worms were scored for gfp expression for 13 generations after segregating away the λN::hrde-1 allele. The percentage of GFP+ (ON) or GFP– (OFF) worms is indicated, N > 30 worms scored in each generation. (G) Color chart showing genetic requirements of inherited silencing triggered by λN::HRDE-1 tethering. The λN::hrde-1; reporter worms were crossed to the indicated mutants. After segregating away λN::hrde-1, reporter worms homozygous for the indicated mutations were scored for GFP expression: ON or OFF, as indicated. N > 30 worms scored for each genotype. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 20 Figure 2. HRDE-1 and NRDE-2 tethering induce heterochromatin formation (A and B) Quantification of H3K9me3 levels near the reporter in the presence or absence of HRDE-1 or NRDE-2 tethering, as determined by chromatin immunoprecipitation (ChIP)- qPCR. P1 and P4 primer sets analyze sequences 5 kb upstream or downstream of the reporter, and P2 and P3 analyze sequences within the reporter, as indicated in the schematic. All quantities were normalized to the level of P1 in reporter control samples. Error bars show the standard deviation from the mean. (C and D) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to HRDE-1 or NRDE-2 tethering, as determined by qPCR. The average quantities relative to wild type (WT) are indicated. Error bars show the standard deviation from the mean. (E and F) Color chart showing the genetic requirements of silencing in the presence of λN::NRDE-2 or λN::HRDE-1. Reporter worms homozygous for the indicated mutations were scored for GFP expression: ON or OFF, as indicated. N > 30 worms scored for each genotype. GFP is ON, but signal is weak (see Figure S2H). Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 21 Figure 3. HRDE-1 and NRDE-2 tethering promote antisense small RNA production (A) Plot showing antisense small RNA reads (per million total reads) mapping to the reporter (indicated below the plot) in the absence of tethering. Only the first nucleotide is counted. Green boxes, GFP coding; blue box, H2B coding; pink boxes, BoxB hairpins. (B–E) Genetic requirements of small RNAs induced by NRDE-2 tethering. Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::NRDE-2 in WT (B), nrde-4 (C), rde-3 (D), or hrde-1 (E) worms. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 22 (F–I) Genetic requirements of small RNAs induced by HRDE-1 tethering. Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::HRDE-1 in WT (F), nrde-2 (G), rde-3 (H), or mut-16 (I) worms. (J–M) Genetic requirements of inherited small RNAs induced by HRDE-1 tethering. Plots showing antisense small RNA reads mapping to the reporter in WT (F), nrde-2 (G), rde-3 (H), or mut-16 (I) worms after segregating λN::HRDE-1. Note that in (G), the y axis is compressed 50% compared with other plots to conserve space. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 23 Figure 4. HRDE-1 guide RNA loading is not required for small RNA amplification (A) Western blot analysis to detect GFP::HRDE-1 and GFP::HRDE-1(Y669E) in worm lysates. Top panel: probed with anti-GFP antibody. GFP::HRDE-1 was indicated. Bottom panel: probed with anti-tubulin antibody as a loading control. (B) Confocal images showing the localization of GFP::HRDE-1(WT) or GFP::HRDE-1(Y669E) with mCherry::GLH-1 as P granule marker. The white dashed lines outline a gonadal arm of the germline. (C) Representative fluorescence (left panels) and differential interference contrast (DIC; right) images showing that λN::HRDE-1(Y669E) silences the BoxB reporter in WT worms (top panels, OFF) but not in rde-3 mutant worms (bottom panels, ON). The images represent 100% of the animals scored, N > 30. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 24 (D) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to HRDE-1(Y669E) tethering, determined by qPCR. The average quantities relative to WT are indicated. Error bars show the standard deviation from the mean. (E and F) Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::HRDE-1(Y669E) in WT (E) or rde-3 (F) worms. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 25 Figure 5. HRDE-1 N-terminal domain promotes small RNA amplification and poly-UG modification (A) Schematic showing HRDE-1 linear domain structure and truncations tested. The subdomains are color coded based on human Ago2 (Figure S5A). The percentage of GFP+ worms (ON) is indicated, N > 30 worms scored in each test. (B) Predicted three-dimensional structures of HRDE-1 N-terminal domain (NTD) and C- terminal domain (CTD). Subdomains as in (A). (C) Color chart indicating the expression (ON) or silencing (OFF) of the reporter in the presence of λN::CTD or λN::NTD and the requirement of nrde-2 or rde-3. N > 30 worms scored for each genotype. (D and E) Plots showing antisense small RNA reads mapping to the reporter in the presence of λN::NTD in WT (D) or rde-3 worms (E). (F) Bar graphs showing the quantification of reporter RNA and pre-mRNA levels in response to NTD tethering, as determined by qPCR. The average quantities relative to the control are indicated. Error bars show the standard deviation from the mean. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 26 (G and H) Analysis of poly-UG modification of reporter RNA in response to tethering in the indicated mutants. Poly-UG PCR products in (G) were cloned and sequenced to identify the precise positions of poly-UG addition (H), indicated by arrowheads. A gsa-1-specific PCR was used as loading control. Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 27 Figure 6. HRDE-1 localizes in Mutator foci, and HRDE-1 tethering caused peri-nuclear accumulation of reporter RNA in nuclear silencing mutants (A and B) Confocal image of live germ cells showing the co-localization of GFP::HRDE-1 with mCherry::GLH-1 (A) and MUT-16::mCherry (B). Each subpanel shows a projected view of a segment of the germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. Yellow arrows point to peri-nuclear foci where HRDE-1 co-localizes with GLH-1 and MUT-16. (C and D) Confocal images of live germ cells showing the co-localization of GFP::HRDE-1(NTD) with mCherry::GLH-1 (C) and MUT-16::mCherry (D). As in (A) and Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 28 (B). (E and F) Confocal images of RNA FISH experiments showing the localization of reporter RNA with mCherry::GLH-1 (left) or MUT-16::mCherry (right) in control worms (E) or in worms exposed to gfp RNAi (F). Magenta, RNA; green, GLH-1 or MUT-16; and blue, DAPI. Each subpanel shows a projected view of a segment of a representative germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image with DNA and GLH-1 or MUT-16 signals (center) or with DNA signal only (right). Yellow arrows point to nuclear RNA foci that likely correspond to transcription sites. (G and H) As in (F) but in hrde-1 (G) or nrde-2 (H) mutant worms. (I) Bar graphs showing the percentage of nuclei from (E)–(H) containing one reporter RNA focus (orange) or two or more reporter RNA foci (light green). Three independent germlines were measured for each condition. Error bars show the standard deviation from the mean. (J) Bar graphs showing the percentage of nuclei from DNA FISH (Figure S6E) containing one reporter DNA focus or two or more reporter DNA foci. Similar to (I). (K–N) Confocal images of RNA FISH showing the localization of reporter RNA with mCherry::GLH-1 (left) or MUT-16::mCherry (right) in the absence (K) or presence (L–N) of HRDE-1 tethering, as indicated. Details as in (E) and (F). (O) Bar graphs showing the percentage of nuclei from (J)–(N) containing one reporter RNA focus (peach) or two or more reporter RNA foci (light green). (P–S) Confocal images of DNA FISH experiments showing the localization of reporter DNA loci in the absence (P) or presence (Q–S) of HRDE-1 tethering, as indicated. Green, DNA FISH signal; blue, DAPI. A projected view of a segment of a representative germline is shown to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. Yellow arrows point to the nuclear DNA signals. (T) Bar graphs showing the percentage of nuclei from DNA FISH experiments in (P)–(S) containing one reporter DNA focus or two or more reporter DNA foci. Details as in (I). Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 29 Figure 7. Model of HRDE-1-mediated self-enforcing mechanism (A and B) Confocal images showing the localization of GFP::HRDE-1 with mCherry::GLH-1 in WT worms (A) and mut-16 mutants (B). Green, GFP::HRDE-1 (left); magenta, mCherry::GLH-1 (middle); merge (right). Each subpanel shows a projected image of a representative pachytene region of the germline to the left, and the nucleus bounded by a dashed box is shown as a single-focal-plane image to the right. (C) Model (see Discussion). Cell Rep. Author manuscript; available in PMC 2023 August 22. Ding et al. Page 30 A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t 8 2 3 5 0 3 _ B A D I F R : ; 0 6 1 6 b a # t a C 7 8 6 0 1 3 _ B A D I F R : ; 3 2 5 7 0 # t a C 0 5 P O : e s a B m r o W 5 1 1 T H : e s a b m r o W m a c b A e t a t s p U C G C C G C S 6 5 9 2 # t a C g n i l a n g i S l l e C s n i a r t s s u r i v d n a l a i r e t c a B 0 5 P O oli: E. C 5 1 1 T H oli: E. C n i l u b u T - i t n A P F G - i t n A 3 e m 9 K 3 H - i t n A s e i d o b i t n A R E I F I T N E D I E C R U O S E C R U O S E R r o T N E G A E R E L B A T S E C R U O S E R Y E K 9 5 0 1 8 0 1 # t a C ) T D I ( s e i g o l o n h c e T A N D d e t a r g e t n I 3 V e s a e l c u N 9 s a C . p . S ® R - t l A s n i e t o r p t n a n i b m o c e r d n a , s e d i t p e p , s l a c i m e h C 4 3 5 2 7 0 1 # t a C 6 5 7 9 L # t a C 4 2 4 9 T # t a C S 3 0 3 0 M # t a C 0 0 2 1 9 0 8 1 # t a C 6 1 6 6 2 # t a C 2 6 1 6 2 # t a C 0 7 2 2 M A # t a C - O R K T O R P R # t a C 6 9 6 2 m a # t a C L 3 7 3 0 M # t a C S 4 0 2 0 M # t a C 5 8 0 0 8 0 8 1 # t a C 0 2 8 6 5 2 1 # t a C S 3 7 2 0 M # t a C 1 0 0 6 4 4 1 1 4 1 1 # t a C 0 0 5 – 7 3 3 P B h c i r d l A - a m g i S h c i r d l A - a m g i S B E N T D I s e i g o l o n h c e T e f i L s e i g o l o n h c e T e f i L e s a t p i r c s n a r T e s r e v e R V I ™ t p i r c S r e p u S x i M r e t s a M n e e r G R B Y S t s a F e d i r o l h c o r d y h e l o s i m a r t e T A N R r c a r t - 9 s a C R P S I R C ) L O Z I R T ( t n e g a e r I R T I e s a N D s e i g o l o n h c e T e f i L s d a e b c i t e n g a m G A g I / e d a r g P I H C e c r e i P s e i g o l o n h c e T e f i L h c i r d l A a m g i S n e g o r t i v n I s e i g o l o n h c e T e f i L c i f i t n e i c S r e h s i F h c i r d l A - a m g i S c i f i t n e i c S r e h s i F B E N B E N B E N e s a t p i r c s n a r T e s r e v e R I I I ™ t p i r c S r e p u S e d i l S k c i h T d e t s o r F n O - e t i R ™ a i d e r p E ) G T P I ( e d i s o t c a l a g o i h t - D β- - l y p o r p o s I e s a r e m y l o p A N D q a T 0 2 n e e w T Q K d e t a c n u r t , 2 e s a g i L A N R 4 T r o t i b i h n I e s a N R e s a R E P U S 1 e s a g i L A N R 4 T A e s a N R K e s a t o r P Cell Rep. Author manuscript; available in PMC 2023 August 22. Ding et al. Page 31 l m t h . n e z - s s i e z / e r a w t f o s / s t c u d o r p / n e / y p o c s o r c i m m o c . s s i e z . w w w / / / : s p t t h S S I E Z n o i t i d e e u l b n o i s r e V o r p 2 N E Z m o c . r o d n a . p l e h . n o i s u f / / : s p t t h s t n e m u r t s n I d r o f x O 4 4 . 0 . 3 . 2 n o i s r e V n o i s u F m o c . t s n i x o . s i r a m i / / : s p t t h s t n e m u r t s n I d r o f x O : e r a w t f o S s i s y l a n A e g a m I y p o c s o r c i M o i . s c o d e h t d a e r . t p a d a t u c / / : s p t t h g r o . n o h t y p . w w w / / : s p t t h t e n . j e g a m i / / : s p t t h g r o . n o h t y P S I B N H N I 1 . 7 . 9 n o i s r e V s i r a m I J e g a m I 3 n o h t y P 1 . 4 n o i s r e v t p a d a t u C L m 0 5 2 – 7 8 7 8 T # t a C R E I F I T N E D I 1 – 5 2 3 1 P B # t a C 9 4 5 2 5 2 # t a C h c i r d l A - a m g i S E C R U O S E C R U O S E R r o T N E G A E R 0 0 1 - X n o t i r T c i f i t n e i c S r e h s i F n o i t u l o S X 0 2 , ) C S S ( e t a r t i C m u i d o S e n i l a S h c i r d l A - a m g i S e d y h e d l a m r o F 0 1 – 1 B H - F M S # t a C d t L s c i m o n e G , h c r a e s o i B C G L r e f f u B n o i t a z i d i r b y H H S I F A N R ® s i r a l l e t S 0 1 – 0 0 2 1 - H # t a C s e i r o t a r o b a L r o t c e V g n i t n u o M e d a f i t n A ® D L E I H S A T C E V I P A D h t i w m u i d e M s y a s s a l a i c r e m m o c l a c i t i r C 9 2 0 0 – 0 1 9 5 # t a C 1 6 5 1 M A # t a C 6 0 9 4 2 0 0 2 # t a C 1 4 6 0 5 4 # t a C s e i g o l o n h c e T e f i L t i k n o i t a l o s I i A N R m ™ a n a V r i m s e i g o l o n h c e T e f i L t i K ™ g n i n o l C A T ™ O P O T . c n I , a n i m u l l I ) s e l c y C 5 7 ( 5 . 2 v t i K t u p t u O h g i H 0 5 5 / 0 0 5 q e S t x e N e r a C a r e S t i K g n i t t o l B t n e c s e n i m u l i m e h C P A ™ r o t c e t e D L P K a t a d d e t i s o p e D ) 6 0 8 4 7 8 A N J R P ( I B C N y d u t s s i h T a t a d g n i c n e u q e s A N R l l a m S 1 S e l b a T y d u t s s i h T 2 S e l b a T 3 S e l b a T 4 S e l b a T y d u t s s i h T y d u t s s i h T y d u t s s i h T s n i a r t s / s m s i n a g r O : s l e d o m l a t n e m i r e p x E s e b o r p H S I F A N R f o t s i L s e b o r p H S I F A N D f o t s i L s e d i t o e l c u n o g i l O f o t s i L s m h t i r o g l a d n a e r a w t f o S s n i a r t s s n a g e l e . C s e d i t o e l c u n o g i l O Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Ding et al. Page 32 l m t h . 2 q e S E D / l m t h / c o i b / 0 . 3 / s e g a k c a p / p j . n e k i r . r o t c u d n o c o i b / / : s p t t h / p o t k s e d - o i d u t s r / d a o l n w o d / o c . t i s o p / / : s p t t h / R E S S A T - I / g r o . p u o r g g n a h z / / : s p t t h g r o . l o m y p / / : s p t t h 2 e i t w o b / t e n . e g r o f e c r u o s . o i b - e i t w o b / / : s p t t h R E I F I T N E D I g r o . t c e j o r p - r . w w w / / : s p t t h 2 4 . l a t e d a e m g n a L E C R U O S g r o . t c e j o r p - R e r a w t f o S t i s o P r o t c u d n o c o i B 3 3 . l a t e g n a Y r e g n i d o r h c S E C R U O S E R r o T N E G A E R 4 5 5 d l i u B n o i s r e V o i d u t s R 0 . 6 2 . 1 n o i s r e v 2 q e S E D 5 . 4 . 2 n o i s r e V 2 e i t w o B 1 . 2 . 4 n o i s r e V R R E S S A T - I s c i h p a r G r a l u c e l o M ) M T ( L O M y P 0 . 5 . 2 n o i s r e V m e t s y S Cell Rep. Author manuscript; available in PMC 2023 August 22. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1016_j.cell.2023.07.039	Phage-assisted evolution and protein engineering yield compact, efﬁcient prime editors Resource Graphical abstract Authors Jordan L. Doman, Smriti Pandey, Monica E. Neugebauer, ..., Jakub Tolar, Mark J. Osborn, David R. Liu Correspondence [email protected] In brief Phage-assisted continuous evolution and protein engineering of prime editors reveals relationships between prime edit type, reverse transcriptase variant, and editing efﬁciency, enabling the development of PE6 reverse transcriptase and Cas9 variants with reduced size and improved editing efﬁciency in cell lines and in mice. Highlights d PE-PACE converts compact, low-activity RTs into efﬁcient prime editors d PegRNA length and secondary structure determine the optimal choice of prime editor d PE6 RT and Cas9 domains can enhance prime editing efﬁciencies beyond that of PEmax d AAV-delivered PE6 editors enable the installation of long, complex edits in vivo Doman et al., 2023, Cell 186, 3983–4002 August 31, 2023 ª 2023 The Authors. Published by Elsevier Inc. https://doi.org/10.1016/j.cell.2023.07.039 ll ll OPEN ACCESS Resource Phage-assisted evolution and protein engineering yield compact, efﬁcient prime editors Jordan L. Doman,1,2,3,5 Smriti Pandey,1,2,3,5 Monica E. Neugebauer,1,2,3 Meirui An,1,2,3 Jessie R. Davis,1,2,3 Peyton B. Randolph,1,2,3 Amber McElroy,4 Xin D. Gao,1,2,3 Aditya Raguram,1,2,3 Michelle F. Richter,1,2,3 Kelcee A. Everette,1,2,3 Samagya Banskota,1,2,3 Kathryn Tian,1,2,3 Y. Allen Tao,1,2,3 Jakub Tolar,4 Mark J. Osborn,4 and David R. Liu1,2,3,6,* 1Merkin Institute of Transformative Technologies in Healthcare, Broad Institute of MIT and Harvard, Cambridge, MA, USA 2Department of Chemistry and Chemical Biology, Harvard University, Cambridge, MA, USA 3Howard Hughes Medical Institute, Harvard University, Cambridge, MA, USA 4Department of Pediatrics, University of Minnesota Medical School, Minneapolis, MN, USA 5These authors contributed equally 6Lead contact Correspondence: [email protected] https://doi.org/10.1016/j.cell.2023.07.039 SUMMARY Prime editing enables a wide variety of precise genome edits in living cells. Here we use protein evolution and engineering to generate prime editors with reduced size and improved efﬁciency. Using phage-assisted evo- lution, we improved editing efﬁciencies of compact reverse transcriptases by up to 22-fold and generated prime editors that are 516–810 base pairs smaller than the current-generation editor PEmax. We discovered that different reverse transcriptases specialize in different types of edits and used this insight to generate reverse transcriptases that outperform PEmax and PEmaxDRNaseH, the truncated editor used in dual- AAV delivery systems. Finally, we generated Cas9 domains that improve prime editing. These resulting ed- itors (PE6a-g) enhance therapeutically relevant editing in patient-derived ﬁbroblasts and primary human T-cells. PE6 variants also enable longer insertions to be installed in vivo following dual-AAV delivery, achieving 40% loxP insertion in the cortex of the murine brain, a 24-fold improvement compared to previous state-of-the-art prime editors. INTRODUCTION Prime editing (PE) can install virtually any substitution, small insertion, or small deletion in the genomes of living cells without requiring double-stranded breaks (DSBs) in DNA or donor DNA templates and thus can correct the vast majority of known path- ogenic mutations.1 PE requires a prime editing guide RNA (pegRNA) and a prime editor protein, which consists of a programmable nickase and a reverse transcriptase (RT). The ﬁrst-generation prime editor (PE1) used the wild-type Moloney murine leukemia virus (M-MLV) RT, while subsequent prime ed- itors (PE2–PE5) use an engineered pentamutant M-MLV RT (Fig- ure 1A).1,2 The pegRNA contains a guide RNA scaffold, a spacer that speciﬁes the target site, a primer binding site (PBS) that is complementary to the target DNA, and a reverse transcriptase template (RTT) that encodes the desired edit. The prime edi- tor,pegRNA complex pairs with one strand of the target genomic DNA and nicks the opposite strand to generate an exposed 30 end that binds the PBS of the pegRNA. The RT engages the re- sulting primer-template complex and initiates reverse transcrip- tion of the RTT. The newly synthesized 30 DNA ﬂap containing the edit is incorporated into the genome, replacing the original DNA sequence and permanently installing the desired edit.1 In the PE3 and PE5 systems, an additional single guide RNA (sgRNA) directs the prime editor to nick the non-edited DNA strand and bias cellular mismatch repair to favor installation of the edit (Figure 1A).1,2 Since the development of PE systems, we and others have improved them by engineering the pegRNA,3–5 prime editor ar- chitecture,2,3,6,7 and cellular DNA repair response to favor desired outcomes.2,8 Twin prime editing (twinPE) and related ‘‘dual-ﬂap’’ methods use two pegRNAs to edit both DNA strands, enabling larger insertions and deletions (>100 base pairs [bp]).9–15 PE and twinPE have been used to install recombi- nase landing sites, enabling targeted gene-sized (>5,000 bp) in- sertions and inversions.9,16 Despite these advances, improving the prime editor protein has proven challenging. The M-MLV RT mutations used in PE2–PE5 systems were identiﬁed over decades of screening for improved RTs,17–20 followed by additional screening to opti- mize mammalian PE efﬁciencies.1 These mutations are critical to the efﬁciency of PE, and few analogous mutations are known for other RTs. Prime editor proteins that use compact RTs could facilitate in vivo prime editor delivery, and different RT enzymes Cell 186, 3983–4002, August 31, 2023 ª 2023 The Authors. Published by Elsevier Inc. 3983 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). ll OPEN ACCESS A C Resource B D E F G I J H Figure 1. Identiﬁcation and engineering of reverse transcriptase enzymes into prime editor candidates (A) Overview of PE systems. All use a prime editor protein consisting of SpCas9(H840A) nickase fused to a reverse transcriptase (RT) enzyme. PE1 uses the wild- type RT from the Moloney murine leukemia virus (M-MLV), while the PE2 system uses an engineered pentamutant variant of the M-MLV RT. PE3 uses an additional single guide RNA (sgRNA) to nick the non-edited strand. PBS = primer binding site. RT template = reverse transcriptase template. (B) Phylogenetic classiﬁcation of RTs tested in this study. Red circles indicate PE-active enzymes. Green circles indicate PE-inactive enzymes. (C) Mammalian activity of 20 different RT enzymes in the prime editing system at endogenous sites in HEK293T cells. (D) Comparison of wild-type Tf1 RT, PE2DRNaseH, and PE2 at three longer, complex PE (HEK3) or twinPE (CCR5 and IDS) edits in HEK293T cells. 3984 Cell 186, 3983–4002, August 31, 2023 (legend continued on next page) Resource ll OPEN ACCESS may support different editing capabilities. All previously reported prime editors that use RTs other than M-MLV RT, however, have shown substantially lower PE efﬁciencies than PE2 even after extensive engineering.3,16,21,22 Further improvement of the high- ly engineered M-MLV RT in PE2 has also proven difﬁcult, as all reported variants of this RT have also yielded minimal improve- ments in mammalian cell PE.16,22,23 Although we reported that Cas9 mutations known to improve nuclease performance can also increase PE efﬁciency,2 mutants of Cas9 identiﬁed specif- ically to improve PE have not yet been reported. In this study, we developed a phage-assisted continuous evo- lution (PACE)24 selection for PE and used evolution and protein engineering to generate PE6a–g variants that are more efﬁcient and/or more easily delivered in vivo than previous state-of-the- art prime editors. PE6 variants synergize with other recent PE ad- vances2,4 to offer cumulative beneﬁts in a variety of contexts, including in patient-derived ﬁbroblasts and primary human T cells. Dual adeno-associated virus (dual-AAV) delivery of PE6 systems achieved 12- to 183-fold improvements in PE efﬁciency compared to previous state-of-the-art systems for the installa- tion of 38- to 42-bp edits in the mouse brain, yielding 62% tar- geted installation of the loxP sequence among transduced cells in the mouse cortex. RESULTS Surveying reverse transcriptase enzymes for prime editing Because only a handful of RTs beyond M-MLV RT have been used for PE,3,16,21,22 we ﬁrst surveyed RTs from diverse phyloge- netic origins and tested 59 enzymes (Table S1) spanning 14 clas- ses (Figure 1B) as prime editors. We compared these editors to PE1, PE2, and PE2DRNaseH (the RNaseH-truncated form of PE2 used for dual-AAV delivery3,21,25–27) for three edits in HEK293T cells. Twenty RTs from four different classes showed detectable PE activity, and nine of these RTs are R500 bp smaller in gene size than M-MLV RT (Figure 1C). However, all PE-compatible RTs exhibited lower editing efﬁciencies than PE2, with the smaller RTs showing especially poor activity (Figures 1C and S1A). These results agree with recent re- ports3,16,21,22 that while diverse RTs can support PE, their wild- type forms do not mediate efﬁcient PE in mammalian cells. The most efﬁcient wild-type RT, Schizosaccharomyces pombe Tf1 retrotransposon28 RT, approached PE2 efﬁciencies at substitution edits but struggled to install a 40-bp loxP insertion edit (Figure 1C). We noted a similar trend for PE2DRNaseH. While the RNaseH domain of MMLV RT is dispensable for PE,21,25,26 our data suggested that PE2DRNaseH might show deﬁciencies at longer, more challenging edits. Indeed, the Tf1- derived editor and PE2DRNaseH performed worse than PE2 at two additional complex edits that use twinPE (Figure S1B). On average, at these three challenging edits, PE2DRNaseH yielded 1.4-fold lower PE efﬁciency than PE2, and wild-type Tf1 per- formed 15-fold worse than PE2 (Figure 1D). These initial ﬁndings identiﬁed three challenges. First, the vast majority of RTs, especially the most compact enzymes, do not support efﬁcient mammalian cell PE for any edit type. Second, even the most active dual-AAV-compatible RTs ((cid:2)1.5 kb in gene size) such as the truncated RT in PE2DRNaseH showed lower editing efﬁcieny compared to the full-length RT in PE2 when installing long, complex edits. Finally, none of the enzymes we evaluated surpassed the editing efﬁciency of PE2. We ﬁrst at- tempted to addess these problems using protein engineering. Rational engineering of reverse transcriptase enzymes We ﬁrst engineered retroviral RTs based on our previous engi- neering of the M-MLV RT to create PE2. The PE2 protein con- tains ﬁve mutations in M-MLV RT (D200N, T306K, W313F, T330P, and L603W) that enhance the enzyme’s in vitro substrate binding, processivity, and thermostability.1,17–20 Installing muta- tions corresponding to each of these PE2 substitutions into RTs from porcine endogenous retrovirus (PERV), koala retrovirus (KoRV), avian reticuloendotheliosis virus (AVIRE), and woolly monkey sarcoma virus (WMSV) increased PE efﬁciencies (Fig- ure S1C). Combining all ﬁve mutations further improved editing by an average of 5.3-fold to 6.8-fold compared to each enzyme’s wild-type counterpart across ﬁve different edits in HEK293T cells (Figures 1E and S1C). We were also interested in engineering Tf1 RT due to its small size and higher baseline performance compared to other wild- type enzymes. Since increasing the afﬁnity between the RT and its DNA,RNA substrate can improve PE efﬁciency,1 we used the structure of a Tf1 homolog, Ty3 RT (Protein DataBank [PDB]: 4OL8), to guide the design of mutations in Tf1 proximal to DNA,RNA substrate and tested their ability to support PE in HEK293T cells (Figure 1F). Five of these mutations (K118R, S118K, I260L, S297Q, and R288Q) improved editing efﬁciency, and combining all ﬁve mutations additively improved mammalian (E) Comparison of prime editors containing engineered retroviral RT variants with their wild-type counterparts in HEK293T cells. Horizontal bars show the mean value. (F) Residues mutated to improve editing of the Tf1 RT prime editor correspond to V188, R118, L258, M281 and V286 (red) in Ty3 RT (blue, PDB: 4OL8). V188 and R118 are in close proximity to the RNA (green) substrate and correspond to K118 and S188 in Tf1, respectively. L258, M281 and V286 are near the DNA (yellow) substrate and correspond to I260, S297 and R288 in Tf1, respectively. (G) Rationally designed Tf1 pentamutant variant (rdTf1) shows improvements in editing over its wild-type counterpart in HEK293T cells. All edits are PE edits, except the AAVS1 site, which is twinPE. (H) Rationally designed Ec48 triple mutant variant (rdEc48) shows improvements in editing over its wild-type counterpart for ﬁve edits in HEK293T cells. (I) Comparison of prime editors containing engineered RT variants with PE2 in HEK293T cells. All edits use single-ﬂap prime editing, except the AAVS1 site, which uses twinPE. (J) Comparison of rdTf1 with PE2 and its wild-type counterpart at three longer, complex PE (HEK3) or twinPE (CCR5 and IDS) edits in HEK293T cells. Dots indicate individual replicates for n = 3 biological replicates (C–E and G–J). Bars reﬂect the mean of n = 3 independent replicates (C, D, G, H, and J). See also Figure S1. Throughout all ﬁgures (Figures 1, 2, 3, 4, 5, 6, 7, and S1–S7), prime editing efﬁciencies shown reﬂect the frequency of the intended prime editing outcome with no indels or other changes at the target site. Cell 186, 3983–4002, August 31, 2023 3985 A D G ll OPEN ACCESS Resource B C E F H I J K Figure 2. Development and validation of a prime editing PACE selection (A) Schematic of PE-PACE selection circuit. Upon infection of E. coli by selection phage (blue), the NpuN intein and NpuC intein (pink) mediate reconstitution of the PE2 prime editor (purple and pink), which engages a pegRNA (dark green) and corrects a frameshift in T7 RNAP (orange) via PE. Functional T7 RNAP then transcribes gIII (light green), which enables SP propagation. (B) Phage replication levels from overnight propagation of empty phage (red), NpuC-PE2-RT phage (purple), and T7-RNAP phage (green) in PE-PACE host cells before pegRNA optimization. 3986 Cell 186, 3983–4002, August 31, 2023 (legend continued on next page) Resource ll OPEN ACCESS editing efﬁciencies. The ﬁnal rationally designed Tf1 variant (rdTf1) showed a 1.8-fold average improvement in PE efﬁciency over wild-type Tf1 in HEK293T cells across seven different edits (Figures 1G, S1D, and S1E). We also used structure-guided engineering to improve the ed- iting efﬁciency of the Escherichia coli Ec48 retron29 RT, which is even smaller than Tf1 RT, but also less active (Figure 1C). Since the structure of a retron RT30 had not been reported at the time, we used AlphaFold231 to predict the structure of Ec48 RT (Fig- ure S1F). Incorporation of T189N in Ec48, the mutation predicted by AlphaFold2 to correspond to D200N in PE2, improved PE efﬁciency by 3-fold on average across six different edits in HEK293T cells (Figures S1G and S1H). Rational engineering us- ing the same structure yielded ﬁve additional mutations (K307R, R378K, L182N, T385R, and R378K) that improved PE efﬁ- ciencies, potentially by improving binding to the DNA or RNA substrates (Figures S1H and S1I). Combining the top-performing mutations yielded rdEc48, which exhibits an 8.6-fold improve- ment in average PE efﬁciency over wild-type Ec48 across six edits in HEK293T cells (Figures 1H and S1J). Despite these substantial improvements, PE efﬁciencies of all six engineered RT enzymes remained lower than those of PE2 (Figure 1I). The most compact engineered RT (rdEc48) exhibited 8-fold lower average editing efﬁciencies than PE2 (Figure 1I). Although rdTf1 approached PE2 levels of editing for several edits noted in Figure 1I, it struggled with longer, more complex edits and performed 1.6-fold worse than PE2 at the same three sites tested in Figure 1D (Figure 1J). To overcome these limitations, we turned to laboratory evolution. Development and validation of a prime editing PACE selection circuit Phage-assisted continuous and non-continuous evolution (PACE and PANCE, respectively)24,32 are methods for highly accelerated laboratory evolution in which the propagation of a modiﬁed bacteriophage is linked to the activity of a protein of in- terest (Figures S2A and S2B). To develop a prime editor PACE (PE-PACE) circuit that links PE activity with phage propagation, we removed the essential phage gene gIII from the phage genome and placed it under the control of a T7 promoter on a plasmid (P1) in host E. coli. A second plasmid (P2) contained a defective T7 RNA polymerase (T7 RNAP) gene with a 1-bp dele- tion frameshift mutation. PE correction of this frameshift enables T7 RNAP production, gIII expression, and phage propagation. In the initial version of our circuit (v1), SpCas9(H840A) nickase was fused to the N-terminal half of the Npu intein (NpuN) and encoded on a separate host plasmid, P3. A C-terminal Npu intein (NpuC) fused to the PE2 RT was encoded on the selection phage, such that intein splicing reconstitutes full-length prime editor after phage infection. Finally, a pegRNA encoding the corrective T7 edit was included on P1. This selection allows the RT, but not the Cas9 nickase domain, to evolve during PACE (Figure 2A). We evaluated this selection circuit by overnight phage propa- gation assays. NpuC-PE2-RT phage only propagated 1.4-fold overnight, indicating the need to optimize the circuit (Figure 2B). Because mammalian PE efﬁciency is heavily inﬂuenced by the choice of pegRNA PBS and RTT,33 we tested 35 pegRNAs and found that overnight propagation levels of NpuC-PE2-RT phage varied 14,000-fold depending on the pegRNA (Figures 2C and S2C). An optimized pegRNA enabled robust (>100-fold) over- night propagation of NpuC-PE2-RT phage. To test the dynamic range of the selection, we generated NpuC-PE1-RT phage and evaluated them in our pegRNA-opti- mized circuit, and we found that NpuC-PE1-RT phage de-en- riched 6.7-fold, while NpuC-PE2-RT phage propagated 140-fold (Figure 2D), establishing that the selection can distin- guish RT variants based on their PE activity. Finally, to verify that the circuit can enrich mutations that enhance PE, we evolved NpuC-PE1-RT phage in PANCE. Eight overnight PANCE passages yielded six converged mutations (Figures 2E including two we previously engineered1 in PE2, and 2F), demonstrating that PANCE can evolve mutations known to enhance mammalian cell PE. High-stringency PE-PACE reveals edit-dependent effects on evolved editors Based on our observation that RTs such as PE2DRNaseH and rdTf1 were deﬁcient when using long RTTs (Figures 1C and 1D), we hypothesized that increasing edit size and RTT length (C) Screen of pegRNAs for the v1 PE-PACE circuit. Overnight propagation values of empty phage (red), NpuC-PE2-RT phage (purple), and T7-RNAP phage (green) are shown. Each point reﬂects the mean value of n = 3 independent biological replicates for a different pegRNA. Individual replicates are shown in Figure S2C. (D) Overnight propagation of empty phage (red), NpuC-PE1-RT phage (light purple), NpuC-PE2-RT phage (dark purple), and T7-RNAP phage (green) in the v1 pegRNA-optimized circuit. (E) PANCE titers for the evolution of NpuC-PE1-RT phage. Gray shading indicates a passage of evolutionary drift, in which phage were supplied gIII in the absence of selection. Titers of four replicate lagoons are shown. (F) Mutation table for NpuC-PE1-RT phage surviving v1 PANCE. Four clones per lagoon (L1-L4, with clones ordered by lagoon) were sequenced. Light purple denotes conserved mutations. Dark purple denotes conserved mutations also present in the previously engineered PE2 RT1. (G) Schematic of the PE-PACE selection for evolution of the whole prime editor, including the Cas9 domain. The P1 plasmid (green) and P3 plasmid (orange) are identical to those used in Figure 2A. (H) PANCE experiment to compare the outcome of selection on v1 and v2 selection circuits. Replicate lagoons were evolved on each (v1, yellow and v2, blue) selection circuit. After 31 passages, clones from each selection were sequenced, and the resulting mutations were compared to generate (I-K). (I) Violin plots showing the number of mutations per clone for the M-MLV domain of whole-editor phage evolved with either the v1 (yellow) or v2 (blue) circuit. Data are shown as individual values, with one dot representing one sequenced phage. The mean value is shown as a dotted line. (J) Predicted positions of mutated residues in M-MLV from v1 (yellow) or v2 (blue) PANCE. The structure is from the highly homologous XMRV (PDB: 4HKQ). (K) Overnight propagation of pools of wild-type RT and evolved RT phage on their cognate or noncognate host-cell selection strains. Phage were from PANCE on the v1 circuit (yellow bars), from PANCE on the v2 circuit (blue bars), or wild-type-PE2 phage (gray bars). Propagation was then measured in the v1 circuit (left) or the v2 circuit (right). Bars reﬂect the mean of n = 3 independent replicates, and dots show individual replicate values (B, D, K). See also Figure S2. Cell 186, 3983–4002, August 31, 2023 3987 ll OPEN ACCESS A Resource B E G C D F H Figure 3. Phage-assisted evolution of compact RTs for prime editing (A) Summary of evolution campaigns for NpuC-Gs RT, NpuC-Ec48 RT, or NpuC-Tf1 RT phage in the v1 (yellow), v2 (blue), and v3 (purple) PE-PACE circuits. Whether an evolution was PANCE or PACE is speciﬁed. PANCE passages (p) or hours of PACE (h) are speciﬁed in parentheses. Arrowheads indicate increases in selection stringency. Mutants characterized in mammalian cells are denoted with a dot and labeled. Additional increases in stringency are in pink. (B) Position of residues in wild-type Gs RT (PDB: 6AR1) that were mutated during evolution. (C) Predicted positions of residues in Ec48 RT that were mutated during evolution. Residues are mapped onto the AlphaFold-predicted structure of Ec48 RT overlayed with the substrate of the XMRV RT (PDB: 4HKQ). (D) Predicted positions of residues in Tf1 RT that were mutated during evolution. Residues are mapped onto the AlphaFold predicted structure of Tf1 RT overlayed with the substrate of the Ty3 RT (PDB: 4OL8). (E) Prime editing using prime editors containing wild-type (gray) Gs, Ec48, and Tf1 RTs, evolved Gs-RT (evoGs, green), evolved Ec48 RT (evoEc48, blue), and evolved Tf1 RT (evoTf1, yellow) in HEK293T cells (n = 3 independent replicates). (F) Comparison of prime editors in the optimized PEmax architecture containing either engineered pentamutant Marathon RT (Marathon penta, red), evoEc48 (blue), or evoTf1 (yellow) with PEmax (gray) in HEK293T cells (n = 3 independent replicates). 3988 Cell 186, 3983–4002, August 31, 2023 (legend continued on next page) Resource ll OPEN ACCESS would increase the stringency of the PE-PACE circuit. We devel- oped a second circuit (v2, Figure S2D) in which a 20-bp insertion, instead of the 1-bp insertion used in the original v1 circuit, is required to enable phage propagation. We also speculated that evolving complete PE proteins, rather than only the RT domain, may yield Cas9 mutations that enhance PE outcomes. We therefore removed the P2 plasmid from the host E. coli and encoded the entire prime editor protein, including the Cas9 nickase domain, on the phage without the use of a host P2 plasmid or split inteins (Figure 2G). To study the effects of the target edit on evolutionary outcomes, we designed a comparative PANCE experiment evolving the same whole-editor PE2 phage using the v1 or v2 circuit (Fig- ure 2H). Since different outcomes can emerge even from identical selection conditions,34 we performed multiple replicates of each selection. After 31 PANCE passages in six v1 lagoons and ﬁve v2 lagoons, we observed that mutations were shared among PANCE replicates for a given edit but differed greatly between la- goons that were required to perform the two different edits (Table S2A; Figures 2H and 2I). Mutations evolved in our v2 circuit were more numerous and also located closer to the polymerase’s active site, whereas residues evolved in the v1 circuit were typi- cally surface exposed (Figures 2I and 2J). These ﬁndings demon- strated that the target edit during PE-PACE strongly affects the re- sulting genotypes, suggesting that the most efﬁcient prime editors may specialize in speciﬁc types of edits. To investigate this possibility, we performed overnight propa- gation of phage evolved in the 1-bp insertion or 20-bp insertion selection on either the matched or mismatched evolution strain. When phage were evaluated in the strain in which they were evolved, their propagation improved compared to starting whole-editor PE2 phage; however, when evolved phage were evaluated in a strain requiring the other edit, they propagated less well than the parental PE2 phage (Figures 2K and S2E). These data further conﬁrmed that prime editors evolved proper- ties that specialize in their respective edits, and thus different prime editors will likely be best for different types of edits. We combined the above insights, as well as other recent PE improvements, to design a v3 PE-PACE circuit that used engi- neered pegRNAs (epegRNAs),4 which broadly improve PE by protecting pegRNAs from cellular degradation, to correct a different 20-bp deletion in T7 RNAP (Figure S2F). We used the v1, v2, and v3 PE-PACE circuits to evolve several different RTs below. Evolution of compact RTs We ﬁrst applied PE-PACE to evolve RTs that are substantially smaller than the PE2 RT, including the Geobacillus stearother- mophilus GsI-IIC intron RT (Gs RT), as well as the Ec48 and Tf1 RTs engineered above (Figure 1). The various evolutionary trajectories pursued are summarized below and in Figure 3A. We began by evolving the weakly active Gs RT (Figure 1C) us- ing 12 passages of PANCE in the v1 circuit, followed by either 100 h in the v1 PACE circuit or 23 passages in the v2 PANCE cir- cuit. Evolution improved phage propagation (Figures S3A–S3C), and sequencing the evolved Gs RT phage showed a high degree of predicted structural convergence (Tables S2B and S2C; PDB: 6AR1)35: each clone harbored mutations (N12D, A16E/V, L17P, L37P/R, R38H, I41N/S, and/or W45R) that are predicted to perturb the interaction between two alpha-helices of Gs RT’s N-terminal extension (Figure 3B). One of these helices protrudes into the major groove of the DNA/RNA duplex substrate, sug- gesting that these mutations may improve substrate binding. We next evolved the compact Ec48 RT (Figure 1C) using 29 passages of v1 PANCE and 23 passages of v2 PANCE. We increased v2 selection stringency by decreasing the expression of T7 RNAP and evolved the phage for 20 additional passages, yielding high levels of convergence (Tables S2D–S2F). Three mu- tations (E60K, E279K, and K318E) are predicted to be proximal to the DNA,RNA substrate (Figure 3C), suggesting that they also may alter substrate binding. Finally, we evolved the Tf1 RT using 29 PANCE passages in the v1 circuit, 23 passages in the v2 circuit, and 25 passages in the v3 circuit. In the v3 circuit, we increased selection stringency by decreasing the PBS length from 7 to 4 nucleotides (nt). Several of the resulting converged mutations (K118R, I128V, K413E, and to the DNA,RNA substrate in the S492N) are proximal AlphaFold-predicted Tf1 structure, while others (P70T, G72V, M102I, and K106R) may interact with the RTT of the pegRNA (Fig- ure 3D; Tables S2G–S2I). Our previous observation that K118R improves PE efﬁciency in HEK293T cells (Figure 1E) validates that at least some of the evolved mutations improve mammalian cell editing outcomes. Collectively, these data demonstrate that PE-PANCE enables the rapid, parallel evolution of improved prime editors and is generalizable to diverse RTs. Mammalian cell characterization of compact evolved RTs We evaluated evolved Gs RT, Ec48 RT, and Tf1 RT variants (evo- Gs, evo-Ec48, and evo-Tf1, respectively) as prime editors in HEK293T cells. Across six different edits at endogenous genomic loci using the PE3 system, evolved RTs greatly outper- formed their wild-type RT counterparts. We observed a 6.2-fold average improvement for evo-Gs, a 22-fold improvement for evo-Ec48, and a 2.7-fold improvement for evo-Tf1 (Figure 3E). Among these RTs, evo-Tf1 offered the highest average editing efﬁciency, and evo-Ec48 was the most compact RT (1.2-kb gene size). We further characterized these two enzymes in the PEmax (G) Prime editing in primary human T-cells at commonly edited test loci (n = 4 independent replicates). Indel-free editing is shown in blue or pink, and indels are shown in gray. (H) Correction of the HEXA 1278insTATC mutation that causes Tay-Sachs disease in a HEK293T cell line model previously engineered to harbor the mutation (left) and in patient-derived ﬁbroblasts (right). n = 3 independent replicates were used for the HEK293T cell line model. n = 2 independent replicates were used for the patient-derived ﬁbroblasts. For B-D, the DNA substrate is green, RNA substrate is yellow, residues mutated following PANCE in the v1 circuit are blue, residues mutated following PANCE in the v2 circuit are red, and residue mutated following PANCE in the v3 circuit is orange. For (E–H), bars show the mean value for the speciﬁed number of replicates, and dots show individual replicate values. See also Figure S3. Cell 186, 3983–4002, August 31, 2023 3989 ll OPEN ACCESS Resource A D F H B C E G I J Figure 4. Development of dual-AAV compatible RT variants for installing long, complex edits (A) Summary of evolution and engineering campaigns used to generate PE6c and PE6d. (B) Conserved mutations from M-MLV RT evolution. The structure of XMRV RT (PDB: 4HKQ), which is highly homologous to M-MLV shows PACE-evolved residues (blue) lie close to the enzyme active site (dark gray) and DNA/RNA duplex substrate (pink/purple). An incoming dNTP, modeled by alignment with PDB: 5TXP, is shown in yellow. Below, pink lines indicate locations in the M-MLV RT at which PACE-evolved mutations truncated the protein. (C) Fold-change in editing efﬁciency relative to PEmax for PEmaxDRNaseH, PE6c, and PE6d in HEK293T cells. Individual replicates are plotted, with n = 3 biological replicates per edit. 3990 Cell 186, 3983–4002, August 31, 2023 (legend continued on next page) Resource ll OPEN ACCESS architecture, which improves codon optimization, linkers, and nuclear localization signals.2 We compared these evolved prime editors to PEmax (2.2 kb) and PEmaxDRNaseH (1.5 kb), as well as the previous state-of-the-art size-minimized (1.2 kb) Mara- thon pentamutant RT engineered by Joung and coworkers21 at six genomic loci using epegRNAs in HEK293T cells. Evo-Ec48 outperformed the engineered Marathon pentamutant21 by 3.7-fold on average and approached PEmax performance levels, averaging 80% of PEmax editing efﬁciencies across the eight edits tested (Figures 3F and S3D). Since evoEc48 is 810 bp smaller in gene size than the engineered M-MLV RT in PE- max, 270 bp smaller than the DRNaseH form of M-MLV, and more efﬁcient than the size-equivalent Marathon pentamutant, we recommend evo-Ec48’s use for PE applications in which the size of the prime editor must be minimized. The use of epegRNAs is important for achieving efﬁcient PE with evo-Ec48 (Figure S3E). We designated the evo-Ec48 RT-derived prime editor as PE6a. Evo-Tf1 on average supported PE levels equal to those of PEmax at the eight edits tested (Figures 3F and S3D). The evo-Tf1 RT-derived prime editor hereafter is designated PE6b. Both PE6a and PE6b are typically less efﬁcient at longer, complex edits (Figure S3F). To examine PE6a and PE6b variants in a therapeutically rele- vant cell type, we compared them to their wild-type RT counter- parts, the Marathon pentamutant, and PEmax in primary human T cells at two loci following electroporation of the corresponding PE mRNA and pegRNA. For a 15-bp deletion at DNMT1, wild-type Ec48 was minimally active (0.22% average editing ef- ﬁciency), and the Marathon pentamutant yielded 3.3% average editing. The similarly sized PE6a supported 47% average edit- ing, a 211-fold improvement over wild-type Ec48 and a 14-fold improvement over the Marathon pentamutant. PE6a performed as well as or better than PEmax (Figure 3G). Similarly, PE6b offered large improvements over its wild-type RT counterpart, yielding an 8-fold improvement in editing efﬁciency over PE us- ing wild-type Tf1, comparable to that of PEmax (Figure 3G). We observed similar trends for a substitution edit at VEGFA. PE6a and PE6b thus can offer editing efﬁciencies similar to those of PEmax (Figure 3G) in primary human T cells. We also evaluated PE6a and PE6b in HEK293T cells harboring the HEXA 1278insTATC mutation that causes Tay-Sachs dis- ease.1,4 Treatment of this cell model with PE6a and PE6b and an epegRNA programmed to delete the pathogenic TATC inser- tion in HEXA yielded 33% and 42% correction, respectively, of the pathogenic mutation. These values are similar to the 41% correction generated by PEmax (Figure 3H). We then electropo- rated either PE6a, PE6b, or PEmax mRNA along with the neces- sary epegRNA and nicking sgRNA into Tay-Sachs disease pa- tient-derived ﬁbroblasts harboring the 1278insTATC mutation. PE6a, PE6b, and PEmax yielded 16%, 53%, and 46% average HEXA correction, respectively—all above the 2% threshold for therapeutic relevance36 (Figure 3H). Overall, these ﬁndings establish that size-minimized, non-M- MLV RTs can approach or exceed PEmax’s editing efﬁciencies while also offering substantially smaller gene sizes (1.2 kb and 1.5 kb for PE6a and PE6b vs. 2.2 kb for PEmax). PE6a and PE6b are the ﬁrst enzymes in a suite of improved PE6 variants (PE6a-g) developed in this study. To simplify nomenclature, we deﬁne PE6 variants as prime editor proteins in the PEmax archi- tecture. When used for PE, the use of a nicking sgRNA is assumed unless stated otherwise, while the use of MLH1dn (which can enhance PE efﬁciency by inhibiting cellular mismatch repair in the PE4 and PE5 systems)2 is not assumed and is spec- iﬁed on a case-by-case basis. Evolution and engineering of highly active AAV- compatible RTs Next, we combined PE-PACE with protein engineering to generate prime editors that are the same size as PEmaxDRNa- seH, but better support long, complex edits. To create a highly active Tf1 RT, we combined mutations in the evolved Tf1 RT (PE6b) with rationally designed mutations used in rdTf1. The re- sulting engineered and evolved Tf1 variant, PE6c, harbors sixteen mutations from evolution and rational engineering (Figure 4A). To create a highly active, truncated M-MLV RT, we evolved the PE2 RT in the v1, v2, and v3 circuits in parallel and compared mu- tations emerging from each evolution (Figure 4A). Interestingly, explicit deletion of the RNaseH domain was not necessary, as many evolved M-MLV RT variants contained mutations such as Q492stop that truncated the RT between its polymerase domain and RNaseH domain (Figure 4B).21,25,26 In addition to these (D) Editing efﬁciencies of PEmaxDRNaseH and PE6d at the HEK3 +1 loxP insertion edit (pink) and the HEK3 +1 FLAG insertion edit (orange) in HEK293T cells. The NUPACK-predicted structures of the RTT and PBS extensions for each edit is shown. (E) Results of a TdT assay on the HEK3 +1 loxP insertion edit in HEK293T cells. The y axis indicates the percentage of total RT products of a given length, and the x axis represents the length of the product in base pairs. PEmaxDRNaseH is shown in gray, and PE6d is shown in blue. The lines are mean values from n = 3 biological replicates. The pink box indicates DNA bases templated by the structured portions of the pegRNA. (F) Editing efﬁciencies of PEmaxDRNaseH (gray) and PE6d (blue) at an example engineered hairpin edit and its corresponding unpinned control in HEK293T cells. The sequence of the RTT is shown, with point mutations in the unpinned control shown in red. The NUPACK-predicted structures of the RTT and PBS extensions for each edit is shown. (G) Relationship between pegRNA RTT/PBS secondary structure and PE6d improvements. The y axis reﬂects the fold-improvement of PE6d over PE- maxDRNaseH. The x axis is the absolute value of the free energy of pegRNA folding as measured by NUPACK. Each dot represents one edit in HEK293T cells that was calculated from the mean values from n = 3 biological replicates. See Figure S4D for individual editing values and edit identities. (H) Comparison of evolved and engineered RTs to PEmaxDRNaseH at typical twinPE edits in HEK293T cells. Solid bars indicate editing efﬁciency. Striped bars indicate indels. (I) TwinPE-mediated insertion of the 38-bp attB sequence into the Rosa26 locus in N2a cells. Indel-free editing is shown in yellow, and indels are shown in gray. (J) PE-mediated insertion of a 42-bp sequence containing loxP into the Dnmt1 locus in N2a cells. Indel-free editing is shown in yellow, and indels are shown in gray. For D, F, and H-J, bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. See also Figure S4. Cell 186, 3983–4002, August 31, 2023 3991 ll OPEN ACCESS RNaseH-truncating mutations and the ﬁve engineered mutations1 already present in PE2 compared to wild-type M-MLV RT, over 20 additional mutations emerged (Tables S2J–S2L). One cluster of mutations emerging from the v2 and v3 evolutions was particularly promising (Figure 4B): T128N, V129A/G, P196S/T/F, N200S/Y, and V223A/M/L/E all lie near the polymerase active site. Addition- ally, we previously installed D200N to create PE2 from the wild- type M-MLV RT,1 and V223 is part of the core YXDD motif that has been implicated in the activities of various RTs.37 We tested evolved and engineered mutations at then combined the most promising candidates to generate an RNa- seH-truncated evolved and engineered M-MLV variant that we designated PE6d (Figure S4A). these residues, Dependence of PE6c, PE6d, and PEmaxDRNaseH performance on RTT secondary structure We compared PE6c, PE6d, and PEmaxDRNaseH—three editors small enough to be compatible with dual-AAV delivery25,26 —as well as full-length PEmax, at several longer prime edits and twinPE edits in HEK293T cells. Importantly, PE6c and PE6d recovered PE efﬁciency for long edits compared to PEmaxDR- NaseH, matching or exceeding PEmax’s editing efﬁciency for all four tested edits (Figure 4C). We noted, however, that PEmaxDRNaseH did not always exhibit deﬁciencies at long edits compared to PEmax, PE6c, and PE6d, and RTT length alone did not fully account for the performance dif- ferences between prime editors. For instance, both the HEK3 +1 FLAG insertion and the HEK3 +1 loxP insertion pegRNAs require the use of a long RTT (58 bp and 74 bp, respectively) and have iden- tical spacer and PBS sequences, but the relative efﬁciency of PE- maxDRNaseH versus PE6d differed substantially between the two edits. While both editors performed comparably at the FLAG inser- tion, PE6d offered 1.9-fold higher editing efﬁciency than PE- maxDRNaseH for the loxP insertion (Figure 4D). To probe this discrepancy, we examined the predicted sec- ondary structure of the two pegRNAs’ 30 extensions using NUPACK38 and found that the FLAG insertion pegRNA 30 exten- sion is predicted to be largely disordered, whereas the loxP insertion 30 extension contains a strong predicted 13-bp hairpin (Figure 4D). A terminal deoxynucleotidyl transferase (TdT) assay1,4 (Figure S4B) further revealed that for the loxP insertion, 30% of products generated by PEmaxDRNaseH were prema- turely truncated at hairpin-templated bases, whereas only 5.8% of products generated by PE6d were prematurely trun- cated at these positions (Figure 4E). As a result, PE6d produced a larger proportion of full-length DNA ﬂaps that contained the entire RTT-encoded sequence (62% of PE6d RT products versus 34% of PEmaxDRNaseH RT products [Figure 4E]). In contrast, at the HEK3 FLAG insertion edit for which the two ed- itors performed similarly, PEmaxDRNaseH and PE6d both mostly produced full-length ﬂaps (70% and 78% of RT products, respectively [Figure S4C]). These data suggest a mechanism for the effect of RTT second- ary structure on editing efﬁciency: RNaseH domain truncation, which decreases enzyme processivity,39 increases the genera- tion of prematurely terminated, unproductive, RT products when faced with a highly structured RTT substrate. The polymer- ase domain mutations in PE6d (and certain other variants) 3992 Cell 186, 3983–4002, August 31, 2023 Resource enhance RT processivity and can compensate for the lack of the RNaseH domain, supporting full-length product formation even when the pegRNA RTT has substantial secondary structure. To test this hypothesis, we engineered a series of pegRNAs predicted to contain long, stable hairpins, as well as ‘‘unpinned’’ control pegRNAs in which 2–4 point mutations strongly disrup- ted pegRNA secondary structure. PE6d outperformed PE- maxDRNaseH when RTTs contained strong hairpins, yielding a 2.3-fold average improvement in editing efﬁciency (Figures 4F and S4D). the two prime editors performed comparably for the corresponding unpinned control RTTs. These results conﬁrm that secondary structure, rather than RTT length alone, determines the relative efﬁciencies of PE6d and PEmaxDRNaseH. In contrast, To establish a simple predictive method to identify which compact PE is best for a given edit, we analyzed many prime edits including the hairpin tests above and compared the rela- tionship between the NUPACK-predicted free energy of RTT and PBS folding and the difference in editing efﬁciency between PE6d and PEmaxDRNaseH. When the predicted free energy of folding was stronger than (cid:3)23 kcal/mol, PE6d offered substan- improvements compared to PEmaxDRNaseH (Figure 4G). tial This relationship provides a useful guideline for when to use PE6d over PEmaxDRNaseH. When the predicted folding free energy of the RTT and PBS was weaker than (cid:3)23 kcal/mol, PE6d tended to yield lower ed- iting efﬁciencies and higher indel frequencies than PEmaxDR- NaseH (Figures 4G and S4E). Upon examining the PE6d-medi- ated indels, we discovered that PE6d catalyzed an increased rate of pegRNA scaffold insertion relative to PEmaxDRNaseH when a short, unstructured RTT was used (Figure S4F). Scaf- fold insertion is a byproduct of PE in which reverse transcrip- tion of the sgRNA scaffold produces undesired bases at the end of the genomic DNA ﬂap1; these extra bases are typically removed by cellular nucleases, but they can impede ﬂap equil- ibration or generate indels, especially if some scaffold nucleo- tides share adventitious homology with the target site. PE var- iants that overcome RTT secondary structure can also increase this type of undesired byproduct, leading to reduced precise editing for short-RTT edits. PE6d is therefore not well suited for most small prime edits. Interestingly, we did not observe general increases in indels (Figures 4H–4J) or scaffold insertion (Figures 4E and S4C) when PE6d was used with a long, struc- tured RTT. We speculate that the RTT itself acts as a barrier to reduce reverse transcription into the sgRNA scaffold. Thus, PE6d and other processive RTs do not generally increase in- dels at the edit types for which they are most useful; instead, increases in scaffold incorporation occur when the RT is more processive than is required for a speciﬁc edit. This discovery yields key insights into PE. For a given edit, there is an optimal level of RT activity that balances successful generation of RTT-templated bases with minimization of reverse transcription into the sgRNA scaffold. This ﬁnding also agrees with our early PACE results and explains why RTs evolved in the v2 selection, which used a long RTT, became less ﬁt in the v1 selection, which uses a short RTT. We performed similar processivity analyses on Tf1 variants PE6b (which is less processive) and PE6c (which is more Resource A B ll OPEN ACCESS C D E F G H Figure 5. Characterization of PE6 variants compared with PEmax (A) Prime editing efﬁciencies of PE6c, PE6d, and PEmax at challenging twinPE edits in HEK293T cells. (B) Edit to indel ratios of PE6c, PE6d, and PEmax at sites shown in (A) in HEK293T cells. (C) Twin prime editing in primary human T-cells at the CCR5 safe harbor locus. Indel-free editing is shown in red, and indels are shown in gray. Bars reﬂect the mean of n = 4 independent replicates. Dots show individual replicate values. (legend continued on next page) Cell 186, 3983–4002, August 31, 2023 3993 ll OPEN ACCESS Resource processive) and found a similar relationship between these two enzymes (Figure S4D). While generally not as active as PE6d, PE6c outperformed PEmaxDRNaseH at most highly structured edits (Figure S4D). PE6b has a level of processivity similar to PE- maxDRNaseH, which makes it a promising candidate for the installation of edits that require a short, unstructured RTT. PE6c and PE6d should also improve most twinPE efﬁciencies, which typically use long RTTs. We therefore compared them to PE- maxDRNaseH at a variety of twinPE edits in HEK293T cells. PE6 variants indeed offered improvements in efﬁciency relative to PE- maxDRNaseH, with PE6c yielding a 1.6-fold average improvement across the ﬁve sites tested (Figure 4H). To minimize potential PCR bias that can arise during sample preparation for large twinPE edits,9 we applied unique molecular identiﬁers (UMI) to quantify a subset of twinPE edits to conﬁrm this improvement (Figure S4G). Importantly, PE6c and PE6d did not substantially alter the editin- g:indel ratio for these twinPE edits. We also examined the ability of PE6 variants to perform longer prime edits in two mouse genomic targets in N2a cells. For the twinPE-mediated insertion of the Bxb1 recombinase attB recog- nition sequence at the murine Rosa26 safe harbor locus, PE- maxDRNaseH generated on average 31% installation of the edit but also yielded an equal number of indels. Conversely, PE6c and PE6d both increased editing efﬁciency and decreased indel rates at this site, with PE6d yielding an 8.6-fold increase in the editing:indel ratio for this edit (Figure 4I). Similarly, we opti- mized a strategy for the PE-mediated installation of a loxP sequence at the murine Dnmt1 locus. Compared to PEmaxDR- NaseH, PE6d enhanced editing efﬁciency by 2.1-fold and increased the editing:indel ratio by 1.7-fold (Figure 4J). These data further support that highly processive RTs do not substan- tially increase indel levels for long, structured RTTs. Overall, these results indicate that among dual-AAV compatible editors, PE6c and PE6d offer substantial improvements over PEmaxDR- NaseH for several types of challenging edits. PE6 variants with different processivities offer improvements over PEmax Next, we compared PE6 variants with PEmax. Given PE6c and PE6d0s enhanced processivity, we wondered if they might offer im- provements over PEmax for longer prime edits. We therefore tested PEmax, PE6c, and PE6d using six 38- to 108-bp insertion twinPE edits at ﬁve loci in HEK293T cells and found that PE6 var- iants improved average editing efﬁciency by 1.4-fold over PEmax across these edits (Figures 5A and S5A) without altering the pre- cise edit:indel ratio (Figures 5B and S5B). We also tested PEmax and PE6 variants for attB insertion at the CCR5 safe harbor locus in primary human T cells. PE6c offered a 1.5-fold improvement in editing efﬁciency relative to PEmax, achieving an average attB insertion efﬁciency of 34% across T cells from four different donors (Figures 5C and S5C). These results conﬁrm that PE6 variants offer substantial im- provements for therapeutically relevant PE. Since we discovered that highly processive RTs can be detri- mental for the installation of edits that use short, unstructured RTTs (Figure S4E), we wondered if the same caveat applied to PE- max. Since PE6b and PEmaxDRNaseH have reduced RT proces- sivity compared to PEmax (as approximated by their lower perfor- mance for long edits), they might improve editing:indel ratios compared to PEmax for small, unstructured edits as a result of reduced pegRNA scaffold incorporation. We compared PE6b, PE- maxDRNaseH, and PEmax for ten edits using short, unstructured RTTs with NUPACK-predicted RTT free energies between 0 and (cid:3)12 kcal/mol. Both PE6b and PEmaxDRNaseH indeed offered more favorable edit:indel proﬁles than PEmax (Figures 5D, S5D, and S5E), and for every edit tested, PEmaxDRNaseH or a PE6 variant offered a higher editing:indel ratio than PEmax (Figure 5E). Examination of the indels for a subset of edits conﬁrmed that PE6b and PEmaxDRNaseH incorporated pegRNA scaffold bases less frequently than PEmax (Figure S5F). Collectively, these data indi- cate that PE6b and PEmaxDRNaseH are well-suited for edits with unstructured RTTs due to their lower processivity, which re- duces scaffold incorporation and improves edit:indel ratios. PE6b and PE6c offer improvements over PEmax for therapeutic edits An expanded set of prime editor options should increase the likelihood of ﬁnding a high-efﬁciency PE approach for speciﬁc therapeutic edits. We tested 77 pegRNAs40 (Table S3) that install disease-associated mutations into endogenous sites in HEK293T cells and transfected them along with plasmids encoding MLH1dn (but no nicking sgRNA) and PEmax, PE6b, or PE6c. On average, PE6b and PE6c modestly outperformed PEmax (Figure 5F; Table S3), but at 16 of the 77 sites tested, Tf1-dervied editors offered substantial improvements over PE- max (1.5-fold–3.1-fold, Figure 5F). We chose several edits for (D) Edit to indel ratios of PE6b and PEmaxDRNaseH normalized to that of PEmax in HEK293T cells. Individual replicates are plotted, with n = 3 biological replicates per edit. Lines reﬂect the mean across all edits and replicates. Individual editing efﬁciencies and indel levels are shown in Figures S5D and S5I. (E) Edit to indel ratios of prime editors at endogenous HEK293T sites. The editor with the highest edit:indel ratio was picked and plotted side-by-side with PEmax for each speciﬁc edit. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. Individual editing efﬁciencies and indel levels are shown in Figures S5D and S5E. (F) Prime editing efﬁciencies of PE6b and PE6c normalized to the editing efﬁciency of PEmax at 77 edits that install a pathogenic allele into endogenous sites in HEK293T cells. No nicking gRNA was used and MLH1dn plasmid was simultaneously transfected with prime editor plasmid for all conditions. All values from n = 3 replicates are shown. Lines reﬂect the mean across all edits and replicates. Prime editing efﬁciencies for edits where PE6b or PE6c outperformed PEmax by more than 1.5-fold are shown on the right. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (G) Correction of pathogenic mutations implicated in Crigler-Najjar Syndrome, Bloom Syndrome, and Pompe disease in HEK293T cell models using PEmax, PEmaxDRNaseH, PE6b, and PE6c. (H) Correction of mutations implicated in Crigler-Najjar Syndrome (UGT1A1) and Bloom Syndrome (RECQL3) in patient-derived ﬁbroblast using PE6c and PEmax. Bars reﬂect the mean of n = 3 independent replicates for treated samples and n = 1–3 replicates of an untreated control for editing (red) and indels (gray). Dots show individual replicate values. For A, B, and G, bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. See also Figure S5. 3994 Cell 186, 3983–4002, August 31, 2023 Resource which PE6b and/or PE6c improved editing efﬁciencies and added nicking guide RNAs that target the non-edited strand to enhance editing efﬁciency. For all of these edits, PE6b or PE6c continued to outperform PEmax without increasing indel levels beyond those of PEmax (Figures S5G and S5H). Similarly, to examine the potential utility of Tf1-derived editors for disease correction, we used Sleeping Beauty transposase41 to integrate pathogenic alleles known to cause glycogen storage disease II (Pompe Disease), Bloom Syndrome, or Crigler-Najjar Syndrome into the genomes of HEK293T cells. We evaluated PE- max, PEmaxDRNaseH, PE6b, and PE6c for their ability to correct each pathogenic mutation. For all three edits, PE6c generated the highest average editing efﬁciency (13–35%), a 2.1-fold average in- crease over PEmax across the three model cell lines (Figure 5G). We also tested PEmax and PE6c in ﬁbroblasts derived from Pompe Disease, Bloom Syndrome, and Crigler-Najjar Syndrome patients. PE6c-mediated improvements in indel-free editing efﬁ- ciencies were more pronounced in these patient-derived ﬁbro- blasts, yielding 1.9-fold–4.5-fold improvement over PEmax (Figures 5H, S5I, and S5J). Collectively, these data show that the PE6 RT variants generated in this study can repeatedly outperform PEmax in a variety of disease-relevant contexts and cell types. Evolution of Cas9 variants for enhanced prime editing During evolutions that used whole-editor phage, the Cas9 domain of the prime editor also acquired dozens of conserved mutations in the v1–v3 circuits (Figures 6A and S6A). Mutations that evolved in the Cas9 domain were dependent on the target used during evo- lution and were distributed across the entire Cas9 protein, without evident hotspots in any location (Tables S2M and S2N). However, evolved Cas9 mutants decreased editing efﬁciencies compared to PE2 in HEK293T cells (Figure 6B). Reversion analysis of evolved Cas9 mutants suggested that a subset of evolved mu- tations were driving lower mammalian cell editing efﬁciencies (Fig- ure S6B). To identify beneﬁcial and detrimental mutations, we dissected the effect of 163 individual Cas9 mutations in PEmaxDR- NaseH for two substitution edits in human and mouse cells (Fig- ure 6C; Table S4). Most mutations that strongly decreased editing efﬁciency at both mammalian targets (K1151E, A1034D, K1003E, and K1014E) are known to decrease the afﬁnity of Cas9 for DNA, or are predicted to do so based on structures of Cas9 complexed with DNA42–46 (Figure S6C; Table S2M). We hypothesized that dur- ing PACE, Cas9 binding to a target gene can decrease the expres- sion of that gene through a bacterial CRISPRi mechanism,47 so high-afﬁnity binding to the corrected T7 RNAP gene after PE can lower ﬁtness. In mammalian cells, however, requirements for DNA binding are likely more stringent due to lower target site con- centration and competing DNA-binding proteins. Therefore, in mammalian cells, PE efﬁciency may suffer from weaker DNA bind- ing by Cas9. Indeed, we conﬁrmed that disrupting Cas9,DNA binding improved PE-PACE circuit activation in a prime editing-in- dependent manner (Figure S6D). Engineering Cas9 variants for enhanced prime editing Having identiﬁed and rationalized the enrichment of detrimental Cas9 mutations, we next combined Cas9 mutations beneﬁcial to PE. The single-mutant Cas9 assays identiﬁed mutants such as H99R, E471K, I632V, D645N, R654C, H721Y, K775R, and ll OPEN ACCESS K918A that maintained or modestly increased mammalian PE efﬁ- ciency (Figure 6C; Table S2N). To create Cas9 variants that can better enhance mammalian PE efﬁciency, we tested these muta- tions in combinations to identify the best-performing evolved and engineered Cas9 variants, designated PE6e-g (Figure 6D). We compared these mutants to parental PEmaxDRNaseH across a wider array of editing conditions and target sites in HEK293T cells and N2a cells (Figures 6D and S6E). At ﬁve of the 13 sites tested, PE6e-g variants improved PE efﬁciency, supporting up to 1.8- fold improvement in average editing efﬁciency compared to PE- maxDRNaseH. This result demonstrates that PE6 Cas9 variants are capable of improving mammalian PE efﬁciency for some edits. For other edits, however, PE6e-g did not improve or even decreased editing efﬁciencies compared to PEmaxDRNaseH (Figures 6D and S6E). In contrast with evolved RT domains, we did not observe a clear relationship between characteristics of the edit and the beneﬁts of different Cas9 mutants. Nevertheless, the location of the PE6 Cas9 mutations suggest potential explana- tions for their site-speciﬁc beneﬁts to PE. The K775R and K918A mutations are located in Cas9’s L1 and L2 linkers, which are involved in R-loop stabilization and also mediate conformational changes in the HNH domain upon DNA binding.48,49 The H721Y mutation appears to impact binding to the sgRNA scaffold (Fig- ure S6F). Therefore, features speciﬁc to a target site’s R-loop or pegRNA may account for the observed site-dependent effects. We recommend screening PE6e-g, in addition to the Cas9 domain in PEmax, when optimizing a PE strategy for a site of interest. If only one Cas9 mutant can be tested in addition to the PEmax Cas9, PE6e is the variant most likely to yield improvements (Figure 6D). Combining PE6 RT and Cas9 mutants To maximize PE efﬁciencies, evolved RT and Cas9 variants can be evaluated separately and then combined. For example, the size-minimized PE6a RT exhibits lower editing efﬁciencies than PEmax at the CXCR4 and IL2RB loci (Figure 6E), but the evolved PE6e Cas9 improves PE efﬁciency at those loci (Figure 6D). Combining these two domains (PE6a/e), restores PE efﬁciency to near-PEmax levels, while maintaining the small size of the PE6a RT (Figure 6E). Additionally, Cas9 and RT domains that both enhance editing efﬁciency for an edit can be combined: the RT domain of PE6c and the Cas9 domain of PE6g improve twin PE efﬁciency for the recoding exon 4 of the PAH gene. When these domains are combined to generate PE6c/g, the ben- eﬁts to editing efﬁciency were additive, yielding a 2.9-fold improvement over PEmaxDRNaseH (Figure 6F). These results demonstrate that PE6 RT domains and Cas9 domains can be treated modularly to overcome deﬁcits in one domain or yield cumulative improvements from both domains. Recommendations and applications of PE6 mutants The suite of prime editors engineered and evolved in this study (PE6a–g) offer improvements in editor size (PE6a and b), RT activity (PE6c and d), and Cas9-dependent editing efﬁciency (PE6e–g). From this set of tools, the choice of prime editor variant for a given application is informed by editor size requirements and characteristics of the desired edit (Figure 6G). We recommend ﬁrst considering size constraints. When editor size must be minimized, PE6a—the smallest prime editor described to date—should be Cell 186, 3983–4002, August 31, 2023 3995 Resource B ll OPEN ACCESS A C E F D G Figure 6. Evolution and engineering of improved Cas9 domains for prime editing, and summary of PE6 recommended use cases (A) Summary of evolution campaigns for whole PE2 phage in the v1 (yellow), v2 (blue), and v3 (purple) circuits. Green shading indicates reversion analysis. PANCE passages (p) or hours of PACE (h) are in parentheses. Arrowheads indicate increases in selection stringency. Mutants characterized in mammalian cells are denoted with a dot and labeled. Additional increases in stringency are in pink. (B) Evaluation of PACE-evolved clones in HEK293T cells. EvoCas9-1 through evoCas9-4 were isolated from low-stringency evolution. EvoCas9-5 and evoCas9-6 were isolated from high-stringency evolution. (C) Assessment of individual Cas9 mutations on prime editing efﬁciency at two test sites. The y axis shows editing efﬁciency at the Pcsk9 +3 C to G / +6 G to C edit in N2a cells. The x axis shows editing efﬁciency for the RNF2 +5 G to T edit in HEK293T cells. Mutants incorporated into ﬁnal Cas9 variants are shown in green. Mutants previously shown to, or structurally predicted to, decrease Cas9 binding are shown in maroon. PEmaxDRNaseH is shown in orange. (D) Comparison of combined Cas9 mutants to PEmaxDRNaseH in HEK293T cells and N2a cells. Editing efﬁciencies of variants are normalized to the editing efﬁciency generated by PEmaxDRNaseH. Individual replicates are plotted, with n = 3 biological replicates per edit. (E) Comparison of PEmax, PE6a, and PE6a/e at two sites in HEK293T cells. (F) Comparison of PEmaxDRNaseH, PE6c, and PE6g in HEK293T cells. (G) Decision tree for selecting a PE6 variant. For secondary structure stability predictions, we recommend the NUPACK prediction tool38 with the RTT/PBS sequence as the input. For B, E, and F, bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. See also Figure S6. 3996 Cell 186, 3983–4002, August 31, 2023 Resource used. If editor size is restricted due to AAV delivery constraints but does not need to be strictly minimized, PEmaxDRNaseH and PE6b–d should be considered. If the target edit uses a pegRNA with a highly structured 30 extension (NUPACK-predicted free en- ergy of (cid:3)23 kcal/mol or more stable for the RTT and PBS) or is a twinPE edit, PE6c and PE6d are likely to be optimal. Conversely, if the target edit utilizes a largely unstructured 30 extension (NUPACK-predicted free energy of folding less stable than (cid:3)23 kcal/mol), PEmaxDRNaseH, PE6b, and PE6c should be examined. Finally, if no size constraints exist, PEmax can also be tested in addition to the four editors just discussed (Figure 6G). If an edit requires an unstructured RTT and scaffold insertion- derived indel levels are high when using PEmax, then PEmaxDR- NaseH and PE6b should be evaluated in order to reduce indels. Conversely, if an edit is a twinPE edit or a challenging PE edit, PE6c and PE6d may offer improvements over PEmax (Figure 6G). Although indel frequencies vary by site and by RT variant, when PE6 editors are applied to their recommended classes of edits, we do not observe any consistent increases in the proportion of in- dels. Regardless of the RT used, screening Cas9 variants from PE6e-g in combination with the optimized RT can further enhance editing efﬁciency (Figure 6G). PE6 variants enable longer and more complex edits in vivo via a dual-AAV delivery system Following the decision tree in Figure 6G, we used PE6 variants to perform long, complex prime edits in vivo. When using efﬁcient dual-AAV systems for in vivo prime editing,3,25–27 editors smaller than PEmax must be used in order for the PE protein, pegRNA, nicking RNA, and their regulatory elements to ﬁt within the pack- aging capacity of two AAVs ((cid:2)5 kb per AAV). Because PE6c and PE6d are the same size as PEmaxDRNaseH but substantially outperform PEmaxDRNaseH at highly structured edits in cell cul- ture, we reasoned that these trends may also facilitate edits requiring structured pegRNAs in vivo after dual-AAV mediated delivery (see STAR Methods for details). We ﬁrst tested if PE6 variants could enable dual-ﬂap PE in vivo, which has not been previously reported. To create a dual-AAV system for twinPE (v3em twinPE-AAV), we began with the archi- tecture described in our recently reported v3em PE-AAV prime editor delivery system25 (Figure 7A). In a universal N-terminal AAV, we encoded the majority of the Cas9 protein fused to an N-terminal Npu split intein. In a second C-terminal AAV, we en- coded a C-terminal Npu split intein fused to the remainder of the prime editor, using either PEmaxDRNaseH, PE6c, or PE6d (Figures 7A and S7A). In the C-terminal virus, we included two epegRNAs that are required for twinPE, instead of an epegRNA and a nicking sgRNA (Figure 7A). These epegRNAs encoded the installation of the Bxb1 integrase attB substrate sequence at the murine Rosa26 safe harbor locus. We also included 1010 vg of a GFP-KASH AAV to mark nuclei from transduced cells. We administered a low dose of both twinPE AAVs (4x1010 vg total, 2x1010 vg per virus) and the GFP AAV (1x1010 vg) via neonatal intracerebroventricular (P0 ICV) injections to C57BL/6 mice. Three weeks later, we isolated nuclei from the mice cortices and analyzed bulk (unsorted) or transduced (GFP-positive) nuclei (Figure S7B). Mice treated with PEmaxDRNaseH AAV showed 0.34% attB installation in bulk cortex and 0.89% attB installation ll OPEN ACCESS in transduced cells (Figure 7B). In comparison, PE6c yielded 4.5% and 5.1% insertion of the attB sequence in bulk and sorted nuclei, respectively (Figure S7C). PE6d generated 7.8% and 10.4% editing in bulk and sorted cells, respectively (Figure 7B). PE6d thus yielded an average 23-fold improvement in bulk cortex editing and an average 12-fold improvement in editing efﬁciency in transduced cells relative to PEmaxDRNaseH. This increase in editing efﬁciency was not accompanied by an increase in indels relative to PEmaxDRNaseH (Figure 7B). These data reinforce that PE strategies that were previously inefﬁcient in vivo can be achieved using PE6 variants, and establish a method for in vivo dual-ﬂap prime editing. We also tested the ability of PE6 variants to mediate large single- ﬂap insertions in vivo. We attempted the installation of a 42-bp loxP sequence at the murine Dnmt1 locus, having observed that PE6d outperformed PEmaxDRNaseH for this edit in cell culture (Fig- ure 4J). We used the v3em PE-AAV25 architecture with either PE- maxDRNaseH or PE6d. We administered PE-AAVs via P0 ICV in- jections using a higher dose of 1x1011 vg total (531010 vg per PE virus) or a lower dose of 231010 vg total (131010 vg per virus) along with a GFP-KASH AAV transduction marker. Three weeks after low-dose injection, loxP insertion in bulk cor- tex tissue was virtually undetectable when PEmaxDRNaseH was used (0.03% average editing [Figure S7D]). Sorting for transduced cells improved PEmaxDRNaseH-mediated average editing to 0.75%. Importantly, mice injected with a low dose of PE6d showed an average of 5.5% loxP insertion in bulk cortex and 17% among transduced cells (Figure S7D) an increase of 183-fold and 23-fold, respectively, compared to PEmaxDRNaseH. PE6d gener- ated just 0.45% indels and 0.25% indels in bulk and transduced cortex, respectively, leading to an editing:indel ratio of 12:1 in bulk cells and 69:1 among transduced cells (Figure S7D). Following the higher dose, PEmaxDRNaseH’s editing efﬁ- ciency remained inefﬁcient, generating 1.7% and 2.4% loxP installation in bulk and transduced cells, respectively (Figure 7C). In contrast, PE6d generated an average of 40% and 62% loxP insertion in bulk and transduced cells, respectively, while main- taining low indel levels (1.6% in bulk tissue and 4.2% in trans- duced cells [Figure 7C]). These results not only represent a large (>23-fold) improvement over PEmaxDRNaseH in both bulk and transduced cells, but also establish a high editing:indel ratio of 23:1 in bulk cells and 14:1 in transduced cells for PE6d. To examine whether the more active RT used in these in vivo experiments increased off-target PE, we analyzed the top ten CHANGEseq-nominated off-target loci for the Dnmt1 pegRNA protospacer26,50 for the high-dose treated animals. For both PE- maxDRNaseH-treated and PE6d-treated animals, we did not detect any off-target modiﬁcations (Figure S7E). These results collectively demonstrate that while PEmaxDRNaseH cannot support the efﬁcient in vivo installation of difﬁcult, structured PE or twinPE edits, PE6 variants make these changes possible without generating substantial indels or off-target edits. DISCUSSION In this study, we addressed three key challenges facing PE. First, we developed PE6a and PE6b, which are 516–810 bp smaller in gene size than the M-MLV RT and can support Cell 186, 3983–4002, August 31, 2023 3997 ll OPEN ACCESS A B Resource C Figure 7. PE6 variants enable longer and more complex prime edits in vivo (A) Schematic showing a dual-AAV delivery system for twinPE (v3em twinPE-AAV). In the N-terminal AAV, production of the N-terminal portion of Cas9 (yellow) fused to an N-terminal Npu split intein (orange) is regulated by the Cbh promoter (green) and the SV40 late polyA signal (tan). In the C-terminal AAV, the C-terminal Npu split intein (dark green) is fused to the remainder of the prime editor (Cas9, yellow and RT, purple). The SV40 late polyA signal (tan), two epegRNAs (light and dark blue), AAV ITRs (black) are also shown. (B) Injection route and twinPE editing efﬁciency of PEmaxDRNaseH and PE6d viruses in the for the twinPE-mediated insertion of a 38-bp attB sequence at murine Rosa26 in the mouse cortex. N- and C- terminal twinPE viruses are administered via ICV injection (4x1010 vg total) along with a GFP-KASH virus. Editing efﬁ- ciencies (light and dark blue) and indel frequencies (black and gray) are shown to the right. Bars reﬂect the mean of n = 3–4 mice. Dots show individual mice. (C) Injection route and PE editing efﬁciency of PEmaxDRNaseH and PE6d viruses for the installation of a 42-bp insertion containing loxP at the Dnmt1 locus in the mouse cortex. (Left) The C-terminal virus is modiﬁed to include one epegRNA and one nicking sgRNA to encode a PE edit as opposed to a twinPE edit. (Right) Editing efﬁciencies (light/dark pink) and indel rates (black/gray). Bars reﬂect the mean of n = 3 mice. Dots show individual mice. See also Figure S7. state-of-the-art PE efﬁciencies. Second, to generate highly active, dual-AAV compatible editors, we used evolution and en- gineering to produce Tf1-derived PE6c and M-MLV-derived PE6d. Third, we developed multiple strategies for improving editing outcomes over those produced by PEmax. For chal- lenging edits such as those requiring highly structured RTTs, PE6c and PE6d can offer beneﬁts over PEmax; and conversely, for short, unstructured RTTs, indels and scaffold insertion prod- ucts generated by PEmax can be reduced by using PEmaxDR- NaseH or PE6b. Finally, both Tf1 RT-derived PE6b and PE6c offer different substrate preferences than M-MLV RT-derived editors and can substantially improve editing over PEmax at several therapeutically relevant loci. Evolved and engineered Cas9 domains in PE6e-g can further enhance PE efﬁciencies at some sites. Recommended use cases for PE6 variants are provided in Figure 6G. In addition to PE6 editors, this study generated insights that deepen our understanding of PE. By examining differences be- tween PE6 variants and PEmaxDRNaseH, we discovered that pegRNA extension folding energy is a determinant of PE efﬁciency. The protospacer-dependent effects from Cas9 mutants that emerged from our selection also raise interesting questions about the target-speciﬁc impact of pegRNA binding and R-loop stabiliza- tion on PE. The PE-PACE platform also enables future investigations. The edit-dependent requirements shown here suggest that bespoke prime editor evolution on speciﬁc high-impact targets could pro- duce optimal PE systems for those targets. PE-PACE could easily be manipulated for target sequence context-speciﬁc selections, which our lab has recently reported for base editing.51 PE-PACE could also be used to improve the PE activity of other Cas9 or RT orthologues.52 The RTs successfully evolved in this study 3998 Cell 186, 3983–4002, August 31, 2023 Resource span four different classes (Group II intron, retron, long terminal repeat retrotransposon, and retrovirus), suggesting that PE- PACE will yield additional advances when applied at scale to the 80,000 reported RT genes in this enzyme superfamily. Finally, PE6c and PE6d enable longer and more complex inser- tions to be effectively installed in vivo via dual-AAV delivery. They offer an order-of-magnitude improvement compared to a previ- ous state-of-the-art editor, PEmaxDRNaseH, and support in vivo dual-ﬂap PE. Even for non-viral delivery methods in which gene size is not strictly limited, PE6a-d could facilitate critical processes such as the in vitro synthesis of editor mRNA or the packaging of editor proteins into liposomes or engineered virus-like particles.53 The installation of insertion edits in the CNS is a particularly difﬁcult challenge in genome editing. Homology-dependent methods such as SLENDR and homology-independent methods such as HITI have been used,54,55 but rely on DSBs that can lead to indels. The efﬁcient editing and low indels achieved in this study, combined with the distinct DNA repair pathways required for PE-based approaches relative to other approaches, suggest PE6 variants will be valuable tools for in vivo editing. Finally, both in vivo edits shown in this study involve the insertion of a recom- binase recognition sequence. These results thus lay the founda- tion for programmable, DSB-free whole gene insertion in vivo when paired with a recombinase and donor DNA. Limitations of the study One remaining challenge is how to easily predict which edits will beneﬁt from the use of each PE6 variant. We have addressed this problem for some variants: for dual-AAV compatible prime editors, the degree of predicted pegRNA secondary structure can be used to determine whether PEmaxDRNaseH or a PE6 variant should be used. For other scenarios, however, guidelines are not as clear. For example, we have demonstrated that Tf1-derived RTs and Cas9 mutants can offer large improvements in editing efﬁciency com- pared to PEmax, but these gains are not observed across all target sites and edits. Library-based studies40,56–58 of RT and Cas9 vari- ants and machine learning models that facilitate a priori prediction of the best PE variant for a given application may further advance our understanding of these editors. Finally, while in vivo twinPE edit- ing efﬁciencies remained lower than in vivo PE editing efﬁciencies (here, 10.4% versus 62%), techniques such as increasing dose or extensively optimizing a twinPE dual AAV architecture may be needed to further enhance in vivo dual-ﬂap PE efﬁciencies. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d KEY RESOURCES TABLE d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d EXPERIMENTAL MODEL AND SUBJECT DETAILS B Mammalian cell culture conditions B Generation of HEK293T models of Tay-Sachs disease ll OPEN ACCESS B Generation of HEK293T model cell lines for Bloom Syn- drome, Crigler-Najjar disease, and Pompe Disease B Isolation and culture of primary human T cells d METHOD DETAILS B General methods and molecular cloning B Phylogenetic tree analysis B Bacteriophage cloning B Preparation of chemically competent cells B Phage-based luciferase assay B Plasmid-based luciferase assay B Overnight propagation assay B Plaquing B Phage-assisted noncontinuous evolution (PANCE) B qPCR determination of PANCE and PACE titers B Phage-assisted continuous evolution (PACE) B Transfection of HEK293T, N2a, and Huh7 cells B HTS sample preparation B HTS analysis B In vitro transcription (IVT) of editor mRNA B Electroporation of patient-derived ﬁbroblasts B Electroporation of primary human T cells B TDT assay and analysis B Secondary structure preduction using NUPACK38 B UMI sample prep and analysis B AAV production B Animals B P0 ventricle injections B Mice tissue collection B Nuclear isolation and sorting B Analysis of off-target editing d QUANTIFICATION AND STATISTICAL ANALYSIS SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.cell. 2023.07.039. ACKNOWLEDGMENTS This work was supported by US National Institutes of Health (NIH) grants UG3AI150551, U01AI142756, R35GM118062, RM1HG009490, R01EB027793, and R01HL56067; the Bill and Melinda Gates Foundation; the St. Jude Collabo- ration Research Consortium, the Friedreich’s Ataxia Accelerator, and the Howard Hughes Medical Institute. J.L.D. is supported by the Hertz Foundation. J.L.D., A.R., P.B.R., and K.A.E. are supported by the NSF Graduate Research Fellowship Program. M.E.N. is supported by the Ruth L. Kirschstein National Research Ser- vice Awards Postdoctoral Fellowship (GM143776-02). M.F.R. received funding from the HHMI Hanna Gray Fellowship. M.J.O. receives funding from the Bill and Melinda Gates Foundation, the Saint Baldrick’s Foundation, and the Kidz1st- Fund. J.T. receives funding from NIH grant 5R01AR063070-08. We thank Travis Blum for helpful discussions. Biorender was used to create ﬁgures. AUTHOR CONTRIBUTIONS S.P. and J.L.D. contributed equally and both designed and performed protein engineering, evolution, and mammalian cell experiments. M.E.N. and K.T. as- sisted with phage-based experiments. M.A., J.R.D., P.B.R., and Y.A.T. pro- duced AAV and performed mouse injections. A.M. performed primary T cell experiments. M.J.O. and J.T. supervised T cell experiments. A.R. generated the phylogenetic tree. X.D.G., M.F.R., S.B., and K.A.E. provided pegRNA and mRNA reagents. D.R.L. supervised the research. J.L.D., S.P., and D.R.L. drafted the manuscript with input from all authors. Cell 186, 3983–4002, August 31, 2023 3999 ll OPEN ACCESS DECLARATION OF INTERESTS J.L.D., S.P., and D.R.L. have ﬁled patent applications on aspects of this work. M.F.R. is an employee of Vertex Pharmaceuticals. J.R.D. is an employee of Prime Medicine. S.B. is an employee of Nvelop Therapeutics. M.J.O. receives compensation as a consultant for Agathos Biologics. D.R.L. is a consultant and equity holder of Beam Therapeutics, Prime Medicine, Pairwise Plants, Chroma Medicine, Resonance Medicine, Exo Therapeutics, and Nvelop Ther- apeutics. The authors have ﬁled patent applications on evolved and/or engi- neered prime editors and methods to generate them. INCLUSION AND DIVERSITY One or more of the authors of this paper self-identiﬁes as an underrepresented ethnic minority in their ﬁeld of research or within their geographical location. One or more of the authors of this paper self-identiﬁes as a gender minority in their ﬁeld of research. One or more of the authors of this paper self-identiﬁes as a member of the LGBTQIA+ community. One or more of the authors of this paper self-identiﬁes as living with a disability. One or more of the authors of this paper received support from a program designed to increase minority repre- sentation in their ﬁeld of research. Received: January 2, 2023 Revised: May 7, 2023 Accepted: July 28, 2023 Published: August 31, 2023 REFERENCES 1. Anzalone, A.V., Randolph, P.B., Davis, J.R., Sousa, A.A., Koblan, L.W., Levy, J.M., Chen, P.J., Wilson, C., Newby, G.A., Raguram, A., and Liu, D.R. (2019). Search-and-replace genome editing without double-strand breaks or donor DNA. Nature 576, 149–157. https://doi.org/10.1038/ s41586-019-1711-4. 2. Chen, P.J., Hussmann, J.A., Yan, J., Knipping, F., Ravisankar, P., Chen, P.-F., Chen, C., Nelson, J.W., Newby, G.A., Sahin, M., et al. (2021). Enhanced prime editing systems by manipulating cellular determinants of editing outcomes. Cell 184, 5635–5652.e29. https://doi.org/10.1016/j. cell.2021.09.018. 3. Liu, B., Dong, X., Cheng, H., Zheng, C., Chen, Z., Rodrı´guez, T.C., Liang, S.-Q., Xue, W., and Sontheimer, E.J. (2022). A split prime editor with un- tethered reverse transcriptase and circular RNA template. Nat. Bio- technol. 40, 1388–1393. https://doi.org/10.1038/s41587-022-01255-9. 4. Nelson, J.W., Randolph, P.B., Shen, S.P., Everette, K.A., Chen, P.J., Anz- alone, A.V., An, M., Newby, G.A., Chen, J.C., Hsu, A., and Liu, D.R. (2022). Engineered pegRNAs improve prime editing efﬁciency. Nat. Biotechnol. 40, 402–410. https://doi.org/10.1038/s41587-021-01039-7. 5. Zhang, G., Liu, Y., Huang, S., Qu, S., Cheng, D., Yao, Y., Ji, Q., Wang, X., Huang, X., and Liu, J. (2022). Enhancement of prime editing via xrRNA motif-joined pegRNA. Nat. Commun. 13, 1856. https://doi.org/10.1038/ s41467-022-29507-x. 6. Velimirovic, M., Zanetti, L.C., Shen, M.W., Fife, J.D., Lin, L., Cha, M., Akinci, E., Barnum, D., Yu, T., and Sherwood, R.I. (2022). Peptide fusion improves prime editing efﬁciency. Nat. Commun. 13, 3512. https://doi. org/10.1038/s41467-022-31270-y. 7. Zong, Y., Liu, Y., Xue, C., Li, B., Li, X., Wang, Y., Li, J., Liu, G., Huang, X., Cao, X., and Gao, C. (2022). An engineered prime editor with enhanced ed- iting efﬁciency in plants. Nat. Biotechnol. 40, 1394–1402. https://doi.org/ 10.1038/s41587-022-01254-w. 8. Ferreira da Silva, J., Oliveira, G.P., Arasa-Verge, E.A., Kagiou, C., Moret- ton, A., Timelthaler, G., Jiricny, J., and Loizou, J.I. (2022). Prime editing ef- ﬁciency and ﬁdelity are enhanced in the absence of mismatch repair. Nat. Commun. 13, 760. https://doi.org/10.1038/s41467-022-28442-1. 9. Anzalone, A.V., Gao, X.D., Podracky, C.J., Nelson, A.T., Koblan, L.W., Ra- guram, A., Levy, J.M., Mercer, J.A.M., and Liu, D.R. (2022). Programmable 4000 Cell 186, 3983–4002, August 31, 2023 Resource deletion, replacement, integration and inversion of large DNA sequences with twin prime editing. Nat. Biotechnol. 40, 731–740. https://doi.org/10. 1038/s41587-021-01133-w. 10. Choi, J., Chen, W., Suiter, C.C., Lee, C., Chardon, F.M., Yang, W., Leith, A., Daza, R.M., Martin, B., and Shendure, J. (2022). Precise genomic de- letions using paired prime editing. Nat. Biotechnol. 40, 218–226. https:// doi.org/10.1038/s41587-021-01025-z. 11. Jiang, T., Zhang, X.-O., Weng, Z., and Xue, W. (2022). Deletion and replacement of long genomic sequences using prime editing. Nat. Bio- technol. 40, 227–234. https://doi.org/10.1038/s41587-021-01026-y. 12. Lin, Q., Jin, S., Zong, Y., Yu, H., Zhu, Z., Liu, G., Kou, L., Wang, Y., Qiu, J.- L., Li, J., and Gao, C. (2021). High-efﬁciency prime editing with optimized, paired pegRNAs in plants. Nat. Biotechnol. 39, 923–927. https://doi.org/ 10.1038/s41587-021-00868-w. 13. Tao, R., Wang, Y., Jiao, Y., Hu, Y., Li, L., Jiang, L., Zhou, L., Qu, J., Chen, Q., and Yao, S. (2022). Bi-PE: bi-directional priming improves CRISPR/ Cas9 prime editing in mammalian cells. Nucleic Acids Res. 50, 6423– 6434. https://doi.org/10.1093/nar/gkac506. 14. Wang, J., He, Z., Wang, G., Zhang, R., Duan, J., Gao, P., Lei, X., Qiu, H., Zhang, C., Zhang, Y., and Yin, H. (2022). Efﬁcient targeted insertion of large DNA fragments without DNA donors. Nat. Methods 19, 331–340. 15. Zhuang, Y., Liu, J., Wu, H., Zhu, Q., Yan, Y., Meng, H., Chen, P.R., and Yi, C. (2022). Increasing the efﬁciency and precision of prime editing with guide RNA pairs. Nat. Chem. Biol. 18, 29–37. https://doi.org/10.1038/ s41589-021-00889-1. 16. Yarnall, M.T.N., Ioannidi, E.I., Schmitt-Ulms, C., Krajeski, R.N., Lim, J., Vil- liger, L., Zhou, W., Jiang, K., Garushyants, S.K., Roberts, N., et al. (2023). Drag-and-drop genome insertion of large sequences without double- strand DNA cleavage using CRISPR-directed integrases. Nat. Biotechnol. 41, 500–512. https://doi.org/10.1038/s41587-022-01527-4. 17. Arezi, B., and Hogrefe, H. (2009). Novel mutations in Moloney Murine Leu- kemia Virus reverse transcriptase increase thermostability through tighter binding to template-primer. Nucleic Acids Res. 37, 473–481. https://doi. org/10.1093/nar/gkn952. 18. Baranauskas, A., Paliksa, S., Alzbutas, G., Vaitkevicius, M., Lubiene, J., Letukiene, V., Burinskas, S., Sasnauskas, G., and Skirgaila, R. (2012). Generation and characterization of new highly thermostable and proces- sive M-MuLV reverse transcriptase variants. Protein Eng. Des. Sel. 25, 657–668. https://doi.org/10.1093/protein/gzs034. 19. Gerard, G.F., Potter, R.J., Smith, M.D., Rosenthal, K., Dhariwal, G., Lee, J., and Chatterjee, D.K. (2002). The role of template-primer in protection of reverse transcriptase from thermal inactivation. Nucleic Acids Res. 30, 3118–3129. https://doi.org/10.1093/nar/gkf417. 20. Kotewicz, M.L., Sampson, C.M., D’Alessio, J.M., and Gerard, G.F. (1988). Isolation of cloned Moloney murine leukemia virus reverse transcriptase lacking ribonuclease H activity. Nucl Acids Res 16, 265–277. https://doi. org/10.1093/nar/16.1.265. 21. Gru¨ newald, J., Miller, B.R., Szalay, R.N., Cabeceiras, P.K., Woodilla, C.J., Holtz, E.J.B., Petri, K., and Joung, J.K. (2022). Engineered CRISPR prime editors with compact, untethered reverse transcriptases. Nat. Biotechnol. https://doi.org/10.1038/s41587-022-01473-1. 22. Lin, Q., Zong, Y., Xue, C., Wang, S., Jin, S., Zhu, Z., Wang, Y., Anzalone, A.V., Raguram, A., Doman, J.L., et al. (2020). Prime genome editing in rice and wheat. Nat. Biotechnol. 38, 582–585. https://doi.org/10.1038/s41587- 020-0455-x. 23. Zong, Y., Liu, Y., Xue, C., Li, B., Li, X., Wang, Y., Li, J., Liu, G., Huang, X., Cao, X., and Gao, C. (2022). Author Correction: An engineered prime editor with enhanced editing efﬁciency in plants. Nat. Biotechnol. 40, 1412. https://doi.org/10.1038/s41587-022-01308-z. 24. Esvelt, K.M., Carlson, J.C., and Liu, D.R. (2011). A system for the contin- uous directed evolution of biomolecules. Nature 472, 499–503. https:// doi.org/10.1038/nature09929. Resource ll OPEN ACCESS 25. Davis, J., Banskota, S., Levy, J.M., Newby, G.A., Wang, X., Anzalone, A.V., Nelson, A.T., Chen, P.J., An, M., Roh, H., et al. Efﬁcient AAV-Mediated in Vivo Prime Editing in Multiple Organs. Submitted 26. Bo¨ ck, D., Rothgangl, T., Villiger, L., Schmidheini, L., Mathis, N., Ioannidi, E., Kreutzer, S., Kontarakis, Z., Rimann, N., Grisch-Chan, H.M., et al. (2021). In vivo prime editing of a metabolic liver disease in mice. Sci. Transl. Med. 14. https://doi.org/10.1126/scitranslmed.abl9238. 41. Izsva´ k, Z., Chuah, M.K.L., Vandendriessche, T., and Ivics, Z. (2009). Efﬁ- cient stable gene transfer into human cells by the Sleeping Beauty trans- poson vectors. Methods 49, 287–297. https://doi.org/10.1016/j.ymeth. 2009.07.001. 42. Anders, C., Niewoehner, O., Duerst, A., and Jinek, M. (2014). Structural basis of PAM-dependent target DNA recognition by the Cas9 endonu- clease. Nature 513, 569–573. 27. Zhi, S., Chen, Y., Wu, G., Wen, J., Wu, J., Liu, Q., Li, Y., Kang, R., Hu, S., Wang, J., et al. (2022). Dual-AAV delivering split prime editor system for in vivo genome editing. Mol. Ther. 30, 283–294. https://www. sciencedirect.com/science/article/abs/pii/S1525001621003658. 43. Chen, J.S., Dagdas, Y.S., Kleinstiver, B.P., Welch, M.M., Sousa, A.A., Har- rington, L.B., Sternberg, S.H., Joung, J.K., Yildiz, A., and Doudna, J.A. (2017). Enhanced proofreading governs CRISPR–Cas9 targeting accu- racy. Nature 550, 407–410. https://doi.org/10.1038/nature24268. 28. Kirshenboim, N., Hayouka, Z., Friedler, A., and Hizi, A. (2007). Expression and characterization of a novel reverse transcriptase of the LTR retrotrans- poson Tf1. Virology 366, 263–276. https://doi.org/10.1016/j.virol.2007. 04.002. 44. Lapinaite, A., Knott, G.J., Palumbo, C.M., Lin-Shiao, E., Richter, M.F., Zhao, K.T., Beal, P.A., Liu, D.R., and Doudna, J.A. (2020). DNA capture by a CRISPR-Cas9–guided adenine base editor. Science 369, 566–571. https://doi.org/10.1126/science.abb1390. 29. Millman, A., Bernheim, A., Stokar-Avihail, A., Fedorenko, T., Voichek, M., Leavitt, A., Oppenheimer-Shaanan, Y., and Sorek, R. (2020). Bacterial Ret- rons Function In Anti-Phage Defense. Cell 183, 1551–1561.e12. https:// doi.org/10.1016/j.cell.2020.09.065. 30. Wang, Y., Guan, Z., Wang, C., Nie, Y., Chen, Y., Qian, Z., Cui, Y., Xu, H., Wang, Q., Zhao, F., et al. (2022). Cryo-EM structures of Escherichia coli Ec86 retron complexes reveal architecture and defence mechanism. Nat. Microbiol. 7, 1480–1489. https://doi.org/10.1038/s41564-022-01197-7. 31. Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., (cid:1)Zı´dek, A., Potapenko, A., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. https://doi.org/10.1038/s41586-021-03819-2. 32. Roth, T.B., Woolston, B.M., Stephanopoulos, G., and Liu, D.R. (2019). Phage-Assisted Evolution of Bacillus methanolicus Methanol Dehydroge- nase 2. ACS Synth. Biol. 8, 796–806. https://doi.org/10.1021/acssynbio. 8b00481. 33. Doman, J.L., Sousa, A.A., Randolph, P.B., Chen, P.J., and Liu, D.R. (2022). Designing and executing prime editing experiments in mammalian cells. Nat. Protoc. 17, 2431–2468. https://doi.org/10.1038/s41596-022-00724-4. 34. Dickinson, B.C., Leconte, A.M., Allen, B., Esvelt, K.M., and Liu, D.R. (2013). Experimental interrogation of the path dependence and stochas- ticity of protein evolution using phage-assisted continuous evolution. Proc. Natl. Acad. Sci. USA 110, 9007–9012. https://doi.org/10.1073/ pnas.1220670110. 35. Stamos, J.L., Lentzsch, A.M., and Lambowitz, A.M. (2017). Structure of a Thermostable Group II Intron Reverse Transcriptase with Template-Primer and Its Functional and Evolutionary Implications. Mol. Cell 68, 926–939.e4. https://doi.org/10.1016/j.molcel.2017.10.024. 36. Flotte, T.R., Cataltepe, O., Puri, A., Batista, A.R., Moser, R., McKenna-Ya- sek, D., Douthwright, C., Gernoux, G., Blackwood, M., Mueller, C., et al. (2022). AAV gene therapy for Tay-Sachs disease. Nat. Med. 28, 251–259. https://doi.org/10.1038/s41591-021-01664-4. 37. Sharma, P.L., Nurpeisov, V., and Schinazi, R.F. (2005). Retrovirus Reverse Transcriptases Containing a Modiﬁed YXDD Motif. Antivir. Chem. Chemo- ther. 16, 169–182. https://doi.org/10.1177/095632020501600303. 38. Zadeh, J.N., Steenberg, C.D., Bois, J.S., Wolfe, B.R., Pierce, M.B., Khan, A.R., Dirks, R.M., and Pierce, N.A. (2011). NUPACK: Analysis and design of nucleic acid systems. J. Comput. Chem. 32, 170–173. https://doi.org/ 10.1002/jcc.21596. 39. Telesnitsky, A., and Goff, S.P. (1993). RNase H domain mutations affect the interaction between Moloney murine leukemia virus reverse transcrip- tase and its primer-template. Proc. Natl. Acad. Sci. USA 90, 1276–1280. https://doi.org/10.1073/pnas.90.4.1276. 45. Nishimasu, H., Ran, F.A., Hsu, P.D., Konermann, S., Shehata, S.I., Doh- mae, N., Ishitani, R., Zhang, F., and Nureki, O. (2014). Crystal Structure of Cas9 in Complex with Guide RNA and Target DNA. Cell 156, 935–949. 46. Slaymaker, I.M., Gao, L., Zetsche, B., Scott, D.A., Yan, W.X., and Zhang, F. (2016). Rationally engineered Cas9 nucleases with improved speciﬁcity. Science 351, 84–88. https://doi.org/10.1126/science.aad5227. 47. Qi, L.S., Larson, M.H., Gilbert, L.A., Doudna, J.A., Weissman, J.S., Arkin, A.P., and Lim, W.A. (2013). Repurposing CRISPR as an RNA-Guided Plat- form for Sequence-Speciﬁc Control of Gene Expression. Cell 152, 1173– 1183. https://doi.org/10.1016/j.cell.2013.02.022. 48. Jiang, F., and Doudna, J.A. (2017). CRISPR–Cas9 Structures and Mecha- nisms. Annu. Rev. Biophys. 46, 505–529. https://doi.org/10.1146/an- nurev-biophys-062215-010822. 49. Zeng, Y., Cui, Y., Zhang, Y., Zhang, Y., Liang, M., Chen, H., Lan, J., Song, G., and Lou, J. (2018). The initiation, propagation and dynamics of CRISPR-SpyCas9 R-loop complex. Nucleic Acids Res. 46, 350–361. https://doi.org/10.1093/nar/gkx1117. 50. Lazzarotto, C.R., Malinin, N.L., Li, Y., Zhang, R., Yang, Y., Lee, G., Cowley, E., He, Y., Lan, X., Jividen, K., et al. (2020). CHANGE-seq reveals genetic and epigenetic effects on CRISPR–Cas9 genome-wide activity. Nat. Bio- technol. 38, 1317–1327. https://doi.org/10.1038/s41587-020-0555-7. 51. Huang, T.P., Heins, Z.J., Miller, S.M., Wong, B.G., Balivada, P.A., Wang, T., Khalil, A.S., and Liu, D.R. (2023). High-throughput continuous evolution of compact Cas9 variants targeting single-nucleotide-pyrimidine PAMs. Nat. Biotechnol. 41, 96–107. https://doi.org/10.1038/s41587-022-01410-2. 52. Liu, P., Liang, S.-Q., Zheng, C., Mintzer, E., Zhao, Y.G., Ponnienselvan, K., Mir, A., Sontheimer, E.J., Gao, G., Flotte, T.R., et al. (2021). Improved prime editors enable pathogenic allele correction and cancer modelling in adult mice. Nat. Commun. 12, 2121. https://doi.org/10.1038/s41467- 021-22295-w. 53. Banskota, S., Raguram, A., Suh, S., Du, S.W., Davis, J.R., Choi, E.H., Wang, X., Nielsen, S.C., Newby, G.A., Randolph, P.B., et al. (2022). Engi- neered virus-like particles for efﬁcient in vivo delivery of therapeutic pro- teins. Cell 185, 250–265.e16. 54. Nishiyama, J., Mikuni, T., and Yasuda, R. (2017). Virus-Mediated Genome Editing via Homology-Directed Repair in Mitotic and Postmitotic Cells in Mammalian Brain. Neuron 96, 755–768.e5. https://doi.org/10.1016/j. neuron.2017.10.004. 55. Suzuki, K., Tsunekawa, Y., Hernandez-Benitez, R., Wu, J., Zhu, J., Kim, E.J., Hatanaka, F., Yamamoto, M., Araoka, T., Li, Z., et al. (2016). In vivo genome editing via CRISPR/Cas9 mediated homology-independent tar- geted integration. Nature 540, 144–149. https://doi.org/10.1038/ nature20565. 40. Mathis, N., Allam, A., Kissling, L., Marquart, K.F., Schmidheini, L., Solari, C., Bala´ zs, Z., Krauthammer, M., and Schwank, G. (2023). Predicting prime editing efﬁciency and product purity by deep learning. Nat. Bio- technol. https://doi.org/10.1038/s41587-022-01613-7. 56. Koeppel, J., Weller, J., Peets, E.M., Pallaseni, A., Kuzmin, I., Raudvere, U., Peterson, H., Liberante, F.G., and Parts, L. (2023). Prediction of prime ed- iting insertion efﬁciencies using sequence features and DNA repair deter- minants. Nat. Biotechnol. https://doi.org/10.1038/s41587-023-01678-y. Cell 186, 3983–4002, August 31, 2023 4001 ll OPEN ACCESS Resource 57. Kim, H.K., Yu, G., Park, J., Min, S., Lee, S., Yoon, S., and Kim, H.H. (2021). Predicting the efﬁciency of prime editing guide RNAs in human cells. Nat. Biotechnol. 39, 198–206. https://doi.org/10.1038/s41587-020-0677-y. 64. Engler, C., Kandzia, R., and Marillonnet, S. (2008). A One Pot, One Step, Precision Cloning Method with High Throughput Capability. PLoS One 3, e3647. https://doi.org/10.1371/journal.pone.0003647. 58. Yu, G., Kim, H.K., Park, J., Kwak, H., Cheong, Y., Kim, D., Kim, J., Kim, J., and Kim, H.H. (2023). Prediction of efﬁciencies for diverse prime editing systems in multiple cell types. Cell 186, 2256–2272.e23. https://doi.org/ 10.1016/j.cell.2023.03.034. 59. Thorrez, L., Vandenburgh, H., Canver, M., Gehrke, J., Farouni, R., Hsu, J., Cole, M., Liu, D., Joung, K., Bauer, D., et al. (2019). CRISPResso2 provides accurate and rapid genome editing sequence analysis. Nat. Biotechnol. 37, 215–216. https://doi.org/10.1038/s41587-019-0043-0. 65. Hubbard, B.P., Badran, A.H., Zuris, J.A., Guilinger, J.P., Davis, K.M., Chen, L., Tsai, S.Q., Sander, J.D., Joung, J.K., and Liu, D.R. (2015). Continuous directed evolution of DNA-binding proteins to improve TALEN speciﬁcity. Nat. Methods 12, 939–942. https://doi.org/10.1038/ nmeth.3515. 66. Miller, S.M., Wang, T., and Liu, D.R. (2020). Phage-assisted continuous and non-continuous evolution. Nat. Protoc. 15, 4101–4127. https://doi. org/10.1038/s41596-020-00410-3. 60. Clement, K., Farouni, R., Bauer, D.E., and Pinello, L. (2018). AmpUMI: design and analysis of unique molecular identiﬁers for deep amplicon sequencing. Bioinformatics 34, i202–i210. https://doi.org/10.1093/bioin- formatics/bty264. 61. Mok, B.Y., Kotrys, A.V., Raguram, A., Huang, T.P., Mootha, V.K., and Liu, D.R. (2022). CRISPR-free base editors with enhanced activity and expanded targeting scope in mitochondrial and nuclear DNA. Nat. Bio- technol. 40, 1378–1387. https://doi.org/10.1038/s41587-022-01256-8. 67. Badran, A.H., and Liu, D.R. (2015). Development of potent in vivo muta- genesis plasmids with broad mutational spectra. Nat. Commun. 6, 8425. https://doi.org/10.1038/ncomms9425. 68. Levy, J.M., Yeh, W.-H., Pendse, N., Davis, J.R., Hennessey, E., Butcher, R., Koblan, L.W., Comander, J., Liu, Q., and Liu, D.R. (2020). Cytosine and adenine base editing of the brain, liver, retina, heart and skeletal mus- cle of mice via adeno-associated viruses. Nat. Biomed. Eng. 4, 97–110. https://doi.org/10.1038/s41551-019-0501-5. 62. Richter, M.F., Zhao, K.T., Eton, E., Lapinaite, A., Newby, G.A., Thuronyi, B.W., Wilson, C., Koblan, L.W., Zeng, J., Bauer, D.E., et al. (2020). Phage-assisted evolution of an adenine base editor with improved Cas domain compatibility and activity. Nat. Biotechnol. 38, 883–891. https:// doi.org/10.1038/s41587-020-0453-z. 69. Mendell, J.R., Al-Zaidy, S.A., Lehman, K.J., McColly, M., Lowes, L.P., Al- fano, L.N., Reash, N.F., Iammarino, M.A., Church, K.R., Kleyn, A., et al. (2021). Five-Year Extension Results of the Phase 1 START Trial of Ona- semnogene Abeparvovec in Spinal Muscular Atrophy. JAMA Neurol. 78, 834–841. https://doi.org/10.1001/jamaneurol.2021.1272. 63. Thuronyi, B.W., Koblan, L.W., Levy, J.M., Yeh, W.-H., Zheng, C., Newby, G.A., Wilson, C., Bhaumik, M., Shubina-Oleinik, O., Holt, J.R., and Liu, D.R. (2019). Continuous evolution of base editors with expanded target compatibility and improved activity. Nat. Biotechnol. 37, 1070–1079. https://doi.org/10.1038/s41587-019-0193-0. 70. Nowak, E., Potrzebowski, W., Konarev, P.V., Rausch, J.W., Bona, M.K., Svergun, D.I., Bujnicki, J.M., Le Grice, S.F.J., and Nowotny, M. (2013). Structural analysis of monomeric retroviral reverse transcriptase in com- plex with an RNA/DNA hybrid. Nucleic Acids Res. 41, 3874–3887. https://doi.org/10.1093/nar/gkt053. 4002 Cell 186, 3983–4002, August 31, 2023 ll OPEN ACCESS Resource STAR+METHODS KEY RESOURCES TABLE REAGENT or RESOURCE Bacterial and virus strains One Shot Mach1 T1 Phage-Resistant Chemically Competent E. coli E. coli S2060 Chemicals, peptides, and recombinant proteins BsaI-HFv2 LguI (SapI) T4 DNA Ligase NEBuilder HiFi DNA assembly master mix Dimethyl sulfoxide Poly(ethylene glycol) 3350 DNaseI (Rnase-free) Magnesium chloride solution Carbenicillin Chloramphenicol Tetracycline Streptomycin L-arabinose Glucose Bluo-gal dNTPs Lipofectamine 2000 TrypLE Proteinase K, recombinant, PCR grade SDS (10% wt/vol) DNAdvance Kit AMPure XP CleanCap Reagent AG N1 -Methylpseudouridine- 50 -Triphosphate LiCl Precipitation Solution (7.5 M) DMEM, high glucose, GlutaMAX supplement Fetal bovine serum L-Glutamine Penicillin-Streptomycin GlutaMAX supplement N-acetyl-L-cysteine Human AB Serum Recombinant Human IL-2 Lymphoprep density gradient medium SOURCE IDENTIFIER Thermo Fisher Scientiﬁc Cat#C862003 Addgene #105064 New England BioLabs Thermo Fisher Scientiﬁc New England BioLabs New England BioLabs Cat#R3733S Cat#ER1932 Cat#M0202S Cat#E2621S Sigma-Aldrich Sigma-Aldrich Cat#D8418-50ML Cat#P4338-500G New England BioLabs Cat#M0303 Sigma-Aldrich Gold Biotechnology Gold Biotechnology Gold Biotechnology Gold Biotechnology Gold Biotechnology Sigma-Aldrich Gold Biotechnology New England BioLabs Thermo Fisher Scientiﬁc Thermo Fisher Scientiﬁc Thermo Fisher Scientiﬁc Cat#M1028-10X1ML Cat#C-103 Cat#C-105 Cat#T-101 Cat#S-150 Cat#A-300 Cat#G7021 Cat#B-673-10 Cat#N0447S Cat#11668019 Cat#12605010 Cat#11668019 Thermo Fisher Scientiﬁc Cat#15553027 Beckman Coulter Beckman Coulter TriLink BioTechnologies TriLink BioTechnologies Cat#A48705 Cat#B23318 Cat#N-7113 Cat#N-1081 Thermo Fisher Scientiﬁc Thermo Fisher Scientiﬁc Cat#AM9480 Cat#10566016 Thermo Fisher Scientiﬁc Corning Thermo Fisher Scientiﬁc Thermo Fisher Scientiﬁc Sigma-Aldrich Valley Biomedical Peprotech STEMCELL Technologies Cat#16000044 Cat#25-005-Cl Cat#15070063 Cat#35050061 Cat#A7250-100G Cat#HP1022HI Cat#200-02 Cat#07801 (Continued on next page) Cell 186, 3983–4002.e1–e13, August 31, 2023 e1 ll OPEN ACCESS Continued REAGENT or RESOURCE Dynabeads Human T-Expander CD3/CD28 X-VIVO(cid:2) 15 Serum-free Hematopoietic Cell Medium Dulbecco0s Modiﬁ–d Eagle0s Medium – low glucose Eagle’s minimal essential Medium (EMEM) Opti-MEM reduced serum medium PEG 8000 PEG-it Virus Precipitation Solution Salt active nuclease 0.9% NaCl BSA Vybrant DyeCycle Ruby EZ-PREP buffer Critical commercial assays Phusion U Multiplex PCR Master Mix Q5 High-Fidelity 2 x Master Mix Phusion Green Hot Start II High-Fidelity DNA Polymerase QIAquick PCR Puriﬁcation Kit QIAquick Gel Extraction Kit QIAGEN Plasmid Plus Midi Kit QIAprep Spin Miniprep Kit Qiagen Plasmid Plus 96 Miniprep Kit EasySep Human T cell Isolation Kit Neon(cid:2) Transfection System QuickExtract(cid:2) DNA Extraction Solution SE Cell Line 4D- Nucleofector X Kit S Illustra TempliPhi 100 ampliﬁcation kit NEB T7 HiScribe Kit AAVpro Titration Kit version 2 Agencourt DNAdvance Kit MiSeq Reagent Kit v2 (300-cycles) MiSeq Reagent Micro Kit v2 (300-cycles) Resource SOURCE Thermo Fisher Scientiﬁc IDENTIFIER Cat#11141D Lonza Cat#BE02-053Q Sigma-Aldrich Cat#D5546 ATCC Cat#30-2003 Thermo Fisher Scientiﬁc Cat#31985070 Sigma-Aldrich System Biosciences ArcticZymes Fresenius Kabi NEB Thermo Fisher Sigma-Aldrich Cat#25322-68-3 Cat#LV825A-1 Cat#70910-202 Cat#918610 Cat#B9000S Cat#V10309 #NUC-101 Thermo Fisher Scientiﬁc Cat#F562L New England BioLabs Cat#M0492L Thermo Fisher Scientiﬁc Cat#F537L QIAGEN QIAGEN QIAGEN QIAGEN QIAGEN STEMCELL Technologies Thermo Fisher Scientiﬁc Lucigen Lonza Cytiva New England BioLabs Clontech/Takara Beckman Coulter Illumina Illumina Cat#28104 Cat#28704 Cat#12943 Cat#27106 Cat#16181 Cat#17951 Cat#MPK1096 Cat# QE09050 Cat#V4XC-1032 Cat#25640010 Cat#E2040S Cat#6233 Cat#V10309 Cat#MS-102-2002 Cat#MS-103-1002 e2 Cell 186, 3983–4002.e1–e13, August 31, 2023 (Continued on next page) Resource Continued REAGENT or RESOURCE Deposited data Amplicon sequencing data Experimental models: Cell lines Human (female): HEK293T Mouse (male): N2a Human (female): HEK293T clone 17 Primary human ﬁbroblast (HEXA) Primary human ﬁbroblast (UGT1A1) Primary human ﬁbroblast (RECQL3) Primary human ﬁbroblast (GAA) Experimental models: Organisms/strains Timed pregnant C57BL/6J mice Oligonucleotides HEXA, 1278ins TATC pegRNA: mAmUmCrCr UrUrCrCrArGrUrCrArGrGrGrCrCrArUrGrUrUrU rUrArGrArGrCrUrArGrArArArUrArGrCrArArGrUrU rArArArArUrArArGrGrCrUrArGrUrCrCrGrUrUrAr UrCrArArCrUrUrGrArArArArArGrUrGrGrCrArCr CrGrArGrUrCrGrGrUrGrCrGrUrArCrCrUrGrArAr CrCrGrUrArUrArUrCrGrUrArUrGrGrCrCrCrUrGr ArCrUrUrCrUrCrUrCrUrCrCrGrCrGrGrUrUrCr UrArUrCrUrArGrUrUrArCrGrCrGrUrUrAr ArArCrCrArArCrUrAmGmAmA VEGFA, +2 G to A pegRNA: mGmAmUrGrUr CrUrGrCrArGrGrCrCrArGrArUrGrArGrUrUrUr UrArGrArGrCrUrArGrArArArUrArGrCrArArGr UrUrArArArArUrArArGrGrCrUrArGrUrCrCr GrUrUrArUrCrArArCrUrUrGrArArArArArGrUrGr GrCrArCrCrGrArGrUrCrGrGrUrGrCrArArUrGrUr GrCrCrArUrCrUrGrGrArGrCrArCrUrCrArUrCrUr GrGrCrCrUrGrCrArGrArArCrArArUrCrUrCrCrGr CrGrGrUrUrCrUrArUrCrUrArGrUrUrArCrGrCr GrUrUrArArArCrCrArArCrUrArGrArAmUmUmU DNMT1, 1–15 deletion pegRNA: mGmAmUrUr CrCrUrGrGrUrGrCrCrArGrArArArCrArGrUrUrUr UrArGrArGrCrUrArGrArArArUrArGrCrArArGrUr UrArArArArUrArArGrGrCrUrArGrUrCrCrGrUrUrAr UrCrArArCrUrUrGrArArArArArGrUrGrGrCr ArCrCrGrArGrUrCrGrGrUrGrCrArGrGrAr GrGrArArGrCrUrGrCrUrArArGrGrArCrUrArGrUrUr CrUrGrCrCrCrUrUrCrUrGrGrCrArCrCrArGrGrAr CrCrUrCrUrUrCrUrCrGrCrGrGrUrUrCrUrArUr CrUrArGrUrUrArCrGrCrGrUrUrArArArCrCrArArCrUr ArGrArAmUmUmU ll OPEN ACCESS SOURCE This paper ATCC ATCC ATCC IDENTIFIER NCBI SRA: BioProject PRJNA916060 Cat#CRL-3216 Cat#CCL-131 Cat#CRL-11268 Coriell Institute Cat#GM00221 Coriell Institute Cat# GM09551 Coriell Institute Cat# GM02085 Coriell Institute Cat# GM20092 Charles River Laboratories Cat#027 Integrated DNA Technologies N/A Integrated DNA Technologies N/A Integrated DNA Technologies N/A (Continued on next page) Cell 186, 3983–4002.e1–e13, August 31, 2023 e3 ll OPEN ACCESS Continued REAGENT or RESOURCE SOURCE IDENTIFIER Resource Integrated DNA Technologies N/A Integrated DNA Technologies N/A Integrated DNA Technologies N/A Integrated DNA Technologies N/A Integrated DNA Technologies N/A Synthego Corporation N/A CCR5, attB insertion pegRNA1: mGmCmUrGr UrGrUrUrUrGrCrGrUrCrUrCrUrCrCrCrGrUrUr UrUrArGrArGrCrUrArGrArArArUrArGrCrArArGr UrUrArArArArUrArArGrGrCrUrArGrUrCrCrGr UrUrArUrCrArArCrUrUrGrArArArArArGrUrGr GrCrArCrCrGrArGrUrCrGrGrUrGrCrArCrGrAr CrGrGrArGrArCrCrGrCrCrGrUrCrGrUrCrGr ArCrArArGrCrCrArGrArGrArCrGrCmAmAmA CCR5, attB insertion pegRNA2: mGmUmArUrGr GrArArArArUrGrArGrArGrCrUrGrCrGrUrUrUrUr ArGrArGrCrUrArGrArArArUrArGrCrArArGrUrUrAr ArArArUrArArGrGrCrUrArGrUrCrCrGrUrUrArUr CrArArCrUrUrGrArArArArArGrUrGrGrCr ArCrCrGrArGrUrCrGrGrUrGrCrArCrGrAr CrGrGrCrGrGrUrCrUrCrCrGrUrCrGrUrCrArGr GrArUrCrArUrGrCrUrCrUrCrArUrUmUmUmC UGT1A1, correction of 13BP deletion Exon 2 pegRNA: mGmCmUrCrUrArGrGrArArUrUr UrGrArArGrCrCrArGrUrUrUrUrArGrArGrCrUr ArGrArArArUrArGrCrArArGrUrUrArArArArUrAr ArGrGrCrUrArGrUrCrCrGrUrUrArUrCrArArCr UrUrGrArArArArArGrUrGrGrCrArCrCrGrArGrUr CrGrGrUrGrCrArCrArArUrUrCrCrArUrGrUrUr CrUrCrCrArGrArArGrCrArUrUrArArUrGrUrArGr GrCrUrUrCrArArArUrUrCrCrUrArCrGrCrGrGr UrUrCrUrArUrCrUrArGrUrUrArCrGrCrGrUrUrAr ArArCrCrArArCrUrAmGmAmA RECQL3, correction of 6-BP del/7BP ins at nt.2281 pegRNA: mUmCmUrGrArGrUrCrArGrUr CrUrUrArUrCrArCrCrGrUrUrUrUrArGrArGrCrUrAr GrArArArUrArGrCrArArGrUrUrArArArArUrArAr GrGrCrUrArGrUrCrCrGrUrUrArUrCrArArCrUrUr GrArArArArArGrUrGrGrCrArCrCrGrArGrUrCrGrGr UrGrCrUrCrCrArGrCrUrArCrArUrArUrCrUrGr ArCrArGrGrUrGrArUrArArGrArCrUrGrCrGrCrGr GrUrUrCrUrArUrCrUrArGrUrUrArCrGrCrGrUr UrArArArCrCrArArCrUrAmGmAmA GAA, correction of 13-bp deletion nt.1456-1468 pegRNA mUmCmGrUrUrGrUrCrCrArGr GrUrArUrGrGrCrCrCrGrUrUrUrUrArGrArGrCr UrArGrArArArUrArGrCrArArGrUrUrArArArArUrAr ArGrGrCrUrArGrUrCrCrGrUrUrArUrCrArArCrUr UrGrArArArArArGrUrGrGrCrArCrCrGrArGrUrCr GrGrUrGrCrUrCrCrUrCrCrCrArCrCrArGrGrCrCr ArGrGrGrCrUrGrUrGrGrGrGrUrUrGrGrUrGrArAr GrUrCrGrGrGrGrArArGrGrCrArGrUrGrGrArGr CrCrGrGrGrCrCrArUrArCrCrUmGmGmA HEXA, nick sgRNA: mUmAmCrCrUrGrAr ArCrCrGrUrArUrArUrCrGrUrAGrUrUrUrUrArGrAr GrCrUrArGrArArArUrArGrCrArArGrUrUrArArArAr UrArArGrGrCrUrArGrUrCrCrGrUrUrArUrCr ArArCrUrUrGrArArArArArGrUrGrGrCrArCrCrGrArGr UrCrGrGrUr GrCrUmUmUmU e4 Cell 186, 3983–4002.e1–e13, August 31, 2023 (Continued on next page) Resource Continued REAGENT or RESOURCE SOURCE IDENTIFIER ll OPEN ACCESS VEGFA, nick sgRNA mGmAmGrCrCrCrAr GrGrGrCrUrGrGrGrCrArCrArGGrUrUrUrUr ArGrArGrCrUrArGrArArArUrArGrCrArArGrUr UrArArArArUrArArGrGrCrUrArGrUrCrCrGrUr UrArUrCrArArCrUrUrGrArArArArArGrUrGrGr CrArCrCrGrArGrUrCrGrGrUr GrCrUmUmUmU DNMT1, nick sgRNA: mCmCmCrUrUrCrArGr CrUrArArArArUrArArArGrGGrUrUrUrUrArGrAr GrCrUrArGrArArArUrArGrCrArArGrUrUrArArArAr UrArArGrGrCrUrArGrUrCrCrGrUrUrArUrCrArAr CrUrUrGrArArArArArGrUrGrGrCrArCrCrGrAr GrUrCrGrGrUr GrCrUmUmUmU UGT1A1, nick sgRNA: mAmUmUrGrCrCrAr UrArGrCrUrUrUrCrUrUrCrUrCrGrUrUrUrUrAr GrArGrCrUrArGrArArArUrArGrCrArArGrUrUrAr ArArArUrArArGrGrCrUrArGrUrCrCrGrUrUrAr UrCrArArCrUrUrGrArArArArArGrUrGrGrCrArCrCrGr ArGrUrCrGrGrUrGrCrUmUmUmU RECQL3, nick sgRNA mAmUmUrCrCrArGr CrUrArCrArUrArUrCrUrGrArCrGrUrUrUrUrArGr ArGrCrUrArGrArArArUrArGrCrArArGrUrUrArAr ArArUrArArGrGrCrUrArGrUrCrCrGrUrUrArUrCr ArArCrUrUrGrArArArArArGrUrGrGrCrAr CrCrGrArGrUrCrGrGrUrGrCrUmUmUmU GAA, nick sgRNA mAmGmCrCrArCrCrArUrGrUr CrCrUrCrCrCrArCrCrGrUrUrUrUrArGrArGrCrUr ArGrArArArUrArGrCrArArGrUrUrArArArArUrAr ArGrGrCrUrArGrUrCrCrGrUrUrArUrCrArArCrUr UrGrArArArArArGrUrGrGrCrArCrCrGrArGr UrCrGrGrUrGrCrUmUmU*mU Recombinant DNA Mutagenesis plasmid MP6 pJC175e pBT114-splitC pBT29-splitD pCMV-PE2 pCMV-PEmax pT7-PEmax pEF1a-MLH1dn pU6-tevopreq1-GG-acceptor pU6-pegRNA-GG-acceptor pCMV-PE6a pCMV-PE6b pCMV-PE6c pCMV-PE6d pCMV-PE6e pCMV-PE6f pCMV-PE6g AAV-PE6c-Rosa26-twinPE AAV-PE6d-Rosa26-twinPE AAV-PEmaxdeltaRNaseH-Rosa26-twinPE AAV-PE6d-Dnmt1-loxP Synthego Corporation N/A Synthego Corporation N/A Synthego Corporation N/A Synthego Corporation N/A Synthego Corporation N/A Addgene Addgene Addgene Addgene Addgene Addgene Addgene Addgene Addgene Addgene This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper This paper #69669 #79219 #138523 #138521 #132775 #174820 #178113 #174824 #174038 #132777 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A (Continued on next page) Cell 186, 3983–4002.e1–e13, August 31, 2023 e5 ll OPEN ACCESS Continued REAGENT or RESOURCE AAV-PEmaxdeltaRNaseH-Dnmt1-loxP Software and algorithms CRISPResso2 Prism Geneious Prime AmpUMI Python 3 Mutato Scaffold insertion analysis TDT analysis RESOURCE AVAILABILITY Resource SOURCE This paper IDENTIFIER N/A Clement et al., 201959 GraphPad Dotmatics Clement et al., 201860 Python Mok et al., 202261 Anzalone et al., 20191 This paper https://github.com/pinellolab/CRISPResso2 https://www.graphpad.com/ https://www.geneious.com/prime/ http://github.com/pinellolab/AmpUMI. https://www.python.org/downloads/ https://hub.docker.com/r/araguram/mutato/ Note S1 Note S2 Lead contact Please direct requests for resources and reagents to lead contact: David R. Liu (D.R.L. [email protected]). Materials availability Plasmids generated in this study are available from Addgene. Additional details are provided in the key resources table. Data and code availability d All sequencing data have been deposited at the NCBI Sequence Read Archive database and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. d All original code is available in Notes S1 and S2. d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS Mammalian cell culture conditions HEK293T (American Type Culture Collection (ATCC), Cat# CRL-3216), Neuro-2a (N2a from ATCC, Cat# CCL-131) and Huh7 (a gift from Erik Sontheimer’s group, originated from ATCC) cells were cultured in Dulbecco’s Modiﬁed Eagle Medium (DMEM) plus GlutaMAX (Thermo Fisher Scientiﬁc) supplemented with 10% (v/v) fetal bovine serum (FBS) (Thermo Fisher Scientiﬁc). Primary Tay Sachs disease patient ﬁbroblast cells were purchased from Coriell Institute (Cat. ID GM00221) and cultured in low-glucose DMEM (Sigma Aldrich) supplemented with 10% (v/v) FBS and 2mM GlutaMAX Supplement (Thermo Fisher Scientiﬁc). All cell lines C with 5% CO2. Cell lines were authenticated by their respective suppliers and tested were incubated, maintained, and cultured at 37 negative for mycoplasma. (cid:4) Generation of HEK293T models of Tay-Sachs disease HEK293T cells homozygous for the HEXA1278TATCins mutation were previously reported.1 HEK293T cells were seeded in a 48-well plate and transfected with 250 ng of a pegRNA plasmid, 83 ng of a nicking sgRNA plasmid, and 750 ng of a PE2-P2A-GFP plasmid programmed to install the HEXA1278TATCins mutation. 3 days after transfection, GFP-positive cells were ﬂow sorted using an LE-MA900 cell sorter (Sony) into a 96-well ﬂat bottom culture well plate. Cells were cultured for 10 days and then analyzed for HEXA1278TATCins mutation installation. Two different clonal, homozygous (100% installation of HEXA1278TATCins) cell lines were used for experiments. Generation of HEK293T model cell lines for Bloom Syndrome, Crigler-Najjar disease, and Pompe Disease Pathogenic gene fragments were generated by examining disease alleles from patient-derived ﬁbroblasts in the Coriell Institute data- base. These gene fragments (300 bp total, ﬂanking the pathogenic mutation) were then ordered as eBlocks (Integrated DNA tech- nologies). These fragments were then cloned into a Sleeping Beauty transposon vector, downstream of a blasticidin resistance gene expression cassette. (The target pathogenic gene itself was not expressed.) 3.2E5 low-passage HEK293T cells were plated in a 6-well dish and transfected with 50 ng of disease allele transposon, 25 ng of transposase, and 725 ng of PUC19 in a total volume of 250 mL using 20 mL lipofectamine 2000 (Thermo Fisher). 48 h after transfection, cells were trypsinized, resuspended in 2 mL of media, and 60 mL of the resuspended cells were plated in a fresh 6-well plate well with media containing 10 mg/mL blasticidin. Cells e6 Cell 186, 3983–4002.e1–e13, August 31, 2023 Resource ll OPEN ACCESS were passaged until a no-transposase negative control had completely died. The heterogeneous pool of cells was then used for transfection with editors to target the disease allele for correction. In the downstream HTS sample preparation, primers speciﬁc for the transposon backbone were used to selectively amplify the knocked-in pathogenic allele, as opposed to the wild-type endog- enous allele. Isolation and culture of primary human T cells Memorial Blood Center (St. Paul, MN) buffy coats were obtained followed by peripheral blood mononuclear cells (PBMC) isolation with Lymphoprep and SepMate tubes (STEMCELL Technologies). CD4+ T-cells were puriﬁed from PBMCs using the EasySep Human CD4 + T cell Isolation Kit (STEMCELL Technologies). T-cells were cultured in X-VIVO TM 15 Serum-free Hematopoietic Cell Medium (Lonza, Basel, Switzerland) supplemented with: 300 IU/mL IL-2 (PeproTech), GlutaMAX (Gibco), N-acetyl-cysteine (Sigma Aldrich), 5% AB human serum (Valley Biomedical), 50 U/mL penicillin and 50 mg/mL streptomycin (Gibco). METHOD DETAILS General methods and molecular cloning The following working concentrations were used for antibiotics (Gold Biotechnology): carbenicillin 50 mg/mL, chloramphenicol 25 mg/ mL, kanamycin 50 mg/mL, tetracycline 10 mg/mL, streptomycin 25 mg/mL. For all cloning experiments, Nuclease-free water (Qiagen) was used, gene blocks were ordered from Integrated DNA Technologies (IDT) and primers were ordered from either IDT or Eton Bio- sciences. All synthetic genes were codon-optimized for human cell expression using GenScript’s algorithm and obtained as gene blocks from either GenScript or IDT. All plasmid construction was done using Gibson assembly. Brieﬂy, for most Gibson cloning, unless otherwise noted, PCR was done using either Phusion U Green Hot Start II DNA polymerase (Thermo Fisher Scientiﬁc) or Phusion Green Hot Start II High-Fidelity DNA polymerase (Thermo Fisher Scientiﬁc). The resulting PCR products were puriﬁed using QIAquick PCR puriﬁcation Kit (Qiagen) and fragments were assembled using NEBuilder HiFi DNA assembly master mix (New England BioLabs) according to the manufacturer’s protocol. Plasmids for mammalian expression of prime editors were cloned into the pCMV- PE2 vector backbone (Addgene #132775) and plasmids used for the in vitro transcription of different prime editor mRNA were cloned into the pT7-PEmax (Addgene #178113) vector backbone. Plasmids for the mammalian expression of pegRNAs, sgRNA, and epegRNAs were cloned as previously described.33 Brieﬂy, vector backbone expressing a guide RNA under the human U6 promoter was digested using BsaI-HFv2 (New England BioLabs) according to the manufacturer’s protocol. The digested fragment was puriﬁed by gel electrophoresis with a 1% agarose gel using QIAquick Gel Extraction Kit (QIAGEN). The BsaI-digested vector backbone was then assembled with eblocks ordered from IDT using NEBuilder HiFi DNA assembly master mix (New England BioLabs) according to the manufacturer’s protocol. Vector backbone pU6-- pegRNA-GG-acceptor (Addgene, #132777) was used for pegRNA and sgRNA cloning and pU6-tevopreQ1-GG-acceptor (Addgene, #174038) was used for epegRNA cloning. Genotypes of mutants are shown in Table S5. All pegRNAs, nicking sgRNAs and epegR- NAs used in this study are provided in the key resources table and Table S6A. PegRNAs designed to install the 77 pathogenic edits into endogenous sites in HEK293T cells were designed using pegRNA spacer and PBS sequences reported previously.40 All epegRNA sequences used to install these edits are provided in Table S3. Fragments assembled after Gibson Assembly were transformed into One Shot Mach1 cells (Thermo Fisher Scientiﬁc) and subse- quently plated in 2 x YT agar with the appropriate antibiotics. Illustra TempliPhi 100 ampliﬁcation kit (Cytiva) was used to amplify plasmid DNA before sending it for Sanger sequencing (Quintara Biosciences). Bacterial clones with the veriﬁed plasmids were grown in 2 x YT media with the appropriate antibiotics. Plasmid DNA used for mammalian cell transfections were isolated using either QIAGEN Plus Midi Kit or Qiagen Plasmid Plus 96 Miniprep Kit while all other plasmids were isolated using QIAprep Spin Miniprep Kit. All isolated plasmid DNA were eluted in nuclease-free water and quantiﬁed using NanoDrop One UV-Vis spectrophotometer (Thermo Fisher Scientiﬁc). Phylogenetic tree analysis RT protein sequences were collected by searching the UniProt database with the BLASTP algorithm using query sequences listed in Table S1. Each individual BLASTP result was ﬁltered to remove duplicate sequences, sequences shorter than 100 residues, and se- quences longer than 1000 residues. To reduce phylogenetic complexity, 9–10 representative sequences were randomly sampled from each ﬁltered BLASTP result. The 543 RT sequences used for downstream phylogenetic analyses are listed in Table S1. Phylo- genetic analyses were performed using Geneious Prime. The MUSCLE algorithm was used to generate a multiple sequence align- ment of all 543 RT sequences. From this sequence alignment, an unrooted tree was generated using the neighbor-joining tree build method with the Jukes-Cantor genetic distance model. Bacteriophage cloning Phage cloning was performed in a two-step manner as previously described.62,63 Brieﬂy, Gibson Assembly was performed to clone a donor plasmid encoding for either the appropriate reverse transcriptase fused to an Npu C-terminal intein or the entire prime editor protein between two LguI (Life Technologies) type IIS restriction sites. Golden Gate assembly64 was performed with the donor plasmid along with two other previously reported plasmids (pBT114-splitC and pBT29-splitD) that each encode for one part of a Cell 186, 3983–4002.e1–e13, August 31, 2023 e7 ll OPEN ACCESS Resource two-part split phage genome. For Golden Gate assembly, all three plasmids were incubated between 30 min and 18 h with LguI enzyme and T4 DNA ligase at 37(cid:4)C. Following assembly, the reaction was transformed into chemicompetent S206065 E. coli host cells that contain plasmid pJC175e. We refer to this strain as S2208. Plasmid pJC175e supplies gIII under the phage shock promoter, enabling activity-independent phage propagation. After transformation, the cloned phage was grown overnight in Davis Rich Medium (DRM) at 37(cid:4)C with the appropriate antibiotics. Bacteria were then centrifuged for 5 min at 8,000 g and plaqued (see below). Individual plaques were picked and grown in DRM until the culture reached late growth phase. Bacteria were centrifuged and the supernatant containing phage was isolated. Colony PCR was performed and sent for sanger sequencing (Quintara Biosciences) to conﬁrm that the phage encoded for the correct insert. Preparation of chemically competent cells Strain S2060 was used in all experiments. Chemically competent cells were prepared as previously described.66 Brieﬂy, an overnight culture of bacteria was diluted 50-fold in 2 x YT media with appropriate antibiotics and grown at 37(cid:4)C, shaking at 230 RPM until the culture reached an optical density (OD600) of 0.4–0.6. Cells were then centrifuged at 4(cid:4)C for 10 min at 4,000g. The supernatant was discarded, and the cell pellets were resuspended in ice-cold TSS solution (LB media supplemented with 5% v/v DMSO, 10% w/v PEG 3350, and 20 mM MgCl2). Resuspended cells were aliquoted, frozen in dry ice and stored at (cid:3)80(cid:4)C until use. Phage-based luciferase assay Phage-based luciferase assays were performed as described previously.63 For each replicate, one colony of the evolution strain was grown overnight to saturation in DRM and appropriate antibiotics and then back-diluted 50-fold into DRM with appropriate antibi- otics. Cultures were grown at 37(cid:4)C with shaking at 230 RPM until cultures reached OD600 = 0.4. The mid-log culture was distributed into a 96-well black clear-bottomed plate (Corning), 135 mL of culture per well. 15 mL of high-titer (1 x1011 pfu/mL) phage were added to each well. The plate was covered with a breathable seal and incubated, shaking at 37(cid:4)C and 230 RPM for 3.5 h. Luminescence and OD600 were measured using a plate reader (TECAN). Values reported are OD600-normalized luminescence. Plasmid-based luciferase assay Strains for plasmid-based luciferase assays were made by transforming chemicompetent S2060 E. coli with all necessary plasmids, recovering in antibiotic-free DRM for 2 h, and then plating on 2x YT agar containing maintenance antibiotics and 100 mM glucose. For each biological replicate, one colony was picked into DRM and grown overnight. The following day, cultures were back-diluted 50-fold into DRM and antibiotics. For induced samples, arabinose was added to a ﬁnal concentration of 20 mM. Cultures were grown shaking at 230 RPM and 37(cid:4)C for 3 h, after which 150 mL were removed, placed into a 96-well black clear-bottomed plate (Corning), and measured for luminescence and OD600 on a plate reader (TECAN). Values reported are OD600-normalized luminescence. Overnight propagation assay For each replicate, a single colony of a host strain was picked and grown overnight in DRM and appropriate antibiotics. Saturated cultures were back-diluted 50-fold into DRM with appropriate antibiotics and grown for (cid:2)2 h, at 37(cid:4)C and 230 RPM until OD reached approximately 0.4. For each phage sample, 1 mL of this mid-log culture was placed into a well of a 96-well deep well plate and then infected with 1E5 total phage. Cultures were grown overnight (37(cid:4)C and 230 RPM), and then centrifuged for 10 min at 3400g. Super- natant containing phage was collected and then plaqued to determine total number of output phage. Fold propagation is the total number of output phage divided by the number of input phage. Plaquing Plaquing was performed as previously described.66 Brieﬂy, a saturated culture of S2208 E. coli was back-diluted 50-fold into DRM containing 50 mg/mL carbenicillin. 2 h later, the mid-log culture (OD = (cid:2)0.5) was used for plaquing. For each phage to be plaqued, three 100-fold serial dilutions of the sample were made using DRM. 10 mL of the original concentrated sample or each serially diluted sample was combined with 100 mL of mid-log 2208 culture. Immediately after mixing the bacteria and the phage, 1 mL of top agar (2:1 ratio of 2x YT media: 2x YT agar, stored at 55(cid:4)C until use) was added to the phage/bacteria solution, mixed quickly, and then immediately plated on 2x YT agar plates containing no antibiotics and 0.04% Bluogal (Gold Biotechnologies). The following day, the number of blue plaques were counted for whichever dilution (either the concentrated sample or one of the 100-fold dilutions) gave a discernable number of blue plaques. This number was then used to calculate the concentration of the phage sample in pfu/mL. For cases where activity-dependent plaquing was used, the relevant selection strain replaced S2208s. Phage-assisted noncontinuous evolution (PANCE) To perform one passage of PANCE, chemicompetent selection strains were transformed with MP6,67 recovered for 2 h in DRM without antibiotics, and then plated on 2x YT agar plates containing maintenance antibiotics for the selection strain, 25 mg/mL chlor- amphenicol, and 100 mM glucose. The following day, (cid:2)10 colonies were selected from the plate, pooled in DRM containing 25 mg/mL chloramphenicol and maintenance antibiotics, and grown to OD 0.5. Arabinose was then added to the mid-long culture to reach a ﬁnal concentration of 20 mM to induce MP6 expression. Immediately after addition of arabinose, 1 mL of this culture per PANCE replicate was infected with 1E5 pfu of phage and then incubated in a 37(cid:4)C shaker at 230 RPM overnight. The following day, cultures e8 Cell 186, 3983–4002.e1–e13, August 31, 2023 Resource ll OPEN ACCESS were centrifuged for 10 min at 3400g and the supernatant containing propagating phage was collected and used to infect the next round of evolution. Phage titer after each round was determined using qPCR (see below), Typically, 20 mL of phage were used to infect the next round of evolution (a 1:50 dilution). If phage titers were exceptionally high (1E7 PFU/mL or greater), then a 1:100, 1:200, or 1:1000 dilution factor was used instead. If titers were exceptionally low (less than 1E5 PFU/mL), a passage of drift was per- formed. For drift passages, 2208s containing MP6 were used instead of selection strains. In drift passages, phage were only allowed to propagate for 6–8 h instead of overnight to minimize recombination-mediated cheating. Once a noticeable change in phage propagation in the selection strain occurred, phage were plaqued using 2208s or the selection strain. Individual plaques were then ampliﬁed by PCR using primers JLD 1311 and JLD 1313 (see Table S6B) and submitted for Sanger sequencing to generate in- puts for Mutato analysis (https://hub.docker.com/r/araguram/mutato). qPCR determination of PANCE and PACE titers Phage titers in PANCE were estimated using qPCR as previously described.66 For each qPCR titer experiment, in addition to phage pools from evolution, a standard phage sample of a known high titer (1X1010 pfu/mL as determined by plaquing) was treated iden- tically to create a standard curve. To titer a phage sample, eight serial 10-fold dilutions of phage were made into DRM (no antibiotics). 25 mL of each serial dilution was heated to 80(cid:4)C for 30 min. Then 5 mL of heat-treated phage we combined with 44.5 mL of 1x DNase buffer and 0.5 mL of DNase (NEB). The DNase mixture was heated to 37(cid:4)C for 20 min and then 95(cid:4)C for 20 min to remove genomes from replication-incompetent polyphage. 1.5 mL of the heat-inactivated DNase mixture was pipetted into a 28 mL Q5 High-ﬁdelity PCR reaction (NEB) containing SYBR Green (Invitrogen) and primers M13-fwd and M13-rev (see Table S6B). qPCR was run on a Biorad CFX96 Real Time system with the following cycling conditions: 98(cid:4)C for 2 min, [98(cid:4)C for 10 s, 60(cid:4)C for 20 s, 72(cid:4)C for 15 s]x40. Cq values for phage of known titer were used to generate a standard curve, and other samples’ Cq values were used to calculate phage titer in pfu/mL. Phage-assisted continuous evolution (PACE) Chemicompetent selection strains were transformed with MP6, recovered for 2 h in DRM without antibiotics, and then plated on 2x YT agar plates containing maintenance antibiotics for the selection strain, 25 mg/mL chloramphenicol, and 100 mM glucose. The following day, colonies were picked into DRM and appropriate antibiotics into wells of the top row of a deep well 96-well plate and serially diluted 5-fold down the rows of the plate. The plate was incubated shaking at 37(cid:4)C and 230 RPM overnight. The next day, wells with an OD600 between 0.1 and 0.9 were pooled, diluted to a total volume of 140 mL in DRM and maintenance antibiotics and grown (37(cid:4)C, 230 RPM) until OD600 reached 0.5. This culture was used to ﬁll an 80 mL chemostat and four 15-mL lagoons. The ﬁlled chemostat and lagoons were inserted into a PACE apparatus. Conﬁguration of the PACE apparatus was identical to pre- viously described setups.66 The ﬂow rate for the chemostat was controlled by a Masterﬂex L/S Digital Drive Pump (Cole-Parmer) us- ing a Masterﬂex L/S Multichannel pump head. Supplement solution for a PACE carboy was made with 500 mL DI water, 59 g Harvard Custom Media C, 50 mL of 0.1M CaCl2, 120 mL of a trace metal solution, 400 mg chloramphenicol pre-dissolved in 3 mL of ethanol, and appropriate maintenance antibiotics for the selection strain (500 ng carbenicillin, 1 g spectinomycin, and 300 mg kanamycin, as needed depending on the PACE strain). The supplement was then combined with a 20 L solution of Harvard Custom Media A to create PACE media. This ﬁnal media was used as input into the chemomstat. The 80 mL chemostat was maintained at OD = (cid:2)0.5, starting with a ﬂow rate of approximately 80 mL/h. The chemostat’s effective ﬂow rate (vol/h) was adjusted throughout the PACE experiment to maintain a constant OD600, either by increasing the ﬂow rate on the pump or by decreasing the chemostat vol- ume by lowering the waste needle. Chemostat waste was collected in a carboy containing bleach. Lagoon ﬂow rates were also controlled by a Masterﬂex L/S Digital Drive Pump (Cole-Parmer) using a Masterﬂex L/S Multichannel pump head. Mid-log culture from the chemostat was used as the input for all lagoons, and lagoon waste was collected in a carboy containing bleach. To achieve MP6 induction in the lagoons but not the chemostat, arabinose was continuously added to each lagoon. 250 mM arabinose was taken up into a 50 mL syringe, and using a six-channel programmable syringe pump (New Era NE-1600), arabinose was pumped into each lagoon (0.6 mL/h of arabinose for a 15 mL/h lagoon ﬂow rate). The PACE apparatus was allowed to equilibrate for 1–12 h before phage infection. To begin the PACE, all pumps were turned off, and a total of 1.5E8 pfu were injected into each lagoon. After 10 min, pumps were turned back on, and (cid:2)400 mL was removed from each lagoon for the t = 0 timepoint. Lagoon ﬂow rates began at 0.5 vol/h. Subse- quent timepoints were taken every 8–24 h, and each phage sample was stored at 4(cid:4)C after removal from the lagoon. Immediately after sample collection, lagoon titers were measured using qPCR. If titers were the same as or higher than the previous timepoint, the ﬂow rate was increased by 0.5 vol/h, and arabinose pump rates were adjusted accordingly. If titers were decreasing, ﬂow rate was held constant. Plaquing was used to determine more accurate titers for reporting in ﬁgures. At the end of the PACE experiment, phage were plaqued in two different strains to check for cheating (S2060s to check for gIII recombinants and S2060s transformed with a pT7-gIII plasmid one to check for T7 recombinants), and ampliﬁed by PCR to check for bands corresponding to typical cheater recombinants using primers JLD 1311 and JLD 1313. If cheating was not detected (i.e., no plaques on cheater strains and no additional bands via PCR), phage were plaqued in either 2208s or the selection strain. Individual plaques were then ampliﬁed by PCR and submitted for Sanger sequencing to generate inputs for Mutato analysis. (https://hub. docker.com/r/araguram/mutato). Cell 186, 3983–4002.e1–e13, August 31, 2023 e9 ll OPEN ACCESS Resource Transfection of HEK293T, N2a, and Huh7 cells All transfections used to evaluate editors in mammalian cells were performed in TC-treated 96-well plates (Corning). For both HEK293T cells and N2a cells, a T-75 ﬂask of cells was washed with PBS, trypsinized using TrypLE Express enzyme (Thermo Fisher Scientiﬁc), and diluted to a concentration of 1.6E5 cells/mL in DMEM (10% FBS, no antibiotics). 100 mL of diluted cells were added to each well of a 96-well plate. 18–24 h after plating, cells were transfected. For unmodiﬁed HEK293T cells, the following conditions were used: 100 ng editor, 40 ng of pegRNA, and 13 ng nicking sgRNA (or, if conducting a twinPE experiment, 40 ng of the other pegRNA) plasmid were combined in a total volume of 6.25 mL Opti-MEM (Thermo Fisher Scientiﬁc) per well. For each well, 0.5 mL of Lipofectamine 2000 (Thermo Fisher Scientiﬁc) was mixed with 5.75 mL OptiMEM and then combined with the DNA mixture. 10 min later, the DNA/lipid mixture was added dropwise to cells. For the HEK293T Tay Sachs model cell line, the following conditions were used: 200 ng editor, 40 ng pegRNA, 13 ng nick- ing sgRNA. For N2a cells, the procedure was the same as HEK293T cells, except the plasmid DNA amounts differed: for PE3, 175 ng editor, 50 ng pegRNA, and 20 ng nicking sgRNA (or, if conducting a twinPE experiment, 50 ng of the other pegRNA) were used. For PE5 experiments in N2as, 100ng of MLH1dn plasmid was added. For the twinPE transfection performed in Huh7 cells, 150,000 cells were plated in poly-D-lysine-coated 24-well plates (Corning) in DMEM plus GlutaMAX supplemented with 10% FBS. After 16–24 h, cells were transfected with 400 ng of prime editor plasmid DNA, and 40 ng of each pegRNA plasmid DNA with 2 mL Lipofectamine 2000 (Thermo Fisher Scientiﬁc), according to the manufacturer’s protocol. HTS sample preparation 72 h following transfection, cells were washed with PBS (Thermo Fisher Scientiﬁc) and lysed for 1 h at 37(cid:4)C in lysis buffer (10 mM Tris- HCl pH 8, 0.05% SDS and 25 mg/mL proteinase K (Thermo Fisher)). Lysate was then heat inactivated at 80(cid:4)C for 30 min 1 mL of lysate was used as an input for PCR1. PCR1 reactions were 25 mL total, using the Phusion Hot Start II kit (Thermo Fisher), 0.75 mL of DMSO, and 0.125 mL of each 100mM primer (sequences listed in Table S6B). PCR1 was performed under the following cycle conditions: 98(cid:4)C for 3 min, [98(cid:4)C 15 s, 61(cid:4)C 30 s, 72(cid:4)C 30 s]x29, 72(cid:4)C 2 min. Exceptions to these cycling conditions include: N2a sites Pcsk9 and Dnmt1 used an annealing temperature of 70(cid:4)C instead of 61(cid:4)C, and for twinPE edits, 25 cycles were performed as opposed to 29, in order to decrease PCR bias. Samples were barcoded in a second PCR reaction (PCR 2). PCR2 reactions were 25 mL total, using the Phusion Hot Start II kit (Thermo Fisher Scientiﬁc), 1.25 mL each of 10 mM Illumina barcoding primers, and 1 mL of PCR1. All PCR2 reactions were performed using the following cycling conditions: 98(cid:4)C for 3 min, [98(cid:4)C 15 s, 61(cid:4)C 30 s, 72(cid:4)C 30 s]x8, 72(cid:4)C 2 min. After PCR2, samples of similar lengths were pooled and gel extracted in a 1% agarose gel using a Qiaquick gel extraction kit (Qiagen). Concentrations of puriﬁed libraries were determined using a Qubit double-stranded DNA high sensitivity kit (Thermo Fisher Scientiﬁc) according to the manu- facturer’s instructions. Libraries were diluted to 4nM and sequenced using a Miseq (Illumina) using an Illumina Miseq v2 Reagent kit or an Illumina Miseq v2 Micro Reagent kit using single read cycles. HTS analysis Samples were demultiplexed with Miseq Reporter (Illumina). CRISPResso2 was used to analyze demultiplexed reads. For samples in which the prime edit was a single base change, samples were aligned to the wild type amplicon in batch mode (see Table S6C), using the following parameters: ‘‘-q 30’’, ‘‘-discard_indel_reads TRUE’’, and ‘‘-qwc’’. The value of the qwc parameter, which deﬁned the portion of the sequence to be analyzed for indels, differed for each amplicon. The qwc interval included 10 bp before the ﬁrst nick of the amplicon (whether that was the prime editing nick site or the PE3 nicking guide nick site) to 10 bp after the second nick of the amplicon (whether that was the prime editing nick site or the PE3 nicking guide nick site). To calculate percent editing, the percent base change was multiplied by an indel correction factor. Percent base changes were found in the CRISPResso2 output ﬁle titled ‘‘Reference.Nucleotide_percentage_summary.txt’’. The indel correction factor was obtained by dividing ‘‘reads aligned’’/‘‘reads aligned all amplicons’’ values in the ‘‘CRISPResso_quantiﬁcation_of_editing_frequency.txt’’ CRISPResso2 output ﬁle. To calculate percent indels, ‘‘Discarded’’ was divided by ‘‘reads aligned all amplicons’’ in the same ﬁle. For samples in which the prime edit was multiple base changes or an insertion or deletion, CRISPResso2 was run in HDR batch mode. Parameters were identical to those described above for single nucleotide changes, but an additional parameter ‘‘e’’ was included, the value of which was the sequence of the desired, edited amplicon. For these types of edits, percent editing was calcu- lated by dividing the HDR-aligned reads/reads aligned all amplicons and then multiplying by 100. Indels were calculated by adding the ‘‘Discarded’’ reads from the reference-aligned sequences and the ‘‘Discarded’’ reads from the HDR-aligned sequences and then dividing that sum by ‘‘reads aligned all amplicons’’. All of these values are found in the ‘‘CRISPResso_quantiﬁcation_of_editing_fre- quency.txt’’ ﬁle when HDR mode is used. To quantify scaffold integration, a custom python script available in Note S1 was used. For each condition, scaffold integration is the percentage of (number of amplicons with scaffold-templated bases)/(number of reads that align to the amplicon). e10 Cell 186, 3983–4002.e1–e13, August 31, 2023 Resource ll OPEN ACCESS In vitro transcription (IVT) of editor mRNA IVT of editor mRNA was performed as described previously.33 Editors were cloned into pT7 expression constructs (example Addg- ene 178113). To generate linear DNA templates for IVT, the pT7-editor plasmids were ampliﬁed by PCR using the Phusion U green multiplex master mix (NEB) using primers IVT-fwd and IVT-rev (Table S6B). PCRs were puriﬁed using the QIAquick PCR puriﬁcation kit (Qiagen) and eluted in water. IVT reactions were performed using a T7 high yield RNA synthesis kit (NEB), following the manufac- turer’s directions with two exceptions: Trilink’s CleanCap reagent AG was added, and the uridine 50 triphosphate in the kit was re- placed with N1-methylpseudouridine 50 triphosphate (Trilink). Each 160 mL reaction used 8 mL 10x reaction buffer, 8 mL 100 mM ATP, 8 mL 100 mM CTP, 8 mL 100 mM GTP, 8 mL 100 mM N1-methylpseudouridine 50 triphosphate, 6.4 mL 100 mM CleanCap AG, 16 mL T7 RNAP mix, and 1 mg of puriﬁed linear template DNA. After assembly, reactions were incubated at 37(cid:4)C for 4 h. Samples were then DNase treated by adding 544 mL water, 80 mL DNase reaction buffer (NEB), and 60 mL DNaseI (NEB) to the IVT reaction. Samples were incubated at 37(cid:4)C for 15 min, and RNA was puriﬁed using a lithium chloride precipitation, following by two washes in 70% ethanol. RNA was resuspended in nuclease-free water, and purity and quality were veriﬁed using a 2% agarose gel stained with SYBER Gold (Thermo Fisher Scientiﬁc). RNA was stored at (cid:3)80 until use. Electroporation of patient-derived ﬁbroblasts An 80% conﬂuent T-75 ﬂask of patient-derived ﬁbroblasts (Coriell) were washed with PBS (Thermo Fisher Scientiﬁc), trypsinized us- ing TrypLE Express enzyme (Thermo Fisher Scientiﬁc), and suspended in 10 mL of media. The following media was used for each patient-derived ﬁbroblast line: low-glucose DMEM (Sigma Aldrich) supplemented with 10% (v/v) FBS and 2mM GlutaMAX Supple- ment (Thermo Fisher Scientiﬁc) for Tay Sachs Disease (ID: GM00221), high-glucose DMEM (Thermo Fisher Scientiﬁc) supplemented with 15% (v/v) FBS and 2mM GlutaMAX Supplement (Thermo Fisher Scientiﬁc) for Pompe Disease (ID: GM20092) and EMEM (ATCC) supplemented with 15% (v/v) FBS for both Crigler-Najjar Syndrome (ID: GM09551) and Bloom Syndrome (ID: GM02085). Cells were transferred to falcon tubes and centrifuged for 5 min at 150 g. During centrifugation, RNA reagents were prepared. For each sample, 1 mL of 1 mg/mL editor mRNA was added to a PCR tube, along with 0.45 mL of a 200 mM HEXA1278ins correction pegRNA solution and 0.6 mL of a 100 mM HEXA1278ins correction nicking sgRNA solution. (See key resources table for sequences of epegRNA and nicking sgRNA). An SE cell line kit (Lonza) was used to perform electroporation. 90.2 mL of SE nucleofector solution was mixed with 19.8 mL of supplement solution to make reconstituted Lonza buffer. Pelleted cells were washed with PBS and resuspended in the reconstituted Lonza buffer. 20 mL of resuspended cells was added to each editor/epegRNA/nicking guide mixture, transferred to a cuvette (Lonza), and electroporated using program CM130 on a Lonza 4D nucleofector with X unit (100,000 cells per electroporation condition). Immediately after electroporation, 80 mL of media was added to each well and incubated at room temperature for 10 min 1 mL of media was aliquoted into each well of a 24 well plate, and all cells were transferred to this plate. Cells grew for 5 days, with a media change at day 3, before lysis and sequencing. Electroporation of primary human T cells T cells were cultured in X-VIVO TM 15 Serum-free Hematopoietic Cell Medium (Lonza, Basel, Switzerland) supplemented with: 300 IU/mL IL-2 (PeproTech, Cranbury, NJ), GlutaMAX (Gibco, Waltham, MA), N-acetyl-cysteine (Sigma Aldrich, St. Louis, MO), 5% AB human serum (Valley Biomedical, Winchester, VA), 50 U/mL penicillin and 50 mg/mL streptomycin (Gibco, Waltham, MA). T-cells were stimulated with a 3:1 ratio of Dynabeads Human T-Expander CD3/CD28 beads (Thermo Fisher Scientiﬁc, Waltham, MA) and cells. At 72 h, the beads were removed and 300,000 T-cells were electroporated with 1 mL (1 mg) of editor mRNA, 1 mL (2 mg) of MLH1dn mRNA, 0.9 mL (100 mM) pegRNA, and 0.6 mL (100 mM) nicking sgRNA using the Neon electroporation system (ThermoFisher) with 10 mL tips and instrument settings of 1,400 V, 10 ms, and 3 pulses. Cells were cultured for 72 h followed by DNA isolation using the QuickExtract DNA Extraction Solution. TDT assay and analysis HEK293T cells were transfected in a 96 well plate as described above using 200 ng of editor and 40 ng of pegRNA. (No nicking guides were used for TDT transfections). 24 h after transfection, cells were lysed using 50 mL of lysis buffer per well (47.5 mL Beckman lysis Buffer (Beckman Coulter), 1.25 mL of 1M DTT, and 1.25 mL of proteinase K (Thermo Fisher). Genomic DNA was puriﬁed using the Beckman bead puriﬁcation kit (Beckman Coulter) and eluted in 40 mL of water. 10 mL of puriﬁed genomic DNA was used in a 50 mL tailing reaction (1X TDT buffer, 0.25 mM CoCl2, 100 mM dGTP, 10 units of terminal transferase, NEB). Samples were incubated at 37(cid:4)C for 30 min and then 70(cid:4)C for 10 min. The tailed DNA was isolated from the reaction mixture using the Beckman bead puri- ﬁcation kit again and eluted in 20 mL of water. 5 mL of puriﬁed tailed DNA was used as input for a 50 mL PCR1 reaction. TDT PCR1 reactions were performed with Phusion U Green Multiplex PCR Master Mix (25 mL), 5 mL of puriﬁed tailed DNA, 19.5 mL of water, and 0.25 mL of 100 mM primers. For TDT assay sequencing, one site-speciﬁc primer and one polyC primer (see Table S6B) were used for PCR1. PCR2 and Miseq were then performed as described above in ‘‘HTS sample preparation’’. To analyze TDT samples, a custom Python script (Note S2) was used to analyze demultiplexed fastq ﬁles. For scaffold insertion plots (Figure S4F), TDT results are plotted as the percentage of total edit-containing ﬂaps of a given length. For plots showing the lengths of RTT-encoded ﬂaps synthesized (Figures 4D and S4C), all RT products (ﬂaps length 1 or more) were counted, regardless of whether or not they contained the entire edit. Because polyG tailing was used, ﬂap lengths corresponding to a ﬂap ending in G are not detected. Cell 186, 3983–4002.e1–e13, August 31, 2023 e11 ll OPEN ACCESS Resource Secondary structure preduction using NUPACK38 Using the ‘‘old’’ NUPACK website (https://old.nupack.org/), the sequence of the pegRNA RTT and PBS was entered as the strand1 sequence using the RNA setting, a temperature of 37(cid:4)C, and default other parameters. This measure of folding free energy does not consider the pegRNA spacer, scaffold, or epegRNA 30 pseudoknot motif, as they are not directly engaged by the RT. UMI sample prep and analysis Unique molecular identiﬁers (UMIs) were applied in a three-step PCR protocol as previously described.9 Brieﬂy, linear ampliﬁcation was ﬁrst performed with 1uL of genomic DNA, Phusion U Green Multiplex PCR Master Mix and 0.1 mM of only the forward primer containing a 15-nt UMI in a 25 mL reaction (eleven cycles of 98(cid:4)C for 1 min, 61(cid:4)C for 25 s and 72(cid:4)C for 1 min). 1.6x AMPure beads (Beckman Coulter) was used to purify the PCR products in 20 mL nuclease-free water, according to the manufacturer’s protocol. For the second PCR, a forward primer that binds to the P5 Illumina adaptor sequence located at the 50 end of the UMI primer was used. This PCR was performed using 2uL of puriﬁed linear DNA, 0.5 mM of each forward and reverse primer and Phusion U Green Multiplex PCR Master Mix for 30 cycles in a 25 mL reaction. In the third PCR, 1 mL of product from the second PCR was ampliﬁed for 10 cycles using Phusion U Green Multiplex PCR Master Mix to add unique Illumina barcodes and adaptors as has been described earlier. The products from the third PCR were then pooled, separated by electrophoresis on a 1% agarose gel and puriﬁed with QIAquick Gel Extraction Kit (QIAGEN). The library was quantiﬁed using Qubit 3.0 Fluorometer (Thermo Fisher Scientiﬁc) and ﬁnally sequenced using the MiSeq Reagent Kit v2 or MiSeq Reagent Micro Kit v2 (Illumina) with 300 single-read cycles. AmpUMI60 was used to UMI deduplicate the raw sequencing reads. The UMI-deduplicated R1s were then analyzed using CRISPResso2 as described earlier.59 AAV production Transfer vectors were designed and generated as previously described (see v3em constructs from Davis et al.25). epegRNA sequences were changed to change the target edit. For transfer vectors using PE6c, further truncation of the Tf1 RT allowed us to minimize prime editor size an additional 100 bp to facilitate AAV packaging. For the single ﬂap loxP insertion single ﬂap edit at the Dnmt1 locus, the 40-bp loxP sequence was inserted, along with 2 additional bp of ﬁller sequence to preserve the frame of the Dnmt1 open reading frame after editing. AAV production was performed as previously described.25,68 HEK293T/17 cells (ATCC) were cultured in DMEM with 10% fetal bovine serum without antibiotics in 150-mm2 dishes (Thermo Fisher Scientiﬁc) and passaged every 2–3 days at 37(cid:4)C with 5% CO2. Cells were split 1:3, 18–22 h before transfection. 5.7 mg AAV genome, 11.4 mg pHelper (Clontech), and 22.8 mg AAV9 rep- cap plasmid were transfected per plate using polyethyleneimine (PEI MAX, Polysciences). Media was exchanged for DMEM with 5% fetal bovine serum the following day. Three days after the media change, cells were harvested using a rubber cell scraper (Corning), pelleted via centrifugation (10 min, 2,000 g) and resuspended in 500 mL hypertonic lysis buffer (40 mM Tris base, 2 mM MgCl2, 500 mM NaCl, and 100 U mL(cid:3)1 salt active nuclease (ArcticZymes)) per plate, and incubated at 37(cid:4)C for 1 h. The media was decanted and combined with 5x solution of poly(ethylene glycol) (PEG) 8000 (Sigma-Aldrich) and NaCl to achieve a ﬁnal con- centration of 8% PEG and 500 mM NaCl. This solution was incubated on ice for 2 h or overnight to facilitate PEG precipitation and then centrifuged (3,200 g, 30 min). The supernatant was discarded, and the pellet was resuspended in 500 mL hypertonic lysis buffer per plate. This was added to the cell lysate, which was either immediately ultracentrifuged or stored at 4(cid:4)C overnight. Cell lysates were ﬁrst clariﬁed by centrifugation at 3,400 g for 10 min and added to Beckman Coulter Quick-Seal tubes using a 16-gauge, 5-inch needle (Air-Tite N165) in a discontinuous gradient of iodixanol. The gradient of iodixanol was formed by sequentially ﬂoating the following layers: 9 mL 15% iodixanol in 500 mM NaCl and 1x PBS-MK (1x PBS with 2.5 mM KCl, and 1 mM MgCl2), 6 mL 25% iodixanol in 1x PBS-MK, and 5 mL each of 40% and 60% iodixanol in 1x PBS-MK. Phenol red was added to a ﬁnal concentration of 1 mg mL(cid:3)1 in the 15, 25, and 60% layers to facilitate layer identiﬁcation. Ultracentrifugation was performed at 58,600 rpm for 2 h 15 min at 18(cid:4)C using a Ti 70 rotor in an Optima XPN-100 Ultracentrifuge (Beckman Coulter). After centrifugation, an 18-gauge needle was used to remove 3 mL of solution from the 40–60% iodixanol interface. This solution was buffer exchanged using PES 100 kD MWCO columns (Thermo Fisher Scientiﬁc) with cold PBS containing 0.001% F-68 and ﬁnally sterile ﬁltered using a 0.22-mm ﬁlter. The ﬁnal concentrated AAV solution was quantiﬁed using qPCR (AAVpro titration kit, Clontech) and stored at 4(cid:4)C until use. Animals All mouse experiments were approved by the Broad Institute Institutional Animal Care and Use Committee and consistent with local, state, and federal regulations (as applicable), including the National Institutes of Health Guide for the Care and Use of Laboratory Animals. For P0 studies, timed pregnant C57BL/6J mice were purchased from Charles River Laboratory. All mice were housed in a room maintained on a 12 h light and dark cycle with ad libitum access to standard rodent diet and water. P0 ventricle injections All in vivo editing experiments were conducted via an ICV injection performed on day P0. P0 ventricle injections were performed as described previously.25,68 Drummond PCR pipettes (5-000-1001-X10) were pulled at the ramp test value of a Sutter P1000 micropi- pette puller and passed through a Kimwipe three times to achieve a tip diameter size of (cid:2)100 mm. To assess ventricle targeting, a small amount of Fast Green dye was added to the AAV injection solution. Using the included Drummond plungers, 4 mL of the injection e12 Cell 186, 3983–4002.e1–e13, August 31, 2023 Resource ll OPEN ACCESS solution was loaded via front ﬁlling. Cryoanestheisa was used to anesthetize the P0 pups. Successful anesthesia was veriﬁed by color and unresponsiveness to bilateral toe pinch. Then, 2mL of the injection solution was injected freehand into each ventricle. Transillu- mination of the head was used to assess ventricle targeting by the spread of Fast Green throughout the ventricles. Genders of mice and viral doses used for in vivo experiments are as follows (M = male, F = female, vg = viral genomes): Low-dose twinPE attB ins: [PEmaxDRNaseH: 3M + 1F, PE6d: 2M +2F, PE6c: 2M + 2F, untreated 3F]. Treated mice received 2E10 vg of each PE virus and 1E10 vg of GFP-KASH virus. Low-dose PE loxP ins. [PEmaxDRNaseH: 2M + 1F, PE6d: 2M + 1F, untreated: 1M]. Treated mice received 1E10 vg of each PE virus and 1E10 vg of GFP-KASH virus. High-dose PE loxP ins. [PEmaxDRNaseH: 3M, PE6d: 2M + 1F, untreated 1M, 2F]. Treated mice received 5E10 vg of each PE virus and 1E10 vg of GFP-KASH virus. We note that the prime editor AAV doses used in these experiments (1.35x1013 total vg/kg to 6.75x1013 total vg/kg) is 1.6-fold–8- fold lower than the 1.1x1014 vg/kg dose used in FDA-approved AAV therapies.69 Mice tissue collection All mice were sacriﬁced by CO2 asphyxiation, and tissues were immediately dissected. To harvest the cortex, hemispheres were ﬁrst split sagittally using a razor blade. The cortex (neocortex + hippocampus) was then isolated using a microspatula. Nuclear isolation and sorting Nuclear isolation and sorting were performed as described previously.25,68 Dissected cortex tissue was ﬁrst homogenized using a glass Dounce homogenizer (Sigma-Aldrich; D8938) with 20 strokes of pestle A followed by 20 strokes of pestle B in 2 mL of ice- cold EZ-PREP buffer ((Sigma-Aldrich). Sample was decanted into a new tube with additional 2 mL of cold EZ-PREP buffer on ice and centrifuged (500g, 4(cid:4)C). The supernatant was decanted, and the nuclei pellet was resuspended in 4 mL of ice-cold Nuclei Sus- pension Buffer (NSB: 100 mg/mL BSA (New England Biolabs) and 3.33 mM Vybrant DyeCycle Ruby (Thermo Fisher) in PBS). The sample was again centrifuged at 500g for 5 min at 4(cid:4)C, the supernatant was decanted, and the nuclei were resuspended in 1 mL of NSB. Samples were passed twice through a 35-mM cell strainer before ﬂow sorting using the Sony MA900 Cell Sorter (Sony Biotechnology) at the Broad Institute ﬂow cytometry core. See Figure S7B for example FACS gating. Nuclei were sorted into DNAdv- ance lysis buffer, and the genomic DNA was puriﬁed according to the manufacturer’s protocol (Beckman Coulter). Analysis of off-target editing Previously identiﬁed murine Dnmt1 off-target sites26,50 were ampliﬁed from either bulk or sorted cells from the mouse cortex. One of the off-target sites did not amplify efﬁciently by PCR. CRISPRESSO was run without an e ﬂag (not in HDR mode), with indels dis- carded, and with a w value of 20. Off-target edits were counted as leniently as possible: percent off-targets was calculated as the sum of indel reads and editing reads divided by the total number of reads aligned for all amplicons x 100. Off-target indels were counted as the number of discarded reads for the sample. To calculate off-target editing events, the pegRNA-encoded sequence was compared to the off-target site. The ﬁrst SNP at which the two sequences differed was used as a marker for off-target editing: all reads containing that SNP were counted as off-target editing events, even if they did not contain the entire loxP insertion. QUANTIFICATION AND STATISTICAL ANALYSIS The number of independent biological replicates and technical replicates for each experiment are described in the ﬁgure legends or the STAR Methods section. Cell 186, 3983–4002.e1–e13, August 31, 2023 e13 Resource Supplemental ﬁgures ll OPEN ACCESS (legend on next page) ll OPEN ACCESS Resource Figure S1. Characterization and engineering of reverse transcriptase enzymes for prime editing, related to Figure 1 (A) Native small RT enzymes demonstrate poor activity in the prime editing system (HEK293T cells, HEK3 +5 G to T edit). RT enzymes engineered in Figure 1 are highlighted in green, and the wild-type M-MLV RT used in the PE1 system is highlighted in black. All other enzymes are in red. Dots reﬂect the mean of n = 3 independent replicates. Of these enzymes that can support detectable mammalian PE activity, 11 are closely related to the M-MLV RT and are encoded by retroviruses, two are encoded by LTR retrotransposons, and seven are bacterial RTs from group-II introns, retrons, or CRISPR-Cas associated systems. (B) Overview of twinPE. The prime editor protein (gray and blue) uses two pegRNAs (dark blue and teal) to target opposite strands of DNA. The prime editor generates two 3’ ﬂaps (red) that are complementary to each other. After these newly synthesized 3’ ﬂaps anneal and the original DNA sequence in the 50 ﬂaps is degraded, the edited sequence in the ﬂaps is permanently installed at the target DNA site. (C) Incorporation of each of the ﬁve mutations analogous to those in PE2 (D200N, T306K, W313F, T330P, and L603W) improves the activity of four retroviral RT enzymes in HEK293T cells. PERV = porcine endogenous retrovirus RT, AVIRE = avian reticuloendotheliosis virus RT, KORV = koala retrovirus RT and WMSV = woolly monkey sarcoma virus RT. Combining all ﬁve mutations together (Penta) further improves the activity of each enzyme. All values from n = 3 independent replicates are shown. Horizontal bars show the mean value. (D) Structure-guided rational engineering of the Tf1 RT identiﬁes ﬁve mutations that improve prime editing in HEK293T cells. The solved structure of the Tf1 RT homolog, Ty3 RT, was used to predict mutations that could increase contacts of the RT with its DNA-RNA substrate (PDB: 4OL8). All values from n = 3 inde- pendent replicates are shown. Horizontal bars show the mean value across all sites and replicates. (E) Combining all mutations identiﬁed from structure-guided rational engineering improves the activity of the Tf1 RT prime editor in HEK293T cells. The ﬁnal rationally designed Tf1 variant (rdTf1) is a combination of ﬁve mutations: K118R, S188K, I260L, R288Q and S297Q. All values from n = 3 independent replicates are shown. Horizontal bars show the mean value. (F) AlphaFold-predicted structure of the Ec48 RT enzyme. The predicted structure aligns well with the RT from the xenotropic murine leukemia virus-related virus (XMRV, PDB: 4HKQ), a close relative of the M-MLV RT.70 (G) Aligning the AlphaFold-predicted structure of the Ec48 RT (blue) with the RT from xenotropic murine leukemia virus-related virus (XMRV, PDB: 4HKQ, yellow), a close relative of the M-MLV RT, suggests that the residue analogous to the D200 residue in M-MLV RT is the T189 residue in Ec48 RT. (H) Structure-guided rational engineering of the Ec48 RT identiﬁes six mutations that improve prime editing in HEK293T cells. An AlphaFold-generated predicted structure of the Ec48 RT was overlayed with the structure of the RT from the xenotropic murine leukemia virus-related virus (XMRV) (PDB: 4HKQ) to perform structure-guided mutagenesis. All values from n = 3 independent replicates are shown. Horizontal bars show the mean value. (I) Positions of residues (red) proximal to the substrate that were mutated to improve the activity of the Ec48 RT prime editor. Residues are mapped onto the predicted AlphaFold structure of the Ec48 RT aligned with the solved substrate of the XMRV RT (PDB: 4HKQ). L182 and T385 are proximal to the DNA substrate (green), R315 and K307 are proximal to the RNA substrate (yellow) and R378 is proximal to both the DNA and RNA rate. (J) Combining the top three mutations identiﬁed from structure-guided engineering improves the activity of the Ec48 RT prime editor in HEK293T cells. The ﬁnal rationally designed Ec48 RT variant (rdEc48) contains three mutations: L182N, T189N and R315K. All values from n = 3 independent replicates are shown. Horizontal bars show the mean value. Resource ll OPEN ACCESS Figure S2. Design and validation of a PE-PACE circuit, related to Figure 2 (A) Summary of phage-assisted continuous evolution (PACE). In both PACE and PANCE, the desired activity of a biomolecule of interest is linked to propagation of a modiﬁed M13 bacteriophage. To achieve this linkage, gIII, a gene required for phage propagation, is moved from the phage genome to a plasmid in host E. coli cells under the control of a gene circuit, such that gIII expression and phage propagation are only possible if the phage contain gene(s) that encode proteins with the desired activity. Simultaneous expression of mutagenic proteins from the MP6 plasmid mutagenizes the phage, including the gene of interest.67 During PACE, continuous dilution of a ﬁxed-volume ‘lagoon’ with fresh host cells selects for rapidly propagating phage encoding molecules that trigger gIII expression (Fig- ure S2A). PANCE uses the same selection strategy, but is implemented using discrete dilution steps every 12–24 h (Figure S2B)32: PANCE thus offers higher sensitivity (lower stringency) and greater ease of parallelization than PACE, with the trade-off of slower evolution. Both methods can complete dozens of (legend continued on next page) ll OPEN ACCESS Resource generations of mutagenesis and selection every 24 h. Host E. coli (gray) harboring relevant selection circuit plasmids (green, pink, and orange) and the muta- genesis plasmid (MP, black) continuously ﬂow into a ﬁxed-volume lagoon (left). Addition of arabinose induces expression of mutagenic genes on the MP. Se- lection phage (blue) harboring an NpuC-RT transgene (purple) infect the E. coli and are mutagenized. If a mutagenized RT is inactive (red, bottom/right), then prime editing does not trigger gIII expression and pIII production, and phage are not able to propagate. These phage encoding inactive RTs are washed out of the lagoon by continuous ﬂow. If a mutagenized RT is active (green, center), then prime editing leads to pIII production, and phage encoding that RT can propagate faster than the rate at which they are diluted out of the lagoon. (B) Summary of phage-assisted non-continuous evolution (PANCE). The same principles shown above in Figure S2A are used in PANCE, except periodic discrete dilution steps instead of continuous ﬂow is used to dilute selection cultures. Mid-log phase cultures of selection E. coli are infected with phage, and arabinose is added to induce mutagenesis (left). After an overnight incubation, cultures are centrifuged to pellet bacteria and allow isolation of propagating phage from the supernatant (middle). A small volume of supernatant (typically a 1:50 dilution factor) is used to infect a fresh lagoon of mid-log selection strains (right). This process is iterated until phage titers stabilize (i.e., when overnight phage propagation is equal to or greater than the dilution factor). (C) Effect of pegRNA optimization on PE2 phage propagation. Overnight propagation of empty phage (native control, red), PE2 phage (purple), and T7 RNAP phage (positive control, green) in strains harboring pegRNAs of different PBS and RTT lengths. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. This data was used to generate Figure 2C. (D) Luciferase assay to screen pegRNAs for the v2 PE-PACE circuit. Selection strains encoding luxAB transcriptionally coupled to gIII were infected with either empty phage (red) or PE2 phage (purple). 4 h after infection, OD600-normalized luminescence was measured as a proxy for circuit activation. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. Strains in which PE2 phage outperformed empty phage were used for v2 evolutions. (E) Overnight propagation of pools of wild-type RT and evolved RT phage on their cognate or noncognate host-cell selection strains. Additional evolved pools of phage are shown here beyond those provided in Figure 2K. Phage were from PANCE on the v1 circuit (yellow bars), from PANCE on the v2 circuit (blue bars), or wild-type-PE2 phage (gray bars). Propagation was then measured in the v1 circuit (left) or the v2 circuit (right). Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (F) Design of v3 circuit and improvements compared to v1 and v2 designs. A long insertion edit (20-bp insertion edit with a 60-bp RTT) was used to select for high- processivity, high-activity prime editors. Unlike v1 and v2 circuits, the v3 pegRNA (gray) targets the noncoding strand of T7 RNAP; this shortens the time between prime editing and wild type T7 RNAP production. In addition to the 20-bp insertion (green) needed to restore the frame of T7 RNAP, the v3 pegRNA also encodes silent PAM edits (maroon) and a seed edit (blue) that prevents subsequent binding and nicking of the edited sequence. Resource ll OPEN ACCESS Figure S3. Evolution and characterization of compact RTs for prime editing, related to Figure 3 (A) Overnight propagation of phage encoding dead M-MLV RT (red), Gs (blue), or PE2 (purple) RTs in the NpuC-RT phage architecture in the pegRNA-optimized v1 PE-PACE circuit. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (B) Phage titers during PANCE of NpuC-Gs-RT phage. Gray shading indicates a passage of evolutionary drift, in which phage were supplied gIII in the absence of selection to allow free mutagenic replication. Titers of four replicate lagoons are shown. (C) PACE of NpuC-Gs-RT phage. The left y axis and pink and blue lines show the SP titer of three different replicate lagoons at various timepoints. The right y axis and dotted gray line show the ﬂow rate in volumes per hour. (D) Indel frequencies for prime editors in the optimized PEmax architecture containing either engineered pentamutant Marathon RT (Marathon penta, red), evoEc48 (blue), or evoTf1 (yellow) with PEmax (gray) in HEK293T cells. Editing frequencies corresponding to this data is in Figure 3F. Bars reﬂect the mean of three independent replicates. Dots show individual replicate values. (E) Performance of PE6a and PE6b in the presence and absence of epegRNAs in HEK293T cells. All values from n = 3 independent replicates are shown. Horizontal bars show the mean value. (F) Comparison of PE6a, PE6b, and PEmax at three longer, complex edits in HEK293T cells. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. ll OPEN ACCESS Resource Figure S4. Development and characterization of highly processive, dual AAV-compatible RTs, related to Figure 4 (A) Editing efﬁciencies of prime editors containing single M-MLV mutants in HEK293T cells. Prime editing efﬁciencies used are the frequency of the intended prime editing outcome with no indels or other changes at the target site. Lines reﬂect the mean of n = 2 independent replicates per edit. Dots show individual replicate values. (B) Overview of the terminal deoxynucleotidyl transferase (TdT) assay for directly sequencing newly reverse-transcribed DNA ﬂaps that have not been incor- porated into the genome. 24 h after treatment with a prime editor and pegRNA, cells are lysed, and DNA is puriﬁed to capture and sequence newly reverse- transcribed DNA before its incorporation into the genome. A terminal transferase enzyme (yellow) adds a polyG sequence to all DNA 30 ends. PCR ampliﬁcation for high-throughput DNA sequencing is performed using a locus-speciﬁc forward primer and a polyC reverse primer. (C) Results of a TdT assay on the HEK3 +1 FLAG insertion edit in HEK293T cells. The y axis indicates the percentage of total RT products of a given length, and the x axis represents the length of the product in base pairs. PEmaxDRNaseH is shown in gray, and PE6d is shown in blue. The lines are mean values from n = 3 biological replicates. (legend continued on next page) Resource ll OPEN ACCESS (D) Editing efﬁciencies of PE6b-d, PEmax, and PEmaxDRNaseH for edits engineered to contain varying levels of secondary structure. ‘‘UC’’ indicates an unpinned control for a corresponding hairpin edit. These values were used to generate the free energy vs. fold improvement plot in Figure 4G. All edits are in HEK293T cells. Individual replicates are shown, with n = 3 replicates per condition. (E) Editing efﬁciencies (left) and indel rates (right) of PE6d (blue) and PEmaxDRNaseH (gray) for a series of prime edits that use short unstructured pegRNAs in HEK293T cells. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (F) Results of a TdT assay on the RNF2 +5 G to T edit in HEK293T cells. Note that the x axis differs from other TdT plots shown in this study: instead of RTT- templated bases correctly installed, it quantiﬁes the number of sgRNA scaffold-templated bases aberrantly installed (for example, x = 1 indicates the addition of one extra scaffold-templated base). The y axis indicates the percentage of edit-containing ﬂaps that have a given number of scaffold-templated bases. For each prime editor, the line reﬂects the mean of n = 3 independent replicates. Pie charts indicate the percentages of edit-containing ﬂaps that either have %2 bp (solid color) or >2 bp (striped) of scaffold-templated bases. Data shown are the mean of three independent biological replicates. (G) Unique molecular identiﬁer (UMI) analysis of prime editing efﬁciencies for twinPE edits in N2a cells (left) and HEK293T cells (middle, right). UMI protocol was applied to remove PCR bias, and trends agree with the data shown in Figure 4. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. ll OPEN ACCESS Resource Figure S5. Comparison of PE6 variants with PEmax, related to Figure 5 (A) Prime editing efﬁciencies of the best performing PE6 variant (either PE6c or PE6d) normalized to the editing efﬁciency of PEmax at sites tested in Figure 5A. All values from n = 3 independent replicates are shown. Editing was performed in HEK293T cells. The horizontal bar shows the mean value. (B) Indel frequencies of PEmax, PE6c, and PE6d at edits tested in Figure 5A. This data was used for Figure 5B. Bars reﬂect the mean of three independent replicates. Editing was performed in HEK293T cells. Dots show individual replicate values. (legend continued on next page) Resource ll OPEN ACCESS (C) Screening PE6 variants for insertion of attB into the CCR5 locus in primary human T cells. Bars reﬂect the mean of n = 4 independent replicates for editing (red) and indels (gray). Dots show individual replicate values. (D) Absolute prime editing efﬁciencies of PE6 variants, PEmaxDRNaseH, and PEmax in HEK293T cells used to plot data for Figures 5D and 5E. Prime editing efﬁciencies used are the frequency of the intended prime editing outcome with no indels or other changes at the target site. Bars reﬂect the mean of three in- dependent replicates. Dots show individual replicate values. (E) Indel frequencies of PE6 variants, PEmaxDRNaseH, and PEmax in HEK293T cells used to plot data for Figures 5D and 5E. Bars reﬂect the mean of three independent replicates. Dots show individual replicate values. (F) Percentage of sequencing reads containing a pegRNA scaffold insertion after prime editing using PE6 variants, PEmaxDRNaseH, and PEmax in HEK293T cells. These reads contribute to the total indel frequency. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (G) Prime editing efﬁciencies for edits where PE6b or PE6c outperformed PEmax using a nicking gRNA. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. Prime editing efﬁciencies used are the frequency of the intended prime editing outcome with no indels or other changes at the target site in HEK293T cells. (H) Indel frequencies of PE6 variant and PEmax at sites shown in Figure 5F in HEK293T cells. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (I) Correction of mutation implicated in Pompe disease in patient-derived ﬁbroblast using PE6c and PEmax. Bars reﬂect the mean of n = 3 independent replicates for editing (red) and indels (gray). Dots show individual replicate values. (J) Distribution of editing outcomes after correction of the pathogenic mutation implicated in Pompe disease in patient-derived ﬁbroblasts using PE6c. The patient was heterozygous. Indel genotypes are shown. Interestingly, many of the indels detected at this site did not contain the silent PAM edit encoded by the pegRNA, suggesting those indels were not RT-templated products. ll OPEN ACCESS Resource Figure S6. Evolution and engineering of Cas9 mutants for PE, related to Figure 6 (A) Representative PACE campaign for the v1 circuit. Different colored lines represent different replicate lagoons. PACE experiments with less than four lagoons shown experienced cheating (activity-independent phage propagation likely from rare gene III recombination onto the SP) or washout (complete loss of viable phage) for one or more lagoons. Top graphs represent the phage titer over a PACE experiment. Bottom graphs show the ﬂow rate at the corresponding time. (B) Reversion analysis of EvoCas9-4 in HEK293T cells. Editing efﬁciency was normalized to the values obtained using PE2. Data are shown as individual data points for n = 3 biological replicates and as the grand mean across the four sites tested. (C) Structural analysis of mutations that harm mammalian prime editing activity. (Left) Structure (PDB: 4UN3) of wild-type Sp Cas9 (gray) bound to its guide RNA (purple) and DNA substrate (yellow/orange). Residue K1151 is shown in dark pink. (Right) Structure (PDB: 4OO8) of wild-type Sp Cas9 (gray) bound to its guide RNA (purple) and DNA substrate (orange). Wild-type residues K1003, K1014, and A1034 are shown in dark pink. (legend continued on next page) Resource ll OPEN ACCESS (D) To test whether mutations that disrupt DNA binding enhanced circuit propagation via mechanisms other than enhancing PE efﬁciency, we transformed E. coli with plasmids encoding a corrected wild-type T7 RNAP, the pegRNA used in the v1 circuit, a gIII-luxAB fusion under the T7 promoter, and either a wild-type or K1151E PE2 mutant under the control of an arabinose-inducible promoter. After induction, OD-normalized luminescence for n = 3 biological replicates were used to measure circuit turn on. This system assessed the effect of each editor on the expression of already-corrected T7 RNAP by luciferase signal. Compared to uninduced bacteria, strains induced to express PE2 exhibited a 2.8-fold lower luciferase signal. Strains induced to express the K1151E mutant, though, showed no reduction in T7 RNAP expression. These ﬁndings support a model in which PE-PACE not only selects for PE activity, but also selects for avoidance of impeding the expression of edited T7 RNAP. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. (E) Prime editing efﬁciencies N2a cells (left, Ctnnb1 through Pcks9) and HEK293T cells (right, CXCR4 through RNF2) used to generate the fold changes reported in Figure 6D. Individual replicates are plotted, with n = 3 biological replicates per edit. (F) Structure (PDB: 4UN3) of Cas9 (gray) bound to its sgRNA (purple). Residue H721, which is mutated to Tyr in evolutions, is shown in green sticks. Dotted lines denote predicted polar contacts between H721 and other atoms. The H721Y mutation is predicted to perturb an interaction between Cas9 and stem loop 2 of the guide RNA scaffold, so its effects may differ depending on the pegRNA used. ll OPEN ACCESS Resource Figure S7. In vivo prime editing with PE6c and PE6d delivered via dual AAV, related to Figure 7 (A) Further truncation of the Tf1 RT allowed us to minimize prime editor size an additional 100 bp to facilitate AAV packaging. Editing (yellow) and indels (gray) are shown for the installation of an attB sequence at the murine Rosa26 locus in N2a cells using either PE6c or a truncated variant of PE6c. Bars reﬂect the mean of n = 3 independent replicates. Dots show individual replicate values. The number below each variant indicates the number of DNA bases that have been deleted from the C-terminal end of the Tf1 gene. (B) Representative ﬂow plots for the isolation of unsorted and sorted nuclei from mouse cortices. Left: scatterplot of all events, gate A set to collect nuclei. Middle: selection of single-nuclei droplets in Gate B, Right: FITC signal was used to collect unsorted cells (Gate C) and transduced, GFP-positive cells (Gate D). (C) TwinPE editing efﬁciency of PEmaxDRNaseH and PE6c viruses in the mouse cortex. N- and C- terminal twinPE viruses are administered via ICV injection (4x1010 vg total) along with a GFP-KASH virus. Editing efﬁciencies (light and dark blue) and indel (black/gray) rates are shown to the right. Bars reﬂect the mean of n = 3–4 mice. Dots show individual mice. (D) Injection route and PE editing (Dnmt1 loxP insertion) efﬁciency of PEmaxDRNaseH and PE6d viruses at a low viral dose (2 x1010 vg total) in the mouse cortex. (Left) The C-terminal virus is modiﬁed to include one epegRNA and one nicking sgRNA to encode a PE edit as opposed to a twinPE edit. (Right) Editing efﬁciencies (light/dark pink) and indel rates (black/gray). Bars reﬂect the mean of n = 3 mice. Dots show individual mice. (legend continued on next page) Resource ll OPEN ACCESS (E) Off-target editing from AAV-treated and untreated mice. Bars reﬂect the mean of n = 3 mice. Dots show individual mice. PE6d bulk (light pink) and transduced (dark pink) values were either less than 0.1% on average or were not statistically signiﬁcant from untreated controls (light gray). For both ns notes, p = 0.08. Analyses were performed with an unpaired t test with Welch correction. The y axis indicates off-target editing and indels summed (see STAR Methods for calculation). OT6 failed to amplify by PCR. All treated samples are from the high AAV dose condition.
10.1016_j.str.2023.03.011	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Structure. Author manuscript; available in PMC 2023 May 15. Published in final edited form as: Structure. 2023 May 04; 31(5): 518–528.e6. doi:10.1016/j.str.2023.03.011. Structure of Anabaena flos-aquae gas vesicles revealed by cryo- ET Przemys1aw Dutka1,2, Lauren Ann Metskas2,7,8, Robert C. Hurt2, Hossein Salahshoor3, Ting-Yu Wang2,4, Dina Malounda1, George J. Lu1,9, Tsui-Fen Chou2,4, Mikhail G. Shapiro1,5,, Grant J. Jensen2,6,10, 1Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, CA 91125, USA 2Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA 91125, USA 3Division of Engineering and Applied Science, California Institute of Technology, Pasadena, CA 91125, USA 4Proteome Exploration Laboratory, Beckman Institute, California Institute of Technology, Pasadena, CA 91125, USA 5Howard Hughes Medical Institute, Pasadena, CA 91125, USA 6College of Physical and Mathematical Sciences, Brigham Young University, Provo, UT 84602, USA 7Present address: Biological Sciences Department, Purdue University, West Lafayette, IN 47907, USA 8Present address: Chemistry Department, Purdue University, West Lafayette, IN 47907, USA 9Present address: Department of Bioengineering, Rice University, Houston, TX 77005, USA 10Lead contact Graphical abstract This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Correspondence: [email protected] (M.G.S.), [email protected] (G.J.J.). AUTHOR CONTRIBUTIONS P.D. conceived experiments, prepared samples, acquired and analyzed data, performed data exploration, drafted the manuscript, and prepared the figures. L.A.M. initiated the project and collected data for Mega GVs. R.C.H. performed mutation screening for GvpA and participated in initial sample preparation and optimization for Mega GVs. H.S. performed finite element simulation and analyzed data. T.-Y.W. performed XLMS experiments and analyzed the data. D.M. expressed and purified GV samples. G.L. participated in initial sample preparation and optimization for Mega GVs. T.-F.C. supervised XLMS experiments. All authors participated in correction of the manuscript. M.G.S. participated in guidance, experimental design, funding, and correction/advising on writing the manuscript. G.J.J. participated in guidance, experimental design, funding, and correction/advising on writing the manuscript. SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.str.2023.03.011. DECLARATION OF INTERESTS The authors declare no competing interests. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 2 In brief Dutka et al. used cryo-ET supported by biochemical data and computational modeling to reveal the conserved structure of Anabaena flos-aquae gas vesicles. The resulting model gives insights into the distinctive mechanical properties of gas vesicles and their assembly. SUMMARY Gas vesicles (GVs) are gas-filled protein nanostructures employed by several species of bacteria and archaea as flotation devices to enable access to optimal light and nutrients. The unique physical properties of GVs have led to their use as genetically encodable contrast agents for ultrasound and MRI. Currently, however, the structure and assembly mechanism of GVs remain unknown. Here we employ cryoelectron tomography to reveal how the GV shell is formed by a helical filament of highly conserved GvpA subunits. This filament changes polarity at the center of the GV cylinder, a site that may act as an elongation center. Subtomogram averaging reveals a corrugated pattern of the shell arising from polymerization of GvpA into a β sheet. The accessory protein GvpC forms a helical cage around the GvpA shell, providing structural reinforcement. Together, our results help explain the remarkable mechanical properties of GVs and their ability to adopt different diameters and shapes. INTRODUCTION A fundamental property of many living organisms is their ability to move within their environment, with single-celled organisms capable of swimming, swarming, and aligning Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 3 with magnetic fields. The molecular machines underlying many of these motility functions have been characterized in detail.1–3 However, the structure underlying one of the oldest evolved forms of motility, flotation, remains more mysterious. Some cyanobacteria, heterotrophic bacteria, and archaea regulate their buoyancy in aquatic environments to access sunlight and nutrients using intracellular flotation devices called gas vesicles (GVs).4,5 These unique protein nanostructures consist of a gas-filled compartment, typically ~100 nm in diameter and ~500 nm in length, enclosed by a ~3-nm-thick protein shell (Figure 1A) that can withstand hundreds of kilopascals of applied pressure.6,7 The interior of the shell is strongly hydrophobic, keeping out water while allowing gas molecules to diffuse in and out on a sub-millisecond timescale.4,5 In addition to their biological significance, GVs are a subject of intense interest for biotechnology. Analogous to fluorescent proteins, opsins, and CRISPR nucleases, GVs’ unusual biophysical properties can be harnessed for other purposes. The gaseous composition of GVs allows them to scatter ultrasound waves, enabling their use as genetically encoded reporters and actuators of cellular function deep in tissues.8–14 Other applications take advantage of GVs’ refractive index, gas permeability, and susceptibility to magnetic fields.15–17 GVs were discovered in the 19th century, but we still have limited knowledge of their structure and assembly. GVs adopt a cylindrical shape with conical caps (Figure 1A). Their components are encoded in operons containing relatively few genes (8–23+, depending on the species).5 One of these genes encodes the main structural protein, GvpA, a small (~8-kDa), highly hydrophobic protein that polymerizes to form the GV shell.4 In some species, the gene cluster contains a secondary structural protein called GvpC, which binds to the exterior of the shell to provide mechanical reinforcement.18 The remaining genes encode proteins whose functions are not well understood, possibly including chaperones, assembly factors, and additional minor shell constituents. GVs are nucleated as bicones that then elongate into a cylindrical shape with low-pitch helical ribs,5,19 but their detailed molecular structure is not known. Here, we apply state-of-the-art cryoelectron tomography (cryo-ET) and subtomogram averaging techniques to GVs from the cyanobacterium Anabaena flos-aquae (Ana). These GVs are among the best studied by biophysicists4,20,21 and the most commonly used in biotechnology applications.13,22,23 We show that the Ana GV shell is formed by a continuous helical filament of repeating GvpA subunits, giving rise to a corrugated cylindrical structure with terminal cones that taper over a conserved distance. Near the middle of the cylinder, the angle of corrugation is inverted, suggesting a potential elongation center for GV biosynthesis. The corrugated shell is externally reinforced by circumferential rods of GvpC. Combining our cryo-ET data with an atomic model of the homologous Bacillus megaterium (Mega) GvpA protein determined in a complementary study,24 we build an integrative model of the Ana GV. This model explains the connection between the GV shell and GvpC and highlights the structural conservation of GVs between diverse species. Finally, we extend our study with biochemistry and computational modeling to corroborate our model and explore its implications for GV engineering. Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. RESULTS Molecular architecture of GVs Page 4 Ana GVs are long, cone-tipped cylinders with diameters of 85 ± 4 nm7 and lengths of 519 ± 160 nm6 (Figures 1A and 1B). Although GVs have apparent helical symmetry, they are prone to deformation in thin ice (Figure S1) and are therefore intractable for cryoelectron microscopy (cryo-EM) helical processing. For this reason, we decided to use cryo-ET. However, cryo-ET analysis of GVs presents its own challenges. We observed that GVs are highly sensitive to electron dose, losing high-resolution features quickly before deflating and shrinking (Video S1). To mitigate this effect, we limited the total electron dose to ~45 electrons/Å2 per tilt series, which is ~2.5 times lower than typically used for high-resolution subtomogram averaging.25,26 We started by examining large-scale structural features. While the diameter and length of GVs have been characterized,7,27 the conical ends and their connection to the cylindrical body are less studied. Close inspection of individual caps in our cryo-tomograms revealed a heterogeneous morphology that deviated from a simple conical structure (Figures 1C and 1D). We observed two elements in the majority of cones: a pointed closed tip and a rounded transition region between the cone and cylinder (Figure 1D). The height of the conical caps was 59 ± 6 nm, independent of cylinder diameter (Figure 1E). The rounding of the base was more pronounced in GVs with larger diameters, so we also examined cryo-tomograms of Mega GVs, whose average diameter is ~30 nm smaller than that of Ana GVs. However, Mega GVs showed similar rounding at the cap transition (Figure S2), suggesting that this is a conserved feature of the structure independent of width. The GvpA spiral reverses polarity in the middle of the cylinder The GV shell consists of a low-pitch helix, running the length of the GV (Figures 2A and 2B). Near the middle of the GV, however, the angle of the helix abruptly inverts. Previously, Waaland and Branton28 noticed that one rib in the middle of the GV cylinder appears to be thicker than the others and suggested that this could be the growth point, where new GvpA subunits are added. Indeed, this abnormal rib was clearly visible in our tomograms (Figure 2A). To obtain a better understanding of the rib architecture in that region, we applied subtomogram averaging, which revealed that the angle of corrugation is opposite above and below the central rib (Figure 2B). This polarity inversion occurs within one rib, and the continuity of the spiral is not broken (Figures 2B and 2C). We were unable to distinguish whether the polarity of GvpA subunits changed relatively gradually within the space of one helical turn or abruptly from one monomer to the next. We also could not tell whether additional proteins are present at the inversion point. By inspecting hundreds of cryo-electron micrographs of GVs from different species (Ana, Mega, and Halobacterium salinarum), we found that the polarity inversion point is a conserved feature (Figure S3). Although in general the inversion point was near the middle of the cylinder, in some cases it was located closer to one end (Figure S3A). If it is the nucleation point, then this suggests that GvpA subunits are not always added symmetrically in both directions. Additionally, we observed some examples where a GV exhibited different Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 5 diameters on either side of the inversion point (Figure S3B). While we saw examples in all three species, it was most frequent and most pronounced in GVs from H. salinarum (Halo). Subtomogram averaging of the GV shell To understand the molecular details of the GV structure, we applied subtomogram averaging to the Ana GV shell in its native state and after biochemically removing the reinforcing protein GvpC to produce “stripped” (AnaS) GVs. Initially, we tried averaging tubular sections of the GVs. However, because of flattening and the low number of particles, the resolution of this approach was limited (Figure 3A). As an alternative, we decided to average only small sections of the shell with randomly seeded particle centers similar to an oversampling method.25,29 This strategy produced a higher number of particles and allowed more rigorous 3D classification to remove distorted particles. With this method, we produced subtomogram averages of native Ana (Figure 3B and S4) and AnaS (Figure S5) GV shells with global resolutions of 7.7 Å and 7.3 Å, respectively (Table S1; Figures S4 and S5). Despite high global resolution, our maps manifested a certain degree of anisotropy with significantly lower resolution in the y direction (Figures S4D and S5D). The particle poses after subtomogram averaging indicate that all particles are oriented outward and consistent with a helical arrangement (Figure S6). Typically, we observed one significant break in the particle poses per GV, which corresponds to the inversion point. However, because of the strong effects of missing wedge artifacts on tubular structures, such as GVs, they typically appear as two disconnected arches. As a result, we observed a fraction of misaligned particles in the direction of the missing wedge. Furthermore, flattening of the GV cylinder and small variability in diameters could lead to inaccurate alignment of some particles, resulting in blurring of the structure, particularly in the y direction, and limiting resolvability of the secondary structures. Although the GV corrugated structure has strong features in the x and z directions, there are no features in the y direction that could aid subtomogram alignment. A visual examination of the maps revealed that, despite the lower resolution, the map for the native Ana GV shell had higher quality (Figures 3F, 3G, and S4C). For this reason, we used the native GV shell map for further interpretation, and the AnaS map was only used to determine the position of GvpC. The subtomogram average revealed a prominent pattern of beveled ribs, giving rise to the corrugated GV shell. The shell was ~4 nm wide at its thickest and only ~1 nm thick in the region between adjacent ribs (Figure 3C). We also observed pores in this region, at the interface between neighboring ribs of the spirals (Figure 3B), likely allowing gas to diffuse in and out of the GV. In contrast to the complex exterior face of the GV shell, the gas-facing interior appeared relatively smooth. Comparing the maps of native Ana and AnaS GVs (lacking GvpC), we noticed a pronounced rod-like structure positioned along the GV ribs that is absent in AnaS (Figures 3C–3E). Previously, various models for GvpC binding to the GV shell have been proposed,30 with most of the field favoring one in which GvpC spans longitudinally across GvpA ribs.13,31 Our structure shows instead that GvpC binds circumferentially to the thickest part of the GV shell, creating a spiral cage around the GV cylinder (Figures 3F–3H). We do not yet know whether the GvpC filament binds the central inversion rib or extends to Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 6 the conical caps, where the decreasing radius of curvature might be prohibitive, or whether it is continuous, as the average would blur away gaps. Conserved assembly of GvpA and its consequences on GV development and mechanics The resolution of our Ana GV density map was sufficient for rigid-body fitting of a homology model of GvpA. Taking advantage of the high degree of conservation of the protein, we used the structure of GvpA2 from Mega solved by helical reconstruction in a contemporaneous study.24 The only substantial difference between GvpA from Ana and Mega is an extended C terminus in the latter (Figure S7), so our homology model was complete and fit well into our cryo-ET density map (Figures 4A and S8). After docking the model to our map, we observed that the fit of α helices is not perfect. It could be due to the limited resolution of our maps or because these helices adopt a slightly different conformation compared with Mega GvpA2. The GvpA spiral is formed by polymerization of individual subunits, resembling the packing of amyloids. All domains of the small GvpA protein play a role in building the GV shell (Figure 4B), packing into a tight structure with only small pores contributing to the remarkable stability of GVs; we find that purified GVs are stable for years at cool or ambient temperature. As mentioned above, the only major difference between Mega GvpA2 and Ana GvpA is the presence of an elongated C terminus (Figure S7). This C terminus was not resolved in a recent structure solved by helical processing,24 presumably because of its flexibility. In our cryo-ET of Mega GVs, we observed additional density on the surface of the shell that is absent from the structures of AnaS and native Ana GV shells (Figure S9). The density was not highly regular but appeared connected. It may be that this extra density belongs to the C terminus of GvpA2, which perhaps plays a role in stabilizing the GV shell. The sequence of GvpA, the major structural protein, is highly conserved in all GV-producing species,33,34 and we think it is likely that its structure is similarly conserved, as evidenced by our ability to fit a model from Mega GvpA224 into the density of Ana GvpA. Remarkably, though, GvpA can assemble into GVs with varying diameters (Figure S10A)7 and morphologies (Figures S10B and S10C). For instance, the largest Halo GVs are ~7 times larger in diameter than the smallest Mega GVs. One key to understanding different morphologies may lie in what appears to be a hinge region located between helix α1 and strand β1 (Figure 4B), where a conserved glycine resides (Figures 4G and S7). Small sequence differences in GvpA have been suggested to contribute to different morphologies of GVs.4 Halo contains two independent GV gene clusters, p-vac and c-vac.5 The sequences of the GvpA encoded by the two clusters are 94% identical (Figure S7), but these cluster can produce GVs with a lemon shape (Figure S10B) or a more typical cylindrical shape with conical caps (Figure S10C). We used ConSurf32 to visualize the evolutionary conservation of GvpA, revealing that the most conserved residues are located in the β sheets and α helices (Figure 4C). In contrast, the N-terminal domain of the protein responsible for interactions between neighboring ribs showed the greatest variability (Figure 4C). Within the generally conserved β strands, the most variable sites were those interacting with the N terminus from the subunit below. This variability in amino acid composition in the domains responsible for holding adjacent ribs Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 7 together might be one factor contributing to differences in the mechanical strength of GVs. Under hydrostatic pressure, GVs can collapse, forming flattened sacs.7 The critical pressure required to collapse GVs varies greatly between species. For example, the hydrostatic collapse pressure threshold of Ana GVs is 587 kPa, while that of Halo GVs is 59 kPa, an order of magnitude lower.6 By EM imaging, we found that Ana GVs collapse without major disruptions to the rib structure (Figure 4D), while collapsed Halo GVs often exhibit major disruption of the rib structure and separation of the GvpA filament (Figure 4E). This supports the idea that the strength of connectivity between ribs varies between species. To test the importance of conserved GvpA residues in GV assembly, we mapped tolerated mutations by screening a scanning site saturation library of GvpA mutants in Escherichia coli engineered to express a hybrid gene cluster encoding the structural proteins GvpA and GvpC from the Ana GV gene cluster and the accessory proteins from the Mega GV gene cluster. GV-producing mutant clones were identified by nonlinear X-wave ultrasound (xAM) (Figures 4F, 4G, and S11). The results largely correlated with observed evolutionary conservation, with the highest number of function-retaining mutations occurring in the evolutionarily variable C-terminal coil (Figure 4G). Interestingly, the only conserved region that tolerated mutations well was helix α2, which is not involved in interactions between monomers but plays a crucial role in GvpC binding (see below). GvpC forms a helical spiral around the GV shell Having identified GvpC in our subtomogram average of the Ana GV shell (Figure 3H), we next investigated how GvpC binds to GvpA and how multiple GvpC proteins might cooperate to strengthen GVs. GvpC is predicted to form an amphipathic α-helical structure composed of a characteristic 33-residue repeating sequence4,35,36 (Figure S12A). Ana GvpC consists of 5 such repeats plus short N and C termini. To build a model of a GV shell decorated with GvpC, we fitted a poly-alanine helix of a length corresponding to one repeating unit into our subtomogram average (Figures 5A and 5B). We found that GvpC binds perpendicular to the surface-exposed α2 helices of GvpA, directly above the hydrophobic pockets (Figures 5B and 5C). Although there is insufficient density to anchor the helix, we predict that GvpC binds to GvpA with its hydrophobic side facing the shell. In addition to being amphipathic, GvpC also has an unequal distribution of charge (Figure S12B). In our model, GvpC binds directly above the negatively charged C terminus of GvpA (Figure S12C). One 33-residue repeating sequence of GvpC interacts with approximately four GvpAs, indicating a GvpC to GvpA ratio of at least 1:20 when saturated. This is close to the previously calculated ratio of 1:25.30 Despite multiple rounds of 3D classification and application of different focus masks, we were unable to resolve the junctions between neighboring GvpC molecules. Instead, GvpC appeared as a continuous helical belt. To get a better understanding of GvpC-GvpC and GvpC-GvpA interactions, we performed chemical cross-linking coupled with mass spectrometry (XLMS) (Table S2). Most of the cross-links we observed were between the N terminus of GvpA and apparently random locations on GvpC (Figure 5D), which is consistent with the close association between the N terminus of GvpA and the GvpA α2 helix in the adjacent rib, where GvpC binds, in our structure (Figure 5A). However, we Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 8 did not observe any cross-links between GvpC and helix α2, potentially because of the unfavorable orientation of the lysines. Among GvpC-GvpC cross-links, the most interesting was between K36 and K174 (Figure 5D). The distance between these residues is ~20 nm, too far for an intramolecular cross-link,37 suggesting that GvpC termini are either closely packed or potentially interact head to tail (Figure S13). To quantify the effect of increasing GvpC occupancy on GV stabilization, we used solid mechanics simulations to estimate the applied pressure at which the GV shell starts to buckle, a parameter relevant to its ability to withstand hydrostatic pressure as well as produce nonlinear signal in ultrasound imaging. We implemented several finite element models of a GV shell, each 500 nm in length and 85 nm in diameter and with a custom density of GvpC molecules. From a continuous belt, representing 100% GvpC, we randomly removed GvpC-length (25-nm) segments of the helix to achieve the desired saturation for each model (Figure 5E). We subjected the outer surface of each GV shell to uniform normal stress, simulating hydrostatic or acoustic pressure, and obtained a critical buckling pressure by linear buckling analysis. We observed a simple linear dependence of buckling on scaffolding protein density (Figure 5F), consistent with previous experimental findings that GvpC level can be utilized to modulate the GV buckling threshold.22 DISCUSSION The GV shell has remarkable mechanical properties; despite being only ~3 nm thick, it is highly stable and can withstand up to hundreds of kilopascals of pressure. This is achieved by tight packing of the GvpA subunits into a low-pitch helix that forms a corrugated cylinder. On the macroscopic level, corrugation is typically used when flexibility is important (e.g., pipes) or to increase durability and strength (e.g., unpressurized cans). One or both of these properties might be similarly important for GV function. Our data indicate that GV cylinders can be significantly deformed without collapsing the structure.7 This elasticity of the GV shell may be crucial for adapting to pressure fluctuations in vivo and enables GVs to be used as contrast agents in high-specificity nonlinear ultrasound imaging.38 We noticed a highly conserved glycine between helix α1 and strand β1 of GvpA. The single hydrogen in the side chain of glycine gives it much more flexibility than other amino acids,39 suggesting that this region may act as a hinge that confers elasticity on the shell structure and lets it adapt to different geometries, such as those observed in terminal cones or the bodies of lemon-shaped GVs. The primary contact between adjacent GvpA subunits is mediated by lateral interactions of antiparallel β strands in an extended sheet, resembling the aggregation of β-amyloids.40,41 Such assemblies are typically stabilized by an extensive network of backbone hydrogen bonding, conferring outstanding strength.42 Such strength is also observed in GVs from diverse species; individual GvpA monomers can only be dissociated from the polymer by harsh chemical treatment.43,44 That backbone interactions are the main force driving subunit polymerization is consistent with the wide range of diameters observed in different species;7 as the curvature of the cylinder changes, the relative positions of backbone residues will be affected much less than those of side chains. We find that GvpA domains involved in forming the GV wall have a low tolerance for mutations, likely because of selective pressure Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 9 to preserve the highly hydrophobic composition of the β sheets and maintain interactions with the linker domain connecting subsequent coils of the helix. Our scanning mutagenesis data largely correlate with results obtained for Halo GVs.45 Interestingly, however, Halo GVs appear to be more tolerant to mutations in the conserved regions, possibly because, unlike Ana or Mega GVs, they are synthesized without turgor pressure in the cells. Stacked ribs of the continuous GvpA polymer are joined by interactions of the coiled N termini from one row of subunits with the β strands of the subunits in the next. We observe that the strength of these inter-rib interactions varies between species, likely related to evolutionary variability in the N-terminal linker. It has been observed previously that the critical collapse pressure of Mega GVs is much higher than that of Ana or Halo GVs,6 likely because of the narrower diameter of Mega GVs.46–48 However, we note that the C terminus of Mega GvpA is longer than in other species, and in our tomograms of Mega GVs, we observed extended irregular surface densities connecting ribs. We suggest that these extra densities correspond to the extended C termini of Mega GvpA2 and may confer additional mechanical strength. Other mechanisms also enhance the strength of the GV shell. Almost all GV gene clusters encode an additional, minor structural protein, GvpC, that binds to the GvpA helical spiral and reinforces the shell;22,49 we find that GvpC binds to the surface-exposed α2 helix of GvpA. In our mutational analysis, this helix was relatively mutation tolerant, suggesting that it has a minimal role in GvpA shell integrity and instead acts primarily as an adapter for GvpC. In contrast to previous models of GvpC spanning ribs, we find that GvpC instead tracks along ribs, forming a spiral cage around the GV cylinder. Our XLMS results indicate close conjunction of GvpC molecules, and even with multiple masking and 3D classification strategies, we never observed discontinuity in the GvpC rod in our subtomogram averages. Although we could not resolve interactions between GvpC N and C termini, we showed previously that their removal leads to a significant drop in critical collapse pressure of Ana GVs.22 Here, we used finite element simulations to quantify the reinforcing effect of GvpC density on GV buckling and find that the degree of strengthening is directly proportional to the amount of GvpC bound. However, full GvpC occupancy is not required for full strengthening, and small gaps in the GvpC cage have a negligible effect on collapse pressure. Even though our work focused on Ana GVs, it is possible that the GvpC binding model is conserved between different species of GVs. Previously, the interaction between GvpA and GvpC was studied in Halo by split-GFP assay50 providing similar results to those obtained in our XLMS analysis. In the initial stage of assembly, GVs grow as bicones until reaching their target diameter; at that point, growth elongates the central section, producing cylinders that can reach several micrometers in length.5,10 The trigger for this transition is unclear. Our data show that the height of mature cones is relatively constant, regardless of GV diameter, indicating that the number of helical turns/height is the measured quantity rather than the number of GvpA subunits. Our observation of a polarity inversion near the middle of the GV suggests that this is the site of cylinder elongation, with individual subunits being incorporated in both directions. In some cases, we observed that the elongation center was located closer to one end of the GV, suggesting a mechanism that does not require GvpA subunits to be Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 10 added symmetrically in both directions. Although GV cylinders typically exhibit a uniform diameter, we documented some examples with different diameters on either side of the elongation center. We observed variations in the shape of conical ends within and between GVs. This suggests that mismatches in GV geometry might arise in the initial bicone growth stage, but further investigation is needed to fully dissect the mechanism of GV morphogenesis. Currently, the method of choice for solving the structure of helical assemblies is helical reconstruction.51,52 However, the large and nonuniform diameter of Ana GVs (~85 nm) and their susceptibility to deformation during cryopreservation present challenges for this approach. Cryo-ET and subtomogram averaging can circumvent these limitations by focusing on smaller and therefore more uniform 3D sections of the object of interest. Subtomogram averaging can reach high resolution in certain favorable cases, such as for large53 or symmetrical26,54 proteins, but for most targets, resolution has remained limited. Here we show that even with a fairly challenging target, recent developments in cryo-ET data collection and subtomogram averaging methods combined with integrative modeling make it possible to obtain a sufficient resolution to dock an atomic model. Our work, together with a complementary study of Mega GVs,24 advances our understanding of the molecular architecture of GVs and may inform further engineering of GVs to serve as genetically encoded contrast agents and biosensors. Limitations of the study Using subtomogram averaging, we determined the structure of the Ana GV protein shell, providing insight into GV morphogenesis and explaining their unusual mechanical properties. Because of the high conservation of GvpA, we were able to build an integrative model of the Ana GV shell using the homologous structure of Mega GvpA2.24 However, the limited resolution of our map only allowed rigid-body fitting. Despite the high homology of GvpA, there might be a subtle difference between the structure of GvpAs from different organisms, reflecting unique proprieties of each GV type. Additionally, we are not able to discern whether there are any conformational changes caused by GvpC binding. Future higher-resolution studies will be necessary to allow for flexible fitting of GvpA models to extend our knowledge on GV evolution and mechanics. Additionally, a better understanding of how GVs are assembled will require biochemical and structural work focusing on the GV initiation and elongation process. STAR★METHODS RESOURCE AVAILABILITY Lead contact—Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Grant J. Jensen ([email protected]). Materials availability—This study did not generate new unique reagents. Data and code availability—The unprocessed tilt series used for the data analysis are available upon request. Representative tomograms for Ana, Mega, and Halo GVs have Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 11 been deposited in the Electron Microscopy Data Bank under accession codes EMDB: EMD-29922, EMD-29925, EMD-29924, EMD-29923. Subtomogram averages for native Ana and AnaS GV shell have been deposited in EMDB under accession codes EMD-29921 and EMD-29916, respectively. The integrative model of Ana GvpA/GvpC has been deposited in the Protein Data Bank (PDB): 8GBS. The XLMS data have been deposited to the ProteomeXchange Consortium with the dataset identifier PXD038631. The code for ultrasound data collection and processing is available upon request. EXPERIMENTAL MODEL AND SUBJECT DETAILS GVs were produced either in native sources, Anabaena flos-aquae (Ana) and Halobacterium salinarum NRC1 (Halo), or expressed heterologously in Rosetta 2(DE3)pLysS Escherichia coli, Bacillus megaterium (Mega). We followed previously published protocols by Lakshmanan et al.6 describing in details bacterial growth conditions specific for production of each GV type investigated here. METHOD DETAILS GV preparation—GVs were isolated as previously described.6 In the final steps of buoyancy purification, the sample buffer was exchanged for 10 mM HEPES, pH 7.5. To obtain GVs stripped of GvpC (AnaS), 6 M urea solution was added to purified native GVs and two additional rounds of buoyancy purification were performed. AnaS GVs were subsequently dialyzed in 10 mM HEPES, pH 7.5. Concentrations were measured by optical density (OD) at 500 nm using a spectrophotometer (NanoDrop ND-1000, Thermo Scientific). Cryo-ET—A freshly purified GV sample was diluted to OD500 = ~20 (Ana and Halo), ~3 (AnaS), or ~1 (Mega) and mixed with 10 nm BSA-coated gold beads. A 3 μL volume of sample was applied to C-Flat 2/2 – 3C grids (Protochips) that were freshly glow-discharged (Pelco EasiGlow, 10 mA, 1 min). GV samples were frozen using a Mark IV Vitrobot (FEI, now Thermo Fisher Scientific) (4°C, 100% humidity, blot force 3, blot time 4 s). Tilt-series were collected on a 300 kV Titan Krios microscope (Thermo Fisher Scientific) equipped with a K3 6k × 4k direct electron detector (Gatan). Multi-frame images were collected using SerialEM 3.39 software55 using a dose-symmetric tilt scheme. Super- resolution movies were acquired at a pixel size of 0.8435 Å (53,000× magnification) with varying defocus from - 1.0 to - 3.5 μm. Tilt-series of Halo and Mega GVs were collected from −60° to 60° with 3° increments. Tilt-series of native Ana GVs were collected in two sessions. The first set was collected from −60° to 60° with 3° increments and the second from −44° to 44° with 4° increments. For AnaS GVs, data were collected from −45° to 45° with 3° increments. Due to the rapid shrinking of GVs during exposure to the electron beam (Video S1), the total accumulated dose in all cases was limited to 45 electrons/Å2. Data collection parameters are summarized in Table S1. Raw movies were binned by a factor of 2 and gain- and motion-corrected on-the-fly using Warp.56 Assembled tilt-series were exported to Dynamo57 for automated alignment using autoalign_dynamo.65 Aligned tilt-series were CTF corrected and full tomograms were either Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 12 reconstructed in Warp with a pixel size of 10 Å or manually aligned and reconstructed using Etomo.66 Subtomogram averaging - inversion point—Sub-volume extraction, alignment, and averaging were performed using the Dynamo software package.57 Particles for subtomogram averaging of the inversion site were manually selected from GVs with a diameter of ~85 nm, yielding a total of 68 particles. Sub-volumes were extracted from 4x binned tomograms with a final pixel size of 6.748 Å and 180x180x180 box size. The initial reference for particle alignment was generated by averaging segments with azimuth-randomized orientations. Due to the low number of particles, subtomogram averaging was not performed according to a gold standard. Instead, convergence of the structure was analyzed by changes in particle shifts and cross-correlation scores. During the final rounds of refinement, a soft cylindrical mask was applied to the central 40% of the GV tube. Subtomogram averaging - GV shell—Subtomogram averaging was carried out using Dynamo,57 Warp,56 Relion-3.1,58 and M,53 software packages. Data transfer between Dynamo and Warp/M was carried out using a set of tools provided by warp2catalogue and dynamo2m.65 Particle selection and initial reference generation were performed using the Dynamo package. Orientations and positions of shell sections were determined using geometrical tools for particle picking in Dynamo.67 Initial estimates of positions and orientations on the GV shell were generated with an interparticle distance of ~150 Å (~3 ribs). Particles were extracted in Dynamo with a pixel size of 10 Å and averaged. After removal of duplicated particles, data was transferred to Warp and subtomograms were reconstructed with a pixel size of 5 Å based on the alignment information from Dynamo. Subtomograms were subsequently refined in RELION, re-reconstructed at 2.5 Å /pixel and 3D classified without alignment in RELION. After 3D classification, several additional rounds of 3D refinement were carried out in RELION. Finally, subtomograms were reconstructed at 1.687 Å /pixel and iteratively refined in RELION and M using a soft- edged mask around ~3 or 4 adjacent ribs. Although we did not see a resolution boost after iterative refinement of the tilt-series parameters in M, subsequent refinement in RELION produced a better-quality reconstruction when applied to particles reconstructed after M refinement. Final maps were post-processed in RELION. The resolution was estimated using a soft-edged mask around ~3–4 adjacent ribs in 3DFSC program.64 The final results are summarized in Figures S4, S5, and Table S1. Model building and validation—Although the density map for AnaS reached a higher overall resolution, individual features were better resolved in the map of native Ana GVs (Figure S4), so all model building was performed using this map. To build the GvpA model, a high-resolution cryo-EM structure of the homologous GvpA2 from B. megaterium (PDB: 7R1C)24 was fitted into the segmented cryo-ET density map corresponding to an individual subunit in UCSF Chimera.60 The GvpA amino acid sequence was rebuilt by manual replacement of mismatched residues in Coot.63 The A. flos-aquae GvpA model was subsequently refined by rigid-body fitting using the Phenix real-space refinement tool.62 The refined GvpA model was used to populate a larger section of the cryo-ET map in UCSF Chimera.60 The multimeric GvpA model was further refined by rigid-body fitting in Phenix Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 13 to maximize fit into the density map. The GvpC model was built as a poly-Ala chain in Coot. The poly-Ala chain corresponds in length to a single 33-residue repeating sequence of GvpC and spans across four subunits of GvpA. The quality of the fit was analyzed by visual inspection and fitting scores from UCSF Chimera (Figure S8). We roughly placed four GvpA subunits at the height of one rib and performed a global search using “fitmap” command in Chimera. Subsequently, we analyzed scores for cross-correlation and fraction inside density for each fit. The three best results with similar fitting scores all fit our density map very well and are only different in that they shift by one subunit along Y (the are essentially all the same “fit”). We obtained similar results with a starting point at the height of other ribs. Negative stain EM—To prepare collapsed GV samples, the purified GV sample was diluted to OD500~ 0.5 and pressurized in a sealed syringe until the solution turned transparent. Three microliters of the target sample was applied to a freshly glow-discharged (Pelco EasiGlow,15 mA, 1 min) Formvar/carbon-coated, 200 mesh copper grid (Ted Pella) for 1 min before blotting. Afterward, the sample was incubated for 1 min with a 0.75% uranyl for-mate solution before blotting and air-dried. Image acquisition was performed using a Tecnai T12 (FEI, Thermo Fisher Scientific) EM at 120 kV, equipped with a Gatan Ultrascan 2 k×2 k CCD. Cross-linking mass spectrometry (XLMS)—The cross-linking procedure was carried out according to the manufacturer’s instructions (Thermo Fisher). In brief, a freshly purified sample of native Ana GVs in 10 mM HEPES, pH 7.5 was mixed with an excess of cross-linker: either DSSO or BS3 (Thermo Fisher). The sample was incubated for 1h at room temperature and subsequently the reaction was quenched with Tris buffer at a final concentration of 20 mM. The crosslinking samples were digested in an S-Trap mcrio spin column (Protifi, USA) according to the manufacturer’s instructions. For trypsin digestion, an additional aliquot of trypsin was added after 24 hours on the S-trap column and the digestion continued for another 24 hours. After elution and drying, peptides were suspended in LCMS-grade water containing 0.2% formic acid and 2% acetonitrile for further LC-MS/MS analysis. LC- MS/MS analysis was performed with an EASY-nLC 1200 (Thermo Fisher) coupled to a Q Exactive HF hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher). Peptides were separated on an Aurora UHPLC Column (25 cm × 75 μm, 1.6 μm C18, AUR2-25075C18A, Ion Opticks) with a flow rate of 0.35 μL/min for a total duration of 43 min and ionized at 1.7 kV in the positive ion mode. The gradient was composed of 6% solvent B (2 min), 6–25% B (20.5 min), 25–40% B (7.5 min), and 40–98% B (13 min); solvent A: 2% ACN and 0.2% formic acid in water; solvent B: 80% ACN and 0.2% formic acid. MS1 scans were acquired at a resolution of 60,000 from 375 to 1500 m/z, AGC target 3e6, and a maximum injection time of 15 ms. The 12 most abundant ions in MS2 scans were acquired at a resolution of 30,000, AGC target 1e5, maximum injection time 60 ms, and normalized collision energy of 28. Dynamic exclusion was set to 30 s and ions with charges +1, +7, +8, and >+8 were excluded. The temperature of the ion transfer tube was 275°C and the S-lens RF level was set to 60. For cross-link identification, MS2 fragmentation spectra were searched Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 14 and analyzed using Sequest and XlinkX node bundled into Proteome Discoverer (version 2.5, Thermo Scientific) against in silico tryptic digested Dolichospermum-flos-aquae GvpA from the Uniprot database. The maximum missed cleavages were set to 2. The maximum parental mass error was set to 10 ppm, and the MS2 mass tolerance was set to 0.05 Da. Variable crosslink modifications were set DSS (K and protein N-terminus, +138.068 Da) for BS3 crosslink and DSSO (K and protein N-terminus, +158.004 Da) for DSSO crosslink, respectively. For BS3 crosslink, the dynamic modifications were set to DSS hydrolyzed on lysine (K, +156.079 Da), oxidation on methionine (M, +15.995 Da), protein N-terminal Met-loss (−131.040 Da), and protein N-terminal acetylation (+42.011 Da). For the DSSO crosslink, the dynamic modifications were set to DSSO hydrolyzed on lysine (K, +176.014 Da), DSSO Tris on lysine (K, +279.078 Da), oxidation on methionine (M, +15.995 Da), protein N-terminal Met-loss (−131.040 Da) and protein N-terminal acetylation (+42.011 Da). Carbamidomethylation on cysteine (C, +57.021 Da) was set as a fixed modification. The false discovery rate (FDR) for crosslinked peptide validation was set to 0.01 using the XlinkX/PD Validator Node and crosslinks with an Xlinkx score greater than 30 were reported here. The raw data have been deposited to the ProteomeXchange Consortium68 via the PRIDE69 partner repository. Scanning site saturation library generation and screening—The scanning site saturation library was constructed via a Gibson assembly-based version of cassette mutagenesis as previously described.70 Briefly, the A. flos-aquae GvpA coding sequence was divided into sections that tiled the gene, and oligos were designed to have a variable middle region with flanking constant regions against which PCR primers with Gibson overhangs were designed. The variable region was designed to sequentially saturate each residue with every amino acid other than the WT at that position, plus a stop codon to produce truncation mutants (i.e., the size of such libraries is 20 [# of amino acids in the protein]). Oligos were synthesized as a pool by Twist Biosciences, and were amplified by 10 cycles of PCR (both to make them double-stranded and to add overhangs for Gibson assembly) using Q5 polymerase (according to the manufacturer’s protocol, but with 5 μM of each primer) and assembled with the rest of the GV gene cluster (i.e., Ana GvpC and Mega GvpR-GvpU) into a pET28a vector via Gibson assembly using reagents from New England Biolabs. Assembled libraries were electroporated into NEB Stable E. coli and grown in Lennox LB with 100 mg/μL kanamycin and 1% glucose. 71 Plasmid DNA was miniprepped (Econospin 96-well filter plate, Epoch Life Science) and verified by Sanger sequencing. Ultrasound-based phenotyping of mutants was performed in BL21-AI (Thermo Fisher) as previously described,23 and all screened mutants were sequenced using the evSeq pipeline.72 Finite element simulation—We first developed a finite element model of a single stripped GV isolated from A. flos-aquae (AnaS). The geometry, adapted from the cryo- EM images, comprises a cylindrical shell with conical ends, with height and diameter, respectively, of 500 nm and 85 nm. The protein wall was idealized as a continuum shell with a thickness of 2.8 nm and a shell density of 1350 kg/m3. The rib-like structure of the gas vesicle wall was mirrored in the computational model by an elastic anisotropic material model, with elastic moduli across and along the principal axis of the GV of 0.98 GPa and 3.92 GPa, respectively.38 In order to simulate the nearly incompressible nature of the protein Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 15 shell, we assigned a Poisson’s ratio of 0.499. We note that the material parameters were not obtained from direct experimental measurements, but rather chosen such that, in addition to falling within a range of parameters consistent with those of protein-based biological materials,73 they effectively replicated the buckling pressures observed experimentally. We next added a helical rod that spirals around the cylindrical portion of the GV shell, modeling the GvpC molecules. We modeled the GvpC rod as a shell of radius 0.6 nm. The helical structure was generated by assigning a pitch of 4.9 nm. The finite element model of the resultant wild-type GV was obtained by discretizing the entire geometry with quadrilateral shell elements of effective side length 1 nm with reduced integration (i.e., S4R elements) in Abaqus (Dassault Systemes Simulia, France). These general-purpose shell elements with only one integration point within each element are capable of capturing both tensile and in-plane bending, and, with a sufficiently fine mesh size, are computationally cost-effective. We subjected the interior surfaces of the GV to an initial pressure of 101 kPa, modeling the inner gas pressure. We further subjected the vertices at both the top and bottom conical ends of the GV to a zero-displacement Dirichlet boundary condition, which prevented rigid body translations and rotations of the entire GV structure. In order to investigate the effect of GvpC density on the buckling pressure, we first computed the total length of the helix where N, D, and z are the total number of turns, the perimeter of the GV cross-section, and the pitch of the helix, respectively. Given the pitch and the length of the cylindrical segment of the GV model, 416.5 nm, the total number of turns was computed as 85. We thus computed the total length of the helix as 22.702 micrometers. Given that the length of GvpC is ~25 nm, about 908 GvpC molecules constituted the helix in our model. We generated six additional finite element models with distinct GvpC saturation levels of 90%, 80%, 60%, 40%, 20%, and 10%, for which we randomly removed about 90, 180, 360, 540, 720, and 810 GvpC units, respectively. We conducted linear buckling analysis (LBA) and solved the corresponding eigenvalue problem to obtain the threshold buckling pressures for each model. We solved this problem using the Lanczos algorithm and obtained the first ten modes of buckling. Unlike the buckling modes (i.e., eigenvectors), which were virtually identical at different levels of GvpC saturation, the buckling pressures (i.e., eigenvalues) were remarkably dependent on the GvpC density, with an almost linear monotonic relation, where decreasing the saturation level decreases the buckling pressure. Figure S14 depicts the buckling modes and pressures for 100%, 60%, 20%, and 0% GvpC saturations. Bioinformatics and visualization—Protein sequence alignment was carried out using Clustal Omega74 and visualized with Jalview.75 Protein conservation analysis was performed using ConSurf.32 Data were visualized using GraphPad Prism, IMOD,59 Chimera,60 and ChimeraX.61 Identified crosslinks were visualized using xiNET.76 QUANTIFICATION AND STATISTICAL ANALYSIS The heights of the GVs’ conical ends (Figure 1E) were manually measured from cryo- electron tomograms using ImageJ. The average height is calculated from 132 conical ends and reported as mean ± SD. Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 16 The subtomogram averages were determined using software listed in the key resources table. Details of the data processing are displayed in Figures S4 and S5, and Table S1. The resolution anisotropy and final FSC curves (Figures S4D and S5D) were determined using the 3DFSC package. The ultrasound data (Figures 4F and S11), XLMS analysis (Figure 5D and Table S2), and finite element simulation (Figures 5E, 5F, and S14) were analyzed using software listed the key resources table. No other statistical analyses were performed. Supplementary Material Refer to Web version on PubMed Central for supplementary material. ACKNOWLEDGMENTS The authors are grateful to Catherine Oikonomou for helpful editorial comments. We thank Songye Chen for assistance with tomography data collection. Electron microscopy was performed in the Beckman Institute Resource Center for Transmission Electron Microscopy at Caltech. The Proteome Exploration Laboratory (PEL) is supported by the Beckman Institute and National Institutes of Health 1S10OD02001301. This work was supported by the National Institutes of Health (R01-AI127401 to G.J.J. and R01-EB018975 to M.G.S.) and the Caltech Center for Environmental Microbial Interactions (CEMI). Related research in the Shapiro Laboratory is supported by the Packard Foundation, the Chan Zuckerberg Initiative, and the Heritage Medical Research Institute. M.G.S. is a Howard Hughes Medical Institute Investigator REFERENCES 1. Komeili A, Li Z, Newman DK, and Jensen GJ (2006). Magnetosomes are cell membrane invaginations organized by the actin-like protein MamK. Science 311, 242–245. 10.1126/ science.1123231. [PubMed: 16373532] 2. Krause DC, Chen S, Shi J, Jensen AJ, Sheppard ES, and Jensen GJ (2018). Electron cryotomography of Mycoplasma pneumoniae mutants correlates terminal organelle architectural features and function. Mol. Microbiol 108, 306–318. 10.1111/mmi.13937. [PubMed: 29470845] 3. Wadhwa N, and Berg HC (2022). Bacterial motility: machinery and mechanisms. Nat. Rev. Microbiol 20, 161–173. 10.1038/s41579-021-00626-4. [PubMed: 34548639] 4. Walsby AE (1994). Gas vesicles. Microbiol. Rev 58, 94–144. [PubMed: 8177173] 5. Pfeifer F (2012). Distribution, formation and regulation of gas vesicles. Nat. Rev. Microbiol 10, 705–715. 10.1038/nrmicro2834. [PubMed: 22941504] 6. Lakshmanan A, Lu GJ, Farhadi A, Nety SP, Kunth M, Lee-Gosselin A, Maresca D, Bourdeau RW, Yin M, Yan J, et al. (2017). Preparation of biogenic gas vesicle nanostructures for use as contrast agents for ultrasound and MRI. Nat. Protoc 12, 2050–2080. 10.1038/nprot.2017.081. [PubMed: 28880278] 7. Dutka P, Malounda D, Metskas LA, Chen S, Hurt RC, Lu GJ, Jensen GJ, and Shapiro MG (2021). Measuring gas vesicle dimensions by electron microscopy. Protein Sci 30, 1081–1086. 10.1002/ pro.4056. [PubMed: 33641210] 8. Shapiro MG, Goodwill PW, Neogy A, Yin M, Foster FS, Schaffer DV, and Conolly SM (2014). Biogenic gas nanostructures as ultrasonic molecular reporters. Nat. Nanotechnol 9, 311–316. 10.1038/nnano.2014.32. [PubMed: 24633522] 9. Bourdeau RW, Lee-Gosselin A, Lakshmanan A, Farhadi A, Kumar SR, Nety SP, and Shapiro MG (2018). Acoustic reporter genes for noninvasive imaging of microorganisms in mammalian hosts. Nature 553, 86–90. 10.1038/nature25021. [PubMed: 29300010] Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 17 10. Farhadi A, Ho GH, Sawyer DP, Bourdeau RW, and Shapiro MG (2019). Ultrasound imaging of gene expression in mammalian cells. Science 365, 1469–1475. 10.1126/science.aax4804. [PubMed: 31604277] 11. Wu D, Baresch D, Cook C, Duan M, Malounda D, Maresca D, Abundo MP, Lee J, Shivaei S, Mittelstein DR, Qiu T, Fischer P, and Shapiro MG (2023). Biomolecular actuators for genetically selective acoustic manipulation of cells. Sci Adv 9, eadd9186. 10.1126/sciadv.add9186. [PubMed: 36812320] 12. Farhadi A, Bedrossian M, Lee J, Ho GH, Shapiro MG, and Nadeau JL (2020). Genetically encoded phase contrast agents for digital holographic microscopy. Nano Lett 20, 8127–8134. 10.1021/acs.nanolett.0c03159. [PubMed: 33118828] 13. Lakshmanan A, Jin Z, Nety SP, Sawyer DP, Lee-Gosselin A, Malounda D, Swift MB, Maresca D, and Shapiro MG (2020). Acoustic biosensors for ultrasound imaging of enzyme activity. Nat. Chem. Biol 16, 988–996. 10.1038/s41589-020-0591-0. [PubMed: 32661379] 14. Bar-Zion A, Nourmahnad A, Mittelstein DR, Shivaei S, Yoo S, Buss MT, Hurt RC, Malounda D, Abedi MH, Lee-Gosselin A, et al. (2021). Acoustically triggered mechanotherapy using genetically encoded gas vesicles. Nat. Nanotechnol 16, 1403–1412. 10.1038/s41565-021-00971-8. [PubMed: 34580468] 15. Shapiro MG, Ramirez RM, Sperling LJ, Sun G, Sun J, Pines A, Schaffer DV, and Bajaj VS (2014). Genetically encoded reporters for hyperpolarized xenon magnetic resonance imaging. Nat. Chem 6, 629–634. 10.1038/nchem.1934. [PubMed: 24950334] 16. Lu GJ, Farhadi A, Szablowski JO, Lee-Gosselin A, Barnes SR, Lakshmanan A, Bourdeau RW, and Shapiro MG (2018). Acoustically modulated magnetic resonance imaging of gas-filled protein nanostructures. Nat. Mater 17, 456–463. 10.1038/s41563-018-0023-7. [PubMed: 29483636] 17. Lu GJ, Chou L-D, Malounda D, Patel AK, Welsbie DS, Chao DL, Ramalingam T, and Shapiro MG (2020). Genetically encodable contrast agents for optical coherence tomography. ACS Nano 14, 7823–7831. 10.1021/acsnano.9b08432. [PubMed: 32023037] 18. Hayes PK, Buchholz B, and Walsby AE (1992). Gas vesicles are strengthened by the outer-surface protein. Arch. Microbiol 157, 229–234. 10.1007/BF00245155. [PubMed: 1510555] 19. Offner S, Ziese U, Wanner G, Typke D, and Pfeifer F (1998). Structural characteristics of halobacterial gas vesicles. Microbiology 144, 1331–1342. 10.1099/00221287-144-5-1331. [PubMed: 9611808] 20. Maley AM, Lu GJ, Shapiro MG, and Corn RM (2017). Characterizing single polymeric and protein Nanoparticles with surface plasmon resonance imaging measurements. ACS Nano 11, 7447–7456. 10.1021/acsnano.7b03859. [PubMed: 28692253] 21. Cai K, Xu B-Y, Jiang Y-L, Wang Y, Chen Y, Zhou C-Z, and Li Q (2020). The model cyanobacteria Anabaena sp. PCC 7120 possess an intact but partially degenerated gene cluster encoding gas vesicles. BMC Microbiol 20, 110. 10.1186/s12866-020-01805-8. [PubMed: 32375647] 22. Lakshmanan A, Farhadi A, Nety SP, Lee-Gosselin A, Bourdeau RW, Maresca D, and Shapiro MG (2016). Molecular engineering of acoustic protein nanostructures. ACS Nano 10, 7314–7322. 10.1021/acsnano.6b03364. [PubMed: 27351374] 23. Hurt RC, Buss MT, Duan M, Wong K, You MY, Sawyer DP, Swift MB, Dutka P, Barturen- Larrea P, Mittelstein DR, Jin Z, Abedi MH, Farhadi A, Dephande R, and Shapiro MG (2023). Genomically mined acoustic reporter genes for real-time in vivo monitoring of tumors and tumor- homing bacteria. Nat. Biotechnol 10.1038/s41587-022-01581-y. 24. Huber ST, Terwiel D, Evers WH, Maresca D, and Jakobi AJ (2023). Cryo-EM structure of gas vesicles for buoyancy-controlled motility. Cell 186, 975–986.e13. 10.1016/j.cell.2023.01.041. [PubMed: 36868215] 25. Peukes J, Xiong X, Erlendsson S, Qu K, Wan W, Calder LJ, Schraidt O, Kummer S, Freund SMV, Kräusslich HG, and Briggs JAG (2020). The native structure of the assembled matrix protein 1 of influenza A virus. Nature 587, 495–498. 10.1038/s41586-020-2696-8. [PubMed: 32908308] 26. Metskas LA, Ortega D, Oltrogge LM, Blikstad C, Laughlin T, Savage DF, and Jensen GJ (2022). Rubisco forms a lattice inside alpha-carboxysomes. Nat. Commun 13, 4863. 10.1038/ s41467-022-32584-7. [PubMed: 35982043] Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 18 27. Walsby AE, and Bleything A (1988). The dimensions of cyanobacterial gas vesicles in relation to their efficiency in providing buoyancy and with-standing pressure. Microbiology 134, 2635–2645. 10.1099/00221287-134-10-2635. 28. Waaland JR, and Branton D (1969). Gas vacuole development in a blue-green alga. Science 163, 1339–1341. 10.1126/science.163.3873.1339. [PubMed: 17807814] 29. Wan W, Clarke M, Norris MJ, Kolesnikova L, Koehler A, Bornholdt ZA, Becker S, Saphire EO, and Briggs JA (2020). Ebola and Marburg virus matrix layers are locally ordered assemblies of VP40 dimers. Elife 9, e59225. 10.7554/eLife.59225. [PubMed: 33016878] 30. Buchholz BE, Hayes PK, and Walsby AE (1993). The distribution of the outer gas vesicle protein, GvpC, on the Anabaena gas vesicle, and its ratio to GvpA. J. Gen. Microbiol 139, 2353–2363. 10.1099/00221287-139-10-2353. [PubMed: 8254305] 31. Maresca D, Lakshmanan A, Abedi M, Bar-Zion A, Farhadi A, Lu GJ, Szablowski JO, Wu D, Yoo S, and Shapiro MG (2018). Biomolecular ultrasound and sonogenetics. Annu. Rev. Chem. Biomol. Eng 9, 229–252. 10.1146/annurev-chembioeng-060817-084034. [PubMed: 29579400] 32. Ashkenazy H, Abadi S, Martz E, Chay O, Mayrose I, Pupko T, and Ben-Tal N (2016). ConSurf 2016: an improved methodology to estimate and visualize evolutionary conservation in macromolecules. Nucleic Acids Res 44, W344–W350. 10.1093/nar/gkw408. [PubMed: 27166375] 33. Englert C, Horne M, and Pfeifer F (1990). Expression of the major gas vesicle protein gene in the halophilic archaebacteriumHaloferax mediterranei is modulated by salt. Mol. Gen. Genet 222, 225–232. 10.1007/BF00633822. [PubMed: 1703266] 34. Griffiths AE, Walsby AE, and Hayes PK (1992). The homologies of gas vesicle proteins. J. Gen. Microbiol 138, 1243–1250. 10.1099/00221287-138-6-1243. [PubMed: 1527496] 35. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, Tunyasuvunakool K, Bates R, Žídek A, Potapenko A, et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. 10.1038/s41586-021-03819-2. [PubMed: 34265844] 36. Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S, and Steinegger M (2022). ColabFold: making protein folding accessible to all. Nat. Methods 19, 679–682. 10.1038/s41592-022-01488-1. [PubMed: 35637307] 37. Merkley ED, Rysavy S, Kahraman A, Hafen RP, Daggett V, and Adkins JN (2014). Distance restraints from crosslinking mass spectrometry: mining a molecular dynamics simulation database to evaluate lysine-lysine distances. Protein Sci 23, 747–759. 10.1002/pro.2458. [PubMed: 24639379] 38. Maresca D, Lakshmanan A, Lee-Gosselin A, Melis JM, Ni Y-L, Bourdeau RW, Kochmann DM, and Shapiro MG (2017). Nonlinear ultrasound imaging of nanoscale acoustic biomolecules. Appl. Phys. Lett 110, 073704. 10.1063/1.4976105. [PubMed: 28289314] 39. Huang Y-H, and Huang C-Y (2018). The glycine-rich flexible region in SSB is crucial for PriA stimulation. RSC Adv 8, 35280–35288. 10.1039/c8ra07306f. [PubMed: 35547063] 40. Liberta F, Loerch S, Rennegarbe M, Schierhorn A, Westermark P, Westermark GT, Hazenberg BPC, Grigorieff N, Fändrich M, and Schmidt M (2019). Cryo-EM fibril structures from systemic AA amyloidosis reveal the species complementarity of pathological amyloids. Nat. Commun 10, 1104. 10.1038/s41467-019-09033-z. [PubMed: 30846696] 41. Berhanu WM, Alred EJ, Bernhardt NA, and Hansmann UH (2015). All-atom simulation of amyloid aggregates. Phys. Procedia 68, 61–68. 10.1016/j.phpro.2015.07.110. 42. Paul TJ, Hoffmann Z, Wang C, Shanmugasundaram M, DeJoannis J, Shekhtman A, Lednev IK, Yadavalli VK, and Prabhakar R (2016). Structural and mechanical properties of amyloid beta fibrils: a combined experimental and theoretical approach. J. Phys. Chem. Lett 7, 2758–2764. 10.1021/acs.jpclett.6b01066. [PubMed: 27387853] 43. Walker JE, and Walsby AE (1983). Molecular weight of gas-vesicle protein from the planktonic cyanobacterium Anabaena flos-aquae and implications for structure of the vesicle. Biochem. J 209, 809–815. 10.1042/bj2090809. [PubMed: 6409075] 44. Belenky M, Meyers R, and Herzfeld J (2004). Subunit structure of gas vesicles: a MALDI-TOF mass spectrometry study. Biophys. J 86, 499–505. [PubMed: 14695294] Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 19 45. Knitsch R, Schneefeld M, Weitzel K, and Pfeifer F (2017). Mutations in the major gas vesicle protein GvpA and impacts on gas vesicle formation in Haloferax volcanii. Mol. Microbiol 106, 530–542. 10.1111/mmi.13833. [PubMed: 28898511] 46. Beard SJ, Handley BA, Hayes PK, and Walsby AE (1999). The diversity of gas vesicle genes in Planktothrix rubescens from Lake Zürich. Microbiology 145, 2757–2768. 10.1099/00221287-145-10-2757. [PubMed: 10537197] 47. Beard SJ, Davis PA, Iglesias-Rodrı Guez D, Skulberg OM, and Walsby AE (2000). Gas vesicle genes in Planktothrix spp. from Nordic lakes: strains with weak gas vesicles possess a longer variant of gvpC. Microbiology 146 (Pt 8), 2009–2018. 10.1099/00221287-146-8-2009. [PubMed: 10931905] 48. Salahshoor H, Yao Y, Dutka P, Nyström NN, Jin Z, Min E, Malounda D, Jensen GJ, Ortiz M, and Shapiro MG (2022). Geometric effects in gas vesicle buckling under ultrasound. Biophys. J 121, 4221–4228. 10.1016/j.bpj.2022.09.004. [PubMed: 36081347] 49. Walsby AE, and Hayes PK (1988). The minor cyanobacterial gas vesicle protein, GVPc, is attached to the outer surface of the gas vesicle. Microbiology 134, 2647–2657. 10.1099/00221287-134-10-2647. 50. Jost A, and Pfeifer F (2022). Interaction of the gas vesicle proteins GvpA, GvpC, GvpN, and GvpO of Halobacterium salinarum. Front. Microbiol 13, 971917. 10.3389/fmicb.2022.971917. [PubMed: 35966690] 51. Egelman EH (2015). Three-dimensional reconstruction of helical polymers. Arch. Biochem. Biophys 581, 54–58. 10.1016/j.abb.2015.04.004. [PubMed: 25912526] 52. He S, and Scheres SHW (2017). Helical reconstruction in RELION. J. Struct. Biol 198, 163–176. 10.1016/j.jsb.2017.02.003. [PubMed: 28193500] 53. Tegunov D, Xue L, Dienemann C, Cramer P, and Mahamid J (2021). Multi-particle cryo-EM refinement with M visualizes ribosome-antibiotic complex at 3.5 Å in cells. Nat. Methods 18, 186–193. 10.1038/s41592-020-01054-7. [PubMed: 33542511] 54. Schur FKM, Obr M, Hagen WJH, Wan W, Jakobi AJ, Kirkpatrick JM, Sachse C, Kräusslich HG, and Briggs JAG (2016). An atomic model of HIV-1 capsid-SP1 reveals structures regulating assembly and maturation. Science 353, 506–508. 10.1126/science.aaf9620. [PubMed: 27417497] 55. Mastronarde DN (2005). Automated electron microscope tomography using robust prediction of specimen movements. J. Struct. Biol 152, 36–51. 10.1016/j.jsb.2005.07.007. [PubMed: 16182563] 56. Tegunov D, and Cramer P (2019). Real-time cryo-electron microscopy data preprocessing with Warp. Nat. Methods 16, 1146–1152. 10.1038/s41592-019-0580-y. [PubMed: 31591575] 57. Castaño-Díez D, Kudryashev M, Arheit M, and Stahlberg H (2012). Dynamo: a flexible, user- friendly development tool for subtomogram averaging of cryo-EM data in high-performance computing environments. J. Struct. Biol 178, 139–151. 10.1016/j.jsb.2011.12.017. [PubMed: 22245546] 58. Zivanov J, Nakane T, Forsberg BO, Kimanius D, Hagen WJ, Lindahl E, and Scheres SH (2018). New tools for automated high-resolution cryo-EM structure determination in RELION-3. Elife 7. 10.7554/eLife.42166. 59. Kremer JR, Mastronarde DN, and McIntosh JR (1996). Computer visualization of three- dimensional image data using IMOD. J. Struct. Biol 116, 71–76. 10.1006/jsbi.1996.0013. [PubMed: 8742726] 60. Pettersen EF, Goddard TD, Huang CC, Couch GS, Greenblatt DM, Meng EC, and Ferrin TE (2004). UCSF Chimera–a visualization system for exploratory research and analysis. J. Comput. Chem 25, 1605–1612. 10.1002/jcc.20084. [PubMed: 15264254] 61. Goddard TD, Huang CC, Meng EC, Pettersen EF, Couch GS, Morris JH, and Ferrin TE (2018). UCSF ChimeraX: meeting modern challenges in visualization and analysis. Protein Sci 27, 14–25. 10.1002/pro.3235. [PubMed: 28710774] 62. Adams PD, Afonine PV, Bunkóczi G, Chen VB, Davis IW, Echols N, Headd JJ, Hung L-W, Kapral GJ, Grosse-Kunstleve RW, et al. (2010). PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D Biol. Crystallogr 66, 213–221. 10.1107/ S0907444909052925. [PubMed: 20124702] Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 20 63. Emsley P, Lohkamp B, Scott WG, and Cowtan K (2010). Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr 66, 486–501. 10.1107/S0907444910007493. [PubMed: 20383002] 64. Tan YZ, Baldwin PR, Davis JH, Williamson JR, Potter CS, Carragher B, and Lyumkis D (2017). Addressing preferred specimen orientation in single-particle cryo-EM through tilting. Nat. Methods 14, 793–796. 10.1038/nmeth.4347. [PubMed: 28671674] 65. Burt A, Gaifas L, Dendooven T, and Gutsche I (2021). A flexible framework for multi-particle refinement in cryo-electron tomography. PLoS Biol 19, e3001319. 10.1371/journal.pbio.3001319. [PubMed: 34437530] 66. Mastronarde DN, and Held SR (2017). Automated tilt series alignment and tomographic reconstruction in IMOD. J. Struct. Biol 197, 102–113. 10.1016/j.jsb.2016.07.011. [PubMed: 27444392] 67. Castaño-Díez D, Kudryashev M, and Stahlberg H (2017). Dynamo Catalogue: geometrical tools and data management for particle picking in subtomogram averaging of cryo-electron tomograms. J. Struct. Biol 197, 135–144. 10.1016/j.jsb.2016.06.005. [PubMed: 27288866] 68. Deutsch B, and Sharma. (2020). The ProteomeXchange consortium in 2020: enabling “big data”approaches in proteomics. Nucleic Acids Mol. Biol 48, D1145–D1152. 10.1093/nar/gkz984. 69. Perez-Riverol Y, Csordas A, Bai J, Bernal-Llinares M, Hewapathirana S, Kundu DJ, Inuganti A, Griss J, Mayer G, Eisenacher M, et al. (2019). The PRIDE database and related tools and resources in 2019: improving support for quantification data. Nucleic Acids Res 47, D442–D450. 10.1093/nar/gky1106. [PubMed: 30395289] 70. Ravikumar A, Arzumanyan GA, Obadi MKA, Javanpour AA, and Liu CC (2018). Scalable, continuous evolution of genes at mutation rates above genomic error thresholds. Cell 175, 1946– 1957.e13. 10.1016/j.cell.2018.10.021. [PubMed: 30415839] 71. Ammar EM, Wang X, and Rao CV (2018). Regulation of metabolism in Escherichia coli during growth on mixtures of the non-glucose sugars: arabinose, lactose, and xylose. Sci. Rep 8, 609. 10.1038/s41598-017-18704-0. [PubMed: 29330542] 72. Wittmann BJ, Johnston KE, Almhjell PJ, and Arnold FH (2022). evSeq: cost-effective amplicon sequencing of every variant in a protein library. ACS Synth. Biol 11, 1313–1324. 10.1021/acssyn- bio.1c00592. [PubMed: 35172576] 73. Gosline J, Lillie M, Carrington E, Guerette P, Ortlepp C, and Savage K (2002). Elastic proteins: biological roles and mechanical properties. Philos. Trans. R. Soc. Lond. B Biol. Sci 357, 121–132. 10.1098/rstb.2001.1022. [PubMed: 11911769] 74. Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, Li W, Lopez R, McWilliam H, Remmert M, Sö ding, J., et al. (2011). Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol. Syst. Biol 7, 539. [PubMed: 21988835] 75. Waterhouse AM, Procter JB, Martin DMA, Clamp M, and Barton GJ (2009). Jalview Version 2—a multiple sequence alignment editor and analysis workbench. Bioinformatics 25, 1189–1191. 10.1093/bioinformatics/btp033. [PubMed: 19151095] 76. Combe CW, Fischer L, and Rappsilber J (2015). xiNET: cross-link network maps with residue resolution. Mol. Cell. Proteomics 14, 1137–1147. 10.1074/mcp.O114.042259. [PubMed: 25648531] Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 21 Highlights • • • • Gas vesicles (GVs) are formed by ~3-nm corrugated protein shells Corrugation reverses at the cylinder midpoint, which may act as an elongation center The protein shell is primarily formed by the conserved major structural protein GvpA GvpC provides extra reinforcement by forming a helical spiral around the GV cylinder Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 22 Figure 1. Molecular architecture of Ana GVs (A) Schematic of an Ana GV with dimensions annotated. (B) Representative slices at the indicated z heights from cryo-ET of an individual GV. Inset: enlargement of the area indicated by the black dashed box. Scale bars, 50 nm. (C) Central tomography slices of two conical GV ends with different morphologies. Scale bars, 50 nm. (D) Enlarged views of the areas indicated by orange (apex) and blue (cone to cylinder transition) dashed boxes in (C). Scale bars, 10 nm. (E) Distribution of the diameters and heights of conical GV ends; n = 132. The orange dashed line indicates the average height of the cones (59 ± 6 nm). Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 23 Figure 2. Polarity inversion point (A) Enlargement of the tomographic slices from Figure 1B (indicated by the orange dashed box) at different z heights. The blue dashed outlines indicate sections where polarity changes. Scale bars, 50 nm. (B) Subtomogram average of the middle region of the GV where the ribs reverse polarity. Arrows denote the rib where polarity is reversed. (C) Enlarged view of the subtomogram average in (B), highlighting the inversion of the helical assembly. Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 24 Figure 3. Cryo-ET structure of the Ana GV shell (A) Initial, low-resolution subtomogram average of a cylindrical GV segment. (B) Orthogonal views of a higher-resolution (7.7 Å) subtomogram average of the native Ana GV shell. (C–E) Cross-sections of the subtomogram averages of the GV shell: (C) native Ana GV, (D) AnaS GV, and (E) superimposed. (F and G) Projections trough the subtomogram average of the native Ana GV. (F) Projection along the GV helical axis. In the right panel color-coded densities corresponding to GvpA and GvpC. (G) Projection trough the neighboring subunits forming GV helix. Bottom: color-coded densities corresponding to GvpA and GvpC. Scale bars, 2 nm. (H) Segmented density map of the native Ana GV, indicating the locations of GvpC. Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 25 Figure 4. Conserved assembly of the GV shell (A) Segmented ~8-Å resolution structure of two adjacent GvpA ribs determined by subtomogram averaging (gray surface), fitted with a homology model of GvpA based on GvpA2 (PDB: 7R1C).24 (B) Domain annotation within an individual GvpA. (C) Conservation analysis of GvpA determined by ConSurf.32 (D and E) Negative-stain EM images of collapsed GVs from (D) Ana and (E) Halo. Arrows indicate separated GvpA filaments. Collapse pressure (CP) is indicated above. Scale bars, 50 nm. (F) Location of tolerated mutation sites (yellow spheres) in the GvpA structure (blue). (G) Map of all tolerated mutations in GvpA. Original sequence colored by conservation score as in (C). Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 26 Figure 5. Mechanical reinforcement of the GV shell by GvpC (A) Segmented ~8-Å resolution subtomogram average of neighboring Ana GvpA monomers connected by GvpC (gray surface) fitted with a model of GvpA and a poly-Ala chain corresponding in length to one repeating sequence of GvpC. (B) Resulting GvpC binding model. (C) GvpC binding site (dashed black box) at the hydrophobic pockets between α2 helices of GvpA. The surface of GvpA is colored by hydrophobicity. (D) Cross-linked sites between GvpA and GvpC identified by mass spectrometry. (E) Finite element shell models of a GV with a length of 500 nm and width of 85 nm and the indicated degree of GvpC saturation. (F) Buckling pressure as a function of GvpC density. The orange line represents a simple linear regression fit. Structure. Author manuscript; available in PMC 2023 May 15. Dutka et al. Page 27 A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t KEY RESOURCES TABLE REAGENT or RESOURCE SOURCE IDENTIFIER Bacterial and virus strains E. coli Rosetta 2(DE3)pLysS Millipore Sigma Cat# 71401–3 E. coli BL21-AI E. coli NEB Stable Thermo Fisher Scientific Cat# C607003 New England Biolabs Cat# C3040H Dolichospermum flos-aquae strain 1403/13F SAMS LIMITED CCAP Cat# 1403/13F Halobacterium sp. NRC-1, Living, Plate Carolina Biological Supply Cat# 154801 Chemicals, peptides, and recombinant proteins Q5 High-Fidelity DNA Polymerase New England Biolabs Cat# M0491L NEBuilder HiFi DNA Assembly Master Mix New England Biolabs Cat# E2621L LB Broth (Lennox) Kanamycin sulfate D-(+)-Glucose Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich SoluLyse™ Bacterial Protein Extraction Reagent Genlantis Cyanobacteria BG-11 Freshwater Solution Sigma-Aldrich 20x PBS, Sterile Teknova Lysozyme from chicken egg white Sigma-Aldrich IPTG, Isopropyl β D 1 thiogalactopyranoside Teknova Chloramphenicol Ampicillin sodium salt Tris-HCl 1 M, pH 7.5 Sodium chloride (NaCl) Magnesium sulfate heptahydrate Casein hydrolysate Gibco™ Bacto™ Yeast Extract Sodium nitrate Potassium phosphate monobasic Sodium metasilicate nonahydrate Sodium carbonate Sodium bicarbonate Citric acid EDTA, disodium salt dihydrate Ferric ammonium citrate Urea Trisodium citrate dihydrate Potassium chloride D-Sorbitol Formic Acid, LC/MS Grade Acetonitrile, LC/MS Grade Sigma-Aldrich Sigma-Aldrich Teknova Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich Gibco Amresco Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich BDH Chemicals Alfa Aesar Amresco Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich Sigma-Aldrich Fisher Scientific Fisher Scientific Cat# L3022–1KG Cat# K1377–25G Cat# G7021–5KG Cat# L200500 Cat# C3061 Cat# P1225 Cat# L6876 Cat# I3305 Cat# C0378 Cat# A0166 Cat# T1075 Cat# S7653 Cat# 230391 Cat# 22090 Cat# 212750 Cat# 0598 Cat# P5655 Cat# S4392 Cat# 71345 Cat# S5761 Cat# BDH7397 Cat# A15161 Cat# 0846 Cat# 51456 Cat# S1804 Cat# 746436 Cat# S6021 Cat# A11750 Cat# A9554 Structure. Author manuscript; available in PMC 2023 May 15. Dutka et al. Page 28 REAGENT or RESOURCE SOURCE IDENTIFIER Water, LC/MS Grade Trypsin, TPCK Treated DSSO, crosslinker BS3, crosslinker Deposited data Fisher Scientific Cat# W64 ThermoFisher Scientific Cat# 20233 ThermoFisher Scientific Cat# A33545 ThermoFisher Scientific Cat# A39266 Cryo-electron tomogram for Ana GV (Figures 1 and 2A) This study Cryo-electron tomogram Mega GV (Figure S2A) This study Cryo-electron tomogram Halo GV, p-vac (Figure S9B) This study Cryo-electron tomogram Halo GV, c-vac (Figure S9C) This study Subtomogram average of the native Ana GV shell This study Subtomogram average of AnaS GV shell Integrative model of Ana GvpA/GvpC XLMS data Atomic model of GvpA2 Oligonucleotides This study This study This study Huber et al.24 EMD-29922 EMD-29925 EMD-29924 EMD-29923 EMD-29921 EMD-29916 PDB 8GBS PXD038631 PDB 7R1C Primers for Gibson assembly Integrated DNA Technologies N/A Mutagenic oligo pool Twist Bioscience N/A Recombinant DNA pST39 plasmid containing pNL29 Mega GV gene cluster Addgene Cat# 91696 Software and algorithms SnapGene MATLAB Abaqus SerialEM Warp Dynamo RELION M IMOD UCSF Chimera UCSF ChimeraX Phenix Coot 3DFSC SnapGene Mathworks Dassault Systmes https://www.snapgene.com/ https://matlab.mathworks.com/ https://www.3ds.com/products-services/simulia/products/ abaqus/ Mastronarde55 https://bio3d.colorado.edu/SerialEM/ Tegunov and Cramer56 http://www.warpem.com/warp/ Castaño-Díez et al.57 https://wiki.dynamo.biozentrum.unibas.ch/w/index.php/ Main_Page Zivanov et al.58 Tegunov et al.53 Kremer et al.59 Pettersen et al.60 Goddard et al.61 Adams et al.62 Emsley et al.63 Tan et al.64 https://relion.readthedocs.io/en/release-3.1/ http://www.warpem.com/warp/ https://bio3d.colorado.edu/imod/ https://www.cgl.ucsf.edu/chimera/index.html https://www.rbvi.ucsf.edu/chimerax/ https://phenix-online.org/ https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/ https://github.com/nysbc/Anisotropy Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Dutka et al. Page 29 REAGENT or RESOURCE dynamo2m autoalign_dynamo Xcalibur Proteome Discoverer SOURCE Burt el al.65 Burt el al.65 IDENTIFIER https://github.com/alisterburt/dynamo2m https://github.com/alisterburt/autoalign_dynamo ThermoFisher Scientific https://www.thermofisher.com/us/en/home.html ThermoFisher Scientific https://www.thermofisher.com/us/en/home.html XlinkX node for Proteome Discoverer ThermoFisher Scientific https://www.thermofisher.com/us/en/home.html Other C-flat Holey Carbon R2/2 300 Mesh Formvar/Carbon 200 mesh, Copper EMS Ted Pella Cat# CF-223C-100 Cat# 01801 Vitrobot Mark IV FEI, Thermo Fisher Scientific Gatan K3 summit camera with post-column energy filter Gatan 300 kV Titan Krios microscope FEI, Thermo Fisher Scientific https://www.thermofisher.com/us/en/home/electron- microscopy/products/sample-preparation-equipment-em/ vitrobot-system.html https://www.gatan.com/products/tem-imaging- spectroscopy/k3-cameras https://www.thermofisher.com/us/en/home/ electron-microscopy/products/transmissionelectron- microscopes.html Tecnai T12 120 kV FEI, Thermo Fisher Scientific Out-of-production EconoSpin 96-Well Filter Plate Epoch Life Sciences Cat# 2020–001 Shaker incubator with cooling, CO2 control and illumination Infors HT Multitron II Benchtop microcentrifuge Eppendorf Cat# 5430R UltraCruz Filter Flask, PES, 0.22 Santa Cruz Biotechnology Cat# sc-359029 1000 mL Filter System, PES Filter Material, 0.22 μm EMD Millipore™ Steritop Threaded Bottle Top Filter CELLTREAT Cat# 229708 EMD Millipore Cat# S2GPT05RE STERICUP-GP500 NanoDrop 2000C series 1-cm-path-length quartz cuvette EMD Millipore Thermo Scientifc Hellma Analytics Cat# S2GPU10RE Cat# ND-2000C Cat# 176.700-QS 18G BLUNT NEEDLE 1.5IN STERILE VWR International Cat# 89134–020 96 well filter plate Square Petri Dish with Grid S-Trap micro spin column Aurora UHPLC Column, 25 cm × 75 μm, 1.6 μm C18 Q Exactive HF hybrid quadrupole-Orbitrap mass spectrometer Bio Basic VWR PROTIFI Ion Opticks SD5006 60872–310 Cat# C02-micro-80 Cat# AUR2–25075C18A-CSI ThermoFisher Scientific N/A Easy-nLC 1200 ThermoFisher Scientific LC140 Structure. Author manuscript; available in PMC 2023 May 15. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1073_pnas.2220528120	RESEARCH ARTICLE \| BIOPHYSICS AND COMPUTATIONAL BIOLOGY OPEN ACCESS PRC2 direct transfer from G-quadruplex RNA to dsDNA has implications for RNA-binding chromatin modifiers Wayne O. Hemphilla,b , and Thomas R. Cecha,b,1 , Anne R. Goodinga,b , Regan Fenskea,b Edited by John Kuriyan, Vanderbilt University, Nashville, TN; received December 2, 2022; accepted May 1, 2023 The chromatin-modifying enzyme, Polycomb Repressive Complex 2 (PRC2), deposits the H3K27me3 epigenetic mark to negatively regulate expression at numerous target genes, and this activity has been implicated in embryonic development, cell differenti- ation, and various cancers. A biological role for RNA binding in regulating PRC2 his- tone methyltransferase activity is generally accepted, but the nature and mechanism of this relationship remains an area of active investigation. Notably, many in vitro studies demonstrate that RNA inhibits PRC2 activity on nucleosomes through mutually antagonistic binding, while some in vivo studies indicate that PRC2’s RNA-binding activity is critical for facilitating its biological function(s). Here we use biochem- ical, biophysical, and computational approaches to interrogate PRC2’s RNA and DNA-binding kinetics. Our findings demonstrate that PRC2-polynucleotide dissoci- ation rates are dependent on the concentration of free ligand, indicating the potential for direct transfer between nucleic acid ligands without a free-enzyme intermediate. Direct transfer explains the variation in previously reported dissociation kinetics, allows reconciliation of prior in vitro and in vivo studies, and expands the potential mechanisms of RNA-mediated PRC2 regulation. Moreover, simulations indicate that such a direct transfer mechanism could be obligatory for RNA to recruit proteins to chromatin. methyltransferase \| polynucleotide \| nucleosomes \| displacement \| exchange PRC2 is a histone methyltransferase (HMTase) that sequentially deposits three methyl groups onto lysine 27 of histone H3 [H3K27me1/2/3; (1–4), reviewed in refs. 5–7], and its activity is crucial for epigenetic silencing during development and cancer (5). How PRC2 is targeted to genetic loci is of considerable interest, given its critical function and abundance of target genes (8). PRC2’s core subunits include the Enhancer of Zeste Homolog 2 (EZH2) catalytic domain, Suppressor of Zeste 12 (SUZ12) scaffold subunit, Embryonic Ectoderm Development (EED) histone tail-binding subunit, and Retinoblastoma-Binding Protein 4 (RBBP4) histone chaperone subunit, and it has addi- tional accessory subunits that define the PRC2.1 and PRC2.2 subtypes and differentially regulate its activity [(9–11), reviewed in ref. 5]. PRC2 binds numerous long noncoding RNAs (lncRNAs) and pre-mRNAs in cell nuclei, and this RNA binding is believed to regulate PRC2’s HMTase activity (12–15). Furthermore, biochemical studies have demon- strated that PRC2 has specificity for G-tracts and G-quadruplex (G4) RNA structures (16), which are ubiquitous in the human transcriptome, consistent with its widespread RNA binding in cells. The nature and mechanism(s) of PRC2 regulation by RNA remain quite controversial. While some studies have proposed a role for RNA in PRC2 recruitment to chromatin (13, 17), others have suggested roles in PRC2 eviction from chromatin and/or inhibition of PRC2 catalytic activity (18–22), and these ideas are not mutually exclusive. Biochemical experiments have convincingly demonstrated that RNA antagonizes PRC2 HMTase activity (18, 20, 22), and that this is mediated by competitive binding with nucleosomes (19, 20). On the other hand, a recent work has demonstrated that the PRC2-RNA inter- action is critical in vivo for maintaining H3K27me3 levels and chromatin occupancy at PRC2 target genes in induced pluripotent stem cells (23). It is prudent to note that the biochemical studies have utilized RNA and nucleosomes in free solution, which is not representative of the chromatin-associated nascent RNA suspected to regulate PRC2 activity in vivo (18, 19, 21, 22). Furthermore, the role(s) of RNA in PRC2 activity could be contextual to chromatin architecture, available PRC2 accessory subunits and protein partners, competing RNA-binding proteins, and/or post-translational modifications. Thus, prior biochemical studies may lack considerations relevant to in vivo function. The direct evidence for a mechanism that can reconcile RNA antagonizing PRC2’s nucleosome binding and HMTase activity in vitro with RNA-mediated PRC2 recruitment in vivo has yet to be reported. Significance Studies of PRC2 in vitro indicate that RNA inhibits its histone methyltransferase (HMTase) activity through mutually antagonistic binding with nucleosomes, but some in vivo studies paradoxically suggest that RNA binding is necessary to facilitate its chromatin occupancy and HMTase activity. Our findings unveil a mechanism for direct exchange of RNA and DNA/ nucleosome on the PRC2 protein complex, which reconciles these prior findings by allowing RNA regulation of PRC2 to be antagonistic or synergistic depending on RNA–nucleosome proximity. Furthermore, there is an increasing awareness that multiple chromatin-associated proteins exhibit regulatory RNA binding activity, and our findings indicate that this “direct transfer” mechanism may be generally required for RNA recruitment of proteins to chromatin. aDepartment of Biochemistry, Author affiliations: BioFrontiers Institute, University of Colorado Boulder, Boulder, CO 80309; and bHHMI, University of Colorado Boulder, Boulder, CO 80309 Author contributions: W.O.H. and T.R.C. designed research; W.O.H. and R.F. performed research; W.O.H. and A.R.G. contributed new reagents/analytic tools; W.O.H. and R.F. analyzed data; and W.O.H. and T.R.C. wrote the paper. interest statement: T.R.C. Competing is a scientific advisor for Storm Therapeutics, Eikon Therapeutics, and SomaLogic. The other authors declare no competing interest. This article is a PNAS Direct Submission. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2220528120/-/DCSupplemental. Published May 30, 2023. PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 1 of 11 Herein, we measure the kinetics of human PRC2’s RNA and DNA binding using biochemical, biophysical, and computational methods. Our findings unexpectedly reveal that PRC2 has the intrinsic ability to exchange one nucleic acid for another without completely dissociating from the first nucleic acid. Such mechanisms have been well-studied for homo-multimeric DNA-binding proteins like lac repressor (24), E. coli catabolite activator protein (CAP) (25), SSB (26), and recA (27) and for the hexameric RNA-binding protein Hfq (28). Historically, this phenomenon has been variously identi- fied as “concentration-dependent dissociation,” “direct trans- fer” (25–27), “facilitated exchange” (29), or “active exchange” (28), and it is related to the sister phenomena of protein movement along DNA (30–34) and “facilitated dissociation” (35, 36). The proteins in these cases are homo-oligomeric, and as others have noted (37), their multiple ligand-binding sites likely facilitate direct transfer by providing a foothold for a second ligand before it displaces a previ- ously bound ligand. We propose that PRC2’s ability to directly transfer from one nucleic acid to another may reconcile the disparate eviction versus recruitment models of previous studies. Furthermore, binding to nascent RNA has been suggested as a general strategy by which tran- scription factors and other DNA-binding proteins are maintained at high local concentrations for recruitment to target genes. Our find- ings indicate that this model may be feasible only if the protein can directly transfer from RNA to DNA without dissociation, suggesting direct transfer capabilities may have general relevance. Results PRC2 Exhibits Direct Transfer Between G4 RNA and dsDNA. Two prior studies from our group (19, 23) determined the dissociation rate constant for a G4 RNA species from PRC2, but they obtained significantly different values despite nearly identical methodologies. One of the few methodological distinctions between these two studies was the concentration of unlabeled competitor RNA used to prevent the rebinding of labeled RNA once it dissociated from PRC2. Long et al. and Wang et al. used a 200- and 2,000-fold excess of competitor RNA over RNA ligand, respectively, both of which should have been sufficient to totally prevent ligand rebinding. We used fluorescence polarization (FP)-based competitive dissociation (FPCD) experiments (Fig. 1) to replicate the Wang/Long studies across a range of competitor RNA concentrations. Unexpectedly, obs) of PRC2 and G4 RNA did the observed dissociation rate (koff not plateau at excess concentrations of competitor, but it instead continued to increase linearly in a competitor concentration- dependent manner (Fig. 2A and SI Appendix, Fig. S1). This result is consistent with the incoming competitor being able to displace the initially bound ligand without free PRC2 as an intermediate, i.e., direct transfer of PRC2 between ligand and competitor (26). Our data are consistent with both the Wang et al. and Long et al. findings given the respective competitor concentrations they used. Since our initial experiments utilized an RNA with 10 G-tracts that could form G4s heterogeneously, we tested a simpler RNA Fig. 1. Experiment and analysis strategy to measure direct transfer kinetics (FPCD Experiments). (1) The minimum amount of PRC2 required for saturated binding is mixed with a trace amount of fluorescently labeled nucleic acid (ligand), then incubated (at 4/25 °C) until thermal and reaction equilibrium. (2) Various concentrations of unlabeled nucleic acid (competitor) are added to the preformed complex to initiate reactions (at 25 °C). (3) The time-course reactions are immediately monitored by fluorescence polarization in a microplate reader (at 25 °C). Potential complexes with their polarization states are shown, and they are labeled with rate constants describing inter-complex transitions. Rate constants associated with a classic competition model are indicated by green boxes, and those additionally necessary for a direct transfer model are indicated by a purple box. The intercomplex transition solely associated with the direct transfer model has an implied unstable ternary complex intermediate. The system of differential equations describing these reactions is given by SI Appendix, Eq. S1. (4) Polarization signals are normalized to the range in polarization signal across all competitor concentrations to give proportion of initial complex remaining. Normalized polarization signals are plotted versus time and fit with one-phase exponential decay regression (SI Appendix, Eq. S3.1). (5) The regression initial slopes obs; SI Appendix, Eq. S3.2) are plotted versus competitor concentration and regressed with custom equations describing the classic competition (SI Appendix, (koff Eq. S4.2) and direct transfer (SI Appendix, Eq. S4.1) models to determine rate constant values. Model fits are compared with the Bayesian Information Criterion (BIC) to determine the appropriate model. 2 of 11 https://doi.org/10.1073/pnas.2220528120 pnas.org Fig. 2. PRC2 exhibits direct transfer kinetics for G4 RNA and dsDNA. FPCD experiments (Fig. 1) were performed with the Wang et al. and Long et al. ligand/ competitor over a range of competitor concentrations (panel A) and to measure direct transfer kinetics for every ligand–competitor combination of a G4 RNA and 60-bp dsDNA (panel B). Data are from representative experiments (of n ≥ 3), where error bars indicate mean ± SD for four technical replicates. Rate constant values from regression can be found in Table 1, additional nomenclature definitions are in SI Appendix, Table S1, and polynucleotide species definitions are in obs, see SI Appendix, Eq. SI Appendix, Table S2. (A) Exponential regression fit lines from each condition (left plot), alongside the observed initial dissociation rates (koff S3.2) as a function of competitor concentration (right plot). Solid line in right plot is a visual aid connecting data means. Raw data are shown in SI Appendix, Fig. S1. (B) Experiments were performed in BB10 buffer. Isotherm, carrier nucleic acid, and fluorophore controls for the RNA-RNA competition experiment (Top Left) can be found in SI Appendix, Fig. S2. Analogous studies with a 50-bp dsDNA can be found in SI Appendix, Fig. S3. sequence containing only four G-tracts and found that it also exhibited direct transfer kinetics (SI Appendix, Fig. S2A). Next, we repeated this experiment at constant room temperature (SI Appendix, Fig. S2B) to interrogate temperature-dependent effects. Then, we repeated the experiment using a carrier poly(A) RNA that does not bind PRC2 to keep total RNA concentration constant in the reactions (SI Appendix, Fig. S2C), so that any non- specific polynucleotide concentration-dependent phenomena (e.g., electrostatic effects) could be ruled out as artifactual explanations. Finally, we repeated the experiment with a different fluorescent label on the ligand molecule (SI Appendix, Fig. S2D) to interrogate interactions with the fluorophore. The data from all experiments were well fit by a regression model allowing direct transfer kinetics but poorly fit by a classic model of competition. We conclude that PRC2 exhibits direct transfer between G4 RNAs. Of particular biological relevance is PRC2’s potential for direct transfer between RNA and chromatin. Prior studies indicate that PRC2 affinity for nucleosomes is entirely mediated by exposed nucleosome linker DNA (19), suggesting comparable-length dsDNA species should be representative of PRC2’s nucleosome PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 3 of 11 binding activity. Thus, we performed FPCD experiments with our simple G4 RNA and a 60-bp dsDNA, using all possible ligand–competitor combinations (Fig. 2B). Notably, our results indicate that direct transfer occurs between all species (Table 1). Experiments with a 50-bp dsDNA species produced qualitatively similar results (SI Appendix, Fig. S3). We also note that prior reports of PRC2 dsDNA and G4 RNA binding affinities (13, 16, 19, 23) are consistent with our corresponding values in Table 1. PRC2 May Have Additional Electrostatic Contacts with dsDNA Not Utilized for G4 RNA. Prior studies indicate that G4 RNA and dsDNA binding to PRC2 are mutually antagonistic (i.e., competitive) (19, 20), which may suggest competition for shared protein-polynucleotide contacts. However, it is not clear to what extent the PRC2 binding surfaces for RNA and DNA may have some unique contacts. If such unique interactions had an electrostatic component, they might be revealed by differential app for salt sensitivity. We therefore used FP to determine Kd G4 RNA and dsDNA at a range of salt concentrations. The experiments demonstrated a much greater influence of ionic strength on PRC2’s binding affinity for dsDNA than for G4 app) versus log([KCl]) RNA (Fig. 3A). Linear regression of log(Kd plots (Fig. 3B) suggests that more salt bridges mediate PRC2 binding to dsDNA (m ≈ 1.4 ± 0.68) versus G4 RNA (m ≈ 0 ± 0.34) (38), which is consistent with the previous conclusion that the PRC2-RNA interaction is not primarily electrostatic (39). It is prudent to note, however, that ionic strength dependence could also be affected by other properties like nucleic acid conformation and divalent ion concentration (38, 40). Thus, a simplistic interpretation of these data suggests that PRC2 has additional ionic contacts with dsDNA that are not utilized during its binding to G4 RNA. Modeling Suggests the PRC2 Direct Transfer Mechanism Allows RNA-Mediated Recruitment to Nucleosomes. Prior studies indicate that RNA inhibits PRC2’s nucleosome DNA binding and HMTase activity (13, 19, 20), while others paradoxically suggest that RNA facilitates PRC2 chromatin occupancy and H3K27me3 deposition (22, 23). To interrogate whether PRC2 direct transfer might reconcile these views, we constructed a reaction scheme of PRC2’s proposed biochemical activity (Fig. 4A). This scheme accounts for classic PRC2 (E) binding to (k1) and dissociation from (k−1) RNA (R) and nucleosome DNA (N), PRC2 mutually antagonistic binding to RNA and nucleosome, the catalytic deficiency of RNA-bound PRC2 (19, 20), and PRC2 catalytic (kcat) methylation of nucleosomes (Nm). In addition, the model includes the direct transfer reactions (kθ) demonstrated by our present studies (Fig. 2). To account for the in vivo proximity between nascent RNA and chromatin, we incorporate a simple Table 1. Rate constants for a variety of protein–ligand interactions Ligand r(GGAA)10[A488] r(G3A2)4[F] r(G3A2)4[A488] ds-[F]d(N)60 ds-d(N)50[F] Buffer BB25 BB10 BB25 BB100 BB200 BB25 BB10 BB25 BB100 BB200 BB10 BB25 T (°C) 25 4 to 25 25 4 to 25 25 4 to 25 25 25 25 4 to 25 25 app (nM) KdP 78 ± 12 n.d. ‡2.3 ± 0.35 n.d. 4.4 ± 0.34 n.d. ‡1.4 ± 0.15 6.0 ± 0.93 8.2 ± 0.69 n.d. 5.1 ± 0.60 4 to 25 n.d. 25 25 25 25 82 ± 13 170 ± 15 #n/a 5.0 ± 0.46 4 to 25 25 n.d. 390 ± 31 Competitor r(GGAA)10 r(GGAA)10 ds-d(N)50 r(G3A2)4 ds-d(N)60 r(G3A2)4 r(G3A2)4 §r(G3A2)4 \| C1 – – – r(G3A2)4 r(G3A2)4 ds-d(N)60 r(G3A2)4 ds-d(N)60 – – – r(G3A2)4 ds-d(N)50 ds-d(N)50 – k−1P (s−1) ,†n.d. 5.2 ± 0.5 (×10−4) \|\|1.4 ± 1.1 (×10−3) 8.3 ± 1.6 (×10−4) \|\|4.5 ± 0.15 (×10−4) 1.7 ± 0.60 (×10−3) 5.6 ± 0.50 (×10−4) ¶4.7 × 10−4 – – – 8.8 ± 0.32 (×10−4) 2.4 ± 0.25 (×10−3) ¶9.1 × 10−5 1.2 ± 0.062 (×10−3) 5.3 ± 2.3 (×10−4) – – – 2.5 ± 0.11 (×10−3) 7.6 ± 0.75 (×10−4) ¶2.8 × 10−4 – kθD (M−1s−1) ,†n.d. 79 ± 6.0 \|\|91 ± 31 30 ± 13 \|\|100 ± 19 660 ± 130 47 ± 17 ¶73 – – – 59 ± 20 170 ± 41 ¶260 67 ± 8.8 150 ± 21 – – – 170 ± 8.2 210 ± 91 ¶340 – app; it’s possible that Kd < Kd Experiments used PRC25m (somatic AEBP2 isoform), not PRC25me (embryonic AEBP2 isoform). †Dissociation completed during initiation-measurement delay (~90 s; λ ≥ 3.3 × 10−2 s−1). ‡Experiment used [Ligand] ≥ 2× KdP §Total polynucleotide concentration was kept constant by serially diluting competitor in a carrier nucleic acid; C1 = r(A)20. ¶Value from single experiment. #Binding too weak to obtain Kd \|\|Weak competitor—manual baseline (from binding curve) used for regression calculations. app) (in the absence of com- Fluorescence polarization-based methodology (Fig. 1 and Materials and Methods) was used to determine the apparent equilibrium dissociation constants (KdP app are from regression petitor), intrinsic dissociation rate constants (k−1P), and direct transfer rate constants (kθD). Values indicate mean ± SD for at least three independent experiments. Kd with a standard (non-quadratic, non-Hill) binding equation (SI Appendix, Eq. S2). Numerical subscript of buffers refers to their variable concentration of salt, and specific buffer definitions can be found in Materials and Methods. Additional nomenclature definition is provided in SI Appendix, Table S1, and polynucleotide sequences are defined in SI Appendix, Table S2. n.d. = not determined; n/a = not applicable. app (> 1 µM). app. 4 of 11 https://doi.org/10.1073/pnas.2220528120 pnas.org Fig. 3. Ionic interactions contribute to PRC2’s dsDNA but not G4 RNA affinities. FP-based equilibrium binding experiments (Materials and Methods) were carried out under various salt concentrations (BBX = X mM KCl) for a G4 RNA and 60-bp dsDNA ligand (no competitor present). Kinetic constant values from regression can be found in Table 1, additional nomenclature definitions are in SI Appendix, Table S1, and polynucleotide species definitions are in SI Appendix, Table S2. (A) Binding curves for indicated PRC2 ligands. Curves are composites of three experiments with four replicates each, where error bars indicate mean ± SD. Solid lines are visual aids connecting the data points. (B) Affinity versus ionic strength plots with app values from regression of data in panel A. Data are Kd composites of all experiments in panel A, where error bars indicate mean ± SD. Solid lines are from linear regression of data on the logarithmic axes shown. Regression values can be found in Materials and Methods or the corresponding text. tuning parameter (α) for the effective molarity between RNA and nucleosome in direct transfer reactions. We note that while effective molarity for these reactions should increase when RNA– nucleosome proximity is increased by tethering (e.g., nascent RNA), the degree of increase would have a complex relationship with other factors such as nascent RNA length, caging effects in condensates, and/or the relative prevalence of free versus nascent RNA. Consequently, the parameter’s effects are best interpreted semi-quantitatively (if not qualitatively), and quantitative conclusions about how RNA length and other factors affect protein activity are outside the scope of our studies. We simulated reactions under this scheme using our empirically determined rate constants for association, dissociation, and direct transfer events (Table 1) and using the previously reported rate constant for EZH2 (PRC2 catalytic subunit) methylation of nucleosomes (41). The results (Fig. 4B and SI Appendix, Fig. S4) indicate that RNA should be antagonistic to PRC2 HMTase activity in free solution (α = 1), as observed experimentally. However, RNA should eventually become synergistic as the RNA–nucleosome effective molarity in direct transfer events increases (e.g., α = 500). As expected, all RNA effects on PRC2 HMTase activity are ablated if the PRC2-RNA complex is unstable (k−1R × 109) (SI Appendix, Fig. S5). Notably, the ability of RNA to boost PRC2 HMTase activity is completely ablated if direct transfer is ablated (kθ = 0) (SI Appendix, Fig. S6). Overall, these data suggest that PRC2’s mutually antagonistic RNA and nucleosome binding could be reconciled with RNA-mediated recruitment of PRC2 to chromatin under some conditions, but only if PRC2 can direct transfer from RNA to nucleosomes. Direct Transfer May Be Generally Required for RNA Recruitment of Chromatin-Associated Proteins. In the case of PRC2 and some other chromatin-associated proteins, RNA and nucleosomal DNA bind mutually antagonistically (i.e., competitively). However, other proteins can stably bind both RNA and chromatin simultaneously. For example, the transcription factor Yin Yang 1 (YY1) binds DNA and RNA independently, and Sigova et al. (42) proposed that its RNA binding keeps YY1 trapped near its DNA binding sites to help recruit it to chromatin DNA. To interrogate this alternative situation of simultaneous binding, we designed a reaction scheme for a hypothetical HMTase enzyme with independent RNA and nucleosome binding activity (Fig. 5A). This scheme accounts for classic protein (E) binding to (k1) and dissociation from (k−1) RNA (R) and nucleosome DNA (N) and the catalytic (kcat) methylation of nucleosomes (Nm). It also utilizes the α tuning parameter for effective molarity (with the same caveats as for Fig. 4A), which in this case applies to ternary complex formation from bimolecular complexes. In addition, the scheme accounts for RNA-mediated suppression of catalysis (β) and potential interplay between nucleosome and RNA binding in the context of ternary complex formation (δ1) and dissociation (δ2). We simulated reactions under this scheme using kinetic con- stants for RNA and DNA binding that are consistent with those reported by Sigova et al. for YY1 (42) and using the same meth- ylation rate constant as for PRC2. As expected, our results (Fig. 5B) support RNA concentration having no effect on activ- ity in free solution (α = 1) when RNA binding is independent of nucleosome binding (δ = 1) and does not affect catalysis (β = 1). In contrast, under the same conditions the simulations show that RNA concentration facilitates catalytic activity as the effec- tive molarity of ternary complex-forming reactions (e.g., due to RNA–nucleosome proximity) increases (α > 1). However, this synergy is easily ablated by even minor RNA-mediated catalytic suppression (β < 1) (SI Appendix, Fig. S7). Importantly, this synergy is completely dependent on formation of the stable ter- nary complex, and preventing its formation (δ1 = 0) ablates RNA-dependent increases in activity (SI Appendix, Fig. S8). Thus, direct translocation between RNA and nucleosome DNA without a free-enzyme intermediate seems necessary to improve activity rate for the alternative situation of independent RNA and nucleosome binding. These data address chromatin binders with catalytic activity, but RNA-binding transcription factors (RBTFs) like YY1 have biological activity that is not catalytic in nature. To interrogate the relevance PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 5 of 11 of our findings to such RBTFs, we eliminated catalytic activity (kcat = 0) from the prior reaction scheme (Fig. 5A) and monitored nucleosome binding. Our results suggest RNA could improve RBTF activity at high effective molarity for ternary complex-forming reac- tions (α > 1) without any detrimental effects in free solution (α = 1) (SI Appendix, Fig. S9A). However, this is dependent on an RBTF’s ability to function on its nucleosome target with RNA co-bound (SI Appendix, Fig. S9B) and on formation of the ternary complex (SI Appendix, Fig. S9C). Consequently, translocation without a free protein intermediate seems necessary for RNA-mediated facilitation of activity for proteins with independent RNA and nucleosome binding, independent of whether the protein acts catalytically or simply by binding DNA. These findings indicate that competitive (PRC2-like; Fig. 4A) and independent (YY1-like; Fig. 5A) RNA and nucleosome bind- ing systems can both have RNA-mediated facilitation of their activity, and that this facilitation is dependent on the ability to translocate between RNA and nucleosomes without a free-protein intermediate. While PRC2 would accomplish this through direct transfer, independent binding systems accomplish this through a stable ternary complex. However, while our PRC2 reaction scheme for direct transfer events (Fig. 4A) doesn’t explicitly iden- tify a ternary complex (Fig. 5A), the existence of an unstable ternary complex intermediate is still implied. Indeed, making the ternary complex for an HMTase with independent binding (Fig. 5A) a million-fold less stable (δ1 = 10−1, δ2 = 105) allows RNA facilitation of activity (SI Appendix, Fig. S10). Interestingly, non-catalytic independent binders (i.e., RBTFs) have their RNA-mediated effects ablated by relatively minor destabilization (δ1 = 1−1, δ2 = 102) of the ternary complex if there is no bias between ligands (δ2N = δ2R) (SI Appendix, Fig. S9D). However, a million-fold ternary complex still allows RNA-mediated recruitment (SI Appendix, Fig. S9E) or inhibition (SI Appendix, Fig. S9F) if the RBTF has a bias toward RNA (δ1 = 10−1, δ2N = 104, δ2R = 106) or nucleosome (δ1 = 10−1, δ2N = 106, δ2R = 104) dissociation from the ternary complex, respectively. Collectively these data suggest that the seemingly distinct PRC2-like and Sigova et al. models for RNA-mediated recruitment to chromatin both rely on translocation between RNA and nucleosome DNA through a ternary complex inter- mediate, and they differ only in the stability of their ternary complexes (i.e., the lifetime of the ternary intermediate). Thus, some form of direct transfer may be generally necessary for RNA-binding chromatin-associated proteins to have their func- tions on chromatin facilitated by RNA. less stable is insufficient that direct transfer alone Our data suggest that direct transfer creates a synergistic rela- tionship between RNA concentration and protein activity, but only if the α value (i.e., RNA–nucleosome proximity) is high, for implying RNA-mediated facilitation of protein function. We next con- sider whether RNA–nucleosome proximity could allow RNA-mediated protein recruitment to chromatin for strictly exclusive binders (i.e., no direct transfer). It might seem that a reservoir of RNA-bound protein directly adjacent to chromatin DNA could increase chromatin occupancy by the protein. However, our results imply that RNA–nucleosome proximity without direct transfer should be insufficient for RNA-mediated facilitation of protein activity (SI Appendix, Fig. S6), though we acknowledge that our simulations only incorporate the α param- eter in the presence of direct transfer. To test this question in a manner that avoids use of the α parameter, we employed single-molecule dynamics (SMD) simulations of mutually exclu- sive protein binding to RNA and nucleosomes tethered together. Fig. 4. Direct transfer allows RNA to boost PRC2 HMTase activity. (A) Reaction scheme of PRC2-like protein (E) binding of RNA (R) and nucleosomes (N) with catalytic activity on nucleosomes (Nm), where conjugates are complexes of the respective reactants. Major protein states are shown in red, additional reactants in purple, and rate constants and tuning parameters in blue. For rate constants, k1 is for association, k−1 is for dissociation, kθ is for direct transfer, and kcat is for catalysis. The α tuning parameter is included for any protein complex transitions where RNA–nucleosome direct transfer is possible. It is an adjustment of effective molarity for direct transfer reactions, meant to account for the spatial proximity of nascent RNA and nucleosome DNA (e.g., nascent RNA), but it has a complex relationship with other factors that warrants qualitative interpretation (see corresponding text). Specific nomenclature definitions are in SI Appendix, Table S1. These reactions are described by the system of differential equations, SI Appendix, Eq. S5. Inter-complex transitions defined by the kθ rate constants are like those shown in Fig. 1, and their removal collapses this scheme to a classic model. (B) Reactions were simulated using SI Appendix, Eq. S5 for the scheme (panel A) to monitor rate of nucleosome methylation (H3K27me3) over time under varying RNA–nucleosome molar ratios (RNA:Nuc), direct transfer effective molarity adjustments (α), and protein concentrations. Black curves represent HMTase time-course reactions in the absence of RNA, and the colored lines represent the effect of increasing RNA concentrations. For simulations, k−1, kθ, and Kd values were taken from Table 1, kcat was taken from prior PRC2 literature, [NT] = 5 nM, [ET] = 0.1–2 × KdN, and other parameter values are indicated; explicit values are provided in Materials and Methods. Limited data from a single protein concentration (2 × KdN) are shown, but the full data set is provided in SI Appendix, Fig. S4. (C) HMTase activity rate data were used to calculate the relative initial rates (V0) for reactions with 8:1 versus 0:1 RNA–nucleosome molar ratios (R:N), across a range of α values. Data used were the same as for panel B. Dotted line is a visual aid for when RNA concentration has no effect on initial reaction rate. 6 of 11 https://doi.org/10.1073/pnas.2220528120 pnas.org Our findings indicate that increasing RNA binding affinity increases RNA occupancy (SI Appendix, Fig. S11A) as expected, but slightly decreases nucleosome occupancy (SI Appendix, Fig. S11B). Similarly, the simulations show increased intermo- lecular distance between nucleosomes and nearby unbound protein (SI Appendix, Fig. S11C) and a reduced concentration of unbound in nucleosome-adjacent solvent space (SI Appendix, protein Fig. S11D), confirming an antagonistic relationship between RNA and nucleosome occupancy. We note that our SMD approach is a first approximation that does not, for example, account for Debye–Hückel effects on short range electrostatics, which could affect the assumption of isotropic diffusion around RNA/nucle- osome. Thus, our collective findings suggest that RNA–nucle- osome proximity alone is not sufficient for RNA-mediated recruitment of mutually exclusive binders to chromatin, with the limitation that more fine-grained interrogations may show otherwise under specific conditions. Discussion Implications for PRC2 Biology. Our biophysical studies indicate that PRC2 is intrinsically capable of direct transfer between G4 RNA and nucleosome-linker-sized dsDNA (Fig. 2), and our computational investigations reveal that this behavior could allow RNA to have either an antagonistic or a synergistic effect on PRC2 activity depending on the relative RNA–nucleosome effective molarity (i.e., RNA–nucleosome proximity) (Fig. 4). These findings provide direct evidence for a mechanism (Fig. 6A) that theoretically allows RNA to facilitate PRC2 HMTase activity under certain conditions, which could reconcile prior perplexing in vitro and in vivo results where RNA was alternatively found to inhibit PRC2 or recruit it to sites of action. We propose that PRC2 binding to nascent RNA (Fig. 6A–1-2) could increase the effective molarity for direct transfer events (Fig. 6A–3), allowing increased chromatin association and H3K27me3 deposition (Fig. 6A–4) relative to an RNA-free (or RNA binding-free) system. While these findings demonstrate that PRC2 direct transfer kinetics might support RNA-mediated recruitment to chromatin under certain conditions, they do not prove its occurrence in vivo. Specifically, we do not know what α parameter values pertain to physiological conditions, the effects of ligand/competitor length on direct transfer kinetics were not robustly explored here, and we only tested one combination of accessory proteins for the PRC2 complex. However, we note that our ligands represent the core G4 RNA structure and average nucleosome linker DNA length. Ideally, one would test in vivo a separation-of-function mutant that prevented direct transfer but retained full RNA and chromatin binding activities. However, we are pessimistic that such a mutant could be obtained, given that direct transfer is likely an intrinsic property of the nucleic acid-binding surfaces of PRC2. Although it is unclear whether direct transfer facilitates RNA-mediated PRC2 recruitment in vivo, we note that the ~1 mM nucleotide concentration of RNA in a human cell nucleus (43) could satisfy the 10 µM RNA competitor concentrations required for substantial direct transfer in our experiments, sug- gesting that PRC2 biology is impacted by flux through direct transfer. The alternative situation, where RNA binding inhibits PRC2 activity, appears to occur in vivo. For example, it explains why many active genes have PRC2 close enough to be captured by ChIP (chromatin immunoprecipitation), yet the PRC2 does not act there (39). Furthermore, RNA inhibition of PRC2 has been Fig. 5. Stable RNA and nucleosome cobinding could boost a protein’s activity. (A) Reaction scheme of a protein (E) binding RNA (R) and nucleosomes (N) independently, with the potential for catalytic activity on nucleosomes (Nm), where conjugates are complexes of the respective reactants. Major protein states are shown in red, additional reactants in purple, and rate constants and tuning parameters in blue. For rate constants, k1 is for association, k−1 for dissociation, and kcat is for catalysis. The α tuning parameter is included for any protein complex transitions where ternary complex formation from a bimolecular complex is possible. It is an adjustment of effective molarity for direct transfer reactions, meant to account for the spatial proximity of nascent RNA and nucleosome DNA (e.g., nascent RNA), but it has a complex relationship with other factors that warrants qualitative interpretation (see corresponding text). The β tuning parameter is effect of bound RNA on catalytic activity. The δ tuning parameters are the effects of bound RNA/nucleosome on ternary nucleosome/RNA binding or dissociation. Specific nomenclature definitions are in SI Appendix, Table. S1. (B) Reactions were simulated using SI Appendix, Eq. S6 for the scheme (panel A) to monitor rate of nucleosome methylation (H3K27me3) over time under varying RNA–nucleosome molar ratios (RNA:Nuc) and effective molarity adjustments (α). Black curves represent HMTase time-course reactions in the absence of RNA, and the colored lines represent the effect of increasing RNA concentrations. For α = 1, all lines overlap. For simulations, Kd values were taken from Sigova et al, kcat was set to the value from PRC2 literature, [NT] = 5 nM, [ET] = 0.1−2 × KdN, β = 0−1, δ = 1, and other parameter values are indicated; explicit values are provided in Materials and Methods. Limited data from a single protein concentration (0.125 × KdN) and β value (β = 1) are shown, but the full data set is provided in SI Appendix, Fig. S7. (C) HMTase activity rate data were used to calculate the relative initial rates (V0) for reactions with 8:1 versus 0:1 RNA–nucleosome molar ratios (R:N), across a range of α values. Data used were the same as for panel B. Dotted line indicates RNA concentration having no effect on initial reaction rate. PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 7 of 11 Fig. 6. A direct transfer model of RNA regulation of PRC2 HMTase activity. (A) Proposed Steps for an RNA Recruitment Model of PRC2. (1) G4-containing nascent RNA at transcriptionally active PRC2 target genes (2) is bound by PRC2, (3) RNA-tethered PRC2 is transferred onto spatially proximal nucleosomes, then (4) PRC2 deposits its H3K27me3 mark. (B) Proposed Mechanisms for the Direct Transfer Step. PRC2 could have shared contacts for G4 RNA and nucleosome DNA binding but allow the ligands to occupy partially associated binding states that permit transient cobinding. Nucleosome DNA could give the appearance of actively disrupting a PRC2–RNA complex (Left) by forming a highly transient ternary intermediate where the PRC2-RNA interaction is destabilized (Middle Top). The unstable ternary intermediates may quickly dissociate to form a more stable PRC2-nucleosome (Right) or PRC2-RNA (Left) complex. shown in cells by Jenner et al. (20, 21). Consistent with these observations, our simulations (Fig. 4) indicate that RNA is indeed antagonistic below a certain threshold of α (i.e., low RNA–nucle- osome proximity), although we have no way of predicting when and where these threshold conditions would be met in vivo. Importantly, our demonstration of PRC2’s direct transfer kinetics means that the recruitment and eviction models need not be mutually exclusive. As such, it is possible that RNA-mediated regulation of PRC2 operates as a “switch,” where predominantly free versus predominantly nascent RNA landscapes around target genes drive PRC2’s relationship with RNA being antagonistic versus synergistic, respectively (44). Biophysical Mechanism for Direct Transfer. These findings demonstrate that PRC2 can translocate directly between polynucleotide species that are biologically relevant. Previous studies of such direct transfer have concerned homo-multimeric proteins, where a ligand bound to one protomer is in position to displace a second ligand bound to a nearby protomer. Therefore, the occurrence of direct transfer with PRC2 was unexpected, and it raises compelling questions about the underlying biophysical mechanism. We consider a model (Fig. 6B) that involves PRC2 ligands, such as RNA and nucleosome DNA, competing for the same or overlapping binding sites. Dynamic motion of the protein and/or the RNA gives a partially dissociated intermediate allowing the nucleosome to bind, forming an unstable ternary complex (Fig. 6B–top-middle). Full dissociation of the RNA ligand then allows full association of the nucleosome ligand (right). We note that it is not yet clear how RNA and nucleosome DNA compete for binding on the surface(s) of PRC2, though some structural insights are emerging (45–47). This lack of defi- nition in the PRC2 binding surfaces and the heterogeneity and complexity of PRC2 limit our ability to critically evaluate this model for PRC2 at this time. It is, for example, alternatively possible that the PRC2 direct transfer via an unstable ternary complex could be facilitated by distinct binding surfaces with mutual negative allosteric regulation. However, we believe several pieces of evidence favor the shared contacts model (Fig. 6B). First, prior work has implicated similar mechanisms in direct transfer kinetics, and suggested that many nucleic acid binding 8 of 11 https://doi.org/10.1073/pnas.2220528120 pnas.org interfaces could have the capacity to support them (26, 35, 37). Indeed, our own concurrent studies (48) provide mechanistic evidence for direct transfer occurring commonly with nucleic acid binding proteins. Second, we also observed PRC2 direct transfer between identical (labeled versus unlabeled) G4 RNA species and between identical dsDNA species (Fig. 2). Unless PRC2 has multiple distinct binding sites with intersite negative allostery for each of these ligands, our observed direct transfer kinetics seem unlikely to be produced without shared contacts. Finally, recent structural and biophysical studies support some degree of overlap in PRC2 contacts with G4 RNA and dsDNA (47). Ultimately, however, elucidating the full-resolution mech- anism that facilitates PRC2 direct transfer will require more structural and biophysical work. Implications for Other Chromatin Modifiers. Accumulating evidence suggests that RNA-binding activity is common among chromatin-associated proteins (14, 49–52), and our simulations raise the possibility that direct transfer might be generally required for RNA-mediated recruitment models of such proteins. Our concurrent studies (48) suggest that the capacity for direct transfer could be quite common among other nucleic acid binding proteins. Furthermore, the kθD/k−1P ratios for direct transfer reactions in Table 1 and our concurrent studies average ~105 M−1, suggesting that flux through a direct transfer pathway (Fig. 6B) exceeds classic dissociation at competitor effective molarities above ~10 µM (48). We note, by the same kθ/k−1 metric, that the 10 µM competitor condition for direct transfer would also apply to proteins like SSB, CAP, and recA that are typically considered to be proficient at direct transfer (25–27). Micromolar effective molarities are likely to be achieved in cells, because the nucleotide concentration of RNA is ~1 mM (~50 µM of a 20-nt RNA) in a cell nucleus and DNA nucleotides are ~10- to 40-fold more concentrated (43, 53); thus, there is potential for biologically relevant flux through a direct transfer (versus classic) pathway in vivo. We note in the specific case of PRC2 that not all RNA in a cell nucleus would form the G4 RNA preferred by PRC2; however, only a minority of nuclear RNA must form a G4 structure to achieve micromolar G4 RNA concentrations, the effective molarity of G4s may be increased in some contexts (e.g., nascent RNA and nucleosome DNA), and PRC2 also exhibits intermediate affinity for many other RNA sequences/structures (16). If direct transfer capability proves to be pervasive among chromatin-associated proteins, then our findings for PRC2 might explain why so many chromatin-associated proteins exhibit RNA-binding activity: intrinsic direct transfer capabil- ity could allow for RNA-mediated regulation. As a recent exam- ple, the RNA-binding domain of CCCTC-binding factor (CTCF) has been proposed to increase its search efficiency for DNA target sites [(54, 55); see also refs. 44 and 56]. In our work, it’s important to distinguish the characteristics of com- petitive binding (PRC2-like) (Fig. 4) versus independent bind- ing (YY1-like) (Fig. 5) for direct transfer. In the former case, RNA can recruit protein under high RNA–nucleosome prox- imity conditions but actively antagonize protein activity under low RNA–nucleosome proximity conditions. In the latter case, while RNA could indeed recruit proteins if there is high RNA– nucleosome proximity, it may be unable to antagonize protein activity if RNA is predominantly in free solution, unless bound RNA affects other nucleosome binding-independent protein function(s). Thus, it is possible that chromatin-associated pro- teins could evolve PRC2-like versus YY1-like direct transfer in response to physiological pressures for tight regulation versus efficient recruitment, respectively. Future in vitro and in vivo studies with a diversity of chromatin-associated proteins are war- ranted to interrogate the prevalence and nature of direct transfer’s role(s) in RNA-mediated regulation of gene expression. Materials and Methods PRC2 Expression and Purification. According to prior methodology (39), we used pFastBac vectors encoding N-terminally MBP-tagged fusions of each of the four core PRC2 subunits (EZH2, SUZ12, EED, and RBBP4), and either the embry- onic (PRC25me) or somatic (PRC25m) isoform of AEBP2 (57), to prepare respective baculovirus stocks for co-infection of Sf9 cells. Then, according to prior meth- odology (16), cell paste containing expressed PRC2 was lysed, clarified, then purified by sequential amylose column chromatography, MBP-tag cleavage, hep- arin column chromatography, and size-exclusion column chromatography. The PRC25me protein was used for all reported experiments, except where otherwise indicated (Table 1). Preparation of Polynucleotides. All oligos were ordered from IDT, and their sequences in IDT syntax are provided (SI Appendix, Table S1). For dsDNA con- structs, complementary oligos ordered from IDT were mixed at 5 µM (ligand) or 300 µM (competitor) each in annealing buffer (50 mM TRIS pH 7.5 at 25 °C, 200 mM NaCl), subjected to a thermocycler program (95 °C for 10-min, 954 °C at 0.5 °C/min, hold at 4 °C) for annealing, then annealing confirmed via Native-PAGE. Concentrations of all ligands were confirmed spectroscopically using manufac- turer-provided extinction coefficients. Binding Buffers. All binding buffers (BB) contained 50 mM TRIS (pH 7.5 at 25 °C), 2.5 mM MgCl2, 0.1 mM ZnCl2, 0.1 mg/mL BSA, 5% v/v glycerol, and 2 mM 2-mercaptoethanol, plus a variable concentration of KCl (10, 25, 100, or 200 mM). Subscript of each binding buffer indicates the concentration of KCl in milli-molarity (e.g., BB25 = 25 mM KCl). FP-Based Kd Determination. Pre-reaction mix was prepared with 5 nM ligand molecule in respective binding buffer (Binding Buffers), then dispensed in 36 µL volumes into the wells of a 384-well black microplate (Corning #3575). PRC2 was prepared at 10X the reported concentrations via serial dilution in binding buffer. Binding reactions were initiated by addition of 4 µL of PRC2 solution to the corresponding prereaction mix, then incubated 30 min at room temperature. Wells with binding buffer only were also included for blanking. Fluorescence polarization readings were then taken for 30 min in 30-s intervals with a TECAN Spark microplate reader (excitation wavelength = 481 ± 20 nm, emission wave- length = 526 ± 20 nm). Each experiment had two or four technical replicates per protein concentration (as indicated), and at least three independent experiments were performed per protein–polynucleotide combination. Raw data were analyzed in R v4.1.1 with the FPalyze function (FPalyze v1.3.1 package; see Data, Materials, and Software Availability). Briefly, polarization versus time data were calculated for each reaction, the last 10 data points for each reac- tion averaged to generate an equilibrium polarization value, and then equilibrium polarization values were plotted as a function of protein concentration. Plot data were regressed with SI Appendix, Eq. S2 to calculate Kd app for the interaction. FPCD Experiments. Pre-reaction mix was prepared with 5 nM ligand molecule app (at 25 °C) in binding buffer (see Binding Buffers), then dis- and PRC2 ≥ 2xKdP pensed in 36 µL volumes into the wells of a 384-well black microplate (Corning #3575). Decoy was prepared at 10X the reported concentrations via serial dilution in binding buffer or carrier polynucleotide (Table 1) at a concentration equal to the highest competitor concentration. Pre-reaction mix and competitor dilutions were then incubated at the indicated temperature to attain thermal and binding equilibrium (4 °C/90 min or 25 °C/30 min). Competitive dissociation reactions were initiated by addition of 4 µL of the respective competitor concentration to the corresponding pre-reaction mix, then fluorescence polarization readings were immediately (the delay between initiation of the first reactions and the first polarization reading was ~90 s) taken at 25 °C for 120 min in 30-s intervals with a TECAN Spark microplate reader (Ex = 481 ± 20 nm, Em = 526 ± 20 nm). Each experiment had 4 technical replicates per competitor concentration, and at PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 9 of 11 least three independent experiments were performed per protein–polynucleotide combination unless otherwise indicated. All reported competition reactions used a PRC2 concentration of 100 nM. Raw data were analyzed in R v4.1.1 with the FPalyze function (FPalyze v1.3.1 package). Briefly, polarization versus time data were calculated for each reaction, each reaction’s polarization data were normalized to the maximum and mini- mum polarization across all reactions, each normalized reaction was fit with an obs exponential dissociation function (SI Appendix, Eq. S3.1) to determine koff obs values were plotted as a function of compet- (SI Appendix, Eq. S3.2), and koff obs subtracted to mitigate itor concentration. Plotted data (with background koff temperature effects on polarization) were regressed (the theoretical background for this approach is thoroughly covered in a separate manuscript) via SI Appendix, Eq. S4.1 and then SI Appendix, Eq. S4.2 with tuning parameters constrained to the SI Appendix, Eq. S4.1 solutions, then the regression models compared with the Bayesian Information Criterion (58) (BIC). Rate constants (k−1P and/or kθD) were determined from the best-performing regression model. If minimum polarization was not reached during competition experiments (e.g., due to a weak competitor), then it was manually defined with minimum polarization data from corresponding binding curve data (FP-Based Kd Determination). app (FP-based Kd Determination). Then, log10{Kd Ionic Strength Dependence. Binding curve data (Fig. 3A) were regressed as app} described to determine Kd versus log10{[KCl]} plots were regressed in R via stats:lm (package:function). Regression indicated m = 5.7 × 10−4 ± 0.34 (slope) and b = −8.5 ± 0.49 (intercept) for the G4 RNA data, and m = 1.4 ± 0.68 and b = −5.2 ± 1.1 for the dsDNA data, where values are the regressions’ estimate ± SE. PRC2 Reaction Scheme Simulations. Reactions (Fig. 4A) were simulated and analyzed in R v4.1.1 with a custom script (Data, Materials, and Software Availability). Briefly, [ET], [NT], [RT], Kd, k−1, kθ, kcat, and α were user-provided. Then, other rate constants and initial conditions were calculated via SI Appendix, Eq. S7, and the system of differential equations (SI Appendix, Eq. S5) was solved by numerical integration. Initial reaction rates (V0) were calculated as the average rate of change in [mT] during the first 5% of each reaction. By default, k−1N = 9.1 × 10−5 s−1, k−1R = 1.7 × 10−3 s−1, kθN = 91, kθR = 170, kθNN = 260, KdN = 5.1 nM, and KdR = 2.3 nM were taken directly from Table 1 (k−1 are from self-competitions), kcat = 10−1 s−1 was taken from PRC2 literature (41), [NT] = 5 nM was chosen arbitrarily, and all other parameter values were varied as indicated. By exception, kθ = 0 for the SI Appendix, Fig. S6 studies, and k−1R = 1.7 × 106 s−1 and KdR = 2.3 M for the SI Appendix, Fig. S5 studies. We note that the k−1R value used was necessarily from BB25 buffer conditions, while other constants were from BB10 buffer conditions, but we also note that our salt dependency data (Fig. 3, Fig. 6, and Table 1) suggest that this produced no meaningful discrepancy. Cobinder Reaction Scheme Simulations. Reactions (Fig. 5A) were simulated and analyzed in R v4.1.1 with a custom script (Data, Materials, and Software Availability). Briefly, [ET], [NT], [RT], Kd, k−1, kcat, α, β, and δ were user-provided. Then, other rate constants and initial conditions were calculated via SI Appendix, Eq. S7, and the system of differential equations (SI Appendix, Eq. S6) was solved by numerical integration. Initial reaction rates (V0) were calculated as the average rate of change in [mT] during the first 5% of each reaction. By default, KdR = 400 nM and KdN = 200 nM were taken from Sigova et al. (42), k1 = 105 M−1 s−1 was selected as a typical on-rate, k−1 = Kd × k1 s−1, kcat = 10−1 s−1 was constrained to the value for PRC2, [NT] = 5 nM was used arbitrarily, δ = 1, and all other param- eter values were varied as indicated. By exception, δ1 = 0 for the SI Appendix, Figs. S7 and S9C studies, δ1 = 10−1 for the SI Appendix, Figs. S9 D–F and S10 studies, δ2 = 102 for the SI Appendix, Fig. S9D studies, δ2R = 106 and δ2N = 104 for the SI Appendix, Fig. S9E studies, δ2R = 104 and δ2N = 106 for the SI Appendix, Fig. S9F studies, δ2 = 105 for the SI Appendix, Fig. S10 studies, and kcat = 0 s−1 for the SI Appendix, Fig. S9 studies. Diagram, Reaction Scheme, and Figure Generation. Diagrams were prepared with BioRender, reaction schemes were prepared with ChemDraw v21.0.0 (Perkin Elmer), tables were prepared with Word (Microsoft), graphs were prepared with R v4.1.1, protein structures were prepared in PyMOL v2.5.2 (Schrodinger), and figures were assembled in PowerPoint (Microsoft). Data, Materials, and Software Availability. GitHub hosts the FPalyze (github.com/whemphil/FPalyze) (59) R package. The custom scripts referenced in these methods are available on GitHub (github.com/whemphil/PRC2_Direct- Transfer_Manuscript) (60). pFastBac vectors for PRC2 expression have been deposited to AddGene (ID #125161-125165) by the Davidovich lab. All lig- ands and competitors are available from IDT via the sequences in SI Appendix, Table S2. Methodology on single-molecule simulations and equations can be found in SI Appendix, Materials and Methods. All other data are included in the manuscript and/or SI Appendix. ACKNOWLEDGMENTS. W.O.H. is supported by the NIH (F32-GM147934). T.R.C. is an investigator of the Howard Hughes Medical Institute. We thank Olke Uhlenbeck, Deborah Wuttke, Halley Steiner, and members of the Cech lab (University of Colorado Boulder) and Freddie Salsbury (Wake Forest University) for stimulating discussion and feedback concerning these studies. 1. 2. 3. 4. 5. 6. 7. 8. 9. R. Cao et al., Role of histone H3 lysine 27 methylation in Polycomb-group silencing. Science 298, 1039–1043 (2002). B. Czermin et al., Drosophila enhancer of Zeste/ESC complexes have a histone H3 methyltransferase activity that marks chromosomal polycomb sites. Cell 111, 185–196 (2002). A. Kuzmichev, K. Nishioka, H. Erdjument-Bromage, P. Tempst, D. Reinberg, Histone methyltransferase activity associated with a human multiprotein complex containing the Enhancer of Zeste protein. Genes. Dev. 16, 2893–2905 (2002). J. Müller et al., Histone methyltransferase activity of a drosophila polycomb group repressor complex. Cell 111, 197–208 (2002). R. Margueron, D. Reinberg, The polycomb complex PRC2 and its mark in life. Nature 469, 343–349 (2011). L. Di Croce, K. Helin, Transcriptional regulation by Polycomb group proteins. Nat. Struct. Mol. Biol. 20, 1147–1155 (2013). J. A. Simon, R. E. Kingston, Occupying chromatin: Polycomb mechanisms for getting to genomic targets, stopping transcriptional traffic, and staying put. Mol. Cell 49, 808–824 (2013). F. Mohn et al., Lineage-specific polycomb targets and de novo DNA methylation define restriction and potential of neuronal progenitors. Mol. Cell 30, 755–766 (2008). D. T. Youmans, A. R. Gooding, R. D. Dowell, T. R. Cech, Competition between PRC2.1 and 2.2 subcomplexes regulates PRC2 chromatin occupancy in human stem cells. Mol. Cell 81, 488–501.e9 (2021). 16. X. Wang et al., Targeting of Polycomb Repressive Complex 2 to RNA by Short Repeats of Consecutive Guanines. Mol. Cell 65, 1056–1067.e5 (2017). 17. J. Yan, B. Dutta, Y. T. Hee, W.-J. Chng, Towards understanding of PRC2 binding to RNA. RNA Biol. 16, 176–184 (2019). 18. S. Kaneko, J. Son, R. Bonasio, S. S. Shen, D. Reinberg, Nascent RNA interaction keeps PRC2 activity poised and in check. Genes. Dev. 28, 1983–1988 (2014). 19. X. Wang et al., Molecular analysis of PRC2 recruitment to DNA in chromatin and its inhibition by RNA. Nat. Struct. Mol. Biol. 24, 1028–1038 (2017). 20. M. Beltran et al., The interaction of PRC2 with RNA or chromatin is mutually antagonistic. Genome. Res. 26, 896–907 (2016). 21. M. Beltran et al., G-tract RNA removes Polycomb Repressive Complex 2 from genes. Nat. Struct. Mol. Biol. 26, 899–909 (2019). 22. C. Cifuentes-Rojas, A. J. Hernandez, K. Sarma, J. T. Lee, Regulatory interactions between RNA and Polycomb Repressive Complex 2. Mol. Cell 55, 171–185 (2014). 23. Y. Long et al., RNA is essential for PRC2 chromatin occupancy and function in human pluripotent stem cells. Nat. Genet. 52, 931–938 (2020), 10.1038/s41588-020-0662-x. 24. T. Ruusala, D. M. Crothers, Sliding and intermolecular transfer of the lac repressor: Kinetic perturbation of a reaction intermediate by a distant DNA sequence. Proc. Natl. Acad. Sci. U.S.A. 89, 4903–4907 (1992). 10. Y. Long et al., Conserved RNA-binding specificity of polycomb repressive complex 2 is achieved by 25. M. G. Fried, D. M. Crothers, Kinetics and mechanism in the reaction of gene regulatory proteins with dispersed amino acid patches in EZH2. eLife 6 (2017). DNA. J. Mol. Biol. 172, 263–282 (1984). 11. A. Petracovici, R. Bonasio, Distinct PRC2 subunits regulate maintenance and establishment of Polycomb repression during differentiation. Mol. Cell 81, 2625–2639.e5 (2021), 10.1016/j. molcel.2021.03.038. 12. C. Davidovich et al., Toward a consensus on the binding specificity and promiscuity of PRC2 for RNA. Mol. Cell 57, 552–558 (2015). 13. C. Davidovich, T. R. Cech, The recruitment of chromatin modifiers by long noncoding RNAs: Lessons from PRC2. RNA 21, 2007–2022 (2015). 26. A. G. Kozlov, T. M. Lohman, Kinetic Mechanism of Direct Transfer of Escherichia coli SSB Tetramers between Single-Stranded DNA Molecules.† Biochemistry 41, 11611–11627 (2002). 27. J. P. Menetski, S. C. Kowalczykowski, Transfer of recA protein from one polynucleotide to another. Kinetic evidence for a ternary intermediate during the transfer reaction. J. Biol. Chem. 262, 2085–2092 (1987). 28. J. Roca, A. Santiago-Frangos, S. A. Woodson, Diversity of bacterial small RNAs drives competitive 14. D. G. Hendrickson, D. R. Kelley, D. Tenen, B. Bernstein, J. L. Rinn, Widespread RNA binding by strategies for a mutual chaperone. Nat. Commun. 13, 2449 (2022). chromatin-associated proteins. Genome Biol. 17, 28 (2016). 15. J. Zhao et al., Genome-wide identification of Polycomb-associated RNAs by RIP-seq. Mol. Cell 40, 939–953 (2010). 29. J. S. Graham, R. C. Johnson, J. F. Marko, Concentration-dependent exchange accelerates turnover of proteins bound to double-stranded DNA. Nucleic Acids Res. 39, 2249–2259 (2011). 10 of 11 https://doi.org/10.1073/pnas.2220528120 pnas.org 30. P. H. von Hippel, A. Revzin, C. A. Gross, A. C. Wang, “Interaction of lac repressor with non-specific DNA binding sites” in Protein-Ligand Interactions (Walter de Gruyter and Co., Berlin, 1975), pp. 270–288. 46. V. Kasinath et al., JARID2 and AEBP2 regulate PRC2 in the presence of H2AK119ub1 and other histone modifications. Science 371, eabc3393 (2021). 31. J. L. Bresloff, D. M. Crothers, DNA-ethidium reaction kinetics: Demonstration of direct ligand transfer 47. J. Song, A. R. Gooding, W. O. Hemphill, V. Kasinath, T. R. Cech, Structural basis for inactivation of between DNA binding sites. J. Mol. Biol. 95, 103–123 (1975). 32. J. Rudolph, J. Mahadevan, P. Dyer, K. Luger, Poly(ADP-ribose) polymerase 1 searches DNA via a PRC2 by G-quadruplex RNA. bioRxiv [Preprint] (2023). https://doi.org/10.1101/2023.02.06.527314 (Accessed 14 February 2023). ‘monkey bar’ mechanism. eLife 7, e37818 (2018). 33. J. Iwahara, M. Zweckstetter, G. M. Clore, NMR structural and kinetic characterization of a homeodomain diffusing and hopping on nonspecific DNA. Proc. Natl. Acad. Sci. U.S.A. 103, 15062–15067 (2006). 34. M. Doucleff, G. M. Clore, Global jumping and domain-specific intersegment transfer between DNA cognate sites of the multidomain transcription factor Oct-1. Proc. Natl. Acad. Sci. U.S.A. 105, 13871–13876 (2008). 48. W. O. Hemphill, C. K. Voong, R. Fenske, J. A. Goodrich, T. R. Cech, RNA- and DNA-binding proteins generally exhibit direct transfer of polynucleotides: Implications for target site search. bioRxiv [Preprint] (2022). https://doi.org/10.1101/2022.11.30.518605 (Accessed 30 November 2022). 49. A. M. Khalil et al., Many human large intergenic noncoding RNAs associate with chromatin- modifying complexes and affect gene expression. Proc. Natl. Acad. Sci. U.S.A. 106, 11667–11672 (2009). 50. L. Skalska et al., Nascent RNA antagonizes the interaction of a set of regulatory proteins with 35. A. Erbaş, J. F. Marko, How do DNA-bound proteins leave their binding sites? The role of facilitated chromatin. Mol. Cell 81, 2944–2959.e10 (2021). dissociation Curr. Opin. Chem. Biol. 53, 118–124 (2019). 36. S. E. Halford, J. F. Marko, How do site-specific DNA-binding proteins find their targets? Nucleic Acids Res. 32, 3040–3052 (2004). 51. H. R. Steiner, N. C. Lammer, R. T. Batey, D. S. Wuttke, An extended DNA binding domain of the estrogen receptor alpha directly interacts with RNAs in vitro. Biochemistry 61, 2490–2494 (2022), 10.1021/acs.biochem.2c00536. 37. C. E. Sing, M. Olvera de la Cruz, J. F. Marko, Multiple-binding-site mechanism explains 52. O. Oksuz et al., Transcription factors interact with RNA to regulate genes. bioRxiv [Prepint] (2022). concentration-dependent unbinding rates of DNA-binding proteins. Nucleic Acids Res. 42, 3783–3791 (2014). https://doi.org/10.1101/2022.09.27.509776 (Accessed 28 September 2022). 53. J. F. Gillooly, A. Hein, R. Damiani, Nuclear DNA content varies with cell size across human cell types. 38. M. T. Record, T. M. Lohman, P. de Haseth, Ion effects on ligand-nucleic acid interactions. J. Mol. Biol. Cold Spring Harb. Perspect. Biol. 7, a019091 (2015). 107, 145–158 (1976). 54. A. S. Hansen, A. Amitai, C. Cattoglio, R. Tjian, X. Darzacq, Guided nuclear exploration increases CTCF 39. C. Davidovich, L. Zheng, K. J. Goodrich, T. R. Cech, Promiscuous RNA binding by Polycomb repressive target search efficiency. Nat. Chem. Biol. 16, 257–266 (2020). complex 2. Nat. Struct. Mol. Biol. 20, 1250–1257 (2013). 55. A. S. Hansen et al., An RNA-binding region regulates CTCF clustering and chromatin looping 40. R. M. Saecker, M. T. Record, Protein surface salt bridges and paths for DNA wrapping. Curr. Opin. (Biophysics, 2018). 10.1101/495432 (April 13, 2023). Struct. Biol. 12, 311–319 (2002). 41. T. J. Wigle et al., The Y641C mutation of EZH2 alters substrate specificity for histone H3 lysine 27 methylation states. FEBS Lett. 585, 3011–3014 (2011). 56. L. Ringrose, S. Chabanis, P.-O. Angrand, C. Woodroofe, A. F. Stewart, Quantitative comparison of DNA looping in vitro and in vivo: chromatin increases effective DNA flexibility at short distances. EMBO J. 18, 6630–6641 (1999). 42. A. A. Sigova et al., Transcription factor trapping by RNA in gene regulatory elements. Science 350, 57. H. Kim, M. B. Ekram, A. Bakshi, J. Kim, AEBP2 as a transcriptional activator and its role in cell 978–981 (2015). 43. A. Khong, R. Parker, The landscape of eukaryotic mRNPs. RNA 26, 229–239 (2020). 44. L. Ringrose, Noncoding RNAs in Polycomb and Trithorax Regulation: A Quantitative Perspective. Annu. Rev. Genet. 51, 385–411 (2017). 45. A. G. Iragavarapu, L. Yao, V. Kasinath, Structural insights into the interactions of Polycomb Repressive migration. Genomics 105, 108–115 (2015). 58. Y. Sakamoto, M. Ishiguro, G. Kitagawa, Akaike Information Criterion Statistics (D. Reidel Publishing Company, ed. 3, 1986). 59. W. Hemphill, FPalyze. GitHub. https://github.com/whemphil/FPalyze. Deposited 16 May 2023. 60. W. Hemphill, PRC2_Direct-Transfer_Manuscript. GitHub. https://github.com/whemphil/PRC2_ Complex 2 with chromatin. Biochem. Soc. Trans. 49, 2639–2653 (2021). Direct-Transfer_Manuscript. Deposited 16 May 2023. PNAS 2023 Vol. 120 No. 23 e2220528120 https://doi.org/10.1073/pnas.2220528120 11 of 11
10.1073_pnas.2220159120	RESEARCH ARTICLE \| CELL BIOLOGY OPEN ACCESS Osteolectin increases bone elongation and body length by promoting growth plate chondrocyte proliferation Jingzhu Zhanga , Liming Dua, Bethany Davisa, Zhimin Gua,1, Junhua Lyua, Zhiyu Zhaoa, Jian Xua , and Sean J. Morrisona,b,c,2 Contributed by Sean J. Morrison; received November 26, 2022; accepted April 21, 2023; reviewed by Roberto Civitelli, Ophir D. Klein, and Francesca V. Mariani Osteolectin is a recently identified osteogenic growth factor that binds to Integrin α11 (encoded by Itga11), promoting Wnt pathway activation and osteogenic differ- entiation by bone marrow stromal cells. While Osteolectin and Itga11 are not required for the formation of the skeleton during fetal development, they are required for the maintenance of adult bone mass. Genome-wide association studies in humans reported a single-nucleotide variant (rs182722517) 16 kb downstream of Osteolectin associated with reduced height and plasma Osteolectin levels. In this study, we tested whether Osteolectin promotes bone elongation and found that Osteolectin-deficient mice have shorter bones than those of sex-matched littermate controls. Integrin α11 deficiency in limb mesenchymal progenitors or chondrocytes reduced growth plate chondrocyte proliferation and bone elongation. Recombinant Osteolectin injections increased femur length in juvenile mice. Human bone marrow stromal cells edited to contain the rs182722517 variant produced less Osteolectin and underwent less oste- ogenic differentiation than that of control cells. These studies identify Osteolectin/ Integrin α11 as a regulator of bone elongation and body length in mice and humans. osteolectin \| integrin α11 \| chondrocyte \| bone elongation \| osteogenesis Height is a polygenic trait (1) determined by the longitudinal growth of limb bones and vertebrae. In juvenile mammals, bones grow longitudinally as a result of the proliferation and osteogenic differentiation of chondrocytes in the growth plate (2). Chondrocyte proliferation and bone growth are systemically promoted by hormones (3) as well as by the activation of Wnt (4, 5) and Hedgehog (6, 7) signaling in chondrocytes. In young mice, skeletal stem cells are present among parathyroid hormone (PTH)– related protein (Pthrp)-CreER and Aggregan-expressing chondrocytes in the resting zone of the growth plate (8). These stem cells proliferate to form columns of chondrocytes in the growth plate that then differentiate into osteoblasts that contribute to bone growth (8). The chondrocytes also give rise to leptin receptor–expressing (LepR+) stromal cells that migrate into the bone marrow metaphysis (8, 9). Dlx-CreER+ perichondrial cells contribute to the formation of diaphyseal bone during fetal and early postnatal develop- ment and give rise to LepR+ cells in diaphysis bone marrow (10). The LepR+ cells that arise from these sources are rare in early postnatal bone marrow but expand in number to account for 0.3% of cells in adult bone marrow, where they are a key source of growth factors for the maintenance of hematopoietic stem cells and restricted hematopoietic progenitors (11–14). The LepR+ cells also include the skeletal stem cells that are the main source of osteoblasts and adipocytes in adult bone marrow (15), as well as restricted osteogenic (16) and adipogenic (17) progenitors. Osteolectin (Clec11a) is an osteogenic growth factor that binds to Integrin α11 (encoded by Itga11), promoting Wnt pathway activation and osteogenic differentiation by bone marrow stromal cells, including LepR+ cells (18, 19). Osteolectin and Integrin α11 are not necessary for the regulation of hematopoiesis or the formation of the skeleton during fetal or early postnatal development but are necessary for the maintenance of adult bone mass, at least in mice (18–20). Osteolectin is synthesized by LepR+ bone marrow stromal cells, hypertrophic chondrocytes, osteoblasts, osteocytes, and periosteal cells (16, 18, 19). Human and mouse Osteolectin are 80% identical at the amino acid level. Recombinant human Osteolectin promotes osteogenic differentiation by human bone marrow stromal cells (18, 19). Osteolectin expression is induced by PTH in mice and humans, and it mediates the osteogenic effects of PTH in mice (20); however, there is not yet functional evidence that Osteolectin regulates the human skeleton in vivo. Osteolectin and Integrin α11 were not previously known to play any role in the regu- lation of bone elongation or height/body length. However, when we analyzed the results from genome-wide association studies of human genetic variants associated with differ- ences in plasma protein levels (21) and height (22–24), we found a single-nucleotide Significance Osteolectin promotes the maintenance of bone mass in adult mice, but there has not been genetic evidence that Osteolectin is functionally important in humans. In the current study, we identified a function for Osteolectin in juvenile mice: the promotion of bone elongation by increasing growth plate chondrocyte proliferation, leading to increased body length. Second, we describe a human genetic variant that suggests Osteolectin also promotes bone elongation in humans. The single-nucleotide variant (rs182722517) that is associated with reduced height and plasma Osteolectin levels in humans is also associated with reduced Osteolectin expression in bone marrow stromal cells. Author contributions: J.Z., J.X., and S.J.M. designed research; J.L. performed research; J.Z., J.L., Z.Z., J.X., and S.J.M. analyzed data; and J.Z. and S.J.M. wrote the paper. J.Z., L.D., B.D., Z.G., and Reviewers: R.C., Washington University in St. Louis School of Medicine; O.D.K., University of California; and F.V.M., University of Southern California. Competing interest statement: UT Southwestern Medical Center has filed a patent application on the use of Osteolectin to promote bone formation. S.J.M. is an advisor for Garuda Therapeutics, Kojin Therapeutics, Frequency Therapeutics, ONA Therapeutics, and Inception Therapeutics. Copyright © 2023 the Author(s). Published by PNAS. This open access article is distributed under Creative Commons Attribution License 4.0 (CC BY). 1Present address: Institute of System Medicine, Chinese Academy of Medical Sciences, Suzhou, Jiangsu Province, 215123, China. 2To whom correspondence may be addressed. Email: [email protected]. This article contains supporting information online at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas. 2220159120/-/DCSupplemental. Published May 22, 2023. PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 1 of 11 variant (rs182722517) 16 kb downstream of Osteolectin that is associated with reduced plasma Osteolectin levels and height. Based on these observations, we set out to test whether Osteolectin promotes bone elongation. Results Osteolectin Promotes Chondrocyte Proliferation and Bone Elongation. To test whether Osteolectin regulates longitudinal bone growth, Osteolectin-deficient mice (18) were analyzed at postnatal day 4 (P4), P14, 4 week (4W), and 8 week (8W) of age. At P4 and P14, Osteolectin-deficient and sex-matched littermate control mice exhibited no differences in body mass, the lengths of femurs or the third lumbar spine (LS3) vertebrae (SI Appendix, Fig. S1 A and B), or femur or vertebra bone parameters (SI Appendix, Fig. S1 C–H). Osteolectin deficiency thus had no effect on the formation or growth of the skeleton prior to P14. At 4 and 8 wk of age, Osteolectin-deficient and sex-matched littermate control mice did not differ in body mass (Fig. 1 A and B) or in femur cortical bone parameters (SI Appendix, Fig. S2 A and B). At 4 wk of age, Osteolectin-deficient and control mice also did not differ in trabecular bone parameters within femurs or vertebrae (SI Appendix, Fig. S2 C and D) or vertebral length (Fig. 1A), but Osteolectin-deficient mice had femurs that were 3 to 4% shorter than those in control mice (Fig. 1A). At 8 wk of age, the Osteolectin-deficient mice had femurs and vertebrae that were 3 to 4% shorter than those of control mice (Fig. 1B) as well as reduced trabecular bone volume, number, and thickness (SI Appendix, Fig. S2 E and F). Thus, Osteolectin is required for bone elongation in juvenile mice, even before detectably contrib- uting to trabecular bone volume. To test whether Osteolectin promotes growth plate chondrocyte proliferation, we administered a 4-h to 2-d pulse (depending on mouse age) of 5-ethynyl-2′-deoxyuridine (EdU) to Osteolectin-deficient and sex-matched littermate control mice to mark dividing chondro- cytes. At P4, P14, and 8 wk of age, we observed no difference in the percentage of Aggrecan+ chondrocytes in the growth plate that incor- porated EdU (SI Appendix, Fig. S1I and Fig. 1 C, D, and F). However, at 4 wk of age, significantly fewer growth plate chondrocytes were EdU+ in Osteolectin-deficient as compared to control mice (SI Appendix, Fig. S1I and Fig. 1E). Cell death appeared to be rare in growth plate chondrocytes at all stages as we were unable to find any TUNEL+ cells (25) in the growth plates of either Osteolectin-deficient or control mice (SI Appendix, Fig. S2 G and H). At all stages (P4, P14, 4 wk, and 8 wk), the number of Aggrecan+ chondrocytes per millimeter of growth plate was similar in Osteolectin-deficient and sex-matched littermate control mice (SI Appendix, Fig. S1I and Fig. 1 G–J). Since growth plate chondrocytes differentiate into osteoblasts (26, 27), this raised the possibility that the decrease in chondrocyte proliferation in Osteolectin-deficient mice at 4 wk of age led to reduced osteogenesis and reduced bone elongation. A B 4W 8W C P4 D P14 E 4W F 8W G P4 H P14 I 4W J 8W Fig. 1. Osteolectin promoted growth plate chondrocyte proliferation and longitudinal bone growth in juvenile mice. (A and B) Body mass, femur length, and LS3 vertebra length in Osteolectin-deficient (Oln−/−) and sex-matched littermate control (Oln+/+) mice at 4 week (4W, A) and 8 week of age (8W, B). Each square/circle represents a different mouse (seven to nine mice per sex per age per genotype in five or six independent experiments per age). (C–F) The percentages of Aggrecan+ growth plate chondrocytes that were EdU+ at P4 (C), P14 (D), 4W (E), and 8W (F) of age. (G–J) Numbers of Aggrecan+ chondrocytes per mm of growth plate in Oln−/− and sex-matched littermate control (Oln+/+) mice at P4 (G), P14 (H), 4W (I), and 8W (J) of age (Three mice per sex per genotype per age in three independent experiments per age for panels C–J). All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using Student’s t tests followed by Holm–Sidak’s multiple comparisons test (femur length in A and LS3 vertebra length in B, and C–J), or two-way ANOVAs followed by Sidak’s multiple comparisons test (other panels in A and B). Integrin α11 Promotes Chondrocyte Proliferation and Bone Elongation. We tested whether the Osteolectin receptor, Integrin α11 (encoded by Itga11), regulates bone elongation by conditionally deleting Itga11 (19) using Prx1Cre. Prx1Cre recombines during fetal development in limb mesenchymal cells but not in axial skeleton cells (including vertebrae) (28). We compared femur bone parameters between Prx1Cre;Itga11fl/ fl mice and sex-matched Itga11fl/fl littermate controls at P4 and P14. Prx1Cre;Itga11fl/fl and Itga11fl/fl controls did not differ in body mass or femur length (SI Appendix, Fig. S3 A and B), or femur cortical (SI Appendix, Fig. S3 C and D) or trabecular bone parameters (SI Appendix, Fig. S3 E and F). Itga11 deficiency thus had no effect on the formation or growth of limb bones prior to P14. At 4 and 8 wk of age, Prx1Cre;Itga11fl/fl mice and sex-matched Itga11fl/fl littermate controls did not differ in body mass or vertebra length (Fig. 2 A and B) or femur cortical bone parameters (SI Appendix, Fig. S4 A and B); however, at 4 wk of age, there was a trend toward shorter femurs in Prx1Cre;Itga11fl/fl mice (Fig. 2A) and at 8 wk of age, femurs were significantly (3 to 4%) shorter in Prx1Cre;Itga11fl/fl as compared to control mice (Fig. 2B). At 4 wk of age, we observed no differences in femur trabecular bone parameters (SI Appendix, Fig. S4C) but at 8 wk of age, we observed significantly reduced femur trabecular bone volume, number, and thickness in 2 of 11 https://doi.org/10.1073/pnas.2220159120 pnas.org A B 4W 8W C P4 D P14 E 4W F 8W 100µm 100µm G P4 H P14 I 4W J 8W Fig. 2. Integrin α11 promoted growth plate chondrocyte proliferation and longitudinal bone growth in juvenile mice. (A and B) Body mass, femur length, and LS3 vertebra length in Prx1Cre;Itga11fl/fl and sex-matched littermate control (Itga11fl/fl) mice at 4 wk (A) and 8 wk (B) of age (Eight mice per sex per age per genotype in five independent experiments per age). (C–J) Prx1Cre;Itga11fl/ fl and littermate control (Itga11fl/fl) mice were administered pulses of EdU for 4 h (P4), 1 d (P14), or 2 d (4W and 8W). The percentages of Aggrecan+ growth plate chondrocytes that were EdU+ at P4 (C), P14 (D), 4W (E), and 8W (F) of age. Numbers of Aggrecan+ chondrocytes per mm of growth plate in Prx1Cre;Itga11fl/fl and sex-matched littermate control (Itga11fl/fl) mice at P4 (G), P14 (H), 4W (I), and 8W (J) of age (Three mice per sex per genotype per age in three independent experiments per age for panels C–J). All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using Mann– Whitney tests followed by Holm–Sidak’s multiple comparisons test (femur length in A), or two-way ANOVAs followed by Sidak’s multiple comparison tests (B and body mass in A), or Student’s t tests followed by Holm–Sidak’s multiple comparison tests (LS3 vertebra length in A and C–J). Prx1Cre;Itga11fl/fl as compared to control mice (SI Appendix, Fig. S4D). Therefore, Integrin α11 is also required for bone elon- gation in juvenile mice. To test whether Integrin α11 promotes growth plate chondrocyte proliferation, we administered a 4-h to 2-d pulse of EdU to Prx1Cre;Itga11fl/fl and sex-matched littermate control mice. At P4, P14, and 8 wk of age, we observed no difference in the percentage of Aggrecan+ growth plate chondrocytes that incorporated EdU (Fig. 2 C, D, and F and SI Appendix, Fig. S3G). However, at 4 wk of age, significantly fewer growth plate chondrocytes were EdU+ in Prx1Cre;Itga11fl/fl as compared to control mice (Fig. 2E and SI Appendix, Fig. S3G). Cell death appeared to be rare in growth plate chondrocytes at all stages in Prx1Cre;Itga11fl/fl and control mice as we rarely observed TUNEL+ cells (SI Appendix, Fig. S4E). The number of Aggrecan+ chondrocytes per millimeter of growth plate was similar in Prx1Cre;Itga11fl/fl and control mice at all stages (Fig. 2 G–J). This suggested that Osteolectin/Integrin α11 signaling pro- motes bone elongation by increasing growth plate chondrocyte proliferation around 4 wk of age. Integrin α11 Acts Cell Autonomously in Chondrocytes to Promote Proliferation. To test whether Integrin α11 acted cell autonomously or noncell autonomously to promote the proliferation of growth plate chondrocytes, we conditionally deleted Itga11 using AggrecanCreER (AcanCreER), which recombines in chondrocytes (29). We initiated recombination at 2 wk of age by injecting AcanCreER;Itga11fl/fl mice and sex-matched Itga11fl/ fl littermate controls with tamoxifen and then analyzed bone parameters at 4 wk and 8 wk of age. At both ages, AcanCreER;Itga11fl/ fl mice and sex-matched Itga11fl/fl littermate controls did not differ in body mass (Fig. 3 A and B), femur cortical bone parameters (SI Appendix, Fig. S5 A and B), or femur or vertebra trabecular bone parameters (SI Appendix, Fig. S5 C–F). At 4 wk of age, there was a significant reduction or a trend toward shorter femurs and LS3 vertebrae in AcanCreER;Itga11fl/fl mice (Fig. 3A). At 8 wk of age, the AcanCreER;Itga11fl/fl mice had femurs and LS3 vertebrae that were significantly (4 to 6%) shorter than those in littermate controls (Fig. 3B). Integrin α11 thus acted within chondrocytes to promote bone elongation. To test whether Integrin α11 promoted bone elongation by promoting the proliferation of growth plate chondrocytes, we analyzed AcanCreER;Itga11fl/fl mice and sex-matched Itga11fl/fl litter- mate controls at 4 and 8 wk of age. At both ages, the number of Aggrecan+ chondrocytes per millimeter of growth plate was similar in AcanCreER;Itga11fl/fl and sex-matched littermate control mice (SI Appendix, Fig. S5 G–I). The percentage of Aggrecan+ chon- drocytes in the growth plate that incorporated a 2-d pulse of EdU was similar in AcanCreER;Itga11fl/fl and littermate controls at 8 wk of age (SI Appendix, Fig. S5G and Fig. 3D) but was lower in AcanCreER;Itga11fl/fl mice at 4 wk of age (SI Appendix, Fig. S5G and Fig. 3C). We did not detect any TUNEL+ chondrocytes in the growth plates of either AcanCreER;Itga11fl/fl or control mice at 4 or 8 wk of age (SI Appendix, Fig. S5J). Integrin α11 thus acted within growth plate chondrocytes to promote their proliferation, tran- siently increasing bone growth in juvenile mice. Although deficiency for either Osteolectin or Itga11 decreased the proliferation of growth plate chondrocytes at 4 wk of age (Figs. 1E and 3C), neither affected the number of Aggrecan+ chondrocytes per mm of growth plate (Fig. 1I and SI Appendix, Fig. S5H). Since growth plate chondrocyte proliferation lengthens bones as a result of the differentiation of chondrocytes into bone (26, 27), we wondered whether decreased chondrocyte proliferation was associated with reduced cortical bone generation. To test this, we generated AcanCreER; Rosa26loxp-tdTomato/+ and AcanCreER; Rosa26loxp-tdTomato/+; Itga11fl/fl mice to trace the bone formed by chondrocytes in the presence and absence of Integrin α11. We treated the mice with tamoxifen at 2 wk of age and analyzed Tomato expression at 8 wk of age (SI Appendix, Fig. S6A). To quantitate the rate at which chondrocytes generated cortical bone, we measured the length of cortical bone in sections that contained Tomato+ bone cells, starting at the growth plate (white lines in SI Appendix, Fig. S6A). AcanCreER; Rosa26loxp-tdTomato/+; Itga11fl/fl mice had a significantly shorter length of Tomato+ cortical bone as PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 3 of 11 A B 4W 8W C 4W D 8W E 8W F n a c e r g g A o t a m o T m - n O l H n a c e r g g A o t a m o T m - n O l 2W 100µm G 100µm 3W Epiphysis Growth Plate Metaphysis 4W Growth Plate Metaphysis I 8W Epiphysis 100µm Epiphysis 100µm Growth Plate Growth Plate Metaphysis Metaphysis Fig. 3. Integrin α11 cell-autonomously promoted the proliferation of growth plate chondrocytes. (A and B) AcanCreER;Itga11fl/fl and sex-matched littermate control (Itga11fl/fl) mice were treated with tamoxifen at 2 wk of age, then body mass, femur length, and LS3 vertebra length were measured at 4 wk (A) and 8 wk (B) of age. Each square/circle represents a different mouse (Six to eight mice per sex per age per genotype in four or five independent experiments per age). (C and D) The percentages of Aggrecan+ growth plate chondrocytes that were EdU+ at 4 wk (C) and 8 wk (D) of age (Three mice per sex per genotype per age in four or three independent experiments per age). (E) The length of cortical bone that arose from chondrocytes since tamoxifen treatment (Three mice per sex per genotype in three independent experiments). (F–I) Representative images of distal femur growth plate from OsteolectinmTomato/+ mice at 2 (F), 3 (G), 4 (H), or 8 (I) wk of age. Dotted white lines indicate the boundary between the growth plate and metaphysis or epiphysis. All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using two-way ANOVAs followed by Sidak’s multiple comparison tests (A and B), Student’s t tests followed by Holm– Sidak's multiple comparison tests (C and D), or Student’s t tests (E). compared to AcanCreER; Rosa26loxp-tdTomato/+ controls (Fig. 3E). This suggested that Integrin α11 signaling promoted bone elongation by increasing growth plate chondrocyte proliferation and cortical bone generation. We also tested whether AcanCreER recombined in perichondrial cells. Perichondrial cells contribute to bone formation in the dia- physis during fetal and early postnatal development (30). To test whether AcanCreER recombined in perichondrial cells, we treated 2-wk-old AcanCreER; Rosa26loxp-tdTomato/+ mice with tamoxifen and then assessed Tomato expression 2 d later. Acan-CreER recom- bined in all or nearly all growth plate chondrocytes but only in a very small number of Periostin+ perichondrial cells (31) (SI Appendix, Fig. S6B). Within the diaphysis, where perichondrial cells give rise to cortical bone, we observed no recombination by AcanCreER (SI Appendix, Fig. S6C). The recombination of Acan-CreER in perichondrial cells, thus, cannot explain the bone growth phenotype we observed. To test whether there is a source of Osteolectin near the growth plate, we examined OsteolectinmTomato/+ reporter mice (16). At 3 and 4 wk of age, we observed Tomato expression by bone lineage cells in the metaphysis and epiphysis, adjacent to the growth plate, as well as by chondrocytes within the growth plate (Fig. 3 G and H). We observed much less Tomato expression within the growth plate at 2 wk of age (Fig. 3F), and Tomato expression within the growth plate appeared to decline between 4 and 8 wk of age (Fig. 3I). Osteolectin expression thus increased within the growth plate at 3 to 4 wk of age, when it increased the proliferation of growth plate chondrocytes. This transient increase in Osteolectin expression within the growth plate at 3 to 4 wk of age may be part of the reason why bone elon- gation occurs more rapidly in juvenile as compared to adult mice. Recombinant Osteolectin Promotes Bone Elongation. With the goal of testing whether increased Osteolectin levels are sufficient to promote bone elongation in vivo, we first identified an effective dose of recombinant Osteolectin by injecting 8-wk-old male wild- type mice with 50 µg/kg/d, 100 µg/kg/d, 200 µg/kg/d, or 400 µg/ kg/d Osteolectin for 1 mo. Control mice were injected daily with diluent (phosphate-buffered saline, PBS). None of these doses had any effect on body mass, femur length, or LS3 vertebra length (Fig. 4 A–C), consistent with the fact that bone elongation is largely complete by 8 wk of age (32). The lower doses (50 µg/ kg/d and 100 µg/kg/d) did not affect cortical bone parameters, but the higher doses (200 µg/kg/d and 400 µg/kg/d) significantly increased femur cortical bone volume and thickness (Fig. 4D). All doses significantly increased trabecular bone volume and number (Fig. 4E). The data thus indicated that 200 µg/kg/d of recombinant Osteolectin increased bone formation in vivo. those of controls (Fig. 4F). The To test the effect of Osteolectin on bone elongation, we injected 2- to 8-wk-old male and female wild-type mice with 200 µg/kg/d of recombinant Osteolectin or PBS. At 8 wk of age, the Osteolectin-treated and sex-matched littermate control mice did not differ in body mass or LS3 vertebra length (Fig. 4F). However, the Osteolectin-treated mice had femurs that were 3 to 4% longer than the Osteolectin-treated mice also had significantly increased trabecular bone volume and number and reduced trabecular spacing as com- pared to control femurs (Fig. 4G). The percentage of Aggrecan+ chon- drocytes in the growth plate that incorporated EdU was higher in the Osteolectin-treated as compared to control mice at 4 wk of age (Fig. 4H) but similar in the Osteolectin-treated and control mice at 8 wk of age (Fig. 4I). Osteolectin treatment also significantly increased the length of cortical bone formed by chondrocytes in AcanCreER; Rosa26loxp-tdTomato/+ mice (Fig. 4J; akin to the experiment femurs from 4 of 11 https://doi.org/10.1073/pnas.2220159120 pnas.org A D E F G B C 8W 8W H 4W I 8W J 8W Fig. 4. Daily injections of recombinant Osteolectin increased bone formation in mice. (A–E) Eight-week-old male wild-type mice were administered daily injections of PBS or recombinant Osteolectin (Oln) at 50 µg/kg/d (50), 100 µg/kg/d (100), 200 µg/kg/d (200), or 400 µg/kg/d (400) for 4 wk. Oln injections did not affect body mass (A), femur length (B), LS3 vertebra length (C), but 200 or 400 µg/kg/d significantly increased cortical bone volume/ total volume and cortical thickness in the mid-femur diaphysis (D). Oln also significantly increased trabecular bone volume/total volume, connectivity density, number, and reduced trabecular spacing in the distal femur metaphysis (E) (Five mice per group in three independent experiments). (F and G) Wild-type mice were administered daily injections of PBS or recombinant Osteolectin (Oln) from 2 to 8 wk of age, then body mass, femur length, and LS3 vertebra length were measured at 8 wk of age (F), and Oln injection significantly increased trabecular bone volume/total volume and trabecular number and reducing trabecular spacing (G) (Six mice per sex per group in three independent experiments). (H and I) Percentages of Aggrecan+ growth plate chondrocytes that incorporated a 2-d pulse of EdU at 4 (H) or 8 (I) wk of age (Three mice per sex per age per group in three or four independent experiments per age). (J) AcanCreER;Rosa26loxp-tdTomato/+ mice were administered daily injections of PBS or Oln from 2 to 8 wk of age, then the length of cortical bone that arose from chondrocytes since tamoxifen treatment was measured as in SI Appendix, Fig. S6A (Three mice per sex per group in three independent experiments). All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using one-way ANOVAs followed by Dunnett’s multiple comparison tests (A–E), Welch’s t test followed by Holm–Sidak’s multiple comparison test (femur length in F), Student’s t tests followed by Holm–Sidak’s multiple comparison tests (trabecular BV/TV and trabecular connectivity density in G–I), Mann–Whitney tests followed by Holm–Sidak’s multiple comparison tests (trabecular spacing in G), and two-way ANOVAs followed by Sidak’s multiple comparison tests (body mass and LS3 vertebra length in F, and other panels in G), or Student’s t tests (J). in Fig. 3E). Recombinant Osteolectin is thus sufficient to promote bone elongation in juvenile mice and appeared to do so by increas- ing the proliferation of growth plate chondrocytes. Osteolectin/Integrin α11 Activates the Wnt Pathway in Chondrocytes. Wnt pathway activation promotes the proliferation and osteogenic differentiation of chondrocytes (33). To test whether Osteolectin/Integrin α11 signaling increases Wnt pathway activation in chondrocytes, we cultured growth plate chondrocytes from 4-wk-old wild-type mice. The Osteolectin-treated cells had increased levels of total β-catenin (Fig. 5A) and active β-catenin (unphosphorylated at GSK-3-dependent sites including Ser33, Ser37, and Thr41) (SI Appendix, Fig. S7A), and Wnt target genes, including Alpl (34), Lef1 (35), and Runx2 (36) (Fig. 5B), as compared to control chondrocytes. The Osteolectin-treated chondrocytes also exhibited increased EdU incorporation as PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 5 of 11 A C G I L N B D E PBS Oln F J K H M O Fig. 5. Osteolectin/Integrin α11 signaling promoted Wnt pathway activation and proliferation in chondrocytes. (A–F) Growth plate chondrocytes from 4-wk-old male wild-type mice were cultured with PBS or Oln for 2 d. β-catenin levels were quantified by western blot (A); Alpl, Lef1, and Runx2 transcripts were quantified by qRT-PCR (B); and EdU incorporation was quantitated by flow cytometry (C). Chondrocytes from these cultures were subcloned into secondary cultures at clonal density (100 cells per well in six-well dishes), and the percentage of cells that formed colonies (D) as well as colony size was assessed (E and F) in the presence and absence of Oln (Three mice per group in panels A–E, 52 and 55 colonies per group in panel F, all in three independent experiments). (G–K) Growth plate chondrocytes from 4-wk-old male AcanCreER;Itga11fl/fl and littermate Itga11fl/fl control mice were cultured with 4-hydroxytamoxifen for 2 d to delete Integrin α11. Osteolectin expression by these cells was confirmed by western blotting of the culture medium (G). In the cells, β-catenin protein levels (H); Alpl, Lef1, and Runx2 transcript levels (I); and EdU incorporation were quantified (J). Chondrocytes from these cultures were subcloned into secondary cultures at clonal density and the number of cells per colony was counted (K) (Three mice per genotype in panels G–K, 48 and 43 colonies per genotype in panel K, all in three independent experiments). (L–O) Growth plate chondrocytes from 4-wk-old male Oln−/− and littermate control mice were cultured for 2 d and then β-catenin protein levels (L); Alpl, Lef1, and Runx2 transcript levels (M); and EdU incorporation were quantified (N). Chondrocytes from these cultures were subcloned into secondary cultures at clonal density and the number of cells per colony was counted (O) (Three mice per genotype in panels L–N, 44 and 43 colonies per genotype in panel O, all in three independent experiments). All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using paired t tests (A, C, D, H, J, L, and N), paired sample two-way ANOVAs followed by Sidak’s multiple comparisons test (B, I, and M), or Student’s t tests (F, K, and O). compared to control cells (Fig. 5C). When the chondrocytes were subcloned into secondary cultures at clonal density, Osteolectin did not affect the percentage of cells that formed colonies (Fig. 5D) but increased the size of colonies (Fig. 5 E and F). Osteolectin did promote osteogenic differentiation by chondrocytes cultured in osteogenic differentiation medium (SI Appendix, Fig. S7B). Osteolectin thus promotes Wnt pathway activation, proliferation, and osteogenic differentiation in chondrocytes. Bone marrow stromal cells secrete Osteolectin into the culture medium and it promotes their differentiation into osteoblasts (19). To test whether chondrocytes secrete Osteolectin that promotes their proliferation, growth plate chondrocytes from 4-wk-old AcanCreER;Itga11fl/fl and sex-matched Itga11fl/fl littermate control mice were cultured with 4-hydroxytamoxifen for 2 d to delete Itga11. Western blotting showed the presence of Osteolectin in the culture medium (Fig. 5G). The Itga11-deficient cultures had significantly lower levels of total β-catenin (Fig. 5H); β-catenin unphosphorylated at GSK-3-dependent sites (SI Appendix, Fig. S7C); Alpl, Lef1, and Runx2 expression (Fig. 5I); EdU incor- poration (Fig. 5J); cells per colony (Fig. 5K); and osteogenic 6 of 11 https://doi.org/10.1073/pnas.2220159120 pnas.org differentiation (SI Appendix, Fig. S7D) as compared to control cultures. Chondrocytes thus secrete Osteolectin and Osteolectin/ Integrin α11 signaling in chondrocytes promotes Wnt pathway activation, proliferation, and osteogenic differentiation. Consistent with this, Osteolectin-deficient growth plate chondro- cytes had significantly lower levels of total β-catenin (Fig. 5L); unphosphorylated β-catenin (SI Appendix, Fig. S7E); Alpl, Lef1, and Runx2 expression (Fig. 5M); EdU incorporation (Fig. 5N); cells per colony (Fig. 5O); and osteogenic differentiation (SI Appendix, Fig. S7F) compared to control chondrocytes. To test whether reduced β-catenin (encoded by Ctnnb1) levels phenocopy Osteolectin or Itga11 deficiency, we treated AcanCreER;Ctnnb1fl/+ and sex-matched Ctnnb1fl/+ littermate control mice with tamoxifen at 2 wk of age and then analyzed them at 4 and 8 wk of age. At both ages, the AcanCreER;Ctnnb1fl/+ mice had femurs and LS3 vertebrae that were significantly (5 to 6%) shorter than those in littermate controls (SI Appendix, Fig. S7 G and H). AcanCreER;Ctnnb1fl/+ mice and littermate controls did not significantly differ in body mass, femur cortical bone volume, or femur or vertebral trabecular bone volume (SI Appendix, Fig. S7 G and H). The percentage of Aggrecan+ chondrocytes in the growth plate that incorporated a 2-d pulse of EdU was similar in AcanCreER;Ctnnb1fl/+ and littermate controls at 8 wk of age (SI Appendix, Fig. S7J) but was lower in AcanCreER;Ctnnb1fl/+ mice at 4 wk of age (SI Appendix, Fig. S7I). A quantitative reduction in b-catenin levels within growth plate chondrocytes, thus, tran- siently reduced chondrocyte proliferation and bone elongation, just as observed in Osteolectin and Itga11 mutant mice. Osteolectin/Integrin α11 signaling in growth plate chondrocytes thus appears to promote the expression of Wnt target genes by increasing b-catenin activity. The rs182722517 Variant Reduces Osteolectin Expression by hBMSCs. Separate genome-wide association studies in humans identified a single-nucleotide variant (rs182722517) that is associated with reduced height (22–24) and plasma Osteolectin levels (21). A B D H C E F G I Fig. 6. The rs182722517 variant reduced Osteolectin expression and osteogenic differentiation by human bone marrow stromal cells. (A) ATAC sequencing of primary human bone marrow stromal cells (hBMSCs) showing a peak of accessible chromatin at chr19:50739310, where the rs182722517 variant is observed (these cells contained the common allele). (B) ATAC sequencing in heterozygous rs182722517 variant–containing or control hBMSCs determined the ratio of reads from the CRISPR-edited allele versus the endogenous allele in each clone. This revealed a significantly lower fraction of reads from the rs182722517 variant– containing allele as compared to the CRISPR-edited control allele, suggesting that the rs182722517 variant reduces chromatin accessibility (Three independently targeted clones per genotype). (C and D) Osteolectin levels were assessed in cell extracts and in culture medium from cultures of homozygous rs182722517 variant–containing or control hBMSCs by western blot (C) or qRT-PCR (D) (Three replicate cultures per clone in three independent experiments). (E−G) Osteogenic differentiation of homozygous rs182722517–containing or control hBMSCs based on Alizarin Red staining in culture (E; Three replicates per clone from three independent experiments), or the area occupied by bone in sections through ossicles that grew in vivo. The sections were stained with hematoxylin and eosin (F) or trichome (G) to identify bone (Two replicates per clone from two independent experiments). (H) Osteolectin produced in, and around, the growth plate activates Integrin α11 signaling in growth plate chondrocytes, promoting chondrocyte proliferation and bone elongation (image generated using BioRender). (I) Schematic summarizing mechanism. All statistical tests were two sided. All data represent mean ± SD. Statistical significance was assessed using a Student’s t test (B) and nested t tests (C−G). PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 7 of 11 Rs182722517 is located at chr19:50739310 [GRCh38 (37)], 16 kb downstream of Osteolectin. The major allele at this locus is G, the rs182722517 variant has an A, and the rs182722517 variant frequency is 0.0032 (38). Rs182722517 is located within a candidate cis-regulatory element, ranging from chr19:50739241-50739456, that is annotated by ENCODE (Expanded Encyclopedias of DNA Elements) (39, 40). To test whether this represents a possible cis- regulatory element in hBMSCs, we performed ATAC sequencing on primary human bone marrow stromal cells (hBMSCs) bearing the major allele and detected a strong peak of accessible chromatin in the region surrounding the rs182722517 locus (Fig. 6A). This intergenic sequence differs in the mouse genome. To test whether the rs182722517 variant alters Osteolectin expression, we introduced this variant into wild-type hBMSCs using CRISPR/Cas9 editing (SI Appendix, Fig. S8A for the editing strategy). Prior to CRISPR editing, we confirmed that the hBMSCs carried the major allele at the chr19:50739310 locus (SI Appendix, Fig. S8B). To increase recombination efficiency, we introduced a C>G mutation in a protospacer adjacent motif (PAM, chr19:50739290) that pre- vented recutting of the already-recombined single-stranded oligo- deoxynucleotide (ssODN) sequence. We isolated three independently targeted hBMSC clones with homozygous G>A mutations at the chr19:50739310 locus and PAM site mutations, three independently targeted hBMSC clones with heterozygous G>A mutations at the chr19:50739310 locus and PAM site mutations, three control clones with homozygous PAM mutations, and three control clones with heterozygous PAM mutations (SI Appendix, Fig. S8B). To test whether the rs182722517 variant affects chromatin accessibility in hBMSCs, we performed ATAC sequencing on het- erozygous variant and control hBMSCs. The PAM mutation did not appear to affect chromatin accessibility as there were approxi- mately equal numbers of reads from the PAM mutation–containing allele and the wild-type allele in control cells that were heterozygous for the PAM mutation (Fig. 6B). Conversely, there were signifi- cantly fewer reads of the rs182722517 variant–containing allele than those of the wild-type allele in cells heterozygous for the rs182722517 variant (Fig. 6B). This suggests that the rs182722517 variant reduced chromatin accessibility in hBMSCs. To test whether the rs182722517 variant affects Osteolectin expression, we compared Osteolectin production by variant and control hBMSCs in culture. Clones that were homozygous for the rs182722517 variant had lower levels of Osteolectin in cell extracts and lower levels of Osteolectin secreted into the culture medium as compared to control clones (Fig. 6C). Osteolectin mRNA levels were also lower in the variant clones (Fig. 6D). The rs182722517 variant thus reduced Osteolectin expression by hBMSCs. We next tested the effect of the rs182722517 variant on osteogenesis by hBMSCs. As noted above, Osteolectin expres- sion by hBMSCs promotes osteogenic differentiation in cul- ture (19). After 21 d in osteogenic culture medium, homozygous variant clones exhibited significantly reduced Alizarin red levels as compared to control clones (Fig. 6E and SI Appendix, Fig. S8C). To test the effect of the rs182722517 variant on osteogenesis in vivo, we mixed homozygous variant or control cells with hydroxyapatite/tricalcium phosphate par- ticles and fibrin gel, then transplanted the suspensions sub- cutaneously into 3- to 4-mo-old NOD.CB17-Prkdcscid Il2rgtm1Wjl/ SzJ (NSG) mice and allowed them to form bony ossicles for 6 wk. The variant clones exhibited significantly reduced bone formation as compared to the control clones (Fig. 6 F and G and SI Appendix, Fig. S8 D and E). The rs182722517 variant thus reduced osteogenic differentiation by hBMSCs in vitro and in vivo. Discussion Osteolectin/Integrin α11 signaling promotes the maintenance of adult bone mass by promoting the differentiation of LepR+ skeletal stem cells into osteoblasts (18, 19). Consistent with this, PTH increases plasma Osteolectin levels in people who exhibit an oste- ogenic response to PTH (20). Moreover, Osteolectin deficiency reduces the osteogenic response to PTH in mice (20). This raised the possibility that Osteolectin mediates much of the effect of PTH on osteogenesis in humans; however, genetic evidence that Osteolectin regulates osteogenesis in humans was lacking. In this study, we found that Osteolectin/Integrin α11 signaling promotes longitudinal bone growth in juvenile mice by increasing Wnt pathway activation and proliferation in growth plate chondrocytes. In humans, the rs182722517 variant is associated with reduced height (22–24) and plasma Osteolectin levels (21), suggesting that Osteolectin promotes bone elongation in humans as well. The rs182722517 variant is located in a candidate cis-regulatory ele- ment (Fig. 6A) that influences chromatin accessibility in hBMSCs (Fig. 6B). The rs182722517 variant reduced Osteolectin expres- sion and osteogenic differentiation by hBMSCs (Fig. 6 C–G). Taken together, the data suggest that Osteolectin/Integrin α11 signaling promotes bone elongation in mice and humans, reveal- ing a function for Osteolectin/Integrin α11 as well as genetic evidence that this function is conserved among mice and humans. Osteolectin is expressed by LepR+ bone marrow stromal cells, osteoblasts, osteocytes, growth plate chondrocytes, and periosteal cells (Fig. 3 F–I). The bone elongation phenotype we report in this study is likely driven by Osteolectin that is expressed by chondro- cytes within the growth plate as well as by osteoblasts/osteocytes in the metaphysis/epiphysis, adjacent to the growth plate (Fig. 3 F–I). The Osteolectin produced in, and around, the growth plate activates Integrin α11 signaling in growth plate chondrocytes (Fig. 6H). Integrin α11 signaling promotes β-catenin activation, increasing the transcription of Wnt target genes, the division of growth plate chondrocytes, and bone elongation (Fig. 6I). These effects appear to be driven by an increase in Osteolectin expression within the growth plate at around 3 wk after birth (Fig. 3 F–I), increasing growth plate chondrocyte proliferation and bone elongation start- ing before 4 wk of age and ending before 8 wk of age. A transient increase in Osteolectin/Integrin α11 signaling in growth plate chondrocytes thus appears to be part of the reason why bone elon- gation is more rapid in juvenile as compared to adult mice. We previously reported that Osteolectin/Integrin α11 signaling in mouse LepR+ bone marrow stromal cell inhibits GSK3, promoting the accumulation of β-catenin and the transcription of Wnt target genes (19). The current study suggests that Osteolectin/Integrin α11 signaling promotes the proliferation and osteogenic differentiation of growth plate chondrocytes through a similar signaling mechanism (Fig. 5 and SI Appendix, Fig. S7). This is also consistent with studies by other groups who showed that increased Wnt pathway activation is necessary and sufficient to promote growth plate chondrocyte proliferation and bone formation (5, 36, 41). A limitation is that human growth plate chondrocytes bearing the rs182722517 variant were not available to us. Consequently, we had to study the effect of the rs182722517 variant on Osteolectin expression and osteogenic differentiation in human bone marrow stromal cells. Skeletal stem/progenitor cells (SSCs) that contribute to bone repair are present in the periosteum on the outside surface of bones (42–44) as well as inside the bone marrow (12, 45–47). We recently compared the functions of Gli1+ periosteal SSCs to LepR+ bone marrow SSCs and found that they make distinct contributions to the maintenance and repair of bones (47). LepR+ bone marrow SSCs 8 of 11 https://doi.org/10.1073/pnas.2220159120 pnas.org are responsible for steady-state osteogenesis and the repair of drill injuries in bone, while Gli1+ periosteal SSCs are responsible for the repair of bicortical fractures as well as the regeneration of LepR+ bone marrow stromal cells at fracture sites. The observation that Osteolectin is expressed in periosteal cells (16) and that Osteolectin deficiency slows fracture repair (18) raises the possibility that Osteolectin may promote osteogenesis by periosteal cells. This will be an important issue to address in future studies, given that past studies have focused on the effect of Osteolectin on LepR+ bone marrow SSCs. Materials and Methods Mice. All mouse experiments complied with all relevant ethical regulations and were performed according to protocols approved by the Institutional Animal Care and Use Committee at UT Southwestern Medical Center (UTSW; protocol 2017-101896). Human bone marrow stromal cells were transplanted into NOD. CB17-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice in ossicle formation assays (see details below). All other mice were maintained on a C57BL/Ka background, including Osteolectin−/− (Oln−/−) mice (18), Prx1Cre mice (28), Itga11flox mice (19), AggrecanCreER(AcanCreER) mice (29), Ctnnb1flox mice (48), Rosa26-CAG-loxp- stop-loxp-tdTomato (Rosaloxp-tdTomato) mice (49), and OsteolectinmTomato (OlnmT) mice (16). To induce Cre recombinase activity, AcanCreER mice were given three intraperitoneal injections of tamoxifen dissolved in corn oil at a dose of 100 mg/ kg body mass/injection over 5 d. Mouse body mass was measured using a Scout SPX123 Precision Balance (Ohaus). All mice were housed in AAALAC-accredited, specific pathogen-free animal care facilities at UTSW. Primers used to genotype mice are shown in the methods section of SI Appendix. In some experiments, recombinant mouse Osteolectin was administered to 2-mo-old wild-type male mice by daily subcutaneous injections for 1 mo, at a dose of 50 µg/kg/d, 100 µg/kg/d, 200 µg/kg/d, or 400 µg/kg/d. One hundred microliters of PBS was injected as vehicle control. Daily Osteolectin injections were also administered subcutaneously to wild-type mice from 2 to 8 wk of age at a dose of 200 µg/kg/d. Osteolectin was purified as previously described (18, 19) by affinity purification after secretion into the culture medium by Osteolectin overexpressing H293 cells. Generation and Culture of Genetically Engineered Human Bone Marrow Stromal Cells. Primary human bone marrow stromal cells (hBMSCs) were acquired from Lonza (PT-2501). The cells were cultured in DMEM (Gibco) with 15% fetal bovine serum (FBS) (Sigma, F0926) at 37 °C in a 5% CO2 incubator, and the medium was refreshed every 2 d. The cells were confirmed to carry the major allele G at the rs182722517 locus. To generate hBMSCs carrying the minor allele A at the variant site, two gRNAs were designed to target a site 22 base pairs upstream of the variant site. Two ssODNs were designed for recombination. The control ssODN carried a C>G mutation at the PAM to prevent recutting after the ssODN had recombined into the genome, thus increasing editing efficiency. The variant ssODN carried this PAM mutation as well as a G>A mutation at the vari- ant site. For transfection, cells were dissociated using TrypLE™ Express Enzyme (Gibco). Approximately 3 × 105z to 5 × 105 cells were transfected with Cas9 ribonucleoprotein complex (200 μM crRNA-tracrRNA complex + 15 μg Cas9) and ssODN (12 μg) following P1 Primary Cell 4D-Nucleofector™ X Kit Protocol FF-104. The sequences of gRNAs and ssODNs are listed below: gRNA1 was 5′-CAC TCA TGT GTC TCT AGT TA_TGG; gRNA2 was 5′-ACT CAT GTG TCT CTA GTT AT_GGG; control ssODN was 5′-TCA CGT TCT CCT CCA CCA TAG CAC AGA GCG TCT AAG GGT GCC ACC CTC TCG CAT AAC TAG AGA CAC ATG AGT GAC AGC AGC AAT GAG CTG TCC CAT CTG CTA GTC GTC GAC ACA GAA GAG C (PAM mutation site is in bold); and rs182722517 variant ssODN was 5′- TCA CGT TCT CCT CCA CCA TAG CAC AGA GCG TCT AAG GGT GCC ACC CTC TCG CAT AAC TAG AGA CAC ATG AAT GAC AGC AGC AAT GAG CTG TCC CAT CTG CTA GTC GTC GAC ACA GAA GAG C (the PAM and the variant mutations are in bold). Transfected cells were cultured at high density for 5 d and then single cells were sorted into 96-well plates and grown for 2 to 3 wk. To genotype the resulting colonies, genomic DNA was extracted using Quick-DNA 96 Kits (Zymo Research) and amplified by PCR using the following primers: forward 5′- CTT CAC GTA TT CAT TCA CGC A and reverse 5′- AAA GTG GAG TCG GTA GGT CA. The PCR products were then subjected to Sanger DNA sequencing using the sequencing primer: 5′- GTC AGA CGT TCA GTT AAG AAC TGC. Mouse Growth Plate Chondrocyte Cultures. Mouse growth plate chondro- cytes were dissected according to published protocols (50, 51), with modifica- tions described below. The cartilage cap of the femoral heads was removed from 4-wk-old mice using a blunt forceps. The cartilage caps were placed in 50 mL conical tubes containing PBS with penicillin and streptomycin (1:10 dilution of HyClone Penicillin-Streptomycin Solution from Cytiva). The cartilage caps were washed twice with PBS and then incubated in 5 mL pronase solution (Sigma Aldrich; 1 mg/mL pronase in DMEM with 5% FBS and 0.5% penicillin and strep- tomycin, sterilized by 0.22 µm filtration) in a 60-mm culture dish for 1 h at 37 °C in a 5% CO2 incubator. Pronase solution was removed, and the cartilage caps were washed once with PBS. They were then digested with liberase solution (Sigma Aldrich; 1 mg/mL liberase in DMEM with 1% FBS, sterilized by 0.22 µm filtration) for 6 h at 37 °C in a 5% CO2 incubator. Chondrocytes were liberated by pipetting the remaining cartilage fragments up and down a dozen times and filtering through a 40-µm cell strainer. The cells were then washed twice with PBS and pelleted by centrifuging at 500× g for 5 min. These cells were then cultured in complete medium (DMEM with 10% FBS plus 1% penicillin and streptomycin) overnight at 37 °C in a 5% CO2 incubator. Floating cells were removed, and viable chondrocytes were allowed to attach to the plastic surface of cell culture dish. Chondrocytes from 4-wk-old AcanCreER;Itga11fl/fl and Itga11fl/fl control mice were cultured with 200 nM 4-hydroxytamoxifen (Sigma Aldrich) in complete medium at 37 °C in a 5% CO2 incubator for 2 d, then washed with complete medium to remove 4-hydroxytamoxifen. The chondrocytes were cultured with 40 ng/mL recombinant Osteolectin (obtained as described below) in complete medium at 37 °C in a 5% CO2 incubator for 2 d. To quantitate proliferation, the chondrocytes were cultured with 10 µM EdU in complete medium for 24 h and then analyzed by flow cytometry. Alternatively, 2 d after 4-hydroxytamoxifen treatment, the chondrocytes were subcloned into secondary cultures at clonal density (100 cells/well in six-well culture dishes) in complete medium for 5 d, and then the colonies were imaged using a Leica DMi1 Inverted Microscope to count colony numbers and cells per colony. Chondrocytes were cultured at 10,000 cells per well in osteogenic differen- tiation medium (StemPro Osteogenesis Differentiation kit, Gibco) in 48-well tis- sue culture plates for 3 wk to induce osteogenic differentiation. These cultures were then stained with Alizarin Red S (EMD Millipore). To quantitate Alizarin red staining, the stained cells were rinsed with PBS and extracted with 10% (w/v) cetylpyridinium chloride in 10 mM sodium phosphate (pH 7.0) for 10 min at room temperature. Alizarin red in the extract was quantitated by optical density measurement at 562 nm. MicroCT Analysis. MicroCT analysis was performed using the settings described previously (18, 19). Mouse femurs were dissected, fixed overnight in 4% paraformal- dehyde (Thermo Fisher Scientific), and stored in 70% ethanol at 4 °C (52). The femurs and lumbar spine vertebrae were scanned at an isotropic voxel size of 3.5 μm and 7 μm, respectively, with a peak tube voltage of 55 kV and current of 0.145 mA (μCT 35; Scanco). A three-dimensional Gaussian filter (s = 0.8) with a limited, finite filter support of one was used to suppress noise in the images, and a threshold of 330 to 1,000 was used to segment mineralized bone from air and soft tissues. Trabecular bone parameters were measured in the femur distal metaphysis. The region was selected from below the distal growth plate where the epiphyseal cap structure completely disappeared and continued for 100 slices toward the proximal end of the femur. Contours were drawn manually a few voxels away from the endocortical surface to define trabecular bone in the metaphysis. Cortical bone parameters were measured by analyzing 100 slices in mid-diaphysis femurs. Vertebral trabecular bone parameters were measured by analyzing the vertebral body of the third lum- bar spine vertebra (LS3). The region started from the top where the vertebral body fully connects to the transverse process and continued 100 slices toward the bottom of the vertebra. For methods related to the immunostaining of bone sections, see methods in SI Appendix. Osteogenic Differentiation of Human Bone Marrow Stromal Cells. To assess osteogenic differentiation, hBMSCs cells were transferred into 48-well plates at confluent cell density (10,000 cells/well). On the second day after plating, the culture medium was replaced with osteogenic differentiation medium (StemPro Osteogenesis Differentiation kit, Gibco). The medium was replaced with fresh medium every 3 d. The cells were cultured for 21 d and their osteogenic differen- tiation was analyzed by staining with Alizarin red S (Sigma Aldrich). The cells were washed with PBS and then fixed in 4% paraformaldehyde (Thermo Fisher Scientific) PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 9 of 11 for 10 min. They were then washed with PBS twice and stained with Alizarin Red S for 15 min, followed by washing with PBS another three times. The cells stained with Alizarin Red S were imaged using a Leica DMi1 Inverted Microscope. To quantitate Alizarin red staining, the stained cells were rinsed with PBS and extracted with 10% (w/v) cetylpyridinium chloride in 10 mM sodium phosphate (pH 7.0) for 10 min at room temperature. Alizarin red in the extract was quantitated by optical density measurement at 562 nm. Ossicle Formation by Human Bone Marrow Stromal Cells. Bone ossi- cle formation by hBMSCs in vivo was assessed as described previously (53). hBMSC clones were cultured in osteogenic differentiation medium (StemPro Osteogenesis Differentiation kit, Gibco) for 5 d. Then 2 × 106 cells per clone were incubated with 40 mg hydroxyapatite/tricalcium phosphate particles (65%/35%, Zimmer Dental), rotating for 2 h at 37 °C. The cell/carrier slurry was centrifuged at 135× g for 5 min and embedded in a fibrin gel by adding 15 µL human fibrinogen (3.2 mg/mL in sterile PBS; EMD Millipore) with 15 µL human thrombin (25 U/mL in sterile 2% CaCl2 in PBS; EMD Millipore). The suspensions were left at room temperature for 10 min to clot and then trans- planted subcutaneously into NSG mice. After 6 wk in vivo, the bony ossicles formed by these cells were harvested and analyzed by cryosectioning of for- malin-fixed, paraffin-embedded specimens followed by hematoxylin and eosin (H&E) staining or trichrome staining. We analyzed three sections from each bony ossicle with a 200-μm distance between sections. The bone area in the sections was determined based on H&E or trichrome staining and morphology. The percentage of bone area in each section was imaged and quantified with Virtual Slide Scanner NanoZoomer 2.0HT (Hamamatsu) and analyzed with NDP. view2 software (Hamamatsu). ATAC Sequencing and Data Analysis. ATAC sequencing was performed based on a published protocol (54) with modifications described below. Briefly, 2 × 104 hBMSCs were washed twice in PBS and resuspended in 500 μL lysis buffer (10 mmol/L Tris-HCl, 10 mmol/L NaCl, 3 mmol/L MgCl2, 0.1% NP-40, pH 7.4). After cell lysis, the nuclei were pelleted by centrifugation at 500× g for 10 min at 4 °C. The nuclei were then resuspended in 50 μL tagmentation mix (Illumina) and incubated at 37 °C for 45 min. The tagmentation reaction was terminated by adding 10 μL 0.2% SDS and incubating at room temperature for 2 min and then at 55 °C for 7 min. TDE1 transposase–tagged DNA was purified using the QIAquick MinElute PCR Purification Kit (Qiagen) and amplified using KAPA HiFi Hotstart PCR Kit (KAPA) with Nextera DNA CD Indexes (Illumina). Libraries were quanti- tated using the double-stranded DNA High-Sensitivity Assay Kit (Invitrogen) on the Qubit fluorometer and the Agilent 2,200 TapeStation and were sequenced on an Illumina Nextseq500 using the 75 bp high-output sequencing kit. Wild-type hBMSCs were sequenced (56.8 M reads) as 75 bp single-end reads using an Illumina NextSeq 500. Control and variant hBMSCs were sequenced (61.8 M ± 4.0 M reads per sample) as 75 bp paired-end reads using an Illumina NextSeq 500. The quality of raw reads was checked using FastQC 0.11.8. Raw reads were trimmed using TrimGalore 0.6.4 and mapped to the Ensembl GRCh38 mouse reference genome version 100 using Bowtie 2.4.1. Mapped reads were quali- ty-filtered using SAMtools 1.12 to keep reads of MAPQ score > 10. In wild-type samples, 72% of raw reads with a >10 MAPQ score were processed by deepTools 3.5.1 to generate bigwig files, and IGV version 2.11.9 was used to browse them and generate panel Fig. 6A. In CRISPR-edited samples, 52.7 ± 1.7% of the raw reads with a >10 MAPQ score were processed by Bowtie2.4.1 to generate bam files, and IGV version 2.11.9 was used to browse them to quantify wild-type allele reads and mutant allele reads. In the six heterozygous control and variant clones, only the reads spanning both the PAM (Chr19:50739290) and the variant loci (Chr19:50739310) that had no mutations at both sites were counted as wild-type allele reads, the reads having only C>G mutations at the PAM were counted as control edited allele reads, and the reads having both C>G mutations at the PAM and G>A mutations at the rs182722517 variant locus were counted as variant allele reads. Total reads were quantitated as the sum of the wild-type reads and the edited allele reads in each clone, and the ratio of the mutant reads to total reads is presented in Fig. 6B. ATAC sequencing data have been submitted to the NCBI Sequence Read Archive, accession number: BioProject PRJNA887449. For methods related to quantitative PCR and western blots, see the methods section of SI Appendix. Statistical Analysis. In each type of mouse experiment, multiple mice were tested in multiple independent experiments performed on different days. Mice were allocated to experiments randomly and samples processed in an arbitrary order, but formal randomization techniques were not used. No formal blinding was applied when performing the experiments or analyzing the data. Sample sizes were not predetermined based on statistical power calculations but were based on our experience with these assays. No data were excluded. Prior to analyzing the statistical significance of differences among genotypes and treatments, we tested whether data were normally distributed and whether variance was similar among groups. To test for normality, we performed the Shapiro–Wilk tests when 3 ≤ n < 20 or D’Agostino Omnibus tests when n ≥ 20. To test whether variability significantly differed among groups, we performed F-tests (for experiments with two groups) or Levene’s median tests (for experi- ments with more than two groups). When the data significantly deviated from normality or variability significantly differed among groups, we log2-transformed the data and tested again for normality and variability. If the transformed data no longer significantly deviated from normality and equal variability, we performed parametric tests on the transformed data. If log2 transformation was not possible or the transformed data still significantly deviated from normality or equal varia- bility, we performed nonparametric tests on the nontransformed data. When data or log2-transformed data were normal and equally variable, statisti- cal analyses were performed using Student’s t tests/paired t tests (when there were two groups), nested t tests (when there were two groups and, in each group, sub- jects had multiple measurements), one-way ANOVAs (when there were more than two groups), and two-way ANOVAs/matched samples two-way ANOVAs (when there were two or more groups with multiple genotypes, treatments, or sexes). When the data or log2-transformed data were normal but unequally variable, statistical analyses were performed using Welch’s t tests (when there were two groups). When the data and log2-transformed data were abnormal or unequally variable, statistical analysis was performed using Mann–Whitney tests (when there were two groups). P-values from multiple comparisons were adjusted using Dunnett’s method after one-way ANOVAs (when comparisons were between a control and all other groups) or Sidak’s method after two-way ANOVAs (when there were more than two groups and planned comparisons). We only performed multiple post hoc tests when statis- tically significant differences were first observed by ANOVAs. Holm–Sidak’s method was used to adjust comparisons following multiple Student’s t tests, Mann–Whitney tests, or Welch’s t tests (when there were two groups with multiple genotypes, treatments, or sexes). All statistical tests were two sided. All data represent mean ± SD. Statistical tests were performed using GraphPad Prism V9.1.0. Data, Materials, and Software Availability. ATAC sequencing data have been submitted to the NCBI Sequence Read Archive, accession number: BioProject PRJNA887449. All study data are included in the article and/or SI Appendix. ACKNOWLEDGMENTS. S.J.M. is a Howard Hughes Medical Institute Investigator, the Mary McDermott Cook Chair in Pediatric Genetics, the Kathryn and Gene Bishop Distinguished Chair in Pediatric Research, the Director of the Hamon Laboratory for Stem Cells and Cancer, and a Cancer Prevention and Research Institute of Texas Scholar. This work was funded by the Josephine Hughes Sterling Foundation to S.J.M., by the Howard Hughes Medical Institute to S.J.M., and by the NIH grant DK111430 to J.X. We thank the Moody Foundation Flow Cytometry Facility for flow cytometry and Megan Mulkey for mouse colony management. We thank Mai Nguyen and Yunhee Cho from the Abcam Cell Engineering team for generating Crispr/Cas9-edited human bone marrow stromal cells. We thank John Shelton and Diana Wigginton from the Histopathology Core at University of Texas Southwestern Medical Center for H&E staining of bone specimens. We thank Denise Ramirez from the University of Texas Southwestern Medical Center Whole-Brain Microscopy Facility for imaging H&E-stained bone specimens. We thank Yoonjung Kim from the Children's Research Institute Sequencing Facility for sequencing. We thank the BioHPC high-performance computing cloud at the University of Texas Southwestern Medical Center for computational resources. Author affiliations: aChildren’s Research Institute, University of Texas Southwestern Medical Center, Dallas, TX 75390; bDepartment of Pediatrics, University of Texas Southwestern Medical Center, Dallas, TX 75390; and cHHMI, University of Texas Southwestern Medical Center, Dallas, TX 75390 10 of 11 https://doi.org/10.1073/pnas.2220159120 pnas.org 1. 2. 3. 4. C. Durand, G. A. Rappold, Height matters-from monogenic disorders to normal variation. Nat. Rev. Endocrinol. 9, 171–177 (2013). I. Gkiatas et al., Factors affecting bone growth. Am. J. Orthop. 44, 61–67 (2015). J. Baron et al., Short and tall stature: A new paradigm emerges. Nat. Rev. Endocrinol. 11, 735–746 (2015). Y. Yang, L. Topol, H. Lee, J. Wu, Wnt5a and Wnt5b exhibit distinct activities in coordinating chondrocyte proliferation and differentiation. Development 130, 1003–1015 (2003). 28. M. Logan et al., Expression of Cre recombinase in the developing mouse limb bud driven by a Prxl enhancer. Genesis 33, 77–80 (2002). 29. S. P. Henry et al., Generation of aggrecan-CreERT2 knockin mice for inducible Cre activity in adult cartilage. Genesis 47, 805–814 (2009). 30. H. M. Kronenberg, The role of the perichondrium in fetal bone development. Ann. N Y Acad. Sci. 1116, 59–64 (2007). 31. J. M. Brown et al., Periostin expression in neoplastic and non-neoplastic diseases of bone and joint. 5. M. Chen et al., Inhibition of beta-catenin signaling causes defects in postnatal cartilage Clin. Sarcoma Res. 8, 18 (2018). 6. 7. 8. development. J. Cell Sci. 121, 1455–1465 (2008). B. St-Jacques, M. Hammerschmidt, A. P. McMahon, Indian hedgehog signaling regulates proliferation and differentiation of chondrocytes and is essential for bone formation. Genes Dev. 13, 2072–2086 (1999). F. Long, X. M. Zhang, S. Karp, Y. Yang, A. P. McMahon, Genetic manipulation of hedgehog signaling in the endochondral skeleton reveals a direct role in the regulation of chondrocyte proliferation. Development 128, 5099–5108 (2001). K. Mizuhashi et al., Resting zone of the growth plate houses a unique class of skeletal stem cells. Nature 563, 254–258 (2018). 32. R. L. Jilka, The relevance of mouse models for investigating age-related bone loss in humans. J. Gerontol. A Biol. Sci. Med. Sci. 68, 1209–1217 (2013). 33. H. Chen et al., Molecular mechanisms of chondrocyte proliferation and differentiation. Front. Cell Dev. Biol. 9, 664168 (2021). 34. L. Topol et al., Wnt-5a inhibits the canonical Wnt pathway by promoting GSK-3-independent beta- catenin degradation. J. Cell Biol. 162, 899–908 (2003). 35. M. Filali, N. Cheng, D. Abbott, V. Leontiev, J. F. Engelhardt, Wnt-3A/beta-catenin signaling induces transcription from the LEF-1 promoter. J. Biol. Chem. 277, 33398–33410 (2002). 36. T. Gaur et al., Canonical WNT signaling promotes osteogenesis by directly stimulating Runx2 gene 9. H. S. Shu et al., Tracing the skeletal progenitor transition during postnatal bone formation. Cell Stem expression. J. Biol. Chem. 280, 33132–33140 (2005). Cell 28, 2122–2136 (2021). 37. V. A. Schneider et al., Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates 10. Y. Matsushita et al., The fate of early perichondrial cells in developing bones. Nat. Commun. 13, the enduring quality of the reference assembly. Genome Res. 27, 849–864 (2017). 7319 (2022). 38. C. Genomes Project et al., A global reference for human genetic variation. Nature 526, 68–74 11. L. Ding, T. L. Saunders, G. Enikolopov, S. J. Morrison, Endothelial and perivascular cells maintain (2015). haematopoietic stem cells. Nature 481, 457–462 (2012). 39. E. P. Consortium, Expanded encyclopaedias of DNA elements in the human and mouse genomes. 12. L. Ding, S. J. Morrison, Haematopoietic stem cells and early lymphoid progenitors occupy distinct Nature 583, 699–710 (2020). bone marrow niches. Nature 495, 231–235 (2013). 40. E. P. Consortium, An integrated encyclopedia of DNA elements in the human genome. Nature 489, 13. H. Oguro, L. Ding, S. J. Morrison, SLAM family markers resolve functionally distinct subpopulations of hematopoietic stem cells and multipotent progenitors. Cell Stem Cell 13, 102–116 (2013). 14. S. Comazzetto et al., Restricted hematopoietic progenitors and erythropoiesis require SCF from leptin receptor+ niche cells in the bone marrow. Cell Stem Cell 24, 477–486 (2019). 57–74 (2012). 41. D. Y. Dao et al., Cartilage-specific beta-catenin signaling regulates chondrocyte maturation, generation of ossification centers, and perichondrial bone formation during skeletal development. J. Bone Miner. Res. 27, 1680–1694 (2012). 15. B. O. Zhou, R. Yue, M. M. Murphy, J. G. Peyer, S. J. Morrison, Leptin-receptor-expressing 42. S. Debnath et al., Discovery of a periosteal stem cell mediating intramembranous bone formation. mesenchymal stromal cells represent the main source of bone formed by adult bone marrow. Cell Stem Cell 15, 154–168 (2014). Nature 562, 133–139 (2018). 43. O. Duchamp de Lageneste et al., Periosteum contains skeletal stem cells with high bone 16. B. Shen et al., A mechanosensitive peri-arteriolar niche for osteogenesis and lymphopoiesis. Nature regenerative potential controlled by Periostin. Nat. Commun. 9, 773 (2018). 591, 438–444 (2021). 44. L. C. Ortinau et al., Identification of functionally distinct Mx1+alphaSMA+ periosteal skeletal stem 17. B. O. Zhou et al., Bone marrow adipocytes promote the regeneration of stem cells and cells. Cell Stem Cell 25, 784–796 (2019). haematopoiesis by secreting SCF. Nat. Cell Biol. 19, 891–903 (2017). 45. C. K. Chan et al., Identification and specification of the mouse skeletal stem cell. Cell 160, 285–298 18. R. Yue, B. Shen, S. J. Morrison, Clec11a/osteolectin is an osteogenic growth factor that promotes the (2015). maintenance of the adult skeleton. ELife 5, e18782 (2016). 46. Y. Matsushita et al., A Wnt-mediated transformation of the bone marrow stromal cell identity 19. B. Shen et al., Integrin alpha11 is an osteolectin receptor and is required for the maintenance of orchestrates skeletal regeneration. Nat. Commun. 11, 332 (2020). adult skeletal bone mass. Elife 8, e42274 (2019). 20. J. Zhang et al., The effect of parathyroid hormone on osteogenesis is mediated partly by osteolectin. Proc. Natl. Acad. Sci. U.S.A. 118, e2026176118 (2021). 21. B. B. Sun et al., Genomic atlas of the human plasma proteome. Nature 558, 73–79 (2018). 22. J. R. Staley et al., PhenoScanner: A database of human genotype-phenotype associations. Bioinformatics 32, 3207–3209 (2016). 47. E. C. Jeffery, T. L. A. Mann, J. A. Pool, Z. Zhao, S. J. Morrison, Bone marrow and periosteal skeletal stem/progenitor cells make distinct contributions to bone maintenance and repair. Cell Stem Cell 29, 1547–1561 (2022). 48. V. Brault et al., Inactivation of the beta-catenin gene by Wnt1-Cre-mediated deletion results in dramatic brain malformation and failure of craniofacial development. Development 128, 1253–1264 (2001). 23. M. A. Kamat et al., PhenoScanner V2: An expanded tool for searching human genotype-phenotype 49. L. Madisen et al., A robust and high-throughput Cre reporting and characterization system for the associations. Bioinformatics 35, 4851–4853 (2019). whole mouse brain. Nat. Neurosci. 13, 133–140 (2010). 24. G. Kichaev et al., Leveraging polygenic functional enrichment to improve GWAS power. Am. J. Hum. 50. J. H. Jonason, D. Hoak, R. J. O’Keefe, Primary murine growth plate and articular chondrocyte Genet. 104, 65–75 (2019). isolation and cell culture. Methods Mol. Biol. 1226, 11–18 (2015). 25. W. Gorczyca, S. Bruno, R. Darzynkiewicz, J. Gong, Z. Darzynkiewicz, DNA strand breaks occurring 51. A. Haseeb, V. Lefebvre, Isolation of mouse growth plate and articular chondrocytes for primary during apoptosis - their early insitu detection by the terminal deoxynucleotidyl transferase and nick translation assays and prevention by serine protease inhibitors. Int. J. Oncol. 1, 639–648 (1992). cultures. Methods Mol. Biol. 2245, 39–51 (2021). 52. M. L. Bouxsein et al., Guidelines for assessment of bone microstructure in rodents using micro- 26. L. Yang, K. Y. Tsang, H. C. Tang, D. Chan, K. S. Cheah, Hypertrophic chondrocytes can become computed tomography. J. Bone Miner. Res. 25, 1468–1486 (2010). osteoblasts and osteocytes in endochondral bone formation. Proc. Natl. Acad. Sci. U.S.A. 111, 12097–12102 (2014). 53. P. Bianco, S. A. Kuznetsov, M. Riminucci, P. Gehron Robey, Postnatal skeletal stem cells. Methods Enzymol. 419, 117–148 (2006). 27. X. Zhou et al., Chondrocytes transdifferentiate into osteoblasts in endochondral bone during 54. X. Liu et al., In situ capture of chromatin interactions by biotinylated dCas9. Cell 170, 1028–1043 development, postnatal growth and fracture healing in mice. PLoS Genet. 10, e1004820 (2014). (2017). PNAS 2023 Vol. 120 No. 22 e2220159120 https://doi.org/10.1073/pnas.2220159120 11 of 11
10.1021_acschembio.3c00413	pubs.acs.org/acschemicalbiology This article is licensed under CC-BY 4.0 Articles Direct Modulators of K‑Ras−Membrane Interactions Johannes Morstein,* Rebika Shrestha, Que N. Van, César A. López, Neha Arora, Marco Tonelli, Hong Liang, De Chen, Yong Zhou, John F. Hancock, Andrew G. Stephen, Thomas J. Turbyville, and Kevan M. Shokat* Cite This: ACS Chem. Biol. 2023, 18, 2082−2093 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: Protein−membrane interactions (PMIs) are ubiq- uitous in cellular signaling. Initial steps of signal transduction cascades often rely on transient and dynamic interactions with the inner plasma membrane leaflet to populate and regulate signaling hotspots. Methods to target and modulate these interactions could yield attractive tool compounds and drug candidates. Here, we demonstrate that the conjugation of a medium-chain lipid tail to the covalent K-Ras(G12C) binder MRTX849 at a solvent-exposed site enables such direct modulation of PMIs. The conjugated lipid tail interacts with the tethered membrane and changes the relative membrane orientation and conformation of K-Ras(G12C), as shown by molecular dynamics (MD) simulation-supported NMR studies. In cells, this PMI modulation restricts the lateral mobility of K-Ras(G12C) and disrupts nanoclusters. The described strategy could be broadly applicable to selectively modulate transient PMIs. ■ INTRODUCTION Bifunctional molecules targeting biological interfaces are emerging therapeutic modalities that are undergoing a rapid expansion (e.g., PROTACS).1−4 To date, the majority of these strategies are focused on the modulation of protein−protein to target protein−membrane interactions, and methods interactions (PMIs) have remained relatively unexplored,5,6 despite their central importance in cellular signaling.7,8 Many targets in cancer signaling (e.g., Ras, PI3K, PKC, AKT) undergo transient and dynamic recruitment to the inner leaflet of the plasma membrane (PM), which could be susceptible to a relatively subtle pharmacological intervention. These targets include K-Ras4b (hereafter simply referred to as K-Ras), which is one of the most widely mutated cancer oncogenes.9−11 The lysine hypervariable region of K-Ras exhibits a patch of residues that aid in transiently associating K-Ras with the PM upon post-translational farnesylation. Inhibition of farnesyla- tion was extensively explored as a therapeutic strategy to inhibit K-Ras function but ultimately failed due to alternative rescued membrane attachment.12 More prenylation that recently, switch II pocket engagement has emerged as a direct strategy to covalently target the mutant allele K-Ras(G12C) giving rise to two clinically approved inhibitors sotorasib and adagrasib (Figure 1A).13−18 Moreover, this strategy has been translated to other mutant alleles K-Ras(G12S),19 K-Ras- (G12R),20 and K-Ras(G12D),21,22 this approach could be quite general. suggesting that by the C-terminal membrane anchor that consists of a farnesylated hexa-lysine polybasic domain. This anchor selectively associates with defined species of phosphatidylser- ine to form nanoclusters, comprising 4−6 K-Ras proteins.23−27 In addition, K-Ras diffusion is distinctive when compared to other paralogs, indicating that the lipid−protein environment that K-Ras explores is unique.11,28,29 Importantly, the specific lipid environment within K-Ras nanoclusters facilitates effector recruitment and activation.30−32 However, the precise mechanism underlying this PMI dependence in effector recruitment is currently unknown. New chemical tools that enable a precise modulation of these PMIs could therefore meet a critical need. Additionally, PMIs may present a therapeutic vulnerability that could be utilized in drug design. A number of monofunctional approaches to target PMIs have previously been reported for lipid clamp domains,8,33,34 and a screening hit for K-Ras with unique membrane-dependent behavior was found to modulate its PMIs in vitro.35 Herein, we attempted the rational design of bifunctional K- Ras(G12C) inhibitors with the capacity to directly modulate K-Ras−membrane interactions (Figure 1B). We envisioned Received: Accepted: Published: August 14, 2023 July 14, 2023 July 31, 2023 K-Ras’ association and interaction with plasma membrane lipids are essential for its function. K-Ras PMIs are mediated © 2023 The Authors. Published by American Chemical Society 2082 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 1. Design and synthesis of direct modulators for K-Ras−membrane interactions. (A) Scheme of direct Ras inhibition. (B) Scheme of a direct Ras inhibitor that simultaneously modulates its membrane interaction. (C) Crystal structure of K-Ras(G12C) in complex with MRTX849 (PDB 6UT0), highlighting the solvent exposed site of MRTX849.17 (D) Chemical structures of lipidated analogues of MRTX849, C5-MRTX, C11-MRTX, C18-MRTX, and the noncovalent control compound C11′-MRTX. (E) Synthesis of lipidated MRTX849 conjugates. that the installation of a second lipid tail on the surface of K- Ras would allow for modulation of PMIs. To this end, we the solvent-exposed site of proposed the modification of known covalent binders of K-Ras(G12C) with lipophilic groups. Effects on PMIs were characterized extensively in vitro and in cellulo. ■ RESULTS AND DISCUSSION Design and Synthesis of Lipid-Conjugated K-Ras- (G12C) the crystal structure of MRTX849 bound to K-Ras(G12C)17 (PDB 6UT0) revealed partial solvent exposure of the pyrrolidine fragment of the Inhibitors. Analysis of covalently bound ligand (Figure 1C).36 We envisioned that this site could be utilized to append lipophilic groups on the surface of K-Ras with the capacity to directly interact with the membrane. To this end, a series of lipid-conjugated MRTX849 analogues with varying lipid chain lengths were designed (Figure 1D). A small-chain lipid (SCL) conjugate with a 5- carbon containing tail (C5-MRTX), a medium-chain lipid (MCL) conjugate with a 11-carbon tail (C11-MRTX), and a long-chain lipid (LCL) conjugate with an 18-carbon containing tail (C18-MRTX) were synthesized. The control compound C11′-MRTX, which replaced the cysteine-reactive acrylamide warhead with a nonreactive saturated analogue was 2083 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 2. Biochemical and cell biological characterization of K-Ras(G12C) inhibitors. (A) LC/MS detection of covalent adducts of respective MRTX849-lipid conjugates to K-Ras(G12C) in vitro after 60 min. (B) Cellular target engagement using a TAMRA-Click assay.40 After 4 h incubation in H358 cells (G12C/WT), cells were harvested and incubated with TAMRA-azide to label the terminal end of lipids with a fluorophore. Pellets were blotted for RAS, and the shift of the upper band is indicative of cellular covalent engagement of K-Ras(G12C). (C) Dynamic light scattering measurement to determine critical aggregation concentration (CAC). Scattering intensity was plotted against logarithmic concentration. The origins of slope were used to identify the CAC as a starting point of aggregation. (D) Thermal stability shift assay using SYPRO Orange and covalently modified K-Ras(G12C) in vitro. (E) Cellular viability assay (CellTiter-Glo) of H358 cells with MRTX849, C5-MRTX, C11- MRTX, and C11′-MRTX (CTRL) after 72 h incubation at varying concentrations. also produced. All compounds were synthesized from a previously described MRTX849 intermediate37 and an N- functionalized prolinol derivative (Figure 1E). C11-MRTX is a Nonaggregating Potent Cellular Inhibitor of K-Ras(G12C). In vitro labeling of recombinant K-Ras(G12C) showed that C5-MRTX and C11-MRTX undergo rapid covalent modification of K-Ras(G12C), while the control compound C11′-MRTX and C18-MRTX do not label K-Ras(G12C) covalently (Figure 2A). To test if these results translate into cellular labeling of K-Ras(G12C), the alkyne moiety at the lipid terminus was utilized for copper- catalyzed azide-alkyne click chemistry38,39 leading to a shift in sodium dodecyl sulfate−polyacrylamide gel electrophoresis. Incubation of H358 (WT/G12C) cells with respective analogues of MRTX for 4 h, subsequent click labeling, SDS electrophoresis, and western blotting revealed effective cellular engagement of K-Ras(G12C) in cells by C5-MRTX and C11- MRTX as observed through a shift in SDS gel electrophoresis (Figure 2B) of the K-Ras band (note: H358 is a heterozygous cell line; partial labeling is observed due to the presence of a wildtype allele). Similar to our intact mass spectrometry experiments, covalent target engagement was not detected for C18-MRTX. We hypothesized that this could be due to an increased propensity of longer lipid tails to form aggregates, which was confirmed by a dynamic light scattering experiment (Figure 2C). Interestingly, the critical aggregation concen- tration of C5-MRTX was lower than that of MRTX849 and C11-MRTX was comparable to MRTX849. By contrast, C18- MRTX exhibited a much lower critical aggregation concen- tration (∼80 nM), which could be limiting its labeling efficiency and bioactivity. MRTX849 engages the switch II pocket of K-Ras(G12C) leading to a marked stabilization of its fold. To assess if our lipid conjugates behave similarly, we used a thermal shift assay with SYPRO Orange (Figure 2D). Notably, MRTX849-, C5- MRTX-, and C11-MRTX-labeled K-Ras(G12C) variants all showed a large thermal shift compared to nonlabeled K- Ras(G12C). At the same time, the shift between the three labeled variants exhibits no detectable differences, suggesting that the lipid tail does not strongly bind to K-Ras(G12C), which is desirable for it to potentially interact with the inner leaflet of the PM. To confirm that C11-MRTX exhibits specific cellular toxicity in K-Ras(G12C)-driven cancer cell lines, we performed a cell viability assay with C11-MRTX and the noncovalent control compound C11′-MRTX (CellTiter- Glo).13,15 C11-MRTX was found to be significantly more potent than the negative control compound (Figure 2E), which verifies that this inhibitor exhibits potent cellular activity despite the MCL conjugation. C11-MRTX Alters the Relative Conformation of K- Ras(G12C) on the PM. To study the capacity of C11-MRTX to modulate K-Ras−membrane interactions, we decided to employ coarse-grained MD simulations with Martini 3 force fields in conjunction with NMR paramagnetic relaxation enhancement (NMR-PRE) for K-Ras(G12C)·MRTX849 and K-Ras(G12C)·C11-MRTX that were chemically tethered to 2084 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 3. MD simulations and NMR-PRE experiments with membrane-tethered K-Ras(G12C). (A) Model of C11-MRTX modified K-Ras(G12C) tethered to a model membrane for MD simulations. (B) MD simulations revealed transient membrane engagement through the MCL of C11- MRTX leading to a novel bianchored conformation of K-Ras(G12C) on the membrane. (C) Ranking of membrane contacts of ligands for simulation with K-Ras(G12C)·MRTX849 (top) and K-Ras(G12C)·C11-MRTX (bottom). (D) Selected peaks from K-Ras(G12C/C118S)· MRTX849 and K-Ras(G12C/C118S)·C11-MRTX on nanodisks with and without the PRE tag Tempo. (E) Structure of K-Ras(G12C) highlighting areas that are moved close to the membrane when bound to C11-MRTX relative to MRTX849 in blue and moved further away in red. (F) NMR-PRE ratios for K-Ras·MRTX849 and K-Ras·C11-MRTX tethered to nanodisks. lipid nanodisks. MD simulations of K-Ras(G12C)·C11-MRTX (Figure 3A) revealed transient binding of the C11-MRTX MCL to the membrane leading to unusual bianchored conformations of K-Ras(G12C) (Figure 3B). To visualize the membrane contacts induced by C11-MRTX relative to ligand−membrane contacts were MRTX849, counted (Figure 3C), showing frequent membrane contacts for for C11-MRTX but near zero membrane contacts the direct 2085 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 4. Single-molecule tracking of K-Ras(G12C)·C11-MRTX in HeLa cells. (A) Schematic of HaloTag-tagged K-Ras used for TIRF single- particle tracking experiment. (B) TIRF image of JF549-chloroalkane labeled HaloTag K-Ras(G12C) in HeLa cells. (C) Representative trajectories of diffusion for labeled K-Ras(G12C) on the inner leaflet of the PM. Colors represent different single-molecule tracks over time. (D−F) Mean- square displacement plots calculated from the trajectories obtained for HaloTag K-Ras(G12C) labeled with 50 pM JF549 treated with no drug (black), 10 μM C11-MRTX (blue), 10 μM C11′-MRTX (green), and 10 μM MRTX849 (orange) for 30 min (D), 1 h (E), and 2 h (F) of compound incubation. In panels (E, F), the orange and green lines partially cover each other. MRTX849. To test these predictions experimentally, we tethered K-Ras(G12C) to nanodisks and conducted protein NMR studies. We observed marked chemical shift perturba- tions comparing K-Ras(G12C) bound to MRTX849 versus C11-MRTX. These shifts occurred on residues of SI, SII, and α3 regions (Figure 3D). Residue 63 from the switch II region had a particularly strong chemical shift response. This was consistent with the MD prediction of regions in K-Ras(G12C) moving into closer proximity of the membrane (marked in blue, Figure 3E). We further conducted NMR-PRE experi- ments which confirmed greater membrane proximity of residues 62 to 66 in the switch II region of K-Ras and an overall decrease in NMR-PRE ratios for β1, α3, and α4 residues (Figure 3F). K-Ras(G12C)·C11-MRTX had a longer rotational correlation time of 22.8 vs 18.4 ns for K-Ras(G12C)· MRTX849, which provided additional support for its closer membrane proximity (Figure S4). C11-MRTX Modulates the Diffusion of PM-Localized K-Ras(G12C) in Live Cells. To study if the PMI modulations observed in MD simulations and NMR experiments translate to live cells, we decided to study the lateral diffusion of labeled live cells.11 K- K-Ras(G12C) on the inner PM leaflet of internal Ras(G12C) diffusion was measured using total reflection microscopy (TIRF) employing a charge-couple 2086 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 5. Nanoclustering of K-Ras(G12C)·C11-MRTX in MDCK cells. (A) TEM image of 4.5 nm gold nanoparticles immunolabeling the GFP- tagged K-Ras(G12C) at a magnification of 100,000X. (B-D) Color-coded TEM images of the gold-labeled GFP-tagged K-Ras(G12C) treated with DMSO (B), C11-MRTX (C), and MRTX849 (D). (E, F) Analysis of PM localization and nanoclustering for GFP-K-Ras(G12C) (E) and GFP-K- Ras(G12D) (F). Error bars indicate mean ± SEM of the at least 15 PM sheet images for each condition. Bootstrap tests evaluated the statistical significance of the Lmax data, while one-way ANOVA calculated the statistical significance of the gold labeling data, with indicating p < 0.05. that device (CCD) camera for fast frame rate acquisition and a bright organic dye covalently linked to HaloTag K-Ras(G12C) overexpressed in HeLa cells (Figure 4A,B). The result demonstrates labeling of K-Ras(G12C) with C11- MRTX leads to marked changes in its dynamic diffusion along the PM. While no clear trends could be observed within 30 min (Figure 4C), C11-MRTX showed a marked reduction in diffusion rates compared to MRTX849 and C11′-MRTX after 1 h (Figure 4D) and further pronounced after 2 h (Figure 4E). We reasoned that the lateral restriction in K-Ras(G12C) mobility along the plasma membrane is a likely effect of the additional membrane contacts established by the C11-lipid tail. We further the subcellular distribution of K-Ras, for example by shifting its localization from the plasma membrane to endomembranes.41 Confocal imaging of GFP-fused K-Ras did not reveal alterations in the subcellular localization of K-Ras (Figure S5). tested if our molecules alter C11-MRTX Disrupts K-Ras(G12C) Nanoclusters. The spatial organization of K-Ras on the inner PM leaflet is critical for its physiological function. Transient nanoclusters were found to be the sites where effectors preferentially interact with K-Ras and are therefore especially critical for its physiological function.9,42 To test the lateral if C11-MRTX affects organization of K-Ras into nanoclusters, we conducted electron microscopy (EM) combined with spatial analysis43 in MDCK cells stably expressing GFP-K-Ras(G12C) or GFP- K-Ras(G12D) as control. Intact 2D PM sheets from cells treated with DMSO vehicle control, 10 μM C11-MRTX, or 10 μM MRTX849 for 2 h were fixed and labeled with 4.5 nm gold nanoparticles conjugated directly to anti-GFP antibody (Figure 5A). The gold particle spatial distributions were quantified using univariate K-functions expressed as L(r) − r. The maximum value of this function, Lmax, can be used as a summary statistic for the extent of nanoclustering. The extent of nanoclustering, L(r) − r, was plotted as a function of the length scale, r. The L(r) − r value of 1 is the 99% confidence the values above which indicate the statistically interval, meaningful clustering. Based on this K-function analysis, the EM images were color-coded to indicate the population distribution of the gold-labeled GFP-K-Ras(G12C). Larger L(r) − r values indicate more clustering (Figure 5B−D). We found that MRTX849 and C11-MRTX treatment both decreased the gold labeling density when compared with control, indicating that MRTX849 and C11-MRTX both reduced the localization of K-Ras(G12C) to the PM. C11- MRTX significantly reduced the Lmax value for GFP-K- Ras(G12C), indicating that C11-MRTX also disrupted the nanoclustering of K-Ras(G12C) (Figure 5E). Both MRTX849 and C11-MRTX had no effect on localization or nano- indicating selectivity for K- clustering of K-Ras(G12D), Ras(G12C) (Figure 5F). Combined, these data demonstrate the ability of the lipidated drug to selectively disrupt the lateral spatial organization of K-Ras(G12C) on the PM, which is function of GTP-bound K- critical Ras.30,31 the physiological for 2087 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles ■ CONCLUDING REMARKS leaflet of Herein, we report a bifunctional chemical approach to directly modulate the interactions between K-Ras(G12C) and the inner the PM. This is achieved through the installation of a C11 medium-chain lipophilic group to the solvent-exposed site of the covalent K-Ras(G12C) inhibitor MRTX849. Medium-chain lipids are common in natural products, occur in drugs (e.g., fingolimod or orlistat), and may present a sweet spot for bioactive amphiphiles due to their capacity to partition, while exhibiting a lower propensity to aggregate compared to longer membrane lipids.44 In our lead molecule C11-MRTX, the conjugated lipid tail establishes new interactions with the inner leaflet of the plasma membrane, resulting in novel bi-anchored conformations of membrane- tethered K-Ras(G12C). Thereby, the nucleotide binding site and switch I/II regions are brought in closer proximity to the PM, as demonstrated through a combination of MD simulations and NMR experiments. In cells, C11-MRTX restricts the lateral mobility of K-Ras which was observed through a marked reduction in diffusion rates. Finally, C11- MRTX was found to disrupt K-Ras(G12C) nanoclusters, which are the sites of Ras effector recruitment and activation and thus essential for signal transmission of noninhibited K- Ras. Combined, these results demonstrate a targeted modulation of protein−membrane interactions. These types of interactions are ubiquitous in early steps of cellular signaling, and our strategy could be translatable to target other signaling or lipid binding factors. PMI modulators could provide useful tools to dissect the function of these interactions and hold promise for the design of novel therapeutic agents. ■ MATERIALS AND METHODS General Methods. Anhydrous solvents were purchased from Acros Organics. Unless specified below, all chemical reagents were purchased from Sigma-Aldrich, Oakwood, Ambeed, or Chemscene. Analytical thin-layer chromatography (TLC) was performed using aluminum plates precoated with silica gel (0.25 mm, 60 Å pore size, 230−400 mesh, Merck KGA) impregnated with a fluorescent indicator (254 nm). TLC plates were visualized by exposure to ultraviolet light (UV). Flash column chromatography was performed with Teledyne ISCO CombiFlash EZ Prep chromatography system, employing prepacked silica gel cartridges (Teledyne ISCO RediSep). Proton nuclear magnetic resonance (1H NMR) spectra were recorded on a Bruker Avance III HD instrument (400/100/376 MHz) at 23 °C operating with the Bruker Topspin 3.1. NMR spectra were processed using Mestrenova (version 14.1.2). Proton chemical shifts are expressed in parts per million (ppm, δ scale) and are referenced to residual protium in the NMR solvent (CHCl3: δ 7.26, MeOD: δ 3.31). Data are represented as follows: chemical shift, multiplicity (s = singlet, d = doublet, t = triplet, q = quartet, dd = doublet of doublets, dt = doublet of triplets, m = multiplet, br = broad, app = apparent), integration, and coupling constant (J) in hertz (Hz). High-resolution mass spectra were obtained using a Waters Xevo G2-XS time-of-flight mass spectrometer operating with Waters MassLynx software (version 4.2). When liquid chromatography−mass spectrometry (LC−MS) analysis of the reaction mixture is indicated in the procedure, it was performed as follows. An aliquot (1 μL) of the reaction mixture (or the organic phase of a mini-workup mixture) was diluted with 100 μL 1:1 acetonitrile/water. 1 μL of the diluted solution was injected onto a Waters Acquity UPLC BEH C18 1.7 μm column and eluted with a linear gradient of 5−95% acetonitrile/water (+0.1% formic acid) over 3.0 min. Chromatograms were recorded with a UV detector set at 254 nm and a time-of-flight mass spectrometer (Waters Xevo G2-XS). Intact Protein Mass Spectrometry. Purified K-Ras variants (4 μM final) were incubated with compounds at 50 or 100 μM (1% v/v DMSO final) in 20 mM HEPES pH 7.5, 150 mM NaCl, 1 mM MgCl2 in a total volume of 150 μL. After the noted time, the samples were analyzed by intact protein LC/MS using a Waters Xevo G2-XS system equipped with an Acquity UPLC BEH C4 1.7 μm column. The mobile phase was a linear gradient of 5−95% acetonitrile/water + 0.05% formic acid. The spectra were processed using QuantLynx, giving the ion counts observed for the most abundant species. TAMRA-Click Assay. This assay was performed as previously described.40 Briefly, cells (500,000 to 1,000,000 cells per well) were seeded into six-well ultralow attachment plates (Corning Costar #3471) and allowed to incubate at 37 °C overnight. Cells were treated with the indicated concentrations of compound combinations and then incubated at 37 °C for the indicated lengths of time. In preparation for sodium dodecyl sulfate−polyacrylamide gel electro- phoresis (SDS−PAGE) and immunoblotting, cells were pelleted at 4 °C at 500 g and washed twice with ice-cold phosphate-buffered saline (PBS). Lysis was conducted, and copper-catalyzed click chemistry was performed by addition of the following to each lysate at the following final concentrations: 1% SDS (20% SDS in water stock), 50 μM TAMRA-N3 (5 mM in DMSO stock), 1 mM TCEP (50 mM in water stock), 100 μM TBTA (2 mM in 1:4 DMSO/t-butyl alcohol stock), and 1 mM CuSO4 (50 mM in water stock). After 1 h at room temperature, the reaction was quenched with 6× Laemmli sample buffer before SDS−PAGE. Dynamic Light Scattering. Measurements were performed using a DynaPro MS/X (Wyatt Technology) with a 55 mW laser at 826.6 nm, using a detector angle of 90°. Histograms represent the average of three data sets. Differential Scanning Fluorimetry. The protein of interest was diluted with HEPES buffer [20 mM HEPES 7.5, 150 mM NaCl, 1 mM MgCl2] to 2 μM. 1 μL of SYPRO Orange (500×) was mixed with 99 μL of protein solution. This solution was dispensed into wells of a white 96-well PCR plate in triplicate (25 μL/well). Fluorescence was measured at 0.5 °C temperature intervals every 30 s from 25 to 95 °C on a Bio-Rad CFX96 qPCR system using the FRET setting. Each data set was normalized to the highest fluorescence, and the normalized fluorescence reading was plotted against temperature in GraphPad Prism 8.0. Tm values were determined as the temper- ature(s) corresponding to the maximum (ma) of the first derivative of the curve. Cell Viability Assay. Cells were seeded into 96-well white flat bottom plates (1000 cells/well) (Corning) and incubated overnight. Cells were treated with the indicated compounds in a seven-point threefold dilution series (100 μL final volume) and incubated for 72 h. Cell viability was assessed using a commercial CellTiter-Glo (CTG) luminescence-based assay (Promega). Briefly, the 96-well plates were equilibrated to room temperature before the addition of diluted CTG reagent (100 μL) (1:4 CTG reagent/PBS). Plates were placed on an orbital shaker for 30 min before recording luminescence using a Spark 20M (Tecan) plate reader. Molecular Simulations. Coordinates of K-Ras bound to MRTX849 were downloaded from the pdb database (6UTO). Missing residues in the HVR were modeled using Modeller,45 as a disordered region.46 The protein was represented using the Martini 347 coarse-grained force field in combination with the structure- based48 approach in order to maintain its secondary structure. The farnesyl group was represented using parameters published before49 and updated in order to keep consistency with Martini 3. MRTX849, C11-MRTX, and GDP molecules were modeled using the method- ology published before,50 and bead types were updated accordingly to match the Martini 3 force field interaction matrix. Harmonic bonds were used to maintain the stability of the ligands in their respective binding regions, a methodology used successfully in the past.2 A membrane lipid bilayer composed of 70:30 POPC/POPS was constructed using the “insane” tool.51 Before insertion of K-Ras, the membrane was pre-equilibrated at 310 K for 100 ns. Protein and ligands were inserted, embedding the farnesyl group into the lipid bilayer and removing overlapping Martini water beads. Systems were charge-neutralized, and ions (Na+, Cl−) were added to mimic a 150 mM ionic strength environment. Before production, system boxes were energy-minimized and trajectories were saved every 2 ns for 2088 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles analysis. Each trajectory (2 total) was run for 30 μs. Simulations were carried out with GROMACS 2018.6,52 using a 20 fs time step for updating forces as recommended in the original publication. Reaction- field electrostatics was used with a Coulomb cutoff of 1.1 nm and dielectric constants of 15 or 0 within or beyond this cutoff, respectively. A cutoff of 1.1 nm was also used for calculating Lennard-Jones interactions, using a scheme that shifts the van der Waals potential to zero at this cutoff. Membranes were thermally coupled to 310 K using the velocity rescaling53 thermostat. Semi- isotropic pressure coupling was set for all systems at 1 bar using a Berendsen54 barostat with a relaxation time of 12.0 ps. DNA for Protein Production of K-Ras4b(1−185) G12C/ C118S. The gene for protein expression of Hs.K-Ras4b(1−185) initially G12C/C118S was generated from a DNA construct synthesized as a Gateway Entry clone (ATUM, Newark, CA). The construct consisted of an Escherichia coli gene-optimized fragment containing an upstream tobacco etch virus (TEV) protease site (ENLYFQ/G), followed by the coding sequence of human K- Ras4b(1−185). An entry clone was transferred to an E. coli destination vector containing an amino terminal His6-MBP (pDest- 566, Addgene #11517) tag by gateway LR recombination (Thermo Scientific, Waltham, MA). The construct generated was R949-x95- 566: His6-MBP-tev-Hs.K-Ras4b(1−185) G12C/C118S. The mem- brane scaffolding protein expression clone (pMSP delH5) was obtained from the group of Gerhard Wagner at Harvard University.55 Protein Expression and Purification. K-Ras4b(1−185) G12C/ C118S was expressed following the protocols described in Travers et al. for 15N/13C incorporation with modifications.49 Specifically, ZnCl2 was omitted and induction after IPTG addition was at 16 °C. Highly deuterated and 15N-labeled K-Ras protein was expressed using the protocols described in Chao et al.56 and purified essentially as outlined in Kopra et al.57 for K-Ras(1−169). pMSP delH5 was expressed and purified as described in Travers et al. NMR-PRE Sample Preparation and NMR Data Collection and Processing. Uniformly 15N/2H-labeled K-Ras(G12C/C118S) was first labeled with 2.5× excess of MRTX849 and C11-MRTX in 20 mM Hepes, pH 7.48, and 150 mM NaCl overnight at room temperature (∼11 h), and excess compounds were removed using a PD10 column equilibrated with 20 mM Hepes, pH 7.0, and 150 mM NaCl. Then, the MRTX849 and C11-MRTX bound K-Ras, concentration between 182 and 195 μM, were tethered to 2× excess of delH5 nanodisks composed of 63.75/30/6.25 POPC/POPS/PE MCC and 57.5/30/6.25/6.25 POPC/POPS/PE MCC/Tempo PC at room temperature overnight, followed by purification on an AKTA FPLC with a Superdex 200 Increase 10/300 column to remove nontethered K-Ras. All lipids were purchased from Avanti Polar Lipids. Empty delH5 nanodisks were made as described in Van et al. with pH 7.0 buffer.58 The final NMR buffer was 20 mM Hepes, pH 7.0, 150 mM NaCl, 0.07% NaN3, and 7.0% D2O. 280 μL of each sample was enclosed in 5 mm susceptibility-matched Shigemi tubes (Shigemi, Allison Park, PA) for NMR data collection. All NMR experiments were acquired on a Bruker AVANCE III HD spectrometer operating at 900 MHz (1H), equipped with a cryogenic triple-resonance probe. The temperature of the sample was regulated at 298 K throughout the experiments. Two-dimensional (2D) 1H,15N- TROSY-HSQC spectra were recorded with 1024 × 128 complex points for the 1H and 15N dimension, respectively, 128 scans, and a recovery delay of 1.5 s for a total collection time of 15 h. All 2D spectra were processed using NMRPipe59 and analyzed using NMRFAM-SPARKY.60 The NMR-PRE ratios were calculated from peak intensities and normalized to 1 (Figure 3F). Chemical shift perturbations (CSP) were calculated using CSP (ppm) = Sqrt((ΔN2/ 25 + ΔH2)/2) (Figure S1). The TROSY spectra for K-Ras·MRTX849 and K-Ras·C11-MRTX tethered to nanodisks without the Tempo PRE tag are shown in Figures S2 and S3, respectively. Expansion of the spectral region for residues 61 to 67 is shown in Figure 3D. To estimate the tumbling time of the K-Ras proteins in solution, 1H/15N-TRACT61 experiments were recorded as a series of one- dimensional (1D) spectra for the α and β states. For the 15N-α state, the relaxation delays were set to 0, 5, 10, 16, 22, 30, 40, 50, 64, 80, 100, 130, 170, and 240 ms. The relaxation delays for the faster- relaxing 15N-β state were set to 0, 1, 2, 4, 7, 11, 15, 20, 26, 32, 39, 47, 56, and 70 ms. Spectra for both the α and β states were recorded in a in an interleaved fashion. Each FID was single experiment accumulated for 1536 scans with a repetition delay between scans of 1.5 s for a total recording time of 18.5 h for both the α and β states. The interleaved spectra were separated in topspin using inhouse written scripts and analyzed using Mestrelab Research Mnova software. Plots showing the fits to calculate the rotational correlation time are shown in Figure S4. K-Ras·MRTX849 Backbone Chemical Shift Assignments. A sample of uniformly 13C,15N-labeled K-Ras bound to MRTX849 (6.4 mM in 20 mM Hepes, pH 7.0, with 150 mM NaCl, 1 mM MgCl2, 1 mM TCEP, 0.07% NaN3, and 7.0% D2O) was used to collect sequence-specific assignments of backbone resonances: two-dimen- sional (2D) 1H,15N-HSQC and three-dimensional (3D) HNCACB, 3D CBCA(CO)NH, 3D HNCA, 3D HN(CA)CO, 3D HNCO spectra, as well as a 3D NOESY 1H,15N-HSQC spectrum with a 100 ms mixing time. The 1H/15N assignments are shown in Figure S6. To increase the resolution of the C α cross-peaks in the 13C dimension of the 3D HNCA spectrum, band-selective shaped pulses (BADCOP) developed by optimal control theory were utilized to decoupled C α from C β nuclei.62 All NMR experiments were acquired on a Bruker AVANCE III HD spectrometer operating at 750 MHz (1H), equipped with a cryogenic triple-resonance probe. The temperature of the sample was regulated at 298 K throughout the experiments. All 3D spectra were recorded using nonuniform sampling (NUS) with sampling rates ranging between 30.5 and 33.3%. All spectra were processed using NMRPipe and analyzed in NMRFAM-SPARKY. The 3D spectra recorded with NUS were reconstructed and processed using the SMILE package available with NMRPipe. Single-Particle Tracking Experiments. HeLa cells were grown in Dulbecco’s modified Eagle medium (DMEM) (Thermo Fisher Scientific) supplemented with 1% 200 mM L-Glutamine and 10% FBS in a 6-well plate. The HaloTag fusion construct of K-Ras4b(G12C) was transiently transfected into each well using Fugene 6 transfection reagent (Promega) and 1.1 μg DNA per well. The protocol for plasmid design is described in Goswami et al.11 On the following day, cells were transferred on to plasma-cleaned coverslips (#1.5, 25 mm). On the day of imaging, the cells were first labeled with the fluorescent JF549 HaloTag ligand (Tocris) and then treated with the compounds. For labeling, the cells were first washed with 3 mL of PBS 3 times, incubated with 50 pM of JF54963 in complete media for 40 min, washed with 3 mL of PBS, and then allowed to recover in complete media for 30 min. For drug treatment, cells were first washed with 3 mL of PBS and then incubated with 10 μM of compound in complete media for the indicated time course. Single-particle tracking experiments were performed on the Nikon NStorm Ti-81 inverted microscope equipped with thermo-electric-cooled Andor iX EMCCD camera (Andor Technologies). During imaging experiments, the cells were maintained at 37 °C and 5% CO2 using a Tokai hit stage incubator (Tokai Hit Co., Ltd., Japan). The JF549 fluorescent molecules were illuminated under TIRF mode with the continuous 561 nm laser line from the Agilent laser module at 15% and imaged with an APO x100 TIRF objective with 1.49 NA (Nikon Japan). A 100 by 100-pixel region (16 × 16 μm2) of interest (ROI) was created in the cytoplasmic region of the PM in a cell and imaged at a frame rate of 10ms/frame for a total of 5000 frames. For each experiment, a minimum of 17 cells were imaged. Single-particle tracking movies were analyzed using the Localizer plugin embedded in Igor Pro software.64 Single particles in each frame were localized as spots based on the eight-way adjacency particle detection algorithm with a generalized likelihood ratio test (GLRT) sensitivity of 30 and a point spread function (PSF) of 1.3 pixels. The position of the PSF was estimated based on a symmetric 2D Gaussian fit function. If the particles persisted for more than 6 frames, they were then linked between consecutive frames into tracks. The particles were allowed a maximum jump distance of 5 pixels and blinking for one frame. For each experiment, tracks from all of the movies were combined into a single Matlab file and used to calculate mean-square displacement 2089 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles plots using a home-written script in Matlab. The plots were created using GraphPad Prism software. FLIM Imaging. In this study, we conducted fluorescence lifetime imaging (FLIM) experiments on doxycycline (Dox)-inducible eGFP- tagged K-Ras4b G12C HeLa cells. To generate the Dox-inducible cell line, HeLa cells (ATCC #CCL-2) were transduced with lentivirus containing the plasmid construct R733-M42-663 (TRE3Gp > eGFP- Hs.K-Ras4b G12C) at an MOI of 1.0. The cells were cultured in DMEM media supplemented with 10× L-Glutamine, 10% fetal bovine serum (complete media), 4 μg/mL of blastocydin, and 1 μg/mL of puromycin. Prior to imaging, the cell media was replaced with complete media containing doxycycline at a concentration of 500 ng/ mL, and drug treatment was administered at 10 μM for at least 2 h. FLIM imaging was performed using an Olympus Fluoview FV1000 inverted confocal microscope equipped with the Picoquant LSM upgrade kit and Picoharp 300 TCSPC module. A picosecond pulsed diode laser for the green channel (LDH-D-C-485) was used to illuminate the samples at a repetition rate of 40 MHz, allowing us to obtain the fluorescence lifetime decay curve. PicoQuant Symphotime 64 software was utilized for fluorescence lifetime fitting and image analysis. The fluorescence decay curve was fitted to a single- component n-Exponential tailfit to calculate the fluorescence lifetime for each pixel. The color scale on the right represents the fluorescence the mean lifetime of each pixel fluorescence lifetime of eGFP-K-Ras G12C was calculated to be approximately 2.6 ns, as depicted in green within the FLIM images.65,66 in the FLIM image. Notably, EM Spatial Analysis. MDCK cells stably expressing GFP-K- Ras(G12C) or GFP-K-Ras(G12D) were maintained in Dulbecco’s modified Eagle medium (DMEM) containing 10% fetal bovine serum (FBS). Cells were treated with DMSO, C11-MRTX, or MRTX849 at a concentration of 10 μM for 2 h, followed by preparation of the cell PM for electron microscopy (EM) analysis. An EM spatial distribution method is used to quantify the extent of K-Ras protein lateral spatial segregation in the inner leaflet of the PM.26,67 Gold grids with basal PM were prepared as described previously.30,68 Briefly, MDCK cells expressing GFP-tagged K-Ras mutants were grown on pioloform and poly-L-lysine-coated gold EM grids. After treatment, intact basal PM sheets attached to the gold grids were fixed with 4% paraformaldehyde and 0.1% glutaraldehyde, labeled with 4.5 nm gold nanoparticles coupled to anti-GFP antibody, and embedded in methyl cellulose containing 0.3% uranyl acetate. Distribution of gold particles on the basal PM sheets was imaged using a JEOL JEM- 1400 transmission electron microscope at 100,000× magnification. The EM images were analyzed using ImageJ software to assign x and y coordinates to gold particles in a 1 μm2 area of interest on the PM sheets. We use Ripley’s K-function to quantify the gold particle distribution and the extent of nanoclustering eqs A and B. (A) (B) where K(r) indicates the univariate K-function for the number of gold particles (n) within a selected area (A), r is the radius or length scale, \|\|·\|\| is the Euclidean distance, the indicator function 1(·) is assigned a −1 is the value of 1 if \|\|xi − xj\|\| ≤ r and a value of 0 otherwise, and wij proportion of the circumference of a circle with center at xi and a radius \|\|xi − xj\|\|. K(r) is linearly transformed to yield a parameter of L(r) − r, which is normalized on the 99% confidence interval (99% C.I.) using Monte Carlo simulations. The maximum value of the L(r) − r function Lmax provides a statistical summary for the extent of nanoclustering. For each treatment condition (DMSO, C11-MRTX, or MRTX849), at least 15 PM sheets were imaged, analyzed, and data pooled. Bootstrap tests were used to calculate the statistical significance of the nanoclustering data, while one-way ANOVA was used to estimate the statistical significance of the gold labeling density as previously described. ■ ASSOCIATED CONTENT *sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.3c00413. Chemical shift perturbation plot; plots of the alpha and beta state signal decay; experimental procedures; and compound characterization by high-resolution mass spectrometry (HRMS) and NMR (PDF) Final video (MP4) ■ AUTHOR INFORMATION Corresponding Authors Johannes Morstein − Department of Cellular and Molecular Pharmacology and Howard Hughes Medical Institute, University of California, San Francisco, California 94158, United States; Email: [email protected] orcid.org/0000-0002-6940-288X; Kevan M. Shokat − Department of Cellular and Molecular Pharmacology and Howard Hughes Medical Institute, University of California, San Francisco, California 94158, United States; Email: [email protected] orcid.org/0000-0001-8590-7741; Authors Rebika Shrestha − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Que N. Van − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States César A. López − Theoretical Biology and Biophysics Group, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, United States; orcid.org/0000-0003-4684-3364 Neha Arora − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States Marco Tonelli − National Magnetic Resonance Facility at Madison, Biochemistry Department, University of Wisconsin- Madison, Madison, Wisconsin 53706, United States Hong Liang − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States De Chen − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Yong Zhou − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States John F. Hancock − Department of Integrative Biology and Pharmacology, McGovern Medical School, University of Texas Health Science Center, Houston, Texas 77030, United States Andrew G. Stephen − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Thomas J. Turbyville − NCI RAS Initiative, Cancer Research Technology Program, Frederick National Laboratory for Cancer Research, Frederick, Maryland 21701, United States Complete contact information is available at: 2090 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles https://pubs.acs.org/10.1021/acschembio.3c00413 Notes The authors declare the following competing financial interest(s): K.M.S. and J.M. are inventors on patents owned by UCSF covering K-Ras targeting small molecules. K.M.S. has consulting agreements for the following companies, which involve monetary and/or stock compensation: Revolution Medicines, Black Diamond Therapeutics, BridGene Bioscien- ces, Denali Therapeutics, Dice Molecules, eFFECTOR Therapeutics, Erasca, Genentech/Roche, Janssen Pharmaceut- icals, Kumquat Biosciences, Kura Oncology, Mitokinin, Nested, Type6 Therapeutics, Venthera, Wellspring Biosciences (Araxes Pharma), Turning Point, Ikena, Initial Therapeutics, Vevo and BioTheryX. ■ ACKNOWLEDGMENTS J.M. thanks the NCI for a K99/R00 award (K99CA277358). K.M.S. thanks NIH grant 5R01CA244550 and the Samuel Waxman Cancer Research Foundation. The authors thank J. from B. Shoichet’s lab for assistance with the O’Connell dynamic light scattering (DLS) measurement. The authors wish to acknowledge C. J. DeHart, J.-P. Denson, P. H. Frank, M. Hong, S. Messing, A. Mitchell, N. Ramakrishnan, W. for cloning, protein Burgan, K. Powell, and T. Taylor expression, protein purification, cell line production, and electrospray ionization mass spectroscopy. The authors thank J. B. Combs, P. Pfaff, and D. M. Peacock for the critical review of the manuscript. The authors also thank Q. Zheng for providing optimized conditions for the Cbz-deprotection step. This project has been funded in whole or in part with Federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. 75N91019D00024. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services nor does the mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. This study made use of the National Magnetic Resonance Facility at Madison, which is supported by NIH grants P41GM136463 and R24GM141526. ■ REFERENCES (1) Alabi, S. B.; Crews, C. M. Major Advances in Targeted Protein Degradation: PROTACs, LYTACs, and MADTACs. J. Biol. Chem. 2021, 296, No. 100647. (2) Békés, M.; Langley, D. R.; Crews, C. M. PROTAC Targeted Protein Degraders: The Past Is Prologue. Nat. Rev. Drug Discovery 2022, 21, 181−200. (3) Schreiber, S. L. The Rise of Molecular Glues. Cell 2021, 184, 3− 9. (4) Kozicka, Z.; Thomä, N. H. Haven’t Got a Glue: Protein Surface Variation for the Design of Molecular Glue Degraders. Cell Chem. Biol. 2021, 28, 1032−1047. (5) Yin, H.; Flynn, A. D. Drugging Membrane Protein Interactions. Annu. Rev. Biomed. Eng. 2016, 18, 51−76. (6) Payandeh, J.; Volgraf, M. Ligand Binding at the Protein−Lipid Interface: Strategic Considerations for Drug Design. Nat. Rev. Drug Discovery 2021, 20, 710−722. (7) Vögler, O.; Barceló, J. M.; Ribas, C.; Escribá, P. V. Membrane Interactions of G Proteins and Other Related Proteins. Biochim. Biophys. Acta, Biomembr. 2008, 1778, 1640−1652. (8) Cho, W.; Stahelin, R. V. Membrane-Protein Interactions in Cell Signaling and Membrane Trafficking. Annu. Rev. Biophys. Biomol. Struct. 2005, 34, 119−151. (9) Zhou, Y.; Hancock, J. F. Ras Nanoclusters: Versatile Lipid-Based Signaling Platforms. Biochim. Biophys. Acta, Mol. Cell Res. 2015, 1853, 841−849. (10) Cox, A. D.; Der, C. J.; Philips, M. R. Targeting RAS Membrane Association: Back to the Future for Anti-RAS Drug Discovery? Clin. Cancer Res. 2015, 21, 1819−1827. (11) Goswami, D.; Chen, D.; Yang, Y.; Gudla, P. R.; Columbus, J.; Worthy, K.; Rigby, M.; Wheeler, M.; Mukhopadhyay, S.; Powell, K.; Burgan, W.; Wall, V.; Esposito, D.; Simanshu, D. K.; Lightstone, F. C.; Nissley, D. V.; McCormick, F.; Turbyville, T. Membrane Interactions of the Globular Domain and the Hypervariable Region of KRAS4b Define Its Unique Diffusion Behavior. eLife 2020, 9, No. e47654. (12) Pass, D. V. M. W. FTase Inhibition Holds Promise for RAS Targeting and Beyond. Cancer, 2018; 90, 2. (13) Ostrem, J. M.; Peters, U.; Sos, M. L.; Wells, J. A.; Shokat, K. M. K-Ras(G12C) Inhibitors Allosterically Control GTP Affinity and Effector Interactions. Nature 2013, 503, 548−551. (14) Ostrem, J. M. L.; Shokat, K. M. Direct Small-Molecule Inhibitors of KRAS: From Structural Insights to Mechanism-Based Design. Nat. Rev. Drug Discovery 2016, 15, 771−785. (15) Janes, M. R.; Zhang, J.; Li, L.-S.; Hansen, R.; Peters, U.; Guo, X.; Chen, Y.; Babbar, A.; Firdaus, S. J.; Darjania, L.; Feng, J.; Chen, J. H.; Li, S.; Li, S.; Long, Y. O.; Thach, C.; Liu, Y.; Zarieh, A.; Ely, T.; Kucharski, J. M.; Kessler, L. V.; Wu, T.; Yu, K.; Wang, Y.; Yao, Y.; Deng, X.; Zarrinkar, P. P.; Brehmer, D.; Dhanak, D.; Lorenzi, M. V.; Hu-Lowe, D.; Patricelli, M. P.; Ren, P.; Liu, Y. Targeting KRAS Mutant Cancers with a Covalent G12C-Specific Inhibitor. Cell 2018, 172, 578−589.e17. (16) Moore, A. R.; Rosenberg, S. C.; McCormick, F.; Malek, S. RAS- Targeted Therapies: Is the Undruggable Drugged? Nat. Rev. Drug Discovery 2020, 19, 533−552. (17) Fell, J. B.; Fischer, J. P.; Baer, B. R.; Blake, J. F.; Bouhana, K.; Briere, D. M.; Brown, K. D.; Burgess, L. E.; Burns, A. C.; Burkard, M. R.; Chiang, H.; Chicarelli, M. J.; Cook, A. W.; Gaudino, J. J.; Hallin, J.; Hanson, L.; Hartley, D. P.; Hicken, E. J.; Hingorani, G. P.; Hinklin, R. J.; Mejia, M. J.; Olson, P.; Otten, J. N.; Rhodes, S. P.; Rodriguez, M. E.; Savechenkov, P.; Smith, D. J.; Sudhakar, N.; Sullivan, F. X.; Tang, T. P.; Vigers, G. P.; Wollenberg, L.; Christensen, J. G.; Marx, M. A. the Clinical Development Candidate MRTX849, a Covalent KRASG12C Inhibitor for the Treatment of Cancer. J. Med. Chem. 2020, 63, 6679−6693. (18) Lanman, B. A.; Allen, J. R.; Allen, J. G.; Amegadzie, A. K.; Ashton, K. S.; Booker, S. K.; Chen, J. J.; Chen, N.; Frohn, M. J.; Goodman, G.; Kopecky, D. J.; Liu, L.; Lopez, P.; Low, J. D.; Ma, V.; Minatti, A. E.; Nguyen, T. T.; Nishimura, N.; Pickrell, A. J.; Reed, A. B.; Shin, Y.; Siegmund, A. C.; Tamayo, N. A.; Tegley, C. M.; Walton, M. C.; Wang, H.-L.; Wurz, R. P.; Xue, M.; Yang, K. C.; Achanta, P.; Bartberger, M. D.; Canon, J.; Hollis, L. S.; McCarter, J. D.; Mohr, C.; Rex, K.; Saiki, A. Y.; Miguel, T. S.; Volak, L. P.; Wang, K. H.; Whittington, D. A.; Zech, S. G.; Lipford, J. R.; Cee, V. J. Discovery of a Covalent Inhibitor of KRASG12C (AMG 510) for the Treatment of Solid Tumors. J. Med. Chem. 2020, 63, 52−65. (19) Zhang, Z.; Guiley, K. Z.; Shokat, K. M. Chemical Acylation of an Acquired Serine Suppresses Oncogenic Signaling of K-Ras(G12S). Nat. Chem. Biol. 2022, 18, 1177−1183. (20) Zhang, Z.; Morstein, J.; Ecker, A. K.; Guiley, K. Z.; Shokat, K. M. Chemoselective Covalent Modification of K-Ras(G12R) with a Small Molecule Electrophile. J. Am. Chem. Soc. 2022, 144, 15916− 15921. (21) Wang, X.; Allen, S.; Blake, J. F.; Bowcut, V.; Briere, D. M.; Calinisan, A.; Dahlke, J. R.; Fell, J. B.; Fischer, J. P.; Gunn, R. J.; Hallin, J.; Laguer, J.; Lawson, J. D.; Medwid, J.; Newhouse, B.; J. M.; Olson, P.; Pajk, S.; Rahbaek, L.; Nguyen, P.; O’Leary, Rodriguez, M.; Smith, C. R.; Tang, T. P.; Thomas, N. C.; Vanderpool, D.; Vigers, G. P.; Christensen, J. G.; Marx, M. A. Identification of MRTX1133, a Noncovalent, Potent, and Selective KRASG12D Inhibitor. J. Med. Chem. 2022, 65, 3123−3133. (22) Vasta, J. D.; Peacock, D. M.; Zheng, Q.; Walker, J. A.; Zhang, Z.; Zimprich, C. A.; Thomas, M. R.; Beck, M. T.; Binkowski, B. F.; Identification of 2091 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Corona, C. R.; Robers, M. B.; Shokat, K. M. KRAS Is Vulnerable to Reversible Switch-II Pocket Engagement in Cells. Nat. Chem. Biol. 2022, 18, 596−604. (23) Zhou, Y.; Hancock, J. F. RAS Nanoclusters Are Cell Surface Transducers That Convert Extracellular Stimuli to Intracellular Signalling. FEBS Lett. 2023, 597, 892−908. (24) Simanshu, D. K.; Philips, M. R.; Hancock, J. F. Consensus on the RAS Dimerization Hypothesis: Strong Evidence for Lipid- Mediated Clustering but Not for G-Domain-Mediated Interactions. Mol. Cell 2023, 83, 1210−1215. (25) Hancock, J. F. Ras Proteins: Different Signals from Different Locations. Nat. Rev. Mol. Cell Biol. 2003, 4, 373−385. (26) Prior, I. A.; Muncke, C.; Parton, R. G.; Hancock, J. F. Direct Visualization of Ras Proteins in Spatially Distinct Cell Surface Microdomains. J. Cell Biol. 2003, 160, 165−170. (27) Plowman, S. J.; Muncke, C.; Parton, R. G.; Hancock, J. F. H- Ras, K-Ras, and Inner Plasma Membrane Raft Proteins Operate in Nanoclusters with Differential Dependence on the Actin Cytoskele- ton. Proc. Natl. Acad. Sci. U.S.A. 2005, 102, 15500−15505. (28) Shrestha, R.; Chen, D.; Frank, P.; Nissley, D. V.; Turbyville, T. J. Recapitulation of Cell-like KRAS4b Membrane Dynamics on Complex Biomimetic Membranes. iScience 2022, 25, No. 103608. (29) Lee, Y.; Phelps, C.; Huang, T.; Mostofian, B.; Wu, L.; Zhang, Y.; Tao, K.; Chang, Y. H.; Stork, P. J.; Gray, J. W.; Zuckerman, D. M.; Nan, X. High-Throughput, Single-Particle Tracking Reveals Nested Membrane Domains That Dictate KRasG12D Diffusion and Trafficking. eLife 2019, 8, No. e46393. (30) Zhou, Y.; Prakash, P.; Liang, H.; Cho, K.-J.; Gorfe, A. A.; Hancock, J. F. Lipid-Sorting Specificity Encoded in K-Ras Membrane Anchor Regulates Signal Output. Cell 2017, 168, 239−251.e16. (31) Zhou, Y.; Wong, C.-O.; Cho, K.; van der Hoeven, D.; Liang, H.; Thakur, D. P.; Luo, J.; Babic, M.; Zinsmaier, K. E.; Zhu, M. X.; Hu, H.; Venkatachalam, K.; Hancock, J. F. Membrane Potential Modulates Plasma Membrane Phospholipid Dynamics and K-Ras Signaling. Science 2015, 349, 873−876. (32) Zhou, Y.; Liang, H.; Rodkey, T.; Ariotti, N.; Parton, R. G.; Hancock, J. F. Signal Integration by Lipid-Mediated Spatial Cross Talk between Ras Nanoclusters. Mol. Cell. Biol. 2014, 34, 862−876. (33) Singaram, I.; Sharma, A.; Pant, S.; Lihan, M.; Park, M.-J.; Pergande, M.; Buwaneka, P.; Hu, Y.; Mahmud, N.; Kim, Y.-M.; Cologna, S.; Gevorgyan, V.; Khan, I.; Tajkhorshid, E.; Cho, W. Targeting Lipid−Protein Interaction to Treat Syk-Mediated Acute Myeloid Leukemia. Nat. Chem. Biol. 2023, 19, 239−250. (34) Katti, S. S.; Krieger, I. V.; Ann, J.; Lee, J.; Sacchettini, J. C.; Igumenova, T. I. Structural Anatomy of Protein Kinase C C1 Domain Interactions with Diacylglycerol and Other Agonists. Nat. Commun. 2022, 13, No. 2695. (35) Fang, Z.; Marshall, C. B.; Nishikawa, T.; Gossert, A. D.; Jansen, J. M.; Jahnke, W.; Ikura, M. Inhibition of K-RAS4B by a Unique Mechanism of Action: Stabilizing Membrane-Dependent Occlusion of the Effector-Binding Site. Cell Chem. Biol. 2018, 25, 1327−1336.e4. (36) Bond, M. J.; Chu, L.; Nalawansha, D. A.; Li, K.; Crews, C. M. Targeted Degradation of Oncogenic KRASG12C by VHL-Recruiting PROTACs. ACS Cent. Sci. 2020, 6, 1367−1375. (37) Chu, L.; Crews, C. M.; Dong, H.; Hornberger, K. R.; Medina, J. R.; Snyder, L.; Wang, J. Compounds and Methods for Targeted Degradation of Kras. WO Patent, WO2021207172A12021. (38) Presolski, S. I.; Hong, V. P.; Finn, M. G. Copper-Catalyzed Azide−Alkyne Click Chemistry for Bioconjugation. Curr. Protoc. Chem. Biol. 2011, 3, 153−162. (39) Thirumurugan, P.; Matosiuk, D.; Jozwiak, K. Click Chemistry for Drug Development and Diverse Chemical−Biology Applications. Chem. Rev. 2013, 113, 4905−4979. (40) Lou, K.; Steri, V.; Ge, A. Y.; Hwang, Y. C.; Yogodzinski, C. H.; Shkedi, A. R.; Choi, A. L. M.; Mitchell, D. C.; Swaney, D. L.; Hann, B.; Gordan, J. D.; Shokat, K. M.; Gilbert, L. A. KRASG12C Inhibition Produces a Driver-Limited State Revealing Collateral Dependencies. Sci. Signal. 2019, 12, No. eaaw9450. (41) Schmick, M.; Vartak, N.; Papke, B.; Kovacevic, M.; et al. KRas Localizes to the Plasma Membrane by Spatial Cycles of Solubilization, Trapping and Vesicular Transport. Cell 2014, 157, 459−471. (42) Kattan, W. E.; Hancock, J. F. RAS Function in Cancer Cells: Translating Membrane Biology and Biochemistry into New Therapeutics. Biochem. J. 2020, 477, 2893−2919. (43) Zhou, Y.; Hancock, J. F. Electron Microscopy Combined with Spatial Analysis: Quantitative Mapping of the Nano-Assemblies of Plasma Membrane-Associating Proteins and Lipids. Biophys. Rep. 2018, 4, 320−328. (44) Morstein, J.; Capecchi, A.; Hinnah, K.; Park, B.; Petit-Jacques, J.; Van Lehn, R. C.; Reymond, J.-L.; Trauner, D. Medium-Chain Lipid J. Am. Conjugation Facilitates Cell-Permeability and Bioactivity. Chem. Soc. 2022, 144, 18532−18544. (45) Eswar, N.; Webb, B.; Marti-Renom, M. A.; Madhusudhan, M. S.; Eramian, D.; Shen, M.-Y.; Pieper, U.; Sali, A. Comparative Protein Structure Modeling Using Modeller Curr. Protoc. Bioinf. 2006, 15, DOI: 10.1002/0471250953.bi0506s15. (46) Ingólfsson, H. I.; Neale, C.; Carpenter, T. S.; Shrestha, R.; López, C. A.; Tran, T. H.; Oppelstrup, T.; Bhatia, H.; Stanton, L. G.; Zhang, X.; Sundram, S.; Di Natale, F.; Agarwal, A.; Dharuman, G.; Kokkila Schumacher, S. I. L.; Turbyville, T.; Gulten, G.; Van, Q. N.; Goswami, D.; Jean-Francois, F.; Agamasu, C.; Chen, D.; Hettige, J. J.; Travers, T.; Sarkar, S.; Surh, M. P.; Yang, Y.; Moody, A.; Liu, S.; Van Essen, B. C.; Voter, A. F.; Ramanathan, A.; Hengartner, N. W.; Simanshu, D. K.; Stephen, A. G.; Bremer, P.-T.; Gnanakaran, S.; Glosli, J. N.; Lightstone, F. C.; McCormick, F.; Nissley, D. V.; Streitz, F. H. Machine Learning−Driven Multiscale Modeling Reveals Lipid- Dependent Dynamics of RAS Signaling Proteins. Proc. Natl. Acad. Sci. U.S.A. 2022, 119, No. e2113297119. (47) Souza, P. C. T.; Alessandri, R.; Barnoud, J.; Thallmair, S.; Faustino, I.; Grünewald, F.; Patmanidis, I.; Abdizadeh, H.; Bruininks, B. M. H.; Wassenaar, T. A.; Kroon, P. C.; Melcr, J.; Nieto, V.; Corradi, V.; Khan, H. M.; Domański, J.; Javanainen, M.; Martinez- Seara, H.; Reuter, N.; Best, R. B.; Vattulainen, I.; Monticelli, L.; Periole, X.; Tieleman, D. P.; de Vries, A. H.; Marrink, S. J. Martini 3: A General Purpose Force Field for Coarse-Grained Molecular Dynamics. Nat. Methods 2021, 18, 382−388. (48) Poma, A. B.; Cieplak, M.; Theodorakis, P. E. Combining the MARTINI and Structure-Based Coarse-Grained Approaches for the Molecular Dynamics Studies of Conformational Transitions in Proteins. J. Chem. Theory Comput. 2017, 13, 1366−1374. (49) Travers, T.; López, C. A.; Van, Q. N.; Neale, C.; Tonelli, M.; Stephen, A. G.; Gnanakaran, S. Molecular Recognition of RAS/RAF Complex at the Membrane: Role of RAF Cysteine-Rich Domain. Sci. Rep. 2018, 8, No. 8461. (50) Potter, T. D.; Barrett, E. L.; Miller, M. A. Automated Coarse- the Martini Force Field and Grained Mapping Algorithm for Benchmarks for Membrane-Water Partitioning. J. Chem. Theory Comput. 2021, 17, 5777−5791. (51) Wassenaar, T. A.; Ingólfsson, H. I.; Böckmann, R. A.; Tieleman, D. P.; Marrink, S. J. Computational Lipidomics with Insane: A Versatile Tool for Generating Custom Membranes for Molecular Simulations. J. Chem. Theory Comput. 2015, 11, 2144−2155. (52) Páll, S.; Abraham, M. J.; Kutzner, C.; Hess, B.; Lindahl, E. Tackling Exascale Software Challenges in Molecular Dynamics Simulations with GROMACS. In Solving Software Challenges for Exascale; Markidis, S.; Laure, E., Eds.; Springer International Publishing: Cham, 2015; pp 3−27. (53) Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007, 126, No. 014101. (54) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola, A.; Haak, J. R. Molecular Dynamics with Coupling to an External Bath. J. Chem. Phys. 1984, 81, 3684−3690. (55) Hagn, F.; Etzkorn, M.; Raschle, T.; Wagner, G. Optimized Phospholipid Bilayer Nanodiscs Facilitate High-Resolution Structure Determination of Membrane Proteins. J. Am. Chem. Soc. 2013, 135, 1919−1925. 2092 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles (56) Chao, F.-A.; Dharmaiah, S.; Taylor, T.; Messing, S.; Gillette, W.; Esposito, D.; Nissley, D. V.; McCormick, F.; Byrd, R. A.; Simanshu, D. K.; Cornilescu, G. Insights into the Cross Talk between Effector and Allosteric Lobes of KRAS from Methyl Conformational Dynamics. J. Am. Chem. Soc. 2022, 144, 4196−4205. (57) Kopra, K.; Vuorinen, E.; Abreu-Blanco, M.; Wang, Q.; Eskonen, V.; Gillette, W.; Pulliainen, A. T.; Holderfield, M.; Härmä, H. Homogeneous Dual-Parametric-Coupled Assay for Simultaneous Nucleotide Exchange and KRAS/RAF-RBD Interaction Monitoring. Anal. Chem. 2020, 92, 4971−4979. (58) Van, Q. N.; López, C. A.; Tonelli, M.; Taylor, T.; Niu, B.; Stanley, C. B.; Bhowmik, D.; Tran, T. H.; Frank, P. H.; Messing, S.; Alexander, P.; Scott, D.; Ye, X.; Drew, M.; Chertov, O.; Lösche, M.; Ramanathan, A.; Gross, M. L.; Hengartner, N. W.; Westler, W. M.; Markley, J. L.; Simanshu, D. K.; Nissley, D. V.; Gillette, W. K.; Esposito, D.; McCormick, F.; Gnanakaran, S.; Heinrich, F.; Stephen, A. G. Uncovering a Membrane-Distal Conformation of KRAS Available to Recruit RAF to the Plasma Membrane. Proc. Natl. Acad. Sci. U.S.A. 2020, 117, 24258−24268. (59) Delaglio, F.; Grzesiek, S.; Vuister, G. W.; Zhu, G.; Pfeifer, J.; Bax, A. NMRPipe: A Multidimensional Spectral Processing System Based on UNIX Pipes. J. Biomol. NMR 1995, 6, 277−293. (60) Lee, W.; Tonelli, M.; Markley, J. L. NMRFAM-SPARKY: Enhanced Software for Biomolecular NMR Spectroscopy. Bioinfor- matics 2015, 31, 1325−1327. (61) Lee, D.; Hilty, C.; Wider, G.; Wüthrich, K. Effective Rotational Correlation Times of Proteins from NMR Relaxation Interference. J. Magn. Reson. 2006, 178, 72−76. (62) Coote, P. W.; Robson, S. A.; Dubey, A.; Boeszoermenyi, A.; Zhao, M.; Wagner, G.; Arthanari, H. Optimal Control Theory Enables Homonuclear Decoupling without Bloch-Siegert Shifts in NMR Spectroscopy. Nat. Commun. 2018, 9, No. 3014. (63) Grimm, J. B.; Muthusamy, A. K.; Liang, Y.; Brown, T. A.; Lemon, W. C.; Patel, R.; Lu, R.; Macklin, J. J.; Keller, P. J.; Ji, N.; Lavis, L. D. A General Method to Fine-Tune Fluorophores for Live- Cell and in Vivo Imaging. Nat. Methods 2017, 14, 987−994. (64) Dedecker, P.; Duwé, S.; Neely, R. K.; Zhang, J. Localizer: Fast, Accurate, Open-Source, and Modular Software Package for Super- resolution Microscopy. J. Biomed. Opt. 2012, 17, No. 126008. (65) Pliss, A.; Zhao, L.; Ohulchanskyy, T. Y.; Qu, J.; Prasad, P. N. Fluorescence Lifetime of Fluorescent Proteins as an Intracellular Environment Probe Sensing the Cell Cycle Progression. ACS Chem. Biol. 2012, 7, 1385−1392. J. P. Shifts in the (66) Li, W.; Houston, K. D.; Houston, Fluorescence Lifetime of EGFP during Bacterial Phagocytosis Measured by Phase-Sensitive Flow Cytometry. Sci. Rep. 2017, 7, No. 40341. (67) Diggle, P. J.; Mateu, J.; Clough, H. E. A Comparison between Parametric and Non-Parametric Approaches to the Analysis of Replicated Spatial Point Patterns. Adv. Appl. Probab. 2000, 32, 331−343. (68) Zhou, Y.; Hancock, J. F.Super-Resolution Imaging and Spatial Analysis of RAS on Intact Plasma Membrane Sheets. In Methods in Molecular Biology; Springer, 2021; Vol. 2262, pp 217−232. 2093 https://doi.org/10.1021/acschembio.3c00413 ACS Chem. Biol. 2023, 18, 2082−2093
10.1002_advs.202300445	RESEARCH ARTICLE www.advancedscience.com Cellular Composition and 5hmC Signature Predict the Treatment Response of AML Patients to Azacitidine Combined with Chemotherapy Guanghao Liang, Linchen Wang, Qiancheng You, Kirk Cahill, Chuanyuan Chen, Wei Zhang, Noreen Fulton, Wendy Stock, Olatoyosi Odenike,* Chuan He,* and Dali Han* Azacitidine (AZA) is a DNA methyltransferase inhibitor and epigenetic modulator that can be an eﬀective agent in combination with chemotherapy for patients with high-risk acute myeloid leukemia (AML). However, biological factors driving the therapeutic response of such hypomethylating agent (HMA)-based therapies remain unknown. Herein, the transcriptome and/or genome-wide 5-hydroxymethylcytosine (5hmC) is characterized for 41 patients with high-risk AML from a phase 1 clinical trial treated with AZA epigenetic priming followed by high-dose cytarabine and mitoxantrone (AZA-HiDAC-Mito). Digital cytometry reveals that responders have elevated Granulocyte-macrophage-progenitor-like (GMP-like) malignant cells displaying an active cell cycle program. Moreover, the enrichment of natural killer (NK) cells predicts a favorable outcome in patients receiving AZA-HiDAC-Mito therapy or other AZA-based therapies. Comparing 5hmC proﬁles before and after ﬁve-day treatment of AZA shows that AZA exposure induces dose-dependent 5hmC changes, in which the magnitude correlates with overall survival (p = 0.015). An extreme gradient boosting (XGBoost) machine learning model is developed to predict the treatment response based on 5hmC levels of 11 genes, achieving an area under the curve (AUC) of 0.860. These results suggest that cellular composition markedly impacts the treatment response, and showcase the prospect of 5hmC signatures in predicting the outcomes of HMA-based therapies in AML. 1. Introduction Acute myeloid leukemia (AML) is an ag- gressive malignancy characterized by a low cure rate and 5-year survival of 30– 35%. The signiﬁcant genetic and cellu- lar heterogeneity of AML contributes to highly variable responses to treatment.[1] Given that epigenetic aberrations arecom- monly observed and implicated in the pathogenesis of AML,[2] there has been an interest in combining hypomethylat- ing agents (HMAs), azacitidine (AZA), and decitabine, with cytotoxic chemotherapy, targeted therapy, or immunotherapy with the goal to improve outcomes for patients with AML.[3] While HMAs can reactivate aberrantly silenced genes, induce antivi- ral innate immune responses, and sen- sitize malignant cells to cytotoxic agents, the mechanism of their anti-leukemic ef- fect is not fully understood.[4] Recent stud- ies have highlighted the importance of malignant cell composition and immune landscape in determining the clinical out- comes of AML.[1c,d,5] However, the exact subsets of malignant cells and immune G. Liang, L. Wang, C. Chen, D. Han Key Laboratory of Genomic and Precision Medicine Beijing Institute of Genomics Chinese Academy of Sciences and China National Center for Bioinformation Beijing 100101, China E-mail: [email protected] G. Liang, L. Wang, C. Chen, D. Han College of Future Technology Sino-Danish College University of Chinese Academy of Sciences Beijing 100049, China The ORCID identiﬁcation number(s) for the author(s) of this article can be found under https://doi.org/10.1002/advs.202300445 © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. DOI: 10.1002/advs.202300445 Q. You, C. He Department of Chemistry and Institute for Biophysical Dynamics The University of Chicago Chicago, IL 60637, USA E-mail: [email protected] Q. You, C. He Howard Hughes Medical Institute Chicago, IL 60637, USA K. Cahill, N. Fulton, W. Stock, O. Odenike Section of Hematology/Oncology Department of Medicine University of Chicago Medicine Chicago, IL 60637, USA E-mail: [email protected] W. Zhang Department of Medicine University of California, San Diego La Jolla, CA 92093, USA Adv. Sci. 2023, 10, 2300445 2300445 (1 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com cells that determine the therapeutic response to these HMAs are unclear. We previously reported a phase 1 clinical trial of AZA treat- ment followed by high-dose cytarabine and mitoxantrone (AZA- HiDAC-Mito) in high-risk AML patients, based on the hypoth- esis that epigenetic priming with a HMA (AZA) would sensi- tize malignant cells to cytotoxic therapy.[6] The overall response rate [(complete remission (CR) + CR with incomplete count re- covery (CRi)] in this phase 1 study was 61% with a low induc- tion death rate of 2.2%. While AZA-HiDAC-Mito trended toward a higher response rate compared to a historical cohort treated with HiDAC-Mito alone,[6–7] the pre-treatment determinants and biomarkers for a treatment strategy including epigenetic priming remain unknown. Although gene expression and epigenetic proﬁling have yet to be adopted routinely in clinical practice, such approaches may help with prognostication and treatment decisions in AML.[8] Cytosine methylation (5mC) is a well-established epigenetic biomarker involved in cancer development and progression.[9] 5mC is maintained by DNA methyltransferases, while the TET family of dioxygenases convert 5mC to 5-hydroxymethylcytosine (5hmC) in an active demethylation process.[10] Increasing evi- dence suggests that 5hmC levels are related to tumorigenesis, in- cluding observations that global 5hmC levels are reduced in vari- ous cancer types.[2b,11] Furthermore, recent studies have demon- strated that AZA treatment aﬀects the cellular level and genomic distribution of 5hmC,[12] which could be used as a robust diag- nostic and prognostic biomarker for broad cancer types.[13] These studies support 5hmC as an ideal candidate for an epigenetic biomarker to predict the outcomes of AZA-HiDAC-Mito therapy. Herein, to elucidate the underlying mechanisms of treatment response for AZA-HiDAC-Mito therapy, we collected samples in a phase 1 clinical study and performed RNA-seq and 5hmC proﬁling. By combining the public single-cell RNA-seq data, we found that responders highly expressed cell-cycle-related genes, which were inferred to be expressed primarily by a subset of Granulocyte-macrophage-progenitor-like (GMP-like) malignant cells. In contrast, hematopoietic stem cell-like (HSC-like) malig- nant cells with low expression of cell-cycle-related genes were more likely to be enriched in non-responders. Moreover, we found that AZA treatment induced gene expression related to NK cell cytotoxicity in responders. In line with this, the pre-treatment W. Zhang Bristol-Myers Squibb San Diego, CA 92121, USA N. Fulton, W. Stock, O. Odenike Comprehensive Cancer Center University of Chicago Medicine Chicago, IL 60637, USA C. He Department of Biochemistry and Molecular Biology The University of Chicago Chicago, IL 60637, USA D. Han Institute for Stem Cell and Regeneration Chinese Academy of Sciences Beijing 100101, China level of NK cells was associated with improved clinical outcome of AZA-HiDAC-Mito therapy. By analyzing samples from patients receiving AZA treatment, compared to those receiving decitabine treatment or standard chemotherapy in the Beat AML cohort, and our historical HiDAC-Mito cohort, we demonstrated the spe- ciﬁc role of NK cells in response to AZA-based treatment. Fur- thermore, AZA exposure induced a dose-dependent alteration in 5hmC after treatment for ﬁve days, and patients with more pronounced changes in 5hmC modiﬁcations exhibited improved survival. We then developed a machine learning prediction model based on 5hmC levels in 11 genes, which accurately predicted treatment response. 2. Results 2.1. Activation of the Cell Cycle Program was Associated with Response to AZA-HiDAC-Mito Therapy A total of 46 patients who received AZA-HiDAC-Mito therapy were enrolled in this study, out of which 41 provided usable RNA and/or 5hmC sequencing data (Figure 1A; Table S1, Supporting Information). Of these patients, 19/46 (41%) achieved complete remission (CR) after treatment, with 9/46 (20%) diagnosed as CR but with incomplete count recovery (CRi), and 18/46 (39%) expe- rienced treatment failure (TF). The overall response rate was 61% (28/46). The results of this trial have been previously published.[6] To identify gene expression programs that may confer sensi- tivity to AZA-HiDAC-Mito treatment, we collected mononuclear cells from bone marrow (BM) and/or peripheral blood (PB) prior to AZA treatment (Day 0) and performed RNA-seq. Twenty-eight patients had pre-treatment material available for RNA-seq. By comparing the gene expression levels between responders (CR + CRi, n = 16) and non-responders (n = 12), we identiﬁed 54 upreg- ulated and 115 downregulated genes in responders (Figure 1B). Unsupervised clustering analysis revealed that these diﬀeren- tially expressed genes (DEGs) were capable of distinguishing re- sponders from non-responders (Figure 1C), suggesting that the transcriptional proﬁle was tightly associated with treatment re- sponse. To shed light on the potential mechanism(s) of treatment response, we performed an extended co-expression network analysis by integrating protein association networks from the STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) database.[14] Brieﬂy, we ﬁrst identiﬁed four gene modules through co-expression analysis, including RAMP3 (signaling receptor activity), EFNA5 (G2M checkpoint), FLT3 (hematopoiesis), EPHB1 (ephrin receptor) for module 1–4, respectively (Figure 1D). Genes within the same co-expression module are highly correlated and probably have similar biolog- ical functions. Next, we extended each gene module by adding ﬁrst-order neighbors in the STRING database to construct a functional network. Enrichment analysis revealed that genes in modules 3 and 4, which were downregulated in responders, are enriched for pathways known to be involved in tumorigen- esis and tumor progression, such as ephrin receptor signaling pathway (Figure 1E).[15] In contrast, for the two modules that were upregulated in responders, we observed that the functional network for module 1 is enriched for cell-cell interaction and Adv. Sci. 2023, 10, 2300445 2300445 (2 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 1. Diﬀerential expression analysis between responders and non-responders of AZA-HiDAC-Mito therapy. A) Diagram of study design, therapeutic strategy, and data analysis workﬂow. RNA and DNA obtained from peripheral blood and bone marrow samples of 46 AML patients receiving AZA-HiDAC- Mito therapy were used in this study (41 patients provided usable DNA/RNA samples). The ﬁgure in the table is the number of patients. B) Volcano plot showing gene expression diﬀerence between responders and non-responders. Thirty-three BM and/or PB samples obtained from 28 patients were used. p values were calculated with the Wald test and adjusted by the Benjamini-Hochberg method. padj, adjusted p value. Top 3 DEGs for both upregulated and downregulated genes were labeled. C) Heatmap showing the expression levels of 169 DEGs in 33 AML samples collected at Day 0. Hierarchical clustering was performed across genes and samples. D) Heatmap showing hierarchical clustering of the pairwise correlations among DEGs in 33 AML samples. DEGs were grouped into four major modules. E) Functional enrichment for genes in each module-related network. The module-related network was obtained from the STRING database by adding the directly interacting genes of the DEGs. The q value was adjusted p value by the Benjamini-Hochberg method. F) GSEA to assess the enrichment of cell cycle signature in responders of AZA-HiDAC-Mito therapy. NES, normalized enrichment score; p value was calculated with permutation test. communication, while genes within the module 2 network are involved in cell cycle and DNA synthesis. Since both AZA and cytarabine are known to interfere with DNA synthesis and preferentially eliminate cycling cells,[16] we reasoned that activation of the cell cycle program may be asso- ciated with response to AZA-HiDAC-Mito therapy. To this end, we performed gene set enrichment analysis (GSEA) to evaluate the enrichment of a curated cell cycle signature in responders compared to non-responders.[17] We found that expression of the cell cycle signature was highly enriched in responders, which Adv. Sci. 2023, 10, 2300445 2300445 (3 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com suggests that patients with activated cell cycle program are sensitive to AZA-HiDAC-Mito therapy (Figure 1F). like malignant subsets can serve as a predictive indicator of re- sponders to AZA-HiDAC-Mito therapy. 2.2. Elevated GMP-Like Malignant Cells with Active Cell Cycle Program Predicted Treatment Response 2.3. AZA Treatment Induced Upregulation of Genes Related to Natural Killer Cell Mediated Cytotoxicity in Responders The cycling status of AML malignant cells is known to be heterogeneous.[16d,18] We next sought to determine the malig- nant subsets that are in active cell cycle and likely sensitive to AZA-HiDAC-Mito therapy. By analyzing the public single-cell RNA sequencing (scRNA-seq) proﬁles of AML samples from 12 patients,[19] we compared six distinct subsets of malignant cells and seven immune cell types (Figure 2A; Figure S1A, Supporting Information). As expected, cell cycle signature was prominently expressed in several malignant subsets (Figure 2B; Figure S1B, Supporting Information). Notably, GMP-like cells exhibited the highest expression of cell cycle signature among the malignant subsets, while HSC-like and monocyte-like (Mono-like) malig- nant cells exhibited the lowest expression of cell cycle signature. Next, we applied the digital cytometry method, CIBERSORTx, to deconvolute the pre-treatment RNA-seq samples and estimate the abundance of each cell type based on the single-cell reference proﬁles (Figure S1C, Supporting Information).[20] We observed a positive correlation between the estimated abundance of GMP- like cells and the overall expression of the cell cycle signature in bulk RNA-seq samples, whereas there were negative correlations for both HSC-like and Mono-like cells (Figure 2C). It is notewor- thy that the HSC-like malignant cells were highly similar to pre- viously deﬁned leukemia stem cells (LSCs), which are known to be in a quiescent and non-dividing state (Figure S1D, Supporting Information).[21] These results indicate that the global cycling sta- tus of AML malignant cells is closely related to the malignant cell composition. Next, we questioned whether the compositions of malignant subsets were linked to treatment responses to AZA-HiDAC-Mito therapy. Remarkably, GMP-like cells were more abundant in re- sponders, while HSC-like cells were enriched in non-responders (Figure 2D,E; Figure S1E, Supporting Information). This obser- vation was further supported by the results of GSEA enrichment analysis conducted on the previously reported GMP-like signa- ture and LSC17 gene signature (Figure 2F,G).[19,21a] We also em- ployed Gene Set Variation Analysis (GSVA) to establish a GMP score based on the GMP-like signature for each sample, and com- pared it with the well-established LSC17 score.[21a] Both signa- tures exhibited AUC = 0.71 in distinguishing responders and non-responders (Figure S1F, Supporting Information). Similar performance was observed when using the relative fraction of GMP-like cells or HSC-likes cells as an indicator (AUC = 0.68 and 0.77, respectively). Furthermore, combining these two cellu- lar fractions using their diﬀerence resulted in superior perfor- mance with an AUC value of 0.83 (Figure 2H). Speciﬁcally, a malignant composition that is GMP-like-dominant predicts treat- ment response, while an HSC-like-dominant malignant compo- sition is associated with treatment failure (Figure S1G, Support- ing Information). Taken together, our results highlight GMP-like cells as the primary malignant subset that is sensitive to AZA- HiDAC-Mito therapy owing to an active cell cycle program, and the diﬀerence in cellular fractions between GMP-like and HSC- Much eﬀort had been made to identify the transcriptional eﬀects upon epigenetic priming by AZA treatment in both solid tumors and hematologic malignancies, providing insight into the mech- anisms by which AZA treatment exerts its eﬀects.[4a,b,22] Never- theless, the in vivo transcriptional eﬀects of AZA treatment in AML and their association to clinical response to an AZA-based therapy are still unclear. To address this, we compared gene ex- pression levels between RNA-seq samples from Day 5 and Day 0. GSEA analysis on KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways revealed that AZA treatment induced up- regulation of multiple pathways related to immune processes and immune activation (Figure 3A). Speciﬁcally, natural killer cell mediated cytotoxicity and T cell receptor signaling pathways were only upregulated in responders, suggesting distinct eﬀects upon AZA treatment between responders and non-responders (Figure 3B; Figure S2A, Supporting Information). We further cal- culated gene-set enrichment scores per sample with GSVA, and observed the pairwise upregulation of natural killer cell mediated cytotoxicity pathway but not T cell receptor signaling pathway in responders (Figure 3C; Figure S2B, Supporting Information). We then mapped the transcriptional changes onto the KEGG path- way using Pathview,[23] and observed a global upregulation of components in natural killer cell mediated cytotoxicity pathway in responders (Figure 3D; Figure S2C, Supporting Information). 2.4. Enrichment of NK Cells Predicted Favorable Clinical Outcomes in AZA-Based Therapies Previous studies reported that AZA treatment facilitated the tu- mor recognition of AML cells by NK cells.[24] Our results pro- vided in vivo evidence to support previous studies and further indicated the involvement of NK cells in determining the treat- ment response to such AZA-based therapy. To test whether the baseline level of NK cells is associated with treatment response, we analyzed the deconvolution result for immune subsets in pre- treatment RNA-seq samples. Notably, responders had a signiﬁ- cantly higher proportion of NK cells (p = 0.0088) (Figure 4A,B; Figure S3A, Supporting Information), which was further sup- ported by the enrichment of a curated NK cell signature with nor- malized enrichment score = 2.227, p = 0.002 (Figure 4C).[25] The core enriched genes included NK cell receptor NCR1 (NKp46), KLRC3, KLRD1, and NK cell cytotoxicity molecules: GZMH, PRF1, GZMA, and NKG7. These ﬁndings underlined an impor- tant role of NK cell abundance and activity in treatment response to AZA-HiDAC-Mito therapy. Furthermore, we tested whether the association of NK cell abundance and treatment response is AZA-speciﬁc. Analyzing RNA-seq samples from a historical cohort of AML patients re- ceiving HiDAC-Mito only therapy revealed that neither the esti- mated abundance of NK cells nor the expression of NK cell sig- nature was correlated with response to HiDAC-Mito only ther- apy (Figure S3B,C, Supporting Information).[6] We also assessed Adv. Sci. 2023, 10, 2300445 2300445 (4 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 2. AML malignant composition correlated with treatment response. A) UMAP visualization of malignant subsets from the public single- cell transcriptome (van Galen et al). Six subsets of malignant cells were included: hematopoietic stem cell-like (HSC-like), progenitor-like (Prog- like), promonocyte-like (ProMono-like), monocyte-like (Mono-like), conventional dendritic cell-like (cDC-like), Granulocyte-macrophage-progenitor-like (GMP-like). B) Boxplot showing the aggregated gene expression of cell cycle signature in each malignant subset per patients in the scRNA-seq dataset. C) Pearson’s correlation between the estimated abundance of malignant subsets and the overall expression of cell cycle signature in pre-treatment RNA-seq samples. D) Boxplot showing the estimated relative abundance of each malignant subset. p values were calculated with two-sided Student’s t-test. Fractions of BM/PB samples from same patients were averaged. E) The relative abundance of each malignant subset in each pre-treatment sam- ple. F,G) GSEA to assess the enrichment of GMP-like signature F) and LSC17 signature G) in responders of AZA-HiDAC-Mito therapy, comparing to non-responders. NES, normalized enrichment score; p value was calculated with permutation test. H) Receiving operating curve (ROC) analysis: Using relative fraction of GMP-like cells, HSC-like cells, or their combination (diﬀerence between the fractions of GMP-like cells and HSC-like cells) to predict responders. Adv. Sci. 2023, 10, 2300445 2300445 (5 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 3. Transcriptional changes upon AZA treatment for 5 days. A) Bubble plot showing the results of GSEA analysis on KEGG pathways by comparing gene expression between Day 5 and Day 0. NES, normalized enrichment score. Top 10 enriched pathways in responders are shown. padj, adjusted p value. NES, normalized enrichment score. B) GSEA to assess the enrichment of Natural killer cell mediated cytotoxicity pathway upon AZA treatment for 5 days in responders (top) and non-responders (bottom). NES, normalized enrichment score; p value was calculated with permutation test. C) Boxplot showing GSVA scores of Natural killer cell mediated cytotoxicity pathway. Paired samples were used in comparison of Day 5 and Day 0. p values were calculated with two-sided paired Student’s t-test. D) Pathview map showing the gene expression changes between Day 5 and Day 0 in responders in Natural killer cell mediated cytotoxicity pathway. The mapped color indicates log2(fold change) of Day 5 versus Day 0. Adv. Sci. 2023, 10, 2300445 2300445 (6 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 4. Immune landscape correlated with treatment response. A) Boxplot showing the estimated relative abundance of each immune subset. p values were calculated with two-sided Student’s t-test. Fractions of BM/PB samples from same patients were averaged. B) The relative abundance of each immune population in each sample from patients prior to AZA treatment. C) GSEA to assess the enrichment of a curated NK cell signature in responders of AZA-HiDAC-Mito therapy, comparing to non-responders. NES, normalized enrichment score; p value was calculated with permutation test. D) Kaplan-Meier curve of overall survival for patients receiving AZA treatment in Beat AML cohort. Patients were equally divided into two groups based on the aggregated expression of a NK cell signature. p value was calculated with a two-tailed log rank test. E) Pearson’s correlation of cellular fractions between HSC-like cells and NK cells in Day 0 samples. F) ROC curve for the performance of classiﬁers based on the cellular fraction of NK cells or the combination of NK cells and GMP-like cells (NK + GMP-like). the expression of known NK cell marker genes NCR1, KLRD1, NKG7, and KLRC3, which were upregulated in responders of AZA-HiDAC-Mito therapy but not HiDAC-Mito therapy, com- pared to non-responders (Figure S3D, Supporting Information). Nevertheless, when analyzing RNA-seq samples from patients receiving AZA treatment in Beat AML cohort,[1c] we found that both upregulation of NK cell signature and elevated NK cell abun- dance is associated with better overall survival (Figure 4D; Figure S3E, Supporting Information). In multivariate Cox regression models, stratifying patients based on both methods retained sig- niﬁcance for overall survival when age, sex, and ELN2017 risk classiﬁcation were considered (Table S2, Supporting Informa- tion). Additionally, for patients receiving decitabine, or standard “7+3” chemotherapy (Cytarabine, Idarubicin) in the Beat AML cohort, we observed no such associations (Figure S3F,G, Support- ing Information). Therefore, our results highlight the involve- ment of NK cells in determining the clinical outcomes of AZA- based therapies, which may link to the in vivo eﬀect of AZA treat- ment in inducing genes related to natural killer cell mediated cy- totoxicity. Adv. Sci. 2023, 10, 2300445 2300445 (7 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com We next asked whether the abundance of NK cells was associ- ated with the malignant composition. We found that the fraction of NK cells inversely correlated with the fraction of HSC-like cells (Figure 4E; Figure S3H, Supporting Information). Therefore, the fractions of NK cells can also be combined with fractions of GMP- like cells to classify responders and non-responders with an AUC = 0.84, which is better than merely using the fractions of NK cells (Figure 4F). Together, our ﬁndings establish that the malignant composition and immune landscape could stratify patients with diﬀerent responses to AZA-HiDAC-Mito therapy. 2.5. AZA Treatment Induced Dose-Dependent 5hmC Changes which were Prognostic AZA treatment is known to aﬀect the genome-wide distribu- tion of DNA 5mC and 5hmC modiﬁcations, which have been widely used for the diagnosis and prognosis of various types of cancer.[13a–d] To understand the epigenetic modulation eﬀect of AZA treatment, we characterized genome-wide 5hmC proﬁles for 120 BM and/or PB samples obtained from 40 patients at Day 0 and/or Day 5 through Nano-hmC-Seal.[26] There were only 19 patients that had both BM and PB samples at Day 0 and Day 5, and we performed diﬀerential analysis on 5hmC levels between Day 0 and Day 5 for each patient. We found that the diﬀerentially hydroxy-methylated genes (DhMGs) between Day 0 and Day 5 were rarely shared among patients, and the intrinsic diﬀerences at 5hmC patterns between patients far outweighed the eﬀects of AZA treatment (Figure S4A,B, Supporting Information). Never- theless, we observed more DhMGs in patients receiving higher dose of AZA, suggesting a dose-dependent epigenetic modula- tion eﬀect of AZA treatment (Figure 5A). We further used Spear- man’s correlation to evaluate the global diﬀerence of 5hmC pro- ﬁles between Day 0 and Day 5 (29 patients with paired samples from BM were included). Indeed, a higher dose of AZA treatment led to a more discriminated 5hmC proﬁle, reﬂecting a higher level of 5hmC alteration upon AZA treatment for 5 days (Figure S4C, Supporting Information). Notably, we found that patients with higher level of 5hmC al- teration upon AZA treatment for 5 days (i.e., higher number of DhMGs or lower correlation between Day 5 and Day 0) had longer overall survival in both univariate and multivariable analysis that incorporated age and sex features (Figure 5B; Figure S4D and Table S3, Supporting Information). Taken together, the epige- netic responsiveness to AZA treatment may be positively associ- ated with patient survival, as reﬂected by the alteration on 5hmC modiﬁcations. 2.6. Predicting the Treatment Response to AZA-HiDAC-Mito via a 5hmC-Based Machine Learning Model Given that 5hmC levels are known to correlate with gene ex- pression levels,[27] we wondered whether responders and non- responders could be also distinguished at 5hmC level in a man- ner similar to the RNA level. We performed diﬀerential hydrox- ymethylation analysis on 62 pre-treatment 5hmC samples col- lected from 40 patients (Figure 1A). The extent of diﬀerences at 5hmC levels positively correlated with diﬀerences at RNA levels, especially for the DEGs (Figure S4E, Supporting Information). Additionally, we evaluated the enrichment of gene signatures for GMP-like cells, HSC-like cells, and NK cells with GSEA analysis based on 5hmC levels, which exhibited consistent enrichment patterns similar to RNA levels (Figure 5C). These data suggest that 5hmC can also be used to distinguish responders and non- responders, similar to RNA, and thus support 5hmC as a candi- date biomarker for prediction of treatment response. In comparison to RNA-based biomarkers, DNA-based bioma- terials are far more stable during collection, handling, and trans- portation. As implicated in tumorigenesis and disease progres- sion, DNA 5hmC modiﬁcations have been widely used for the di- agnosis and prognosis of various types of cancer.[13] We thus tried to identify a 5hmC gene signature for prediction of treatment re- sponse by employing the eXtreme Gradient Boosting (XGBoost) machine learning algorithm to build a classiﬁer model. Since the 5hmC proﬁles from samples collected at Day 0 and Day 5 were highly analogous for the same patient at genome-wide level (Figure S4B, Supporting Information), we included both Day 0 and Day 5 samples to enlarge the sample size. The 5hmC pro- ﬁles were divided into a train set and test set based on sequencing batches (80 samples from 22 patients sequenced in the ﬁrst batch were used as the train set; 40 samples from 18 patients sequenced in the second batch were used as the test set). We initially trained an XGBoost model with all genes on the train set and evaluated its performance with patient-based ﬁve-fold cross validation. The receiver operating characteristic (ROC) curve showed that the 5hmC XGBoost classiﬁer achieved AUC = 0.71 in cross vali- dation (Figure 5D; Figure S4F, Supporting Information). By eval- uating the F score for each gene, we identiﬁed 142 genes that contributed to the model (Figure 5E). Top contributing genes in- cluded S-phase kinase-associated protein 1 (SKP1), the compo- nent of SKP1-CUL1-F-box-protein (SCF) complex that is involved in the proteolysis of cell cycle regulators.[28] Using t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduc- tion, the 5hmC levels of these 142 genes separate the respon- ders and non-responders (Figure 5F). To further obtain a 5hmC biomarker set with the best performance, we selected the genes with the highest contribution to the model to re-build the clas- siﬁer with train set samples. The signature composed from the top 11 contributing genes (including SKP1, WNT8A, CYP2E1, and NBPF9) achieved the best performance in cross validation (Figure 5G; Figure S4G,H, Supporting Information), with an AUC of 0.911 (speciﬁcity = 87.1% and sensitivity = 87.8%). The test set also achieved a high AUC of 0.86 (AUC = 0.90 and AUC = 0.82 for Day 0 and Day 5 samples in the test set, respectively; Figure S4I, Supporting Information). The high accuracy in pre- dicting Day 0 samples suggests that our 11-gene 5hmC signature could serve as a promising pre-treatment biomarker to reﬁne pa- tient selection of AZA-HiDAC-Mito therapy. Given the ease of collection and less invasive properties, pe- ripheral blood samples are generally considered as a preferred source for biomarker development in clinical applications. We thus evaluated the agreement between PB and BM samples in predicting treatment response. When performing unsupervised hierarchical clustering on pre-treatment 5hmC samples based on the contributing genes in XGBoost model, we found that BM and PB samples from same patient exhibited high concordance (Figure S5A, Supporting Information). Most paired BM and PB Adv. Sci. 2023, 10, 2300445 2300445 (8 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com Figure 5. Identiﬁcation of 5hmC gene signature for prediction of treatment response to AZA-HiDAC-Mito therapy. A) Boxplot showing the number of identiﬁed DhMGs between Day 0 and Day 5 for patients treated with diﬀerent AZA doses. Nineteen patients with paired Day 0 and Day 5 samples from both BM and PB were selected (4 samples for each patient). p values were calculated with Wilcoxon rank sum test. B) Kaplan-Meier survival curve for overall survival of 19 patients with paired Day 0 and Day 5 samples from both BM and PB. Patients were divided into two groups based on the median of identiﬁed DhMGs number. p value was calculated with a two-tailed log rank test. C) 5hmC-based GSEA to assess the enrichment of GMP-like, HSC-like, NK cell signature in responders of AZA-HiDAC-Mito therapy, comparing to non-responders. NES, normalized enrichment score; p value was calculated with permutation test. D) ROC curve for the performance of the XGBoost classiﬁer in cross validation for patients receiving AZA-HiDAC-Mito therapy. The model was trained on 80 samples. E) Bar graph showing F scores of top contributing genes in the 5hmC-based XGBoost model. F) t-SNE (t-distributed stochastic neighbor embedding) plot of samples from responders and non-responders based on 5hmC proﬁles of the top 142 contributing genes from the 5hmC-based XGBoost model. G) ROC curve for the performance of the XGBoost classiﬁer based on the 11-gene-5hmC signature. The model was evaluated using patient-based ﬁve-fold cross validation (80 samples from 22 AML patients). 40 samples from 18 AML patients were used for testing. Adv. Sci. 2023, 10, 2300445 2300445 (9 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com samples clustered closely to each other, and successfully clus- tered into the responder group or non-responder group. More- over, as for the 11-gene-5hmC model, we observed comparable AUC values for PB and BM samples in the test set (AUC = 0.84 and 0.86, respectively; Figure S5B, Supporting Information). These results collectively support the feasibility of PB samples in predicting treatment response to AZA-HiDAC-Mito therapy us- ing the established 11-gene 5hmC signature. 3. Discussion As the treatment of AML moves toward a subset speciﬁc ap- proach with targeted agents and combination regimens, identi- fying patients who may respond better to an HMA-based ther- apy remains an unmet need. Previous studies have analyzed DNA methylation proﬁling as a biomarker for response to HMA- based treatment in patients with myelodysplastic syndrome or AML. However, these studies have produced mixed results and none have established an epigenetic predictor of responsive- ness to HMAs.[29] These studies focused on global methyla- tion using long interspersed nuclear element (LINE) methyla- tion, methylation patterns in speciﬁc tumor suppressor genes, and an aberrantly hypermethylated gene signature, but overall there was no reliable signiﬁcant predictor of treatment response. Of note, these studies included other HMA combinations or HMA monotherapy for multiple cycles, whereas our study uti- lized AZA as epigenetic priming prior to cytotoxic therapy. In a phase 1 study of epigenetic priming with decitabine prior to cytotoxic therapy with cytarabine/daunorubicin in patients with AML, pre-treatment and post-priming DNA methylation levels of CDKN2B, LINE1, and HISTH2AA were not predictive of treat- ment response.[30] While overall 5hmC levels are found to vary among pa- tients with AML and show an inverse correlation with patient survival,[31] the feasibility of using 5hmC proﬁles to predict the responsiveness to a particular treatment has yet to be assessed. To our knowledge, our work is the ﬁrst eﬀort to identify a 5hmC predictive biomarker for treatment response in AML. Although 5hmC proﬁling is investigational and not yet a part of the clinical pre-treatment evaluation of patients with AML, it is a quick and sensitive method, which requires only a limited amount of ge- nomic DNA.[26] Using 5hmC proﬁling of peripheral blood and/or bone marrow biopsy samples in AML patients treated with AZA- HiDAC-Mito in a phase 1 clinical trial, we identiﬁed a pre- treatment 11-gene 5hmC signature as a predictive biomarker to identify patients who may beneﬁt from AZA-HiDAC-Mito. Due to small sample size, we were not able to investigate the eﬀect of cytogenetics and pathogenic mutations in the current model, and it would be important to incorporate these prognostic features along with the 5hmC signature in a larger prospective study. In addition to biomarker identiﬁcation, we also revealed mech- anistic insights into the therapeutic response of AML patients to AZA-HiDAC-Mito. Among responders, we found an increased expression of genes involved in the cell cycle and DNA synthe- sis, suggesting that increased numbers of actively cycling cells may be associated with eﬀective AZA-HiDAC-Mito response. In line with this, GMP-like cells were speculated as a dominant pro- liferating malignant subset that was associated with treatment response. We further uncovered the in vivo eﬀect of AZA treat- ment in inducing genes related to NK cell mediated cytotoxic- ity in responders. Clarifying whether this is a direct or indirect eﬀect of AZA towards NK cells requires successor studies. Fur- thermore, analysis of patients receiving an AZA-based regimen or non-AZA-based regimen revealed a unique role of NK cells in determining the response to AZA treatment. A combination of cellular fractions of GMP-like cells and NK cells can better predict the treatment response to AZA-HiDAC-Mito therapy, suggest- ing a combined eﬀect of tumor-intrinsic state and immune mi- croenvironment in governing the therapeutic response of AML patients. 4. Conclusions Collectively, our ﬁndings show that cellular compositions are as- sociated with treatment responses, and DNA 5hmC patterns in an 11-gene signature can be used as a pre-treatment biomarker for AZA-HiDAC-Mito therapy, which may help select patients who beneﬁt from this regimen. The potential of this 5hmC gene signature in predicting treatment response merits validation in larger prospective trials as well as studies involving other novel HMA-based combinations. 5. Experimental Section Study Subjects: Detailed phase 1 trial design methods for this study had been reported.[6] The study population included patients age ≥ 18 years with high-risk AML and Eastern Cooperative Oncology Group (ECOG) performance status 0–2. AML was deﬁned by the 2008 crite- ria of the World Health Organization (WHO).[32] Patients with high-risk disease were included and deﬁned as therapy related-AML (t-AML), re- lapsed/refractory AML (RR-AML), de novo AML in patients age ≥ 60 years, AML arising from myelodysplastic syndrome (MDS-AML), myeloprolifer- ative neoplasms in blast phase (MPN-BP), and chronic myelomonocytic myeloid leukemia (CMML-AML). This single-center trial was registered at www.clinicaltrials.gov as NCT01839240. All participants provided written informed consent. Trial Design: Cohorts of three patients were treated in a 3 + 3 dose escalation scheme. Patients received AZA at 37.5 mg m−2, 50 mg m−2, or 75 mg m−2 by subcutaneous administration (SC) or intravenous therapy (IV) once daily on Days 1–5 followed by cytarabine 3000 mg m−2 given IV over 4 h followed by mitoxantrone 30 mg m−2 given IV over 1 h once each on Day 6 and Day 10. The maximum dose of AZA to be explored was capped at 75 mg m−2. Cytarabine and mitoxantrone dose reductions were made for patients age ≥ 70 by 33% to 2000 mg m−2 of cytarabine and 20 mg m−2 of mitoxantrone. A research related bone marrow aspirate was performed pre-treatment/prior to AZA administration (Day 0) and after AZA administration (Day 5). Mononuclear cells were extracted and sam- ples were cryopreserved for future analysis. To evaluate the eﬃcacy of this regimen, a nadir marrow biopsy was performed on Day 17 and a biopsy to assess remission status was done within 2 weeks of hematologic recovery (deﬁned as absolute neutrophil count (ANC) ≥ 1000 per μL and platelet count ≥ 100 000 per μL), but no later than Day 42. Response criteria for complete remission (CR), CR with incomplete count recovery (CRi), and treatment failure (TF) were deﬁned according to the 2010 ELN Working Group recommendations.[33] The overall response rate was deﬁned as CR + CRi, and these patients were deﬁned as responders to treatment. Over- all survival was deﬁned as time from treatment to time of death. The data cutoﬀ date was November 1, 2017. The clinical results of the trial had been published.[6] DNA and RNA Isolation: DNA was extracted from bone marrow or peripheral blood using the Gentra Puregene Cell kit (Qiagen, Valencia, Adv. Sci. 2023, 10, 2300445 2300445 (10 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com CA) according to the manufacturer’s directions. RNA was extracted us- ing TRIzol Reagent (Thermo Fisher Scientiﬁc, Waltham, MA) according to the manufacturer’s directions. RNA Sequencing: mRNA was extracted from 1 μg of total RNA by using Dynabeads mRNA Direct kit (Ambion). For each sample, 20 ng of mRNA was used for library construction by using TruSeq stranded mRNA sam- ple preparation kit (Illumina). Libraries were sequenced on Illumina Hiseq 4000. Nano-hmC-Seal: Nano-hmC-Seal (5hmC-seq) was performed on 120 bone marrow (BM) and peripheral blood (PB) samples from 40 pa- tients collected at Day 0 and/or Day 5, as previously described with mi- nor changes.[26] Libraries were prepared with KAPA Hyperplus kit (KAPA KK8515) using extracted genomic DNA from patient BM or PB mononu- clear cells. Brieﬂy, 50 ng genomic DNA in 14 μL H2O was fragmented at 37 °C for 20 min by addition of 2 μL of 10x KAPA Fragmentation Buﬀer and 4 μL of KAPA Fragmentation Enzyme. The fragmented DNA was end- polished at 65 °C for 30 min by adding 2.8 μL of End Repair & A-Tailing Buﬀer and 1.2 μL of End Repair & A-Tailing Enzyme Mix. Three microliters of 1.5 μm Adapter (Bioo Scientiﬁc NOVA-514103) were added followed by 12 μL of Ligation Buﬀer and 4 μL DNA Ligase. The mixture was incubated at 20 °C for 1 h. Libraries were then puriﬁed by DNA Clean and Concentrator kit (Zymo D4013) and eluted in 20 μL H2O. 𝛽-GT labeling was then per- formed by addition of 0.85 μL self-synthesized 3 mm N3-UDG and 2.5 μL of T4-𝛽GT (Thermo EO0831) at 37 °C for 2 h. Azide labeled DNA libraries were then puriﬁed by DNA Clean and Concentrator kit (Zymo D4013) and eluted in 30 μL H2O. Libraries were further biotinylated by addition of 1 μL 4.5 mm (Sigma 760 749) DBCO-PEG4-Biotin and incubated at 37 °C for 2 h. Biotinylated DNA libraries were then puriﬁed by DNA Clean and Con- centrator kit (Zymo D4013) and eluted in 30 μL H2O. The biotinylated DNA was further enriched by 5 μL of M-270 Streptavidin beads (Thermo 65 305) and incubated at room temperature for 30 min. The beads were washed 3 times with Wash Buﬀer (5 mm Tris-HCl (pH 7.5); 0.5 mm EDTA; 1 m NaCl; 0.05% Tween 20) and resuspended in 20 μL H2O. Libraries were am- pliﬁed with on-bead PCR by addition of 5 μL Primer Mix (KAPA KK8515) and 25 μL of Enzyme Mix (KAPA KK8515) with following condition (98 ˚C 30 s; 98 ˚C 15 s; 60 ˚C 30 s; 72 ˚C 30 s; Repeat 14 cycles; 72 ˚C 1 min). Post-ampliﬁcation cleanup was performed by adding 0.9x Ampure beads (Beckman Coulter A63880); beads were washed twice with 80% ethanol and eluted in 50 μL H2O. Libraries were sequenced on Illumina NextSeq 500. RNA-Seq Data Processing: The quality control for raw sequence data was performed by FASTQC version 0.11.8.[34] The reads were then aligned to the UCSC hg19 reference genome by STAR-2.5.3 software.[35] Gene counts were analyzed by HOMER software.[36] BM and PB samples col- lected at Day 0 from patients treated with AZA-HiDAC-Mito were used to perform diﬀerential analysis. Diﬀerentially expressed genes between re- sponders and non-responders were detected by DESeq2.[37] The thresh- old of diﬀerentially expressed genes was set to p.adj ≤ 0.1 and \|log2 Fold- Change\| ≥ 0.5. Clustering analysis was performed with pheatmap package version 1.0.12. Annotation and genome ﬁles (Homo sapiens UCSC hg19) were downloaded from iGenomes. Gene Module and Pathway Analysis: For gene module analysis, Pear- son’s correlation coeﬃcients were ﬁrst calculated between each pair of diﬀerentially expressed genes (DEGs) based on log2-scaled normalized expression by variance stabilizing transformation (vst) method. A hier- archical clustering based on the Euclidean distance was then employed to separate genes into four modules. STRING database was utilized to extend gene modules by adding direct interacting genes that had a mean expression over 3 Transcripts Per Million (TPM).[14] Functional en- richment analysis was next performed with Metascape for each module network.[38] For GSEA analysis, clusterProﬁler was utilized.[39] Pathview package was used for visualization of transcriptional changes in indi- cated pathways. GSVA package was used for calculating gene-set enrich- ment scores per sample with default settings except for “mx.diﬀ = F”. For patients receiving AZA-HiDAC-Mito therapy, log2-scaled TPM expression was used; for public Beat AML cohort, the normalized expression ma- trix from https://biodev.github.io/BeatAML2 was used. The LSC17 scores were calculated per sample as the sum of the log2-transformed TPM val- ues for the 17 genes weighted by the regression coeﬃcients, as described previously.[21a] Digital Cytometry: Gene expression deconvolution was performed on CIBERSORTx web portal with default setting. In brief, reference signa- ture matrix was built by CIBERSORTx based on gene expression of 13,653 cells belonging to six malignant subsets (HSC-like, Prog-like, GMP-like, ProMono-like, Mono-like, cDC-like), and 7 non-leukemic immune popula- tions, including Mature B cell (B), Conventional dendritic cell (cDC), Cy- totoxic T Lymphocyte (CTL), Monocyte, Plasma cell (Plasma), Naïve T cell (T), and Natural Killer cell (NK). Bulk RNA-seq data were normalized with TPM and then deconvoluted using S-mode batch correction and relative mode. The inferred fractions were scaled to a sum of 1 for malignant sub- sets or immune populations, respectively. Survival Analysis: Kaplan-Meier survival analysis was calculated with survival R package (version 3.1-7) and visualized by the survminer R pack- age (version 0.4.6). In multivariable analysis, age and sex features were incorporated into the Cox regression models. For Beat AML cohort, the available ELN2017 risk classiﬁcation was also considered. 5hmC-Seq Data Analysis: The quality control for raw sequence data was performed by FASTQC version 0.11.8.[34] The 5hmC reads were then mapped to the UCSC hg19 reference genome by STAR-2.5.3 software with parameter “–alignIntronMax 1 –alignEndsType EndToEnd”. The de- duplication was performed using the parameter “-tbp 5” in makeTagDi- rectory of HOMER software, and the gene counts matrix was generated by the HOMER’s analyzeRepeats. DESeq2 was utilized to identify diﬀeren- tially hydroxy-methylated genes (DhMGs) upon AZA treatment (Day 5 vs Day 0) under threshold p value <0.01 and \|log2 FoldChange\| ≥ 0.5. Machine Learning Based on 5hmC-Seq Data: A total of 120 BM/PB samples were collected from 40 patients for 5hmC proﬁling. The ﬁrst batch of sequenced samples consisted of 80 samples obtained from 22 patients (47 samples from 13 responders and 33 samples from 9 non-responders), which were selected at random and without consideration of their treat- ment response status, to serve as the train set. The remaining 40 samples obtained from 18 patients were sequenced in the second batch and served as the test set (28 samples from 12 responders and 12 samples from 6 non-responders). The responders were set as case observations (positive label), and non-responders were set as control observations (negative la- bel) for the machine learning algorithm. The detailed patient id and clinical information of both train set and test set can be found in Table S1 (Sup- porting Information). Rlog-normalized 5hmC levels for each gene were used to build classiﬁer by XGBoost with python API (version 3.6.6). Prob- abilities estimation was then generated by “predict_proba” method. The performance was evaluated by patient-based ﬁve-fold cross validation. The importance of each gene (F-score) was calculated by “get_fscore” func- tion. Top 11 genes with highest F score were selected to rebuild the clas- siﬁer. To set up a negative control for the machine learning models, the response status of patients were randomly shuﬄed in the train set, and then the XGBoost models were trained on either all genes or just the top 11 contributing genes. Statistical Analyses: All statistical analyses were performed in R 3.6.0 software. The Pearson’s correlation was used unless speciﬁed otherwise. For comparison of responders and non-responders, two-tailed unpaired Student’s t tests or Wilcoxon rank sum tests were performed. For compar- ison of paired Day 0 and Day 5 samples, two-tailed paired Student’s t tests were used. For Kaplan-Meier survival curves, the p values were calculated using two-tailed log rank tests. Multivariable Cox models for overall sur- vival were used to adjust for potential confounders including age and sex. Statistical signiﬁcance was set at p < 0.05. All boxplots indicate median (center), 25th and 75th percentiles (boundaries of the box), and minimum and maximum (whiskers). Ethical Statement: This study was reviewed and approved by the insti- tutional review board at the University of Chicago (IRB 12–0111). Supporting Information Supporting Information is available from the Wiley Online Library or from the author. Adv. Sci. 2023, 10, 2300445 2300445 (11 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com Acknowledgements This work was supported by grants from the National Natural Science Foundation of China (32121001, 91853132, 31922017), CAS Hundred Tal- ent Program (to D.H.), the K. C. Wong Education Foundation (GJTD-2019- 08), the International Partnership Program of Chinese Academy of Sci- ences (153F11KYSB20210006) and the Ludwig Center for metastasis at the University of Chicago. C.H. is an investigator of the Howard Hughes Medical Institute. K.C. was supported through National Cancer Institute T32 CA 9566-35. Conﬂict of Interest C.H. is a scientiﬁc founder, a member of the scientiﬁc advisory board and equity holder of Aferna Bio, Inc. and AccuaDX Inc., a scientiﬁc cofounder and equity holder of Accent Therapeutics, Inc., and a member of the sci- entiﬁc advisory board of Rona Therapeutics; O.O has served on advisory boards convened by ABBVIE, Celgene/BMS, CTIBiopharma, Novartis, Im- pact Biomedicines; W.S. has served on advisory boards convened by Agios, Amgen, Astra Zeneca, Beam, Glaxo Smith Kline, Jazz, Kite, Kronos, Kura, Newave, Pﬁzer, Pluristem, Servier, Syndax. Author Contributions G.L., L.W., Q.Y., and K.C. contributed equally to this work. O.O., C.H., and D.H. designed the study and supervised the research. Q.Y. and N.F. per- formed experiments. K.C., O.O., and W.S. collected samples. G.L., L.W., and C.C. performed data analysis. D.H., G.L., Q.Y., and C.H. wrote the manuscript with input from L.W., K.C., O.O., W.S., W. Z., and C.C. All au- thors discussed the results and commented on the manuscript. Data Availability Statement The data that support the ﬁndings of this study are openly available in Gene expression Ominibus (GEO) at https://www.ncbi.nlm.nih.gov/geo, reference number 152431 and Genome Sequence Archieve (GSA) at https: //ngdc.cncb.ac.cn/gsa-human, reference number HRA000372. Keywords 5hmC, acute myeloid leukemia (AML), azacitidine, biomarkers, machine learning Received: January 19, 2023 Revised: May 12, 2023 Published online: June 4, 2023 [1] a) H. Kantarjian, T. Kadia, C. DiNardo, N. Daver, G. Borthakur, E. Jabbour, G. Garcia-Manero, M. Konopleva, F. Ravandi, Blood Cancer J 2021, 11, 41; b) E. Papaemmanuil, M. Gerstung, L. Bullinger, V. I. Gaidzik, P. Paschka, N. D. Roberts, N. E. Potter, M. Heuser, F. Thol, N. Bolli, G. Gundem, P. Van Loo, I. Martincorena, P. Ganly, L. Mudie, S. McLaren, S. O’Meara, K. Raine, D. R. Jones, J. W. Teague, A. P. Butler, M. F. Greaves, A. Ganser, K. Dohner, R. F. Schlenk, H. Dohner, P. J. Campbell, N. Engl. J. Med. 2016, 374, 2209; c) D. Bottomly, N. Long, A. R. Schultz, S. E. Kurtz, C. E. Tognon, K. Johnson, M. Abel, A. Agarwal, S. Avaylon, E. Benton, A. Blucher, U. Borate, T. P. Braun, J. Brown, J. Bryant, R. Burke, A. Carlos, B. H. Chang, H. J. Cho, S. Christy, C. Coblentz, A. M. Cohen, A. d’Almeida, R. Cook, A. Danilov, www.advancedscience.com K. T. Dao, M. Degnin, J. Dibb, C. A. Eide, I. English, et al., Cancer Cell 2022, 40, 850; d) A. G. X. Zeng, S. Bansal, L. Jin, A. Mitchell, W. C. Chen, H. A. Abbas, M. Chan-Seng-Yue, V. Voisin, P. van Galen, A. Tierens, M. Cheok, C. Preudhomme, H. Dombret, N. Daver, P. A. Futreal, M. D. Minden, J. A. Kennedy, J. C. Y. Wang, J. E. Dick, Nat. Med. 2022, 28, 1212. [2] a) A. H. Shih, O. Abdel-Wahab, J. P. Patel, R. L. Levine, Nat. Rev. Can- cer 2012, 12, 599; b) M. Ko, Y. Huang, A. M. Jankowska, U. J. Pape, M. Tahiliani, H. S. Bandukwala, J. An, E. D. Lamperti, K. P. Koh, R. Ganetzky, X. S. Liu, L. Aravind, S. Agarwal, J. P. Maciejewski, A. Rao, Nature 2010, 468, 839; c) M. E. Figueroa, O. Abdel-Wahab, C. Lu, P. S. Ward, J. Patel, A. Shih, Y. Li, N. Bhagwat, A. Vasanthakumar, H. F. Fernandez, M. S. Tallman, Z. Sun, K. Wolniak, J. K. Peeters, W. Liu, S. E. Choe, V. R. Fantin, E. Paietta, B. Lowenberg, J. D. Licht, L. A. Godley, R. Delwel, P. J. Valk, C. B. Thompson, R. L. Levine, A. Melnick, Cancer Cell 2010, 18, 553. [3] a) C. Riether, T. Pabst, S. Hopner, U. Bacher, M. Hinterbrandner, Y. Banz, R. Muller, M. G. Manz, W. H. Gharib, D. Francisco, R. Bruggmann, L. van Rompaey, M. Moshir, T. Delahaye, D. Gandini, E. Erzeel, A. Hultberg, S. Fung, H. de Haard, N. Leupin, A. F. Ochsenbein, Nat. Med. 2020, 26, 1459; b) C. D. DiNardo, K. Pratz, V. Pullarkat, B. A. Jonas, M. Arellano, P. S. Becker, O. Frankfurt, M. Konopleva, A. H. Wei, H. M. Kantarjian, T. Xu, W. J. Hong, B. Chyla, J. Potluri, D. A. Pollyea, A. Letai, Blood 2019, 133, 7; c) N. Daver, G. Garcia-Manero, S. Basu, P. C. Boddu, M. Alfayez, J. E. Cortes, M. Konopleva, F. Ravandi-Kashani, E. Jabbour, T. Kadia, G. M. Nogueras- Gonzalez, J. Ning, N. Pemmaraju, C. D. DiNardo, M. Andreeﬀ, S. A. Pierce, T. Gordon, S. M. Kornblau, W. Flores, Z. Alhamal, C. Bueso-Ramos, J. L. Jorgensen, K. P. Patel, J. Blando, J. P. Allison, P. Sharma, H. Kantarjian, Cancer Discov 2019, 9, 370; d) W. Sun, T. Triche, Jr., J. Malvar, P. Gaynon, R. Sposto, X. Yang, H. Bittencourt, A. E. Place, Y. Messinger, C. Fraser, L. Dalla-Pozza, B. Salhia, P. Jones, A. S. Wayne, L. Gore, T. M. Cooper, G. Liang, Blood 2018, 131, 1145. [4] a) K. B. Chiappinelli, P. L. Strissel, A. Desrichard, H. Li, C. Henke, B. Akman, A. Hein, N. S. Rote, L. M. Cope, A. Snyder, V. Makarov, S. Budhu, D. J. Slamon, J. D. Wolchok, D. M. Pardoll, M. W. Beckmann, C. A. Zahnow, T. Merghoub, T. A. Chan, S. B. Baylin, R. Strick, Cell 2015, 162, 974; b) D. Roulois, H. Loo Yau, R. Singhania, Y. Wang, A. Danesh, S. Y. Shen, H. Han, G. Liang, P. A. Jones, T. J. Pugh, C. O’Brien, D. D. De Carvalho, Cell 2015, 162, 961; c) H. C. Tsai, H. Li, L. Van Neste, Y. Cai, C. Robert, F. V. Rassool, J. J. Shin, K. M. Harbom, R. Beaty, E. Pappou, J. Harris, R. W. Yen, N. Ahuja, M. V. Brock, V. Stearns, D. Feller-Kopman, L. B. Yarmus, Y. C. Lin, A. L. Welm, J. P. Issa, I. Minn, W. Matsui, Y. Y. Jang, S. J. Sharkis, S. B. Baylin, C. A. Zahnow, Cancer Cell 2012, 21, 430; d) P. Frost, J. L. Abbruzzese, B. Hunt, D. Lee, M. Ellis, Cancer Res. 1990, 50, 4572; e) C. Hu, X. Liu, Y. Zeng, J. Liu, F. Wu, Clin Epigenetics 2021, 13, 166. J. Vadakekolathu, M. D. Minden, T. Hood, S. E. Church, S. Reeder, H. Altmann, A. H. Sullivan, E. J. Viboch, T. Patel, N. Ibrahimova, S. E. Warren, A. Arruda, Y. Liang, T. H. Smith, G. A. Foulds, M. D. Bailey, J. Gowen-MacDonald, J. Muth, M. Schmitz, A. Cesano, A. G. Pockley, P. J. M. Valk, B. Lowenberg, M. Bornhauser, S. K. Tasian, M. P. Rettig, J. K. Davidson-Moncada, J. F. DiPersio, S. Rutella, Sci. Transl. Med. 2020, 12, eaaz0463. [5] [6] K. E. Cahill, Y. H. Karimi, T. G. Karrison, N. Jain, M. Green, H. Weiner, N. Fulton, S. Kadri, L. A. Godley, A. S. Artz, H. Liu, M. J. Thirman, M. M. Le Beau, M. E. McNerney, J. Segal, R. A. Larson, W. Stock, O. Odenike, Blood Adv 2020, 4, 599. [7] S. M. Larson, N. P. Campbell, D. Huo, A. Artz, Y. Zhang, D. Gajria, M. Green, H. Weiner, C. Daugherty, O. Odenike, L. A. Godley, E. Hyjek, S. Gurbuxani, M. Thirman, D. Sipkins, K. Van Besien, R. A. Larson, W. Stock, Leuk Lymphoma 2012, 53, 445. [8] D. Pan, R. Rampal, J. Mascarenhas, Blood Adv 2020, 4, 970. Adv. Sci. 2023, 10, 2300445 2300445 (12 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com [9] a) S. Y. Shen, R. Singhania, G. Fehringer, A. Chakravarthy, M. H. A. Roehrl, D. Chadwick, P. C. Zuzarte, A. Borgida, T. T. Wang, T. Li, O. Kis, Z. Zhao, A. Spreaﬁco, T. D. S. Medina, Y. Wang, D. Roulois, I. Ettayebi, Z. Chen, S. Chow, T. Murphy, A. Arruda, G. M. O’Kane, J. Liu, M. Mansour, J. D. McPherson, C. O’Brien, N. Leighl, P. L. Bedard, N. Fleshner, G. Liu, et al., Nature 2018, 563, 579; b) S. Gkountela, F. Castro-Giner, B. M. Szczerba, M. Vetter, J. Landin, R. Scherrer, I. Krol, M. C. Scheidmann, C. Beisel, C. U. Stirnimann, C. Kurzeder, V. Heinzelmann-Schwarz, C. Rochlitz, W. P. Weber, N. Aceto, Cell 2019, 176, 98. [10] a) S. Ito, A. C. D’Alessio, O. V. Taranova, K. Hong, L. C. Sowers, Y. Zhang, Nature 2010, 466, 1129; b) M. Tahiliani, K. P. Koh, Y. Shen, W. A. Pastor, H. Bandukwala, Y. Brudno, S. Agarwal, L. M. Iyer, D. R. Liu, L. Aravind, A. Rao, Science 2009, 324, 930. [11] B. Thienpont, J. Steinbacher, H. Zhao, F. D’Anna, A. Kuchnio, A. Ploumakis, B. Ghesquiere, L. Van Dyck, B. Boeckx, L. Schoonjans, E. Hermans, F. Amant, V. N. Kristensen, K. Peng Koh, M. Mazzone, M. Coleman, T. Carell, P. Carmeliet, D. Lambrechts, Nature 2016, 537, 63. [12] Y. Nakauchi, A. Azizi, D. Thomas, M. R. Corces, A. Reinisch, R. Sharma, D. Cruz Hernandez, T. Kohnke, D. Karigane, A. Fan, D. Martinez-Krams, M. Staﬀord, S. Kaur, R. Dutta, P. Phan, A. Ediriwickrema, E. McCarthy, Y. Ning, T. Phillips, C. K. Ellison, G. D. Guler, A. Bergamaschi, C. J. Ku, S. Levy, R. Majeti, Blood Cancer Dis- covery 2022, 3, 346. [13] a) B. C. Chiu, Z. Zhang, Q. You, C. Zeng, E. Stepniak, P. M. Bracci, K. Yu, G. Venkataraman, S. M. Smith, C. He, W. Zhang, Blood Adv 2019, 3, 2790; b) J. Cai, L. Chen, Z. Zhang, X. Zhang, X. Lu, W. Liu, G. Shi, Y. Ge, P. Gao, Y. Yang, A. Ke, L. Xiao, R. Dong, Y. Zhu, X. Yang, J. Wang, T. Zhu, D. Yang, X. Huang, C. Sui, S. Qiu, F. Shen, H. Sun, W. Zhou, J. Zhou, J. Nie, C. Zeng, E. K. Stroup, X. Zhang, B. C. Chiu, et al., Gut 2019, 68, 2195; c) M. A. Applebaum, E. K. Barr, J. Karpus, J. Nie, Z. Zhang, A. E. Armstrong, S. Uppal, M. Sukhanova, W. Zhang, A. Chlenski, H. R. Salwen, E. Wilkinson, M. Dobratic, R. Grossman, L. A. Godley, B. E. Stranger, C. He, S. L. Cohn, JCO Precis Oncol 2019, 3, 1.; d) X. Tian, B. Sun, C. Chen, C. Gao, J. Zhang, X. Lu, L. Wang, X. Li, Y. Xing, R. Liu, X. Han, Z. Qi, X. Zhang, C. He, D. Han, Y. G. Yang, Q. Kan, Cell Res. 2018, 28, 597; e) C. X. Song, S. Yin, L. Ma, A. Wheeler, Y. Chen, Y. Zhang, B. Liu, J. Xiong, W. Zhang, J. Hu, Z. Zhou, B. Dong, Z. Tian, S. S. Jeﬀrey, M. S. Chua, S. So, W. Li, Y. Wei, J. Diao, D. Xie, S. R. Quake, Cell Res. 2017, 27, 1231. [14] D. Szklarczyk, A. L. Gable, D. Lyon, A. Junge, S. Wyder, J. Huerta- Cepas, M. Simonovic, N. T. Doncheva, J. H. Morris, P. Bork, L. J. Jensen, C. V. Mering, Nucleic Acids Res. 2019, 47, D607. [15] E. B. Pasquale, Nat. Rev. Cancer 2010, 10, 165. [16] a) L. H. Li, E. J. Olin, T. J. Fraser, B. K. Bhuyan, Cancer Res. 1970, 30, 2770; b) L. H. Li, E. J. Olin, H. H. Buskirk, L. M. Reineke, Cancer Res. 1970, 30, 2760; c) D. Kufe, D. Spriggs, E. M. Egan, D. Munroe, Blood 1984, 64, 54; d) Y. Saito, N. Uchida, S. Tanaka, N. Suzuki, M. Tomizawa-Murasawa, A. Sone, Y. Najima, S. Takagi, Y. Aoki, A. Wake, S. Taniguchi, L. D. Shultz, F. Ishikawa, Nat. Biotechnol. 2010, 28, 275. J. Cuzick, G. P. Swanson, G. Fisher, A. R. Brothman, D. M. Berney, J. E. Reid, D. Mesher, V. O. Speights, E. Stankiewicz, C. S. Foster, H. Moller, P. Scardino, J. D. Warren, J. Park, A. Younus, D. D. Flake, 2nd, S. Wagner, A. Gutin, J. S. Lanchbury, S. Stone, G. Transatlantic Prostate, Lancet Oncol. 2011, 12, 245. [17] [18] Y. Guan, B. Gerhard, D. E. Hogge, Blood 2003, 101, 3142. [19] P. van Galen, V. Hovestadt, M. H. Wadsworth Ii, T. K. Hughes, G. K. Griﬃn, S. Battaglia, J. A. Verga, J. Stephansky, T. J. Pastika, J. Lombardi Story, G. S. Pinkus, O. Pozdnyakova, I. Galinsky, R. M. Stone, T. A. Graubert, A. K. Shalek, J. C. Aster, A. A. Lane, B. E. Bernstein, Cell 2019, 176, 1265. [20] A. M. Newman, C. B. Steen, C. L. Liu, A. J. Gentles, A. A. Chaudhuri, F. Scherer, M. S. Khodadoust, M. S. Esfahani, B. A. Luca, D. Steiner, M. Diehn, A. A. Alizadeh, Nat. Biotechnol. 2019, 37, 773. [21] a) S. W. Ng, A. Mitchell, J. A. Kennedy, W. C. Chen, J. McLeod, N. Ibrahimova, A. Arruda, A. Popescu, V. Gupta, A. D. Schimmer, A. C. Schuh, K. W. Yee, L. Bullinger, T. Herold, D. Gorlich, T. Buchner, W. Hiddemann, W. E. Berdel, B. Wormann, M. Cheok, C. Preudhomme, H. Dombret, K. Metzeler, C. Buske, B. Lowenberg, P. J. Valk, P. W. Zandstra, M. D. Minden, J. E. Dick, J. C. Wang, Nature 2016, 540, 433; b) Y. Saito, H. Kitamura, A. Hijikata, M. Tomizawa-Murasawa, S. Tanaka, S. Takagi, N. Uchida, N. Suzuki, A. Sone, Y. Najima, H. Ozawa, A. Wake, S. Taniguchi, L. D. Shultz, O. Ohara, F. Ishikawa, Sci. Transl. Med. 2010, 2, 17ra9. [22] a) A. Unnikrishnan, E. Papaemmanuil, D. Beck, N. P. Deshpande, A. Verma, A. Kumari, P. S. Woll, L. A. Richards, K. Knezevic, V. Chandrakanthan, J. A. I. Thoms, M. L. Tursky, Y. Huang, Z. Ali, J. Olivier, S. Galbraith, A. G. Kulasekararaj, M. Tobiasson, M. Karimi, A. Pellagatti, S. R. Wilson, R. Lindeman, B. Young, R. Ramakrishna, C. Arthur, R. Stark, P. Crispin, J. Curnow, P. Warburton, F. Roncolato, et al., Cell Rep. 2017, 20, 572; b) K. K. Leung, A. Nguyen, T. Shi, L. Tang, X. Ni, L. Escoubet, K. J. MacBeth, J. DiMartino, J. A. Wells, Proc Natl Acad Sci U S A 2019, 116, 695. [23] W. Luo, C. Brouwer, Bioinformatics 2013, 29, 1830. [24] a) J. Cany, M. W. H. Roeven, J. S. Hoogstad-van Evert, W. Hobo, F. Maas, R. Franco Fernandez, N. M. A. Blijlevens, W. J. van der Velden, G. Huls, J. H. Jansen, N. P. M. Schaap, H. Dolstra, Blood 2018, 131, 202; b) A. B. Raneros, A. Minguela, R. M. Rodriguez, E. Colado, T. Bernal, E. Anguita, A. V. Mogorron, A. C. Gil, J. R. Vidal-Castineira, L. Marquez-Kisinousky, P. D. Bulnes, A. M. Marin, M. C. G. Garay, B. Suarez-Alvarez, C. Lopez-Larrea, OncoTargets Ther. 2017, 8, 31959. J. Cursons, F. Souza-Fonseca-Guimaraes, M. Foroutan, A. Anderson, F. Hollande, S. Hediyeh-Zadeh, A. Behren, N. D. Huntington, M. J. Davis, Cancer Immunol. Res. 2019, 7, 1162. [25] [26] D. Han, X. Lu, A. H. Shih, J. Nie, Q. You, M. M. Xu, A. M. Melnick, R. L. Levine, C. He, Mol. Cell 2016, 63, 711. [27] X. L. Cui, J. Nie, J. Ku, U. Dougherty, D. C. West-Szymanski, F. Collin, C. K. Ellison, L. Sieh, Y. Ning, Z. Deng, C. W. T. Zhao, A. Bergamaschi, J. Pekow, J. Wei, A. V. Beadell, Z. Zhang, G. Sharma, R. Talwar, P. Arensdorf, J. Karpus, A. Goel, M. Bissonnette, W. Zhang, S. Levy, C. He, Nat. Commun. 2020, 11, 6161. [28] K. I. Nakayama, K. Nakayama, Nat. Rev. Cancer 2006, 6, 369. [29] a) T. E. Fandy, J. G. Herman, P. Kerns, A. Jiemjit, E. A. Sugar, S. H. Choi, A. S. Yang, T. Aucott, T. Dauses, R. Odchimar-Reissig, J. Licht, M. J. McConnell, C. Nasrallah, M. K. Kim, W. Zhang, Y. Sun, A. Murgo, I. Espinoza-Delgado, K. Oteiza, I. Owoeye, L. R. Silverman, S. D. Gore, H. E. Carraway, Blood 2009, 114, 2764; b) G. Garcia-Manero, H. M. Kantarjian, B. Sanchez-Gonzalez, H. Yang, G. Rosner, S. Verstovsek, M. Rytting, W. G. Wierda, F. Ravandi, C. Koller, L. Xiao, S. Faderl, Z. Estrov, J. Cortes, S. O’Brien, E. Estey, C. Bueso-Ramos, J. Fiorentino, E. Jabbour, J. P. Issa, Blood 2006, 108, 3271; c) L. Shen, H. Kantarjian, Y. Guo, E. Lin, J. Shan, X. Huang, D. Berry, S. Ahmed, W. Zhu, S. Pierce, Y. Kondo, Y. Oki, J. Jelinek, H. Saba, E. Estey, J. P. Issa, J. Clin. Oncol. 2010, 28, 605; d) M. Cross, E. Bach, T. Tran, R. Krahl, N. Jaekel, D. Niederwieser, C. Junghanss, G. Maschmeyer, H. K. Al-Ali, Onco Targets Ther 2013, 6, 741; e) J. P. Issa, G. Garcia-Manero, F. J. Giles, R. Mannari, D. Thomas, S. Faderl, E. Bayar, J. Lyons, C. S. Rosenfeld, J. Cortes, H. M. Kantarjian, Blood 2004, 103, 1635. J. M. Scandura, G. J. Roboz, M. Moh, E. Morawa, F. Brenet, J. R. Bose, L. Villegas, U. S. Gergis, S. A. Mayer, C. M. Ippoliti, T. J. Curcio, E. K. Ritchie, E. J. Feldman, Blood 2011, 118, 1472. [30] [31] L. I. Kroeze, M. G. Aslanyan, A. van Rooij, T. N. Koorenhof-Scheele, M. Massop, T. Carell, J. B. Boezeman, J. P. Marie, C. J. Halkes, T. de Witte, G. Huls, S. Suciu, R. A. Wevers, B. A. van der Reijden, J. H. Jansen, E. L. Group, Gimema, Blood 2014, 124, 1110. Adv. Sci. 2023, 10, 2300445 2300445 (13 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH www.advancedsciencenews.com www.advancedscience.com [32] J. W. Vardiman, J. Thiele, D. A. Arber, R. D. Brunning, M. J. Borowitz, A. Porwit, N. L. Harris, M. M. Le Beau, E. Hellstrom-Lindberg, A. Teﬀeri, C. D. Bloomﬁeld, Blood 2009, 114, 937. [35] A. Dobin, C. A. Davis, F. Schlesinger, J. Drenkow, C. Zaleski, S. Jha, P. Batut, M. Chaisson, T. R. Gingeras, Bioinformatics 2013, 29, 15. [33] H. Dohner, E. H. Estey, S. Amadori, F. R. Appelbaum, T. Buchner, A. K. Burnett, H. Dombret, P. Fenaux, D. Grimwade, R. A. Larson, F. Lo- Coco, T. Naoe, D. Niederwieser, G. J. Ossenkoppele, M. A. Sanz, J. Sierra, M. S. Tallman, B. Lowenberg, C. D. Bloomﬁeld, L. European, Blood 2010, 115, 453. [34] S. Andrews, Krueger F, Seconds-Pichon A, Biggins F, Wingett S., Babraham Inst 2015. [36] S. Heinz, C. Benner, N. Spann, E. Bertolino, Y. C. Lin, P. Laslo, J. X. Cheng, C. Murre, H. Singh, C. K. Glass, Mol. Cell 2010, 38, 576. [37] M. I. Love, W. Huber, S. Anders, Genome Biol. 2014, 15, 550. [38] Y. Zhou, B. Zhou, L. Pache, M. Chang, A. H. Khodabakhshi, O. Tanaseichuk, C. Benner, S. K. Chanda, Nat. Commun. 2019, 10, 1523. [39] G. Yu, L.-G. Wang, Y. Han, Q.-Y. He, OMICS 2012, 16, 284. Adv. Sci. 2023, 10, 2300445 2300445 (14 of 14) © 2023 The Authors. Advanced Science published by Wiley-VCH GmbH
10.1021_acscentsci.3c00547	http://pubs.acs.org/journal/acscii This article is licensed under CC-BY 4.0 Article Molecular Crowding Facilitates Ribozyme-Catalyzed RNA Assembly Saurja DasGupta,* Stephanie Zhang, and Jack W. Szostak* Cite This: ACS Cent. Sci. 2023, 9, 1670−1678 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: Catalytic RNAs or ribozymes are considered to be central to primordial biology. Most ribozymes require moderate to high concentrations of divalent cations such as Mg2+ to fold into their catalytically competent structures and perform catalysis. However, undesirable effects of Mg2+ such as hydrolysis of reactive RNA building blocks and degradation of RNA structures are likely to undermine its beneficial roles in ribozyme catalysis. Further, prebiotic cell-like compartments bounded by fatty acid membranes are destabilized in the presence of Mg2+, making ribozyme function inside prebiotically relevant protocells a significant challenge. Therefore, we sought to identify conditions that would enable ribozymes to retain activity at low concentrations of Mg2+. Inspired by the ability of ribozymes to function inside crowded cellular environments with <1 mM free Mg2+, we tested molecular crowding as a potential mechanism to lower the Mg2+ concentration required for ribozyme-catalyzed RNA assembly. Here, we show that the ribozyme-catalyzed ligation of phosphorimidazolide RNA substrates is significantly enhanced in the presence of the artificial crowding agent polyethylene glycol. We also found that molecular crowding preserves ligase activity under denaturing conditions such as alkaline pH and the presence of urea. Additionally, we show that crowding-induced stimulation of RNA-catalyzed RNA assembly is not limited to phosphorimidazolide ligation but extends to the RNA-catalyzed polymerization of nucleoside triphosphates. RNA-catalyzed RNA ligation is also stimulated by the presence of prebiotically relevant small molecules such as ethylene glycol, ribose, and amino acids, consistent with a role for molecular crowding in primordial ribozyme function and more generally in the emergence of RNA-based cellular life. ■ INTRODUCTION The catalytic repertoire of RNA lies at the foundation of the RNA world hypothesis, which posits that early life used RNA as both the genetic material and enzymes (ribozymes).1 The ability of single-stranded RNA molecules to assume a wide range of folded structures endows them with functions such as molecular recognition and catalysis, suggesting that folded RNA structures would have been essential to early life. RNA assembly processes (ligation and polymerization) that generate complex folded RNA structures were therefore likely to have played an important role in the propagation and evolution of the earliest living cells. Ribozymes usually require divalent cations like Mg2+ to access their functional folds and perform catalysis. Mg2+ facilitates RNA folding by partially neutralizing the negatively charged RNA backbone and often participates in catalytic interactions within the ribozyme active site.2−4 to RNA function, Mg2+ can also be Although essential detrimental. Mg2+ catalyzes RNA backbone hydrolysis, thereby disrupting functional the structures. hydrolysis of intrinsically reactive RNA building blocks such as phosphorimidazolides that would have been important for primordial RNA assembly.5−7 Additionally, Mg2+ is generally detrimental to the integrity of prebiotic cell-like compartments bounded by fatty acids, which are commonly used models of primordial cell membranes. This incompatibility between It also accelerates ribozyme function and the stability of protocell membranes poses a significant challenge for efficient RNA catalysis within fatty acid protocells.5 RNA assembly would have driven primordial genetics and generated the catalytic diversity required to sustain RNA-based primordial life; therefore, ribozymes that catalyze RNA ligation or polymerization were crucial to primordial biology. Such ribozymes have been identified through in vitro evolution.5 Ligase and polymerase ribozymes that use 5′-triphosphorylated oligoribonucleotides and nucleoside triphosphates as sub- respectively, exhibit high Mg2+ requirements. For strates, example, the Mg2+ concentration at which the half-maximum ligation rate was achieved, [Mg2+]1/2, of the first of its kind, class I ligase is 70−100 mM,8 and polymerase ribozymes derived from the class I ligase have an optimal [Mg2+] of ∼200 mM.9,10 We previously reported ribozymes that catalyze the ligation of RNA oligomers 5′ activated with a prebiotically plausible, 2-aminoimidazole (2AI) moiety. 2AI-activated RNA Received: May 1, 2023 Published: August 3, 2023 © 2023 The Authors. Published by American Chemical Society 1670 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 1. Stimulation of ribozyme activity of ligase 1 at 1−2 mM Mg2+ in the presence of ethylene glycol and PEGs. (A) Schematic of ribozyme- catalyzed ligation of a 2-aminoimidazole-activated RNA substrate. (B) Catalytic ligation is undetectable at 1 mM Mg2+ in a solution without any crowder but is rescued in crowded solutions. (C) Ligation yields after 3 h in the absence and presence of crowding agents at the indicated concentrations. (D) Ligation rates in the absence and presence of crowding agents. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and the indicated concentrations of MgCl2. Reactions contained additives (EG, PEG 200−8000) as indicated. None indicates the absence of crowders. monomers/oligomers are useful substrates for nonenzymatic RNA assembly; therefore, these “2AI-ligase” ribozymes provide continuity between chemical and enzymatic RNA ligation.6 low Mg2+ Most of these 2AI-ligases were inefficient at concentrations (<4 mM); however, we identified a single ligase sequence that had a significantly lower Mg2+ requirement ([Mg2+]1/2 ≈ 0.9 mM).11 Although ribozymes with reduced Mg2+ requirements clearly exist, they are apparently relatively uncommon in the RNA sequence space. We have therefore searched for a more general solution that would have enabled ribozymes to operate in low-Mg2+ environments such as freshwater ponds or within protocells bounded by prebiotic fatty acids.12 Mechanisms that stimulate ribozyme activity at low [Mg2+] would lower the evolutionary threshold for the emergence of such molecules in the RNA world. Although ribozymes usually require moderate to high Mg2+ concentrations to function in vitro, naturally occurring ribozymes have evolved to function in the presence of 0.5−1 mM free Mg2+ within cellular environments.13 This lower Mg2+ requirement is thought to be a consequence of the crowded cellular environment. In addition to cellular structures like organelles, the intracellular milieu is crowded with molecules that range from biopolymers like nucleic acids and proteins to smaller molecules including amino acids, nucleotides, sugars, amines, and alcohols, which collectively occupy up to 30% of the cellular volume.13,14 The presence of these molecules introduces a variety of physical and chemical forces that alter the properties of cellular RNAs.15−17 Volume excluded by macromolecules decreases the conformational entropy of unfolded RNA (an effect commonly referred to as “macro- molecular crowding”) and consequently promotes RNA folding and RNA function. Unfavorable interactions between the solvent-exposed RNA backbone and low-MW species in the cellular milieu also induce folding to minimize these interactions. A decrease in dielectric constant may favor RNA− Mg2+ association due to the diminished solvation of free Mg2+, which can stimulate RNA folding and catalysis. A decrease in water activity caused by cosolutes may favor the formation of RNA folds with reduced solvent-exposed surface area that is accompanied by water release. Investigations into RNA structure and function in solutions artificially crowded with cosolutes like polyethylene glycol (PEG) have revealed favorable effects of crowding on RNA function.17 Biophysical studies using small-angle X-ray scattering (SAXS) and single-molecule Förster resonance energy transfer (smFRET) demonstrated that molecular crowding induces RNA folding. This effect is most pronounced in the low-Mg2+ regime, where folded structures are not usually predominant.18−20 Enhanced folding in crowded solutions is often reflected in modest to significant increases in catalytic rates.17 Ribozymes in the RNA world may have evolved in similarly crowded environments within either primitive cellular compartments or confined microspaces on the Earth’s surface, which may have allowed them to function at low concentrations of Mg2+.17,21 Here, we demonstrate the beneficial effects of molecular crowding on ribozyme-catalyzed RNA assembly, which includes the stimulation of ribozyme ligase activity at low millimolar concentrations of Mg2+ and the preservation of ribozyme activity under harsh reaction conditions such as alkaline pH or urea-induced denaturation. We propose that the stabilization of catalytic RNA folds in prebiotic crowded environments could provide a general means of enabling ribozyme-catalyzed RNA assembly in diverse environments including those with low availability of Mg2+. 1671 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 2. Crowding decreases the Mg2+ requirement for RNA-ligase activity. Mg2+ dependence on ligation rates of the ligase 1 ribozyme (A) in the absence of crowders and (B−D) in the presence of (B) 10% (w/v) EG, (C) 30% (w/v) PEG 200, and (D) 19% (w/v) PEG 1000. (E) Crowding agents reduce the [Mg2+]1/2 values for the rate of ribozyme ligation. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and the indicated concentrations of MgCl2. Reactions contained additives (EG, PEG 200, or PEG 1000) as indicated. ■ RESULTS AND DISCUSSION Crowding Rescues RNA-Catalyzed RNA Ligation at Low Mg2+ Concentrations. To test the effect of crowding on RNA-catalyzed RNA assembly, we chose a ligase ribozyme (henceforth, ligase 1) (Figure 1A, Table S1), previously identified by in vitro selection, that catalyzes the template- directed ligation of a primer strand to a 2AI-activated oligonucleotide. This ribozyme exhibited significantly reduced product yields at Mg2+ concentrations below 4 mM.6 For example, ligation proceeded to ∼30% in 3 h at 4 mM Mg2+, but yields were reduced to 8%, 2%, and 1% at 3, 2, and 1 mM Mg2+, respectively. This ribozyme exhibits a corresponding reduction in activity in the low-Mg2+ regime with only 5−15- fold rate enhancement over background at 2−3 mM Mg2+ compared to the ∼300-fold enhancement observed at 10 mM Mg2+.6 We used polyethylene glycol (PEG) to generate a crowded environment in vitro. PEG is chemically inert and available in a wide range of MWs, which allowed us to simulate the presence of a variety of small molecules and biopolymers that could have been present in prebiotic milieus. We also included ethylene glycol (EG) in our studies in addition to PEGs of various MWs (PEG 200, PEG 400, PEG 1000, PEG 8000). EG can be synthesized abiotically22,23 and is one the larger molecules detected in interstellar medium.24,25 We first screened various concentrations of EG, PEG 200, PEG 400, PEG 1000, and PEG 8000 to identify optimal crowding conditions for ligase 1 activity in the presence of 1 mM Mg2+ and 100 mM Tris-HCl, pH 8 (Figure S1). We observed remarkable ligation rescue in the presence of EG and both low- and high-MW PEGs. Ligation yield rose from barely detectable levels in the absence of crowding agents to about 20% and 50% after 3 h at 1 and 2 mM Mg2+, respectively, in the presence of 10% (w/v) EG. Similar stimulation in ligation was observed in 30% (w/v) PEG 200, 30% (w/v) PEG 400, 19% (w/v) PEG 1000, and 19% (w/v) PEG 8000 at 1 mM Mg2+ with ∼50% ligation after 3 h, which is comparable to the 60% ligation observed in solution at 10 mM Mg2+ with no crowding agents (Figure 1B and 1C). Ligation rates in the presence of crowders at low Mg2+ (from 0.7 to 1.3 h−1) were also comparable to the rate observed in the absence of crowders at 10 mM Mg2+ (∼1.5 h−1) (Figure 1D). Ligation yield decreased with an increase in the concentration of EG. This trend is different from other PEG-based crowders which exhibit better ligation at higher concentrations (Figure S1). This difference between EG and PEGs could be due to the mechanism by which these crowders effect RNA structure. EG cannot exclude significant volume due to its small size and must act through direct interactions with the RNA backbone or through solvent effects which increase the association between RNA and Mg2+. Therefore, the crowding effects observed are likely enthalpic, in contrast to the entropic contributions from PEGs, especially ones with moderate to high MWs. To understand the attenuated Mg2+ dependence of ribozyme-ligase activity, we measured ligation rates as a function of Mg2+ concentration in the presence of 10% EG (low-MW additive), 30% PEG 200 (low-MW additive), and 19% PEG 1000 (high-MW additive). [Mg2+]1/2 was signifi- cantly lowered in the presence of crowding agents (Figure 2, Figure S2), consistent with the enhanced ligation yield observed at low Mg2+ concentrations. A 3-fold reduction in [Mg2+]1/2 was observed in 10% EG, while 30% PEG 200 and 19% PEG 1000 caused a ∼10-fold reduction (Figure 2E). While all three crowders (EG, PEG 200, PEG 1000) supported ligase 1 activity at lower concentrations of Mg2+, maximal rates were achieved at submillimolar Mg2+ with PEG 200 and PEG 1000 and at ∼2 mM Mg2+ with EG. Because EG shows optimal activity at 2 mM Mg2+, all experiments with EG (except for the screening experiment in Figure S1 and the ligation experiment at 55 °C) were performed at 2 mM Mg2+. Previous studies have found a decrease in Mg2+ requirement for ribozyme activity to accompany a decrease in Mg2+ requirement for folding in both the group II intron20 and 1672 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 3. Molecular crowding counteracts loss of ligase 3 ribozyme activity under denaturing conditions. Crowding rescues the loss of ligation activity induced by (A) molar concentrations of urea and (B) alkaline pH. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0, 1 mM MgCl2. Reactions contained additives (PEG 200, PEG 1000, or PEG 8000) and urea (1 or 2.5 M) as indicated. the HDV26 ribozymes, supporting a role of crowding in facilitating the formation of catalytically relevant folds. An alternative explanation to the induction of RNA folding is that the addition of cosolutes like PEG may alter solution properties such as dielectric constant and water activity which result in greater association between RNA and Mg2+. An RNA-bound Mg2+ ion may activate the nucleophile at the site of ligation or stabilize the transition state.27 We tested nonenzymatic ligation in the presence of EG and various PEGs at 1 mM Mg2+ using a FAM-labeled primer (corresponding to the 3′ end sequence of the ligase downstream of the linker, Table S1), the 2AI-activated RNA substrate, and an appropriate RNA template. Ligation yields were unaffected in the presence of EG or PEGs (Figure S3), supporting the importance of ribozyme structure in crowding-induced rate enhancement. To ask if the crowding-induced stimulation of ribozyme- catalyzed phosphorimidazolide ligation was specific to ligase 1 or was more general, we tested the activity of another ligase ribozyme (henceforth, ligase 2) identified from our previous in vitro selection experiment (Table S1).6 Although distinct in sequence and structure, ligase 2 exhibited a similar response to crowding as ligase 1. While ∼6% ligation was observed in the absence of crowding agents after 3 h at 1 mM Mg2+, crowding increased ligation yields up to ∼60%, which was comparable to the ligation yield at 10 mM Mg2+ in the absence of crowding agents. The rates of ligase 2-catalyzed ligation followed a similar trend (Figure S4). Crowding Protects Ligase Ribozyme from Denatura- tion. Since crowding promotes the formation of compact RNA folds, we wondered if molecular crowding could protect ribozymes from unfolding under denaturing conditions at the low Mg2+ concentrations that are compatible with fatty acid- based protocell membranes. As ligase 1 and ligase 2 are inactive at low Mg2+ in the absence of crowding agents, these ribozymes cannot be used to capture the detrimental effects of denaturants or the protective effects of crowding in the presence of denaturants under these low-Mg2+ conditions. Therefore, we used a previously reported 2AI-ligase (hence- forth, ligase 3) that is functional under these conditions for the following experiments.11 First, we tested RNA ligation by ligase 3 in the presence of urea, which is an effective denaturant of RNA and also an important precursor molecule in the prebiotic syntheses of ribonucleotides and amino acids.28 As ligation rates in the background of 1 mM Mg2+ expected, respectively, decreased by ∼6-fold in the presence of 1 M urea, and ligation was further reduced in the presence of 2.5 M urea (Figure 3A, Figure S5A). Next, we tested the stabilizing effects of PEG 200, a low-MW crowder, and PEG 1000, a high-MW crowder, in the presence of urea. Ligation in 1 M urea was restored upon addition of 30% PEG 200 and 19% PEG 1000 (Figure 3A). Ligase 3 was even active in 2.5 M urea in the presence of PEG 200 and PEG 1000 with rate enhancements of 25-fold and 33- fold, relative to solutions without crowding agents. Interestingly, EG did not show any ligation rescue under these partially denaturing conditions (Figure S5B). Ribozyme activity in the presence of molar concentrations of urea is consistent with the stabilization of compact, solvent- excluded RNA tertiary structures by crowding agents.26 We suggest that polymeric crowders such as polypeptides or polyesters or even “proto-peptides” such as depsipeptides that contain a mixture of amide and ester linkages, if present in sufficient concentrations in prebiotic environments, could have shielded catalytic RNA structures from nonspecific denatura- tion by molecules such as urea and formamide.29 Alkaline pH, which can be beneficial for certain prebiotic processes such as the synthesis of sugars28 and RNA strand separation,30 is detrimental to the chemical stability of RNA. However, compact folded RNAs are more resistant to alkaline degradation than their unfolded counterparts. Encouraged by the protective effect of crowding in the presence of urea, we measured the activity of ligase 3 at pH 10 and pH 11 in crowded solutions. No ligation was observed at pH 11 in the presence or absence of crowders. A small amount of ligated product was detected at pH 10 in the absence of crowding agents with an 11-fold reduction in reaction rate relative to that at pH 8 (kobs values of 0.1 h−1 at pH 10 vs 1.1 h−1 at pH 8). We tested ligation at pH 10 with different crowders. Low-MW crowders like EG and PEG 200 showed no benefit; however, the loss of ligase activity at pH 10 was less pronounced in the presence of high-MW crowders PEG 1000 and PEG 8000 with only a 2.6-fold and 2.3-fold reduction in kobs, respectively, relative to their values at pH 8 (Figure 3B, Figure S6A). This represents a 4−5-fold rate enhancement ribozyme- catalyzed ligation at pH 10 upon crowding (Figure 3B). Although ligase activity was rescued in the presence of crowders, crowding had a minimal effect on the extent of RNA degradation at pH 10 or pH 11. Therefore, the beneficial effect of crowders may result from the protection of the catalytic fold from disruption at alkaline pH or by preserving for 1673 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Figure 4. RNA ligation catalyzed by the ligase 1 ribozyme is stimulated in the presence of prebiotic molecules. (A) Ligation yields after 3 h at 1 mM Mg2+ in the presence of ribose and prebiotic amino acids. (B) Ligation rates at 1 mM Mg2+ in the presence of ribose and prebiotic amino acids. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0 and 1 mM or 10 mM MgCl2. Reactions contained additives (3.8% (w/v) D-ribose, 2.5 mM individual amino acid, or 2.5 mM amino acid mixture) as indicated. base-pairing interactions between the substrate, template, and ribozyme. We asked whether crowding could have had a similar protective function during fluctuating temperature cycles on the early Earth. Only a modest enhancement in ligation rates by ligase 3 was observed at 55 °C in the presence of EG, PEG 400, PEG 1000, and PEG 8000 (Figure S6B and S6C). The lack of substantial benefit from crowding at high temperatures is consistent with UV melting experiments with the ligase 1 ribozyme, which revealed a negligible increase (ΔTm = 0.5 °C) in its thermal stability in the presence of high-MW crowder, PEG 1000 (Figure S7A and S7B). EG, on the other hand, caused a 4 °C decrease in Tm (Figure S7A and S7B). A similar decrease in Tm value in the presence of EG has been observed with the hammerhead ribozyme, which we speculate could be due to a destabilization of base-paired helices.31 Crowding Stimulates RNA-Catalyzed RNA Polymer- ization at Low Mg2+ Concentrations. Although ribozymes that catalyze the template-directed polymerization of nucleo- side phosphorimidazolides have not yet been reported, polymerase ribozymes that use NTPs as substrates have been evolved from the class I ligase ribozyme.9,10,32 These ribozymes generally require 50−200 mM Mg2+, which makes them incompatible with fatty acid vesicle-based models for primitive cells. Tagami et al. demonstrated modest polymerase function at 10 mM free Mg2+ in the presence of lysine decapeptide (K10), which enabled RNA-catalyzed RNA polymerization within Mg2+-resistant 1-palmitoyl-2-oleoylphosphatidylcholine (POPC) vesicles.33 Similarly, Takahashi et al. demonstrated the addition of up to 5 nucleotides by the tC9Y polymerase ribozyme in the presence of 10 mM Mg2+ upon addition of 20% PEG 200.34 We tested the ability of the 38−6 polymerase ribozyme10 to extend a 10 nt RNA primer on a 21 nt RNA template in the presence of 5 mM Mg2+ in solutions containing low- or high-MW PEGs. Negligible extension beyond +4 was observed in the absence of crowding agents; however, small amounts of full-length products (+11) were detected in the presence of PEG 200 or PEG 1000 after 24 h. The prominent +1 extension product increased from 24% without crowding agents to 33% and 40% in the presence of PEG 200 and PEG 1000, respectively. While only 26% of the primer was extended in the absence of crowding agents, 37% and 43% of the primer was extended in the presence of PEG 200 and PEG 1000, respectively (Figure S8). Enhancement of ribozyme polymer- ase activity at low millimolar Mg2+ underscores the generality of the beneficial effects of crowded environments on ribozyme- catalyzed RNA assembly. Interestingly, molecular crowding has also been found to enhance the polymerization of NTPs35 and dNTPs36 by biologically derived protein polymerases, which further supports the role of crowding in facilitating nucleic acid assembly. Prebiotically Relevant Small Molecules Enable Ribo- zyme-Catalyzed RNA Ligation at Low Mg2+. While our observations on the effects of molecular crowding agents on ribozyme activity are promising, the above results were obtained with prebiotically irrelevant synthetic PEG molecules with the exception of EG. Therefore, we explored the potential of prebiotically relevant small molecules for stimulating ribozyme-ligase activity. Considering the importance of simple sugars in a pre-RNA/RNA world and the stabilizing effect of ribose on fatty acid membranes, we decided to explore the effect of ribose on ligase 1 ribozyme activity.28,37,38 We also tested a subset of amino acids thought to be available on early Earth as products of prebiotic synthetic pathways such as the cyanosulfidic protometabolic reaction network.39 Ribose at 2% (w/v) and 3.8% (w/v) increased ligation yield from ∼1% to ∼11% and ∼26% after 3 h in the presence of 1 mM Mg2+ with kobs values of ∼0.4 and ∼0.5 h−1, respectively (Figure S9, Figure 4). We also screened the amino acids glycine, alanine, proline, leucine, serine, and aspartic acid at 2.5, 5, 10, and 20 mM concentrations for their ability to stimulate ligation at 1 mM Mg2+. All of the above amino acids were found to stimulate ligation regardless of their concentrations with yields of 25−35% after 3 h (Figure S10). As lower concentrations are prebiotically more likely in most microenvironments, we measured the yield and rate of ligase 1-catalyzed RNA ligation in the presence of 2.5 mM of each amino acid and a mixture of all six amino acids at a total concentration of 2.5 mM (Figure 4). The presence of amino acids both individually and as a mixture rescued ligation rates to within a factor of 1.4−3.8 of that observed at 10 mM Mg2+ without any additive (Figure 4B). As ligase 1 exhibits negligible ligation at 1 mM Mg2+ even in the presence of high concentrations of Na+ (300 mM),6 low in reactions concentrations of monovalent counterions containing 2.5−20 mM amino acids are unlikely to cause 1674 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article this pronounced rate stimulation, and the amino acids must be playing a direct role. The mechanism for ribozyme activation at low Mg2+ by ribose or amino acids is not clear. Aliphatic alcohols such as methanol, ethanol, propanol, 2-methoxyethanol, and propane- 1,3-diol stimulate hammerhead catalysis at 1 mM Mg2+ by decreasing the dielectric constant of the solution, thereby enhancing interactions between the ribozyme and Mg2+.31 Ribose-mediated enhancement of ribozyme-catalyzed RNA ligation could be a result of similar solution-level effects. The beneficial effect of amino acids toward ribozyme activity has It was been previously observed for RNA self-cleavage. proposed that the increase in ribozyme activity resulted from structural compaction of the RNA, which allowed greater sampling of its catalytic fold.40 This assertion was supported by thermal denaturation and SAXS studies. Amino acids may stimulate RNA folding by altering solvent properties like dielectric constant or water activity.17 Additionally, as amino acids can weakly chelate Mg2+, the chelated amino acids may form a layer on the RNA surface, increasing the local concentration of Mg2+, which may lead to improved folding and catalysis.40,41 Regardless, this ability of prebiotic small molecules to facilitate ribozyme-catalyzed RNA assembly presents a “systems”-level solution for lowering the Mg2+ requirement for this central process in primordial biochemis- try. ■ CONCLUSION results the emergence of RNA-based cellular The crucial role of Mg2+ in both nonenzymatic- and ribozyme- catalyzed RNA replication coupled with its ability to accelerate RNA degradation and destabilize fatty acid protocells presents a puzzle for life. Therefore, exploring scenarios that mitigate this “Mg2+ problem” is of critical importance. The low Mg2+ requirement for natural ribozymes that function within crowded cellular environments inspired us to study molecular crowding as a general solution to the Mg2+ problem in the context of ribozyme-catalyzed RNA assembly. Our show a dramatic stimulation of ribozyme-catalyzed assembly of 2AI- activated RNA oligomers and nucleoside triphosphates at low millimolar Mg2+ by prebiotically relevant amino acids, ribose, ethylene glycol, and polyethylene glycols of various MWs (200−8000). The beneficial effects of amino acids, ribose, and ethylene glycol are especially notable since these molecules can be synthesized abiotically and therefore were likely to have been present in early Earth environments. The 3−10-fold lower Mg2+ requirement for ligase ribozymes in the presence of such solutes likely stems from enhanced RNA folding in “crowded” solutions as the corresponding nonenzymatic ligation reaction was not affected by crowding. Stimulation of catalytic activity in the presence of molecular crowding has been reported for other ribozymes.17 Since the crowding- induced enhancement of RNA assembly was largely independent of crowder size, ribozyme folding could be favored by an interplay of both enthalpic forces arising from interactions between the RNA surface and the crowder and entropic forces arising from volume exclusion.17 In most cases where both low-MW and high-MW crowders affect macro- molecular function, it is extremely difficult to delineate the individual contributions of volume exclusion and the various enthalpic forces that are always at play.17 Further studies may help isolate the effects of these distinct thermodynamic forces. We demonstrated that in addition to enabling RNA ligation in the low-Mg2+ regime, crowding offers modest to significant protection to ligase ribozymes under various denaturing conditions relevant to early Earth environments. The ability to function under conditions that favor the disruption of RNA secondary structure could have been important for rapid RNA- catalyzed RNA replication, which requires the separation of newly synthesized RNA strands from their RNA templates while preserving catalytic RNA structures. Efficient RNA assembly, at low Mg2+ concentrations, presents a path to reconcile ribozyme function with the stability of protocell membranes made of fatty acids. Protocells crowded with prebiotic small molecules like sugars, alcohols, and amines and polymeric species such as short oligonucleo- tides or polypeptides could potentially support a wide range of the low Mg2+ concentrations ribozyme activities under is required for maintaining membrane integrity. This particularly interesting in the context of our earlier observation that prebiotically relevant small molecules including ribose also reduce RNA leakage from fatty acid vesicles.11 The combined effect of enhancing ribozyme function under low Mg2+ conditions and stabilizing protocell membranes against Mg2+ suggests a potential role for these prebiotic molecules that is separate from their roles as components of the building blocks of life. By providing a general mechanism to activate RNA low Mg2+ concentration, molecular crowding catalysis at expands the range of environments in which ribozymes can function to less salty environments such as freshwater bodies12 and increases the likelihood of the emergence of active ribozymes from the RNA sequence space.21 Suboptimal sequences that would otherwise not be selected in low-Mg2+ environments could emerge in crowded milieus, potentially creating neutral mutational pathways that would facilitate ribozyme evolution and therefore increase the catalytic diversity of the RNA world. ■ EXPERIMENTAL PROCEDURES the 3′ end of RNA Preparation and Substrate Activation. Ribozymes were prepared by in vitro transcription of PCR-generated dsDNA templates containing 2′-O-methyl modifications to reduce transcriptional heterogeneity at the RNA42 (Table S1). Transcription reactions contained 40 mM Tris-HCl (pH 8), 2 mM spermidine, 10 mM NaCl, 25 mM MgCl2, 10 mM dithiothreitol (DTT), 30 U/mL RNase inhibitor murine (NEB), 2.5 U/mL thermostable inorganic pyrophosphatase (TIPPase) (NEB), a 4 mM concentration of each NTP, 30 pmol/mL DNA template, and 1U/μL T7 RNA Polymerase (NEB) and were incubated for 3 h at 37 °C. DNA template was digested by DNase I (NEB) treatment, and RNA was extracted with phenol−chloroform−isoamyl alcohol (PCI), ethanol precipitated, and purified by denaturing PAGE. Ligation templates, FAM-labeled primers, and ssDNA were purchased from Integrated DNA Technologies. The 5′-monophosphorylated oligonucleotide corresponding to the substrate sequence was activated by incubating it with 0.2 M 1-ethyl-3-(3 dimethylaminopropyl) carbodiimide (HCl salt) and 0.6 M 2-aminoimidazole (HCl salt, pH adjusted to 6) for 2 h at room temperature. The reaction was washed with water in Amicon Ultra spin columns (3 kDa cutoff) 4−5 times (200 μL of water per wash) and purified by reverse-phase analytical HPLC using a gradient from 98% to 75% 20 mM TEAB (triethylamine bicarbonate, pH 8) versus acetonitrile over 40 min.6 1675 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article Ligation Assays. Ligation reactions contained 1 μM ribozyme, 1.2 μM RNA template, and 2 μM 2-AI-activated RNA substrate in 100 mM Tris-HCl pH 8.0, the indicated concentrations of MgCl2, and crowding agents. All reactions were performed at room temperature unless mentioned otherwise. Aliquots were quenched with 5 volumes of quench buffer (8 M urea, 100 mM Tris-Cl, 100 mM boric acid, 100 mM EDTA) and analyzed by denaturing PAGE. Gels were stained using SYBR Gold,43 imaged on an Amersham Typhoon RGB instrument (GE Healthcare), and analyzed in Image- Quant IQTL 8.1. Intensities corresponding to the ligated product were normalized to account for the difference in size between the 95 nt precursor band and the 111 nt product band. Kinetic data were nonlinearly fitted to the modified first- order rate equation, y = A(1 − e−kx), where A represents the fraction of active complex, k is the first-order rate constant, x is time, and y is the fraction of ligated product in GraphPad Prism 9. For nonenzymatic ligation, a 5′-FAM-labeled RNA primer corresponding to the last 8 nt of the ribozyme sequence was used instead of the ribozyme, and the gel was directly imaged. Ligation Assays under Denaturing Conditions. All ligation assays under denaturing conditions were performed with the ligase 3 ribozyme, which retains activity under low [Mg2+]. Ligation at High pH. Ribozyme and template were heated at 95 °C for 2 min in the absence of any buffer and cooled to room temperature. CAPS buffer (pH 10 or 11) was added to a final concentration of 100 mM in the absence or presence of crowding agents (19% PEG 1000 or 19% PEG 8000) and 1 mM MgCl2. The substrate was added immediately after the addition of MgCl2 to initiate ligation. Ligation at High Temperatures. Reactions with or without crowding agents (10% ethylene glycol, 30% PEG 400, 19% PEG 1000, or 19% PEG 8000) were incubated at 55 °C after initiating ligation by adding the substrate. Ligation in the Presence of Urea. A 10 M concentration of urea was added to final concentrations of 1 or 2.5 M after refolding in the presence of crowding agents (30% PEG 200 or 19% PEG 1000) and 1 mM MgCl2 to minimize degradation at high temperatures required for refolding. The substrate was added immediately after the addition of MgCl2 to initiate ligation. Ribozyme-Catalyzed NTP Polymerization Assays. A FAM-labeled RNA primer (80 nM), RNA template (100 nM), and polymerase ribozyme (100 nM) were heated in the absence and presence of crowding agents and 25 mM Tris·HCl pH 8 at 80 °C for 30 s and cooled to 17 °C at a gradient of 0.1 °C/s. MgCl2 was added to final concentrations of 5 and 200 followed by a 0.5 mM concentration of each NTP. mM, Reactions were incubated at 17 °C for 24 h, and 1 μL aliquots were quenched with 7 μL of quench buffer (8 M urea, 100 mM Tris-Cl, 100 mM boric acid, 100 mM EDTA containing 5 μM DNA oligo complementary to template). Reactions were analyzed by denaturing PAGE. Gels were imaged on an Amersham Typhoon RGB instrument (GE Healthcare) and analyzed in ImageQuant IQTL 8.1. UV Melting Analysis of Ligase Ribozyme. UV melting experiments were performed to determine the thermal stability of the ligase 1 ribozyme in the absence of presence of low- and high-MW crowding agents according to the protocol used by Struslon et al.44 Briefly, 0.5 μM ribozyme was incubated at 95 °C for 2 min in 10 mM sodium cacodylate buffer (pH 7) and refolded in the absence or presence of crowding agents (10% ethylene glycol or 19% PEG 1000) in the presence of 1 mM MgCl2 by heating the solution to 55 °C for 10 min followed by cooling to room temperature for 10 min. A Cary UV−vis multicell Peltier spectrophotometer was used for melting experiments. Absorbance was recorded at 260 nm every minute between 20 and 90 °C. Data was normalized with respect to “buffer only” sample in each case, which contained all components in the experimental sample except RNA. Derivative plots of normalized data (dA/dT) vs T) and melting temperatures (Tm) were obtained by the instrument’s default software. ■ ASSOCIATED CONTENT sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscentsci.3c00547. Supplementary table with RNA oligonucleotides used in this work; identification of optimal crowding conditions for ligase 1-catalyzed RNA ligation; representative gels illustrating the Mg2+ dependence of ligase 1 ribozyme- catalyzed RNA ligation in the absence and presence of crowding agents; nonenzymatic ligation is not influenced by molecular crowding; ligase 2 activity at low Mg2+ concentrations is rescued by crowding agents; effect of molecular crowding on ligase 3-catalyzed RNA ligation in the presence of urea; effect of molecular crowding on ligase 3-catalyzed RNA ligation under alkaline pH and high temperature; effect of molecular crowding on the thermal stability of the ligase 1 ribozyme; RNA-catalyzed polymerization of NTPs at low Mg2+ concentration; ribozyme-catalyzed RNA ligation is stimulated in the presence of ribose; ribozyme-catalyzed RNA ligation is stimulated in the presence of prebiotically relevant amino acids (PDF) Transparent Peer Review report available (PDF) ■ AUTHOR INFORMATION Corresponding Authors Saurja DasGupta − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Howard Hughes Medical Institute, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Genetics, Harvard Medical School, Boston, Massachusetts 02115, United States; 0000-0002-9064-9131; Email: dasgupta@ molbio.mgh.harvard.edu orcid.org/ Jack W. Szostak − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Howard Hughes Medical Institute, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Genetics, Harvard Medical School, Boston, Massachusetts 02115, United States; Department of Chemistry and Chemical Biology, Harvard University, Cambridge, Massachusetts 02138, United States; Present Address: Howard Hughes Medical Institute, Department of Chemistry, University of Chicago, Chicago, Illinois 60637, United States.; 1203; Email: [email protected] orcid.org/0000-0003-4131- 1676 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 http://pubs.acs.org/journal/acscii Article ACS Central Science Author Stephanie Zhang − Department of Molecular Biology, Center for Computational and Integrative Biology, Massachusetts General Hospital, Boston, Massachusetts 02114, United States; Department of Chemistry and Chemical Biology, Harvard University, Cambridge, Massachusetts 02138, United States; Present Address: Department of Pathology, Brigham and Women’s Hospital, Boston, Massachusetts 02115, United States Complete contact information is available at: https://pubs.acs.org/10.1021/acscentsci.3c00547 Notes The authors declare no competing financial interest. ■ ACKNOWLEDGMENTS J.W.S. is an Investigator of the Howard Hughes Medical Institute. This work was supported in part by a grant from the Simons Foundation (290363) to J.W.S. ■ REFERENCES (1) Gilbert, W. The RNA World. Nature 1986, 319, 618. (2) DasGupta, S.; Piccirilli, J. A. The Varkud Satellite Ribozyme: A Journey through Biochemistry, Crystallography, and Thirty-Year Computation. Acc. Chem. Res. 2021, 54 (11), 2591−2602. (3) Lee, K. Y.; Lee, B. J. Structural and Biochemical Properties of Novel Self-Cleaving Ribozymes. Molecules 2017, 22 (4), 678. (4) Ren, A.; Micura, R.; Patel, D. J. Structure-based mechanistic insights into catalysis by small self-cleaving ribozymes. Curr. Opin Chem. Biol. 2017, 41, 71−83. (5) Joyce, G. F.; Szostak, J. W. Protocells and RNA Self-Replication. Cold Spring Harb. Perspect. Biol. 2018, 10 (9), a034801. (6) Walton, T.; DasGupta, S.; Duzdevich, D.; Oh, S. S.; Szostak, J. W. In vitro selection of ribozyme ligases that use prebiotically plausible 2-aminoimidazole-activated substrates. Proc. Natl. Acad. Sci. U. S. A. 2020, 117 (11), 5741−5748. (7) AbouHaidar, M. G.; I. G. Non-enzymatic RNA Ivanov, hydrolysis promoted by the combined catalytic activity of buffers and magnesium ions. Z. Naturforsch C J. Biosci 1999, 54 (7−8), 542− 548. (8) Bartel, D. P.; Szostak, J. W. Isolation of new ribozymes from a large pool of random sequences [see comment]. Science 1993, 261 (5127), 1411−1418. (9) Attwater, J.; Wochner, A.; Holliger, P. In-ice evolution of RNA polymerase ribozyme activity. Nat. Chem. 2013, 5 (12), 1011−1018. (10) Tjhung, K. F.; Shokhirev, M. N.; Horning, D. P.; Joyce, G. F. An RNA polymerase ribozyme that synthesizes its own ancestor. Proc. Natl. Acad. Sci. U. S. A. 2020, 117 (6), 2906−2913. (11) DasGupta, S.; Zhang, S. J.; Smela, M. P.; Szostak, J. W. RNA- Catalyzed RNA Ligation within Prebiotically Plausible Model Protocells. Chem. Eur. J. 2023, No. e202301376. (12) Maurer, S. The Impact of Salts on Single Chain Amphiphile Membranes and Implications for the Location of the Origin of Life. Life (Basel) 2017, 7 (4), 44. (13) Leamy, K. A.; Assmann, S. M.; Mathews, D. H.; Bevilacqua, P. C. Bridging the gap between in vitro and in vivo RNA folding. Q. Rev. Biophys. 2016, 49, No. e10. (14) Fulton, A. B. How crowded is the cytoplasm? Cell 1982, 30 (2), 345−347. (15) Minton, A. P. The influence of macromolecular crowding and macromolecular confinement on biochemical reactions in physio- logical media. J. Biol. Chem. 2001, 276 (14), 10577−10580. (16) Zhou, H. X.; Rivas, G.; Minton, A. P. Macromolecular crowding and confinement: biochemical, biophysical, and potential physio- logical consequences. Annu. Rev. Biophys 2008, 37, 375−397. (17) DasGupta, S. Molecular crowding and RNA catalysis. Org. Biomol Chem. 2020, 18 (39), 7724−7739. (18) Dupuis, N. F.; Holmstrom, E. D.; Nesbitt, D. J. Molecular- crowding effects on single-molecule RNA folding/unfolding thermo- dynamics and kinetics. Proc. Natl. Acad. Sci. U. S. A. 2014, 111 (23), 8464−8469. (19) Paudel, B. P.; Rueda, D. Molecular crowding accelerates ribozyme docking and catalysis. J. Am. Chem. Soc. 2014, 136 (48), 16700−16703. (20) Paudel, B. P.; Fiorini, E.; Borner, R.; Sigel, R. K. O.; Rueda, D. S. Optimal molecular crowding accelerates group II intron folding and maximizes catalysis. Proc. Natl. Acad. Sci. U. S. A. 2018, 115 (47), 11917−11922. (21) Saha, R.; Pohorille, A.; Chen, I. A. Molecular crowding and early evolution. Orig Life Evol Biosph 2014, 44 (4), 319−324. (22) Karsili, T. N. V.; Fennimore, M. A.; Matsika, S. Electron- induced origins of prebiotic building blocks of sugars: mechanism of self-reactions of a methanol anion dimer. Phys. Chem. Chem. Phys. 2018, 20 (18), 12599−12607. (23) Liu, Z.; Wu, L. F.; Kufner, C. L.; Sasselov, D. D.; Fischer, W. W.; Sutherland, J. D. Prebiotic photoredox synthesis from carbon dioxide and sulfite. Nat. Chem. 2021, 13 (11), 1126−1132. (24) Rivilla, V. M.; Beltrán, M. T.; Cesaroni, R.; Fontani, F.; Codella, C.; Zhang, Q. Formation of ethylene glycol and other complex organic molecules in star-forming regions. Astron. Astrophys. 2017, 598, A59. (25) Fedoseev, G.; Cuppen, H. M.; Ioppolo, S.; Lamberts, T.; Linnartz, H. Experimental evidence for glycolaldehyde and ethylene glycol formation by surface hydrogenation of CO molecules under dense molecular cloud conditions. Mon. Not. R. Astron. Soc. 2015, 448 (2), 1288−1297. (26) Strulson, C. A.; Yennawar, N. H.; Rambo, R. P.; Bevilacqua, P. C. Molecular crowding favors reactivity of a human ribozyme under ionic conditions. Biochemistry 2013, 52 (46), 8187− physiological 8197. (27) Shechner, D. M.; Grant, R. A.; Bagby, S. C.; Koldobskaya, Y.; Piccirilli, J. A.; Bartel, D. P. Crystal structure of the catalytic core of an RNA-polymerase ribozyme. Science 2009, 326 (5957), 1271−1275. (28) Yadav, M.; Kumar, R.; Krishnamurthy, R. Chemistry of Abiotic Nucleotide Synthesis. Chem. Rev. 2020, 120 (11), 4766−4805. (29) Frenkel-Pinter, M.; Haynes, J. W.; Mohyeldin, A. M.; C, M.; Sargon, A. B.; Petrov, A. S.; Krishnamurthy, R.; Hud, N. V.; Williams, L. D.; Leman, L. J. Mutually stabilizing interactions between proto- peptides and RNA. Nat. Commun. 2020, 11 (1), 3137. (30) Mariani, A.; Bonfio, C.; Johnson, C. M.; Sutherland, J. D. pH- Driven RNA Strand Separation under Prebiotically Plausible Conditions. Biochemistry 2018, 57 (45), 6382−6386. (31) Nakano, S.; Kitagawa, Y.; Yamashita, H.; Miyoshi, D.; Sugimoto, N. Effects of Cosolvents on the Folding and Catalytic Activities of the Hammerhead Ribozyme. Chembiochem 2015, 16 (12), 1803−1810. (32) Horning, D. P.; Joyce, G. F. Amplification of RNA by an RNA polymerase ribozyme. Proc. Natl. Acad. Sci. U. S. A. 2016, 113 (35), 9786−9791. (33) Tagami, S.; Attwater, J.; Holliger, P. Simple peptides derived from the ribosomal core potentiate RNA polymerase ribozyme function. Nat. Chem. 2017, 9 (4), 325−332. (34) Takahashi, S.; Okura, H.; Sugimoto, N. Bisubstrate Function of RNA Polymerases Triggered by Molecular Crowding Conditions. Biochemistry 2019, 58 (8), 1081−1093. (35) Takahashi, S.; Okura, H.; Chilka, P.; Ghosh, S.; Sugimoto, N. Molecular crowding induces primer extension by RNA polymerase through base stacking beyond Watson-Crick rules. RSC Adv. 2020, 10 (55), 33052−33058. (36) Takahashi, S.; Herdwijn, P.; Sugimoto, N. Effect of Molecular Crowding on DNA Polymerase Reactions along Unnatural DNA Templates. Molecules 2020, 25 (18), 4120. (37) Furukawa, Y.; Chikaraishi, Y.; Ohkouchi, N.; Ogawa, N. O.; Glavin, D. P.; Dworkin, J. P.; Abe, C.; Nakamura, T. Extraterrestrial 1677 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678 ACS Central Science http://pubs.acs.org/journal/acscii Article ribose and other sugars in primitive meteorites. Proc. Natl. Acad. Sci. U. S. A. 2019, 116 (49), 24440−24445. (38) Meinert, C.; Myrgorodska, I.; de Marcellus, P.; Buhse, T.; Nahon, L.; Hoffmann, S. V.; d’Hendecourt, L. L. S.; Meierhenrich, U. J. Ribose and related sugars from ultraviolet irradiation of interstellar ice analogs. Science 2016, 352 (6282), 208−212. (39) Wu, L. F.; Sutherland, J. D. Provisioning the origin and early evolution of life. Emerg Top Life Sci. 2019, 3 (5), 459−468. (40) Yamagami, R.; Bingaman, J. L.; Frankel, E. A.; Bevilacqua, P. C. Cellular conditions of weakly chelated magnesium ions strongly promote RNA stability and catalysis. Nat. Commun. 2018, 9 (1), 2149. (41) Yamagami, R.; Huang, R.; Bevilacqua, P. C. Cellular Concentrations of Nucleotide Diphosphate-Chelated Magnesium Ions Accelerate Catalysis by RNA and DNA Enzymes. Biochemistry 2019, 58 (38), 3971−3979. (42) Kao, C.; Zheng, M.; Rudisser, S. A simple and efficient method to reduce nontemplated nucleotide addition at the 3 terminus of RNAs transcribed by T7 RNA polymerase. RNA 1999, 5 (9), 1268− 1272. (43) Guillen, D.; Schievelbein, M.; Patel, K.; Jose, D.; Ouellet, J. A simple and affordable kinetic assay of nucleic acids with SYBR Gold gel staining. PLoS One 2020, 15 (3), No. e0229527. (44) Strulson, C. A.; Boyer, J. A.; Whitman, E. E.; Bevilacqua, P. C. Molecular crowders and cosolutes promote folding cooperativity of RNA under physiological ionic conditions. RNA 2014, 20 (3), 331− 347. 1678 https://doi.org/10.1021/acscentsci.3c00547 ACS Cent. Sci. 2023, 9, 1670−1678
10.1016_j.cell.2023.05.028	A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t HHS Public Access Author manuscript Cell. Author manuscript; available in PMC 2023 August 18. Published in final edited form as: Cell. 2023 June 22; 186(13): 2765–2782.e28. doi:10.1016/j.cell.2023.05.028. DNA hypomethylation silences antitumor immune genes in early prostate cancer and CTCs Hongshan Guo1,2,11,13, Joanna A. Vuille1,13, Ben S. Wittner1, Emily M. Lachtara1, Yu Hou3,4,11, Maoxuan Lin1,4, Ting Zhao1,5, Ayush T. Raman1,4, Hunter C. Russell1, Brittany A. Reeves1, Haley M. Pleskow1,6, Chin-Lee Wu1,5, Andreas Gnirke4, Alexander Meissner4,7, Jason A. Efstathiou1,6, Richard J. Lee1,8, Mehmet Toner9,10, Martin J. Aryee1,5,12, Michael S. Lawrence1,4,5, David T. Miyamoto1,6,, Shyamala Maheswaran1,9,, Daniel A. Haber1,2,8,14,* 1.Massachusetts General Hospital Cancer Center, Harvard Medical School, Charlestown, MA 02129, USA. 2.Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA. 3.Evergrande Center for Immunologic Diseases, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA. 4.Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA. 5.Department of Pathology, Massachusetts General Hospital, Harvard Medical School, Boston, MA 02114, USA. 6.Department of Radiation Oncology, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA 02129, USA. 7.Department of Genome Regulation, Max Planck Institute for Molecular Genetics, Berlin 14195, Germany. This work is licensed under a Creative Commons Attribution 4.0 International License, which allows reusers to distribute, remix, adapt, and build upon the material in any medium or format, so long as attribution is given to the creator. The license allows for commercial use. Correspondence: [email protected] (D.T.M.), [email protected] (S.M.), [email protected] (D.A.H.). Author contributions H.G., J.A.V., D.T.M., S.M. and D.A.H. conceived the project, provided leadership for the project and drafted the manuscript. H.G., J.A.V., Y.H., T.Z., H.C.R., B.A.R., H.M.P., C.W., J.A.E., R.J.L., M.T. and D.T.M. conducted all the experiments. H.G., B.S.W., E.M.L., M.L., A.T.R., A.G., A.M., M.S.L., and M.J.A analyzed all the data. All authors reviewed and edited the manuscript. Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Declaration of interests Massachusetts General Hospital (MGH) has applied for patents regarding the CTC-iChip technology and CTC detection signatures. M.T., S.M. and D.A.H. are cofounders and have equity in Tell-Bio, which is not related to this work. The interests of these authors were reviewed and managed by MGH and Mass General Brigham (MGB) in accordance with their conflict of interest policies. All other authors declare no competing interests. Inclusion and Diversity We support inclusive, diverse, and equitable conduct of research. Key Ressource Table (see document) A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 2 8.Department of Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA 02114, USA. 9.Department of Surgery, Massachusetts General Hospital, Harvard Medical School, Boston, MA 02114, USA. 10.Center for Engineering in Medicine and Shriners Hospital for Children, Harvard Medical School, Boston, MA 02114, USA. 11.Present address: Bone Marrow Transplantation Center, First Affiliated Hospital, Zhejiang University School of Medicine and Liangzhu Laboratory, Zhejiang University Medical Center, Hangzhou, 310012, China. 12.Present address: Department of Data Science, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA 02114, USA. 13.These authors contributed equally. 14.Lead contact. Summary Cancer is characterized by hypomethylation-associated silencing of large chromatin domains, whose contribution to tumorigenesis is uncertain. Through high-resolution genome-wide single-cell DNA methylation sequencing, we identify 40 core domains that are uniformly hypomethylated from earliest detectable stages of prostate malignancy through metastatic Circulating Tumor Cells (CTCs). Nested among these repressive domains are smaller loci with preserved methylation that escape silencing and are enriched for cell proliferation genes. Transcriptionally silenced genes within the core hypomethylated domains are enriched for immune-related genes; prominent among these is a single gene cluster harboring all five CD1 genes that present lipid antigens to NKT cells, and four IFI16-related interferon-inducible genes implicated in innate immunity. Re-expression of CD1 or IFI16 murine orthologs in immunocompetent mice abrogates tumorigenesis, accompanied by activation of anti-tumor immunity. Thus, early epigenetic changes may shape tumorigenesis, targeting co-located genes within defined chromosomal loci. Hypomethylation domains are detectable in blood specimens enriched for CTCs. Graphical Abstract Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 3 In Brief Analysis of circulating tumor cluster cells reveals how DNA hypomethylation during early prostate tumorigenesis silences immune surveillance genes, while sparing proliferation-associated genes. Keywords DNA hypomethylation; prostate cancer; circulating tumor cells; immune surveillance; single-cell sequencing Introduction Cancer is characterized by two primary changes at the level of DNA methylation1–4. Focal hypermethylation of CpG islands, often located within gene regulatory regions, results in gene silencing, a well-established mechanism for inactivation of tumor suppressor genes5–7. In addition, long-range hypomethylated regions, Partially Methylated Domains (PMDs), coincide with nuclear Lamina-Associated Domains (LADs) and Large Organized Chromatin lysine (K) (LOCK) domains8–10. These chromosomal loci are large (>100 kb), gene-poor, correlated with late-replicating DNA, and topologically associated with nuclear lamina. Repetitive sequences and retro-elements residing within PMDs may be de-repressed in cancer, but the rare protein encoding genes are silenced11. Two repression-associated chromatin modifications are evident: H3K9me3 is abundant within hypomethylated blocks, while H3K27me3 denotes their boundaries12,13. Conflicting models have suggested that Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 4 hypomethylated blocks are either a direct consequence of cell transformation14, or an incidental result of excessive cell proliferation13,15. The functional consequences of hypomethylation-associated gene silencing, and potential selection pressures that shape such domains, are not well understood. A recent study of advanced colon cancers proposed an intrinsic tumor suppressive mechanism that may counter cell proliferation13, although genome-wide hypomethylation is extensive in advanced cancers and may not reveal specific targets contributing to early tumorigenesis. Prostate cancer is noteworthy for its characteristically slow evolution from precancerous lesions with low levels of cell proliferation to more invasive, and ultimately metastatic malignancy. Localized prostate cancer may be classified as indolent (Gleason score (GS) 6) or clinically significant (GS≥7) based on histological grade, reflecting differences in differentiation, proliferative index, and metastatic potential16,17. GS6 tumors are often safely monitored without therapy, while the more aggressive GS7 and higher tumors are resected surgically or treated with radiation in combination with androgen deprivation therapy. GS8–10 denotes poorly differentiated tumors with an adverse prognosis and high propensity for metastasis. Multiple heterogeneous foci of early tumors are often dispersed throughout the prostate gland, complicating bulk molecular characterization and necessitating careful dissection with single-cell analytic strategies. Conversely, advanced metastatic prostate cancer predominantly affects bone, making it difficult to perform biopsies to study disseminated tumor deposits. Circulating tumor cells (CTCs), comprising potential metastatic precursors isolated from the bloodstream, thus enable single-cell analysis of advanced prostate cancer. Immune checkpoint blockade (ICB) is generally ineffective in treating prostate cancer18–21, possibly reflecting the stroma-rich, immunosuppressive environment of primary prostate cancer, but tumor cell autonomous mechanisms may also contribute, in both primary and metastatic disease. Epigenetic changes affecting expression of immune regulatory genes and modulating the responsiveness of prostate cancer to immunological therapies have not been characterized. In addition to their biological significance, cancer-associated methylation changes are of considerable molecular diagnostic interest for blood-based cancer detection. These rely primarily on CpG island-enriched methylation within short DNA fragments (170 bp) circulating in plasma, a fraction of which are tumor-derived (ctDNA)22–24. However, among patients with localized prostate cancers, only 11.2% are detectable using plasma CpG island hypermethylation assays25, leading us to ask whether the large genomic coverage provided by hypomethylated domains within CTCs may provide complementary information. To address these questions, we first established genome-wide, high-resolution single-cell bisulfite sequencing of hypomethylated domains within individual prostate CTCs from multiple patients and cancer cell lines, identifying 40 core PMDs, shared across metastatic prostate cancers. The timing of DNA hypomethylation during prostate tumorigenesis reveals that core PMDs are hypomethylated as early as indolent GS6 tumors, identifying a single predominant genomic locus, the CD1A-IFI16 gene cluster, encompassing the entire family of CD1 lipid antigen presentation genes and multiple interferon-inducible genes implicated in innate immunity. Early hypomethylation-mediated gene silencing points to specific tumorigenic pathways with both biological and diagnostic implications. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Results Page 5 Identification of shared core PMDs and PMIs across single metastatic prostate cancer cells To characterize genome-wide DNA methylation features of single metastatic prostate cancer cells, we enriched CTCs from five patients with castration-resistant prostate cancer, all with multiple bone metastases and disease refractory to hormonal therapy and performed individual cell micromanipulation and single-cell sequencing26,27(Table S1, see Methods). We compared 44 single CTCs with 40 single cells from four prostate cancer cell lines (LNCaP, VCaP, PC3 and 22Rv1) and two non-transformed prostate epithelial cell lines (Human Prostate Epithelial Cells (HPrEC) and Benign Prostate Hypertrophy cells (BPH-1)). HPrECs represent normal prostate epithelium, while BPH-1 cells share luminal cell features with cancer precursors28–31. As control for contaminating blood cells within CTC-enriched clinical specimens, we compared single prostate cells with 13 microfluidic-processed single leukocytes (WBCs) from four age-matched healthy men. To confirm the identity of single CTCs, we adapted single-cell multiomics sequencing to enable separation of nucleus from cytoplasm in individual cells, subjecting the former to single-cell whole genome bisulfite sequencing (scBS-seq)32 and the latter to single-cell RNA-seq (SMART-seq2)33 (Figure 1A, see Methods). On average, we detected 9 million CpG sites for each single-cell DNA methylation sequencing sample, and 5,790 genes (RPM>0) for each single-cell RNA-seq library (Figures S1A and S1B). Transcriptomes of prostate CTCs confirm the expression of expected lineage-specific and epithelial transcripts, and absence of hematopoietic markers (Figures 1B and S1C). Unsupervised hierarchical clustering analysis of all single-cell RNA- seq data reveals three distinct clusters: leukocytes, normal prostate, and prostate cancer (including CTCs and prostate cancer cell lines) (Figure S1D). In addition to transcriptional confirmation, all prostate CTCs demonstrate extensive DNA copy number variations (CNV) inferred from single-cell DNA methylation sequencing (see Methods). These CNV patterns are matched with those inferred from cytoplasmic RNA-seq from the same single cells (Figures 1C, S1E and S1F). As controls, HPrEC cells and WBCs show normal diploid copy numbers (Figure 1C, see Methods). As a final test, principal component analysis (PCA) of promoter methylation patterns readily distinguishes all tumor cells from normal controls (Figure S2A). Taken all together, we applied highly stringent criteria, including both transcriptional and DNA copy number confirmation, to nominate 38/44 (86.4%) initially selected CTCs as bona fide prostate CTCs for detailed single-cell genomic analyses. We quantified methylation levels of individual cells by binning the genome into 100 kb windows: the methylation distribution of normal cells is unimodal, with a single peak near 80% methylation, whereas virtually all tumor samples exhibit a bimodal distribution, with a varying number of hypomethylated regions (Figures 1D–1F and S2B, see Methods). Overall, DNA hypomethylation constitutes 20–40% of the genome in patient-derived prostate CTCs and prostate cancer cell lines, but <2.5% in normal prostate cells or blood cells (Figure S2C). In contrast to individual CpG islands (CGIs), which often demonstrate focal hypermethylation around gene regulatory regions, the hypomethylated regions in prostate tumor cells span very large gene-poor regions, consistent with previously described PMDs. In total, we identified 1,496 PMDs with a mean size of 1.2 Mb (range 250 kb to 9.2 Mb) across the prostate cancer genome, a number consistent with previous measurements Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 6 based on bulk tumor sequencing in multiple advanced cancers8,12,34 (Figure S2D, Table S2). Notably, on a chromosome-wide view and with the high resolution afforded by single-cell methylation analysis, some PMDs are punctuated by smaller regions, where DNA methylation is retained (Figures 1D–1F). We call these Preserved Methylation Islands (PMIs, see defining criteria in Methods) (Figure S2D, Table S2). In contrast to large gene- poor PMDs, the 1,412 PMIs interspersed within hypomethylated domains are gene-rich, with sharp methylation boundaries that bracket a single gene or a small group of genes (mean PMI size 1.3Mb; range 30.8 kb to 11.1 Mb) (Figures 1F and 1G). The identification of PMIs raises the possibility that selection pressures may preserve methylation, and potentially gene expression, at a small number of genes nested within PMDs. As demonstrated in other cancers12,13,35, PMDs are gene-poor and have strong enrichment of some endogenous retroviral elements (ERVs), notably Long Terminal Repeats (LTRs). In contrast, PMIs in prostate cancer are gene-rich with relative absence of long interspersed nuclear elements (LINEs) and LTRs (Figures 1G and 1H). Previous studies show that PMDs in breast and colon cancers exhibit depletion of active chromatin marks (H3K4me1/3, H3K27ac, H3K36me3) and enrichment of repressive histone modifications, including H3K9me3 at the center of the domains and H3K27me3 at their borders12,13. To confirm these chromatin changes in prostate cancer, we used cultured cell lines, to analyze chromatin landscapes using ChIP assays. Analysis of prostate cancer cells (LNCaP and 22Rv1) confirms the differential positioning of repressive H3K9me3 marks at the center and H3K27me3 at the border of hypomethylated domains (Figures 2A, 2B and S3A). However, direct comparison of cancer cells with non-transformed prostate epithelial and basal cells (HPrEC and BPH-1) at the same PMDs indicates that changes associated with malignancy primarily relate to H3K27me3 deposition. Indeed, Cut and Run assays show profound enrichment of H3K27me3 at PMD borders in cancer cells compared with normal cells, whereas central H3K9me3 marks are abundant at these loci, but invariant between normal and cancer cells (Figures 2B–2D and S3A–S3C). Thus, hypomethylation-associated gene silencing in cancer cells is primarily correlated with the acquisition of H3K27me3 histone modification flanking these chromosomal domains. In contrast, genes within PMIs show strong enrichment for activation (H3K4me1/3, H3K27ac and H3K36me3) and absence of repression (H3K27me3 and H3K9me2/3) (Figure 2A). At the single-cell level, both PMDs and PMIs show substantial intra-patient and inter-patient heterogeneity (Figures 2E–2G and S3D–S3F), leading us to define common domains shared across all single prostate cancer cells that may identify common and hence functionally significant pathways. Of 1,496 PMDs, only 40 (2.7%) are universally hypomethylated, with the mean quantile normalized methylation level <25%, across cells from all four patients with metastatic prostate cancer and four prostate cancer cell lines (Figure S2D, Table S2, see Methods). The 40 core PMDs have a mean size of 2.5 Mb (range 353.4 kb to 7.7 Mb) and encompass 143 protein-encoding genes, a gene density of 1.44 gene/Mb. Hypomethylation associated with cell proliferation is thought to be more rapid in loci that have reduced CpG content15,36. Indeed, we note that the core prostate PMDs exhibit reduced CpG residue content, compared with other PMDs across the genome (P<0.0066, Figure S3G), providing a possible explanation for their universal hypomethylation. In the same single prostate cancer cells, analysis of the 1,412 PMIs for intersection across all prostate cancer patients and Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 7 prostate cancer cell lines identifies 44 core PMIs (Figure S2D, Table S2, see Methods). Core PMIs have a mean size of 371.7 kb (range 27 kb to 1.9 Mb) and harbor 255 protein- encoding genes, with a gene density of 15.6 genes/Mb (Figure S3H). Our single-cell analysis of prostate cancer cells identifies a small fraction of PMDs that are universally shared, which we describe as core PMDs, and it also reveals that interspersed within these large PMDs are small gene-rich islands with preserved DNA methylation, that we call PMIs. Hypomethylation of core PMDs is an early event in prostate tumorigenesis DNA hypomethylation progresses during cancer evolution to ultimately encompass large regions of the non-coding and gene-poor genome within advanced cancers37. By analogy with early genetic driver mutations, however, non-random epigenetic silencing may play an important role in initiating tumorigenesis, with selection pressures guiding recurrent early events. Having defined core PMDs shared across single metastatic prostate cancer cells, we sought to identify genomic loci that are consistently subject to early silencing during tumorigenesis. Given the characteristic admixture of tumor and stromal cells in localized prostate cancer, we obtained frozen tissue sections from prostatectomy specimens and purified single nuclei for molecular analysis. Tumor origin of individual nuclei was confirmed by CNV inferred from whole genome bisulfite sequencing, and we computed a CNV score (absolute DNA copy number changes per Mb) to complement Gleason histological scoring, as an independent measure of tumor progression (Figure 3A, see Methods). In addition to Gleason histological scoring of localized prostate cancer, we computed a CNV score (absolute DNA copy number changes per Mb) to quantify genomic instability in single nuclei from different prostatectomy samples, as an independent measure of tumor progression. In total, we profiled 38 primary tumor nuclei from five patients with low grade (GS6) prostate cancer, 62 nuclei from another five patients with high grade (GS≥8) disease, and 78 normal prostate cells from adjacent tissue sections, comparing these with the 38 CTCs from patients with metastatic disease (Table S1). Inferred CNV from our high resolution single nucleus analysis identifies Chr8p loss (containing NKX3– 1, BMP1, FGFR1 genes and multiple microRNAs) as one of the earliest genetic events in prostate tumorigenesis, shared by >43% of cancer cells in GS6 tumors (Figure S4A). Early allelic loss of this locus has been reported in prostate cancer38–41. Interestingly, GS6 prostate cancer cells with Chr8p loss show more hypomethylation across PMDs, pointing to coordinated early timing of CNV and hypomethylation (Figure S4B). At the single-cell level across different tumors, hypomethylation at prostate PMDs exhibits less heterogeneity than do hypermethylated CpG promoter regions (Figures S4C and S4D). Remarkably, core PMDs initially defined by their universal hypomethylation in metastatic prostate cancer cells show profound enrichment at the earliest stages of tumorigenesis. In early GS6 tumors, 77.5% (31/40) core PMDs are hypomethylated, compared with only 8% (115/1,456) of non-core PMDs (Figure 3B). Indeed, mean quantitative methylation levels within core PMDs decline from 78.4% (normal prostate), to 70.4% (GS6), 57.2% (GS8), and 20.2% in metastatic CTCs. Comparable methylation levels across all prostate PMDs decline more slowly: 82.2% (normal prostate), 80.9% (GS6), 74.7% (GS8) and 57.6% Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 8 (CTCs) (Figure 3C). By contrast, methylation at interspersed core PMIs shows little change from normal prostate nuclei to GS6, GS8, and metastatic prostate CTCs. Compared with hypomethylation of large chromosomal domains, focal hypermethylation of CpG islands within gene regulatory regions increases gradually from 27.5% (normal prostate), to 30.7% (GS6), 31.9% (GS8), and 34.3% (CTCs) (Figure S4E), as does aneuploidy measured by CNV score (Figure S4F). We observed no confounding correlation (FDR>0.1) between CNV and DNA methylation for core PMDs (Figure S4G). Our observations of accelerated progressive demethylation of core PMDs in early prostate cancer are confirmed by analysis of TCGA prostate cancer methylation array data stratified by Gleason Score (Figure 3D), as well as whole genome bisufite sequencing in primary and metastatic prostate tumors34,42 (Figure S4H). Core PMIs show preserved methylation patterns independent of Gleason Score (Figures 3D and S4H). Taken together, core PMDs begin to lose DNA methylation within indolent GS6 prostate cancers, one of the earliest identifiable lesions in prostate tumorigenesis. This early timing explains their universal hypomethylation in advanced cancers, compared with more heterogeneous hypomethylation domains that emerge during subsequent tumor progression. Silencing of immune-related genes within core PMDs and persistent expression of proliferative genes within PMIs To address the functional consequences of early DNA hypomethylation, we identified protein-encoding genes localized to core PMDs that display loss of expression across the large prostate cancer TCGA database39. Among the 143 protein-coding genes residing within the 40 core PMDs, 68 (48%) are consistently and significantly differentially expressed between normal prostate and primary prostate tumors, with 61 (90%) suppressed and 7 (10%) induced in cancer. Remarkably, 12/61 (20%) silenced genes within core prostate PMDs are immune-related. GSEA analysis reveals lipid antigen processing and presentation (P<1.96E-13) and cellular response to interferon (P<2.74E-5) as the two most highly enriched pathways (Figure 3E, see Methods). Conversely, of the 255 protein- encoding genes within the 44 core PMIs, 161 (63.1%) are comparably expressed in prostate cancer and normal prostate tissues in the same TCGA database. The top GSEA pathways all relate to cell proliferation, including E2F targets (P<0.000975) and DNA repair (P<0.00116) (Figures 3F, S4I–J). As control, GSEA pathway analysis does not identify statistically significant enrichment among core PMD-derived genes that are not expressed or not silenced in prostate cancer, or among core PMI-derived genes without preserved expression. Thus, identifying early and consistent changes in DNA methylation in prostate cancer cells points to silencing of immune-related genes, with selective sparing of genes encoding proliferative drivers, as initial steps in prostate tumorigenesis. PMD-associated silencing of the CD1A-IFI16 gene cluster A remarkable feature of core PMD-associated gene silencing is targeting of the entire CD1 family of lipid antigen presentation genes (CD1A, CD1B, CD1C, CD1D and CD1E) and four interferon inducible genes of the Pyrin and HIN domain (PYHIN) family involved in immune sensing of non-self DNA (IFI16, AIM2, PYHIN1 and MNDA). These genes are clustered within the same core hypomethylation block at chromosome 1q23.1 (hereafter, Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 9 CD1A-IFI16 block), consistent with a single genomic locus playing a major role in integrating these two immune recognition pathways (Figure 4A). The CD1 gene family encodes MHC class I-like molecules that exclusively present non-peptides (e.g. glycolipids) to Natural Killer-T (NKT) cells, a rare subset of T cells implicated in both innate and adaptive immunity43–45. The CD1 pathway is primarily implicated in innate immunity to infectious agents, although a possible role for lipid antigens in anti-tumor immunity is also postulated46,47. Among interferon-inducible genes, IFI16 is highly expressed in normal prostate cells: it is reported to bind non-self dsDNA in both nucleus and cytoplasm in a DNA length-dependent manner, recruiting STING and further activating interferon signaling48. DNA methylation of the CD1A-IFI16 locus declines early and rapidly, scoring as the 14th earliest across all genome-wide PMDs measured at GS6 (Figure 4B). Heterogeneity in hypomethylation at CD1A-IFI16 is evident within single prostate cancer cells at early stage GS6 tumors, progressively increasing in both fraction of tumor cells and degree of hypomethylation within individual tumor cells as they evolve to GS8 and ultimately to metastatic CTCs (Figures 4C and S5A). This early and progressive loss of DNA methylation at the CD1A-IFI16 locus, compared with the slower rate of demethylation genomewide, is also evident in analysis of public databases of primary and metastatic prostate cancer34,42 (Figure S5B). Analysis of TCGA prostate cancer data stratified by Gleason Score further confirms early progressive loss of methylation within the CD1A-IFI16 locus (Figure S5C), and the associated transcriptional downregulation of the encoded genes as early as GS6 tumors (Figures 4D). The accelerated decline in DNA methylation at CD1A-IFI16 is not driven by gene copy number changes, as confirmed by comparing single nuclei with or without CNV at this locus (Figure S5D). Early DNA hypomethylation at the CD1A-IFI16 locus is not restricted to prostate cancer. DNA methylation datasets at defined stages of cancer progression are available for both colon and thyroid cancers9, both of which demonstrate earlier and more progressive demethylation of CD1A-IFI16, when compared to other core PMDs (Figure S5E). Furthermore, analysis of methylation profiles in a TCGA cohort including more than 1,000 samples spanning 33 cancer types (https://portal.gdc.cancer.gov) identifies the CD1A-IFI16 locus as consistently hypomethylated in 23 different cancers (Figures S5F–G). Across all 33 cancer types, CD1A-IFI16 demonstrates the greatest degree of DNA hypomethylation compared with all other core PMDs (Figure 4E), and 19 of the 33 cancers show a significant correlation between hypomethylation of this locus and reduced RNA expression of CD1A- IFI16 resident genes (Figure S5H). Early and profound DNA hypomethylation at CD1A- IFI16 is thus a consistent feature across multiple cancers. Along with DNA hypomethylation of the CD1A-IFI16 locus, we observed the expected enrichment for H3K27me3 chromatin marks, comparing prostate cancer versus normal prostate cell lines, together with suppression of the encoded genes within that locus (Figures S6A, B and S6C, Table S3). Extending this analysis to nuclei from microdissected GS6 and GS8 tumors using ultra-low-input native ChIP-seq (ULI-NChIP), we observe marked progressive enrichment of H3K27me3 at the CD1A-IFI16 locus in early GS6 tumors compared with normal prostate epithelium (Figures S6D–E), whereas other PMDs show increased H3K27me3 only at GS8 (Figure S6F). Finally, within high purity TCGA prostate Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 10 samples (tumor purity >0.5 inferred by ABSOLUTE algorithm), all five lipid antigen presentation genes and three of the four PYHIN interferon inducible genes are suppressed in primary prostate tumors (n=188) compared with normal prostate (n=14) (Figure S6G). The suppression of CD1A-IFI16 gene expression is observed at the earliest timepoint of DNA hypomethylation (GS6), and it persists as DNA hypomethylation progresses, suggesting a potential threshold effect. Thus, immune-related genes within the CD1A-IFI16 cluster are among the earliest targets of cancer hypomethylation-induced transcriptional silencing. Functional recapitulation of hypomethylation-associated silencing at CD1A-IFI16 locus To investigate the functional relationship between DNA methylation, repressive chromatin marks and expression of PMD-resident genes, we applied the DNA demethylating agent 5-azacytidine (5 μM) to the human prostate epithelial cells (BPH-1), in which the CD1A- IFI16 locus shows normal DNA methylation levels (Figure 4A). Global DNA methylation declines by 4.9 % after 24 hrs of 5-azacytidine, and by 37.7% after 5 days of drug exposure, compared with DMSO controls (Figure 4F), with the CD1A-IFI16 locus showing progressive DNA demethylation upon 5-azacytidine treatment (Figure S7A). Bisulfite treatment and Sanger sequencing confirms gradual demethylation at CD1A-IFI16 (DMSO: 75.9%, day5: 40.8%) (Figure S7B). Ectopically-induced demethylation is accompanied by marked increase of the chromatin silencing mark H3K27me3, as shown by quantitative imaging of nuclei (7.18-fold increase after 5 days) (Figures 4G and S7C), along with H3K9me3 (Figures S7D–E), and associated with reduced expression of CD1 (Figure 4H). Thus, DNA hypomethylation appears to trigger the recruitment of chromatin suppressive marks at the CD1A-IFI16 locus, along with repression of the resident genes. We then tested the converse model, using an inhibitor of the EZH2 methyltransferase, GSK126, to suppress H3K27me3 in prostate cancer cells, in which the CD1A-IFI16 locus is hypomethylated and silenced. Treatment of three prostate cancer cell lines (22Rv1, LNCaP and VCaP) with GSK126 results in loss of global H3K27 trimethylation, associated with a dramatic increase in expression of all the genes within the CD1A-IFI16 locus (Figures 4I–J and S7F–G). Together, these observations further support the role of chromatin silencing marks in repressing coding genes within the CD1A-IFI16 locus and other PMDs. Re-expression of lipid antigen presentation or interferon-inducible genes restores anti- tumor immunity in a mouse model To explore the potential significance of CD1A-IFI16 silencing, we tested the consequences of restored expression in a murine model of early prostate tumorigenesis. The mouse prostate cancer cell line Myc-CaP is derived from a genetically engineered model with prostate-specific expression of a c-Myc transgene driving androgen-dependent tumorigenesis49. Single-cell methylation sequencing of Myc-CaP cells shows uniform hypomethylation of two chromosomal loci syntenic with the single human CD1A-IFI16 locus, and encompassing the two murine lipid antigen presentation genes (Cd1d1 and Cd1d2) and the orthologous PYHIN interferon inducible genes (Ifi204, Aim2, Pyhin1 and Mnda), respectively (Figures S8A and S8B). Repressive H3K27me3 and H3K9me3 marks are enriched at the Cd1d and interferon inducible genes (Figures S8A and S8B). The major CD1 murine ortholog Cd1d1 and the IFI16 murine ortholog Ifi204 are repressed Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 11 in Myc-CaP tumor cells, compared with normal prostate tissues dissected from isogenic FVB mice (Figure 5A). We ectopically expressed Cd1d1 (16.1-fold) or Ifi204 (4.1-fold) in Myc-CaP cells by lentiviral transduction, achieving levels comparable to those of normal mouse prostate (Figure 5A and Table S3, see Methods). Cell surface localization of restored Cd1d1 is evident using both flow cytometry and confocal microscopy (Figures S8C and S8D). Ectopic expression of Cd1d1 in Myc-CaP cells does not alter proliferation in vitro, but these cells fail to produce tumors in isogenic immune competent FVB mice, when inoculated either subcutaneously or by direct intraprostatic injection (Figures 5B, 5C and S8E). This effect is dependent upon immune cell activation, since inoculation of the same Cd1d1-expressing Myc-CaP cells into immunodeficient NSG mice does not suppress their ability to give rise to primary tumors (Figure 5D). Cd1d specifically mediates the presentation and activation of lipogenic antigens to NKT cells, a rare T cell subpopulation expressing Cd40lg and Icos (http://rstats.immgen.org/Skyline/skyline.html)50, and tumors from Cd1d1-restored Myc-CaP cells in FVB immune competent mice show increased expression of Cd40lg (2.8-fold; P=0.0063) and Icos (3.2-fold; P=0.00023) compared with controls (Figure S8F and Table S3). Flow cytometric analysis of tumor immune infiltrates in Cd1d1-restored tumors indicates more abundant Cd1d-restricted NKT cells (P=0.0042), along with increased binding to the high affinity synthetic NKT cell ligand alpha-Galactosyl Ceramide (α-GalCer) tetramer and an increase in the CD69 marker of NKT cell activation (P=0.0099) (Figures 5E and S9A–C). To test the consequences of restored Cd1d1 expression in another mouse isogenic tumor model, we restored its expression in the LLC-1 lung epidermoid carcinoma model, which does not express Cd1d1. Ectopic expression of Cd1d1 in LLC-1 reduces tumor growth upon subcutaneous inoculation into immune competent isogenic C57BL/6 mice, despite unaltered in vitro proliferation (Figures S8G–J). We then tested the effect of restored expression in Myc-CaP cells of Ifi204, the murine ortholog of the interferon inducible gene IFI16. Re-expression Ifi204 also suppresses Myc- CaP tumorigenesis in immune competent FVB mice, without any anti-proliferative effect in vitro (Figures 5B and 5C). This effect is not evident in immune deficient NSG mice, pointing to an immunological effect (Figure 5D). Tumors derived from Ifi204-expressing Myc-CaP cells in FVB mice show no difference in the total number of CD4+, CD8+ T cells or in the expression of general marker of T cell activation (Figures 5F, S9D and S9E). However, compared to parental controls, Ifi204-reconstituted tumors have a dramatic reduction in expression of the co-inhibitory receptor PD-1 within CD8+ T cells (P=0.00042), along with an increase in the functional intracellular cytokine TNFα (P=0.0374), all consistent with activated CD8+ T cell cytotoxic function (Figures 5F, S9F and S9G). No change is evident in expression of other co-inhibitory receptors (TIGIT, LAG3 or TIM3) or cytokine (IFNγ) in CD8+ T cells (Figures S9H–S9K). Thus, ectopically restored expression of either CD1 or IFI16 murine orthologs in cancer cells with DNA hypomethylation-induced silencing suppresses tumor formation, a finding only evident in immune competent mice, and associated with evidence of selectively increased anti-tumor activity. Our results indicate that at least two distinct immune populations are impaired by silencing of the CD1A-IFI16 locus (NKT cells modulated Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 12 by Cd1d1 and cytotoxic CD8+ T cells affected by Ifi204), suggesting a complex immune- modulatory function of this multigene locus in tumorigenesis. Detection of CTC-derived DNA hypomethylation in blood specimens using Nanopore sequencing While our study was focused on the characterization of early methylation changes in prostate tumorigenesis and their potential biological consequences, we also note the recent application of CpG island hypermethylation as a blood-based diagnostic assay for early cancer detection24,25. Genome-wide screening for changes in DNA methylation may be more sensitive than mutation-based assays, particularly in tumors like prostate cancers, which do not harbor well defined recurrent driver mutations. Nonetheless among all cancers tested, early prostate cancer shows one of the lowest detection rates (11.2%), using screening for CpG island hypermethylation25. CTCs are shed into the blood by invasive localized prostate cancers long before they establish metastases51–53, raising the possibility that they may provide an orthogonal assay for early cancer detection. Given the specificity of DNA hypomethylation domains in cancer cells and their large genomic size, we reasoned that they may provide high sensitivity and quantitative signal for cancer detection, following CTC enrichment in blood specimens. For such blood-based rare cell signal detection studies, we applied a screen for all prostate PMDs, rather than the much smaller number of core PMDs, so as to increase coverage to a large fraction of the prostate cancer genome. Oxford Nanopore long-read native sequencing typically produces sequencing reads up to 100 kb, and directly identifies methylated CpG residues (5mC), without requiring bisulfite conversion in library preparation54,55. In its current configuration, Nanopore signal analysis does not readily identify 5-hydroxymethyl cytosines (5hmC), which are considerably less abundant than 5mC, and are also not distinguished from 5mC in conventional bisulfite sequencing. Indeed, Nanopore sequencing of the VCaP prostate cancer cell line clearly defines DNA hypomethylation domains, which faithfully recapitulate those identified in these cells using standard bisulfite sequencing (Figures 6A and 6B).In contrast to the short Illumina sequencing reads (usually harboring <5 CpG sites per read), mathematical modeling indicates that the long reads generated by Nanopore sequencing would empower detection with significantly higher precision for rare signal (Figures 6C and 6D, see Methods). We therefore processed 10 ml blood specimens from patients with either localized or metastatic prostate cancer, using microfluidic enrichment to deplete leukocytes (104-fold depletion), but without further CTC purification or individual CTC micromanipulation (Figure 6E, see Methods). While 23 age-matched healthy donors (HDs) show minimal DNA hypomethylation signal (<0.6%), 6 out of 7 (86%) patients with metastatic prostate cancer have significant signal (from 0.62% to 11.08% of sequencing reads, P=0.00011), as do 6/16 (37.5%) patients with localized prostate cancer (from 0.62% to 2.29% of sequencing reads, P=0.004) (Figures 6F and 6G,Tables S4 and S5). Thus, long-range hypomethylated domains are universal characteristics of prostate cancer and they are detectable from rare CTCs in patient-derived blood specimens. The simplicity and cost effectiveness of Nanopore sequencing raises the possibility of hypomethylation-based cancer detection. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Discussion Page 13 Using single-cell DNA methylation analysis, ranging from indolent low grade localized prostate cancer to metastatic CTCs, we annotated at high resolution the shared hypomethylation domains that constitute core PMDs, along with interspersed islands with preserved methylation, that we identify here as PMIs. PMDs are known to be associated with the peripheral and transcriptionally silenced B compartment of the nucleus13,56, raising the possibility that PMIs loop into the active A compartment regions, and hence are spatially distinct from the surrounding silenced chromatin. Given intercellular heterogeneity, the denotation of core PMDs was derived from the intersection of PMDs across many single cells from multiple independent prostate cancers. However, these core PMDs also stand out by virtue of their detection in the earliest low grade prostate cancers (GS6), leading to the suggestion that they are driven by early selective pressures in tumorigenesis, and explaining their universal silencing in advanced prostate cancers. Indeed, silencing within core hypomethylation domains appear to target immune-related genes, including a single chromosomal locus containing the entire family of CD1 genes and a cluster of interferon- inducible genes. PMIs, in contrast, preserve expression of proliferation-associated genes implicated in cell-cycle and DNA damage repair pathways. DNA methylation changes may thus convey a selective advantage in prostate cancer development, suppressing expression of genes contributing to immune surveillance of nascent tumors, while shielding neighboring genes that enhance cell proliferation. Such selective pressures could drive the very early targeting of the immune-rich CD1A-IFI16 locus, as demonstrated by in vivo reconstitution experiments in mouse models. While early PMDs, like the CD1A-IFI16 locus, may emerge solely from selection pressures favoring proliferating prostate cells that escape immune surveillance, it is also possible that such loci have intrinsic properties favoring early loss of DNA methylation. Early hypomethylation of core PMDs The model that hypomethylation-associated gene silencing occurs early and favors tumorigenesis differs conceptually from a hypothesis proposed from a study of advanced colon cancers, whereby hypomethylation might serve an intrinsic tumor suppressor mechanism, restraining uncontrolled cell proliferation13. Of note, the colon cancer study analyzed bulk tumor material, encompassing cancer cells together with reactive stroma and immune cells, and it therefore excluded from analysis immune-related genes, whose cell-of- origin is confounded by whole-tumor sequencing. Single-cell level analysis thus allows assignment of all changes in DNA methylation to the appropriate cell type. Most important, however, is our definition of a small subset of PMDs, annotated as core PMDs (2.7% of all PMDs), that appear early in tumorigenesis and are shared uniformly across multiple independent tumors. The identification of early cancer drivers targeted by epigenetic silencing is likely to differ from the contribution of additional PMD-encoded genes that are silenced during subsequent cancer progression, as DNA hypomethylation extends across major portions of the genome. Compared with the small number of core PMDs identified in early cancers, the very large fraction of the cancer genome that is hypomethylated in advanced tumors may thus reflect distinct selection pressures, as well as bystander effects affecting gene-poor PMDs and the derepression of repetitive elements. While our Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 14 study was centered on prostate cancer, the relevance of core PMDs extends to other cancers, as illustrated by TCGA analyses showing their consistent early hypomethylation across multiple tumors, in contrast to most PMDs which show considerable inter-tumor heterogeneity. Indeed, TCGA methylation data shows that the CD1A-IFI16 locus to have the strongest difference in DNA methylation between 33 different cancers and their normal tissue counterparts. This specific locus, encoding immune-related genes that have not been previously nominated as critical cancer genes, thus appears to be a consistent target of epigenetic silencing in the early stages of tumorigenesis. Our functional assays using the demethylating agent 5-azacytidine and the EZH2 inhibitor GSK126 support the recruitment of chromatin silencing marks to hypomethylated PMDs as a mechanism of transcriptional silencing. However, further studies will be required to better understand the selectivity of PMD hypomethylation across the genome, and both genomic structure and selection pressures that distinguish core PMDs from more global demethylation. The CD1A-IFI16 immune gene cluster The CD1A-IFI16 locus is unique in encompassing the entire gene family of CD1 genes, which together mediate lipid antigen presentation, together with the IFI16 class of interferon-inducible genes. It is well established that genes that are co-located within a single genomic locus may be targeted during tumorigenesis by either chromosomal deletions or amplification events, a single genetic event that may mediate simultaneous loss-of-function or gain-of-function among physically clustered genes. Conceptually, the hypomethylation silencing of the CD1A-IFI16 locus during early prostate tumorigenesis may accomplish a similar function, suppressing T cell recognition of lipid antigens as well as double stranded DNA sensing, as part of a single epigenetic event affecting both alleles. Such a potent selective pressure could explain the early and frequent targeting of this locus in cancer. The 1q23.1 genomic locus has been linked in germline association studies to neurodegenerative disease and autoimmune diseases57,58, and immunological pathways regulated by its resident genes have been linked to innate immunity against infectious pathogens. The potential roles of these genes in immune surveillance of early cancers will require further functional analyses. Alterations in antigen presentation pathways constitute the most critical mechanisms by which tumors evade both innate and therapeutic immune activation59,60. In this respect, the presentation of lipid antigens to NKT cells, a highly specialized subpopulation of T cells, is of particular interest, given potential therapeutic implications. Within prostate cancer, the silencing of CD1A-IFI16 genes is also noteworthy in that it points to tumor cell-intrinsic factors contributing to the escape from immune surveillance, in addition to the proposed immunosuppressive effects of the tumor microenvironment. Diagnostic implications Finally, from a cancer diagnostic standpoint, blood-based detection of early invasive cancers remains a major technological challenge. For prostate cancer, it requires the ability to distinguish between indolent lesions associated with non-specific elevations in serum PSA and more aggressive cancers that may have similar serum PSA levels but warrant therapeutic intervention. Early invasive prostate cancers shed CTCs into the circulation long before metastases are established51–53, and while these rare early CTCs may not be sufficient to Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 15 cause dissemination, they can serve as potential biomarkers of invasive disease. Microscopic imaging of very rare CTCs in the bloodstream is challenging, hence there is a need for sensitive and quantitative molecular readouts applied to CTC-enriched blood specimens. While this study was not designed to formally test Nanopore sequencing of PMDs as a quantitative molecular surrogate of CTCs, it suggests that such long-range DNA sequencing strategies may complement current approaches that rely on hypermethylation of CpG islands within short ctDNA fragments. Such approaches may also enhance tissue-of-origin determinations, given the information content inherent in such long-range genomic analyses. Limitations of the study Our study suggests that early hypomethylation of core PMDs in prostate cancer differentially silences immune surveillance-associated genes, while sparing genes that mediate cell proliferation. While we find shared patterns of core PMDs across multiple different cancers, it is also possible that distinct tumor types will target alternative biologically relevant pathways. Additional studies in different early stage cancers will be required to distinguish shared hypomethylation targets from those showing tissue-specific patterns, and additional patient-derived samples will need to be analyzed within each tumor type. The potential roles in immune surveillance of lipid antigen presentation genes and IFI16-related double stranded DNA sensing genes deserves further functional analyses using additional experimental systems to define their relevance in early tumorigenesis, as well as their potential relevance for anti-cancer therapy. Finally, the potential utility of PMD detection in blood-based cancer diagnostics will require further validation in larger numbers of diverse clinical specimens. STAR Methods Resource availability Lead Contact—Further information required to reanalyze the data reported in this paper and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Daniel A. Haber ([email protected]). Material Availability—Plasmids generated in this study are available upon written request. Data and Code availability • • All raw and processed sequencing data in this study, including single-cell DNA methylation sequencing, single-cell RNA-seq, ChIP-seq, Cut and Run assay and Nanopore sequencing, have been deposited to the NCBI Gene Expression Omnibus (GEO) database under accession GSE208449. All data are publicly available as of the date of publication. This paper analyses existing, publicly available data or available upon request to the authors. These accession numbers for the datasets are listed in the key resources table. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 16 • • • This paper does not report original code. All the scripts and mathematical algorithms used in this study will be available from the corresponding authors upon request. All the versions of software packages used in this study are listed in the key resource table and noted in the data analysis method accordingly. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Experimental model Clinical Specimens—All patient samples were collected in this study after written informed consent, in accordance with Institutional Review Board (IRB) protocols (DF/HCC 05–300, 11–497, 13–217 or 14–375). For the CTC cohort, 10–20 ml of blood was drawn from patients with a diagnosis of metastatic prostate cancer, localized prostate cancer, or age-matched males without a diagnosis of cancer at Massachusetts General hospital (MGH). For the localized tumor tissue cohort, all samples were acquired from either core biopsies or surgical resection of untreated localized prostatic adenocarcinoma (Gleason scores 6 and 8) from patients at MGH. In cases with the lowest grade tumors (Gleason score 6), normal prostate tissue was also identified in the tissue specimen by a Genito-Urinary (GU) specialized pathologist and used as a source of matched normal prostate cells. Both normal and tumor tissue samples were de-identified, snap frozen and sectioned. Only tumor sections with >80% tumor content, as assessed by a specialized GU pathologist were used in this study. The clinical data of the patients with metastatic prostate cancer enrolled in the single-cell CTC analysis and patients with resected localized prostate cancer used for single nucleus analysis are described in Table S1. The clinical data of the patients with localized prostate cancer and metastatic prostate cancer enrolled in Nanopore sequencing anlysis of CTC-enriched blood are described respectively in Table S4 and Table S5. Cell culture—Human prostate cancer cell lines (LNCaP, VCaP, PC3 and 22Rv1), murine prostate cancer line (Myc-CaP), normal cultured prostate epithelial cells (HPrEC), benign prostatic hypertrophy cells (BPH-1) and murine Lewis lung carcinoma cells (LLC-1) were all obtained from ATCC, after authentication by short tandem repeat (STR) profiling. All cell lines used in the paper were derived from male mice or male human patients. They were cultured in the following media at 37°: RPMI-1640 (ATCC) medium supplemented with 10% FBS (Gibco) and 1X Pen/Strep (Gibco) (for LNCaP, VCaP, PC3, 22Rv1 and BPH-1 cells); Prostate Epithelial Cell medium (ATCC) with 6 nM L-glutamine (ATCC), 0.4% Extract P (ATCC), 1.0 mM Epinephrine (ATCC), 0.5 ng/ml rh-TGFα (ATCC), 100ng/ml hydrocortisone hemisuccinate (ATCC), 5 mg/ml rh-Insulin (ATCC), 5 mg/ml Apo-transferrin (ATCC), 33 μM Phenol red (ATCC) and 1X Pen/Strep/Ampho Solution (ATCC) (for HPrEC cells); DMEM high glucose medium (Gibco) with 10% FBS (Gibco) and 1X Pen/Strep (Gibco) (for Myc-CaP cells and LLC-1 cells). All the cell lines used in this study were checked for mycoplasma every 4 months using Mycoalert kit (Lonza). Mouse xenograft assays—All animal experiments were carried out in accordance with approved protocols by the MGH Subcommittee on Research Animal Care (IACUC). All Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 17 the mice used in this study were maintained under a 12/12 h light/dark cycle in MGH animal facility. 6–8 weeks old FVB male mice (Jackson Laboratory, Strain#001800) or 6–8 weeks old male immunodeficient NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ) mice (Jackson Laboratory, Strain#005557) were used for intraprostatic injection or subcutaneous injection of Myc-CaP cells stably expressing luciferase and mCherry. 6–8 weeks old C57BL/6 female mice (Jackson Laboratory, Strain#000664) were used for subcutaneous injection of LLC-1 cells stably expressing luciferase. Littermates of the same sex were randomly assigned to experimental groups. For intraprostatic inoculation, mice were first anesthetized using isoflurane, and a 1 cm skin incision was performed along the midline of the abdomen to expose the inner muscle layer, which was then separated. The tip of seminal vesicle was raised gently with forceps to expose the anterior lobe of the prostate gland. 50,000 Myc-CaP cells 1:1 mixed with Matrigel (v/v) (total volume: 30 μl) were slowly injected into the prostate lobe. All the tissues were then returned into the abdomen, and continuous sutures were used to close the inner muscle layer, followed by separate skin closure. For subcutaneous injections, mice were anesthetized, and 50,000 Myc-CaP cells or 1,000,000 LLC-1 cells 1:1 mixed with Matrigel (v/v) (total volume: 100 μl) were injected into the flank. Tumor cell-derived bioluminescent signal was quantified every other day for the Myc-CaP cells and 3 times a week for the LLC-1 for mice after either orthotopic injection or subcutaneous injection. At 2–3 weeks after inoculation, mice were sacrificed and tumors were harvested for flow cytometry and RNA extraction for the Myc-CaP experiments. Method Details CTC isolation—CTCs were isolated from fresh blood specimens drawn from patients with prostate cancer, following negative depletion of leukocytes using the microfluidic CTC-iChip as reported previously26,27. Briefly, 10–20 ml of whole blood specimens were incubated with biotinylated antibody cocktails against CD45 (R&D Systems, clone 2D1), CD66b (AbD Serotec, clone 80H3), and CD16 (BD Biosciences), followed by incubation with Dynabeads MyOne Streptavidin T1 (Invitrogen) for magnetic labeling and depletion of leukocytes. After CTC-iChip processing, the CTC-enriched product was further stained with FITC-conjugated antibody against EpCAM (Cell Signaling Technology, clone VU1D9) and PE-conjugated antibody against CD45 (BD Biosciences, clone HI30). Single CTCs (FITC positive and PE negative) or white blood cells (WBCs, FITC negative and PE positive) were individually picked into PCR tubes containing 5 μl RNA/DNA lysis buffer using micromanipulator (Eppendorf TransferMan NK 2) and snap-frozen in liquid nitrogen. In total, 38 CTCs from 5 different patients (GU114, GU169, GU181, GU216 and GURa15) with metastatic prostate cancer were individually picked, sequenced and lineage-confirmed based on transcriptome and DNA copy number. One patient sample (GU169) had only one CTC, and it was therefore excluded from some downstream analyses focused on the four patients with multiple CTCs. Nuclei isolation from frozen tumor sections—Tumor tissue sections with high tumor content (>80%) and adjacent normal tissue section were micro-dissected and transferred into a pre-chilled Dounce homogenizer containing ice-cold 1 ml 1X HB buffer (0.26 M sucrose, 30 mM KCl, 10 mM MgCl2, 20 mM Tricine-KOH, 1 mM DTT, 0.5 mM Spermidine, 0.15 mM Spermine, 0.3% NP-40 and 1X complete protease inhibitor). Tissue Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 18 was homogenized with ~10 strokes of “A” loose pestle, followed by another ~10 strokes of “B” tight pestle. The tissue homogenate was then filtered using a 70 μm strainer and pelleted by centrifugation. Nuclear pellets were resuspended and purified by density gradient centrifugation (top layer: 25% Iodixanol solution; middle layer: 30% Iodixanol solution; bottom layer: 40% Iodixanol solution). The nuclear band at the interface of 30% and 40% Iodixanol solutions was collected into a new Eppendorf tube and washed twice with ice-cold 1X PBS. 20% of the purified nuclei were used to isolate single nuclei using fluorescence- activated cell sorting (FACS) for single-cell DNA methylation analysis, while the remaining 80% of the nuclei were subjected to ChIP-seq analysis. Western Blot—Cells or tumor tissues were lysed in Laemmli buffer (Sigma) and cleared. Protein concentration was determined using DC protein assay (Bio-rad). Proteins (25 μg) were separated on precast NuPAGE 4–12% Bis-Tris protein gels (ThermoFisher), and transferred onto nitrocellulose membranes (Bio-Rad). After blocking with 5% BSA buffer for 1 hour at room temperature, membranes were incubated with primary antibodies overnight at the recommended concentrations. HRP conjugated secondary antibodies (1:10,000; Bio-rad; Cat#5196–2504) were applied, and ultra-sensitive autoradiography film (Amersham) was used to detect the chemiluminescence signal. Primary antibodies used are H3K27me3 (1:1,000, Invitrogen Cat#MA5–11198) and H3 total (1:1,000, Abcam Cat#1791). 5-Azacytidine treatment, bisulfite sequencing and staining of chromatin marks —The human prostate epithelial cell line BPH-1 was cultured in the presence of 5 μM 5- azacitidine (Selleck, #S1782). At serial time points (days 0, 1, 4 and 5), cells were collected for DNA extraction, confocal microscopy, or flow cytometric analysis. DMSO-treated cells were used as control at each time point. To quantify 5-azacitidine-induced demethylation at the genomewide level, we used the whole genome bisulfite sequencing (WGBS). Briefly, DNA ws extracted from BPH-1 cells upon 5-azacitidine treatment, 1 μg genomic DNA was used to sonicate into 300–500 bp fragments, DNA was end-polished, A-tailed and ligated with pre-methylated adaters before bisulfite conversion using EZ DNA methylation kit (Zymo, #D5001), bisulfite-converted DNA was amplified and sample index was introduced during amplification. To quantify 5-azacytidine-induced demethylation at the CD1A-IFI16 locus, DNA extracted from BPH-1 cells treated with 5-azacitidine was subjected to bisulfite conversion using EZ DNA methylation kit (Zymo, #D5001), and bisulfite-converted DNA was used for PCR amplification, applying bisulfite-specific PCR primers covering the human CD1A-IFI16 locus (see Table S3). PCR products were purified by 1% agarose gel and cloned using the Zero blunt PCR cloning kit (ThermoFisher, #K270020). 10 individual bacterial clones were randomly picked for Sanger sequencing. Sequencing data were analyzed and shown using online tool QUMA (http://quma.cdb.riken.jp/)61. Nuclear accumulation of H3K27me3 was stained with H3K27me3 antibody (1:1000 dilution; CST#9733), in 5-azacytidine-treated cells. Images were acquired using a Zeiss LSM710 Lase Scanning Confocal and were quantified by quantitative image analysis of cells (ImageJ). Flow cytometry was also performed at serial time points on BD LSRFortessa machine to assess CD1d expression using human CD1d-APC antibody (1:100 dilution; BioLegend#350308, clone: 51.1). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 19 EZH2 inhibitor treatment—Human prostate cancer cell lines (22Rv1, LNCaP and VCaP) were cultured in the presence of the small molecule EZH2 inhibitor GSK126 (Selleckchem, #S7061) at the indicated concentration (0, 5 or 10μM). After 6 days of treatment, protein and RNA were harvested, for quantitation of H3K27me3 and total H3, using Western blotting and expression of individual genes within the CD1A-IFI16 locus by real time qPCR. Paired single-cell DNA methylation and RNA-seq—For these experiments, we used either single CTCs or WBCs individually picked from fresh blood specimens after CTC enrichment, and single cells from cultured prostate cell lines (either picked or FACS-sorted). These were subjected to paired single-cell DNA methylation and RNA-seq analysis to obtain the transcriptomes and DNA methylomes from the same single cells33,62. Briefly, single cells were first lysed in 5 μl DNA/RNA lysis buffer; 0.5 μl Magnetic MyOne Carboxylic Acid Beads (Invitrogen, Cat#65011) were then added to each single cell lysate to facilitate segregation of nucleus versus cytoplasm. After centrifugation and magnetic separation, the supernatant (containing cytoplasmic RNA) was transferred into a new tube for single-cell RNA-seq amplification using the SMART-seq2 protocol63, while the pellet (aggregated beads with the intact nucleus) was resuspended in DNA methylation lysis buffer and subjected to single-cell whole genome methylation sequencing using the scBS-seq protocol64. Single nuclei sorted from the frozen primary prostate tumor sections were also subjected to the scBS-seq procedure. MNase native ChIP-seq—Purified nuclei from frozen tissue sections were subjected to MNase native ChIP-seq following the ULI NChIP procedure, as published elsewhere65. Briefly, nuclei were suspended in Nuclear Isolation Buffer (Sigma) supplemented with 1% TritonX 100, 1% Deoxycholate and 1X complete protease inhibitor. Chromatin was digested by MNase enzyme (NEB, 1:10 diluted) at 21°C for 7.5 min, and further diluted in Complete Immunoprecipitation Buffer, with 1X complete protease inhibitor. 2 μl ChIP- grade H3K27me3 (Active motif, Cat#39155) or H3K9me3 (Abcam, Cat#ab8898) antibody was incubated with the digested chromatin overnight at 4°C. DNA was then purified using protease K digestion followed by phenol-chloroform extraction. ChIP-seq sequencing libraries were prepared using NEBNext Ultra II DNA Library Prep Kit (NEB, Cat#E7645L). Cut and Run Assay—H3K27me3 and H3K9me3 Cut and Run assays were performed with cultured prostate cell lines (LNCaP, 22Rv1, BPH-1, HPrEC and Myc-CaP), using the CUT&RUN Assay kit (CST, Cat#86652S). Briefly, 100,000 freshly cultured prostate cells were collected and incubated with Concanavalin A Magnetic Beads. 2 μl ChIP-grade H3K27me3 (Active motif, Cat#39155) or H3K9me3 (Abcam, Cat#ab8898) or IgG (CST, Cat#66362S) antibody was added to the cell: bead suspension and incubated overnight at 4°C. 1.5 μl pAG-MNase enzyme was then added to the tube, which was rotated for 1 h at 4°C, followed by activation of pAG-MNase using 3 μl cold Calcium Chloride. The activation reaction was stopped and DNA was further diluted and collected for phenol-chloroform extraction. Cut and Run sequencing libraries were constructed using NEBNext Ultra II DNA Library Prep Kit (NEB, Cat#E7645L). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 20 Next generation sequencing—All the single-cell RNA-seq, single-cell DNA methylation, MNase ChIP-seq, Cut and Run samples and WGBS samples were molecularly barcoded, pooled together and sequenced on a HiSeq X sequencer to obtain 150 bp pair- ended reads (Novogene). RNA extraction, reverse transcription and quantitative PCR (qPCR)—RNA extracted from cultured prostate cells was prepared using the RNeasy Mini kit (QIAGEN) with DNase I digestion on the column. To extract RNA from mouse tumor tissues, these were first dissected to remove connective tissue and fat, and washed extensively with 1X PBS to remove excessive blood or necrotic tissues. Tumors were then homogenized in RLT RNA lysis buffer using a Dounce homogenizer, and passed through a QIAshredder column (QIAGEN). RNA from normal prostate of FVB mice were prepared following a similar method. RNA from tissue homogenate was extracted using the RNeasy Mini kit (QIAGEN) with DNase I digestion on the column. cDNA was synthesized from 50–200 ng RNA using SuperScript III One-Step qRT-PCR kit (Invitrogen). qPCR was performed using the primers listed in Table S3. CD1d expression measurement by flow cytometry—Cell surface protein expression of CD1d in human and mouse prostate cells was assessed by flow cytometry. Cells were first trypsinized, and 500,000 cells were used for staining with antibody against CD1d at 4°C for 20 min, followed by washing and quantitation using a BD LSRFortessa machine, and data were analyzed using FlowJo software (v10.4; https://www.flowjo.com/). Antibodies used were as follows: for human prostate cell lines, APC conjugated anti-human CD1d (BD#563505, clone: CD1d42) and APC-conjugated isotype control (BD#555751); for Myc- CaP cells, anti-mouse CD1d (Bio X Cell #BE0179, clone 20H2) and the isotype control (Bio X Cell #BE0088), and secondary antibody anti-rat IgG conjugated with APC (Invitrogen #A10540). Plasmid construction—A lentiviral murine Cd1d1 expression construct (pLenti- Cd1d1-mGFP, Cat#MR226027L4) and its matched control construct (pLenti-C-mGFP, Cat#PS100093) were obtained from Origene. Murine Ifi204 expression vector (pLenti- Ifi204-Myc-DDK-Puro, Cat#MR222527L3), together with its control vector (pLenti-C- Myc-DDK-Puro, Cat#PS100092) were also purchased from Origene, and the puromycin selection cassette of these two Origene plasmids were replaced by blasticidin from lentiCRISPRv2-blast plasmid (Addgene#98293) using NEBuilder HiFi DNA Assembly Cloning kit (NEB, Cat#E5520S). For the LLC-1 experiment, the murine Cd1d1 was cloned into the receiving vector N174-MCS (Addgene#81061) with the restriction enzymes EcoR1 and Mlu1, using the FastDigest protocol of Thermo Scientific. All final construct sequences were confirmed by Sanger sequencing. Plasmids generated in this study are available upon written request. Lentiviral transduction—Early passage 293T cells were transfected with Cd1d1 or Ifi204 lentiviral constructs, together with pMD2.G (Addgene#12259) and psPAX2 (Addgene#12260) packaging plasmids using Lipofectamine 2000 reagent (Invitrogen). 48– 72 h after transfection, culture medium (containing lentiviral particles) was collected, Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 21 filtered and concentrated using LentiX concentrator (Clontech). Concentrated virus was added to the Myc-CaP cells in presence of polybrene (Santa Cruz, 8 μg/ml as final concentration) overnight. FACS was used to select GFP positive cells as marker of Cd1d1 construct transduction in the Myc-CaP cells. The LLC-1 cells transduced with the Cd1d1 cloned in the the N174-MCS vector were selected using G418 (Sigma Aldrich #G8168) at 400 μg/mL for 4–6 days. To obtain stable Ifi204 overexpression, 10 μg/ml blasticidin (InvivoGen) was added to the medium for 5–7 days selection. Tumor immune infiltration assayed by flow cytometry—Mouse tumors generated by intraprostatic injection of control or Cd1d1-expressing Myc-CaP cells were dissected and washed to remove blood, fat and connective tissues. Tumor tissues were further mashed and digested in 5 ml digestion buffer (RPMI1640, 2.5 mg/ml collagenase D, 0.1 mg/ml DNase I) at 37°C for 30 min. Tissue digestion was stopped by adding another 5 ml RPMI1640 with 2% FBS, and then filtered through 70 μm strainers. The tissue cell suspension was obtained in the same way for tumors generated by subcutaneous injection of control or Ifi204 expressing Myc-CaP cells. To stain for NKT cell infiltration in prostate tumors with control or Cd1d1 expression, the single- cell suspension was first blocked with rat anti-mouse CD16/CD32 blocking reagent (BD#553142, Clone: 2.4G2) at 4°C for 30 min, followed by mouse NKT surface antibody cocktail staining at 4°C for another 30 min. The mouse NKT surface antibodies used in this study were: BV510-viability dye (BD#564406), APC-α-GalCer-mCD1d Tetramer (TetramerShop#MCD1d-001), BV711-CD69 (BioLegend#104537, clone: H1.2F3), PerCP- Cy5.5-TCRβ (BioLegend#109228, clone: H57–597), BV605-CD3e (BioLegend#100351, clone: 145–2C11) and BUV395-NK1.1 (BD#564144, clone: PK136). Cells obtained from mouse tumors with control or Ifi204 expression were split into two fractions, with the first fraction stained using a panel of mouse T cell surface antibody cocktails: BV510-viability dye (BD#564406), PerCP-Cy5.5-TCRβ (Biolegend#109228, clone: H57–597), BV711-CD8 (Biolegend#100759, clone: 53–6.7), BV650-CD4 (Biolegend#100546, clone: RM4–5), FITC-CD44 (Biolegend#103006, clone: IM7), PE-Cy7-PD-1 (Biolegend#109110, clone: RMP1–30), BV421-TIM3 (BD#747626, clone: 5D12), APC-TIGIT (Biolegend#156106, clone: 4D4/mTIGIT) and BV785-LAG3 (Biolegend#125219, clone:C9B7W). The second fraction was used to stain for surface and intracellular cytokines by first activating cells with Cell Stimulation Cocktail (eBioscience#00–4970-93) together with Protein Transport Inhibitor Cocktail (eBioscience#00–4980) in 37°C cell culture incubator for 4 h. The cells were then stained for surface antigens before fixation, and subsequently processed for intracellular cytokine staining using BD Fixation/Permeabilization Solution Kit (BD#554714). Antibody cocktails used for surface and intracellular cytokine staining were: BV510-viability dye (BD#564406), PerCP-Cy5.5-TCRβ (Biolegend#109228, clone: H57–597), FITC-CD44 (Biolegend#103006, clone: IM7), PE-TNFα (Biolegend#506306, clone: MP6-XT22), BV650-CD4 (Biolegend#100546, clone: RM4–5), BV711-CD8 (Biolegend#100759, clone: 53–6.7) and BV605-IFNγ (Biolegend#505840, clone: XMG1.2). All flow cytometry was done on the BD LSRFortessa machine, and data were analyzed using FlowJo software (v10.4; https://www.flowjo.com/). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 22 Multiplex Oxford Nanopore native sequencing—Blood samples from either healthy donors or patients with localized or metastatic prostate cancer were subjected to CTC-ichip enrichment (104-fold leukocyte depletion)26,27. The enriched CTCs (ranging from 0.1% to 1% purity, admixed with residual leukocytes) were subjected to high molecule weight (HMW) DNA extraction using the HMW DNA extraction kit (QIAGEN), and then prepared for Oxford Nanopore sequencing using the rapid barcoding kit (Nanopore#SQK-RBK004). For each sequencing run, 11 blood samples (either from healthy donors or cancer patients), together with 1 lambda DNA (unmethylated control), were uniquely barcoded and pooled together. Sequencing was performed using a Nanopore MinION device with R9.4 flowcell for 48 h, per manufacturer instructions. Single-cell and bulk RNA-seq data analysis—Raw fastq reads generated from HiSeq X sequencer were first cleaned using TrimGalore (v0.4.3) (https://github.com/FelixKrueger/ TrimGalore) to remove the adapter-polluted reads and reads with low sequencing quality. Cleaned reads were aligned to the human (hg19) or mouse (mm9) genome using Tophat (v2.1.1)66. PCR duplicates were further removed using samtools (v1.3.1)67, gene counts were computed using HTseq (v0.6.1)68, gene expression level (FPKM) was further calculated using cufflinks (v2.1.1)66. Gene expression matrix was subjected to R (v3.1.2) or Prism9 for graphics. Single-cell and bulk DNA methylation sequencing data analysis—Raw fastq reads from both the single-cell and bulk DNA methylation sequencing were first trimmed using TrimGalore (v0.4.3) (https://github.com/FelixKrueger/TrimGalore), and cleaned reads were aligned to the human hg19 or mouse mm9 genome (in silico bisulfite converted) using Bismark tool (v0.17.0)69. Samtools (v1.3.1)67 was used to remove PCR duplicates, and CpG methylation calls were extracted using the Bismark methylation extractor69. 0.1% lambda DNA was spiked in, prior to bisulfite treatment, for each sample to assess the bisulfite conversion efficiency. Only samples with more than 4 million unique CpG sites covered at least once and with a bisulfite conversion rate > 98% were used in this study. TCGA methylation array data reanalysis—Prostate DNA methylation datasets from TCGA analyzed by Illumina Infinium Human Methylation 450 K BeadChip were downloaded from the National Cancer Institute’s GDC Data Portal (https:// portal.gdc.cancer.gov) for 502 tumor samples and 50 normal samples. CpG site-level methylation files (beta value, txt format) were first converted to hg19 coordinates using UCSC lift-over tool (https://genome.ucsc.edu/cgi-bin/hgLiftOver) for the downstream analysis. The data were binned to a fixed set of 10 kb nonoverlapping genomic windows by computing the average fraction methylation within each bin in each sample. Bins were excluded if they lacked coverage (i.e., had no probes on the Illumina Infinium Human Methylation 450 K BeadChip array) or had a mean normal-tissue methylation level, averaged across all the normal samples, of <70%. For each sample, the global methylation level was calculated as the fraction of bins having methylation >50%. The methylation level at the CD1A-IFI16 locus for each sample was calculated as the fraction of bins in the range chr1:158,130,000–158,340,000 (hg19) having methylation >50%. The gene expression data and clinical information of TCGA PRAD samples, including Gleason score, tumor stage Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 23 and others, were all downloaded from cbioportal (https://www.cbioportal.org/). Tumor purity was calculated using ABSOLUTE algorithm70. DNA Methylation 450 K BeadChip datasets for other cancer types were also downloaded from the National Cancer Institute’s GDC Data Portal (https://portal.gdc.cancer.gov) and CpG site-level methylation files (beta value, txt format) were also converted to hg19 coordinates using UCSC lift-over tool (https:// genome.ucsc.edu/cgi-bin/hgLiftOver) for the downstream analysis. Genomic element enrichment analysis—For analytical purposes, a promoter region was defined based on the relative position to a transcription start site (TSS): 1,500 bp upstream and 500 bp downstream. The annotations of TSS, exon, intron, intragenic regions, CpG islands (CGIs), repetitive elements and UCSC gap regions were all downloaded from UCSC genome table browser (https://genome.ucsc.edu/cgi-bin/hgTables)71. Enrichment analysis on different genomic elements was calculated using the Bioconductor package regioneR (v1.18.1) with overlapPermTest function72. DNA copy number analysis inferred by single-cell DNA methylation sequencing data—Single-cell DNA methylation sequencing reads were first aligned to the genome using Bismark. Uniquely aligned reads were extracted into a bed file and subsequently submitted to Ginkgo online tool73, http://qb.cshl.edu/ginkgo) to infer the DNA copy number, using 5 Mb as the bin size. The processed integer copy number data from the Ginkgo website (SegCopy.tsv) was used to calculate the DNA Copy Number Variation (CNV) score. Given an assignment of a copy number to all the locations in a diploid genome, we define a CNV score for any given single cells as follows. Let ci be the copy number at the ith location of the genome. CNV score is then defined to be the average over all i in the genome of the absolute value of (ci-2). DNA copy number analysis inferred by single-cell RNA-seq data—Single-cell RNA-seq reads were aligned to human genome using TopHat, and large-scale chromosomal copy number alterations were determined by InferCNV (https://github.com/broadinstitute/ infercnv). MNase ChIP-seq and Cut and Run data analysis—ChIP-seq and Cut and Run reads were first trimmed by Trim Galore (v0.4.3) (https://github.com/FelixKrueger/TrimGalore) and then mapped to the human or mouse genome using BWA men74. Duplicated reads were marked by sambamba75 and further removed using samtools67. MACS2 (v2.0.10)76 was used to call the peaks and deepTools77 were used to compute the ChIP-seq or Cut and Run signal around prostate PMDs. Determination of Partially Methylated Domains (PMDs)—The human genome was first binned into 100 kb windows placed at 200 bp offsets. Windows that intersected CGIs or UCSC gap regions were discarded. For each source (i.e., single CTCs from patients with prostate cancer, single WBCs from healthy donors, single cells from normal prostate or prostate cancer cell lines or normal prostate tissues42, the per-source methylation level of each window was calculated by taking the average over all cells from that source of the methylation level of the CpG sites within the given window. For each source the distribution of the per-source methylation level of the 100 kb windows was plotted. Normal cells Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 24 showed a unimodal distribution, while prostate cancer cells showed a bimodal distribution. A threshold for hypomethylation determination was set at the lowest point of the valley in the histogram of the bimodal distribution for each prostate cancer patient or prostate cell line; if the distribution was unimodal, the threshold was set to 60%. The windows with methylation level lower than threshold were defined as hypomethylation windows and overlapping hypomethylation windows were merged into per-source PMDs. The 250 kb minimal length threshold was then applied to the per-source PMDs. The union of the per-source PMDs for all single CTCs from four prostate cancer patients (GU114, GU216, GURa15 and GU181) and for all single cells from four prostate cancer cell lines (LNCaP, VCaP, 22Rv1 and PC3) was defined as the total prostate PMDs (1,496 in total). Chromatin mark and genome element enrichment analyses were performed on these PMDs. To identify the genes that reside in the most consistently hypomethylated PMDs across all prostate cancer specimens analyzed (i.e., intersection), we quantile-normalized the DNA methylation levels for all PMDs among all CTCs from four prostate cancer patients (GU114, GU216, GURa15 and GU181) and all single cells from four prostate cancer cell lines (LNCaP, VCaP, 22Rv1 and PC3) and only used the PMDs (annotated as core prostate PMDs) with their averaged quantile-normalized DNA methylation level less than 25% across these 8 sources to extract the genes. Determination of Preserved Methylation Islands (PMIs)—After identification of PMDs for each of the eight sample sources [CTCs from four prostate cancer patients (GU114, GU216, GURa15 and GU181) and single cells from four prostate cancer cell lines (LNCaP, VCaP, 22Rv1 and PC3)], we defined small interspersed islands (“gaps”) with preserved methylation (sample source PMIs) using the following criteria: (1) every PMI is flanked by defined PMDs in each given source; (2) length of each PMI should be >30 kb and <3 Mb. Total prostate PMIs were defined by taking the union of sample source PMIs across 8 sources (1,412 in total), while core prostate PMIs (44 in total) were defined by requiring the uniformity across sample sources: the genomic location of given PMI is overlapped in all 8 sample sources. Differential gene expression and hypergeometric gene set enrichment analysis (hGSEA)—Differential gene expression between TCGA prostate normal tissue and primary tumors was determined as follows: We started by considering the genes that reside in the most hypomethylated PMDs [as described in the section titled “Determination of partially methylated domains (PMDs)”]. Of those, genes with 95th percentile of normalized FPKM values less than 1 were discarded. A two-tailed variance-equal t-test was performed on each of the remaining genes. The p-values from those t-tests were used to generate a false-discovery rate (FDR) estimate for each gene by the Benjamini- Hochberg method. We considered genes for which the FDR estimate was less than 0.1 to be differentially expressed between normal prostate and prostate tumor samples. hGSEA was performed to determine the gene set and pathway enrichment using the phyper R function as reported elsewhere26. All gene sets and pathways evaluated in this study were obtained from MSigDB (v7.2) from the Broad Institute. Differential gene expression and hGSEA for genes in PMIs was performed in the same way. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 25 Heterogeneity assessment—Consistent with a previous publication26, means of correlation coefficients and jackknife estimates were used to assess the heterogeneity within and between subsets of samples. Nanopore data analysis—Nanopore sequencing reads (format: fast5) generated by Nanopore MinION device were first converted into fastq files using ONT Albacore software (v2.3.1) (https://nanoporetech.com/community). Demultiplexing was also performed during fast5 to fastq conversion. DNA methylation information was extracted from both fast5 and fastq files using Nanopolish software (v0.10.2) (https://github.com/nanoporetech/ nanopolish). Nanopolish output files (albacore_output.sorted.bam and methylation_calls.tsv) were used for downstream analysis. Every nanopore run was spiked in with lambda DNA, which was used as the negative control to assess the fidelity of Nanopore sequencing. To estimate CTC-derived hypomethylation signal in each Nanopore sequencing sample, stringent criteria were applied: (1) each Nanopore read should be long enough to harbor at least 30 CpG sites with confident methylation calls after Nanopolish; (2) the number of Nanopore reads aligned to prostate PMDs (pre-determined among CTCs isolated from 4 prostate cancer patients and 4 prostate cancer cell lines using single-cell whole genome bisulfite sequencing) should be no fewer than 300 for metastatic patients or no fewer than 400 for localized patients; (3) methylation level of spike-in lambda DNA in each run should be <1%. Following application of these criteria, microfluidic processed (leukocyte- depleted) blood samples from seven patients with metastatic prostate cancer, six patients with localized prostate cancer. Since we required different number of Nanopore reads in the prostate PMDs for metastatic patients and localized patients, 23 age-matched healthy donors were validated for analysis in the metastatic cohort, and 21 were validated for localized cohort. In-silico mathematical modeling of Nanopore sequencing in detecting rare signal—To assess the ability to detect large hypomethylated domains in rare circulating tumor cells, we performed an analysis using Nanopore reads from a normally methylated non-cancer cell line (HUES64) with 1% in-silico spiked-in reads from a cancer cell line (HCT116). We assessed the ability to determine the correct cell line of origin for reads that aligned to predefined HCT116 PMDs based on their average methylation level by quantifying the precision and sensitivity of read classification using the PRROC78. Methylation was averaged across each read, considering only CpG sites that fall within PMDs and excluding those within CpG islands. Illustration—Illustrations were created with BioRender.com. Quantification and Statistical Analysis Statistical analyses for all experiments are described in the figure legends and the method details. Statistical analyses were performed using R (version 3.1.2) and GraphPad Prism 9.0. Supplementary Material Refer to Web version on PubMed Central for supplementary material. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 26 Acknowledgments We thank L. Libby for technical support; J. Fung for flow cytometry assistance. We thank R. Manguso, D. Sen and all lab members in Haber/Maheswaran lab for discussions. This work was supported by grants from National Institute of Health (2RO1CA129933 to D.A.H, U01EB012493 to M.T., D.A.H., S.M., U01CA268933 to M.T., R01CA259007 to D.T.M., 5P41EB002503 to M.T.), Howard Hughes Medical Institute (to D.A.H.), ESSCO Breast Cancer Research Fund (to S.M.), Prostate Cancer Foundation (to D.T.M., R.J.L., and J.A.E.), Cygnus Montanus Foundation founded by the Svanberg Family (to D.T.M.), Breast Cancer Research Foundation (to D.A.H.), National Foundation for Cancer Research (to D.A.H.) and Max Planck Institute (to A.M.) References 1. Feinberg AP, and Vogelstein B (1983). Hypomethylation distinguishes genes of some human cancers from their normal counterparts. Nature 301, 89–92. 10.1038/301089a0. [PubMed: 6185846] 2. Goelz SE, Vogelstein B, Hamilton SR, and Feinberg AP (1985). Hypomethylation of DNA from benign and malignant human colon neoplasms. Science 228, 187–190. 10.1126/science.2579435. [PubMed: 2579435] 3. Baylin S, and Bestor TH (2002). Altered methylation patterns in cancer cell genomes: cause or consequence? Cancer Cell 1, 299–305. 10.1016/s1535-6108(02)00061-2. [PubMed: 12086841] 4. Baylin SB, and Jones PA (2016). Epigenetic Determinants of Cancer. Cold Spring Harb Perspect Biol 8. 10.1101/cshperspect.a019505. 5. Esteller M (2002). CpG island hypermethylation and tumor suppressor genes: a booming present, a brighter future. Oncogene 21, 5427–5440. 10.1038/sj.onc.1205600. [PubMed: 12154405] 6. Yegnasubramanian S, Kowalski J, Gonzalgo ML, Zahurak M, Piantadosi S, Walsh PC, Bova GS, De Marzo AM, Isaacs WB, and Nelson WG (2004). Hypermethylation of CpG islands in primary and metastatic human prostate cancer. Cancer Res 64, 1975–1986. 10.1158/0008-5472.can-03-3972. [PubMed: 15026333] 7. Esteller M, Corn PG, Baylin SB, and Herman JG (2001). A gene hypermethylation profile of human cancer. Cancer Res 61, 3225–3229. [PubMed: 11309270] 8. Berman BP, Weisenberger DJ, Aman JF, Hinoue T, Ramjan Z, Liu Y, Noushmehr H, Lange CP, van Dijk CM, Tollenaar RA, et al. (2011). Regions of focal DNA hypermethylation and long-range hypomethylation in colorectal cancer coincide with nuclear lamina-associated domains. Nat Genet 44, 40–46. 10.1038/ng.969. [PubMed: 22120008] 9. Timp W, Bravo HC, McDonald OG, Goggins M, Umbricht C, Zeiger M, Feinberg AP, and Irizarry RA (2014). Large hypomethylated blocks as a universal defining epigenetic alteration in human solid tumors. Genome Med 6, 61. 10.1186/s13073-014-0061-y. [PubMed: 25191524] 10. Hansen KD, Timp W, Bravo HC, Sabunciyan S, Langmead B, McDonald OG, Wen B, Wu H, Liu Y, Diep D, et al. (2011). Increased methylation variation in epigenetic domains across cancer types. Nat Genet 43, 768–775. 10.1038/ng.865. [PubMed: 21706001] 11. Yegnasubramanian S, Haffner MC, Zhang Y, Gurel B, Cornish TC, Wu Z, Irizarry RA, Morgan J, Hicks J, DeWeese TL, et al. (2008). DNA hypomethylation arises later in prostate cancer progression than CpG island hypermethylation and contributes to metastatic tumor heterogeneity. Cancer Res 68, 8954–8967. 10.1158/0008-5472.CAN-07-6088. [PubMed: 18974140] 12. Hon GC, Hawkins RD, Caballero OL, Lo C, Lister R, Pelizzola M, Valsesia A, Ye Z, Kuan S, Edsall LE, et al. (2012). Global DNA hypomethylation coupled to repressive chromatin domain formation and gene silencing in breast cancer. Genome Res 22, 246–258. 10.1101/gr.125872.111. [PubMed: 22156296] 13. Johnstone SE, Reyes A, Qi Y, Adriaens C, Hegazi E, Pelka K, Chen JH, Zou LS, Drier Y, Hecht V, et al. (2020). Large-Scale Topological Changes Restrain Malignant Progression in Colorectal Cancer. Cell 182, 1474–1489 e1423. 10.1016/j.cell.2020.07.030. [PubMed: 32841603] 14. Hansen KD, Sabunciyan S, Langmead B, Nagy N, Curley R, Klein G, Klein E, Salamon D, and Feinberg AP (2014). Large-scale hypomethylated blocks associated with Epstein-Barr virus- induced B-cell immortalization. Genome Res 24, 177–184. 10.1101/gr.157743.113. [PubMed: 24068705] Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 27 15. Zhou W, Dinh HQ, Ramjan Z, Weisenberger DJ, Nicolet CM, Shen H, Laird PW, and Berman BP (2018). DNA methylation loss in late-replicating domains is linked to mitotic cell division. Nat Genet 50, 591–602. 10.1038/s41588-018-0073-4. [PubMed: 29610480] 16. Epstein JI, Egevad L, Amin MB, Delahunt B, Srigley JR, Humphrey PA, and Grading C (2016). The 2014 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma: Definition of Grading Patterns and Proposal for a New Grading System. Am J Surg Pathol 40, 244–252. 10.1097/PAS.0000000000000530. [PubMed: 26492179] 17. Epstein JI, Zelefsky MJ, Sjoberg DD, Nelson JB, Egevad L, Magi-Galluzzi C, Vickers AJ, Parwani AV, Reuter VE, Fine SW, et al. (2016). A Contemporary Prostate Cancer Grading System: A Validated Alternative to the Gleason Score. Eur Urol 69, 428–435. 10.1016/j.eururo.2015.06.046. [PubMed: 26166626] 18. Fay EK, and Graff JN (2020). Immunotherapy in Prostate Cancer. Cancers (Basel) 12. 10.3390/ cancers12071752. 19. Sfanos KS, Bruno TC, Maris CH, Xu L, Thoburn CJ, DeMarzo AM, Meeker AK, Isaacs WB, and Drake CG (2008). Phenotypic analysis of prostate-infiltrating lymphocytes reveals TH17 and Treg skewing. Clin Cancer Res 14, 3254–3261. 10.1158/1078-0432.CCR-07-5164. [PubMed: 18519750] 20. Wang S, He Z, Wang X, Li H, and Liu XS (2019). Antigen presentation and tumor immunogenicity in cancer immunotherapy response prediction. Elife 8. 10.7554/eLife.49020. 21. Venturini NJ, and Drake CG (2019). Immunotherapy for Prostate Cancer. Cold Spring Harb Perspect Med 9. 10.1101/cshperspect.a030627. 22. Liu L, Toung JM, Jassowicz AF, Vijayaraghavan R, Kang H, Zhang R, Kruglyak KM, Huang HJ, Hinoue T, Shen H, et al. (2018). Targeted methylation sequencing of plasma cell-free DNA for cancer detection and classification. Ann Oncol 29, 1445–1453. 10.1093/annonc/mdy119. [PubMed: 29635542] 23. Haber DA, and Velculescu VE (2014). Blood-based analyses of cancer: circulating tumor cells and circulating tumor DNA. Cancer Discov 4, 650–661. 10.1158/2159-8290.CD-13-1014. [PubMed: 24801577] 24. Liu MC, Oxnard GR, Klein EA, Swanton C, Seiden MV, and Consortium C (2020). Sensitive and specific multi-cancer detection and localization using methylation signatures in cell-free DNA. Ann Oncol 31, 745–759. 10.1016/j.annonc.2020.02.011. [PubMed: 33506766] 25. Klein EA, Richards D, Cohn A, Tummala M, Lapham R, Cosgrove D, Chung G, Clement J, Gao J, Hunkapiller N, et al. (2021). Clinical validation of a targeted methylation-based multi-cancer early detection test using an independent validation set. Ann Oncol 32, 1167–1177. 10.1016/ j.annonc.2021.05.806. [PubMed: 34176681] 26. Miyamoto DT, Zheng Y, Wittner BS, Lee RJ, Zhu H, Broderick KT, Desai R, Fox DB, Brannigan BW, Trautwein J, et al. (2015). RNA-Seq of single prostate CTCs implicates noncanonical Wnt signaling in antiandrogen resistance. Science 349, 1351–1356. 10.1126/science.aab0917. [PubMed: 26383955] 27. Ozkumur E, Shah AM, Ciciliano JC, Emmink BL, Miyamoto DT, Brachtel E, Yu M, Chen PI, Morgan B, Trautwein J, et al. (2013). Inertial focusing for tumor antigen-dependent and -independent sorting of rare circulating tumor cells. Sci Transl Med 5, 179ra147. 5/179/179ra47 [pii] 10.1126/scitranslmed.3005616. 28. Taylor RA, Toivanen R, Frydenberg M, Pedersen J, Harewood L, Australian Prostate Cancer B, Collins AT, Maitland NJ, and Risbridger GP (2012). Human epithelial basal cells are cells of origin of prostate cancer, independent of CD133 status. Stem Cells 30, 1087–1096. 10.1002/ stem.1094. [PubMed: 22593016] 29. Stoyanova T, Cooper AR, Drake JM, Liu X, Armstrong AJ, Pienta KJ, Zhang H, Kohn DB, Huang J, Witte ON, and Goldstein AS (2013). Prostate cancer originating in basal cells progresses to adenocarcinoma propagated by luminal-like cells. Proc Natl Acad Sci U S A 110, 20111–20116. 10.1073/pnas.1320565110. [PubMed: 24282295] 30. Lee SO, Tian J, Huang CK, Ma Z, Lai KP, Hsiao H, Jiang M, Yeh S, and Chang C (2012). Suppressor role of androgen receptor in proliferation of prostate basal epithelial and progenitor cells. J Endocrinol 213, 173–182. 10.1530/JOE-11-0474. [PubMed: 22393245] Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 28 31. Hayward SW, Dahiya R, Cunha GR, Bartek J, Deshpande N, and Narayan P (1995). Establishment and characterization of an immortalized but non-transformed human prostate epithelial cell line: BPH-1. In Vitro Cell Dev Biol Anim 31, 14–24. 10.1007/BF02631333. [PubMed: 7535634] 32. Smallwood SA, Lee HJ, Angermueller C, Krueger F, Saadeh H, Peat J, Andrews SR, Stegle O, Reik W, and Kelsey G (2014). Single-cell genome-wide bisulfite sequencing for assessing epigenetic heterogeneity. Nat Methods 11, 817–820. 10.1038/nmeth.3035. [PubMed: 25042786] 33. Bian S, Hou Y, Zhou X, Li X, Yong J, Wang Y, Wang W, Yan J, Hu B, Guo H, et al. (2018). Single-cell multiomics sequencing and analyses of human colorectal cancer. Science 362, 1060– 1063. 10.1126/science.aao3791. [PubMed: 30498128] 34. Zhao SG, Chen WS, Li H, Foye A, Zhang M, Sjostrom M, Aggarwal R, Playdle D, Liao A, Alumkal JJ, et al. (2020). The DNA methylation landscape of advanced prostate cancer. Nat Genet 52, 778–789. 10.1038/s41588-020-0648-8. [PubMed: 32661416] 35. Aran D, Toperoff G, Rosenberg M, and Hellman A (2011). Replication timing-related and gene body-specific methylation of active human genes. Hum Mol Genet 20, 670–680. 10.1093/hmg/ ddq513. [PubMed: 21112978] 36. Hermann A, Goyal R, and Jeltsch A (2004). The Dnmt1 DNA-(cytosine-C5)-methyltransferase methylates DNA processively with high preference for hemimethylated target sites. J Biol Chem 279, 48350–48359. 10.1074/jbc.M403427200. [PubMed: 15339928] 37. Johnstone SE, Gladyshev VN, Aryee MJ, and Bernstein BE (2022). Epigenetic clocks, aging, and cancer. Science 378, 1276–1277. 10.1126/science.abn4009. [PubMed: 36548410] 38. Bucay N, Sekhon K, Majid S, Yamamura S, Shahryari V, Tabatabai ZL, Greene K, Tanaka Y, Dahiya R, Deng G, and Saini S (2016). Novel tumor suppressor microRNA at frequently deleted chromosomal region 8p21 regulates epidermal growth factor receptor in prostate cancer. Oncotarget 7, 70388–70403. 10.18632/oncotarget.11865. [PubMed: 27611943] 39. Cancer Genome Atlas Research, N. (2015). The Molecular Taxonomy of Primary Prostate Cancer. Cell 163, 1011–1025. 10.1016/j.cell.2015.10.025. [PubMed: 26544944] 40. Bethel CR, Faith D, Li X, Guan B, Hicks JL, Lan F, Jenkins RB, Bieberich CJ, and De Marzo AM (2006). Decreased NKX3.1 protein expression in focal prostatic atrophy, prostatic intraepithelial neoplasia, and adenocarcinoma: association with gleason score and chromosome 8p deletion. Cancer Res 66, 10683–10690. 10.1158/0008-5472.CAN-06-0963. [PubMed: 17108105] 41. Bova GS, Carter BS, Bussemakers MJ, Emi M, Fujiwara Y, Kyprianou N, Jacobs SC, Robinson JC, Epstein JI, Walsh PC, and et al. (1993). Homozygous deletion and frequent allelic loss of chromosome 8p22 loci in human prostate cancer. Cancer Res 53, 3869–3873. [PubMed: 7689419] 42. Yu YP, Ding Y, Chen R, Liao SG, Ren BG, Michalopoulos A, Michalopoulos G, Nelson J, Tseng GC, and Luo JH (2013). Whole-genome methylation sequencing reveals distinct impact of differential methylations on gene transcription in prostate cancer. Am J Pathol 183, 1960–1970. 10.1016/j.ajpath.2013.08.018. [PubMed: 24113458] 43. Van Kaer L, Wu L, and Joyce S (2016). Mechanisms and Consequences of Antigen Presentation by CD1. Trends Immunol 37, 738–754. 10.1016/j.it.2016.08.011. [PubMed: 27623113] 44. Brennan PJ, Brigl M, and Brenner MB (2013). Invariant natural killer T cells: an innate activation scheme linked to diverse effector functions. Nat Rev Immunol 13, 101–117. 10.1038/nri3369. [PubMed: 23334244] 45. Godfrey DI, MacDonald HR, Kronenberg M, Smyth MJ, and Van Kaer L (2004). NKT cells: what’s in a name? Nat Rev Immunol 4, 231–237. 10.1038/nri1309. [PubMed: 15039760] 46. Hix LM, Shi YH, Brutkiewicz RR, Stein PL, Wang CR, and Zhang M (2011). CD1d-expressing breast cancer cells modulate NKT cell-mediated antitumor immunity in a murine model of breast cancer metastasis. PLoS One 6, e20702. 10.1371/journal.pone.0020702. [PubMed: 21695190] 47. McEwen-Smith RM, Salio M, and Cerundolo V (2015). The regulatory role of invariant NKT cells in tumor immunity. Cancer Immunol Res 3, 425–435. 10.1158/2326-6066.CIR-15-0062. [PubMed: 25941354] 48. Connolly DJ, and Bowie AG (2014). The emerging role of human PYHIN proteins in innate immunity: implications for health and disease. Biochem Pharmacol 92, 405–414. 10.1016/ j.bcp.2014.08.031. [PubMed: 25199457] Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 29 49. Watson PA, Ellwood-Yen K, King JC, Wongvipat J, Lebeau MM, and Sawyers CL (2005). Context-dependent hormone-refractory progression revealed through characterization of a novel murine prostate cancer cell line. Cancer Res 65, 11565–11571. 10.1158/0008-5472.CAN-05-3441. [PubMed: 16357166] 50. Shen H, Gu C, Liang T, Liu H, Guo F, and Liu X (2020). Unveiling the heterogeneity of NKT cells in the liver through single cell RNA sequencing. Sci Rep 10, 19453. 10.1038/s41598-020-76659-1. [PubMed: 33173202] 51. Stott SL, Lee RJ, Nagrath S, Yu M, Miyamoto DT, Ulkus L, Inserra EJ, Ulman M, Springer S, Nakamura Z, et al. (2010). Isolation and characterization of circulating tumor cells from patients with localized and metastatic prostate cancer. Sci Transl Med 2, 25ra23. 2/25/25ra23 [pii] 10.1126/ scitranslmed.3000403. 52. Seiden MV, Kantoff PW, Krithivas K, Propert K, Bryant M, Haltom E, Gaynes L, Kaplan I, Bubley G, DeWolf W, and et al. (1994). Detection of circulating tumor cells in men with localized prostate cancer. J Clin Oncol 12, 2634–2639. [PubMed: 7527455] 53. Miyamoto DT, Lee RJ, Kalinich M, LiCausi JA, Zheng Y, Chen T, Milner JD, Emmons E, Ho U, Broderick K, et al. (2018). An RNA-Based Digital Circulating Tumor Cell Signature Is Predictive of Drug Response and Early Dissemination in Prostate Cancer. Cancer Discov 8, 288– 303. 10.1158/2159-8290.CD-16-1406. [PubMed: 29301747] 54. Simpson JT, Workman RE, Zuzarte PC, David M, Dursi LJ, and Timp W (2017). Detecting DNA cytosine methylation using nanopore sequencing. Nat Methods 14, 407–410. 10.1038/nmeth.4184. [PubMed: 28218898] 55. Rand AC, Jain M, Eizenga JM, Musselman-Brown A, Olsen HE, Akeson M, and Paten B (2017). Mapping DNA methylation with high-throughput nanopore sequencing. Nat Methods 14, 411– 413. 10.1038/nmeth.4189. [PubMed: 28218897] 56. Lieberman-Aiden E, van Berkum NL, Williams L, Imakaev M, Ragoczy T, Telling A, Amit I, Lajoie BR, Sabo PJ, Dorschner MO, et al. (2009). Comprehensive mapping of long-range interactions reveals folding principles of the human genome. Science 326, 289–293. 10.1126/ science.1181369. [PubMed: 19815776] 57. Wang ZT, Chen SD, Xu W, Chen KL, Wang HF, Tan CC, Cui M, Dong Q, Tan L, Yu JT, and Alzheimer’s Disease Neuroimaging I (2019). Genome-wide association study identifies CD1A associated with rate of increase in plasma neurofilament light in non-demented elders. Aging (Albany NY) 11, 4521–4535. 10.18632/aging.102066. [PubMed: 31295725] 58. Liu H, Xing Y, Guo Y, Liu P, Zhang H, Xue B, Shou J, Qian J, Peng J, Wang R, et al. (2016). Polymorphisms in exon 2 of CD1 genes are associated with susceptibility to Guillain-Barre syndrome. J Neurol Sci 369, 39–42. 10.1016/j.jns.2016.07.029. [PubMed: 27653862] 59. Schreiber RD, Old LJ, and Smyth MJ (2011). Cancer immunoediting: integrating immunity’s roles in cancer suppression and promotion. Science 331, 1565–1570. 10.1126/science.1203486. [PubMed: 21436444] 60. Waldman AD, Fritz JM, and Lenardo MJ (2020). A guide to cancer immunotherapy: from T cell basic science to clinical practice. Nat Rev Immunol 20, 651–668. 10.1038/s41577-020-0306-5. [PubMed: 32433532] 61. Kumaki Y, Oda M, and Okano M (2008). QUMA: quantification tool for methylation analysis. Nucleic Acids Res 36, W170–175. 10.1093/nar/gkn294. [PubMed: 18487274] 62. Hou Y, Guo H, Cao C, Li X, Hu B, Zhu P, Wu X, Wen L, Tang F, Huang Y, and Peng J (2016). Single-cell triple omics sequencing reveals genetic, epigenetic, and transcriptomic heterogeneity in hepatocellular carcinomas. Cell Res 26, 304–319. 10.1038/cr.2016.23. [PubMed: 26902283] 63. Picelli S, Faridani OR, Bjorklund AK, Winberg G, Sagasser S, and Sandberg R (2014). Full-length RNA-seq from single cells using Smart-seq2. Nat Protoc 9, 171–181. 10.1038/nprot.2014.006. [PubMed: 24385147] 64. Clark SJ, Smallwood SA, Lee HJ, Krueger F, Reik W, and Kelsey G (2017). Genome-wide base-resolution mapping of DNA methylation in single cells using single-cell bisulfite sequencing (scBS-seq). Nat Protoc 12, 534–547. 10.1038/nprot.2016.187. [PubMed: 28182018] Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 30 65. Brind’Amour J, Liu S, Hudson M, Chen C, Karimi MM, and Lorincz MC (2015). An ultra-low- input native ChIP-seq protocol for genome-wide profiling of rare cell populations. Nat Commun 6, 6033. 10.1038/ncomms7033. [PubMed: 25607992] 66. Trapnell C, Roberts A, Goff L, Pertea G, Kim D, Kelley DR, Pimentel H, Salzberg SL, Rinn JL, and Pachter L (2012). Differential gene and transcript expression analysis of RNA-seq experiments with TopHat and Cufflinks. Nat Protoc 7, 562–578. 10.1038/nprot.2012.016. [PubMed: 22383036] 67. Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R, and Genome Project Data Processing, S. (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079. 10.1093/bioinformatics/btp352. [PubMed: 19505943] 68. Anders S, Pyl PT, and Huber W (2015). HTSeq--a Python framework to work with high-throughput sequencing data. Bioinformatics 31, 166–169. 10.1093/bioinformatics/btu638. [PubMed: 25260700] 69. Krueger F, and Andrews SR (2011). Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 27, 1571–1572. 10.1093/bioinformatics/btr167. [PubMed: 21493656] 70. Carter SL, Cibulskis K, Helman E, McKenna A, Shen H, Zack T, Laird PW, Onofrio RC, Winckler W, Weir BA, et al. (2012). Absolute quantification of somatic DNA alterations in human cancer. Nat Biotechnol 30, 413–421. nbt.2203 [pii] 10.1038/nbt.2203. [PubMed: 22544022] 71. Karolchik D, Hinrichs AS, Furey TS, Roskin KM, Sugnet CW, Haussler D, and Kent WJ (2004). The UCSC Table Browser data retrieval tool. Nucleic Acids Res 32, D493–496. 10.1093/nar/ gkh103. [PubMed: 14681465] 72. Gel B, Diez-Villanueva A, Serra E, Buschbeck M, Peinado MA, and Malinverni R (2016). regioneR: an R/Bioconductor package for the association analysis of genomic regions based on permutation tests. Bioinformatics 32, 289–291. 10.1093/bioinformatics/btv562. [PubMed: 26424858] 73. Garvin T, Aboukhalil R, Kendall J, Baslan T, Atwal GS, Hicks J, Wigler M, and Schatz MC (2015). Interactive analysis and assessment of single-cell copy-number variations. Nat Methods 12, 1058–1060. 10.1038/nmeth.3578. [PubMed: 26344043] 74. Li H, and Durbin R (2009). Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760. 10.1093/bioinformatics/btp324. [PubMed: 19451168] 75. Tarasov A, Vilella AJ, Cuppen E, Nijman IJ, and Prins P (2015). Sambamba: fast processing of NGS alignment formats. Bioinformatics 31, 2032–2034. 10.1093/bioinformatics/btv098. [PubMed: 25697820] 76. Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, and Liu XS (2008). Model-based analysis of ChIP-Seq (MACS). Genome Biol 9, R137. 10.1186/gb-2008-9-9-r137. [PubMed: 18798982] 77. Ramirez F, Ryan DP, Gruning B, Bhardwaj V, Kilpert F, Richter AS, Heyne S, Dundar F, and Manke T (2016). deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res 44, W160–165. 10.1093/nar/gkw257. [PubMed: 27079975] 78. Grau J, Grosse I, and Keilwagen J (2015). PRROC: computing and visualizing precision- recall and receiver operating characteristic curves in R. Bioinformatics 31, 2595–2597. 10.1093/ bioinformatics/btv153. [PubMed: 25810428] Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 31 Highlights 1. 2. 3. 4. 40 core hypomethylated domains across prostate CTCs arise early in tumorigenesis. Hypomethylation silences immune-related genes, sparing adjacent proliferation genes. The CD1A-IFI16 immune locus is consistently silenced by hypomethylation in cancer. Hypomethylated domains detected in CTC-enriched blood in localized prostate cancer. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 32 Figure 1. Partially Methylated Domains (PMDs) and Preserved Methylation Islands (PMIs) in single metastatic prostate cancer cells. (A) Schematic of CTC enrichment (104-fold leukocyte depletion), and paired DNA methylation sequencing (nucleus) and RNA-seq (cytoplasm) from individual prostate CTCs. (B) Confirmation of CTC identity using stringent RNA expression thresholding of prostatic lineage and epithelial versus leukocyte markers. Maximum log10 (RPM) expression of epithelial (KRT7, KRT8, KRT18, KRT19, EPCAM) and prostatic markers (AR, KLK3, FOLH1, AMACR) are plotted against leukocyte markers (CD45, CD16, CD37, CD53, CD7, CD66b). Only confirmed CTCs without WBC contamination (red crosses) were used in analyses. (C) Representative DNA copy number variation (CNV) analysis in individual CTCs from two patients, compared with a diploid normal prostate epithelial cell (HPrEC) and a healthy donor-derived leukocyte. Single-cell DNA methylation sequencing data was used to infer DNA copy number. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 33 (D) IGV representation (hg19) of DNA methylation spanning chromosome 8, showing extensive PMDs (yellow) across 37 individual CTCs from four patients (GU114, GU216, GU181 and GURa15), and 17 cells from prostate cancer cell lines (LNCaP, PC3, VCaP, 22Rv1). As controls, 4 normal bulk prostate tissues (N.P.), 36 cells from two prostate epithelial cell lines (HPrEC, BPH-1) and normal leukocytes (WBCs) are shown. Normal methylation level (blue). (E-F) Higher resolution of chromosome 8 in IGV, showing precise PMD boundaries shared across individual CTCs and prostate cancer cell lines (panel E), with magnified view of the nested PMI, bracketing a few genes, with precise boundaries of preserved methylation flanked by profound hypomethylation (panel F). (G-H) Components of coding genes and classes of repeats differentially enriched in PMDs versus PMIs (panel G), with differences among subtypes of repeats (panel H). ns, not significant; P<0.05; *P<0.01, assessed by permutation test. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 34 Figure 2. Acquired chromatin marks in prostate cancer PMDs and nomination of shared core PMDs. (A) Differential enrichment of chromatin marks within prostate cancer PMDs and PMIs. Annotated chromatin marks from ChIP-seq dataset of PC3 cells in ENCODE (https:// www.encodeproject.org/). ns, not significant; P<0.05; P<0.01, assessed by permutation test. (B) Line plots showing differential enrichment of silencing chromatin marks at PMDs across the genome in prostate cancer cells (LNCaP; 3 biological replicates, red lines), compared with cultured benign prostatic hyperplasia cells (BPH-1; 2 biological replicates, green lines) and normal prostate epithelial cells (HPrEC; 2 biological replicates, blue lines). Across the genome, prostate cancer cells acquire H3K27me3, with highest levels at the boundaries of PMDs (left panel), whereas H3K9me3 enrichment towards the center of PMDs is not altered between cancer and non-transformed prostate cells (right panel). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 35 (C) Boxplot showing enrichment of Cut and Run signal for H3K27me3, but not H3K9me3, across prostate cancer PMDs between LNCaP cells and non-transformed cell lines (HPrEC and BPH-1). Pvalue, one-tailed Student’s t-test. (D) IGV track showing representative cancer-associated PMD (DNA hypomethylation: yellow), with pronounced enrichment of H3K27me3 at PMD borders in cancer cells (LNCaP: red) versus non-transformed cells (HPrEC: blue, BPH-1: green), whereas PMD- centered H3K9me3 occupancy is unaltered. (E) Inter- and intra-patient heterogeneity of PMDs among single CTCs from four prostate cancer patients (red) and single cells from prostate cancer cell lines. Mean Jaccard index indicates heterogeneity, with higher mean score indicating less heterogeneity among samples. Error bar, mean with 95% confidence interval (CI). (F-G) IGV representation of total PMDs and core PMDs at chromosome 3 locus, across 8 sample sources (4 patients and 4 prostate cancer cell lines). Total PMDs (blue) are the union of PMDs defined in each sample source, while core PMDs (black) are shared across all 8 sample sources (panel F); representation of PMDs from the single-cell components of an individual sample source (22 CTCs from patient GU181) showing a core PMD shared across all sample sources (black) and neighboring non-core PMDs that are shared by >90% CTCs in this patient, but not across different sample sources (panel G). See Figure S2D and Methods for criteria in core PMD and PMI designation. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 36 Figure 3. Demethylation of core PMDs during early prostate tumorigenesis suppresses immune- related genes, while core PMIs spare proliferation genes. (A) Schematic showing prostate tumor microdissection, single nucleus isolation and single- cell DNA methylation sequencing. (B) Ranking of methylation level at 40 core PMDs (red dots) among all 1,496 total PMDs, as a function of timeline from normal prostate, to localized (GS6; GS8) and metastatic cancer (CTCs), showing early demethylation of core PMDs. Within normal prostate, all 40 core PMDs have methylation level >75%, and 31 are hypomethylated as early as GS6. (C) Quantitation of demethylation as a function of Gleason Score (GS). Demethylation of core PMDs (red curve) precedes that of other PMDs (magenta) within microdissected prostate tumor cells and in CTCs. In contrast, core PMIs nested between PMDs (blue) show minimal DNA methylation changes during tumorigenesis. Error bar, mean with SEM. Statistical analysis of DNA methylation curves utilizing longitudinal linear mixed effects model, by which tumor progression x methylation domains was tested. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 37 (D) Quantitation of demethylation as a function of GS in TCGA prostate cancer methylation array data, showing early and progressive loss of methylation of core PMDs (red curve), with an attenuated trend for other PMDs (magenta). The core PMIs (blue) display stable DNA methylation pattern during prostate tumorigenesis. Statistical analysis as for panel C. (E-F) Gene set enrichment analysis (GSEA) of genes residing within core PMDs and downregulated in primary prostate cancer (E), and of genes residing within core PMIs with gene expression preserved (up-regulated and not significantly changed) in primary prostate cancer (F), compared with normal prostate. (FDR <0.1; two-tailed Student’s t-test with FDR correction). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 38 Figure 4. Correlation of DNA demethylation at the CD1A-IFI16 locus with accumulation of chromatin silencing marks and reduced gene expression. (A) IGV of single-cell DNA methylation at the CD1A-IF16 genomic locus, including five lipid antigen presentation and four interferon inducible genes. Tumor cells (37 single CTCs from four prostate cancer patients (red) and 17 single cells from four prostate cancer cell lines (green)) exhibit marked hypomethylation at this locus (shaded yellow), while normal samples (4 bulk normal prostate tissues, 37 single cells from normal prostate cell lines and leukocytes (blue)) show a preserved DNA methylation (shaded blue). (B) Heatmap (upper panel; hypomethylation shaded yellow) and matched quantitative scatter plots (lower panel) of single-cell DNA methylation levels within all 1,496 prostate cancer PMDs, showing progression from normal prostate to localized prostate cancer (GS6, GS8) and metastatic CTCs. The CD1A-IFI16 locus (dashed vertical red line) shows early and profound demethylation, starting at GS6, with its rank number across all PMDs at each tumor stage shown in parentheses (red). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 39 (C) IGV screenshot of single-cell DNA methylation data showing progressive demethylation of CD1A-IFI16 locus (box with red dashed line) from normal prostate cells to localized (GS6 and GS8) and metastatic prostate cancer (CTCs). Heterogeneity of hypomethylation (shaded yellow) across single cells is evident at GS6, becoming more prevalent at GS8, and uniform in CTCs . (D) Plots showing suppressed expression of lipid antigen presentation and interferon inducible genes within the CD1A-IFI16 locus, during transition from normal prostate to low-grade GS6, with persistent silencing in higher grade GS7, 8 and 9 cancers (TCGA dataset). Error bar, mean with SEM. (E) Analysis of 33 different tumor types (TCGA) for DNA methylation differences at core prostate cancer PMDs, compared with corresponding normal tissues. 30 of 35 (86%) evaluable PMDs are hypomethylated across all tumor types (red circles), with the CD1A- IFI16 locus having the strongest hypomethylation. (F) Histograms of DNA methylation level within 100kb windows (200bp offsets) across the genome in normal prostate cells (BPH-1), following 5-azacytidine treatment (days 1 and 5), compared with DMSO control. (G) Quantitation of H3K27me3-related fluorescence intensity within single-cell nuclei (confocal microscopy). Error bar, mean with SEM. P-value, two-tailed Student’s t-test. (H) Sequential reduction in CD1d protein expression in normal prostate cells (BPH-1) treated with 5-azacytidine, compared with DMSO control. Representative flow cytometry (left panel); median fluorescence intensity (right panel). Error bar, mean with SEM. P-value, two tailed Student’s t-test. (I-J) Western blot showing reduced H3K27 trimethylation in 22Rv1 cells treated with EZH2 inhibitor GSK126 for 6 days (panel H); qPCR of genes within the CD1A-IFI16 cluster show induced expression (panel I), while non-PMD resident control genes (PP1A, HPRT and β-actin) remain unchanged. P-value, Tukey’s multiple comparison tests, where GSK126 treatment conditions (red bars) were compared to controls (blue bar). n.s. not significant; **P<0.0001. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 40 Figure 5. Restoring expression of genes within CD1A-IFI16 syntenic locus abrogates tumorigenesis in an immunocompetent mouse prostate cancer model. (A) Plots quantifying Cd1d1 and Ifi204 mRNA in the murine prostate tumor cell line Myc-CaP, which have silenced the syntenic genes (blue), compared to normal prostate cells from 4 isogenic mice FVB (orange). Ectopic expression of murine Cd1d1 (CD1D ortholog, green) and Ifi204 (IFI16 ortholog, red) is comparable to that of normal prostate. Error bar, mean with SEM. (B) Overexpression (OE) of Cd1d1 or Ifi204 in Myc-CaP cells does not alter in vitro proliferation compared with controls. Error bar, mean with SD. (C) Overexpression of either Cd1d1 (green) or Ifi204 (red) in Myc-CaP cells (mCherry- luciferase tagged) suppresses tumorigenesis in isogenic immunocompetent FVB mice. Mock-transfected control tumors are shown as control (blue). Tumor size quantified by luciferase imaging (representative images). Error bar, mean with SEM. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 41 (D) Myc-CaP cells engineered as in (C) show no difference in tumor growth in immune- deficient NSG mice. Error bar, mean with SEM. (E) Flow cytometry of Cd1d-restored Myc-CaP tumors in FVB mice, showing recruitment of CD1d-restricted NKT cells (marked by α-GalCer CD1d Tetramer) and activated NKT cells (marked by CD69), compared with controls. Error bar, mean with SD. (F) Flow cytometry of Ifi204-restored Myc-CaP tumors in FVB mice, showing unaltered infiltration of total CD4+ and CD8+ T cells, but reduced immune infiltration by PD-1+ CD8+ T cells and increased presence of TNFα+ CD8+ T cells, compared with controls. Error bar, mean with SD. P-values, two-tailed Student’s t-test; ns, not significant. Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 42 Figure 6. Detection of CTC-derived DNA hypomethylation in blood specimens using Nanopore sequencing. (A) IGV screenshot showing concordance of DNA hypomethylation measurements between Oxford Nanopore native sequencing of bulk VCaP cells [B], compared with Illumina bisulfite sequencing of three single VCaP cells (#1, #2, #3). DNA methylation across entire chromosome 4 is shown (hypomethylation in shaded yellow). (B) Scatter plot showing high Pearson correlation (r=0.81) between Nanopore native sequencing and Illumina bisulfite sequencing. (C-D) Mathematical modeling showing minimal precision using short reads (average 5 CpG sites per read) for detection of hypomethylated DNA domains. Modest improvement in detection is provided by interrogating predetermined PMDs, instead of whole genome (panel C). Significantly improved precision is predicted using Nanopore long read sequencing (10 or 50 CpGs per read). Highest predicted accuracy by combining Nanopore long reads (>10 CpG sites per read) with selected analysis-predetermined PMD regions (panel D). Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 43 (E) Schematic of microfluidic CTC enrichment (followed by direct Nanopore sequencing of bulk cells (approximatly 0.1% CTC purity). HMW, high molecular weight. (F-G) Scatter plot quantitation of hypomethylation signal by Nanopore sequencing, comparing leukocyte-depleted blood samples from patients with either metastatic (panel F) or localized prostate cancer before surgical resection or radiation therapy (panel G), versus healthy age-matched male donors (HDs). Error bar denotes mean with SEM. P-value assessed by two-tailed Student’s t-test. Dotted lines indicate thresholds of hypomethylation signal that encompass all healthy donors tested, with the fraction of cancer patients with hypomethylation signal above that threshold considered positive. Cell. Author manuscript; available in PMC 2023 August 18. Guo et al. Page 44 Key resources table REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies Mouse anti-CD45 biotinylated (clone 2D1) R&D Systems Cat# BAM1430; RRID:AB_356874 Mouse anti-CD66b (clone 80H3) Bio-Rad MCA216T; RRID:AB_2291565 Mouse anti-human CD16 biotinylated (clone 3G8) BD Biosciences Cat#555405; RRID:AB_395805 AF488-conjugated mouse anti-human EpCAM (clone VU1D9) Cell Signaling Technology Cat#5198; RRID:AB_10692105 PE-conjugated mouse antibody anti-CD45 (clone HI30) BD Biosciences Cat#560975; RRID:AB_2033960 Rabbit anti-histone H3K27me3 (Western blot) Thermo Fisher Scientific Cat#MA5–11198; RRID:AB_2899176 Rabbit anti-histone H3K27me3 (ChIP and CUT&RUN) Active motif Cat#39155; RRID:AB_2561020 Rabbit anti-histone H3K9me3 (ChIP) Abcam Cat#ab8898; RRID:AB_306848 Rabbit anti-IgG control (clone DA1E) (CUT&RUN) Cell Signaling Technology Cat#66362; RRID:AB_2924329 Rabbit anti-histone H3 total Abcam Cat#1791; RRID:AB_302613 Rabbit anti-H3K27me3 (clone C36B11) (immunofluorescence) Cell Signaling Technology CST#9733; RRID:AB_2616029 APC conjugated mouse anti-human CD1d (FACS) BioLegend Cat#350308; RRID:AB_10642829 APC conjugated mouse anti-human CD1d (clone CD1d42) BD Biosciences BD#563505; RRID:AB_2738246 APC-conjugated isotype control BD Biosciences BD#555751 Rat inVivoMab anti-mouse CD1d (clone 20H2) (FACS for Myc- CaP cells) Rat InVivoPlus anti-mouse isotype control (clone HRPN) (FACS for Myc-CaP cells) Bio X Cell #BE0179; RRID:AB_10949293 Bio X Cell #BE0088; RRID:AB_1107775 APC conjugated goat anti-rat IgG (H+L) Thermo Fisher Scientific Cat#A10540 Rat anti-mouse CD16/CD32 blocking reagent (Clone: 2.4G2) BD Biosciences Cat#553142; RRID:AB_394657 BV510-viability dye APC-α-GalCer-mCD1d Tetramer BV711-conjugated anti-mouse CD69 (clone: H1.2F3) BD Biosciences BD#564406; RRID:AB_2869572 TetramerShop BioLegend Cat#MCD1d–001 Cat#104537; RRID:AB_2566120 PerCP-Cy5.5-conjugated anti-mouse TCRβ (clone: H57–597) Biolegend Cat#109228; RRID:AB_1575173 BV605-conjugated anti-mouse CD3e (clone: 145–2C11) BUV395-conjugated anti-mouse NK1.1 (clone: PK136) BV711- conjugated anti-mouse CD8a (clone: 53–6.7) BV650- conjugated anti-mouse CD4 (clone: RM4–5) FITC- conjugated anti-mouse CD44 (clone: IM7) PE-Cy7- conjugated anti-mouse PD-1 (clone: RMP1–30) BV421-conjugated anti-mouse TIM3 (clone: 5D12) APC- conjugated anti-mouse TIGIT (clone: 4D4/mTIGIT) BV785- conjugated anti-mouse LAG3 (clone:C9B7W) PE- conjugated anti-mouse TNFα (clone: MP6-XT22) BV650- conjugated anti-mouse CD4 (clone: RM4–5) BV605-conjugated anti-mouse IFNγ (clone: XMG1.2) Biological samples BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend BioLegend Cat#100351; RRID:AB_2565842 Cat#564144 Cat#100759; RRID:AB_2563510 Cat#100546; RRID:AB_2562098 Cat#103006; RRID:AB_312957 Cat#109110; RRID:AB_572017 Cat#747626 Cat#156106; RRID:AB_2750515 Cat#125219; RRID:AB_2566571 Cat#506306; RRID:AB_315427 Cat#100546; RRID:AB_2562098 Cat#505840; RRID:AB_2734493 Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 45 REAGENT or RESOURCE Healthy donors for blood samples Blood samples from patients with a diagnosis of localized of metastatic prostate cancer SOURCE This paper This paper Localized tumor tissue cohort (core biopsies or surgical resection) This paper Chemicals, peptides, and recombinant proteins 5-azacitidine GSK126 MNase enzyme (micrococcal nuclease) G418 Blasticidin Cell Stimulation Cocktail Protein Transport Inhibitor Cocktail Dynabeads MyOne Streptavidin T1 Critical commercial assays EZ DNA methylation kit Zero blunt PCR cloning kit Magnetic MyOne Carboxylic Acid Beads NEBNext Ultra II DNA Library Prep Kit CUT&RUN Assay kit RNeasy Mini kit SuperScript III One-Step qRT-PCR kit NEBuilder HiFi DNA Assembly Cloning kit IDENTIFIER N/A N/A N/A Cat#S1782 Cat#S7061 Cat# M0247S Cat#G8168 Cat#ant–bl–05 Cat#00–4970–93 Cat#00–4980 Cat#65–601 Cat#D5001 Cat#K270020 Cat#65011 Cat#E7645L Cat#74104 Cat#11732020 Cat#E5520S Cat#554714 Cat#67563 Cat#SQK–RBK004 Selleck Selleckchem NEB Sigma Aldrich InvivoGen eBioscience eBioscience Invitrogen Zymo ThermoFisher Invitrogen NEB QIAGEN Invitrogen NEB QIAGEN Nanopore Cell Signaling Technology Cat#86652S BD Fixation/Permeabilization Solution Kit BD Biosciences HMW DNA extraction kit Rapid Barcoding Kit Deposited data Raw and analyzed data Human reference genome NCBI build 37, GRCh37 (hg19) Illumina Infinium Human Methylation 450 K BeadChip DNA Methylation 450 K BeadChip datasets This paper GEO: GSE208449 Genome Reference Consortium https://www.ncbi.nlm.nih.gov/ assembly/GCF_000001405.13/ National Cancer Institute’s GDC Data Portal National Cancer Institute’s GDC Data Portal https://portal.gdc.cancer.gov https://portal.gdc.cancer.gov TCGA (PRAD samples) CBioPortal https://www.cbioportal.org/ Methylation profiles (TCGA cohorts, 33 cancer types) TCGA Research Network https://portal.gdc.cancer.gov Genome annotations (TSS, exon, intron, intragenic regions, CpG islands (CGIs), repetitive elements and UCSC gap regions) - UCSC genome table browser Karolchik et al, 2004 https://genome.ucsc.edu/cgi-bin/ hgTables DNA methylation datasets (colon and thyroid) Timp et al, 2014 GEO: GSE53051 DNA methylation of normal prostate tissues and primary prostate tumors Yu et al., 2013 Obtained from authors. https://doi.org/ 10.1016/j.ajpath.2013.08.018 DNA methylation of metastatic prostate tumors Zhao et al., 2020 dbGAP: phs001648 Experimental models: Cell lines Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 46 REAGENT or RESOURCE SOURCE IDENTIFIER Human prostate cancer cell line (LNCaP, clone FGC) Human prostate cancer cell line (VCaP) Human prostate cancer cell line (PC3) Human prostate cancer cell line (22Rv1 ) Murine prostate cancer line (Myc-CaP) Normal cultured prostate epithelial cells (HPrEC) Benign prostatic hypertrophy cells (BPH-1) Murine Lewis lung carcinoma cells (LLC-1) Experimental models: Organisms/strains ATCC ATCC ATCC ATCC ATCC ATCC Sigma-Aldrich ATCC CRL–1740 CRL–2876 CRL–1435 CRL–2505 CRL–3255 PCS–440–010 SCC256 CRL–1642 Mouse: FVB mice Jackson Laboratory Strain#001800 Mouse: NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ Jackson Laboratory Strain#005557 Mouse: C57BL/6 Oligonucleotides Primers for qRT-PCR Primers for Bisulfite PCR Recombinant DNA pLenti-murine Cd1d1-mGFP pLenti-C-mGFP pLenti-Ifi204-Myc-DDK-Puro pLenti-C-Myc-DDK-Puro lentiCRISPRv2-blast N174-MCS pMD2.G psPAX2 Software and algorithms QUMA ImageJ FlowJo software (v10.4) Trim Galore (v0.4.3) Tophat (v2.1.1) Samtools (v1.3.1) HTseq (v0.6.1) Cufflinks (v2.1.1) R (v3.1.2) Graph Prism 9 Jackson Laboratory Strain#000664 This paper This paper Origene Origene Origene Origene Addgene Addgene Addgene Addgene Table S3 Table S3 Cat#MR226027L4 Cat#PS100093 Cat#MR222527L3 Cat#PS100092 Cat#98293; RRID:Addgene_98293 Cat#81061; RRID:Addgene_81061 Cat#12259; RRID:Addgene_12259 Cat#12260; RRID:Addgene_12260 Kumaki et al., 2008 http://quma.cdb.riken.jp/ https://imagej.nih.gov/ij/ BD Bioscience https://www.flowjo.com/ Babraham Bioinformatics https://github.com/FelixKrueger/ TrimGalore Trapnell et al., 2012 https://github.com/infphilo/tophat Li et al., 2009 http://samtools.sourceforge.net/ Anders et al., 2015 https://htseq.readthedocs.io/en/master/ Trapnell et al., 2012 https://github.com/cole-trapnell-lab/ cufflinks R Core Team, 2021 https://www.R-project.org/ GraphPad https://www.graphpad.com/ Bismark tool (v0.17.0) Krueger and Andrews, 2011 UCSC lift-over tool Hinrichs et al., 2006 https://github.com/FelixKrueger/ Bismark https://genome.ucsc.edu/cgi-bin/ hgLiftOver Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t Guo et al. Page 47 REAGENT or RESOURCE ABSOLUTE algorithm Molecular Signatures Database (MSigDB) (v7.2) SOURCE IDENTIFIER Carter et al., 2012 Broad Institute & UC San Diego Subramanian, Tamayo et al., 2005 Liberzon et al., 2011 http://software.broadinstitute.org/ cancer/cga/absolute_download https://www.gsea-msigdb.org/gsea/ msigdb/index.jsp Bioconductor package regioneR (v1.18.1) with overlapPermTest function Gel et al., 2016 https://www.bioconductor.org/ packages/release/bioc/html/ regioneR.html Ginkgo InferCNV (V 1.10.1) BWA men Sambamba MACS2 (v2.0.10) DeepTools phyper R function ONT Albacore software (v2.3.1) Nanopolish software (v0.10.2) PRROC R-package BioRender Other Lipofectamine 2000 reagent LentiX concentrator Polybrene Nanopore MinION device with R9.4 flowcell HRP conjugated secondary antibodies Laemmli buffer Garvin et al., 2015 http://qb.cshl.edu/ginkgo Tickle et al., 2019 https://github.com/broadinstitute/ infercnv Li and Durbin, 2009 https://github.com/lh3/bwa Tarasov et al., 2015 https://github.com/biod/sambamba Zhang et al., 2008 https://github.com/macs3-project/ MACS Ramirez et al., 2016 https://github.com/deeptools/deepTools Johnson et al., 1992 https://www.R-project.org/ Oxford Nanopore Technologies Simpson et al., 2017 Grau et al., 2015 https://nanoporetech.com/community https://github.com/nanoporetech/ nanopolish https://cran.r-project.org/web/ packages/PRROC/index.html BioRender https://www.biorender.com/ Invitrogen Clontech Labs Santa Cruz Oxford Nanopore Technologies Bio-rad Sigma Cat#1668019 NC0448638 sc–134220 FLO–MIN106D Cat#5196–2504 S3401–10VL Cell. Author manuscript; available in PMC 2023 August 18. A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t A u t h o r M a n u s c r i p t
10.1016_j.jbc.2023.105073	RESEARCH ARTICLE Mitochondrial double-stranded RNA triggers induction of the antiviral DNA deaminase APOBEC3A and nuclear DNA damage , Rémi Buisson4,5 Received for publication, May 8, 2023, and in revised form, June 27, 2023 Published, Papers in Press, July 19, 2023, https://doi.org/10.1016/j.jbc.2023.105073 Chloe Wick1, Seyed Arad Moghadasi1, Jordan T. Becker1 Elodie Bournique4,5 From the 1Department of Biochemistry, Molecular Biology and Biophysics, University of Minnesota, Minneapolis, Minnesota, USA; 2Department of Biochemistry and Structural Biology, University of Texas Health San Antonio, San Antonio, Texas, USA; 3Department of Life and Environmental Sciences, University of Cagliari, Cittadella Universitaria di Monserrato, Monserrato, Cagliari, Italy; 4Department of Biological Chemistry, School of Medicine, and 5Center for Epigenetics and Metabolism, Chao Family Comprehensive Cancer Center, University of California Irvine, Irvine, California, USA; 6Howard Hughes Medical Institute, University of Texas Health San Antonio, San Antonio, Texas, USA , and Reuben S. Harris1,2,6,* , Elisa Fanunza2,3 , Sunwoo Oh4,5 , Reviewed by members of the JBC Editorial Board. Edited by Karin Musier-Forsyth APOBEC3A is an antiviral DNA deaminase often induced by virus infection. APOBEC3A is also a source of cancer mutation in viral and nonviral tumor types. It is therefore critical to identify factors responsible for APOBEC3A upregulation. Here, we test the hypothesis that leaked mitochondrial (mt) double- stranded (ds)RNA is recognized as foreign nucleic acid, which triggers innate immune signaling, APOBEC3A up- regulation, and DNA damage. Knockdown of an enzyme responsible for degrading mtdsRNA, the exoribonuclease polynucleotide phosphorylase, results in mtdsRNA leakage into the cytosol and induction of APOBEC3A expression. APO- BEC3A upregulation by cytoplasmic mtdsRNA requires RIG-I, MAVS, and STAT2 and is likely part of a broader type I interferon response. Importantly, although mtdsRNA-induced APOBEC3A appears cytoplasmic by subcellular fractionation its induction triggers an overt DNA damage experiments, response characterized by elevated nuclear γ-H2AX staining. Thus, mtdsRNA dysregulation may induce APOBEC3A and contribute to observed genomic instability and mutation sig- natures in cancer. The apolipoprotein B mRNA editing catalytic polypeptide- like 3 (APOBEC3 or A3) family of proteins comprises seven members in humans (1). As single-stranded (ss)DNA cytosine deaminases, these enzymes normally function as antiviral factors capable of inhibiting virus replication, suppressing infectivity, and blocking pathogenesis (2). However, this potent DNA editing activity can also be directed at the human genome in cancer and cause mutations in chromosomal DNA (3–5). A3-catalyzed genomic C-to-U deamination events become immortalized as C-to-T transition and C-to-G trans- version mutations, most frequently in TCA and TCT trinu- cleotide motifs. Collectively, these single base substitution (SBS) mutation patterns in cancer are known as SBS2 and * For correspondence: Reuben S. Harris, [email protected]. SBS13 or, more simply, as the “APOBEC mutation signature.” The APOBEC mutation signature is found in over 70% of cancers and can be the largest fraction of somatic variation in many individual tumors and tumor types (6, 7). APOBEC3A (A3A) and APOBEC3B (A3B) are the most likely sources of APOBEC signature mutations in cancer (most recently addressed by (8, 9)). Both enzymes are potent ssDNA cytosine deaminases that intrinsically prefer TC motifs due to identical loop regions that engage the thymine nucleobase immediately upstream of a target cytosine (10, 11). Ectopic expression of both enzymes inﬂicts APOBEC signature mu- tations in model bacteria and yeast systems, the chicken cell line HAP1 (3, 9, 12–16). line DT40, and the human cell Recently, CRISPR knockout studies have shown that both enzymes contribute to ongoing mutagenesis in human cancer cell lines, with A3A accounting for a larger fraction of the overall APOBEC signature (8). Importantly, each of these human enzymes is capable of catalyzing mutagenesis and promoting tumor formation in mice, which demonstrates that this mutational process is capable of uniquely driving carci- nogenesis (and is not simply a passenger phenomenon despite the fact that most APOBEC signature mutations are likely to be aphenotypic) (17–21). A3A expression is suppressed in most normal human tissues (22–24). However, consistent with its function as an antiviral innate immune factor, its transcription can be induced by viral infection (25–27). For instance, human papillomavirus infec- tion of normal immortalized keratinocytes or human tonsillar epithelial cells, human polyomavirus infection of human uro- thelium, and human cytomegalovirus infection of decidual tissues are all reported to trigger increased expression of A3A (26, 28–30). Furthermore, consistent with antiviral function, A3A is induced by type I interferons (IFNs) in multiple cell types including monocytes, macrophages, and dendritic cells (22, 31–34). This pathway is initiated by IFN binding to its cell surface receptor, JAK/STAT signal transduction, and STAT2 binding the A3A promoter and transcriptional activation (25). J. Biol. Chem. (2023) 299(9) 105073 1 © 2023 THE AUTHORS. Published by Elsevier Inc on behalf of American Society for Biochemistry and Molecular Biology. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). APOBEC3A upregulation by mitochondrial dsRNA However, it is important to note that infection by other viruses such as the lentivirus HIV-1 and the herpesvirus Epstein–Barr virus fails to induce A3A expression (35, 36). Moreover, most cancer types with an APOBEC signature SBS2 and SBS13 and A3A expression lack viral etiologies (7, 23, 24, 37–39). It is to understand nonviral therefore of considerable interest mechanisms of A3A upregulation. Extrinsic nucleic acids from dead cells and intrinsic nucleic acids from chromosome missegregation (micronuclei) and aberrant endogenous virus and transposon activity can, like viral nucleic acids, activate nucleic acid sensors and trigger strong IFN responses including A3A upregulation (31, 40–42). Mito- chondria are another potential source of endogenous immu- nostimulatory nucleic acids (43–45). For instance, bidirectional transcription of mitochondrial genes can result in double- stranded (ds)RNA, which is normally recycled by a degrado- some comprising the exoribonuclease polynucleotide phos- phorylase (PNPase) and the ATP-dependent RNA helicase SUPV3L1 (46–50). Knockdown of either component of this complex results in accumulation of mitochondrial dsRNA (mtdsRNA) (45, 51). Moreover, PNPase depletion additionally allows mtdsRNA to escape into the cytosol (45, 52). Cytosolic mtdsRNA is then free to engage the RNA sensors RIG-I and MDA5 and potentiate an IFN response (45). Therefore, a combination of genetic, biochemistry, and cell biology ap- proaches is used here to test the hypothesis that mtdsRNA can be mistaken as foreign and trigger a virus-like innate immune response that leads to A3A induction and nuclear DNA damage. Results Mitochondrial and nuclear dsRNA trigger A3A upregulation To test the hypothesis that mtdsRNA leads to an induction of A3A expression, the breast epithelial cell line MCF10A was transfected with siRNAs to deplete the mitochondrial exori- bonuclease PNPase and the RNA helicase SUPV3L1 and immunoﬂuorescent microscopy was used to quantify dsRNA. Strong cytoplasmic staining with the dsRNA-speciﬁc mono- clonal antibody J2 was observed in PNPase- and SUPV3L1- depleted cells after membrane permeabilization with 0.2% triton-X100 (45) (Fig. 1A). A stringent 0.2% digitonin per- meabilization protocol yielded similar results (Fig. S1A). The majority of the dsRNA signal in these conditions appeared coincident with mitochondria as indicated by overlapping staining with MitoTracker (Red CMXRos). Interestingly, a milder 0.02% digitonin protocol, which permeabilizes only the plasma membrane (and not mitochondrial or nuclear mem- branes (53)), indicated that only PNPase depletion selectively triggers cytosolic dsRNA accumulation (Fig. 1B; additional images in Fig. S1B). In comparison, when using the same 0.02% digitonin treatment to preferentially permeabilize the cytoplasmic membrane, SUPV3L1 depletion did not lead to signiﬁcant mtdsRNA leakage into the cytosol (Fig. 1B; addi- images in Fig. S1B). As a negative control, non- tional digitonin-permeabilized cells staining (images in Fig. S1C). Quantiﬁcation of imaging results from conﬁrmed the 0.2% and 0.02% digitonin experiments showed little J2 2 J. Biol. Chem. (2023) 299(9) 105073 signiﬁcant overlap between dsRNA and MitoTracker staining (Fig. S1D) and signiﬁcant numbers of dsRNA foci accumu- lating in PNPase-depleted cells (Fig. S1E). To assess if knockdown of PNPase and subsequent release of dsRNA into the cytosol causes an IFN response in MCF10A cells, the expression of the interferon-stimulated gene ISG15 was measured as an indicator of a type I IFN production. PNPase knockdown, but not SUPV3L1 knock- down, resulted in strong upregulation of both ISG15 and A3A (Fig. 1, C and D). The two isoforms of A3A beginning at Met1 and Met13 are both evident, consistent with a transcriptional induction mechanism. Indeed, A3A mRNA levels increased 15- to 20-fold through PNPase knockdown in comparison with a nontargeting siRNA (Fig. 1D). A3B mRNA levels were also induced signiﬁcantly, but other A3 mRNAs appeared un- changed (Fig. 1D; quantiﬁcation of all A3 mRNAs in Fig. S2A). Similar results for A3A and A3B were obtained in the lung carcinoma epithelial cell line A549 but not in HeLa cells, which are defective in interferon synthesis (Fig. S2, B and C). Taken together, these data indicated that leakage of mito- chondrial dsRNA into the cytosol leads to a strong upregula- tion of A3A and a weaker but still signiﬁcant induction of A3B. To determine if dsRNA of a nonmitochondrial origin might also lead to A3A induction, the RNA regulatory protein TAR DNA-binding protein 43 (TDP-43) was knocked down, which is known to result in cytoplasmic RNA polymerase III tran- script accumulation (54, 55). Thus, TDP-43 was depleted from MCF10A cells and, as anticipated from this prior literature, this knockdown caused an accumulation of dsRNA puncta in the cytoplasm (Fig. 1, A and B). Importantly, this dsRNA signal showed little overlap with mitochondrial staining by Mito- Tracker Red CMXRos. However, similar to depletion of PNPase above, immunoblot and reverse transcription-quanti- tative PCR (RT-qPCR) experiments showed a >10-fold in- crease in A3A levels following TDP-43 depletion (Fig. 1, C and D). It is not clear why TDP-43 depletion results in higher A3A protein levels in comparison with PNPase depletion, despite similar fold-induction at the mRNA level and similarly high IFN responses as assessed by ISG15 levels. Nevertheless, despite this additional protein-level curiosity, these results combined to demonstrate that an accumulation of cytosolic dsRNA from mitochondrial or nuclear origins leads to a robust induction of A3A expression. Cytosolic sensing of mitochondrial dsRNA requires the RNA sensor RIG-I To determine the RNA sensor responsible for A3A upre- gulation in response to mtdsRNA accumulation in the cyto- plasm, MCF10A cells were codepleted of PNPase and candidate RNA sensors and then A3A levels were quantiﬁed as above. In comparison with the induction of A3A observed in cells depleted for PNPase, codepletion of PNPase and the cytosolic RNA sensor RIG-I prevented A3A upregulation. In contrast, treatment with siRNAs against MDA5 had no sig- niﬁcant effect (Fig. 2, A and B). To further substantiate these knockdown results, MCF10A cells engineered by CRISPR- A Triton B 0.02% Digitonin APOBEC3A upregulation by mitochondrial dsRNA C D Figure 1. Leaked mitochondrial dsRNA triggers A3A upregulation. A and B, immunoﬂuorescence microscopy images of MCF10A cells treated with siCtrl, siPNPase, siSUPV3L1, or siTDP-43 for 72 h and permeabilized with (A) 0.2% Triton X-100 or (B) 0.02% digitonin after which they were stained with the dsRNA- binding antibody J2. Mitochondria were stained with MitoTracker, and nuclei were stained with Hoechst (the scale bar represents 10 μm). C, immunoblot analysis of the indicated proteins expressed in MCF10A cells treated with siCtrl, siPNPase, siSUPV3L1, or siTDP-43 for 72 h. Tubulin was used as a loading control. All subpanels are from the same representative blot. D, Reverse transcription-quantitative PCR analysis of A3 mRNA levels in MCF10A cells after treatment with siCtrl, siPNPase, siSUPV3L1, or siTDP-43 for 72 h. Expression refers to A3 mRNA fold change relative to the negative control (set to 1) normalized to TBP. Mean values ± SEM of three independent experiments (p ≤ 0.05, p ≤ 0.01, *p ≤ 0.001 by Student’s t test and not shown if insigniﬁcant). Cas9 to lack RIG-I also demonstrated that this sensor is required for A3A induction by cytoplasmic dsRNA (Fig. 2, C and D). As anticipated from prior work (55), RIG-I null MCF10A cells also failed to induce A3A following TDP-43 depletion (Fig. 2E). MAVS is an adaptor in RNA sensing that typically functions downstream of RIG-I (56, 57). To further test whether A3A induction is dependent on the sensing of dsRNA, RNAi ex- periments were done to investigate the involvement of MAVS. As above for RIG-I experiments, codepletion of PNPase and MAVS reduced A3A expression to uninduced levels and MAVS-null clones showed a complete abrogation of A3A induction following knockdown of PNPase (Fig. 2, A, B, F and G). These results combined to further demonstrate that A3A is J. Biol. Chem. (2023) 299(9) 105073 3 APOBEC3A upregulation by mitochondrial dsRNA A C F B D E G Figure 2. The RIG-I/MAVS axis is required for upregulating A3A in response to endogenous mitochondrial dsRNA. A, RT-qPCR analysis of A3A after treatment with siCtrl, siPNPase, and codepletions of siPNPase with siCtrl, siRIG-I, siMDA5, and siMAVS for 72 h in MCF10A cells. Expression refers to mRNA fold change relative to the negative control (which was set to 1) and was normalized to TBP. Mean values ± SEM of three independent experiments (p ≤ 0.05 by Student’s t test and not shown if insigniﬁcant). B, immunoblot analysis of A3A in MCF10A cells treated with siCtrl, siPNPase, and codepletions of siPNPase with siCtrl, siRIG-I, siMDA5, and siMAVS. Tubulin was used as a loading control. C and D, RT-qPCR and immunoblot analysis of A3A mRNA and protein levels, respectively, in control or RIG-I KO MCF10A cells following siCtrl or siPNPase treatment. Expression refers to mRNA fold change relative to the negative control (which was set to 1) and was normalized to TBP. Mean values ± SEM of three independent experiments (*p ≤ 0.001 by Student’s t test and not shown if insigniﬁcant). E, RT-qPCR analysis of A3A mRNA levels, respectively, in control or RIG-I KO MCF10A cells following siCtrl or siTDP-43 treatment. Expression refers to mRNA fold change relative to the negative control (which was set to 1) and was normalized to TBP. Mean values ± SEM of three 4 J. Biol. Chem. (2023) 299(9) 105073 upregulated through the sensing of cytosolic mtdsRNA by the RIG-I/MAVS pathway. A3A upregulation by cytosolic mtdsRNA requires STAT2 Accumulation of immunostimulatory dsRNA can trigger a wide array of cellular responses, and therefore a set of IFN- responsive genes was analyzed to determine whether the ca- nonical IFN pathway is involved. We found that ﬁve canonical IFN-stimulated genes (ISG15, IFI44, DDX60, MX1, and OAS1 (58)) were all signiﬁcantly upregulated following siPNPase treatment (Fig. 3A). Genes encoding the inﬂammatory cyto- kines tumor necrosis factor alpha and interleukin 6 were also induced by PNPase knockdown (Fig. S4). The expression of speciﬁc IFN genes was not examined, but several are also bona ﬁde ISGs and induction is anticipated based on prior reports (e.g., IFN-β in ref. (45)). To conﬁrm that A3A is upregulated through a type I IFN response, knockdown of the IFN-α/β receptor in siPNPase-treated cells effectively reduced A3A expression levels to those of the control (Fig. 3, B and C). IFNAR1 depletion was conﬁrmed by RT-qPCR in these ex- periments because available commercial antibodies did not work in our hands (Fig. 3D). These results indicated that, following activation of RIG-I and MAVS, A3A induction oc- curs through a type-I IFN-dependent signaling pathway. The most IFN-dependent likely mediators of a type-I response based on the prior literature (25, 59, 60) are the IFN- inducible transcription factors STAT1 and STAT2. These factors were therefore depleted from MCF10A cells with siRNA, and the effect of PNPase knockdown was examined as described above. Interestingly, only STAT2 (and not STAT1) depletion was able to block A3A induction by PNPase knockdown (Fig. 3, B and C; independent STAT1 knockdown results in Fig. S3). This important result was conﬁrmed using STAT2-knockout MCF10A clones, where A3A is no longer inducible by PNPase knockdown (Fig. 3, E and F). We also extended these results to another cell line using a completely orthologous approach. A549 cells were transfected with vectors expressing the Zika virus proteins NS2A and NS4B as tools to block the JAK-STAT signaling cascade that occurs following IFN induction. NS2A mediates the degrada- tion of STAT1 and STAT2, and NS4B suppresses the phos- phorylation of STAT1 (61, 62). A549 cells were transfected concurrently with siPNPase and plasmids encoding FLAG- tagged NS2A and NS4B (Fig. 3G). A3A upregulation was eliminated by the addition of NS2A. In contrast, transfection of NS4B into the cells had little effect with A3A levels still rising 15- to 20-fold after PNPase knockdown. As NS2A (but not NS4B) interferes with STAT2 activation, these data sup- port the knockdown and knockout results above showing that A3A induction requires STAT2. Thus, activation of RIG-I/ MAVS by endogenous dsRNA causes a type I IFN response that induces A3A via STAT2. APOBEC3A upregulation by mitochondrial dsRNA A3A induction by mtdsRNA triggers a DNA damage response To investigate the kinetics of A3A induction by mtdsRNA leakage, A3A, A3B, and PNPase mRNA expression levels were analyzed every 24 h over a 4-day period following PNPase depletion (Fig. 4A). This analysis revealed that A3A expression peaks, approximately 15-fold, at around 72 h after siRNA transfection and quickly recovers to 2-fold induction by 96 h. A3B mRNA levels peak with similar kinetics, although only around 3-fold, roughly plateauing between 48 and 72 h post transfection, and A3B mRNA levels may also persist slightly longer. In the same time course, PNPase mRNA levels are depleted maximally by 72 h post transfection and begin to recover by 96 h (Fig. 4A). These results indicate that A3A (and A3B) mRNA levels correlate inversely with PNPase levels (and thereby also with cytosolic mtdsRNA levels) and are likely to be transient in nature. To determine where mtdsRNA-induced A3A protein ac- cumulates within cells, PNPase was depleted from MCF10A, subcellular fractionation was used to separate nuclear and cytoplasmic components, and immunoblots were done to detect relevant proteins. Phorbol 12-myristate 13-acetate was used as a positive control to induce A3A and A3B, as shown previously (63–65). This biochemical approach showed that the majority of mtdsRNA-inducible A3A is localized to the cytoplasm, with tubulin as a positive control (Fig. 4B). In comparison, the majority of A3B localizes to nuclear fractions, with histone H3 as a positive control (Fig. 4B). Cytosolic localization of IFNα-induced endogenous A3A has been re- ported for another cell line (THP1), and nuclear localization of endogenous A3B has been reported for MCF10A and a multitude of cell lines by many groups (63, 66–68). as above Last, we asked whether the A3A protein induced under these conditions of PNPase depletion/cytosolic mtdsRNA accumulation is capable of inﬂicting nuclear DNA damage. from This was done by depleting PNPase MCF10A cells and then using immunoﬂuorescence micro- scopy to visualize and quantify the DNA damage marker γ- H2AX. Interestingly, PNPase depletion causes strong increases in both pan-nuclear and focused γ-H2AX staining including a doubling of the number of γ-H2AX foci (Fig. 4, C and D; quantiﬁcation in Fig. S5). An independent experiment with doxorubicin as a positive control conﬁrmed this result and suggested that the overall level of DNA damage inﬂicted by A3A is less than that caused by this chemotherapeutic (Fig. S5). Importantly, MCF10A cells engineered by CRISPR to lack endogenous A3A demonstrated that the majority of these nuclear γ-H2AX foci are dependent upon this enzyme (Fig. 4, C and D), despite the majority of protein localizing to the cytosol as described above. A3A knockout was conﬁrmed by immunoblot and by sequencing the gRNA-binding site where each allele has multiple mutations including a frameshift mutation (Fig. 4, E and F). These experiments combined to independent experiments (p ≤ 0.001 by Student’s t test and not shown if insigniﬁcant). F and G, RT-qPCR and immunoblot analysis of A3A mRNA and protein levels, respectively, in control or MAVS KO MCF10A cells following siCtrl or siPNPase treatment. Expression refers to mRNA fold change relative to the negative control (which was set to 1) and was normalized to TBP. Mean values ± SEM of three independent experiments (p ≤ 0.01 by Student’s t test and not shown if insigniﬁcant). RT-qPCR, reverse transcription-quantitative PCR. J. Biol. Chem. (2023) 299(9) 105073 5 APOBEC3A upregulation by mitochondrial dsRNA A C F B D E G Figure 3. A3A is upregulated via a STAT2-dependent interferon response. A, RT-qPCR analysis of a panel of interferon-responsive genes (ISG15, IFI44, DDX60, MX1, OAS1) after siPNPase treatment of MCF10A cells. Expression refers to mRNA log2 fold change relative to the negative control (which was set to 0) and was normalized to TBP. Mean values ± SEM of three independent experiments (p ≤ 0.05, *p ≤ 0.01 by Student’s t test and not shown if insigniﬁcant). B, RT-qPCR analysis of A3A in MCF10A cells treated with siCtrl, siPNPase, and siPNPase in combination with siCtrl, siIFNAR1, siSTAT1, and siSTAT2 (p ≤ 0.05 by Student’s t test and not shown if insigniﬁcant). C, immunoblot analysis of MCF10A cells treated with siCtrl, siPNPase, and codepletions of siPNPase in combination with siIFNAR1, siSTAT1, and siSTAT2. D, RT-qPCR analysis of IFNAR1 in MCF10A cells treated with siCtrl or siIFNAR1 and siPNPase. Mean values ± SEM of three independent experiments (p ≤ 0.01 by Student’s t test). E and F, RT-qPCR and immunoblot analysis of A3A mRNA and protein levels, respectively, in control or STAT2 KO MCF10A cells following siCtrl or siPNPase treatment. Expression refers to mRNA fold change relative to the negative 6 J. Biol. Chem. (2023) 299(9) 105073 indicate that cytosolic mtdsRNA accumulation leads to a strong A3A-dependent DNA damage response. Discussion that Here, we report the cytoplasmic accumulation of endogenous dsRNA of mitochondrial origin triggers a strong increase in the expression of A3A and a slight increase in the expression of A3B. While it has been previously reported that foreign and synthetic nucleic acids are able to trigger the in- duction of A3A through a type-I IFN response (22, 25, 31, 33, 69, 70), our results are the ﬁrst to examine how dysregulation of endogenous dsRNA may act as a natural source of immu- nostimulatory nucleic acids and lead to strong upregulation of A3A. We show that the upregulation of A3A by endogenous dsRNA is dependent on the RIG-I/MAVS signaling axis and proceeds through a type I IFN response in a STAT2- dependent manner. Moreover, upregulated A3A, although almost entirely cytoplasmic, is also able to cause chromosomal DNA damage as evidenced by elevated γ-H2AX staining. Taken together, these results support a model in which a breach in mitochondrial integrity can leak dsRNA into the triggers RIG-I/MAVS/STAT2-dependent cytosol, which upregulation of the IFN response including A3A expression and, importantly, DNA damage (Fig. 5). This pathway could be directly relevant to cells with mitochondrial dsRNA leakage as well as to bystander cells due to the auto/paracrine nature of the IFN response. These observations may help explain the periodic (also called episodic) occurrence of APOBEC3 signature mutations in cancer cell lines, which were shown recently to involve A3A (8, 71). The chromosomal DNA damage observed following knockdown of PNPase and accumulation of mtdsRNA is surprising given that the bulk of induced A3A protein is cytoplasmic. In fact, our subcellular fractionation experiments indicate no detectable A3A in the nucleus of PNPase-depleted cells. This observation is consistent with a prior report of endogenous A3A localization to the cytoplasm following IFN- α treatment of the cell line THP1 (67). However, given the strong genetic dependence of γ-H2AX accumulation here on A3A following PNPase knockdown and cytoplasmic mtdsRNA accumulation, we hypothesize that a low level of induced A3A is able to diffuse through nuclear pores (due to its small size), deaminate single-stranded regions of chromosomal DNA, and trigger DNA breaks as evidenced by elevated levels of nuclear γ-H2AX foci. In addition to the robust A3A upregulation observed upon knockdown of PNPase, A3B was also signiﬁcantly induced, although to a much lower extent. A3B has also been found to deaminate genomic DNA, and it is also a major source of mutations in cancer and is found at much higher levels in the nuclear compartment of a wide range of tumors and cancer lines (22–24, 63, 66). Although the majority of DNA cell APOBEC3A upregulation by mitochondrial dsRNA damage observed here following PNPase depletion is depen- dent on A3A, A3B is predominantly nuclear with direct access to chromosomal DNA and, thus, also able to contribute to the overall landscape of APOBEC signature mutations observed in cancer. Here, we propose that sporadic induction of A3A caused by mitochondrial stress and cytoplasmic dsRNA accumulation over the course of human lifetime may contribute to the overall burden of DNA damage and mutation accumulation in cancer. To investigate the volatility of A3A upregulation due to mitochondrial dysfunction, the kinetics of A3A induction following the depletion of PNPase were investigated. The in- duction of A3A is transient in nature with A3A peaking at 72 h after knockdown of PNPase and returning from a 15-fold to a 2-fold induction after an additional 24 h. Thus, the transient induction of A3A by the dysregulation of endogenous nucleic acids could be responsible for some of the proposed episodic bursts of A3A mutagenesis observed in cancer (8, 71). In the context of both A3A and A3B, episodic mutagenesis by A3A may cause “mutational ﬂares” and A3B may contribute to a continuous “mutational smolder,” which together account for the overall landscape of APOBEC signature mutations in cancer. Because of its potent deaminase activity and capacity to damage the genome, A3A is tightly regulated and only induced in response to infection, inﬂammation, and other stresses to the cell including mitochondrial dysfunction as shown here. Thus, the transient upregulation of A3A during these condi- tions could lead to nuclear DNA damage and mutation accumulation. However, A3B, which is nuclear and often expressed at much higher levels in tumors, may result in a continuous but slower accumulation of APOBEC3 signature mutations to the nuclear genome over time. Thus, A3A and A3B can together explain the bulk of the overall APOBEC mutation signature and the mechanism described here through endogenous dsRNA may be particu- larly relevant to tumor types with nonviral, nonchronic, or otherwise unclear etiologies. cancer across Experimental procedures Cell culture MCF10A cells were cultured in Dulbecco’s modiﬁed Eagle’s medium/F12 (Thermo Fisher Scientiﬁc #11320033) supple- mented with 5% horse serum (Sigma-Aldrich #H1270), 20 ng/ ml EGF (Peprotech #AF-100-15), 0.5 μg/ml hydrocortisone (Sigma #H0888), 100 ng/ml cholera toxin (Sigma #C8052), 10 μg/ml insulin (Sigma #91077C), and 1% penicillin/strep- tomycin (Thermo Fisher Scientiﬁc #15140122). HeLa and A549 cells were cultured in Dulbecco’s modiﬁed Eagle’s medium (Thermo Fisher Scientiﬁc #SH30022FS) supple- mented with 10% fetal bovine serum (Life Technologies #1043702) and 1% penicillin/streptomycin (Thermo Fisher Scientiﬁc #15140122). Cells were maintained at 37 (cid:1)C and 5% control (which was set to 1) and was normalized to TBP. Mean values ± SEM of three independent experiments (p ≤ 0.01 by Student’s t test and not shown if insigniﬁcant). G, RT-qPCR analysis of A3A in A549 cells treated with siCtrl/siPNPase and transfected with plasmids encoding NS2A, NS4B, or GFP. Expression refers to mRNA fold change relative to the negative control (which was set to 1) and was normalized to TBP. Mean values ± SEM of two independent experiments (p ≤ 0.05, p ≤ 0.01 by Student’s t test and not shown if insigniﬁcant). RT-qPCR, reverse transcription-quantitative PCR. J. Biol. Chem. (2023) 299(9) 105073 7 APOBEC3A upregulation by mitochondrial dsRNA A B C E D F Figure 4. DNA damage induced by the upregulation of PNPase is A3A dependent. A, Reverse transcription-quantitative PCR analysis of A3A, A3B, and PNPase in MCF10A cells treated with siCtrl/siPNPase after 24, 48, 72, and 96 h (mean ± SEM of three independent experiments). B, immunoblot analysis of whole cell, cytoplasmic, and nuclear fractions of MCF10A cells treated with 10 nM siCtrl/siPNPase for 72 h or DMSO/25 ng/ml PMA for 24 h (C and D) representative immunoﬂuorescence microscopy images of γ-H2AX in control or A3A knockout cells treated with siCtrl/siPNPase with quantiﬁcation of the number of γ-H2AX foci per nucleus (the scale bar represents 10 μm; mean ± SEM of n > 50 cells per condition; p ≤ 0.05 by Student’s t test and not shown if insigniﬁcant). E, immunoblot of A3A in two control (lacZ) clones and in an A3A knockout clone following 24 h of stimulation with 25 ng/ml PMA. F, DNA sequence of the CRISPR-disrupted A3A alleles in MCF10A cells (5/10 sequenced plasmids had allele 1 and 5/10 allele 2). Frameshift-induced premature stop codons are highlighted in yellow, and insertions, deletions, and substitutions are shown in red. DMSO, dimethyl sulfoxide; PMA, phorbol 12-myristate 13- acetate. 8 J. Biol. Chem. (2023) 299(9) 105073 APOBEC3A upregulation by mitochondrial dsRNA Figure 5. Working model for A3A upregulation by endogenous dsRNA. Mitochondrial dsRNAs that accumulate inappropriately in the cytosol are sensed by RIG-I, which signals through the adaptor protein MAVS and leads to a type I interferon response, induction of A3A by STAT2, and chromosomal DNA damage. Three distinct panels are shown to illustrate the fact that interferon signaling can act in both cis and trans (autocrine and paracrine). CO2. MCF10A cells were purchased from Horizon, and A549 cells, HeLa cells, and 293T cell lines were obtained from the American Type Culture Collection (ATCC). RNA interference See Table S1 for all oligonucleotide sequences including siRNA sequences. Duplex siRNAs (IDTDNA) were resus- pended at 20 μM in nuclease-free duplex buffer (IDTDNA #11-01-03-01), and cells were treated at a ﬁnal concentration of 10 nM. siRNAs were reverse transfected using Lipofect- amine RNAiMAX (Thermo Fischer Scientiﬁc #13778150) in OptiMEM (Thermo Fischer Scientiﬁc #31985062). RNAiMAX was used at a ratio of 5 μl to 1 μl of 20 μM siRNA. To transfect plasmid DNA and siRNAs concurrently, TransIT-X2 (Mirus #MIR6000) was used to transfect the plasmid DNA and RNAiMAX was used to transfect the siRNA 24 h later. Transfections of siRNAs were completed in antibiotic-free medium for 72 h before harvesting. siRNA transfection efﬁ- ciency was assessed using TYE 563 (IDTDNA #51-01-20-19), and knockdown of desired proteins was evaluated via immu- noblot and RT-qPCR analysis. CRISPR knockout cells MCF10A cells engineered by CRISPR to lack MAVS and STAT2 were described recently (25, 72). See Table S1 for all oligonucleotide sequences including gRNAs. The construct encoding the gRNA was created by cloning the gRNA into the LentiCRISPR1000 (73) plasmid via Golden Gate cloning using the Esp3I sites. Virus was created using HEK-293T cells (ATCC) transfected with LentiCRISPR1000 plasmids encoding the gRNA, gag, and VSVG. The gRNA for the RIG-I knockouts was 50- GCGCCTGGACAATGGCACCT-30, and the gRNA the A3A knockouts was 50- GAAAAACAACAAGG for GCCCAA-3’. MCF10As were then transduced and selected with puromycin (1 μg/ml) (Gold Biotechnology #P-600-500) after 24 h. Surviving cells were single cell cloned in a 96-well plate and grown until 80% conﬂuent. Cells were maintained in 1 μg/ml puromycin for all subsequent passages. Knockout of the target gene was veriﬁed by immunoblot and pJet sequencing (Thermo Fisher Scientiﬁc #K1231) of the target region. After harvesting genomic DNA from the cells, primers were used to amplify 200 bp surrounding the target sequence on each end (50-GATGCTCGGTGTGGTAGGAG-30 and 50-CCCTGAGTCCTCAGATCCCA-30 for A3A), which was then cloned into a pJet vector using the CloneJet PCR Kit (Thermo Fisher Scientiﬁc #K1231). Ten different plasmids were conﬁrmed using Sanger DNA sequencing (GeneWiz) for each gene. Quantitative reverse-transcription PCR (Roche Life See Table S1 for all oligonucleotide sequences including PCR primers. Total RNA was extracted from cells using the High Pure RNA Isolation Kit Science the manufacturer’s #11828665001) per instructions. The total RNA was transcribed into cDNA in a 20-μl reaction using 50 μM random hexamer primers (50-NNNNNN-30) (IDTDNA), 1 mM dNTPs (Millipore Sigma #DNTP-RO), 20 U transcriptor Science transcriptase #3531317001), and 20 U protector RNase inhibitor (Roche Life Science #3335399001). Quantitative PCR was carried out on a LightCycler 480 II (Roche Life Science) in technical triplicate using SsoFast Eva Green Supermix (Bio-Rad #1725200). (Roche Life reverse Immunoﬂuorescence microscopy See Table S2 for information on all primary and secondary antibodies. Cells were ﬁxed in 4% formaldehyde (Thermo Fisher Scientiﬁc #28906) for 15 min and permeabilized using PBS containing 0.2% Triton X-100 (Sigma-Aldrich #T8787). J. Biol. Chem. (2023) 299(9) 105073 9 APOBEC3A upregulation by mitochondrial dsRNA Cells were blocked in immunoﬂuorescence microscopy blocking solution (2.8 μM KH2PO4, 7.2 μM K2HPO4, 5% goat serum, 5% glycerol, 1% gelatin from cold water ﬁsh, 0.04% sodium azide, pH 7.2) with 0.1% Triton X-100 for 1 h at room temperature. Cells were incubated overnight at 4 (cid:1)C in primary antibody, which was diluted in immunoﬂuorescence blocking buffer. Incubation of cells with ﬂuorophore-conjugated sec- ondary antibody diluted in immunoﬂuorescence blocking buffer was completed for 2 h at room temperature, and nuclei were stained with Hoechst 33342 (1 μg/ml) (Thermo Fisher Scientiﬁc #PI62249). Images were collected at 20× magniﬁca- tion (or 10× magniﬁcation for Fig. 1B) using a Cytation 5 Cell Imaging Multi-Mode Reader (BioTek) or an EVOS FL Cell Imaging System (Thermo Fisher Scientiﬁc). For the γ-H2AX images, a Cytation 5 Cell Imaging Multi-Mode Reader was used to image ﬁve slices that were 2 μM apart, which were then combined using a maximum intensity projection to create the ﬁnal image. Immunoﬂuorescence microscopy with differential permeabilization Immunoﬂuorescence microscopy with differential per- meabilization was conducted in a manner similar to the immunoﬂuorescence microscopy protocol listed above. How- ever, instead of permeabilizing the cells with PBS containing 0.2% Triton X-100, cells were permeabilized with 0.2% digi- tonin (Sigma-Aldrich #D141) to ensure permeabilization of all membranes, or 0.02% digitonin to permeabilize only the plasma membranes, or 0% digitonin to permeabilize none of the membranes. Blocking was completed in the same immu- noﬂuorescence microscopy blocking solution but without the added 0.1% Triton X-100. The rest of the immunoﬂuorescence microscopy is the same as the immunoﬂuorescence micro- scopy protocol for the permeabilization of all membranes with Triton X-100. Immunoblotting See Table S2 for information on all primary and secondary antibodies. Cells were lysed in 2.5× RSB (125 mM Tris HCl, 20% glycerol, 7.5% SDS, 5% β-mercaptoethanol, 250 mM DTT, 0.05% Orange G, pH 6.8) and boiled for 10 min. Lysates were run on a 4 to 20% gradient SDS-PAGE gel (Bio-Rad #3450033) and then transferred to a PVDF membrane (Millipore #IPFL00010). Membranes were blocked in 5% milk in PBS for 1 h at room temperature. Primary antibody was diluted in 5% milk in PBS and applied overnight at 4 (cid:1)C. Blots were then incubated in secondary antibody in 5% milk in PBS supplemented with 0.1% Tween 20 (Thermo Fisher Scientiﬁc #BP337-500) and 0.02% SDS (Thermo Fisher Scientiﬁc #419530010) for 1 h at room temperature. Blots using the antibody 5210-87-13 (66) for A3A and A3B, as well as anti- bodies against MAVS, STAT2, ISG15, and SUPV3L1, utilized an HRP-labeled anti-rabbit secondary antibody (Jackson ImmunoResearch #111-035-144), which was visualized using SuperSignal West Femto Maximum Sensitivity Substrate 10 J. Biol. Chem. (2023) 299(9) 105073 (Thermo Fisher Scientiﬁc #PI34095). Blots were imaged on the LI-COR Odyssey Fc imaging system (LI-COR Biosciences). Subcellular fractionation Approximately 106 cells were pelleted for 2 min at 3000 RPM and then washed in 150 μl cold PBS. Cells were pelleted again at 3000 RPM for 2 min. The supernatant was then decanted, and the pellet was resuspended in 90 μl ice-cold 0.1% IGEPAL CA-630 (Sigma-Aldrich #I8896). This resus- pension was the whole-cell fraction, and a portion (30 μl) of the resuspension was removed and treated with RSB. The remaining resuspension was pelleted at 3000 RPM for 5 min at 4 (cid:1)C. The supernatant was the cytosolic fraction, and a portion (30 μl) of the resuspension was removed and treated with RSB. The pellet was then washed in 80 μl ice-cold 0.1% IGEPAL CA-630 (Sigma-Aldrich #I8896) and spun again at 3000 RPM for 5 min. The supernatant was discarded, and the pellet was resuspended in 10 μl HED buffer (25 mM Hepes, 15 mM EDTA, 1 mM DTT, 10% glycerol, pH 7.4). This resuspension was the nuclear fraction and was subsequently treated with RSB. Chemicals and inhibitors Cells were treated with MitoTracker CMXRos (Thermo Fisher Scientiﬁc #M7512) for 30 min at a concentration of 500 nM in order to stain the mitochondria prior to ﬁxation. 3p-hpRNA/LyoVec (Invivogen #tlrl-hprnalv) was used at 1 μg/ ml for 16 h to stimulate and screen for RIG-I in the control and RIG-I knockout cells. PMA (Sigma-Aldrich #P1585), a known inducer of A3A and A3B (63–65), was used at 25 ng/ml over 24 h. Doxorubicin (Sigma-Aldrich #D1515) was used as a positive control for DNA damage response by treating cells for 24 h at a concentration of 1 μM. Data availability All relevant data are contained within the main article or supplemental information. Please email [email protected] with requests for raw data or reagents. Supporting information. information—This article contains supporting Acknowledgments—We thank Harris helpful feedback during these studies. laboratory members for Author contributions—C. W., S. A. M., J. T. B., R. S. H. conceptu- alization; S. A. M., J. T. B., E. F., S. O., E. B., R. B. methodology; C. W., S. A. M., J. T. B. formal analysis; C. W. investigation; E. F., S. O., E. B., R. B. resources; C. W., R. S. H. writing – original draft; C. W., S. A. M., J. T. B., E. F., S. O., E. B., R. B., R. S. H. writing – review & editing; S. A. M., J. T. B., R. S. H. supervision; R. B., R. S. H. funding acquisition. Funding and additional information—These studies were supported by NCI, National Institutes of Health P01 CA234228 (to R. S. H.), NIAID, National Institutes of Health R37 AI064046 (to R. S. H.), NCI, National Institutes of Health R37 CA252081 (to R. B.), and a Recruitment of Established Investigators Award from the Cancer Prevention and Research Institute of Texas (CPRIT RR220053 to R. S. H.). J. T. B. received partial salary support from the National Institute for Allergy and Infectious Diseases (F32-AI147813). C. W. received part-time support from the University of Minnesota Undergraduate Research Opportunities Program (UROP). C. W. is the Marvin and Christine Ballard Scholar, the Leon Snyder Scholar, and the Harold Paul Morris Memorial Scholarship holder. S. O. is a Dr Lorna Calin Scholar and was supported by the Faculty Mentor Program from the University of California, Irvine. R. S. H. is the Ewing Halsell President’s Council Distinguished Chair at University of Texas San Antonio and an Investigator of the Howard Hughes Medical Institute. The content is solely the responsibility of the authors and does not necessarily represent the ofﬁcial views of the National Institutes of Health. Conﬂict of interest—The authors declare that they have no conﬂicts of interest with the contents of this article. Abbreviations—The abbreviations used are: APOBEC3, apolipo- protein B mRNA editing catalytic polypeptide-like 3; IFN, inter- feron; polynucleotide IFN-stimulated phosphorylase; SBS, single base substitution. PNPase, gene; ISG, References 1. Salter, J. D., Bennett, R. P., and Smith, H. C. (2017) The APOBEC protein family: United by structure, divergent in function. Trends Biochem. Sci. 41, 578–594 2. Refsland, E. W., and Harris, R. S. (2013) The APOBEC3 family of retro- element restriction factors. Curr. Top. Microbiol. Immunol. 371, 1–27 3. Green, A. M., Landry, S., Budagyan, K., Avgousti, D. C., Shalhout, S., Bhagwat, A. S., et al. (2016) APOBEC3A damages the cellular genome during DNA replication. Cell Cycle 15, 998–1008 4. Hoopes, J. I., Cortez, L. M., Mertz, T. M., Malc, E. P., Mieczkowski, P. A., and Roberts, S. A. (2016) APOBEC3A and APOBEC3B preferentially deaminate the lagging strand template during DNA replication. Cell Rep. 14, 1273–1282 5. Cervantes-Gracia, K., Gramalla-Schmitz, A., Weischedel, J., and Chah- wan, R. (2021) APOBECs orchestrate genomic and epigenomic editing across health and disease. Trends Genet. 37, 1028–1043 6. Bergstrom, E. N., Luebeck, J., Petljak, M., Khandekar, A., Barnes, M., Zhang, T., et al. (2022) Mapping clustered mutations in cancer reveals APOBEC3 mutagenesis of ecDNA. Nature 602, 510–517 7. Alexandrov, L. B., Kim, J., Haradhvala, N. J., Huang, M. N., Tian Ng, A. W., Wu, Y., et al. (2020) The repertoire of mutational signatures in hu- man cancer. Nature 578, 94–101 8. Petljak, M., Dananberg, A., Chu, K., Bergstrom, E. N., Striepen, J., von Morgen, P., et al. (2022) Mechanisms of APOBEC3 mutagenesis in hu- man cancer cells. Nature 607, 799–807 9. [preprint] Jarvis, M., Carpenter, M. A., Temiz, N. A., Brown, M., Richards, K. A., Argyris, P. P., et al. (2022) Mutational impact of APO- BEC3B and APOBEC3A in a human cell line. bioRxiv. https://doi.org/10. 1101/2022.04.26.489523 10. Shi, K., Carpenter, M. A., Banerjee, S., Shaban, N. M., Kurahashi, K., Salamango, D. J., et al. (2017) Structural basis for targeted DNA cytosine deamination and mutagenesis by APOBEC3A and APOBEC3B. Nat. Struct. Mol. Biol. 24, 131–139 11. Kouno, T., Silvas, T. V., Hilbert, B. J., Shandilya, S. M. D., Bohn, M. F., Kelch, B. A., et al. (2017) Crystal structure of APOBEC3A bound to single-stranded DNA reveals structural basis for cytidine deamination and speciﬁcity. Nat. Commun. 8, 15024 12. Harris, R. S., Petersen-Mahrt, S. K., and Neuberger, M. S. (2002) RNA editing enzyme APOBEC1 and some of its homologs can act as DNA mutators. Mol. Cell. 10, 1247–1253 APOBEC3A upregulation by mitochondrial dsRNA 13. DeWeerd, R. A., Németh, E., Póti, Á., Petryk, N., Chen, C.-L., Hyrien, O., et al. (2022) Prospectively deﬁned patterns of APOBEC3A mutagenesis are prevalent in human cancers. Cell Rep. 38, 110555 14. Roberts, S. A., Sterling, J., Thompson, C., Harris, S., Mav, D., Shah, R., et al. (2012) Clustered mutations in yeast and in human cancers can arise from damaged long single-strand DNA regions. Mol. Cell. 46, 424–435 15. Carpenter, M. A., Li, M., Rathore, A., Lackey, L., Law, E. K., Land, A. M., et al. (2012) Methylcytosine and normal cytosine deamination by the J. Biol. Chem. 287, foreign DNA restriction enzyme APOBEC3A. 34801–34808 16. Langenbucher, A., Bowen, D., Sakhtemani, R., Bournique, E., Wise, J. F., Zou, L., et al. (2021) An extended APOBEC3A mutation signature in cancer. Nat. Commun. 12, 1602 17. Law, E. K., Levin-Klein, R., Jarvis, M. C., Kim, H., Argyris, P. P., Car- penter, M. A., et al. (2020) APOBEC3A catalyzes mutation and drives carcinogenesis in vivo. J. Exp. Med. 217, e20200261 18. Law, E. K., Sieuwerts, A. M., LaPara, K., Leonard, B., Starrett, G. J., Molan, A. M., et al. (2016) The DNA cytosine deaminase APOBEC3B promotes tamoxifen resistance in ER-positive breast cancer. Sci. Adv. 2, e1601737 19. [preprint] Caswell, D. R., Mayekar, M. K., Gui, P., Law, E. K., Vokes, N. I., Ruiz, C. M., et al. (2022) The role of APOBEC3B in lung tumour evo- lution and targeted therapy resistance. bioRxiv. https://doi.org/10.1101/ 2020.12.18.423280 20. Boumelha, J., de Carné Trécesson, S., Law, E. K., Romero-Clavijo, P., Coelho, M. A., Ng, K. W., et al. (2022) An immunogenic model of KRAS- mutant lung cancer enables evaluation of targeted therapy and immu- notherapy combinations. Cancer Res. 82, 3435–3448 21. [preprint] Durfee, C., Temiz, N. A., Argyris, P. P., Alsoe, L., Carracedo Huroz, S., Alonso de la Vega, A., et al. (2022) APOBEC3B-driven tumors in vivo manifest signature mutations, heterogeneity, and evidence for metastases. bioRxiv. https://doi.org/10.1101/2022.04.26.489523 22. Refsland, E. W., Stenglein, M. D., Shindo, K., Albin, J. S., Brown, W. L., and Harris, R. S. (2010) Quantitative proﬁling of the full APOBEC3 mRNA repertoire in lymphocytes and tissues: implications for HIV-1 restriction. Nucl. Acids Res. 38, 4274–4284 23. Burns, M. B., Lackey, L., Carpenter, M. A., Rathore, A., Land, A. M., Leonard, B., et al. (2013) APOBEC3B is an enzymatic source of mutation in breast cancer. Nature 494, 366–370 24. Burns, M. B., Temiz, N. A., and Harris, R. S. (2013) Evidence for APO- BEC3B mutagenesis in multiple human cancers. Nat. Genet. 45, 977–983 25. Oh, S., Bournique, E., Bowen, D., Jalili, P., Sanchez, A., Ward, I., et al. (2021) Genotoxic stress and viral infection induce transient expression of APOBEC3A and pro-inﬂammatory genes through two distinct pathways. Nat. Commun. 12, 4917 26. Weisblum, Y., Oiknine-Djian, E., Zakay-Rones, Z., Vorontsov, O., Hai- mov-Kochman, R., Nevo, Y., et al. (2017) APOBEC3A is upregulated by human cytomegalovirus (HCMV) in the maternal-fetal interface, acting as an innate anti-HCMV effector. J. Virol. https://doi.org/10.1128/JVI. 01296-17 27. Milewska, A., Kindler, E., Vkovski, P., Zeglen, S., Ochman, M., Thiel, V., et al. (2018) APOBEC3-mediated restriction of RNA virus replication. Sci. Rep. 8, 5960 28. Warren, C. J., Xu, T., Guo, K., Grifﬁn, L. M., Westrich, J. A., Lee, D., et al. (2015) APOBEC3A functions as a restriction factor of human papillo- mavirus. J. Virol. 89, 688–702 29. Warren, C. J., Westrich, J. A., Doorslaer, K. V., and Pyeon, D. (2017) Roles of APOBEC3A and APOBEC3B in human papillomavirus infection and disease progression. Viruses. https://doi.org/10.3390/v9080233 30. Baker, S. C., Mason, A. S., Slip, R. G., Skinner, K. T., Macdonald, A., Masood, O., et al. (2022) Induction of APOBEC3-mediated genomic damage in urothelium implicates BK polyomavirus (BKPyV) as a hit-and- run driver for bladder cancer. Oncogene 41, 2139–2151 31. Stenglein, M. D., Burns, M. B., Li, M., Lengyel, J., and Harris, R. S. (2010) APOBEC3 proteins mediate the clearance of foreign DNA from human cells. Nat. Struct. Mol. Biol. 17, 222–229 32. Zhe, W., Kousho, W., Kouichi, K., Satoru, A., Guangyan, L., Miki, K., et al. (2014) APOBEC3 deaminases induce hypermutation in human J. Biol. Chem. (2023) 299(9) 105073 11 APOBEC3A upregulation by mitochondrial dsRNA papillomavirus 16 DNA upon beta-interferon stimulation. J. Virol. 88, 1308–1317 the accumulation of double-stranded RNA in vivo. PLoS Genet. 15, e1008240 33. Koning, F. A., Newman, E. N. C., Kim, E.-Y., Kunstman, K. J., Wolinsky, S. M., and Malim, M. H. (2009) Deﬁning APOBEC3 expression patterns in human tissues and hematopoietic cell subsets. J. Virol. 83, 9474–9485 34. Peng, G., Lei, K. J., Jin, W., Greenwell-Wild, T., and Wahl, S. M. (2006) Induction of APOBEC3 family proteins, a defensive maneuver underlying interferon-induced anti-HIV-1 activity. J. Exp. Med. 203, 41–46 35. Cheng, A. Z., Yockteng-Melgar, J., Jarvis, M. C., Malik-Soni, N., Borozan, I., Carpenter, M. A., et al. (2019) Epstein-Barr virus BORF2 inhibits cellular APOBEC3B to preserve viral genome integrity. Nat. Microbiol. 4, 78–88 36. Hultquist, J. F., Lengyel, J. A., Refsland, E. W., LaRue, R. S., Lackey, L., Brown, W. L., et al. (2011) Human and rhesus APOBEC3D, APOBEC3F, APOBEC3G, and APOBEC3H demonstrate a conserved capacity to restrict Vif-deﬁcient HIV-1. J. Virol. 85, 11220–11234 37. Alexandrov, L. B., Nik-Zainal, S., Wedge, D. C., Aparicio, S. A. J. R., Behjati, S., Biankin, A. V., et al. (2013) Signatures of mutational processes in human cancer. Nature 500, 415–421 38. Nik-Zainal, S., Alexandrov, L. B., Wedge, D. C., Van Loo, P., Greenman, C. D., Raine, K., et al. (2012) Mutational processes molding the genomes of 21 breast cancers. Cell 149, 979–993 39. Roberts, S. A., Lawrence, M. S., Klimczak, L. J., Grimm, S. A., Fargo, D., Stojanov, P., et al. (2013) An APOBEC cytidine deaminase mutagenesis pattern is widespread in human cancers. Nat. Genet. 45, 970–976 40. Bartsch, K., Knittler, K., Borowski, C., Rudnik, S., Damme, M., Aden, K., et al. (2017) Absence of RNase H2 triggers generation of immunogenic micronuclei removed by autophagy. Hum. Mol. Genet. 26, 3960–3972 41. Lövgren, T., Eloranta, M.-L., Båve, U., Alm, G. V., and Rönnblom, L. (2004) Induction of interferon-α production in plasmacytoid dendritic cells by immune complexes containing nucleic acid released by necrotic or late apoptotic cells and lupus IgG. Arthritis Rheum. 50, 1861–1872 42. Gázquez-gutiérrez, A. N. A., Witteveldt, J., Heras, S. R., and Macias, S. (2021) Sensing of transposable elements by the antiviral innate immune system. RNA. https://doi.org/10.1261/rna.078721.121 52. Wang, G., Chen, H.-W., Oktay, Y., Zhang, J., Allen, E. L., Smith, G. M., et al. (2010) PNPASE regulates RNA import into mitochondria. Cell 142, 456–467 53. Niklas, J., Melnyk, A., Yuan, Y., and Heinzle, E. (2011) Selective per- meabilization for the high-throughput measurement of compartmented enzyme activities in mammalian cells. Anal. Biochem. 416, 218–227 54. Yu, C.-H., Davidson, S., Harapas, C. R., Hilton, J. B., Mlodzianoski, M. J., Laohamonthonkul, P., et al. (2020) TDP-43 triggers mitochondrial DNA release via mPTP to activate cGAS/STING in ALS. Cell 183, 636–649.e18 55. Dunker, W., Ye, X., Zhao, Y., Liu, L., Richardson, A., and Karijolich, J. (2021) TDP-43 prevents endogenous RNAs from triggering a lethal RIG- I-dependent interferon response. Cell Rep. 35, 108976 56. Ishikawa, H., Ma, Z., and Barber, G. N. (2009) STING regulates intra- cellular DNA-mediated, type I interferon-dependent innate immunity. Nature 461, 788–792 57. Ishikawa, H., and Barber, G. N. (2008) STING is an endoplasmic retic- ulum adaptor that facilitates innate immune signalling. Nature 455, 674–678 58. Schoggins, J. W., Wilson, S. J., Panis, M., Murphy, M. Y., Jones, C. T., Bieniasz, P., et al. (2011) A diverse range of gene products are effectors of the type I interferon antiviral response. Nature 472, 481–485 59. Darnell, J. E. J., Kerr, I. M., and Stark, G. R. (1994) Jak-STAT pathways and transcriptional activation in response to IFNs and other extracellular signaling proteins. Science 264, 1415–1421 60. Stark, G. R., and Darnell, J. E. J. (2012) The JAK-STAT pathway at twenty. Immunity 36, 503–514 61. Fanunza, E., Grandi, N., Quartu, M., Carletti, F., Ermellino, L., Milia, J., et al. (2021) INMI1 Zika virus NS4B antagonizes the interferon signaling by suppressing STAT1 Phosphorylation. Viruses. https://doi.org/10.3390/ v13122448 62. Fanunza, E., Carletti, F., Quartu, M., Grandi, N., Ermellino, L., Milia, J., et al. (2021) Zika virus NS2A inhibits interferon signaling by degradation of STAT1 and STAT2. Virulence 12, 1580–1596 43. Guo, Y., Gu, R., Gan, D., Hu, F., Li, G., and Xu, G. (2020) Mitochondrial inﬂammation activation via cGAS–STING DNA drives noncanonical signaling pathway in retinal microvascular endothelial cells. Cell Com- mun. Signal. 18, 172 63. Leonard, B., McCann, J. L., Starrett, G. J., Kosyakovsky, L., Luengas, E. M., Molan, A. M., et al. (2015) The PKC/NF-κB signaling pathway induces APOBEC3B expression in multiple human cancers. Cancer Res. 75, 4538–4547 44. West, A. P., Khoury-Hanold, W., Staron, M., Tal, M. C., Pineda, C. M., Lang, S. M., et al. (2015) Mitochondrial DNA stress primes the antiviral innate immune response. Nature 520, 553–557 45. Dhir, A., Dhir, S., Borowski, L. S., Jimenez, L., Teitell, M., Rötig, A., et al. (2018) Mitochondrial double-stranded RNA triggers antiviral signalling in humans. Nature 560, 238–242 46. Dressaire, C., Pobre, V., Laguerre, S., Girbal, L., Arraiano, C. M., and Cocaign-Bousquet, M. (2018) PNPase is involved in the coordination of mRNA degradation and expression in stationary phase cells of Escher- ichia coli. BMC Genomics 19, 848 64. Roelofs, P. A., Goh, C. Y., Chua, B. H., Jarvis, M. C., Stewart, T. A., McCann, J. L., et al. (2020) Characterization of the mechanism by which the RB/E2F pathway controls expression of the cancer genomic DNA deaminase APOBEC3B. Elife. https://doi.org/10.7554/eLife.61287 65. Siriwardena, S. U., Perera, M. L. W., Senevirathne, V., Stewart, J., and Bhagwat, A. S. (2019) A tumor-promoting phorbol ester causes a large increase in APOBEC3A expression and a moderate increase in APO- line without BEC3B expression in a normal human keratinocyte cell increasing genomic uracils. Mol. Cell. Biol. https://doi.org/10.1128/MCB. 00238-18 47. Wang, D. D.-H., Shu, Z., Lieser, S. A., Chen, P.-L., and Lee, W.-H. (2009) Human mitochondrial SUV3 and polynucleotide phosphorylase form a 330-kDa heteropentamer to cooperatively degrade double-stranded RNA with a 3’-to-5’ directionality. J. Biol. Chem. 284, 20812–20821 66. Brown, W. L., Law, E. K., Argyris, P. P., Carpenter, M. A., Levin-Klein, R., Ranum, A. N., et al. (2019) A rabbit monoclonal antibody against the antiviral and cancer genomic dna mutating enzyme APOBEC3B. Anti- bodies. https://doi.org/10.3390/antib8030047. Basel, Switzerland 48. Nagaike, T., Suzuki, T., Katoh, T., and Ueda, T. (2005) Human mito- chondrial mRNAs are stabilized with polyadenylation regulated by mitochondria-speciﬁc poly(A) polymerase and polynucleotide phos- phorylase. J. Biol. Chem. 280, 19721–19727 49. Borowski, L. S., Dziembowski, A., Hejnowicz, M. S., Stepien, P. P., and Szczesny, R. J. (2013) Human mitochondrial RNA decay mediated by PNPase-hSuv3 complex takes place in distinct foci. Nucl. Acids Res. 41, 1223–1240 50. Szczesny, R. J., Borowski, L. S., Brzezniak, L. K., Dmochowska, A., Gewartowski, K., Bartnik, E., et al. (2010) Human mitochondrial RNA turnover caught in ﬂagranti: involvement of hSuv3p helicase in RNA surveillance. Nucl. Acids Res. 38, 279–298 51. Pajak, A., Laine, I., Clemente, P., El-Fissi, N., Schober, F. A., Maffez- zini, C., et al. (2019) Defects of mitochondrial RNA turnover lead to 67. Land, A. M., Law, E. K., Carpenter, M. A., Lackey, L., Brown, W. L., and Harris, R. S. (2013) Endogenous APOBEC3A DNA cytosine deaminase is cytoplasmic and nongenotoxic. J. Biol. Chem. 288, 17253–17260 68. Manjunath, L., Oh, S., Ortega, P., Bouin, A., Bournique, E., Sanchez, A., et al. (2023) APOBEC3B drives PKR-mediated translation shutdown and protects stress granules in response to viral infection. Nat. Commun. 14, 820 69. Stopak, K. S., Chiu, Y.-L., Kropp, J., Grant, R. M., and Greene, W. C. (2007) Distinct patterns of cytokine regulation of APOBEC3G expression and activity in primary lymphocytes, macrophages, and dendritic cells. J. Biol. Chem. 282, 3539–3546 70. Peng, G., Greenwell-Wild, T., Nares, S., Jin, W., Lei, K. J., Rangel, Z. G., et al. (2007) Myeloid differentiation and susceptibility to HIV-1 are linked to APOBEC3 expression. Blood 110, 393–400 12 J. Biol. Chem. (2023) 299(9) 105073 APOBEC3A upregulation by mitochondrial dsRNA 71. Petljak, M., Alexandrov, L. B., Brammeld, J. S., Price, S., Wedge, D. C., Grossmann, S., et al. (2019) Characterizing mutational signatures in human cancer cell lines reveals episodic APOBEC mutagenesis. Cell 176, 1282–1294.e20 72. Feng, X., Tubbs, A., Zhang, C., Tang, M., Sridharan, S., Wang, C., et al. (2020) ATR inhibition potentiates ionizing radiation-induced interferon response via cytosolic nucleic acid-sensing pathways. EMBO J. 39, e104036 73. Carpenter, M. A., Law, E. K., Serebrenik, A., Brown, W. L., and Harris, R. S. (2019) A lentivirus-based system for Cas9/gRNA expression and subsequent removal by Cre-mediated recombination. Methods 156, 79–84 J. Biol. Chem. (2023) 299(9) 105073 13
10.1016_j.isci.2021.102204	UC San Diego UC San Diego Previously Published Works Title Sensitivity of the mangrove-estuarine microbial community to aquaculture effluent Permalink https://escholarship.org/uc/item/3229890c Journal iScience, 24(3) ISSN 2589-0042 Authors Erazo, Natalia G Bowman, Jeff S Publication Date 2021-03-01 DOI 10.1016/j.isci.2021.102204 Copyright Information This work is made available under the terms of a Creative Commons Attribution- NonCommercial-NoDerivatives License, available at https://creativecommons.org/licenses/by-nc-nd/4.0/ Peer reviewed eScholarship.org Powered by the California Digital Library University of California iScience ll OPEN ACCESS Article Sensitivity of the mangrove-estuarine microbial community to aquaculture efﬂuent Natalia G. Erazo, Jeff S. Bowman [email protected] HIGHLIGHTS In near-intact mangrove forests, we observed the presence of nitrogen ﬁxers Calothrix could play a role in increasing nitrogen inventories via nitrogen ﬁxation Disturbed sites were correlated with increased nitrogen and reduction in diversity Disturbed sites were dominated by nitriﬁers, denitriﬁes, and sulfur- oxidizing bacteria Erazo & Bowman, iScience 24, 102204 March 19, 2021 ª 2021 The Authors. https://doi.org/10.1016/ j.isci.2021.102204 iScience ll OPEN ACCESS Article Sensitivity of the mangrove-estuarine microbial community to aquaculture efﬂuent Natalia G. Erazo1,3,4,* and Jeff S. Bowman1,2,3 SUMMARY Mangrove-dominated estuaries host a diverse microbial assemblage that facili- tates nutrient and carbon conversions and could play a vital role in maintaining ecosystem health. In this study, we used 16S rRNA gene analysis, metabolic infer- ence, nutrient concentrations, and d13C and d15N isotopes to evaluate the impact of land use change on near-shore biogeochemical cycles and microbial community structures within mangrove-dominated estuaries. Samples in close proximity to 3(cid:1); lower in mi- active shrimp aquaculture were high in NH4 crobial community and metabolic diversity; and dominated by putative nitriﬁers, denitriﬁes, and sulfur-oxidizing bacteria. Near intact mangrove forests we observed the presence of potential nitrogen ﬁxers of the genus Calothrix and or- der Rhizobiales. We identiﬁed possible indicators of aquaculture efﬂuents such as Pseudomonas balearica, Ponitmonas salivibrio, family Chromatiaceae, and genus Arcobacter. These results highlight the sensitivity of the estuarine-mangrove mi- crobial community, and their ecosystem functions, to land use changes. (cid:1), and PO4 +, NO3 (cid:1) NO2 INTRODUCTION Mangrove forests are among the most productive ecosystems in the world, harbor signiﬁcant biodiversity, and provide numerous ecosystem services (Ewel et al., 1998). These forests aid in the exchange of carbon and nutrients with the coastal marine environment (Robertson et al., 2011), with an estimated export of 10% of the marine dissolved organic matter to adjacent ecosystems (Dittmar and Lara, 2001). These forests act as carbon sinks by sequestering CO2, help stabilize coastlines, and support coastal ﬁsheries by acting as nursery grounds for a range of marine species (Kathiresan and Bingham, 2001). Despite their ecological and economic importance they have suffered severe losses in the past years (Duke et al., 2007). Although deforestation rates have declined (Friess et al., 2020), mangrove forests are still threatened by pollution, overextraction, conversion to aquaculture, agriculture, and the overall degradation of the environment (Lovelock et al., 2004; Reef et al., 2010; Friess et al., 2019). A key driver of the reduction in mangrove forest area is the expansion of shrimp aquaculture. Within Ecuador, the expansion of aquaculture exceeds the global trend with deforestation rates higher than 80% (Hamilton and Lovette, 2015). Here, shrimp aquaculture has grown to a $1.3 billion industry by 2012 and represents the second largest component of the Ecuadorian economy after fossil fuels (Hamilton and Lovette, 2015). Shrimp aquaculture efﬂuent is associated with the input of excess nutrients to adjacent coastal ecosystems; consequently, it can lead to changes in microbial community structure, biogeochemical cycles, and eutrophication (Maher et al., 2016; Rosentreter et al., 2018). Changes in nutrient ﬂuxes can indi- rectly alter the redox state of the water column and sediment. This can shift mangrove forests from acting as sinks to sources of greenhouse gases such as CO2, nitrous oxide, and methane (Maher et al., 2016). Microorganisms (here meaning single-celled members of the domains bacteria, archaea, and eukarya) are a key component of the mangrove forest and are present in the sediment, the water column, and as bioﬁlms on mangrove roots (Vazquez et al., 2000; Holguin et al., 2001). These microbes interact with mangroves as co-dependent ecosystem engineers and are responsible for many of the biogeochemical processes attrib- uted to mangrove forests (Holguin et al., 2006; Reis et al., 2017; Shiau and Chiu, 2020). Mangrove forest productivity, for example, is dependent on the microbial recycling mechanisms that keep nitrogen and 1Scripps Institution of Oceanography, UC San Diego, 8622 Kennel Way, La Jolla, CA 92037, USA 2Center for Microbiome Innovation, UC San Diego, La Jolla, CA, USA 3Center for Marine Biodiversity and Conservation, UC San Diego, La Jolla, CA, USA 4Lead contact Correspondence: [email protected] https://doi.org/10.1016/j.isci. 2021.102204 iScience 24, 102204, March 19, 2021 ª 2021 The Authors. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). 1 ll OPEN ACCESS iScience Article Table 1. Environmental properties for high, intermediate, and low disturbed mangrove forests Disturbance Low Phosphate (mM)a 0.23 G 0.23 Nitrate+nitrite (mM)a 0.46 G 0.54 Ammonia (mM)a 0.39 G 0.36 Chlorophyll (mg L-1)a 11.52 G 5.86 Intermediate 0.33 G 0.33 0.87 G 0.46 1.77 G 0.60 8.80 G 2.58 High 2.41 G 1.01 9.91 G 8.75 12.79 G 7.50 p Valuef 9.10 3 1011 8.2 3 10 (cid:1)12 2.20 3 10 (cid:1)16 30.75 G 23.52 1.70 3 10 (cid:1)6 d13C (range)b (cid:1)18.45, (cid:1)27.76 (cid:1)18.49, (cid:1)29.00 (cid:1)27.01, (cid:1)32.08 – d15N (range)b 0.36, 11.08 0.54, 8.84 0.73, 5.86 – Samples (n)c, d, e 89 34 29 – aMean value. bLow and high values provided. cLow disturbance (Cayapas-Mataje = 88, Muisne = 1). dIntermediate disturbance (Cayapas-Mataje = 33, Muisne = 1). eHigh disturbance (Muisne = 29). fp Value (Kruskal-Wallis test). other nutrients within the system (Alongi, 1994). Because of the dependence of ecosystem functions on mi- crobes, microbes can be used as sensitive indicators of environmental change and stress. The planktonic microbial community in mangrove forests has been understudied when compared with the sediment community (Gomes et al., 2011; Imchen et al., 2017; Zhang et al., 2017; Gong et al., 2019). In this study, we evaluated the impact of land use change (mangrove forest converted to aquaculture) on micro- bial community structure and key biogeochemical parameters in the water column. We tested the hypoth- esis that shrimp aquaculture facilities are correlated with increased nitrogen inputs, altered microbial struc- ture, and alpha diversity. We identiﬁed speciﬁc microbial taxa that were differentially present between more and less perturbed sites associated with different levels of nutrient enrichment due to land use change. These taxa can be further developed as indicators of perturbation and mangrove forest health. The observed changes in the microbial community structure of the more and less disturbed sites high- lighted the sensitivity of the mangrove forest to aquaculture efﬂuent, with implications for coastal biogeo- chemical cycling and carbon and nitrogen subsidies to adjacent ecosystems. RESULTS Physicochemical properties (cid:1)1 for phosphate, 9.91 G 8.75 mmol L (cid:1)1 for nitrate + nitrite, 12.79 G 7.50 mmol L The disturbed sites (Muisne) were associated with higher levels of ammonia, nitrate + nitrite, phosphate, and chlorophyll a near aquaculture efﬂuent sites (Figure 1). The mean concentrations were 2.41 G (cid:1)1 for ammonia, 1.01 mmol L (cid:1)1 for chlorophyll a (Table 1 and Figure 2). These biogeochemical parameters were and 30.75 G 23.52 mg L (cid:1)6, respectively) in signiﬁcantly lower (Kruskal-Wallis test, p = 9.1 3 10 (cid:1)1 for phosphate, 0.46 G the low disturbance forest (Cayapas-Mataje) with values of 0.23 G 0.23 mmol L (cid:1)1 chlorophyll 0.54 mmol L a. Areas of intermediate disturbance were found around limited aquaculture facilities where the mean con- (cid:1)1 for nitrate + nitrite, 1.77 G centrations were 0.33 G 0.33 mmol L 0.60 mmol L (cid:1)1 for nitrate + nitrite, 0.39 G 0.36 mmol L (cid:1)1 for phosphate, 0.87 G 0.46 mmol L (cid:1)1 for ammonia, and 8.80 G 2.58 mg L (cid:1)1 for chlorophyll (Table 1, Figure 2). (cid:1)1 ammonia, and 11.52 G 5.86 mg L (cid:1)16, 1.7 3 10 (cid:1)12, 2.2 3 10 (cid:1)11, 8.2 3 10 C and N isotope values ranged from (cid:1)18.45 to (cid:1)27.76&d13C in the low disturbed sites, (cid:1)18.94 to (cid:1)29.00&d13C in the intermediate disturbed sites, and (cid:1)27.01 to (cid:1)32.08&d13C in the high disturbed sites (Table 1, Figure 2). The d15N values ranged from 0.36 to 11.08& in the low disturbed sites, 0.54 to 8.84& in the intermediate disturbed sites, and 0.73 to 5.86& in the high disturbed sites (Table 1, Figure 2). The N (cid:1)1; for low and intermediate distur- value for the high disturbed sites ranged from (cid:1)43.68 to (cid:1)4.44 mmol L (cid:1)1 (Figure 2). We identiﬁed higher N:P ratios associated bance sites it ranged from (cid:1)28.10 to 0.21 mmol L with high disturbance and lower ratios with low disturbance sites, and we observed a negative correlation (cid:1)8) and 16S rRNA gene copy number (Spearman’s with genome size (Spearman’s rho = (cid:1)0.46, p = 9.3 3 10 (cid:1)6) (Figure 2). The taxa most associated with smaller predicted genomes were Can- rho = (cid:1)0.5, p = 1.7 3 10 didatus Dependentiae (1.14 Mb), Candidatus Nasuia deltocephalinicola (1.12 Mb), and Candidatus Pelagi- bacter sp. IMCC9063 (1.28 Mb). The taxa most associated with larger predicted genomes were genera 2 iScience 24, 102204, March 19, 2021 ll OPEN ACCESS iScience Article A B C D Figure 1. Map of study site in coastal Ecuador (A) Study site in Esmeraldas, Ecuador, South America. (B) Location of the two ecological reserves: Cayapas-Mataje (CM) and Muisne (M). (C and D) (C) Map of land use changes in CM and (D) map of land use changes in M; green shows mangrove forest cover, pink shows shrimp aquaculture cover, and yellow circles show sampling locations. The base maps were generated from data obtained in Hamilton (2020). Calothrix (12.05 Mb), Oscillatoria acuminata (7.80 Mb), Moorea producens PAL-8-15-08-1 (9.71 Mb), San- daracinus amylolyticus (10.33 Mb), and Singulisphaera acidiphila (9.76 Mb). Alpha diversity For the bacterial community, the inverse Simpson’s indicator of diversity was signiﬁcantly lower in the highly disturbed sites when compared with the intermediate and low sites with mean G SD values of 36.08 G 26.41, (cid:1)9) (Figure 3). The mean diversity 30.05 G 17.56, and 56.73 G 19.82 respectively, (Kruskal-Wallis, p = 3.5 3 10 for the archaeal community was 5.00 G 1.34 for high, 6.38 G 2.29 for intermediate, and 6.85 G 2.87 for low distur- bance sites, and low and intermediate disturbance sites had signiﬁcant higher diversity than high disturbance (cid:1)6) (Figure 3). Alpha diversity for the archaeal community was lower than for sites (Kruskal-Wallis, p = 1.4 3 10 the bacterial community. Low disturbance sites had higher diversity than intermediate disturbance sites for the bacterial community, but no difference was observed between low and intermediate sites for the archaeal community (Figure 3). We also evaluated the predicted metabolic diversity for the bacterial community; the mean metabolic diversity for low disturbance was 244.01 G 8.72 (mean G SD); for intermediate disturbance, was 235.22 G 8.02; and for high disturbance, was 237.87 G 9.29. The low disturbance sites had higher metabolic diversity (Kruskal-Wallis, p = 2.2 3 10 (cid:1)7) when compared with intermediate and high disturbance sites. Differentiated abundance of bacterial and archaeal communities and metabolic pathways Unique reads are represented at the strain (closest completed genome or [CCG]) or clade level (closest estimated genome [CEG]) depending on the point of placement by paprica. The bacterial community iScience 24, 102204, March 19, 2021 3 ll OPEN ACCESS iScience Article A D B E C F Figure 2. Biogeochemical and bacterial signatures (A–C) (A) Nitrogen (ammonia and nitrate + nitrite) and phosphate species concentrations, (B) mean of genome size versus N:P ratio, (C) mean of number of 16S copies versus N:P ratio and Spearman’s correlation. (D and E) (D) N* value and (E) chlorophyll values of three levels of disturbance. Kruskal-Wallis test and p values with Dunn post-test. p < 0.01, p < 0.001. (F) d13C and d15N isotopic signatures. composition was dominated by the class Actinobacteria: Rhodoluna lacicola (CEG), Actinobacteria bacte- rium IMCC26256 (CCG), Acidimicrobium ferrooxidans DSM 10331 (CCG); family Pelagibacteraceae: Can- didatus Pelagibacter sp. IMCC9063 (CCG), Candidatus Puniceispirillum marinum IMCC1322 (CCG), Candi- family Flavobacteriaceae: Kordia sp. SMS9 (CCG), datus Pelagibacter ubique HTCC1062 (CCG); Owenweeksia hongkongensis DSM 17368 (CCG); cyanobacteria: Synechococcus sp. WH 7803 (CCG); and family Rhodobacteraceae: Thalassococcus sp. S3 (CCG) and Sulﬁtobacter sp. AM1-D1(CCG). The archaeal community was dominated by the most abundant class Thermoplasmata: Candidatus Methano- massiliicoccus intestinalis Issoire-Mx1 (CCG), class Methanococci: Methanococcales (CEG), and phylum Thaumarchaeota (Figure S1). Our DESeq2 results identiﬁed 333 amplicon sequence variants or ASVs that were signiﬁcantly different between sites separated by level of disturbance. Here we focus on the top 60 most abundant differentially present ASVs that were signiﬁcantly differentially present across our entire dataset (Figure 4). Members of Chromatiaceae bac- (cid:1)9), genus Delf- terium 2141T.STBD.0c.01a (CCG) (p = 2.02 3 10 (cid:1)7), Steroidobacter denitri- tia (CEG) (p = 1.03 3 10 (cid:1)8) were the most ﬁcans (CEG) (p = 6.17 3 10 signiﬁcantly most abundant taxa in the high disturbed site than in the low disturbed site. Cyanobacteria such as (cid:1)29), and M. producens PAL-8-15-08-1 (CCG) (p = 2.45 3 10 (cid:1)13), and Pseudomonas balearica DSM 6083 (CCG) (p = 2.52 3 10 (cid:1)14), Arcobacter nitroﬁgilis DSM 7299 (CCG) (p = 1.31 3 10 (cid:1)7), order Nostococales (CEG) (p = 7.84 3 10 (cid:1)19), family Planctomycetes (CEG) (p = 1.62 3 10 4 iScience 24, 102204, March 19, 2021 iScience Article ll OPEN ACCESS A B C Figure 3. Alpha diversity (A–C) (A) Bacterial community diversity, (B) archaeal diversity, and (C) metabolic diversity for the three levels of disturbance using InvSimpson metric. Kruskal- Wallis test and p values with Dunn post-test; *p < 0.001. Cyanobium gracile PCC 6307 (CCG) (p = 2.41 3 10 low disturbance sites were characterized by a higher abundance of SAR11 (CEG) (p = 2.17 3 10 dobacteraceae (CEG) (p = 2.30 3 10 yloceanibacter (CEG) (p = 1.42 3 10 (p = 9.12 3 10 othrix sp. NIES-4071 (CCG) (p = 1.79 3 10 (cid:1)38) were also more abundant in the high disturbed sites. The (cid:1)14), family Rho- (cid:1)40), family Flavobacteriaceae (CEG) (p = 8.26 3 10 (cid:1)28), and genus Meth- (cid:1)8). Oscillatoria species such as Oscillatoria nigroviridis PCC 7112 (CCG) (cid:1)7) were more abundant in the low and intermediate disturbed sites as well as cyanobacteria Cal- (cid:1)20) (Figure 4). For domain archaea we identiﬁed a total of seven (CEG) taxa that were the most abundant and differentially (cid:1)13) was associated with high disturbed samples. Candi- present. Candidatus Korarchaeota (p = 1.09 3 10 (cid:1)68), genus Meth- datus Mancarchaeum acidiphilum (p = 4.28 3 10 (cid:1)56) were more abundant anomassiliicoccus (p = 5.08 3 10 in low disturbed sites (Figure 4). (cid:1)28), and genus Methanococcales (p = 3.79 3 10 (cid:1)40), genus Nitrosopumilus (p = 2.92 3 10 A correspondence analysis (CA) of bacterial and archaeal community structures depicted the dissimilar relationship of samples for bacteria and archaea in terms of level of disturbance associated with aquacul- ture (Figure 5). For bacteria, the ﬁrst axis explained 30.6%, and the second axis, 18.3%. The top contributing taxa to the difference were Betaproteobacteria (cos2 = 0.86), Acidothermus cellulolyticus 11B (cos2 = 0.83), and S. denitriﬁcans (cos2 = 0.91) (Figure 5). For the archaeal community, the ﬁrst dimension accounted for 19.9% and the second dimension accounted for 11.8% of variability. Among the top contributors to the two dimensions were class Thermoplasmata (cos2 = 0.61), Candidatus Methanomassiliicoccus intestinalis (cos2 = 0.73), and Candidatus Mancarchaeum acidiphilum (cos2 = 0.63) (Figure 5). The results of our ANO- SIM test showed that the bacterial and archaeal communities were signiﬁcantly different for low and high disturbance mangrove forests (R = 0.52 and p value = 0.001, R = 0.45 and p value = 0.001). We also observed clear association of location of samples with ammonia concentration in dimension 1 (Spearman’s rho = (cid:1)9) for bacteria, and for archaea 0.56, p = 1.4 3 10 only dimension 2 showed a signiﬁcant correlation (Spearman’s rho = 0.49, p = 7.2 3 10 (cid:1)10) and dimension 2 (Spearman’s rho = 0.54, p = 1.2 3 10 (cid:1)5) (Figure S2). A canonical correspondence analysis (CCA) was further performed to examine the relationships between metabolic pathways and environmental factors. This showed that the biogeochemical parameters associ- ated with nitrogen species, phosphate, N:P, chlorophyll a, and d13C and d15N together accounted for 20% of the variability in the metabolic pathways. Nitrogen, phosphorus, and chlorophyll were factors that inﬂu- enced the metabolic pathways in the high disturbance sites. The ﬁrst dimension accounted for 26.1%, and the second dimension accounted for 9.1% of the variability. Here, the top contributors’ predicted meta- bolic pathways of dimethylsulfoniopropionate (DMSP) degradation III methylation (cos2 = 0.61) and glycine betaine (GBT) degradation I (cos2 = 0.62) were associated with low and intermediate disturbance. Taxa associated with DMSP degradation III methylation were Candidatus Puniceispirillum marinum IMCC1322 and Thalassococcus sp. S3, and for GBT degradation, taxa were Alphaproteobacterium HIMB59, cyano- bacteria, and Pelagibacteraceae. Other metabolic pathways with high contributions were arsenate iScience 24, 102204, March 19, 2021 5 ll OPEN ACCESS A B iScience Article Figure 4. Microbial and archaeal signatures of disturbance (A and B) (A) Differentially abundant bacterial taxa (top 60) in high, intermediate, and low disturbance result from DESeq2 analysis; (B) differentially abundant archaeal taxa result from DESeq2. Samples and taxa were clustered using Bray-Curtis dissimilarity distance. detoxiﬁcation (cos2 = 0.75) and methylphosphonate degradation (cos2 = 0.83), both associated with high disturbance sites (Figure 6). Taxa associated with these pathways were Erythrobacter atlanticus and Can- didatus Puniceispirillum marinum IMCC1322 for arsenate detoxiﬁcation and Starkeya novella DSM 506 (or- der Rizobiales) and Oceanicola sp. 3 for methylphosphonate degradation. Weighted gene correlation network analysis (cid:1)17, pink: r = 0.86, p = 2 3 10 (cid:1)50) Moorea producens PAL-8-15-08-01 (CCG) (r = 0.89, p = 5.18 3 10 Weighted gene correlation network analysis (WGCNA) found clusters of highly correlated taxa across samples. We related these clusters to ammonia and nitrate + nitrite to better understand the impact of aquaculture efﬂuent on microbial community structure. We identiﬁed eight major modules or subnetworks. Each module was as- signed a particular color (Figure S3). The blue and pink modules were positively correlated with ammonia and (cid:1)43). The yellow module was negatively nitrate + nitrite (blue: r = 0.64, p = 6.00 3 10 (cid:1)6) (Figure S3). Taxa associated with the pink correlated with ammonia, nitrate, and nitrite (r = (cid:1)0.42, p = 6.00 3 10 (cid:1)53), Actinobacteria bacterium module (Figure S4) included Sulfurivermis fontis (CEG) (r = 0.90, p = 1.45 3 10 (cid:1)53), Candidatus Methylopumilus planktonicus (CEG) (r = 0.89, p = IMCC26256 (CCG) (r = 0.90, p = 9.62 3 10 (cid:1)50), Phycisphaera mikurensis 1.98 3 10 (cid:1)30), and NBRC 102666 (r = 0.85, p = 1.19 3 10 (cid:1)30). All these taxa were signiﬁcantly correlated with Steroidobacter denitriﬁcans (CEG) (r = 0.79, p = 2.31 3 10 ammonia, and with nitrate + nitrite (Table 2). Taxa most strongly associated with the blue module consisted of (cid:1)8), A. cellulolyticus 11B (CCG) (r = 0.69, p = C. bacterium 2141T.STBD.0c.01a (CCG) (r = 0.49, p = 8.88 3 10 (cid:1)14), Pontimonas salivibrio (CEG) (r = 0.59, p = 3.37 3 10 (cid:1)20) (Table 2, Figure S4). The taxa 1.52 3 10 that were most negatively correlated with ammonia in the yellow module were: O. acuminata PCC 6304 (CCG) (cid:1)8), Candidatus pelagi- (cid:1)3, Synechococcus sp. WH 7803 (CCG) (r = (cid:1)0.49, p = 4.25 3 10 (r = (cid:1)0.34, p = 9.57 3 10 (cid:1)7), and Coraliomargarita akajimensis DSM 45221 (CCG) (r = bacter sp. IMCC9063 (CCG) (r = (cid:1)0.37, p = 1.17 3 10 (cid:1)0.36, p = 7.51 3 10 (cid:1)20), S. denitriﬁcans (CEG) (r = 0.61, p = 3.62 3 10 (cid:1)15), M. producens PAL-8-15-08-01 (CCG) (r = 0.69, p = 5.07 3 10 (cid:1)40), Cyanobium gracile PCC 6307 (CCG) (r = 0.78, p = 7.76 3 10 (cid:1)6) (Table 2, Figure S4). We further explored taxa that correlated with salinity to better understand the impact of tide on the micro- (cid:1)8) bial community structure. The red module was positively correlated with salinity (r = 0.47, p = 7.00 3 10 6 iScience 24, 102204, March 19, 2021 iScience Article A C B D ll OPEN ACCESS E Figure 5. Bacterial and archaeal community structure (A–E) Correspondence analysis (CA) ordination of (A) the bacterial community for samples that cluster in ordination space have similar community compositions, whereas those that are dispersed are less similar. (B) Square cosine components for samples; large value of cos2 shows a relatively large contribution to the total distance for bacterial community. (C) CA ordination for archaeal community. (D) Square cosine components for samples for archaeal community. (E) Contribution of top 10 taxa with highest cos2 values for archaeal community (see Figure S2). (Figure S3). Taxa associated with the red module included Acidimicrobium ferrooxidans DSM 10331 (CCG) (cid:1)9), Synechococcus sp. (cid:1)10), Haliglobus japonicus (CEG) (r = 0.52, p = 2.66 3 10 (r = 0.53, p = 8.19 3 10 (cid:1)8), Candidatus Puniceispirillum marinum IMCC1322 (CCG)) (r = CC9605 (CCG) (r = 0.50, p = 2.28 3 10 0.47, p = 4.56 3 10 (cid:1)7), and Prochlorococcus marinus str. MIT 9301 (r = 0.40, p = 1.98 3 10 (cid:1)4) (Table 3). We analyzed the Hellinger-transformed enzyme level output from paprica to better understand the enzy- matic potential of those CEG and CCG that were correlated with ammonia. We found 35 enzymes associ- ated with the nitrogen cycle (Figure 6). The enzyme nitrogenase EC 1.18.6.1 had a mean value of 0.23 G 0.05 for the low disturbed sites, signiﬁcantly higher than that in the intermediate (0.17 G 0.06) and high (cid:1)10) (Figure S5). Nitrate reductase EC 1.7.99.4 had a mean value (0.17 G 0.05) disturbance sites (p = 1.2 3 10 of 0.23 G 0.05 for low disturbance site, 0.45 G 0.14 for intermediate disturbance site, and 0.44 G 0.13 for high disturbance site, and the nitrate reductase value was signiﬁcantly higher in the high disturbance sites (cid:1)14). The same was observed with nitrate reductase NADH EC 1.7.1.4 with a mean of 0.13 G (p = 8.7 3 10 0.05 for low disturbed sites, 0.23 G 0.14 for intermediate disturbance, and 0.23 G 0.13 for highly disturbed (cid:1)15) (Figures 6 andS5). The taxa that were associated with nitrogenase were Methylocella sites (p = 2 3 10 silvestris BL2, genus Calothrix, and Synechococcus sp. CC9605. For nitrate reductase members of the Be- taproteobacteria, Desulfococcus oleovorans Hxd3, and P. mikurensis NBRC 102666 were found to contribute to enzyme abundance. The taxa that were associated with nitrate reductase NADH were A. cellulolyticus 11B and members of the Rhodobacteraceae (Table 4). DISCUSSION Mangrove forests are experiencing a high degree of perturbation through nutrient enrichment, pollution, and deforestation. Shrimp aquaculture efﬂuent in particular is associated with the input of excess nutrients to mangrove forests. In this study we found that shrimp aquaculture efﬂuent is associated with changes in microbial community structure with likely consequences for biogeochemical cycles and mangrove forest health. Previous work suggests that for intensive shrimp farming, 2.22 km2 of mangrove forest is required iScience 24, 102204, March 19, 2021 7 ll OPEN ACCESS A B iScience Article Figure 6. Metabolic pathways and nitrogen cycle enzyme indicators for levels of disturbance (A) CCA ordination for metabolic pathways showing top four pathways with cos2 ranging from 0.6–0.8. Large value of cos2 shows a relatively large contribution to the total distance for bacterial metabolic prediction. (B) Heatmap of key nitrogen cycle enzymes (Bray-Curtis distance) for the bacterial community (see Figure S5). to remove efﬂuent from one pond of 0.01 km2, whereas 0.20 km2 is required for less-intensive farming from one pond of 0.01 km2 (Robertson and Phillips, 1995). As of 2014 in the Muisne region there were 20.47 km2 of shrimp farms and 12.06 km2 of mangrove forests, indicative of an intensive farming system. Cayapas- Mataje had 11.04 km2 of shrimp aquaculture farms and 302.05 km2 of mangrove forest, suggesting less intensive farming (Figure 1) (Hamilton, 2020). As the areal extent of shrimp aquaculture increases so does the volume of the efﬂuent, elevating the ﬂux of ammonia and nitrate to the surrounding ecosystem. Based on our observations we found that microbial communities in mangrove forests are signiﬁcantly altered by this perturbation. The bacterial communities in our mangrove systems were characterized by members of the Pelagibacter- aceae, Flavobacteriaceae, Rhodobacteraceae, Actinobacteria, and cyanobacteria (Figure 4, Figure S1). The archaeal community was dominated by members of the Thermoplasmata, Thaumarchaeota, and Methanococcales (Figure 4, Figure S1). This was in accordance with other studies that have identiﬁed Rhodobacteraceae, SAR86 clade, Actinobacteria, and Flavobacteriaceae, and Thaumarchaeota as the most abundant taxonomic groups (Dhal et al., 2020). Rhodobacteraceae has been found to be dominant in mangrove-dominated estuaries, and members of this family are associated with marine phytoplankton blooms where they play a role in transformations of derived phytoplankton organic matter (Ghosh et al., 2010; Simon et al., 2017). The presence of Actinobacteria has been documented previously in mangrove ecosystems (Azman et al., 2015; Gong et al., 2019), and it has been suggested that they could play a role in carbon cycling by decomposing the plant biomass including refractory lignins (Scott et al., 2010). Thau- marchaeota are the most abundant archaea in the surface ocean (Santoro et al., 2015), and Thermoplas- mata have been found in mangrove ecosystems (Zhang et al., 2019). Both these groups play an important role in the nitrogen cycle by carrying out the oxidation of ammonia in nitriﬁcation (Santoro et al., 2015; Zhang et al., 2019). Both bacterial and archaeal communities were less diverse at our more disturbed sites. This pattern extended to predicted metabolic diversity (Figure 3). We hypothesize that this reduction in diversity could cause reductions in ecosystem functions. This has been observed in previous mangrove forest studies, for example, where lower microbial diversity was associated with a reduction in microbial productivity in sites with high levels of deforestation, sewage, and ﬁshing activities (Carugati et al., 2018). Our results showed differences in biogeochemical parameters between sites at varying levels of distur- bance (Figure 2). In particular, nitrogen was a driver of the microbial community structure leading to segre- gation into three clusters of disturbance in the CA analysis based on our ANOSIM test and signiﬁcantly 8 iScience 24, 102204, March 19, 2021 iScience Article ll OPEN ACCESS Table 2. Signiﬁcant correlated taxa with ammonia and nitrate+nitrite result from WGCNA Taxon Thermogutta terrifontisf Acidothermus cellulolyticus 11B Oceanicola sp. D3 Moorea producens PAL-8-15-08-1 Moorea producens PAL-8-15-08-1 Actinobacteria bacterium IMCC26256 Steroidobacter denitriﬁcans Aureitalea sp. RR4-38 Pontimonas salivibrio Pontimonas salivibrio Acidothermus cellulolyticus 11Bf Candidatus Xiphinematobacter sp. Idaho grape Synechococcus sp. CB0101 Candidatus Cyclonatronum proteinivorum Candidatus Cyclonatronum proteinivorum Halioglobus paciﬁcus Synechococcus sp. WH 8101 Rhodoluna lacicola Actinobacteria bacterium IMCC26256 Thiohalobacter thiocyanaticus Actinobacteria bacterium IMCC26256 Chromatiaceae bacterium 2141T.STBD.0c.01a Wenzhouxiangella marina Thiolapillus brandeum Candidatus Puniceispirillum marinum IMCC1322 Thermogutta terrifontis Candidatus Pelagibacter sp. IMCC9063 Sulfurivermis fontisd Actinobacteria bacterium IMCC26256 Candidatus Methylopumilus planktonicus Moorea producens PAL-8-15-08-1 Owenweeksia hongkongensis DSM 17368 Phycisphaera mikurensis NBRC 102666e Thermogutta terrifontis Candidatus Planktophila vernalis Actinobacteria bacterium IMCC26256 Candidatus Xiphinematobacter sp. Idaho grape Candidatus Pelagibacter sp. IMCC9063 Owenweeksia hongkongensis DSM 17368 Syntrophus aciditrophicus SB Candidatus Pelagibacter sp. IMCC9063 Phycisphaera mikurensis NBRC 102666e Map IDa Module color GS.Nitrogenb 0.87 0.94 0.97 0.82 0.82 0.88 0.91 0.92 0.98 0.98 0.94 0.88 0.98 0.87 0.87 565 0.98 0.98 0.88 0.92 0.88 0.95 0.98 0.94 0.97 0.87 0.91 0.86 0.88 0.96 0.82 0.89 0.8 0.87 0.95 0.88 0.88 0.91 0.89 0.84 0.92 0.8 Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Blue Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink Pink 0.74 0.69 0.69 0.69 0.67 0.62 0.61 0.61 0.59 0.59 0.58 0.58 0.58 0.57 0.56 0.56 0.54 0.53 0.53 0.51 0.50 0.49 0.48 0.45 0.45 0.44 0.40 0.90 0.90 0.89 0.89 0.87 0.85 0.85 0.85 0.84 0.83 0.82 0.81 0.80 0.79 0.79 p.GS.Nitrogenc 6.35 3 10 3.37 3 10 (cid:1)20 (cid:1)25 4.17 3 10 5.07 3 10 2.62 3 10 8.57 3 10 3.62 3 10 5.70 3 10 5.75 3 10 1.52 3 10 1.29 3 10 (cid:1)20 (cid:1)20 (cid:1)18 (cid:1)15 (cid:1)14 (cid:1)14 (cid:1)13 (cid:1)15 (cid:1)12 1.80 3 10 (cid:1)12 3.36 3 10 9.00 3 10 2.06 3 10 2.19 3 10 3.96 3 10 (cid:1)12 (cid:1)12 (cid:1)11 (cid:1)11 (cid:1)10 1.58 3 10 1.60 3 10 (cid:1)9 (cid:1)9 1.94 3 10 9.01 3 10 (cid:1)12 (cid:1)12 8.88 3 10 1.48 3 10 2.19 3 10 3.59 3 10 (cid:1)8 (cid:1)7 (cid:1)6 (cid:1)6 4.27 3 10 1.29 3 10 (cid:1)6 (cid:1)4 1.45 3 10 9.62 3 10 (cid:1)53 (cid:1)53 1.98 3 10 5.18 3 10 4.75 3 10 1.19 3 10 2.24 3 10 9.07 3 10 1.34 3 10 (cid:1)50 (cid:1)50 (cid:1)46 (cid:1)40 (cid:1)40 (cid:1)40 (cid:1)39 7.44 3 10 (cid:1)37 1.31 3 10 (cid:1)34 3.97 3 10 3.58 3 10 1.73 3 10 1.89 3 10 (cid:1)33 (cid:1)32 (cid:1)31 (cid:1)30 (Continued on next page) iScience 24, 102204, March 19, 2021 9 iScience Article ll OPEN ACCESS Table 2. Continued Taxon Steroidobacter denitriﬁcans Cyanobium gracile PCC 6307 Cyanobium gracile PCC 6307 Halioglobus japonicus Halomicronema hongdechloris C2206 Thermogutta terrifontis Mariniﬁlaceae bacterium SPP2 Steroidobacter denitriﬁcans Aureitalea sp. RR4-38 Halioglobus paciﬁcus Synechococcus sp. WH 7803 Candidatus Methylopumilus planktonicus Flavobacteriaceae bacterium Thiolapillus brandeum Owenweeksia hongkongensis DSM 17368 Thermogutta terrifontis Candidatus Pelagibacter sp. IMCC9063 Coraliomargarita akajimensis DSM 45221 Oscillatoria acuminata PCC 6304 Acidothermus cellulolyticus 11Bf Map IDa Module color GS.Nitrogenb 0.91 0.98 0.98 0.91 0.9 0.87 0.85 0.91 0.92 0.94 0.99 0.96 0.91 0.94 0.89 0.87 0.92 0.89 0.81 0.94 Pink Pink Pink Pink Pink Pink Pink Pink Yellow Yellow Yellow Yellow Yellow Yellow Yellow Yellow Yellow Yellow Yellow Yellow 0.79 0.78 0.72 0.56 0.53 0.53 0.47 0.40 (cid:1)0.57 (cid:1)0.50 (cid:1)0.49 (cid:1)0.47 (cid:1)0.45 (cid:1)0.42 (cid:1)0.39 (cid:1)0.38 (cid:1)0.37 (cid:1)0.36 (cid:1)0.34 (cid:1)0.34 p.GS.Nitrogenc 2.31 3 10 (cid:1)30 7.76 3 10 2.75 3 10 2.91 3 10 7.85 3 10 9.21 3 10 3.13 3 10 1.07 3 10 4.73 3 10 1.58 3 10 4.25 3 10 4.55 3 10 2.36 3 10 3.22 3 10 2.45 3 10 7.07 3 10 1.17 3 10 2.84 3 10 6.15 3 10 9.57 3 10 (cid:1)30 (cid:1)23 (cid:1)11 (cid:1)10 (cid:1)10 (cid:1)7 (cid:1)4 (cid:1)12 (cid:1)8 (cid:1)8 (cid:1)7 (cid:1)6 (cid:1)5 (cid:1)4 (cid:1)4 (cid:1)3 (cid:1)3 (cid:1)3 (cid:1)3 aMap ID phylogenetic classiﬁcation. Value = 1 represents a perfect placement on the tree. bGS = Pearson correlation to ammonia and nitrate + nitrite. cp.GS = p-adjusted value (Bonferroni correction) for correlation to ammonia and nitrate+nitrite. dRepresents presence of nitrogenase enzyme EC.1.18.61. eRepresents presence of nitrate reductase enzyme EC.1.7.99.4. fRepresents presence of nitrate reductase enzyme EC.1.7.1.4. correlated with ammonia concentrations (Figure 5, Figure S2). We note that the variance explained by the ﬁrst and second dimensions in our ordination analyses is relatively low (30.6% and 19.9% for the bacterial and archaeal communities, respectively). We attribute this to the complexity associated with the mangrove ecosystem and the large number of physical, chemical, and biological factors that could impact changes in the microbial community. We found a strong connection between N:P ratio and genome size among planktonic bacteria across study sites (Figure 2). Generally, smaller predicted genomes and lower 16S rRNA gene copy number was asso- ciated with higher N:P ratios, whereas larger predicted genomes and higher 16S rRNA gene copy number was associated with lower N:P ratios. The differences in genome sizes between communities associated with different levels of disturbance suggest differing ecological strategies. Studies suggest that generalists possess larger genomes in contrast to the smaller genomes in more specialized microbes (Sriswasdi et al., 2017; Willis and Woodhouse, 2020). This falls from the generalist requirement for a larger gene repertoire to boost activity in multiple environmental conditions and to cope with different stressors associated with a broad physicochemical niche (such as low levels of nitrogen and tidal ﬂuctuations in mangrove-dominated estuaries). The low disturbance sites showed a higher metabolic diversity and larger genomes, which we interpret as a more generalist microbial community. Taxa with larger genomes included Planctomycetes such as Singulisphaera acidiphila. This taxon has been found in other wetland ecosystems (Kulichevskaya et al., 2008; Dedysh and Ivanova, 2019), and it has been shown to play an important role in degradation of plant-derived polymers such as pectin and xylan (Dedysh and Ivanova, 2019). The S. acidiphila genome en- codes several dozen proteins that do not belong to any of the currently carbohydrate-active enzymes, but the enzymes display a distant relationship to glycosyltransferases and carbohydrate esterases, suggesting that this taxon has a diverse glycolytic and carbohydrate metabolic potential (Dedysh and Ivanova, 2019). Other taxa included Sandaracinus amylolyticus. This taxon has been found in association with plant 10 iScience 24, 102204, March 19, 2021 iScience Article ll OPEN ACCESS Table 3. Signiﬁcant correlated taxa with salinity result from WGCNA Taxon Acidimicrobium ferrooxidans DSM 10331 Kordia sp. SMS9 Halioglobus japonicus Synechococcus sp. CC9605 Synechococcus sp. RCC307 Candidatus Puniceispirillum marinum IMCC1322 Salipiger profundus Halioglobus paciﬁcus Roseovarius mucosus Acidimicrobium ferrooxidans DSM 10331 Candidatus Pelagibacter ubique HTCC1062 Owenweeksia hongkongensis DSM 17368 Prochlorococcus marinus str. MIT 9301 Synechococcus sp. KORDI-100 Sulfurivermis fontis Map IDa Module color GS.Salinityb 0.94 0.91 0.93 1.00 1.00 0.98 0.82 0.95 0.96 0.82 1.00 0.89 1.00 1.00 0.87 Red Red Red Red Red Red Red Red Red Red Red Red Red Red Red 0.53 0.53 0.52 0.50 0.47 0.47 0.46 0.46 0.44 0.41 0.40 0.40 0.40 0.39 0.39 aMap ID phylogenetic classiﬁcation. Value = 1 represents a perfect placement on the tree. bGS = Pearson correlation to salinity. cp.GS = p-adjusted value (Bonferroni correction) for correlation to salinity. p.GS.Salinityc 8.19 3 10 1.49 3 10 (cid:1)10 (cid:1)9 2.66 3 10 2.28 3 10 4.36 3 10 4.56 3 10 8.37 3 10 1.02 3 10 7.04 3 10 9.08 3 10 1.10 3 10 1.64 3 10 (cid:1)9 (cid:1)8 (cid:1)7 (cid:1)7 (cid:1)7 (cid:1)6 (cid:1)6 (cid:1)5 (cid:1)4 (cid:1)4 1.98 3 10 2.07 3 10 3.93 3 10 (cid:1)4 (cid:1)4 (cid:1)4 residues (Mohr et al., 2012), in coral ecosystems (Rubio-Portillo et al., 2016), and it is known to survive in poor nutrient conditions by developing desiccation-resistant spores (Mohr et al., 2012). We also observed larger genomes in cyanobacteria including members of the genus Calothrix, genus Os- cillatoria, and M. producens PAL-8-15-08-1. Cyanobacteria are known to have large genomes with low cod- ing density and a high level of gene duplication; it has been proposed that the large non-protein-coding sequences contribute to the genome expansion and metabolic ﬂexibility observed in diazotrophs (nitrogen ﬁxers) that are associated with nitrogen-limited environments (Sargent et al., 2016). The high diversity of cyanobacteria observed in mangrove ecosystems suggests that they play a key role in the ecosystem. Rele- vant functions associated with cyanobacteria include nitrogen and carbon ﬁxation and the production of herbivory-defense molecules and plant growth-promoting substances (Alvarenga et al., 2015). In the disturbed sites the parasite C. Dependentiae accounted for much of the decrease in genome size. Studies have found that C. Dependentiae infects a wide range of protists, including heterotrophs and phytoplankton (Deeg et al., 2019). Other studies have shown that C. Dependentiae is associated with free-living ameba, suggesting that it could be an endosymbiont (Delafont et al., 2015). C. Dependentiae has very limited metabolic capability, lacks complete biosynthetic pathways for various essential cellular building blocks, and has protein motifs to facilitate eukaryotic host interactions (Yeoh et al., 2016; Deeg et al., 2019). C. Nasuia deltocephalinicola was also identiﬁed as having a small genome. C. Nasuia delto- cephalinicola is an obligate symbiont of plant phloem-feeding pest insects, and its main role is to provide essential amino acids that the host can neither synthesize nor obtain in sufﬁcient quantities from a plant diet (Bennett and Moran, 2013). The increase in the concentration of nitrogen species associated with aquaculture efﬂuent could further select for specialist microbes with reduced metabolic potential and lower diversity in nitrogen-processing enzymes. Our nitrogen isotope values are consistent with this, showing reduced variability for the highly disturbed sites (Table 1, Figure 2). Previous work in mangrove systems (Bernardino et al., 2018) associated reduced isotopic variability with a loss of trophic diversity. Higher variability in stable carbon isotopes has also been observed in salt marshes due to contribution dominated by allochthonous material derived from the phytoplankton community (Boschker et al., 1999). The larger variation in the isotopic signal observed in the low and intermediate disturbance sites suggests that these pristine systems contain a more diverse iScience 24, 102204, March 19, 2021 11 ll OPEN ACCESS iScience Article Table 4. Number of enzymes copies for nitrogenase and nitrate reductase enzymes and top 10 associated taxa Taxon Methylocella silvestris BL2 Genus Calothrix Synechococcus sp. CC9605 Family Rhodobacteraceae Oscillatoria nigroviridis PCC 7112 Acidothermus cellulolyticus 11B Phycisphaera mikurensis NBRC 102666 Rhodopirellula baltica SH 1 Desulfococcus oleovorans Hxd3 Class Betaproteobacteria Nitrogenase EC.1.18.6.1 Nitrate reductase EC.1.7.99.4 Nitrate reductase NADH EC.1.7.1.4 16,984 15,186 38,937 0 0 0 0 0 0 0 4,246 3,796 0 0 5,418 0 8,288 10,956 4,773 17,102 0 0 0 1,273 5,418 23,205 0 21,912 0 0 trophic food web as result of a wide range of metabolic and ﬁxation pathways, and environmental condi- tions in the mangrove-estuarine ecosystem (Boschker and Middelburg, 2002). The low N:P ratios we observed in the low disturbance sites suggest that the system is N limited. Pristine mangrove forests tend to be N limited, although nutrients are not uniformly distributed within the mangrove ecosystem and they can switch from N to P limitation. It has been shown that mangrove trees within fringe and tidally exposed zones tend to be N limited (Feller et al., 2003). One way mangrove trees cope with N limitation is through associations with diazotrophs that play a crucial role in N cycling within the mangrove forest (Holguin et al., 1992). Here we showed that the biological nitrogen ﬁxation signal, conﬁrmed by the N value (the linear combination of nitrate and phosphate that eliminates the effect of nitriﬁcation; thus, the remaining variability can be explained by nitrogen ﬁxation and denitriﬁcation) (Gruber and Sarmiento, 1997) and nitrogenase EC.1.18.6.1 abundance, were higher at low disturbance sites in contrast to high disturbance sites (Figures 2, 6, and S5). The microbial denitriﬁcation signal was further conﬁrmed by negative N* values in the highest disturbance sites (Figure 2) (Gruber and Sarmiento, 1997). Because excess nitrate is being introduced into the system via aquaculture efﬂuent, we expect denitriﬁca- tion rates to be high. Conversely, the lowest disturbance sites have a slight positive N* consistent with our identiﬁcation of putative diazotrophs such as genus Calothrix, genus Oscillatoria, and taxa of the order Rhi- zobiales such as M. silvestris (Essien et al., 2008; Liu et al., 2019). GBT degradation I was one of the major pathways contributing to the differences observed between low and high disturbed sites (Figures 6 and S3). GBT is an important source of nitrogen in oligotrophic systems, acts as an organic osmolyte, and plays an important role in phytoplankton-bacteria interactions (Becker et al., 2019; Jones et al., 2019; Zecher et al., 2020). The intertidal coastal mangrove ecosystem experiences daily ﬂuctuations in a range of environmental conditions, including water levels and salinity. Organisms living in this dynamic environment cope with changing environmental conditions by synthesizing a range of organic and inorganic osmolytes including GBT. The results from WGCNA showed that Pelagibactera- ceae taxa correlated with salinity (Table 3) and primary contributors of the GBT degradation I pathway. This suggests that osmolyte production is an important adaptation to salinity intrusions from oceanic waters into the mangrove environment, and GBT could be an additional pool of organic N within this system. Shrimp aquaculture impacts the water quality in adjacent mangrove forests by creating eutrophic condi- tions that can lead to anoxia. Eutrophic conditions were evident through high levels of nutrients and chlo- rophyll a (Figure 2, Table 1). Although we did not measure oxygen concentrations, we observed taxa indic- ative of hypoxic or anoxic conditions. These included purple sulfur bacteria (PSB), such as family Chromaticeae, and sulfur-oxidizing bacteria (SOB), such as genus Sulfurivermis (Figure 4, Table 2). PSB use sulﬁde, elemental sulfur, and thiosulfate as electron donors in anoxygenic photosynthesis and have been shown to play an important role in regime shifts from oxygenated to anoxic conditions (Diao et al., 2018). PSB ﬂourish in micro-aerobic conditions oxidizing sulﬁde into sulfate (Diao et al., 2018). As the oxy- gen inﬂux is reduced below a critical threshold, sulfate-reducing bacteria (SRB) and PSB can take over and outcompete the SOB. This suggests a more anoxic regime in the high disturbance site, allowing for PSB groups and SRB to become more abundant. 12 iScience 24, 102204, March 19, 2021 iScience Article ll OPEN ACCESS Based on our WGCNA analysis we also found nitrate-reducing bacteria (NRB)—indicative of reduced ox- ygen availability—that strongly correlated with the level of ammonia, nitrate, and nitrite. Putative NRB taxa included P. mikurensis and S. denitriﬁcans (Table 2). In addition, we also saw a microbial signature associated with dissimilatory nitrate reduction to ammonium (DNRA) with the presence of genus Acido- thermus, and anaerobic ammonium oxidation (annamox) with the presence of Planctomycetes (Thermo- gutta terrifontis) (Table 2); the presence of the genes involved in these pathways (denitriﬁcation, annamox, DNRA) were inferred by paprica, although further work is needed to conﬁrm the presence and activity of these enzymes. Overall, as nitrate and ammonia inputs increased with aquaculture efﬂuent the relative abundance of NRB increased. We identiﬁed speciﬁc microbes that can be used as sensitive indicators of aquaculture impacts. These included P. balearica (Figure 4), which has been associated with other contaminated wetland systems, sug- gesting that this taxon could be a potential bio-indicator of a disturbed mangrove ecosystem (Salva` -Serra et al., 2017). Similar studies have also identiﬁed aquaculture efﬂuent as a source of pathogens to the coastal ecosystem (Garren et al., 2009). In the disturbed site we saw the presence of members of the genus Arco- bacter (Figure 4). These bacteria have been identiﬁed in coral systems exposed to aquaculture efﬂuents, and have been associated with feces (human, porcine, and bovine) and with sewage-contaminated waters (Garren et al., 2009). PSB taxa such as family Chromaticeae have also been shown to be potential bio-in- dicators for anthropogenic contamination associated with other agriculture efﬂuent systems (Mohd-Nor et al., 2018). P. salivibrio in the order Micrococcales was elevated at the disturbed sites. This taxon has been isolated from high-salinity systems and aquaculture farms (Jang et al., 2013); high salinity levels have been associated with shrimp aquaculture efﬂuent due to high evaporation in the ponds (Barraza- Guardado et al., 2013). Previous studies have shown that taxa in the Micrococcales order are part of the core microbiome signal in shrimp ponds (Chen et al., 2017). Thus, P. salivibrio is a sensible indicator of shrimp aquaculture efﬂuent. Further work is needed to establish robust spatiotemporal baselines of micro- bial indicators of aquaculture to effectively monitor biogeochemical changes and health of the mangrove forests. Aquaculture could impact the health of mangrove ecosystems involving the direct loss of mangrove forests, efﬂuent associated with high levels of nutrients, and the development of anoxic and sulfuric water condi- tions (Robertson and Phillips, 1995). Aquaculture efﬂuent released into mangrove forests may be seques- tered and processed by bacteria. However, processing efﬁciency could change with increasing input. High organic loadings, for example, may shift the balance from aerobic to anaerobic systems (Lønborg et al., 2020). Anaerobic systems are less efﬁcient in nutrient cycling. The signals of SOB, SRB, denitriﬁers, and po- tential pathogenic taxa associated with the perturbed site suggest that aquaculture efﬂuent is playing a role in shifting the microbial community to a more pathogenic and less nutrient efﬁcient community that could impact the health of the mangrove forest. Conclusion In this study, we showed the impacts of aquaculture efﬂuent on the microbial community structure in mangrove forests and identiﬁed microbial signals associated with NRB, PSB, and SRB taxa that could have impacts in nutrient cycling. The high level of nutrients in the perturbed sites were associated with changes in microbial community structure that could impact ecosystem functions. In the low disturbance sites, we saw that the presence of Calothrix species and nitrogen ﬁxers could be important in increasing nitrogen inventories via nitrogen ﬁxation. Denitriﬁcation reduces excess inorganic nitrogen concentration, and in the highly disturbed sites we saw the presence of NRB-associated microbes. Nutrient cycling in mangrove habitats is a balance between nutrient inputs, availability, and internal cycling, and the changes in microbial community structure we see in disturbed sites could be indicators of biogeochemical changes. The results of the study highlight the sensitivity of the mangrove-estuarine microbial community to aqua- culture efﬂuent, and the impacts of land use changes could be ampliﬁed by climate change such as chang- ing precipitation patterns, heat, and rising sea level with severe consequences for the ecosystem. Limitations of the study Our analysis was based on comparison between sites of low, intermediate, and high disturbance in two mangrove systems in coastal Ecuador. Although ammonia concentration is a good proxy for disturbance from shrimp aquaculture efﬂuent, quantiﬁcation of land use changes, and the hydrological connections be- tween aquaculture facilities and our sampling sites was beyond the scope of the current work. We iScience 24, 102204, March 19, 2021 13 ll OPEN ACCESS iScience Article considered salinity, macronutrient concentrations, and isotopes in our analysis, but anticipate that other variables not considered here are contributing to differences in microbial community structure. These include physical processes such as tides and hydrology. The complexity of these environments is evident in our CCA and CA analyses, which capture a relatively small amount of variability in the ﬁrst two dimensions (see discussion section). Other limitations of note are typical of microbial community structure analyses. These include primer bias and a dependence on relative rather than absolute abundance. Resource availability Lead contact Further information and requests for resources should be directed to and will be fulﬁlled by the lead con- tact, Natalia Erazo ([email protected]). Materials availability This study did not generate new unique reagents. Data and code availability The data that support the ﬁndings of this study and sequences were submitted to the NCBI sequence read archive (SRA) under BioProject ID: PRJNA633714. Code for analysis is available on github repository: https://github.com/galud27. METHODS All methods can be found in the accompanying Transparent methods supplemental ﬁle. SUPPLEMENTAL INFORMATION Supplemental information can be found online at https://doi.org/10.1016/j.isci.2021.102204. ACKNOWLEDGMENTS N.G.E. was supported by Organization of American States and SENESCYT fellowships. J.S.B. was sup- ported by a grant from the Simons Foundation Early Career Investigator in Marine Microbiology program. We would like to thank Jesse Wilson and Avishek Dutta for helpful discussion on methods and interpreta- tion, Jesse Estacio for ﬁeld work assistance, and Fernando Rivera for assistance with logistics and permits. Samples were collected under the Ecuador environmental permit (MAE-UAFE-DPAE-2017-2009-E). We would like to thank the platform (mindthegraph.com) used here to create the graphical abstract. AUTHOR CONTRIBUTIONS Conceptualization, N.G.E. and J.S.B.; Methodology, N.G.E. and J.S.B; Investigation, N.G.E. and J.S.B; Writing – Original Draft, N.G.E.; Writing – Review & Editing, J.S.B.; Funding Acquisition and Supervision, J.S.B. DECLARATION OF INTERESTS The authors declare no competing interests. Received: July 30, 2020 Revised: September 25, 2020 Accepted: February 15, 2021 Published: March 19, 2021 REFERENCES Alongi, D.M. (1994). The role of bacteria in nutrient recycling in tropical mangrove and other coastal benthic ecosystems. Hydrobiologia 285, 19–32. Alvarenga, D.O., Rigonato, J., Branco, L.H.Z., and Fiore, M.F. (2015). Cyanobacteria in 14 iScience 24, 102204, March 19, 2021 mangrove ecosystems. Biodivers. Conserv. 24, 799–817. Azman, A.S., Othman, I., Velu, S.S., Chan, K.G., and Lee, L.H. (2015). Mangrove rare actinobacteria: taxonomy, natural compound, and discovery of bioactivity. Front.Microbiol. 6, 856. Barraza-Guardado, R.H., Arreola-Liza´ rraga, J.A., Lo´ pez-Torres, M.A., Casillas-Herna´ ndez, R., Miranda-Baeza, A., Magallo´ n-Barrajas, F., and Ibarra-Ga´ mez, C. (2013). Efﬂuents of shrimp farms and its inﬂuence on the coastal ecosystems of Bahı´a de Kino, Mexico. Sci.WorldJ. 2013, 306370. iScience Article Becker, J.W., Hogle, S.L., Rosendo, K., and Chisholm, S.W. (2019). Co-culture and biogeography of Prochlorococcus and SAR11. ISME J. 13, 1506–1519. Essien, J.P., Antai, S.P., and Benson, N.U. (2008). Microalgae biodiversity and biomass status in Qua Iboe Estuary mangrove swamp, Nigeria. Aquat. Ecol. 42, 71–81. Bennett, G.M., and Moran, N.A. (2013). Small, smaller, smallest: the origins and evolution of ancient dual symbioses in a phloem-feeding insect. Genome Biol. Evol. 5, 1675–1688. Ewel, K.C., Twilley, R.R., and Ong, J.E. (1998). Different kinds of mangrove forests provide different goods and services. Glob. Ecol. Biogeogr. Lett. 7, 83. Bernardino, A.F., Gomes, L.E.O., Hadlich, H.L., Andrades, R., and Correa, L.B. (2018). Mangrove clearing impacts on macrofaunal assemblages and benthic food webs in a tropical estuary. Mar. Pollut.Bull. 126, 228–235. Boschker, H.T.S., de Brouwer, J.F.C., and Cappenberg, T.E. (1999). The contribution of macrophyte-derived organic matter to microbial biomass in salt-marsh sediments: stable carbon isotope analysis of microbial biomarkers. Limnol. Oceanogr. 44, 309–319. Boschker, H.T., and Middelburg, J.J. (2002). Stable isotopes and biomarkers in microbial ecology. FEMS Microbiol. Ecol. 40, 85–95. Carugati, L., Gatto, B., Rastelli, E., Lo Martire, M., Coral, C., Greco, S., and Danovaro, R. (2018). Impact of mangrove forests degradation on biodiversity and ecosystem functioning. Sci. Rep. 8, 13298. Chen, W.-Y., Ng, T.H., Wu, J.H., Chen, J.W., and Wang, H.C. (2017). Microbiome dynamics in a shrimp grow-out pond with possible outbreak of acute hepatopancreatic necrosis disease. Sci. Rep. 7, 9395. Dedysh, S.N., and Ivanova, A.A. (2019). Planctomycetes in boreal and subarctic wetlands: diversity patterns and potential ecological functions. FEMS Microbiol. Ecol. 95, 227. Deeg, C.M., Zimmer, M.M., George, E.E., Husnik, F., Keeling, P.J., and Suttle, C.A. (2019). Chromulinavorax destructans, a pathogen of microzooplankton that provides a window into the enigmatic candidate phylum dependentiae. PLoS Pathog. 15, e1007801. Delafont, V., Samba-Louaka, A., Bouchon, D., Moulin, L., and He´ chard, Y. (2015). Shedding light on microbial dark matter: a TM6 bacterium as natural endosymbiont of a free-living amoeba. Environ.Microbiol. Rep. 7, 970–978. Dhal, P.K., Kopprio, G.A., and Ga¨ rdes, A. (2020). Insights on aquatic microbiome of the Indian Sundarbans mangrove areasJ.-S. Hwang, ed. 15, e0221543. Diao, M., Huisman, J., and Muyzer, G. (2018). Spatio-temporal dynamics of sulfur bacteria during oxic-anoxic regime shifts in a seasonally stratiﬁed lake. FEMS Microbiol. Ecol. 94, 40. Feller, I.C., McKee, K.L., Whigham, D.F., and O’Neill, J.P. (2003). Nitrogen vs. phosphorus limitation across an ecotonal gradient in a mangrove forest. Biogeochemistry 62, 145–175. Friess, D.A., Rogers, K., Lovelock, C.E., Krauss, K.W., Hamilton, S.E., Lee, S.Y., Lucas, R., Primavera, J., Rajkaran, A., and Shi, S. (2019). ‘The state of the world’s mangrove forests: past, present, and future’. Annu. Rev. Environ. Resour. 44, 89–115. Friess, D.A., Yando, E.S., Abuchahla, G.M.O., Adams, J.B., Cannicci, S., Canty, S.W.J., Cavanaugh, K.C., Connolly, R.M., Cormier, N., and Dahdouh-Guebas, F. (2020). Mangroves give cause for conservation optimism, for now. Curr. Biol. 30, R153–R154. Garren, M., Raymundo, L., Guest, J., Harvell, C.D., and Azam, F. (2009). Resilience of coral- associated bacterial communities exposed to ﬁsh farm efﬂuent. PLoS One 4, e7319. Ghosh, A., Dey, N., Bera, A., Tiwari, A., Sathyaniranjan, K.B., Chakrabarti, K., and Chattopadhyay, D. (2010). Culture independent molecular analysis of bacterial communities in the mangrove sediment of Sundarban, India. Saline Syst. 6, 1. Gomes, N.C., Cleary, D.F., Calado, R., and Costa, R. (2011). Mangrove bacterial richness. Commun.Integr. Biol. 4 (4), 419–423. Gong, B., Cao, H., Peng, C., Per(cid:1)culija, V., Tong, G., Fang, H., Wei, X., and Ouyang, S. (2019). High- throughput sequencing and analysis of microbial communities in the mangrove swamps along the coast of Beibu Gulf in Guangxi, China. Sci. Rep. 9, 9377. Gruber, N., and Sarmiento, J.L. (1997). Global patterns of marine nitrogen ﬁxation and denitriﬁcation. Glob.Biogeochem. Cycles 11, 235–266. Hamilton, S.E. (2020). Assessing 50 Years of Mangrove Forest Loss along the Paciﬁc Coast of Ecuador: A Remote Sensing Synthesis. Coastal Research Library (Springer), pp. 111–137. Hamilton, S.E., and Lovette, J. (2015). Ecuador’s mangrove forest carbon stocks: a spatiotemporal analysis of living carbon holdings and their depletion since the advent of commercial aquacultureB. Ruttenberg, ed. 10, e0118880. Dittmar, T., and Lara, R.J. (2001). Do mangroves rather than rivers provide nutrients to coastal environments south of the Amazon River? Evidence from long-term ﬂux measurements. Mar. Ecol. Prog.Ser. 213, 67–77. Holguin, G., Gonzalez-Zamorano, P., de-Bashan, L.E., Mendoza, R., Amador, E., and Bashan, Y. (2006). Mangrove health in an arid environment encroached by urban development-a case study. Sci. Total Environ. 363 (1–3), 260–274. Duke, N.C., Meynecke, J.-O., Dittmann, S., Ellison, A.M., Anger, K., Berger, U., Cannicci, S., Diele, K., Ewel, K.C., Field, C.D., et al. (2007). A world without mangroves? Science 317, 41b–42b. Holguin, G., Guzman, M.A., and Bashan, Y. (1992). Two new nitrogen-ﬁxing bacteria from the rhizosphere of mangrove trees: their isolation, identiﬁcation and in vitro interaction with ll OPEN ACCESS rhizosphere Staphylococcus sp. FEMS Microbiol. Lett. 101 (3), 207–216. Holguin, G., Vazquez, P., and Bashan, Y. (2001). The role of sediment microorganisms in the productivity, conservation, and rehabilitation of mangrove ecosystems: an overview. Biol. Fertil. Soils 33, 265–278. Imchen, M., Kumavath, R., Barh, D., Avezedo, V., Ghosh, P., Viana, M., and Wattam, A.R. (2017). Searching for signatures across microbial communities: metagenomic analysis of soil samples from mangrove and other ecosystems. Sci. Rep. 7, 1–13. Jang, G.I., Cho, Y., and Cho, B.C. (2013). Pontimonas salivibrio gen. nov., sp. nov., a new member of the family Microbacteriaceae isolated from a seawater reservoir of a solar saltern. Int. J. Syst. Evol.Microbiol. 63, 2124–2131. Jones, H.J., Kro¨ ber, E., Stephenson, J., Mausz, M.A., Jameson, E., Millard, A., Purdy, K.J., and Chen, Y. (2019). A new family of uncultivated bacteria involved in methanogenesis from the ubiquitous osmolyte glycine betaine in coastal saltmarsh sediments. Microbiome 7, 1–11. Kathiresan, K., and Bingham, B.L. (2001). Biology of mangroves and mangrove ecosystems. Adv. Mar. Biol. 40, 81–251. Kulichevskaya, I.S., Ivanova, A.O., Baulina, O.I., Bodelier, P.L., Damste´ , J.S., and Dedysh, S.N. (2008). Singulisphaera acidiphila gen. nov., sp. nov., a non-ﬁlamentous, Isosphaera-like planctomycete from acidic northern wetlands. Int. J. Syst. Evol.Microbiol. 58, 1186–1193. Liu, M., Huang, H., Bao, S., and Tong, Y. (2019). Microbial community structure of soils in Bamenwan mangrove wetland. Sci. Rep. 9, 8406. Lønborg, C., Carreira, C., Jickells, T., and A´ lvarez- Salgado, X.A. (2020). Impacts of global change on ocean dissolved organic carbon (DOC) cycling. Front. Mar. Sci. 7, 466. Lovelock, C.E., Feller, I.C., Mckee, K.L., Engelbrecht, B.M.J., and Ball, M.C. (2004). The effect of nutrient enrichment on growth, photosynthesis and hydraulic conductance of dwarf mangroves in Panama´ . Funct. Ecol. 18, 25–33. Maher, D.T., Sippo, J.Z., Tait, D.R., Holloway, C., and Santos, I.R. (2016). Pristine mangrove creek waters are a sink of nitrous oxide. Sci. Rep. 6, 25701. Mohd-Nor, D., Ramli, N., Sharuddin, S.S., Hassan, M.A., Mustapha, N.A., Amran, A., Sakai, K., Shirai, Y., and Maeda, T. (2018). Alcaligenaceae and Chromatiaceae as reliable bioindicators present in palm oil mill efﬂuent ﬁnal discharge treated by different biotreatment processes. Ecol. Indicators 95, 468–473. Mohr, K.I., Garcia, R.O., Gerth, K., Irschik, H., and Mu¨ ller, R. (2012). Sandaracinus amylolyticus gen. nov., sp. nov., a starch-degrading soil myxobacterium, and description of Sandaracinaceae fam. nov. Int. J. Syst. Evol.Microbiol. 62, 1191–1198. Reef, R., Feller, I.C., and Lovelock, C.E. (2010). Nutrition of mangroves. Tree Physiol. 30, 1148– 1160. iScience 24, 102204, March 19, 2021 15 ll OPEN ACCESS iScience Article Reis, C.R.G., Nardoto, G.B., and Oliveira, R.S. (2017). Global overview on nitrogen dynamics in mangroves and consequences of increasing nitrogen availability for these systems. Plant and Soil 410, 1–19. Robertson, A.I., Alongi, D.M., and Boto, K.G. (2011). Food Chains and Carbon Fluxes (American Geophysical Union (AGU)), pp. 293–326. Robertson, A.I., and Phillips, M.J. (1995). Mangroves as ﬁlters of shrimp pond efﬂuent: predictions and biogeochemical research needs. Hydrobiologia 295, 311–321. Rosentreter, J.A., Maher, D.T., Erler, D.V., Murray, R.H., and Eyre, B.D. (2018). ‘Methane emissions partially offset ‘‘blue carbon’’ burial in mangroves’. Sci. Adv. 4, eaao4985. Rubio-Portillo, E., Santos, F., Martı´nez-Garcı´a, M., de Los Rı´os, A., Ascaso, C., Souza-Egipsy, V., Ramos-Espla´ , A.A., and Anton, J. (2016). Structure and temporal dynamics of the bacterial communities associated to microhabitats of the coral O culina patagonica. Environ.Microbiol. 18, 4564–4578. Salva` -Serra, F., Jakobsson, H.E., Busquets, A., Gomila, M., Jae´ n-Luchoro, D., Seguı´, C., Aliaga- Lozano, F., Garcı´a-Valde´ s, E., Lalucat, J., Moore, E.R., and Bennasar-Figueras, A. (2017). Genome sequences of two naphthalene-degrading strains of Pseudomonas balearica, isolated from polluted marine sediment and from an oil reﬁnery site. Genome Announc 5, e00116–e00117. Santoro, A.E., Dupont, C.L., Richter, R.A., Craig, M.T., Carini, P., McIlvin, M.R., Yang, Y., Orsi, W.D., Moran, D.M., and Saito, M.A. (2015). ‘Genomic and proteomic characterization of ‘‘CandidatusNitrosopelagicus brevis’’: an ammonia-oxidizing archaeon from the open ocean’. Proc. Natl. Acad. Sci. USA 112, 1173– 1178. Sargent, E.C., Hitchcock, A., Johansson, S.A., Langlois, R., Moore, C.M., LaRoche, J., Poulton, A.J., and Bibby, T.S. (2016). Evidence for polyploidy in the globally important diazotroph Trichodesmium. FEMS Microbiol.Lett. 363, 244. Scott, J.J., Budsberg, K.J., Suen, G., Wixon, D.L., Balser, T.C., and Currie, C.R. (2010). Microbial community structure of leaf-cutter ant fungus gardens and refuse dumpsC.-H. Yang, ed. 5, e9922. Shiau, Y.J., and Chiu, C.Y. (2020). Biogeochemical processes of C and N in the soil of mangrove forest ecosystems. Forests 11, 492. Simon, M., Scheuner, C., Meier-Kolthoff, J.P., Brinkhoff, T., Wagner-Do¨ bler, I., Ulbrich, M., Klenk, H.P., Schomburg, D., Petersen, J., and Go¨ ker, M. (2017). Phylogenomics of Rhodobacteraceae reveals evolutionary adaptation to marine and non-marine habitats. ISME J. 11, 1483–1499. Sriswasdi, S., Yang, C.C., and Iwasaki, W. (2017). Generalist species drive microbial dispersion and evolution. Nat. Commun. 8, 1–8. Vazquez, P., Holguin, G., Puente, M.E., Lopez- Cortes, A., and Bashan, Y. (2000). Phosphate- solubilizing microorganisms associated with the rhizosphere of mangroves in a semiarid coastal lagoon. Biol. Fertil. Soils 30, 460–468. Willis, A., and Woodhouse, J.N. (2020). ‘Deﬁning cyanobacterial species: diversity and description through genomics. Crit. Rev. Plant Sci. 39, 101–124. Yeoh, Y.K., Sekiguchi, Y., Parks, D.H., and Hugenholtz, P. (2016). Comparative genomics of candidate phylum TM6 suggests that parasitism is widespread and ancestral in this lineage. Mol. Biol. Evol. 33, 915–927. Zecher, K., Hayes, K.R., and Philipp, B. (2020). Evidence of interdomain ammonium cross- feeding from methylamine- and Glycine betaine- degrading Rhodobacteraceae to diatoms as a widespread interaction in the marine phycosphere. Front.Microbiol. 11, 533894. Zhang, C.J., Pan, J., Duan, C.H., Wang, Y.M., Liu, Y., Sun, J., Zhou, H.C., Song, X., and Li, M. (2019). Prokaryotic diversity in mangrove sediments across southeastern China fundamentally differs from that in other biomes. mSystems 4, e00442- 19. Zhang, Y., Yang, Q., Ling, J., Van Nostrand, J.D., Shi, Z., Zhou, J., and Dong, J. (2017). Diversity and structure of diazotrophic communities in mangrove rhizosphere ,revealed by high- throughput sequencing. Front.Microbiol. 8, 2032. 16 iScience 24, 102204, March 19, 2021 iScience, Volume 24 Supplemental information Sensitivity of the mangrove-estuarine microbial community to aquaculture efﬂuent Natalia G. Erazo and Jeff S. Bowman A B Disturbance high intermediate low 1 0.8 0.6 0.4 0.2 0 Disturbance Coraliomargarita akajimensis DSM 45221 Candidatus Carsonella ruddii Candidatus Pelagibacter sp. IMCC9063 Thiolapillus brandeum Synechococcus sp. WH 7803 Synechococcaceae Candidatus Dependentiae Candidatus Pelagibacter ubique HTCC1062 Acidimicrobium ferrooxidans DSM 10331 Pelagibacteraceae Kordia sp. SMS9 Bacteria <prokaryotes> Flavobacteriaceae bacterium Candidatus Methylopumilus planktonicus Synechococcus sp. CC9605 Geminocystis Sulfurivermis fontis Cyanobacteria Cyanobacteria Candidatus Vesicomyosocius okutanii HA Chromatiaceae bacterium 2141T.STBD.0c.01a Actinobacteria bacterium IMCC26256 Thalassococcus Thalassococcus sp. S3 Celeribacter indicus Flavobacteriaceae Owenweeksia hongkongensis DSM 17368 Pelagibacteraceae Sulfitobacter sp. AM1−D1 Owenweeksia hongkongensis DSM 17368 Rhodoluna lacicola Pelagibacteraceae Candidatus Pelagibacter sp. IMCC9063 Candidatus Methylopumilus planktonicus Candidatus Puniceispirillum marinum IMCC1322 alpha proteobacterium HIMB59 Sulfurivermis fontis Roseovarius mucosus Halioglobus pacificus Acidimicrobium ferrooxidans DSM 10331 Disturbance high intermediate low 1 0.8 0.6 0.4 0.2 0 Disturbance (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:3)(cid:49)(cid:76)(cid:87)(cid:85)(cid:82)(cid:86)(cid:82)(cid:83)(cid:88)(cid:80)(cid:76)(cid:79)(cid:88)(cid:86)(cid:3)(cid:68)(cid:71)(cid:85)(cid:76)(cid:68)(cid:87)(cid:76)(cid:70)(cid:88)(cid:86) (cid:55)(cid:75)(cid:68)(cid:88)(cid:80)(cid:68)(cid:85)(cid:70)(cid:75)(cid:68)(cid:72)(cid:82)(cid:87)(cid:68) (cid:39)(cid:76)(cid:68)(cid:73)(cid:82)(cid:85)(cid:68)(cid:85)(cid:70)(cid:75)(cid:68)(cid:72)(cid:68) (cid:55)(cid:75)(cid:68)(cid:88)(cid:80)(cid:68)(cid:85)(cid:70)(cid:75)(cid:68)(cid:72)(cid:82)(cid:87)(cid:68) (cid:55)(cid:36)(cid:38)(cid:46)(cid:3)(cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:80)(cid:76)(cid:70)(cid:85)(cid:82)(cid:69)(cid:76)(cid:68)(cid:79)(cid:72)(cid:86) (cid:55)(cid:36)(cid:38)(cid:46)(cid:3)(cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:55)(cid:36)(cid:38)(cid:46)(cid:3)(cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:55)(cid:36)(cid:38)(cid:46) (cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:55)(cid:36)(cid:38)(cid:46) (cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:55)(cid:36)(cid:38)(cid:46) (cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:55)(cid:75)(cid:68)(cid:88)(cid:80)(cid:68)(cid:85)(cid:70)(cid:75)(cid:68)(cid:72)(cid:82)(cid:87)(cid:68) (cid:55)(cid:36)(cid:38)(cid:46) (cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:3)(cid:49)(cid:76)(cid:87)(cid:85)(cid:82)(cid:86)(cid:82)(cid:83)(cid:88)(cid:80)(cid:76)(cid:79)(cid:88)(cid:86)(cid:3)(cid:68)(cid:71)(cid:85)(cid:76)(cid:68)(cid:87)(cid:76)(cid:70)(cid:88)(cid:86) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:3)(cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:80)(cid:68)(cid:86)(cid:86)(cid:76)(cid:79)(cid:76)(cid:76)(cid:70)(cid:82)(cid:70)(cid:70)(cid:88)(cid:86)(cid:3)(cid:76)(cid:81)(cid:87)(cid:72)(cid:86)(cid:87)(cid:76)(cid:81)(cid:68)(cid:79)(cid:76)(cid:86)(cid:3)(cid:44)(cid:86)(cid:86)(cid:82)(cid:76)(cid:85)(cid:72)(cid:16)(cid:48)(cid:91)(cid:20) (cid:55)(cid:36)(cid:38)(cid:46)(cid:3)(cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:3)(cid:49)(cid:76)(cid:87)(cid:85)(cid:82)(cid:86)(cid:82)(cid:83)(cid:88)(cid:80)(cid:76)(cid:79)(cid:88)(cid:86)(cid:3)(cid:68)(cid:71)(cid:85)(cid:76)(cid:68)(cid:87)(cid:76)(cid:70)(cid:88)(cid:86) (cid:3)(cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:80)(cid:76)(cid:70)(cid:85)(cid:82)(cid:69)(cid:76)(cid:68)(cid:79)(cid:72)(cid:86) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:3)(cid:49)(cid:76)(cid:87)(cid:85)(cid:82)(cid:86)(cid:82)(cid:83)(cid:88)(cid:80)(cid:76)(cid:79)(cid:88)(cid:86)(cid:3)(cid:68)(cid:71)(cid:85)(cid:76)(cid:68)(cid:87)(cid:76)(cid:70)(cid:88)(cid:86) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:17)(cid:49)(cid:76)(cid:87)(cid:85)(cid:82)(cid:86)(cid:82)(cid:83)(cid:88)(cid:80)(cid:76)(cid:79)(cid:88)(cid:86)(cid:3)(cid:68)(cid:71)(cid:85)(cid:76)(cid:68)(cid:87)(cid:76)(cid:70)(cid:88)(cid:86) (cid:55)(cid:36)(cid:38)(cid:46)(cid:3)(cid:74)(cid:85)(cid:82)(cid:88)(cid:83) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:17)(cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:80)(cid:68)(cid:86)(cid:86)(cid:76)(cid:79)(cid:76)(cid:76)(cid:70)(cid:82)(cid:70)(cid:70)(cid:88)(cid:86)(cid:17)(cid:76)(cid:81)(cid:87)(cid:72)(cid:86)(cid:87)(cid:76)(cid:81)(cid:68)(cid:79)(cid:76)(cid:86)(cid:17)(cid:44)(cid:86)(cid:86)(cid:82)(cid:76)(cid:85)(cid:72)(cid:17)(cid:48)(cid:91)(cid:20)(cid:17)(cid:20)(cid:3) (cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:70)(cid:82)(cid:70)(cid:70)(cid:68)(cid:79)(cid:72)(cid:86) (cid:38)(cid:68)(cid:81)(cid:71)(cid:76)(cid:71)(cid:68)(cid:87)(cid:88)(cid:86)(cid:17)(cid:48)(cid:72)(cid:87)(cid:75)(cid:68)(cid:81)(cid:82)(cid:80)(cid:68)(cid:86)(cid:86)(cid:76)(cid:79)(cid:76)(cid:76)(cid:70)(cid:82)(cid:70)(cid:70)(cid:88)(cid:86)(cid:17)(cid:76)(cid:81)(cid:87)(cid:72)(cid:86)(cid:87)(cid:76)(cid:81)(cid:68)(cid:79)(cid:76)(cid:86)(cid:17)(cid:44)(cid:86)(cid:86)(cid:82)(cid:76)(cid:85)(cid:72)(cid:17)(cid:48)(cid:91)1 Figure S1. Top abundant microbial community (bacteria and archaea). Heatmap for the most abundant bacteria (A) and archaea (B) taxa. Samples were clustered using Bray-Curtis dissimilarity distance and normalized (Hellinger transformation) abundance. Related to Figure 4. A R = 0.56, p = 1.4e−10 B 1.5 R = 0.54, p = 1.2e−09 ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1 CA dimension 1 2 3 4 ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ● ● R = 0.2, p = 0.075 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1 0 ] M µ [ 4 H N −1 ● ● ● ● ● ● C 1.5 1.0 0.5 0.0 ] M µ [ 4 H N −0.5 ● ● −1.0 1.0 0.5 0.0 −0.5 −1.0 1.5 1.0 0.5 0.0 ] M µ [ 4 H N D ] M µ [ 4 H N ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −4 −2 0 CA dimension 2 R = 0.49, p = 7.2e−05 ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −0.5 −1.0 ● ● ● ● −2.0 −1.5 −1.0 CA dimension 1 ● ● −0.5 0.0 −6 −4 ● ● −2 CA dimension 2 0 Figure S2. Microbial community structure and association to disturbance levels. CA dimension 1 and dimension 2 vs ammonia concentrations for bacteria (A, B) and for archaea (C, D) (Spearman correlation). Related to Figure 5. (cid:48)(cid:82)(cid:71)(cid:88)(cid:79)(cid:72)(cid:239)(cid:87)(cid:85)(cid:68)(cid:76)(cid:87)(cid:3)(cid:85)(cid:72)(cid:79)(cid:68)(cid:87)(cid:76)(cid:82)(cid:81)(cid:86)(cid:75)(cid:76)(cid:83)(cid:86) green(cid:3) (cid:239)(cid:19)(cid:17)(cid:23) (cid:11)(cid:22)(cid:72)(cid:239)(cid:19)(cid:26)(cid:12) (cid:19)(cid:17)(cid:19)(cid:21)(cid:25) (cid:11)(cid:19)(cid:17)(cid:26)(cid:12) (cid:239)(cid:19)(cid:17)(cid:19)(cid:26)(cid:23) (cid:11)(cid:19)(cid:17)(cid:23)(cid:12) (cid:19)(cid:17)(cid:19)(cid:26)(cid:28) (cid:11)(cid:19)(cid:17)(cid:22)(cid:12) (cid:19)(cid:17)(cid:19)(cid:19)(cid:25)(cid:25) (cid:11)(cid:19)(cid:17)(cid:28)(cid:12) (cid:19)(cid:17)(cid:22)(cid:25) (cid:11)(cid:23)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:19)(cid:23)(cid:25) (cid:11)(cid:19)(cid:17)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:21) (cid:11)(cid:19)(cid:17)(cid:19)(cid:21)(cid:12) (cid:239)(cid:19)(cid:17)(cid:21)(cid:22) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:23)(cid:12) (cid:19)(cid:17)(cid:22)(cid:27) (cid:11)(cid:20)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:20) turquoise(cid:3) (cid:239)(cid:19)(cid:17)(cid:24)(cid:22) (cid:11)(cid:22)(cid:72)(cid:239)(cid:20)(cid:21)(cid:12) (cid:239)(cid:19)(cid:17)(cid:23)(cid:26) (cid:11)(cid:21)(cid:72)(cid:239)(cid:19)(cid:28)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:26) (cid:11)(cid:23)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:25) (cid:11)(cid:25)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:28) (cid:11)(cid:26)(cid:72)(cid:239)(cid:19)(cid:26)(cid:12) (cid:19)(cid:17)(cid:20)(cid:21) (cid:11)(cid:19)(cid:17)(cid:20)(cid:12) (cid:19)(cid:17)(cid:23)(cid:23) (cid:11)(cid:20)(cid:72)(cid:239)(cid:19)(cid:27)(cid:12) (cid:19)(cid:17)(cid:19)(cid:25)(cid:24) (cid:11)(cid:19)(cid:17)(cid:23)(cid:12) (cid:19)(cid:17)(cid:24)(cid:21) (cid:11)(cid:26)(cid:72)(cid:239)(cid:20)(cid:21)(cid:12) (cid:19)(cid:17)(cid:22)(cid:25) (cid:11)(cid:25)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) yellow(cid:3) (cid:239)(cid:19)(cid:17)(cid:24)(cid:23) (cid:11)(cid:20)(cid:72)(cid:239)(cid:20)(cid:21)(cid:12) (cid:239)(cid:19)(cid:17)(cid:24)(cid:20) (cid:11)(cid:21)(cid:72)(cid:239)(cid:20)(cid:20)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:24) (cid:11)(cid:27)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:23)(cid:21) (cid:11)(cid:27)(cid:72)(cid:239)(cid:19)(cid:27)(cid:12) (cid:239)(cid:19)(cid:17)(cid:23)(cid:21) (cid:11)(cid:27)(cid:72)(cid:239)(cid:19)(cid:27)(cid:12) (cid:19)(cid:17)(cid:21)(cid:20) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:27)(cid:12) (cid:19)(cid:17)(cid:24) (cid:11)(cid:23)(cid:72)(cid:239)(cid:20)(cid:20)(cid:12) (cid:19)(cid:17)(cid:19)(cid:22)(cid:21) (cid:11)(cid:19)(cid:17)(cid:26)(cid:12) (cid:19)(cid:17)(cid:22)(cid:22) (cid:11)(cid:22)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:19)(cid:17)(cid:20)(cid:27) (cid:11)(cid:19)(cid:17)(cid:19)(cid:22)(cid:12) k r o w t e n b u s / e l u d o M blue(cid:3) pink(cid:3) (cid:19)(cid:17)(cid:25)(cid:23) (cid:11)(cid:25)(cid:72)(cid:239)(cid:20)(cid:28)(cid:12) (cid:19)(cid:17)(cid:26)(cid:20) (cid:11)(cid:21)(cid:72)(cid:239)(cid:21)(cid:23)(cid:12) (cid:19)(cid:17)(cid:24)(cid:24) (cid:11)(cid:23)(cid:72)(cid:239)(cid:20)(cid:22)(cid:12) (cid:19)(cid:17)(cid:25)(cid:23) (cid:11)(cid:21)(cid:72)(cid:239)(cid:20)(cid:27)(cid:12) (cid:19)(cid:17)(cid:25)(cid:23) (cid:11)(cid:27)(cid:72)(cid:239)(cid:20)(cid:28)(cid:12) (cid:239)(cid:19)(cid:17)(cid:20)(cid:21) (cid:11)(cid:19)(cid:17)(cid:20)(cid:12) (cid:239)(cid:19)(cid:17)(cid:25)(cid:21) (cid:11)(cid:22)(cid:72)(cid:239)(cid:20)(cid:26)(cid:12) (cid:239)(cid:19)(cid:17)(cid:20)(cid:23) (cid:11)(cid:19)(cid:17)(cid:19)(cid:27)(cid:12) (cid:239)(cid:19)(cid:17)(cid:23)(cid:28) (cid:11)(cid:20)(cid:72)(cid:239)(cid:20)(cid:19)(cid:12) (cid:239)(cid:19)(cid:17)(cid:19)(cid:28)(cid:25) (cid:11)(cid:19)(cid:17)(cid:21)(cid:12) (cid:19)(cid:17)(cid:27)(cid:27) (cid:11)(cid:21)(cid:72)(cid:239)(cid:23)(cid:28)(cid:12) (cid:19)(cid:17)(cid:26)(cid:26) (cid:11)(cid:24)(cid:72)(cid:239)(cid:22)(cid:20)(cid:12) (cid:19)(cid:17)(cid:28)(cid:23) (cid:11)(cid:22)(cid:72)(cid:239)(cid:25)(cid:28)(cid:12) (cid:19)(cid:17)(cid:25)(cid:26) (cid:11)(cid:26)(cid:72)(cid:239)(cid:21)(cid:20)(cid:12) (cid:19)(cid:17)(cid:27)(cid:25) (cid:11)(cid:21)(cid:72)(cid:239)(cid:23)(cid:24)(cid:12) (cid:19)(cid:17)(cid:20)(cid:25) (cid:11)(cid:19)(cid:17)(cid:19)(cid:24)(cid:12) (cid:239)(cid:19)(cid:17)(cid:23)(cid:25) (cid:11)(cid:23)(cid:72)(cid:239)(cid:19)(cid:28)(cid:12) (cid:239)(cid:19)(cid:17)(cid:20)(cid:26) (cid:11)(cid:19)(cid:17)(cid:19)(cid:23)(cid:12) (cid:239)(cid:19)(cid:17)(cid:25)(cid:25) (cid:11)(cid:23)(cid:72)(cid:239)(cid:21)(cid:19)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:25) (cid:11)(cid:24)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) brown(cid:3) (cid:19)(cid:17)(cid:20)(cid:22) (cid:11)(cid:19)(cid:17)(cid:20)(cid:12) (cid:239)(cid:19)(cid:17)(cid:21)(cid:20) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:28)(cid:12) (cid:239)(cid:19)(cid:17)(cid:20)(cid:27) (cid:11)(cid:19)(cid:17)(cid:19)(cid:22)(cid:12) (cid:239)(cid:19)(cid:17)(cid:21)(cid:26) (cid:11)(cid:26)(cid:72)(cid:239)(cid:19)(cid:23)(cid:12) (cid:239)(cid:19)(cid:17)(cid:21)(cid:24) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:21)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:21) (cid:11)(cid:26)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:19)(cid:17)(cid:20)(cid:20) (cid:11)(cid:19)(cid:17)(cid:21)(cid:12) (cid:19)(cid:17)(cid:21)(cid:21) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:25)(cid:12) (cid:19)(cid:17)(cid:20)(cid:27) (cid:11)(cid:19)(cid:17)(cid:19)(cid:22)(cid:12) (cid:239)(cid:19)(cid:17)(cid:25)(cid:28) (cid:11)(cid:21)(cid:72)(cid:239)(cid:21)(cid:21)(cid:12) black(cid:3) (cid:239)(cid:19)(cid:17)(cid:20)(cid:22) (cid:11)(cid:19)(cid:17)(cid:20)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:26) (cid:11)(cid:22)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:21) (cid:11)(cid:25)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:27) (cid:11)(cid:21)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:27) (cid:11)(cid:21)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:24) (cid:11)(cid:27)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:19)(cid:17)(cid:21)(cid:25) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:20)(cid:12) (cid:19)(cid:17)(cid:21)(cid:25) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:20)(cid:12) (cid:19)(cid:17)(cid:24)(cid:23) (cid:11)(cid:28)(cid:72)(cid:239)(cid:20)(cid:22)(cid:12) (cid:239)(cid:19)(cid:17)(cid:19)(cid:19)(cid:20)(cid:21) (cid:11)(cid:20)(cid:12) (cid:19)(cid:17)(cid:24) (cid:19) (cid:239)(cid:19)(cid:17)(cid:24) red (cid:239)(cid:19)(cid:17)(cid:21)(cid:25) (cid:11)(cid:19)(cid:17)(cid:19)(cid:19)(cid:20)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:21) (cid:11)(cid:26)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:21) (cid:11)(cid:24)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:20) (cid:11)(cid:20)(cid:72)(cid:239)(cid:19)(cid:23)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:23) (cid:11)(cid:21)(cid:72)(cid:239)(cid:19)(cid:24)(cid:12) (cid:239)(cid:19)(cid:17)(cid:22)(cid:25) (cid:11)(cid:24)(cid:72)(cid:239)(cid:19)(cid:25)(cid:12) (cid:19)(cid:17)(cid:21) (cid:11)(cid:19)(cid:17)(cid:19)(cid:20)(cid:12) (cid:19)(cid:17)(cid:20)(cid:24) (cid:11)(cid:19)(cid:17)(cid:19)(cid:25)(cid:12) (cid:19)(cid:17)(cid:24)(cid:24) (cid:11)(cid:22)(cid:72)(cid:239)(cid:20)(cid:22)(cid:12) (cid:19)(cid:17)(cid:23)(cid:26) (cid:11)(cid:28)(cid:72)(cid:239)(cid:20)(cid:19)(cid:12) (cid:239)(cid:20) Chl Phosphate Nitrate+Nitrite Total Nitrogen Ammonia N:P (cid:49)(cid:13) 15N 13C Salinity Figure S3. Microbial community and environmental variables. Weighted Gene Correlation Network Analysis (WGCNA) was used to identify subnetworks (or modules) of bacteria that correlated with environmental variables. Pearson correlation coefficients for subnetworks that were significantly correlated with environmental variables are shown in the top number (ρ value) and the number in the parentheses is the p-value for each relationship. Positive relationship is in red and negative relationship is in blue. Related to Table 2 & 3. Module membership vs. Taxa significance cor=0.73, p=5.6e−10 A n e g o r t i N r o f e c n a c i i f i n g s a x a T 8 . 0 6 . 0 4 . 0 2 . 0 0 . 0 0.2 0.4 0.6 0.8 Module Membership in blue module C Module membership vs. Taxa significance cor=0.53, p=6.3e−05 n e g o r t i N r o f e c n a c i f i n g i s a x a T 5 . 0 4 . 0 3 . 0 2 . 0 1 . 0 0 . 0 B n e g o r t i N r o f e c n a c i f i n g s i a x a T y t i n i l a S r o f e c n a c i f i n g s a x a T i Module membership vs. Taxa significance cor=0.87, p=4.9e−11 9 0 . 7 0 . 5 0 . 3 0 . 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Module Membership in pink module D Module membership vs. Taxa significance cor=0.38, p=0.013 5 . 0 4 . 0 3 . 0 2 . 0 1 . 0 0.2 0.4 0.6 0.8 0.4 0.5 0.6 0.7 0.8 0.9 Module Membership in yellow module Module Membership in red module Figure S4. WGCNA modules. Module membership of taxa in the blue (A), pink (B) and yellow (C) subnetworks (or modules) which strongly correlated with ammonia and nitrate + nitrite. Module membership of taxa in red subnetwork (D) which strongly correlated with salinity. Related to Table 2 & 3. A B Kruskal−Wallis, p = 1.2e−10 * * 0.4 0.3 0.2 . . . . . 1 6 8 1 1 E e s a n e g o r t i N 0.1 low intermediate high . . . . . 4 9 9 7 1 E e s a t c u d e r . e t a r t i N . 4 . 1 . 7 . 1 . E H D A N . e s a t c u d e r . e t i r t i N 0.8 0.6 0.4 0.2 0.5 0.4 0.3 0.2 0.1 0.0 Kruskal−Wallis, p = 8.7e−14 * * low intermediate high Kruskal−Wallis, p = 2e−15 * * low intermediate ● high Figure S5. Metabolic pathways. (A) Contribution of top taxa from CCA ordination analysis and cos2 values. (B) Nitrogenase EC 1.18.6.1, Nitrate reductase EC 1.7.99.4 and Nitrite reductase NADH EC 1.7.1.4 normalized (Hellinger transformation) abundance. Kruskal-Wallis test and p-values with Dunn post-test, **denotes p-value<0.001. Related to Figure 6. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Supplementary Information Transparent Methods Study sites, sample collection, and physiochemical parameter measurements The study was conducted in two ecological reserves along the coast of Ecuador (Fig. 1). The Cayapas-Mataje Ecological Reserve, located within Esmeraldas province along the Colombian border (1° 17’ 02.14’’ N, 78° 54’ 22.29’’ W), encompasses 302.05 km2 of largely non-disturbed mangrove forests. This reserve is located in the delta formed by the estuary of the Cayapas- Santiago-Mataje rivers, and it is part of what used to be a continuous mangrove belt that ranged from the central area of the Colombia Pacific coast to the south area of Esmeraldas in Ecuador. Cayapas-Mataje is considered one of the most pristine mangrove ecosystems along the Pacific coast of the Americas (Hamilton, 2020a). The dominant mangrove species is Rhizophora mangle, representing 98% of all the mangrove area (Hamilton, 2020a). Traditional uses, such as artisanal fishing and cockle picking are still practiced, and only 2% of mangrove forest area loss is attributed to aquaculture (Hamilton, 2020b). The Muisne Ecological Reserve, also located in the province of Esmeraldas (0° 36’ 41.81’’ N, −80° 1’ 14.36’’ W), is highly perturbed by aquaculture (Fig. 1). The site compromises the delta of the Muisne River and numerous smaller rivers and contains a total of 12.06 km2 of mangrove forests. The species composition is 71% Rhizophora mangle, 1% Avicennia germinans, and 28% Languncularia racemose (Hamilton, 2020a). Muisne has been severely impacted by shrimp aquaculture, accounting for 36% of mangrove loss. Only 1% of the remaining mangrove forest is protected (Hamilton, 2020b). 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 Water samples were taken from the surface (0.5 m depth) along a proximity gradient to the mangrove trees. Samples were grouped by level of disturbance based on the concentration of ammonia in the water column: Low = < 1 μM, Intermediate = 1 – 3 μM, High = > 3 μM. Similar ammonia ranges have been identified in previous studies exposed to aquaculture effluent (Robertson and Alongi, 1992); however, reported values in the literature can vary depending on spatial parameters and aquaculture land expansion (Cifuentes et al., 1996; Barraza-Guardado et al., 2013a, Samocha et al., 2004). Here we also take into account the area of shrimp aquaculture in the two ecological sites. Muisne was identified as highly disturbed, and all the samples were taken near shrimp aquaculture facilities (N = 29) with high levels of ammonia with the exception of two samples near the mouth of the estuarine. The site has 20.47 km2 of shrimp farms and 12.06 km2 of mangrove forests for an approximate 2:1 ratio of aquaculture to mangrove forest (Hamilton, 2020b). Cayapas-Mataje has 11.04 km2 of shrimp aquaculture farms and 302.05 km2 of mangrove forest for a 1:27 ratio of aquaculture to mangrove forest (Fig. 1) (Hamilton, 2020b). Thus, samples that were collected along mangrove forests in Cayapas-Mataje (no presence of aquaculture) were characterized as a low disturbance with lower levels of ammonia, and we included one sample from Muisne with low level of ammonia (N = 89). Within Cayapas-Mataje, there’s a smaller presence of shrimp aquaculture facilities and the samples that were collected near the shrimp facilities were classified as intermediate disturbance with intermediate levels of ammonia in addition to one sample from Muisne with intermediate ammonia (N = 34). For DNA samples, approximately 400 ml of water was filtered through a sterile 47 mm 0.2 µm Supor filter (Pall) directly from 0.5 m depth using a peristaltic pump. The filter was immediately stored on ice and transferred to long term storage at –80 °C within 8 hours. Chlorophyll a concentration was measured with an Aquaflash handheld active fluorometer (Turner 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 Designs) following the manufacturer’s instructions. Temperature, salinity, and turbidity were measured using a YSI ProDss (Xylem). For nutrient analysis, 50 ml of water was filtered through a combusted GF/F filter (Whatman), frozen immediately after collection, and stored at –80 °C. Samples were sent to the UC Santa Barbara Marine Institute and analyzed by flow injection analysis following standard protocols ( Lachat instrument methods: 31-107-04-1A, 31-107-06-5A, 31-115-01-3A). For CHN and isotope analysis, 50 ml of water was filtered through a combusted GF/F filter, and filters were wrapped into a tin envelope (Costech). Samples were analyzed by EA-IRMS at the Scripps Institute of Oceanography Isotope Facility yielding percent carbon and nitrogen by mass, as well as δ13C and δ15N following standard methods (Pestle, Crowley and Weirauch, 2014). The reference materials used were NBS-19 and NBS-18, and IAEA N1 and the analytical precisions were +/- 0.3 to 0.5 for C and 0.7 to 1.0 for N. The standards used for δ13C and δ15N calculations were the Pee Dee Belemnite and atmospheric N2, respectively. DNA extraction and sequencing DNA was extracted using the DNAeasy PowerWater DNA extraction kit (Qiagen). Extracted DNA was quantified using the Qubit HS DNA quantification kit (Invitrogen) and then quality checked by gel electrophoresis and PCR amplification of the 16S rRNA gene using primers 515F and 806R (Walters et al., 2015) for bacteria and archaea. High quality extracted DNA was submitted to the Argonne National Laboratory sequencing center for amplification and library preparation with the same primer set, followed by 2 x 151 paired-end sequenced on the Illumina Miseq platform. Sequences were submitted to NCBI Bio project accession number: PRJNA633714. 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 Sequence analysis Illumina Miseq reads were demultiplexed using the ‘iu-demultiplex’ command in Illumina utils. Demultiplexed reads were quality controlled and denoised using the ‘FilterandTrim’ and ‘dada’ commands within the R package dada2 (Benjamin J Callahan et al., 2016), and assembled with the ‘mergePairs’ command. The final merged reads had mean quality scores >30, and the non-redundant fasta files of the generated unique reads produced by dada2 were used as an input for the paprica pipeline for microbial community structure and metabolic inference (https://github.com/bowmanjeffs/paprica). The paprica method for determining microbial community structure differs from OTU clustering methods in that it relies on the placement of reads on a phylogenetic tree created from the 16S rRNA gene reads from all completed bacterial and archaeal genomes in Genbank (Bowman and Ducklow, 2015). Because the metabolic potential of each phylogenetic edge on the reference tree is known, paprica generates a reasonable estimate of genome sizes, gene content, and metabolic pathways for the organisms of origin of each read. To estimate metabolic potential, a phylogenetic tree of the 16S rRNA genes from each completed genome was generated. For each internal node on the reference tree we determined a “consensus genome”, defined as all genomes shared by all members of the clade originating from the node, and predict the metabolic pathways present in the consensus and complete genomes (Bowman and Ducklow, 2015). Unique sequences (referred to as amplicon sequence variants or ASVs) and estimated gene abundances were normalized according to predicted 16S rRNA gene copy number prior to downstream analysis. The paprica community structure results are described in terms of closest estimated genomes (CEGs; for phylogenetic placements to non-terminal edges) and closest completed genomes (CCGs; for placements to terminal edges). CCGs are names according to their 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 lowest consensus taxonomic ranking, while CEGs are named according to their closest relative on the phylogenetic reference tree. Diversity and statistics analysis The alpha diversity index, inverse of Simpson, for ASVs was calculated using the phyloseq package in R (McMurdie and Holmes, 2013) following methods described in Callahan (Ben J. Callahan et al., 2016). Kruskal-Wallis tests were performed to test differences among groups in the vegan package in R (Oksanen et al., 2019). For the biogeochemical parameters, we used the Kruskal-Wallis test to test differences among groups, and the Spearman correlation to evaluate relationships between N:P ratio, genome size, and 16S rRNA gene copy number. We determined N in disturbed and less disturbed sites; this is a measure of nitrogen vs. phosphorus availability based on the Redfield ratio (N:P = 16:1) (Gruber and Sarmiento, 1997), and we calculated based on nutrient concentrations using the following equation (Wilson, Abboud and Beman, 2017): (1): 𝑁 ∗ = (𝑁𝑂! " + 𝑁𝑂 $ " + 𝑁𝐻% &) − 16 × 𝑃𝑂% !" We used correspondence analysis (CA) to quantify taxon contributions to the sample ordination. This method allowed us to determine the degree of correspondence between sites and species, and which taxa were associated with gradients of disturbance. We performed CA on Hellinger- transformed data such that each value represents a contribution to the Pearson's χ2 (chi-squared) statistic computed for the data (Legendre P., 1998). We also calculated a cos2 value that describes the contribution of each taxa to the major axes of disturbance (Kuramae et al., 2012). Analysis of similarity (ANOSIM) was used to assess significant differences with respect to level of disturbance. This nonparametric permutation procedure uses the rank similarity matrix underlying 114 115 116 117 118 119 120 121 122 an ordination plot to calculate an R test statistic, and it was calculated using the vegan package in R (Oksanen et al., 2019). We examined association of levels of disturbance by a Spearman correlation between ammonia concentrations and dimensions 1 and 2 of CA analysis. To examine the impact of environmental variables associated to aquaculture outflow on the estimated metabolic pathways we performed a canonical correspondence analysis (CCA) to the metabolic output generated in paprica to restrict the sample ordination to nitrogen, phosphate, and isotopic signals to better understand the impact of aquaculture outflow on microbial metabolic potential. The cos2 value was used to determine the contribution of key metabolic pathways to the major axis. The ordinations were performed in R using the factoMiner and CA package (Husson et al., 123 2020). 124 125 126 127 128 129 130 131 132 133 134 135 136 To identify unique reads differentially present between disturbed and non-disturbed sites we used DESeq2 (Michael I Love, Huber and Anders, 2014), following the methods of Webb et al. (2019). DESeq2 performs differential abundance analysis based on the negative binomial/Gamma-Poisson distribution. The default settings were used, which estimates size factors with the median ratio method (Michael I. Love, Huber and Anders, 2014), followed by estimation of dispersion. Next, a Wald test for generalized linear model coefficients was used to test for significance of coefficients, considering size factors and dispersion. The p-values were attained by the Wald test and corrected for multiple testing using the Benjamini and Hochberg method (Michael I. Love, Huber and Anders, 2014). The most abundant bacterial and archaeal taxa that were significantly differentially present were further examined to identify potential microbial markers of shrimp aquaculture effluent. To determine the role of differentially abundant microbes in nutrient cycling, we utilized the BioCyc database (Karp et al., 2019) in combination with the paprica output to assess the potential for genes coding enzymes associated with nitrogen fixation and denitrification. Enzymes included with our assessment included: nitrogenase; EC 1.18.6.1, EC 1,19.6.1, nitrate reductase; EC 17.99.4, EC 1.7.1.1, EC 1.7.1.2, EC 1.9.6.1, EC 1.7.2.2, and nitrite reductase; EC1.7.2.1, EC 1.7.2.2, EC 1.7.1.4. To identify modules of highly correlated taxa we used Weighted Gene Correlation Network Analysis (WGCNA) (Langfelder and Horvath, 2008), following the methods of Wilson et al. (2018). A signed adjacency measure for each pair of features (unique reads) was calculated by raising the absolute value of their Pearson correlation coefficient to the power of parameter p. The value p = 8 was used for each global network to optimize the scale-free topology network fit. This power allows the weighted correlation network to show a scale free topology where key nodes are highly connected with others. The obtained adjacency matrix was then used to calculate the topological overlap measure (TOM), which, for each pair of features, considers their weighted pairwise correlation (direct relationships) and their weighted correlations with other features in the network (indirect relationships). For identifying subnetworks or ‘modules’ a hierarchical clustering was performed using a distance based on the TOM measure. This resulted in the definition of several subnetworks, each represented by its first principal component. A subnetwork is the association between the subnetworks and a given trait that is measured by the pairwise relationships (correlations) between the taxa. To find correlations between subnetworks and environmental factors, Pearson’s correlation coefficients were calculated between the considered environmental factor and the respective principal components. P-values were adjusted using Bonferroni method. All procedures were applied to Hellinger-transformed abundances. 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 References 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 Barraza-Guardado, R. H. et al. (2013) ‘Effluents of shrimp farms and its influence on the coastal ecosystems of Bahía de Kino, Mexico.’, TheScientificWorldJournal, 2013, p. 306370. Bowman, J. S. and Ducklow, H. W. (2015) ‘Microbial Communities Can Be Described by Metabolic Structure: A General Framework and Application to a Seasonally Variable, Depth- Stratified Microbial Community from the Coastal West Antarctic Peninsula’, PLOS ONE. Edited by C. Moissl-Eichinger, 10(8), p. e0135868. Callahan, Ben J. et al. (2016) ‘Bioconductor workflow for microbiome data analysis: From raw reads to community analyses [version 1; referees: 3 approved]’, F1000Research, 5. Callahan, Benjamin J et al. (2016) ‘DADA2: High-resolution sample inference from Illumina amplicon data.’, Nature methods, 13(7), pp. 581–3. Cifuentes, L. A. et al. (1996) ‘Isotopic and Elemental Variations of Carbon and Nitrogen in a Mangrove Estuary’, Estuarine, Coastal and Shelf Science, 43(6), pp. 781–800. Gruber, N. and Sarmiento, J. L. (1997) ‘Global patterns of marine nitrogen fixation and denitrification’, Global Biogeochemical Cycles, 11(2), pp. 235–266. Hamilton, S. E. (2020a) ‘Introduction to Coastal Ecuador’, in Coastal Research Library. Springer, pp. 69–110. Hamilton, S. E. (2020b) ‘Assessing 50 Years of Mangrove Forest Loss Along the Pacific Coast of Ecuador: A Remote Sensing Synthesis’, in Coastal Research Library. Springer, pp. 111–137. Husson, F. et al. (2020) Package ‘FactoMineR’ Title Multivariate Exploratory Data Analysis and Data Mining. [Online] Available at: http://factominer.free.fr Karp, P. D. et al. (2019) ‘The BioCyc collection of microbial genomes and metabolic pathways’, 20(4), pp. 1085–1093. Kuramae, E. E. et al. (2012) ‘Soil characteristics more strongly influence soil bacterial communities than land-use type’, FEMS Microbiology Ecology, 79(1), pp. 12–24. Langfelder, P. and Horvath, S. (2008) ‘WGCNA: an R package for weighted correlation network analysis.’, BMC bioinformatics, 9, p. 559. Legendre P., L. L. F. J. (1998) Numerical Ecology, Volume 24 - 2nd Edition, Elsevier Science. Love, Michael I, Huber, W. and Anders, S. (2014) ‘Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2’, Genome Biology, 15(12), p. 550. McMurdie, P. J. and Holmes, S. (2013) ‘phyloseq: An R Package for Reproducible Interactive Analysis and Graphics of Microbiome Census Data’, PLoS ONE. Edited by M. Watson, 8(4), p. e61217. Oksanen, J. et al. (2019) Package ‘vegan’ Title Community Ecology Package Version 2.0-8. [Online] Available at: http://CRAN.R-project.org/package=vegan Pestle, W. J., Crowley, B. E. and Weirauch, M. T. (2014) ‘Quantifying Inter-Laboratory Variability in Stable Isotope Analysis of Ancient Skeletal Remains’, PLoS ONE. Edited by L. Bondioli, 9(7), p. e102844. Robertson, A. I. and Alongi, D. M. (eds) (1992) Tropical Mangrove Ecosystems. Washington, D. C.: American Geophysical Union (Coastal and Estuarine Studies). Samocha, T. M. et al. (2004) ‘Characterization of intake and effluent waters from intensive and semi-intensive shrimp farms in Texas’, Aquaculture Research, 35(4), pp. 321–339. Walters, W. et al. (2015) ‘Transcribed Spacer Marker Gene Primers for Microbial Community Surveys’, mSystems, 1(1), pp. e0009-15. Webb, S. J. et al. (2019) ‘Impacts of Zostera eelgrasses on microbial community structure in San Diego coastal waters’, Elem Sci Anth, 7(1), p. 11. Wilson, J., Abboud, S. and Beman, J. M. (2017) ‘Primary Production, Community Respiration, and Net Community Production along Oxygen and Nutrient Gradients: Environmental Controls and Biogeochemical Feedbacks within and across “Marine Lakes”’, Frontiers in Marine Science, 4. Wilson, J. M., Litvin, S. Y. and Beman, J. M. (2018) ‘Microbial community networks associated with variations in community respiration rates during upwelling in nearshore Monterey Bay, California’, Environmental Microbiology Reports, 10(3), pp. 272–282. 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
10.1016_j.jbc.2023.105109	RESEARCH ARTICLE Structural insight into G-protein chaperone-mediated maturation of a bacterial adenosylcobalamin-dependent mutase , Daphne A. Faber2 Received for publication, May 23, 2023, and in revised form, July 20, 2023 Published, Papers in Press, July 28, 2023, https://doi.org/10.1016/j.jbc.2023.105109 Francesca A. Vaccaro1 Dallas R. Fonseca5 From the 1Department of Chemistry, and 2Department of Biological Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA; 3Graduate Program in Biophysics, Harvard University, Cambridge, Massachusetts, USA; 4Department of Biology, 5Amgen Scholar Program, and 6Howard Hughes Medical Institute, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA , and Catherine L. Drennan1,4,6,* , Gyunghoon Kang1 , Gisele A. Andree1 , David A. Born3,4 , Marco Jost1 , Reviewed by members of the JBC Editorial Board. Edited by Joan B. Broderick G-protein metallochaperones are essential for the proper maturation of numerous metalloenzymes. The G-protein chaperone MMAA in humans (MeaB in bacteria) uses GTP hydrolysis to facilitate the delivery of adenosylcobalamin (AdoCbl) to AdoCbl-dependent methylmalonyl-CoA mutase, an essential metabolic enzyme. This G-protein chaperone also facilitates the removal of damaged cobalamin (Cbl) for repair. Although most chaperones are standalone proteins, isobutyryl- CoA mutase fused (IcmF) has a G-protein domain covalently attached to its target mutase. We previously showed that dimeric MeaB undergoes a 180(cid:1) rotation to reach a state capable of GTP hydrolysis (an active G-protein state), in which so-called switch III residues of one protomer contact the G- nucleotide of the other protomer. However, it was unclear whether other G-protein chaperones also adopted this conformation. Here, we show that the G-protein domain in a fused system forms a similar active conformation, requiring IcmF oligomerizes both upon Cbl IcmF oligomerization. damage and in the presence of the nonhydrolyzable GTP analog, guanosine-5’-[(β,γ)-methyleno]triphosphate, forming supramolecular complexes observable by mass photometry and EM. Cryo-EM structural analysis reveals that the second pro- tomer of the G-protein intermolecular dimer props open the mutase active site using residues of switch III as a wedge, allowing for AdoCbl insertion or damaged Cbl removal. With the series of structural snapshots now available, we now describe here the molecular basis of G-protein–assisted AdoCbl-dependent mutase maturation, explaining how GTP binding prepares a mutase for cofactor delivery and how GTP hydrolysis allows the mutase to capture the cofactor. * For correspondence: Catherine L. Drennan, [email protected]. Present addresses for: David A. Born, Beam Therapeutics, Cambridge, Massachusetts, 02142, USA; Gyunghoon Kang, NanoImaging Services, 4940 Carroll Canyon Road, San Diego, California, 92121, USA; Dallas R. Fonseca, Department of Plant and Microbial Biology, University of Min- nesota, Twin Cities, St Paul, Minnesota, USA; Marco Jost, Department of Microbiology, Harvard Medical School, Boston, Massachusetts, 02115, USA. for With 30 to 50% of the proteome predicted to be metal- loproteins, proper maturation of metalloproteins is a vital and nontrivial biological process (1). Metallochaperones are essential the correct maturation of metalloproteins, ensuring that valuable metallocofactors are delivered efﬁ- ciently and with minimal toxicity and degradation (2–4). A prominent class of metallochaperones is composed of the guanine nucleotide–binding proteins (G-proteins) belonging to the SIMIBI (signal recognition particle, MinD, and BioD) class of P-loop NTPases (5). In the absence of their target protein, G-protein metallochaperones generally have low GTP hydrolysis activity. However, target protein binding stimulates GTPase activity by (cid:3)100-fold (6, 7). Some members of this class, such as UreG and HypB, directly bind their metal- locofactors and then use GTP hydrolysis to facilitate the maturation of their targets, urease and hydrogenase, respec- tively (8–11). Others, like MeaB (MMAA in humans), do not bind directly to their metallocofactor, which is coenzyme B12 (50-deoxyadenosylcobalamin or AdoCbl) in the case of MeaB/ MMAA, but still use GTP hydrolysis to deliver AdoCbl to the enzyme target, methylmalonyl-CoA mutase (MCM) (7, 12). In particular, an adenosyltransferase (ATR) adenylates cob(II) alamin using ATP, and MeaB/MMAA facilitates the transfer of AdoCbl from ATR to the cobalamin(Cbl)-binding domain of MCM (13, 14). Typically, G-protein metallochaperones are standalone proteins, with the notable exception of AdoCbl- dependent isobutyryl-CoA mutase fused (IcmF), in which the G-protein metallochaperone exists as a domain of the target enzyme (15). Mutations or deletions in the genes encoding metallochaperones can impair metalloprotein function in vivo and lead to disease in humans. For example, in humans, mu- tations to the genes for MCM, MMAA, or any other chaper- ones involved in B12 trafﬁcking can result in methylmalonic aciduria, an inborn error of metabolism (14, 16). The AdoCbl cofactor is essential for the chemically chal- lenging carbon skeletal rearrangements that are performed by mutases (17). For MCM, the radical reservoir of the cobalt– carbon bond of AdoCbl catalyzes the 1,2-rearrangement of (R)-methylmalonyl-CoA to succinyl-CoA (Fig. 1A) (18). IcmF J. Biol. Chem. (2023) 299(9) 105109 1 © 2023 THE AUTHORS. Published by Elsevier Inc on behalf of American Society for Biochemistry and Molecular Biology. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Role of G-protein dimerization in metalloenzyme maturation Figure 1. Adenosylcobalamin-dependent mutase reactions and structures. A, AdoCbl-dependent mutase reactions. B, methylmalonyl-CoA mutase (PDB 4REQ) (26) consists of a Cbl-binding domain (yellow ribbons) and substrate-binding domain (green ribbons) with the AdoCbl cofactor (red sticks) binding at the interface. The inactive subunit consists of a substrate-binding domain (light green ribbons) and Cbl-binding domain (tan ribbons) which cannot perform catalysis. Cartoon representation of MCM is colored according to the ribbon drawings. C, the G-protein chaperone MeaB (PDB 8DPB) (33) (purple ribbons) has conserved P-loop GTPase motifs as labeled and an additional switch III motif (red orange) implicated in GTP hydrolysis. Inset: GMPPCP and the Mg2+ ion interact with the phosphate-binding loop (purple sticks: K68 and S69), switch I region (yellow sticks: D92, D105, R108), switch II region (green sticks: E154), and the switch III region from the other protomer (red-orange sticks: K188, Q185, D182). Residues labeled with “B” are from the other protomer. D, MeaB conformational change. Inactive state: PDB 2QM7 (32); Active state: PDB 8DPB (33). Switch III region of the light purple MeaB protomer (chain B) rearranges from being solvent-exposed in the “inactive state” to contacting GMPPCP in the “active state”. E, Cupriavidus metallidurans IcmF (PDB 4XC6) (29) contains two protomers each consisting of a G-protein domain (blue) and mutase domain (substrate-binding domain (green) and Cbl-binding domain (orange)) connected by a linker (pink). AdoCbl (red sticks) is bound at the active site which is located at the interface of the Cbl-binding domain and substrate-binding domain. The switch III residues are highlighted in red-orange. AdoCbl, adenosylcobalamin; Cbl, cobalamin; GMPPCP, guanosine-5’-[(β,γ)-methyleno] triphosphate; IcmF, isobutyryl-CoA mutase fused; MCM, methylmalonyl-CoA mutase. isobutyryl-CoA to n- catalyzes the 1,2-rearrangement of butyryl-CoA, as well as pivalyl-CoA and isovaleryl-CoA, using AdoCbl (Fig. 1A) (15, 19, 20). The homolytic cleavage of the cobalt–carbon bond of AdoCbl generates cob(II)alamin and a highly reactive 50-deoxyadenosyl radical species (21). After catalysis, these species must come together to regenerate 2 J. Biol. Chem. (2023) 299(9) 105109 Role of G-protein dimerization in metalloenzyme maturation AdoCbl (Fig. S1); however, if the 50-deoxyadenosine moiety is lost and/or the cob(II)alamin species is oxidized to hydrox- ocobalamin (OHCbl), the reformation of AdoCbl is prevented, inactivating the enzyme (20). In addition to these, AdoCbl is also susceptible to inactivation by photolysis of the cobalt– carbon bond (22). To restore activity, the G-protein metal- lochaperone facilitates the removal of the damaged cofactor and the insertion of a new cofactor from ATR (12, 23, 24). In the active conformation of all known AdoCbl-dependent is positioned at the interface of a Cbl- mutases, AdoCbl binding Rossmann domain and a substrate-binding TIM bar- to afford the generation of substrate-radical species rel (Figs. 1B and S1) (25–27). In this active state, AdoCbl is sequestered. Thus, AdoCbl delivery to the mutase requires a transient opening and closing of the mutase structure, a conformational change that is believed to be facilitated by MeaB/MMAA (24, 28, 29). Consistent with their classiﬁcations as P-loop NTPases, MeaB/MMAA and the G-protein domain of IcmF all have a P- loop, a base-speciﬁcity loop, and two conserved switch regions for signal transduction: switch I (residues 92–108 in MeaB) and switch II (residues 154–158 in MeaB) (Fig. 1C) (6, 30–32). Unique to MeaB and MMAA is a so-called switch III region (31), which was identiﬁed through investigation of residues associated with methylmalonic aciduria (16). In vitro, the substitution of MeaB switch III residues Lys188, Gln185, or Asp182 with alanine (Fig. 1C) reduces the stimulatory effect in GTP hydrolysis afforded by mutase binding, that the GTPase-accelerating protein (GAP) activity, and also leads to an uncoupling of GTP hydrolysis from AdoCbl transfer (31). In the absence of its target mutase, the switch III region of MeaB points toward solution (Fig. 1D). However, following the binding of MeaB to the Cbl-binding domain of MCM in the presence of a nonhydrolyzable analog of GTP, guanosine-5’- [(β,γ)-methyleno]triphosphate (GMPPCP), MeaB undergoes a 180(cid:1) rotation that results in the repositioning of switch III residues directly into the GTP-binding site (Fig. 1D) (33). Importantly, it is the switch III residues of the neighboring protomer that contact GTP in this active conformation, indi- cating the importance of the dimeric structure of MeaB to its function (Fig. 1, C and D) (33). is, The recent structure of GMPPCP-MeaB bound to the Cbl- binding domain of MCM (GMPPCP-MeaB:MCMCbl) led to a proposed molecular mechanism for MeaB function (Fig. 2A) (33). In this proposal, the association of GTP-MeaB with the Cbl-binding domain of MCM leads to a conformational change of MeaB from an inactive to active state. The MeaB active state stabilizes an open MCM conformation, allowing MCM to receive AdoCbl from ATR. GTP hydrolysis causes MeaB to undergo a conformational change back to the inactive state. This conformational change of MeaB is proposed to destabilize the open MCM conformation, causing MCM to close and capture AdoCbl inside of the enzyme (Fig. 2A). This proposed mechanism (Fig. 2A) is supported by crystal structures of GDP-MeaB alone and GMPPCP-MeaB in the presence of MCMCbl that together reveal the conformational gymnastics involved in the conversion of inactive and active MeaB states (Fig. 1D) (32, 33). The connection between GTP hydrolysis and AdoCbl capture, displayed in Figure 2A (IV to V), is also supported by biochemical and structural data. Brieﬂy, MCM is unable to capture AdoCbl in the presence of a nonhydrolyzable GTP analog, consistent with the proposal in Figure 2A that a GTP hydrolysis–driven MeaB confor- mational change is needed for MCM to close and thus capture AdoCbl (19, 34). Furthermore, when the active state of MeaB is destabilized by substitutions of switch III resi- dues, such that MeaB can convert to the inactive state in- dependent of GTP hydrolysis, AdoCbl capture is uncoupled from MeaB’s GTPase activity (28, 31). What is missing in terms of experimental support of Figure 2A are structures of MCM and MeaB showing that an active MeaB conformation stabilizes an open MCM state in which the Cbl-binding domain of the mutase is positioned away from the substrate-binding domain (I, III, IV in Fig. 2A). There is no structure of MeaB bound to an intact MCM; only a structure of MeaB is bound to the Cbl-binding domain of MCM (MCMCbl). Thus, part of in Figure 2A is based on structural superpositions of GMPPCP- MeaB:MCMCbl with the crystal structure of IcmF from Cupriavidus metallidurans (29), which contains all relevant structural units: the metallochaperone G-protein domain, the Cbl-binding domain, and the substrate-binding TIM barrel and shares a high level of structural similarity (Figs. 1E, S2 and S3). There is, however, the question as to whether IcmF is a good model system for a dimeric MeaB:MCM complex given that the G-protein domains of IcmF are monomeric and located on opposite sides of this large enzyme structure (Fig. 1E). IcmF is a dimer, but the dimeric interface is not composed of the G-protein domains (Fig. 1E). Thus, if dimerization of the G-protein domains is essential for AdoCbl delivery in IcmF, as it is for MeaB, then IcmF would need to form transient higher-order oligomers (Fig. 2B). If oligomers do not form, then IcmF must use a distinct molecular mechanism from MeaB and not what is shown in Figure 2B. Also, if IcmF is a good model system for MeaB:MCM, then IcmF should also employ a switch III, a question that has not been investigated. the mechanistic proposal By system. employing In this study, we investigate whether IcmF, a fused G-pro- tein system, uses the same molecular mechanism as a stand- alone G-protein site-directed mutagenesis and enzyme assays, we establish that switch III residues are relevant in IcmF from C. metallidurans, and by employing negative stain EM and mass photometry, we show that IcmF oligomerizes as would be expected for a conserved molecular mechanism. With support for a conserved mecha- nism, we go on to use IcmF to obtain the missing structural snapshot, a mutase bound to a G-protein in a state competent for AdoCbl transfer. With these new data, we describe a consensus molecular mechanism for metallochaperone- assisted AdoCbl-dependent mutase maturation. J. Biol. Chem. (2023) 299(9) 105109 3 Role of G-protein dimerization in metalloenzyme maturation Figure 2. Proposed steps for loading AdoCbl into the mutase active site. A, GTP binding to MeaB leads to the formation of the MeaB “active state”. This conformational change of MeaB opens up one subunit of the MCM heterodimer for AdoCbl delivery from ATR. GTP hydrolysis returns MeaB to the “inactive state,” closing the MCM subunit and capturing AdoCbl. B, GTP binding to the G-protein domain of IcmF leads to the formation of a higher order oligomeric state of IcmF, generating a dimer interface analogous to the “active state” of MeaB. Oligomerization of IcmF opens one promoter of the IcmF homodimer for AdoCbl delivery from ATR. GTP hydrolysis breaks apart the higher order oligomer of IcmF, closing the IcmF protomer and trapping the AdoCbl. AdoCbl, adenosylcobalamin; ATR, adenosyltransferase; IcmF, isobutyryl-CoA mutase fused; MCM, methylmalonyl-CoA mutase. Results Substitutions of the switch III regions of IcmF decrease GTPase activity, establishing the relevance of switch III in the IcmF system To assess if the residues in IcmF that are analogous to the switch III residues of MeaB affect GTPase activity of IcmF as they do in MeaB, we substituted Q341 (Q185 in MeaB) and K344 (K188 in MeaB) with alanine residues and measured GTPase activity (Table 1, Figs. 3 and S3) (31). The GTPase −1) activity of C. metallidurans wt IcmF (kcat = 3.13 ± 0.18 min is comparable to previously reported values for IcmF from C. metallidurans and Geobacillus kaustophilus (Table 1) (15, 20). Substituting the switch III residue Q341 with alanine lowers the catalytic efﬁciency 2.6-fold mainly through a decrease in kcat (Table 1). Substituting the switch III residue K344 with alanine lowers the GTPase activity to undetectable levels (Table 1). The observed decrease in GTPase activity of the Q341A and K344A variants is consistent with the observed decrease in the GTPase activity of MeaB switch III variants, validating the importance of the switch III residues in the fused system (31). IcmF forms higher order oligomers in the presence of GTP and nonhydrolyzable GTP analogs To understand if the fused system also utilizes a G-protein dimer arrangement similar to MeaB and other members of the SIMIBI G-proteins (5), we analyzed the oligomeric state of C. metallidurans wt IcmF in solution in the presence and absence of various G-nucleotides. If the fused system’s active state is comparable to the nonfused system, IcmF should form a higher order oligomeric state in the presence of GTP or a nonhydrolyzable analog due to the association of one G-pro- tein domain of an IcmF molecule with the G-protein domain of another. GTP hydrolysis should break apart these higher order oligomers, returning IcmF to the dimeric state that was visualized in the crystal structures (Fig. 1E) (29, 35). Thus, we would expect to see more higher-order oligomers with GMPPCP, a nonhydrolyzable analog of GTP. Additionally, we would expect to see only dimers under inactive GTPase con- ditions, that is, with no nucleotide or with GDP. Importantly, our negative stain EM and mass photometry data are consis- tent with these predictions (Fig. 4). Negative stain EM analysis indicates that with no nucleotide or in the presence of 500 μM 4 J. Biol. Chem. (2023) 299(9) 105109 Role of G-protein dimerization in metalloenzyme maturation Table 1 Kinetic parameters for GTPase activity of wt and switch III variants of IcmF and MeaB Enzyme IcmFa Geobacillus kaustophilus IcmF MeaB MeaB + MCM Q341A IcmF Q185A MeaB + MCM K344A IcmF K188A MeaB + MCM Km GTP μM 2.4 ± 0.9 40 ± 8 51 ± 3 N.R. N.R. 2.2 ± 0.8 N.R. N.D.c N.R. Vmax μM/min 1.568 N.R.b N.R. N.R. N.R. 0.533 N.R. N.D. N.R. kcat −1 min 3.13 ± 0.18 18 ± 1.3 10 ± 1 0.039 ± 0.003 4.20 ± 0.21 1.07 ± 0.07 0.17 ± 0.03 N.D. 0.14 ± 0.02 a Unless noted, all IcmF GTPase activity data is from Cupriavidus metallidurans. b N.R.: Not reported. c N.D.: Not detected. Catalytic efﬁciency kcat/Km μM min −1 1.3 0.45 0.19 N. R. N.R. 0.49 N.R. N.D. N.R. Reference This study (20) (20) (31) (31) This study (31) This study (31) GDP, IcmF does not form any observable higher-order oligo- mers and remains dimeric (Fig. 4, A and B). In the presence of 500 μM GTP, the percentage of wt IcmF in a dimeric state decreases and higher order oligomers are observed (Fig. 4C). With 500 μM of the nonhydrolyzable GTP analog, GMPPCP, long chains of IcmF protomers are observed, with all visible IcmF protomers comprising these ﬁlamentous-like chains (Fig. 4D). These supramolecular structures are consistent with IcmF dimers interacting with other IcmF dimers through the surface-exposed G-protein domains on either end of the long IcmF molecule (see Fig. 1E). Using mass photometry to evaluate the oligomers, we also ﬁnd that in the presence of 500 μM GDP, the primary state of IcmF is dimeric (80%), consistent with the negative stain EM (Fig. 4E) and the crystal structure (Fig. 1E) (29). In the presence of 500 μM GTP, the percentage of IcmF in a dimeric state is decreased from 80% to 60% and higher order oligo- mers are observed (22% of particles are tetramers and 7.5% of particles are hexamers). The dimensions and heterogeneity in length of the GMPPCP-generated IcmF supramolecular complexes precluded accurate quantiﬁcation by mass photometry. The identity of the Cbl cofactor inﬂuences IcmF’s oligomeric state According to the mechanism shown in Figure 2B, we expect that the G-protein domain of IcmF only dimerizes when IcmF needs to open for removal of a damaged cofactor or installa- tion of a new one. Thus, we investigated whether a correlation exists between the identity of the Cbl cofactor (active AdoCbl or inactive cob(II)alamin or OHCbl) and the oligomeric state of IcmF. We employed negative stain EM exclusively for our studies with Cbl, as the Cbl visual spectra interfered with the light scattering method utilized in mass photometry, pre- venting the quantitative analysis of oligomeric state by mass photometry that was possible for GDP and GTP. Negative stain EM grids were prepared both under red light (“dark”) and under white light (“light”). Samples prepared under red light should have intact AdoCbl, and experiments conducted in the light should have inactivated OHCbl cofactor due to photolysis of AdoCbl and oxidation of the resulting cob(II)alamin to OHCbl. We also used OHCbl in these studies, which represent the fully inactivated and oxidized AdoCbl degradation product. We ﬁnd that with no nucleotide or with GDP that the in dark or light or identity of the Cbl present (AdoCbl Figure 3. Michaelis–Menten kinetic analysis of wt IcmF and switch III variant Q341A IcmF. A, the GTP hydrolysis activity of wt IcmF. B, the GTP hy- drolysis activity of the Q341A IcmF variant is lower than that of wt IcmF. Each reaction was performed with 0.5 μM of the enzyme. The data for each curve represent the average ± S.D. of at least three replicates. IcmF, isobutyryl-CoA mutase fused. J. Biol. Chem. (2023) 299(9) 105109 5 Role of G-protein dimerization in metalloenzyme maturation Figure 4. Negative stain EM and mass photometry analysis of the oligomeric state of wt IcmF in the absence and presence of G-nucleotides. A, negative stain EM image of 20 ng/μl wt IcmF without any nucleotide. Representative IcmF dimer is boxed in tan. B, 20 ng/μl wt IcmF in the presence of 500 μM of GDP. Representative IcmF dimer is boxed in tan. C, 20 ng/μl wt IcmF in the presence of 500 μM GTP. Representative IcmF dimers are boxed in tan, and supramolecular complexes are circled in teal. D, 20 ng/μl wt IcmF in the presence of 500 μM GMPPCP. A representative supramolecular complex is circled in teal. E, mass photometry analysis of wt IcmF without any nucleotide or in the presence of 500 μM GDP or 500 μM GTP indicate the presence of a greater proportion of higher order oligomers in the presence of GTP. The data for each condition represents the mean ± S.D of at least four replicates. The dimensions and heterogeneity in length of the GMPPCP-generated IcmF supramolecular complexes precluded accurate quantiﬁcation by mass photometry. GMPPCP, guanosine-5’-[(β,γ)-methyleno]triphosphate; IcmF, isobutyryl-CoA mutase fused. OHCbl) does not alter C. metallidurans IcmF oligomeriza- tion by any appreciable extent. IcmF appears to be dimeric in all cases (Fig. 5, A and B). When GMPPCP is used, supra- molecular complexes are observed in all cases. However, more supramolecular complexes are apparent for samples prepared in the light, when more AdoCbl has been photo- lyzed, and in the presence of OHCbl, than for AdoCbl sam- ples prepared in the dark (Fig. 5C). We have previously noted that GMPPCP is more effective at inducing complex forma- tion between MeaB and MCMCbl in comparison to another nonhydrolyzable GTP analog, guanosine-5’-[(β,γ)-imido] triphosphate (GMPPNP) (33). Thus, we wondered if usage of the weaker oligomerization-agonist GMPPNP would allow us to observe smaller effects on oligomerization that are due to the nature of the Cbl cofactor. We ﬁnd qualitatively that there are very few, if any, IcmF supramolecular complexes in the presence of GMPPNP when AdoCbl is intact (dark sample) (Fig. 5D). Photolysis of AdoCbl leads to more supramolecular complex formation but not as many as in the GMPPNP + OHCbl sample (Fig. 5D). Together, these data indicate that IcmF oligomerization depends more strongly on the identity of the G-nucleotide than the identity of the Cbl cofactor, but IcmF with a damaged Cbl is more prone to oligomerization than IcmF with an intact AdoCbl. Thus, both the G- 6 J. Biol. Chem. (2023) 299(9) 105109 nucleotide identity and the Cbl identity shift the oligomeric state equilibrium, but to different degrees. Cryo-EM data reveal a G-protein dimer and an open conformation of the mutase To further investigate the structures of IcmF’s supramo- lecular complexes, we prepared cryogenic EM grids and collected datasets of C. metallidurans wt IcmF with GMPPCP and butyryl-CoA (wt IcmF + GMPPCP) and Q341A IcmF with GTP (Q341A IcmF + GTP). For the wt IcmF + GMPPCP dataset, the supramolecular complexes were ﬁrst manually picked on the micrographs, extracted using helical parameters, and processed as a single particle dataset. Signal subtraction was utilized to remove the extra density that was due to the extraction of single particles from a continuous helical particle (Fig. S4) (36). Local B factor postprocessing was performed to yield a 6.7-Å resolution reconstruction (Fig. 6A, Table S1) (37). For the Q341A IcmF + GTP dataset, we performed 3D single particle reconstructions, using repeated classiﬁcations for both 2D and 3D steps to separate out heterogeneity (Fig. S5). Local B factor postprocessing was also performed to yield a 4.6-Å resolution reconstruction (Fig. 6C, Table S1). For each the previously solved crystal structure of reconstruction, Role of G-protein dimerization in metalloenzyme maturation Figure 5. Negative stain EM analysis of the oligomeric state of IcmF when bound to various states of the cobalamin cofactor in the presence and absence of G-nucleotides. A, representative negative stain EM images of 20 ng/μl wt IcmF in the presence of AdoCbl incubated in the dark (left), AdoCbl exposed to light (middle), and OHCbl (right). None of the conditions form supramolecular complexes. B, representative negative stain EM images of 20 ng/μl wt IcmF in the presence of AdoCbl and GDP incubated in the dark (left), AdoCbl and GDP exposed to light (middle), and OHCbl and GDP (right). None of the conditions form supramolecular complexes. C, representative negative stain EM images of 20 ng/μl wt IcmF in the presence of AdoCbl and GMPPCP incubated in the dark (left), AdoCbl and GMPPCP exposed to light (middle), and OHCbl and GMPPCP (right). All the states contain supramolecular complexes; however, the states with inactivated cofactor (AdoCbl exposed to light or OHCbl) have more supramolecular complexes. The concentration of all cofactors and nucleotides added was 500 μM. D, representative negative stain EM images of 20 ng/μl wt IcmF in the presence of AdoCbl and GMPPNP incubated in the dark (left), AdoCbl and GMPPNP exposed to light (middle), and OHCbl and GMPPNP (right). Only the states that have inactivated cofactor (AdoCbl exposed to light or OHCbl) and a nonhydrolyzable analog contain supramolecular complexes. The concentration of all cofactors and nucleotides added was 500 μM. Each negative stain condition was repeated three times. AdoCbl, adenosylcobalamin; GMPPCP, guanosine-5’-[(β,γ)-methyleno]triphosphate; GMPPNP, guanosine-5’-[(β,γ)-imido]triphosphate; IcmF, isobutyryl-CoA mutase fused. J. Biol. Chem. (2023) 299(9) 105109 7 Role of G-protein dimerization in metalloenzyme maturation Figure 6. 3D cryogenic EM helical reconstruction of the supramolecular complexes of IcmF indicates the formation of a G-protein interdimer interface in solution. A, 6.7-Å resolution single particle reconstruction of the supramolecular complexes of wt IcmF in the presence of 500 μM GMPPCP and 500 μM butyryl-CoA (wt IcmF + GMPPCP) with four IcmF protomers modeled into the EM map. The unmodeled extra density represents additional density from the continuous supramolecular complexes that cannot accommodate a full IcmF protomer. B, two of the four protomers from wt IcmF + GMPPCP structure are shown. Each protomer is from a different IcmF homodimer (the homodimer is not shown for simplicity). The majority of the intermolecular contacts are made by the G-protein domains. Coloring: G-protein domain in teal, the substrate-binding domain in green, and the linker region in pink. Inset: The mutase active site is open indicated by the red lines. C, 4.6-Å resolution single particle reconstruction of the complexes of Q341A IcmF in the presence of 500 μM GTP (Q341A IcmF + GTP) with three IcmF protomers modeled into the EM map. D, two of the three protomers from Q341A IcmF + GTP structure are shown. Each protomer is from a different IcmF homodimer (the homodimer is not shown for simplicity). As in B, the majority of the intermolecular contacts are made by the G-protein domains. Colored as described in B. Top inset: The mutase active site is open indicated by the red lines. Bottom inset: Switch III residues (red) from one protomer contact the GDP bound to the second protomer across the G-protein domain: G-protein domain interface. GDP and Mg2+ ion are shown against the EM map. GMPPCP, guanosine-5’-[(β,γ)-methyleno]triphosphate; IcmF, isobutyryl-CoA mutase fused. C. metallidurans IcmF with GDP and AdoCbl bound was used as the initial docking model (29). In both reconstructions, one IcmF dimer contacts the other IcmF dimer through intermolecular interactions made by the G-protein domains (cyan in Fig. 6, B and D). The G-protein domain interface adopts the same conformation as observed for MeaB in the GMPPCP-MeaB:MCMCbl complex structure. Again, we observe switch III residues positioned at the inter- face adjacent to the GTP-binding site (Figs. S6A and S7). For the Q341A IcmF + GTP reconstruction, there is clear density for the guanosine and at least two phosphates of GTP in all the nucleotide-binding sites (Fig. 6D). However, due to the lower resolution of the wt IcmF + GMPPCP reconstruction, there is no clear density for GMPPCP in the nucleotide-binding site. For both reconstructions, we observed an open conformation of the mutase active site (Fig. 6, B and D). Overlaying previously determined closed (Fig. 7A) and open (Fig. 7C) conformations of IcmF onto the structure of Q341A IcmF + GTP reveals that the G-protein interdimer interface clashes with the closed conformation (Fig. 7B). Speciﬁcally, the switch III region (residues 333–344) and a helix (purple in Fig. 7B, residues 363–380) of one protomer clash with two 8 J. Biol. Chem. (2023) 299(9) 105109 helices of the substrate-binding domain of the other protomer (helix 1, residues 942–965, and helix 2, residues 974–995 in Fig. 7B); these clashes are not observed when the mutase is in an open conformation (Fig. 7D). Thus, in addition to switch III residues residing at the interface created by the G-protein domains, they also participate in propping open the mutase (Fig. 7, B and D). These structures are consistent with the biochemical solution state data, showing that supramolecular complexes form under conditions that require loading or unloading of Cbl. Overall, the cryo-EM reconstructions have trapped the conformation of G-protein domain that wedges open the active site of the mutase domain for cofactor to be loaded and unloaded. Discussion Metallochaperones ensure that valuable, and often highly reactive, metallocofactors are delivered efﬁciently to their target enzymes. The molecular basis for successful delivery is often enigmatic, especially for metallochaperones like MeaB that do not bind the metallocofactor directly. Because the complexes formed by the metallochaperone and target enzyme Role of G-protein dimerization in metalloenzyme maturation Figure 7. The G-protein domain interface wedges open the active site of the mutase domain of IcmF. A, ribbon drawing of the closed conformation of IcmF (PDB 4XC6) (29) with the substrate-binding domain (green) and Cbl-binding domain (orange), G-protein domain (blue) and linker (pink). Red lines indicate there is no gap between the domains. B, overlay of the closed conformation of IcmF (PDB 4XC6) (29) with the Q341A IcmF + GTP structure. Left inset: overlay reveals a clash between the helices of the substrate-binding domain of IcmF (dark green, helix 1 residues: 942–965; helix 2 residues: 974–995) with the switch III region (red-orange, residues 333–344) or helix (purple, residues 363–380) from the neighboring IcmF protomer. The distance between helix 2 and the Cbl-binding domain is 9.9 Å. Right inset: the same orientation as on left as surface representation but without the clashing G-protomer. C, ribbon drawing of the open conformation of IcmF (PDB 4XC6) (29) colored as in A. Red lines indicate there is a gap between the domains. D, overlay of the open conformation of IcmF (PDB 4XC6) (29) with the Q341A IcmF + GTP structure. Left inset: overlay shows no clashing between the helices of the substrate- binding domain of IcmF (dark green, helix 1 residues: 942–965; helix 2 residues: 974–995) with the switch III region (red-orange, residues 333–344) or helix (purple, residues 363–380) from the neighboring IcmF protomer. The distance between helix 2 and the Cbl-binding domain is 22.2 Å. Right inset: the same orientation as on left as surface representation but without the wedging G-protomer. Cbl, cobalamin; IcmF, isobutyryl-CoA mutase fused. are transient and can sample more than one conformational state, understanding the molecular basis of metalloenzyme maturation and/or cofactor repair involves the difﬁcult task of trapping transient protein:protein complexes in multiple conformational states. In this study, we use cryo-EM to cap- ture a long-awaited structure of the active conformation of an AdoCbl metallochaperone in complex with a target mutase and use mutagenesis and mass photometry to understand the assembly/disassembly of active cofactor-transfer complexes. Our data support the existence of a conserved molecular mechanism for AdoCbl transfer between the fused IcmF sys- tem and the standalone MeaB:MCM system and provide the snapshots needed to understand the molecular basis of metallochaperone-assisted AdoCbl transfer. Previous studies showed that the bacterial AdoCbl metal- lochaperone MeaB is always dimeric and uses a dramatic 180(cid:1) conformational change to switch between active (GTP-bound) and inactive (GDP-bound) states (Fig. 1D) (32, 33). In contrast, we previously showed that the fused metallochaperone:mutase is system IcmF has a chaperone G-protein domain that monomeric in the inactive state (Fig. 1E) (29), and we show here that IcmF uses oligomerization to switch between its inactive (GDP-bound) states and the active (GTP-bound) state. Evidence for the coupling of GTP binding to an IcmF oligo- meric state change comes from negative stain EM, cryo-EM, and mass photometry (Figs. 4 and 6). Importantly, regardless of whether the active conformation of the G-protein domain is formed through oligomerization (IcmF) or through a confor- mational change of an obligate dimer (MeaB/MMAA), the interface between protomers is the same (Fig. S7). The G-protein:G-protein interface is formed in both sys- tems by switch III residues of one protomer and the G-nucleotide and switch I region of the other protomer (Figs. 1C, inset, and 6D, inset) (33). This G-protein:G-protein interface is small, which allows for modest changes, such as loss of a phosphate group due to GTP hydrolysis, to shift the (or oligomeric state) equilibrium between conformational active and inactive G-protein states. Insight into the molecular basis for this conformational/oligomeric state shift comes from structural comparisons of GMPPCP-bound MeaB:MCMCbl (33) and GDP-bound MeaB (32), which indicate that GTP hydrolysis would bring about the loss of the contacts made by J. Biol. Chem. (2023) 299(9) 105109 9 Role of G-protein dimerization in metalloenzyme maturation chain-B switch III residues K188 and Q185 (MeaB numbering) to the chain-A terminal phosphate of GMPPCP and loss of the interchain salt bridge made by switch III residue D182 (chain- B) and switch I residue R108 (chain-A, MeaB numbering) (Fig. 1C inset). In terms of the salt bridge, structural com- parisons suggest that the repositioning of intradomain MeaB residues D92, E154, and R108 (Fig. 1C, inset) as a result of GTP hydrolysis and Mg2+ loss in turn repositions R108, breaking the interchain salt bridge (see Fig. S8). Collectively, the loss of the terminal phosphate of GTP results in a loss of all contacts that stabilize the active state of MeaB. Although the resolution of the cryo-EM IcmF structures is too low for a detailed analysis of protein:G-nucleotide interactions, we do ﬁnd that the switch III residues including K344 (K188 in MeaB) and Q341 (Q185 in MeaB) of one IcmF protomer are in direct contact with the G-nucleotide bound to another IcmF protomer. This arrangement in IcmF is just what one would expect for a conserved mechanism in which the binding/ orientation of one G-protein protomer is based on the G-nucleotide–bound state of another G-protein protomer. Despite the modest resolution of the cryo-EM structures determined in this study, establishing the positioning of the individual domains of IcmF was straightforward (Fig. 6). The resulting model shows that the active G-protein:G-protein interface stabilizes an open conformation of the mutase that is ready for AdoCbl delivery. It was previously proposed through structural superimpositions that the active G-protein confor- mation would stabilize the Cbl-binding domain away from the substrate-binding domain for cofactor loading in MeaB:MCM (Fig. 2A) and similarly in IcmF (Fig. 2B) (33) and here we see that this is in fact the case (Fig. 6). The second protomer of the G-domain dimer is wedged between the Cbl-binding domain and the substrate-binding domain, holding the mutase open (Fig. 7D, left inset). The closed conformation of the mutase appears incapable of interacting with the active G-protein state based on the structural superimpositions of IcmF structures (Fig. 7B, left inset). We previously predicted that the formation of the active state of MCM-bound MeaB would lead to a clash between a helix of MeaB (equivalent to the IcmF helix shown left inset) in purple in Fig. 7B, left inset) and a helix of the substrate- binding domain (helix 1 in Fig. 7B, if MCM remained in a closed state (33). This prediction is consistent with our current structures. Additionally, structural superim- positions suggest that the switch III region (red in Fig. 7B, left inset) also would make unfavorably close interactions with a substrate-binding domain helix (helix 2 in Fig. 7B, left inset). This observation provides an additional role for switch III, acting as a molecular wedge that is sensitive to the identity of the G-nucleotide state. The wedge is secured with GTP bound, allowing AdoCbl to be delivered to an open structure, and the wedge is loosened by GTP hydrolysis, allowing the Cbl-binding domain to close, trapping a delivered AdoCbl (Fig. 2B). Chemical logic dictates that a metallochaperone delivery/ repair system should only open a target enzyme for cofactor delivery in the apo-state of enzyme and/or following cofactor damage, as it is wasteful to replace a working cofactor. The structure of IcmF suggests that the Cbl itself may regulate the open/closed equilibrium of the mutase (Fig. 8). In particular, IcmF structures show that the adenosyl moiety of AdoCbl rea- ches across the boundary between the Cbl-binding domain to the substrate-binding domain, thereby securing the domains together (29) and shifting the conformational equilibrium to- ward a closed mutase state. Loss of the adenosyl moiety due to photolysis or oxidative damage is expected to loosen the connection between domains, facilitating the movement of the Cbl-binding domain and shifting the conformational equilib- rium toward the open mutase state (25–27, 29). The cryo-EM structure presented here shows that an open conformation of IcmF creates a pocket that can be ﬁlled by a neighboring IcmF’s G-protein domain and, in particular, its wedge helix (purple residues 363–380) and its switch III region (Fig. 7D). Thus, loss of AdoCbl or damage to AdoCbl should shift the conforma- tional equilibrium of IcmF to an open state, facilitating the binding of a G-domain of a neighboring IcmF molecule (Fig. 8) and starting the repair process. In addition to data showing that switch III residues play a role in the removal of damaged Cbl (31), our negative stain EM data using GMPPNP and GMPPCP show more extensive IcmF oligomerization when AdoCbl is Figure 8. Cartoon depicting equilibrium between open and closed states of IcmF mutase. The equilibrium is expected to shift to the right in the presence of a bound AdoCbl due to the adenosyl moiety of AdoCbl, which juts into the substrate-binding domain (green), securing the Cbl-binding domain (orange) against the substrate-binding domain. The equilibrium is expected to shift to the left in the case of AdoCbl damage or loss. Without the adenosyl moiety jutting across the mutase interface, the open conformation is more favorable, which in turn would facilitate IcmF oligomerization as the open mutase conformation has a preformed binding pocket for the G-protein “wedge.” The second protomer of each IcmF molecule is not shown for simplicity. AdoCbl, adenosylcobalamin; Cbl, cobalamin; IcmF, isobutyryl-CoA mutase fused. 10 J. Biol. Chem. (2023) 299(9) 105109 Role of G-protein dimerization in metalloenzyme maturation subjected to photolysis or replaced with OHCbl than for intact AdoCbl (Fig. 5, C and D). We expect that in vivo, the confor- mational equilibrium shifts toward the open mutase structure due to Cbl damage or AdoCbl loss will be necessary for oligo- merization, limiting the wasteful replacement of a working cofactor. In terms of maturation, the formation of the supra- molecular complexes made up of apo-IcmF molecules with GMPPCP (Fig. 4D, right) invokes an assembly line model in which a chain of open mutases can be ﬁlled in succession by an ATR that moves down the chain, delivering AdoCbl as it is synthesized. Future experiments aimed at establishing the length of apo-IcmF complexes in vivo would provide insight into this hypothesis. We currently do not know if IcmF forms an oligomer longer than a tetramer in vivo nor do we know how ATR interacts with IcmF. A structure of these IcmF supramo- lecular complexes or of an IcmF tetramer in the presence of ATR would be highly valuable. Collectively, the structural and biochemical data obtained previously (6, 29, 32, 33) and presented here indicate that the only signiﬁcant difference between the standalone system (Fig. 2A) and the fused IcmF (Fig. 2B) is whether the active G-protein state is formed via a conformational change or via oligomerization. It is not unprecedented for signaling proteins to employ both methods (conformational change and oligomerization) to switch between active and inactive states. For example, members of the CAP family of transcription factors can use either ligand binding to induce dimerization that affords DNA association or ligand binding to induce a conformational change that affords DNA association (38, 39). The use of two different methods for creating an active G-protein state makes sense if one considers that G- protein chaperones are designed to be poor GTPases in the absence of their target protein to prevent unwanted GTP hy- drolysis. Since the G-protein and target protein are fused in IcmF, unwanted GTP hydrolysis would be a large problem if the active G-protein state was easily formed. For MeaB, MCM is its GAP. For the G-protein domain of IcmF, however, a second IcmF protomer is the GAP, and our cryo-EM data show that residues required for GTP hydrolysis (K344 for example) are on the sec- ond protomer (Fig. 6). Thus, the active site for GTP hydrolysis is missing in IcmF’s resting homodimeric state, which is the state of wt IcmF that is visible in the absence of GTP or GTP analogs (Fig. 4). If the G-protein domains of IcmF were present as an obligate dimer, allowing a conformational change to create an active site capable of GTP hydrolysis, chemical logic suggests that the unwanted GTP hydrolysis would be a more substantial problem. IcmF is a poor GTPase because its active site residues are not part of its resting structure. The presence of fused systems has not been reported for other members of the SIMIBI class of P-loop G-proteins, such as UreG and HypE, which also rely on binding a G-nucleotide to form the active conformation of the G-protein. Instead, UreG and HypE resemble MeaB (40); however, the degree to which these metallochaperones utilize the same molecular mechanisms is unclear. Each metallochaperone system appears to have a different number of accessory proteins, which are commonly of unknown function and unknown structure, and if structures exist, the structures are often of inactive states or isolated states not of the protein:protein complex that is responsible for maturation. Our results here suggest that cryo- EM is likely to be crucial for obtaining structures of transient protein complexes, affording new snapshots of the metal- lochaperone delivery and repair processes. We hope that the studies presented here will beneﬁt metalloenzyme applications in industry and in medicine by providing structural and mechanistic insight into one system in detail. With the cryo- EM resolution revolution, we expect that this work is the beginning of what will be an exciting decade for this ﬁeld. Experimental procedures All chemicals, solvents, and reagents were purchased from Sigma-Aldrich unless otherwise noted. Plasmids Name pET28a_cmIcmF pET28a_cmIcmF_Q341A pET28a_cmIcmF_K344A Features N-terminal His-tag, thrombin cleavage site, T7 protomer, KanR N-terminal His-tag, thrombin cleavage site, T7 protomer, KanR N-terminal His-tag, thrombin cleavage site, T7 protomer, KanR Source Ref. (29) This study This study Cloning The plasmid containing the WT C. metallidurans icmF gene as described previously (29) was used as the template for site- directed mutagenesis. Site-directed variants were generated using a Quikchange II XL site-directed mutagenesis kit (Agi- lent) using the following primers. Name IcmF_Q341A_F IcmF_Q341A_R IcmF_K344A_F IcmF_ K344A_R Sequence (50 to 30) GCGCGGCCAGCGCGCTCGA GAAGATCGAC GTCGATCTTCTCGAGCGCGCTGG CCGCGC GCGAGCCAGCTCGAGGCGATCGA CATGCTCGACTTCGC GCGAAGTCGAGCATGTCGATCG CCTCGAGCTGGCTGGC Mutations generated according to the Quikchange protocol were conﬁrmed by Sanger sequencing (Genewiz) using the following primers for full sequencing overlap. Name IcmF_seq_1 IcmF_seq_2 IcmF_seq_3 IcmF_seq_4 IcmF_seq_5 Direction Forward Forward Forward Forward Reverse Sequence (50 to 30) GCGCAACTGATTACCGCG CAAGCAGGTGCAGCGCAA CGTGTTCGCGTTCAAGCG GAAGCCGGTGCGAATCCG CCACATCGCCAGCAGCTTG Protein expression and puriﬁcation Cell growth and puriﬁcations of wt IcmF, Q341A IcmF, and K344A IcmF from C. metallidurans were conducted following J. Biol. Chem. (2023) 299(9) 105109 11 Role of G-protein dimerization in metalloenzyme maturation the same procedure described here. An overnight culture of 100 ml of lysogeny broth medium (Fisher BioReagents) sup- plemented with 50 μg/L kanamycin (GoldBio) was inoculated from a single colony of Escherichia coli BL21 T7 Express competent cells (New England Biolabs) transformed with the appropriate gene and grown at 37 (cid:1)C with shaking. The overnight starter culture was used to inoculate 1 L of lysogeny broth supplemented with 50 μg/L kanamycin at 37 (cid:1)C. The 1 L culture was placed at 16 (cid:1)C with shaking when A600 reached (cid:3)0.5 to 0.6. After 2 h, the 1 L culture was induced with a ﬁnal concentration of 0.1 mM IPTG (GoldBio) and grown for 10 h to 12 h at 16 (cid:1)C with shaking. Cells were harvested by centrifugation (5000g, 4 (cid:1)C, 20 min) and ﬂash frozen in liquid N2 before being stored in a −80 (cid:1)C freezer for future use. Cells from 2 L of cell culture were resuspended in 80 mL of lysis buffer (50 mM Hepes pH 7.5, 500 mM NaCl, 20 mM imidazole) supplemented with one cOmplete EDTA-free pro- tease inhibitor tablet (Roche), 1 mM of PMSF, 1 mM of ben- zamidine HCl, and benzonase nuclease. Cells were lysed by ultrasoniﬁcation, and cell lysates were clariﬁed by centrifuga- tion (28,000g, 30 min, 4 (cid:1)C). Clariﬁed lysate was passed through a 0.2 μm ﬁlter before being loaded onto a 5 ml Ni-NTA column (GE Healthcare) equilibrated with lysis buffer using an fast protein liquid chromatography (FPLC) system (BioRad NGC System). The column was washed with 10 column volumes of lysis buffer and 10 column volumes of 50 mM Hepes pH 7.5, 500 mM NaCl, and 40 mM imidazole using an FPLC. Protein was eluted with 50 mM Hepes pH 7.5, 500 mM NaCl, 200 mM imidazole using an FPLC with a ﬂow rate of 4 mL/min. Elution fractions were buffer exchanged into 50 mM Hepes pH 7.5 and then concentrated in a 50 kDa MWCO centrifugal ﬁlter. The concentrated fractions were loaded onto a MonoQ 10/100 anion exchange chromatography column (Cytiva) prepped with 50 mM Hepes pH 7.5 and 5 mM NaCl. Protein was eluted with a linear gradient from 10% to 80% of 50 mM Hepes pH 7.5 and 500 mM NaCl with a ﬂow rate of 2 mL/min. The protein eluted in a sharp peak at around 250 mM to 300 mM NaCl. There was a second peak immediately following the ﬁrst peak, corre- sponding to aggregated/truncated protein. The eluted protein was concentrated in a 50 kDa MWCO centrifugal ﬁlter. The concentrated fractions of full-length IcmF were loaded onto a Superdex 200 16/60 size-exclusion chromatography (SEC) column (GE Healthcare) equilibrated with SEC buffer (20 mM Hepes pH 8, 50 mM NaCl) and eluted with a ﬂow rate of 1 mL/ min. Elution fractions from SEC were concentrated in a 50 kDa MWCO centrifugal ﬁlter. Purity was assessed by a 4 to 20% (w/ v) SDS-PAGE (Bio-Rad). The concentration of the IcmF monomer was determined by UV/Vis absorbance at 280 nm −1, determined using an extinction coefﬁcient of 84,600 M using the ProtParam tool (41). Protein samples at a concen- tration of (cid:3)10 mg/mL (80 μM) in SEC buffer were ﬂash frozen in liquid N2 and stored in a −80 (cid:1)C freezer for future use. −1 cm (Molecular Probes) was used for all GTPase assays following the manufacturer’s instructions with the following modiﬁcations. The assay reactions (200 μL) were prepared excluding the enzyme (wt IcmF, Q341A IcmF, K344A IcmF, or no enzyme SEC buffer control) and incubated at room temperature (23 (cid:1)C) for 5 min before initiating the assay to control for contaminating inorganic phosphate in the GTP. After incubation, the assays were initiated with the addition of 0.5 μM enzyme or 10 μL of IcmF SEC buffer. After 7 s of initial mixing, the absorbance at 360 nm for each assay reaction was recorded every 8 s with 2 s of mixing in between readings for 15 min total using a SpectraMax Plus 384 microplate reader (Molecular Dimensions). The ab- sorbances were converted into concentration of inorganic phosphate using the standard curve generated according to the manufacturer’s directions. The initial rates were calculated for each assay reaction and subtracted from the average initial rates of the no enzyme control of the corresponding concentration of GTP. The Michaelis–Menten parameters (kcat, Km, and Vmax) were generated using a nonlinear regression by Prism v9.4.1 (Graphpad). Reported values ± SD are the results of at least three independent experiments (Table 1). Mass photometry Mass photometry (interferometric scattering mass spec- trometry) was performed on a Refeyn instrument using AcquireMP v2.4.1 and DiscoverMP v2.4.2. All movies were taken for a length of 60 s using the default parameters. The contrasts were converted into molecular weights using the standard curve generated from a sample of NativeMark Un- stained Protein Standard (Novex by life Technologies). Gaussian curves were ﬁt to each histogram distribution, and the mass (kDa), sigma (kDa), and normalized counts were determined using the PhotoMol software (https://spc.embl- hamburg.de/app/photoMol) (42). Each percentage of total counts ± SD reported are the results of at least three inde- pendent experiments. All samples that contained any additives also contained 500 μM MgCl2. Samples were incubated with 500 μM GDP or no nucleotide for 15 min on ice at a con- centration of 80 ng/μl prior to the ﬁnal dilution to 8 ng/μL on the instrument and data recording. Samples were incubated with 500 μM GMPPCP for 15 min on ice at a concentration of 160 ng/μl prior to the ﬁnal dilution to 16 ng/μL on the in- strument and data recording. Samples containing 500 μM GTP were initially diluted to 160 ng/μL before the ﬁnal dilution to 16 ng/μL in SEC buffer (20 mM Hepes pH 8, 50 mM NaCl) supplemented with 500 μM GTP on the instrument and data recording. Samples containing the substrate butyryl-CoA (500 μM) and/or various nucleotides (GDP or GMPPCP) were incubated for 15 min on ice at a concentration of 200 ng/ μL prior to the ﬁnal dilution to 20 ng/μL on the instrument and data recording. Reported values ± SD are the results of at least three independent experiments. GTPase assays The GTPase activity of wt IcmF, Q341A IcmF, and IcmF K344A was determined in the presence of various concentrations of GTP (2–1500 μM) (Roche). The EnzChek phosphate assay kit Negative stain EM specimen preparation and imaging wt IcmF, Q341A IcmF, or K344A IcmF was thawed on ice and diluted to 20 ng/μL in IcmF SEC buffer (20 mM Hepes pH 12 J. Biol. Chem. (2023) 299(9) 105109 Role of G-protein dimerization in metalloenzyme maturation 8, 50 mM NaCl). Each of the samples except samples with GTP were incubated for 30 min on ice with the corresponding nucleotide (GDP or GMPPCP) before grid preparation. The ﬁnal concentration of any additives (nucleotides: GDP, GTP, or GMPPCP; cofactor: AdoCbl or OHCbl; MgCl2) in the samples were 500 μM. For the samples containing GTP, the GTP was added, and the protein solution immediately applied to the grid. For the samples containing AdoCbl and exposed to light, after incubation for 30 min in the dark, the protein so- lution was then exposed to a white light for 15 min before application on the grid. Carbon-coated 300 mesh copper EM grids (Electron Mi- croscopy Services) were glow discharged for 1 min at −15 mA. An aliquot (5 μl) of the protein solution was applied to the grid; after approximately 1 min, the solution was blotted and immediately replaced with solution of 2% uranyl acetate (VWR). The stain solution was blotted and replaced twice, then allowed to stand for 1 min before the ﬁnal blot, and then was dried. All blotting was done manually using ﬁlter paper (Whatman, grade 40). The specimens were imaged with an AMT Nanosprint5 camera on a FEI Morgagni electron mi- croscope operated at 80 kV. Images were collected at 18,000× magniﬁcation. Cryo-EM grid preparation Services) was The grids used to investigate the supramolecular complex of IcmF in the presence of GMPPCP (wt IcmF + GMPPCP) were prepared as follows: 0.2 mg/mL graphene oxide suspension was prepared using molecular grade water. The suspension was centrifuged at 300g for 1 min to remove large aggregates. A Quantifoil 1.2 to 1.3 Cu 300 mesh holey-carbon grid (Electron Microscopy glow discharged at −40 mA at 0.1 bar for 2 min before application of 3 μL of the graphene oxide suspension. The suspension was incubated for 1 min before excess suspension was blotted away (Whatman, grade 40). The grid was washed twice on the graphene oxide suspension side and once on the backside of the grid in mo- lecular grade water and dried. The graphene oxide–covered grid was plunged on a Thermo Fisher Scientiﬁc Vitrobot (Mk IV) cryo-plunger. The ﬁnal protein solution contained 2 μM wt IcmF, 500 μM GMPPCP, and 500 μM butyryl-CoA incubated in SEC buffer for at least 15 min before applica- tion. The sample (5 μL) was applied to the grids that were blotted for 5 s with a blot force of 10 (Whatman ﬁlter paper #1) before plunging into liquid ethane and transferring to storage grids. The temperature and humidity inside the Vitrobot chamber were set to 8 (cid:1)C and 95%, respectively. The grids used to investigate the conformation of Q341A IcmF in the presence of GTP (Q341A IcmF + GTP) were prepared as follows: a Quantifoil R 1.2 to 1.3 Cu 300 mesh holey-carbon grid (Electron Microscopy Services) was glow discharged at −15 mA at 0.039 bar for 1 min before application of the protein solution. The ﬁnal protein solution contained 2 μM Q341A IcmF, 500 μM GTP in SEC buffer. The sample (5 mL was applied to the grids that were blotted for 3 s with a blot force of 10 (Whatman ﬁlter paper #1) before plunging into liquid ethane and transferring to storage grids. The tempera- ture and humidity inside the Vitrobot chamber were set to 8 (cid:1)C and 95%, respectively. Cryo-EM data collection Data were collected at the MIT.nano Center for Automated Cryogenic Electron Microscopy at the Massachusetts Institute of Technology on an FEI Talos Arctica G2 Cryo 200 kV transmission electron microscope equipped with a Falcon 3EC camera. The data collection parameters for the IcmF + GMPPCP grid were as follows: 73,000× magniﬁcation resulting in a pixel size of 2.043 Å, 14 frames, 10.5 e-/Å2/frame dose, and defocus range 1.2 to 3.1 μm. The dataset contained 602 movies. The data collection parameters for the Q341A IcmF + GTP grid were as follows: 92,000× magniﬁcation resulting in a pixel size of 1.5998 Å, 14 frames, 14.7 e-/Å2/frame dose, and defocus range 1.2 to 3.1 μm. The dataset contained 673 movies. These parameters are summarized in Table S1. Cryo-EM data processing and model reﬁnement Cryo-EM data processing of the datasets was carried out using a combination of Relion 4.0-beta (43) and CryoSPARC v3.3.2 (36) and is summarized in Fig. S4. For the wt IcmF + GMPPCP dataset, individual frames of dose-fractionated ex- posures were aligned and summed using Relion’s imple- mentation of MotionCor2 (44) and the defocus of the summed frames was estimated using Relion’s implementation of CTFFind4 (45). The start-end coordinates for the locations of the helical supramolecular complexes were manually deter- mined for all micrographs. Using these coordinates, 510,941 particles were extracted using a box size of 128 pixels (257.8 Å), tube diameter of 200 Å, one asymmetric unit, and a helical rise of 10 Å (Fig. S4). These particles were imported in CryoSPARC for the rest of the data processing. The particles were subjected to one round of initial reference-free 2D clas- siﬁcation with a mask of 200 Å to generate 50 class averages. After removing the classes that visually did not look like par- ticles, another round of reference-free 2D classiﬁcation was performed with the remaining 459,669 particles to generate 50 2D class averages. The 439,589 particles selected after the second round of 2D classiﬁcation were used to generate two ab initio initial reference-free models using no imposed symme- try. The initial models were subjected to heterogenous reﬁnement. The class consisting of 363,613 intact particles was subjected to homogenous reﬁnement. After homogenous reﬁnement, the aligned particles were subjected to a local reﬁnement using CryoSPARC’s new implementation. These aligned particles were checked for duplicate particles. The remaining 165,181 aligned particles were subjected to signal subtraction to remove signal from the ends of the supramo- lecular complexes that were artifacts of processing helical data as single particles using a manually generated mask. These subtracted particles were then subjected to a local reﬁnement using CryoSPARC’s new implementation. These aligned par- ticles were then checked for duplicate particles. The ﬁnal 138,956 particles were subject to one more round of local J. Biol. Chem. (2023) 299(9) 105109 13 Role of G-protein dimerization in metalloenzyme maturation reﬁnement. CryoSPARC’s dynamic masking was utilized for all reﬁning steps, excluding the signal subtraction. Combination of the two half-maps along with local B-factor adjustment was performed on the COSMIC2 server’s implementation (46) of LocSpiral (37) with a low pass ﬁlter of 15 Å, bandwidth of band pass ﬁlter of 8 Å, and an initial binarization threshold of 0.679. The Fourier shell correlation (FSC) plots were generated by CryoSPARC. The ﬁnal masked resolution at FSC = 0.143 was 6.7 Å (Fig. S9A). For the Q341A IcmF + GTP dataset, all data processing was performed in CryoSPARC v3.3.2 (36) and is summarized in Fig. S5. First, individual frames of dose-fractionated exposures were aligned and summed using patch motion correction. Next, the defocus of the summed frames was estimated using CryoSPARC’s patch contrast transfer function estimation. To generate the coordinates of the particles, the blob picker was used with minimum particle diameter of 200 Å and maximum particle diameter of 350 Å. Using a normalized cross-corre- lation score above 0.160 and local power between −8,352 and 242,864, 180,271 particles were extracted with a box size of 196 pixels. These particles were subjected to one round of initial reference-free 2D classiﬁcation to generate 200 2D class averages. After removing classes that visually did not look like particles, another round of reference-free 2D classiﬁcation was performed with the remaining 74,767 particles to generate 100 2D class averages. The 70,582 particles selected after the sec- ond round of 2D classiﬁcation were used to generate three ab initio initial reference-free models using no imposed symme- try. For the following heterogenous reﬁnement, two of the ab initio models (one containing intact complex and one con- taining junk particles) were supplied. Only the 59,335 particles associated with the model containing intact complex were subject to a homogenous reﬁnement. These particles were then subjected to a local reﬁnement using CryoSPARC. Cry- oSPARC’s dynamic masking was utilized for all reﬁning steps. Combination of the two half-maps along with local B-factor adjustment was performed on the COSMIC2 server’s imple- mentation (46) of LocSpiral (37) with a low pass ﬁlter of 15 Å, bandwidth of band pass ﬁlter of 8 Å, and an initial binarization threshold of 0.878. The FSC plots were generated by Cry- oSPARC. The ﬁnal masked resolution at FSC = 0.143 was 4.6 Å (Fig. S9B). For model building and reﬁnement, one protomer of the dimeric IcmF (chain B from PDB 4XC6, residues 22–1093) was segmented into two fragments that were manually docked into the maps resulting from the reconstruction of the wt IcmF + GMPPCP and Q341A IcmF + GTP datasets, respectively (29). All the ligands and water molecules were removed from the fragments. Despite the low to modest resolution, side chains were retained in the fragments. The ﬁrst fragment consisted of the Cbl-binding domain and the G-protein domain (residues 22–442); the second fragment consisted of substrate-binding domain and the linker region (residues 443–1093). For the wt IcmF + GMPPCP reconstruction, four complete protomers (eight fragments) of IcmF were manually docked in the map in ChimeraX v1.4 (47). Using Phenix real-space reﬁnement, the resulting model was subjected to one round of reﬁnement 14 J. Biol. Chem. (2023) 299(9) 105109 consisting of rigid body reﬁnement, with each fragment reﬁned as an individual rigid body. There was no clear density for substrate or GMPPCP and, therefore, these ligands were not modeled in. For the Q341A IcmF + GTP reconstruction, three copies of a complete protomer of IcmF (six fragments) were manually docked into the map in ChimeraX v1.4. Although the map appeared to contain two IcmF homodimers, the density was only good enough to place three protomers (six fragments) and no atoms were modeled into a fourth protomer. Within the modeled protomers, however, there was density present for residues 1,011 to 1,018, which were previously disordered in the crystal structures, and those residues were added manually by model building in Coot (48). Additionally, there was clear In the density for GDP at each nucleotide-binding site. nucleotide-binding site for chain A, there was enough density to model in a magnesium ion in addition to the GDP. Using Phenix real-space reﬁnement (49), the docked models were subjected to one round of reﬁnement consisting ﬁrst of simulated annealing and then rigid body reﬁnement, with each copy of each fragment of IcmF deﬁned as an individual rigid body. The second round of real-space reﬁnement was carried out on the model with rigid body reﬁnement and minimiza- tion. Molecular building and reﬁning software packages were compiled by SBGrid (50). Data availability Atomic coordinates have been deposited in the Protein Data Bank under accession codes 8SSL and 8STA. The cryo-EM density maps have been deposited in the Electron Micro- scopy Data Bank under accession number EMD-40751 and EMD-40758. Supporting information—This article contains supporting informa- tion (51, 52). Acknowledgments—The authors thank Rachel Waller for help with producing the IcmF switch III variants, Dr Andrew Grassetti for help with preparation of the graphene oxide–coated grids, Dr Edward Brignole for cryo-EM data collection help, and Dr Max Wilkinson and Dr Joseph H. Davis for valuable discussions on cryo-EM processing. Specimens were prepared and imaged at the Automated Cryogenic Electron Microscopy Facility in MIT.nano on a Talos Arctica microscope, which was a gift from the Arnold and Mabel Beckman Foundation. Author contributions—C. L. D. supervision; F. A. V., D. A. F., G. A. A., D. A. B., G. K., D. R. F., and M. J. investigation; F. A. V. and C. L. D. formal analysis; F. A. V. and C. L. D. writing–original draft; F. A. V. and C. L. D. writing–review and editing. Funding and additional information—This work was supported by National Institutes of Health (NIH) grant R35 GM126982 (C. L. D.), a Ruth L. Kirschstein Predoctoral Individual National Research Service Award from NIH F31 GM131648 (F. A. V.), and a NIH Molecular Biophysics Training Grant (T32 GM008313) that sup- ported D. A. B. Support was also provided by an MIT School of Science Fellowship (F. A. V.), the MIT UROP ofﬁce (D. A. F.), a David H. Koch Graduate Fellowship (G. K.), an MIT Poitras pre- Role of G-protein dimerization in metalloenzyme maturation doctoral fellowship (M. J.), and the Amgen Scholars Program (D. R. F.). The content is solely the responsibility of the authors and does not necessarily represent the ofﬁcial views of the National Institutes of Health. Conﬂict of interest—C. L. D. is a Howard Hughes Medical Investi- gator. M. J. consults for Gate Biosciences and Evozyne. The other authors declare that they have no conﬂicts of interest with the contents of this article. Abbreviations—The abbreviations used are: AdoCbl, adenosylco- balamin; ATR, adenosyltransferase; Cbl, cobalamin; GAP, GTPase- accelerating protein; GMPPCP, guanosine-5’-[(β,γ)-methyleno] triphosphate; GMPPNP, guanosine-5’-[(β,γ)-imido]triphosphate; isobutyryl-CoA mutase fused; MCM, methylmalonyl-CoA IcmF, signal size-exclusion chromatography; SIMIBI, mutase; SEC, recognition particle, MinD, and BioD. References 1. Maret, W. (2018) Metallomics: the science of biometals and bio- metalloids. In: Arruda, M. A. Z., ed. Metallomics: The Science of Bio- metals, Springer International Publishing, Cham: 1–20 2. Capdevila, D. A., Edmonds, K. A., and Giedroc, D. P. (2017) Metal- lochaperones and metalloregulation in bacteria. Essays Biochem. 61, 177–200 3. Rosenzweig, A. C. (2002) Metallochaperones: bind and deliver. Chem. Biol. 9, 673–677 4. O’Halloran, T. V., and Culotta, V. C. (2000) Metallochaperones, an intra- cellular shuttle service for metal ions. J. Biol. Chem. 275, 25057–25060 5. Leipe, D. D., Wolf, Y. I., Koonin, E. V., and Aravind, L. (2002) Classiﬁ- cation and evolution of P-loop GTPases and related ATPases. J. Mol. Biol. 317, 41–72 6. Froese, D. S., Kochan, G., Muniz, J. R., Wu, X., Gileadi, C., Ugochukwu, E., et al. (2010) Structures of the human GTPase MMAA and vitamin B12-dependent methylmalonyl-CoA mutase and insight into their com- plex formation. J. Biol. Chem. 285, 38204–38213 7. Padovani, D., Labunska, T., and Banerjee, R. (2006) Energetics of inter- action between the G-protein chaperone, MeaB, and B12-dependent methylmalonyl-CoA mutase. J. Biol. Chem. 281, 17838–17844 8. Lee, M. H., Mulrooney, S. B., Renner, M. J., Markowicz, Y., and Hau- singer, R. P. (1992) Klebsiella aerogenes urease gene cluster: sequence of ureD and demonstration that four accessory genes (ureD, ureE, ureF, and ureG) are involved in nickel metallocenter biosynthesis. J. Bacteriol. 174, 4324–4330 9. Fong, Y. H., Wong, H. C., Yuen, M. H., Lau, P. H., Chen, Y. W., and Wong, K.-B. (2013) Structure of UreG/UreF/UreH complex reveals how urease accessory proteins facilitate maturation of Helicobacter pylori urease. PLoS Biol. 11, e1001678 10. Gasper, R., Scrima, A., and Wittinghofer, A. (2006) Structural insights into HypB, a GTP-binding protein that regulates metal binding. J. Biol. Chem. 281, 27492–27502 11. Maier, T., Jacobi, A., Sauter, M., and Böck, A. (1993) The product of the hypB gene, which is required for nickel incorporation into hy- drogenases, is a novel guanine nucleotide-binding protein. J. Bacteriol. 175, 630–635 12. Takahashi-Iñiguez, T., González-Noriega, A., Michalak, C., and Flores, M. E. (2017) Human MMAA induces the release of inactive cofactor and restores methylmalonyl-CoA mutase activity through their complex formation. Biochimie 142, 191–196 13. Padovani, D., Labunska, T., Palfey, B. A., Ballou, D. P., and Banerjee, R. (2008) Adenosyltransferase tailors and delivers coenzyme B12. Nat. Chem. Biol. 4, 194–196 14. Padovani, D., and Banerjee, R. (2009) A G-protein editor gates coenzyme B12 loading and is corrupted in methylmalonic aciduria. Proc. Natl. Acad. Sci. U. S. A. 106, 21567–21572 15. Cracan, V., Padovani, D., and Banerjee, R. (2010) IcmF is a fusion between the radical B12 enzyme isobutyryl-CoA mutase and its G-protein chap- erone. J. Biol. Chem. 285, 655–666 16. Dempsey-Nunez, L., Illson, M. L., Kent, J., Huang, Q., Brebner, A., Watkins, D., et al. (2012) High resolution melting analysis of the MMAA gene in patients with cblA and in those with undiagnosed methylmalonic aciduria. Mol. Genet. Metab. 107, 363–367 17. Banerjee, R. (2003) Radical carbon skeleton rearrangements: catalysis by coenzyme B12 -dependent mutases. Chem. Rev. 103, 2083–2094 18. Korotkova, N., Chistoserdova, L., Kuksa, V., and Lidstrom, M. E. (2002) Glyoxylate regeneration pathway in the methylotroph Methylobacterium extorquens AM1. J. Bacteriol. 184, 1750–1758 19. Li, Z., Kitanishi, K., Twahir, U. T., Cracan, V., Chapman, D., Warncke, K., et al. (2017) Cofactor editing by the G-protein metallochaperone domain regulates the radical B12 enzyme IcmF. J. Biol. Chem. 292, 3977–3987 20. Cracan, V., and Banerjee, R. (2012) Novel coenzyme B12-dependent interconversion of isovaleryl-CoA and pivalyl-CoA. J. Biol. Chem. 287, 3723–3732 21. Mansoorabadi, S. O., Padmakumar, R., Fazliddinova, N., Vlasie, M., Banerjee, R., and Reed, G. H. (2005) Characterization of a succinyl-CoA radical-Cob(II)alamin spin triplet intermediate in the reaction catalyzed mutase. by Biochemistry 44, 3153–3158 adenosylcobalamin-dependent methylmalonyl-CoA 22. Sension, R. J., Cole, A. G., Harris, A. D., Fox, C. C., Woodbury, N. W., Lin, S., et al. (2004) Photolysis and recombination of adenosylcobalamin bound to glutamate mutase. J. Am. Chem. Soc. 126, 1598–1599 23. Takahashi-Íñiguez, T., García-Arellano, H., Trujillo-Roldán, M. A., and Flores, M. E. reactivation of human methylmalonyl-CoA mutase by MMAA protein. Biochem. Biophys. Res. Commun. 404, 443–447 (2011) Protection and 24. Padovani, D., and Banerjee, R. (2006) Assembly and protection of the chaperone. enzyme, methylmalonyl-CoA mutase, by its radical Biochemistry 45, 9300–9306 25. Mancia, F., Smith, G. A., and Evans, P. R. (1999) Crystal structure of substrate complexes of methylmalonyl-CoA mutase. Biochemistry 38, 7999–8005 26. Mancia, F., and Evans, P. R. (1998) Conformational changes on substrate binding to methylmalonyl CoA mutase and new insights into the free radical mechanism. Structure 6, 711–720 27. Mancia, F., Keep, N. H., Nakagawa, A., Leadlay, P. F., McSweeney, S., Rasmussen, B., et al. (1996) How coenzyme B12 radicals are generated: the crystal structure of methylmalonyl-coenzyme A mutase at 2 Å resolution. Structure 4, 339–350 28. Ruetz, M., Campanello, G. C., McDevitt, L., Yokom, A. L., Yadav, P. K., Watkins, D., et al. (2019) Allosteric regulation of oligomerization by a B12 trafﬁcking G-protein is corrupted in methylmalonic aciduria. Cell Chem. Biol. 26, 960–969 29. Jost, M., Cracan, V., Hubbard, P. A., Banerjee, R., and Drennan, C. L. (2015) Visualization of a radical B12 enzyme with its G-protein chap- erone. Proc. Natl. Acad. Sci. U. S. A. 112, 2419–2424 30. Lofgren, M., Koutmos, M., and Banerjee, R. (2013) Autoinhibition and signaling by the switch II motif in the G-protein chaperone of a radical B12 enzyme. J. Biol. Chem. 288, 30980–30989 31. Lofgren, M., Padovani, D., Koutmos, M., and Banerjee, R. (2013) A switch III motif relays signaling between a B12 enzyme and its G-protein chap- erone. Nat. Chem. Biol. 9, 535–541 32. Hubbard, P. A., Padovani, D., Labunska, T., Mahlstedt, S. A., Banerjee, R., and Drennan, C. L. (2007) Crystal structure and mutagenesis of the metallochaperone MeaB: into the causes of methylmalonic aciduria. J. Biol. Chem. 282, 31308–31316 insight 33. Vaccaro, F. A., Born, D. A., and Drennan, C. L. (2023) Structure of metallochaperone in complex with the cobalamin-binding domain of its target mutase provides insight into cofactor delivery. Proc. Natl. Acad. Sci. U. S. A. 120, e2214085120 34. Mascarenhas, R., Ruetz, M., McDevitt, L., Koutmos, M., and Banerjee, R. (2020) Mobile loop dynamics in adenosyltransferase control binding and reactivity of coenzyme B12. Proc. Natl. Acad. Sci. U. S. A. 117, 30412–30422 J. Biol. Chem. (2023) 299(9) 105109 15 Role of G-protein dimerization in metalloenzyme maturation 35. Jost, M., Born, D. A., Cracan, V., Banerjee, R., and Drennan, C. L. (2015) Structural basis for substrate speciﬁcity in adenosylcobalamin-dependent isobutyryl-CoA mutase and related acyl-CoA mutases. J. Biol. Chem. 290, 26882–26898 44. Zheng, S. Q., Palovcak, E., Armache, J.-P., Verba, K. A., Cheng, Y., and Agard, D. A. (2017) MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat. Methods 14, 331–332 36. Punjani, A., Rubinstein, J. L., Fleet, D. J., and Brubaker, M. A. (2017) cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 37. Kaur, S., Gomez-Blanco, J., Khalifa, A. A. Z., Adinarayanan, S., Sanchez- Garcia, R., Wrapp, D., et al. (2021) Local computational methods to improve the interpretability and analysis of cryo-EM maps. Nat. Com- mun. 12, 1240 38. Townsend, P. D., Jungwirth, B., Pojer, F., Bußmann, M., Money, V. A., Cole, S. T., et al. (2014) The crystal structures of apo and cAMP-bound GlxR from Corynebacterium glutamicum reveal structural and dynamic changes upon cAMP binding in CRP/FNR family transcription factors. PLoS One 9, e113265 39. Lazazzera, B. A., Beinert, H., Khoroshilova, N., Kennedy, M. C., and Kiley, P. J. (1996) DNA binding and dimerization of the FeS-containing FNR protein from Escherichia coli are regulated by oxygen. J. Biol. Chem. 271, 2762–2768 40. Vaccaro, F. A., and Drennan, C. L. (2022) The role of nucleoside triphosphate hydrolase metallochaperones in making metalloenzymes. Metallomics 14, mfac030 45. Rohou, A., and Grigorieff, N. (2015) CTFFIND4: defocus estimation from electron micrographs. 216–221 fast and accurate J. Struct. Biol. 192, 46. Cianfrocco, M. A., Wong-Barnum, M., Youn, C., Wagner, R., and Leschziner, A. (2017) COSMIC2: a science gateway for cryo-electron microscopy structure determination. In Proceedings of the Practice and Experience in Advanced Research Computing 2017 on Sustainability, Success and Impact. Association for Computing Machinery, New Orleans, LA 47. Pettersen, E. F., Goddard, T. D., Huang, C. C., Meng, E. C., Couch, G. S., Croll, T. I., et al. (2021) UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 48. Emsley, P., Lohkamp, B., Scott, W. G., and Cowtan, K. (2010) Features and development of Coot. Acta Crystallogr. Sect D: Biol. Crystallogr. 66, 486–501 49. Adams, P. D., Afonine, P. V., Bunkóczi, G., Chen, V. B., Davis, I. W., Echols, N., et al. (2010) PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. Sect D: Biol. Crystallogr. 66, 213–221 41. Wilkins, M. R., Gasteiger, E., Bairoch, A., Sanchez, J.-C., Williams, K. L., Appel, R. D., et al. (1999) Protein identiﬁcation and analysis tools in the ExPASy server. Methods Mol. Biol. 112, 531–552 50. Morin, A., Eisenbraun, B., Key, J., Sanschagrin, P. C., Timony, M. A., Ottaviano, M., et al. (2013) Collaboration gets the most out of software. Elife 2, e01456 42. Niebling, S., Veith, K., Vollmer, B., Lizarrondo, J., Burastero, O., Schiller, J., et al. (2022) Biophysical screening pipeline for cryo-EM grid prepa- ration of membrane proteins. Front. Mol. Biosci. 9, 535 51. Dowling, D. P., Croft, A. K., and Drennan, C. L. (2012) Radical use of Rossmann and TIM barrel architectures for controlling coenzyme B12 chemistry. Annu. Rev. Biophys. 41, 403–427 43. Kimanius, D., Dong, L., Sharov, G., Nakane, T., and Scheres, S. H. W. (2021) New tools for automated cryo-EM single-particle analysis in RELION-4.0. Biochem. J. 478, 4169–4185 52. Sievers, F., Wilm, A., Dineen, D., Gibson, T. J., Karplus, K., Li, W., et al. (2011) Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Mol. Syst. Biol. 7, 539 16 J. Biol. Chem. (2023) 299(9) 105109
10.1016_j.xgen.2023.100356	Article Schizophrenia-associated somatic copy-number variants from 12,834 cases reveal recurrent NRXN1 and ABCB11 disruptions Graphical abstract Authors Eduardo A. Maury, Maxwell A. Sherman, Giulio Genovese, ..., Jonathan Sebat, Eunjung A. Lee, Christopher A. Walsh Correspondence [email protected]. edu In brief Maury et al. leveraged blood-derived SNP-array data across 12,834 schizophrenia cases and 11,648 controls to explore somatic copy-number variants (sCNVs). They found higher early- developmental sCNV incidence in cases compared with controls, along with speciﬁc intragenic events in NRXN1 and ABCB11 that could potentially contribute to SCZ disease. Highlights d Somatic copy-number variants are more common in SCZ cases than in controls d Recurrent somatic deletions of NRXN1 exons 1–5 in SCZ cases d Recurrent intragenic deletions of ABCB11 in SCZ cases d ABCB11 is speciﬁcally enriched in a subset of dopaminergic neurons in human brain Maury et al., 2023, Cell Genomics 3, 100356 August 9, 2023 ª 2023 The Author(s). https://doi.org/10.1016/j.xgen.2023.100356 ll ll OPEN ACCESS Article Schizophrenia-associated somatic copy-number variants from 12,834 cases reveal recurrent NRXN1 and ABCB11 disruptions Eduardo A. Maury,1,2,3 Maxwell A. Sherman,4 Giulio Genovese,3,5,6 Thomas G. Gilgenast,7 Tushar Kamath,6,8 S.J. Burris,6 Prashanth Rajarajan,9 Erin Flaherty,9 Schahram Akbarian,9 Andrew Chess,9 Steven A. McCarroll,3,5 Po-Ru Loh,3,4 Jennifer E. Phillips-Cremins,7 Kristen J. Brennand,9,10 Evan Z. Macosko,6,11 James T.R. Walters,12 Michael O’Donovan,12 Patrick Sullivan,13 Psychiatric Genomic Consortium Schizophrenia and CNV workgroup16, Brain Somatic Mosaicism Network16 Jonathan Sebat,14 Eunjung A. Lee,1,3 and Christopher A. Walsh1,3,15,17,* 1Division of Genetics and Genomics, Manton Center for Orphan Disease, Boston Children’s Hospital, Boston, MA, USA 2Bioinformatics & Integrative Genomics Program and Harvard/MIT MD-PHD Program, Harvard Medical School, Boston, MA, USA 3Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, MA, USA 4Brigham and Women’s Hospital, Division of Genetics & Center for Data Sciences, Boston, MA, USA 5Department of Genetics, Harvard Medical School, Boston, MA, USA 6Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard, Cambridge, MA, USA 7Department of Bioengineering, University of Pennsylvania, Philadelphia, PA, USA 8Harvard Graduate Program in Biophysics, Harvard University, Cambridge, MA, USA 9Nash Family Department of Neuroscience, Friedman Brain Institute, Department of Genetics & Genomics, Icahn Institute of Genomics and Multiscale Biology, Department of Psychiatry, Pamela Sklar Division of Psychiatric Genomics, Icahn School of Medicine of Mount Sinai, New York, NY, USA 10Departments of Psychiatry and Genetics, Yale School of Medicine, New Haven, CT, USA 11Massachusetts General Hospital, Department of Psychiatry, Boston, MA, USA 12MRC Centre for Neuropsychiatric Genetics and Genomics, Division of Psychiatry and Clinical Neurosciences, Cardiff University, Cardiff, Wales 13Department of Genetics, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA 14University of California San Diego, Department of Psychiatry, Department of Cellular & Molecular Medicine, Beyster Center of Psychiatric Genomics, San Diego, CA, USA 15Howard Hughes Medical Institute, Boston Children’s Hospital, Boston, MA, USA 16Further details can be found in the supplemental information 17Lead contact *Correspondence: [email protected] https://doi.org/10.1016/j.xgen.2023.100356 SUMMARY While germline copy-number variants (CNVs) contribute to schizophrenia (SCZ) risk, the contribution of so- matic CNVs (sCNVs)—present in some but not all cells—remains unknown. We identiﬁed sCNVs using blood- derived genotype arrays from 12,834 SCZ cases and 11,648 controls, ﬁltering sCNVs at loci recurrently mutated in clonal blood disorders. Likely early-developmental sCNVs were more common in cases (0.91%) than controls (0.51%, p = 2.68e(cid:1)4), with recurrent somatic deletions of exons 1–5 of the NRXN1 gene in ﬁve SCZ cases. Hi-C maps revealed ectopic, allele-speciﬁc loops forming between a potential cryptic promoter and non-coding cis-regulatory elements upon 50 deletions in NRXN1. We also observed recurrent intragenic deletions of ABCB11, encoding a transporter implicated in anti-psychotic response, in ﬁve treat- ment-resistant SCZ cases and showed that ABCB11 is speciﬁcally enriched in neurons forming mesocortical and mesolimbic dopaminergic projections. Our results indicate potential roles of sCNVs in SCZ risk. INTRODUCTION De novo and rare germline copy-number variants (gCNVs) contribute to up to 5.1%–5.5% of schizophrenia (SCZ) cases, with relatively large effect sizes.1 These gCNVs are usually in- herited or represent de novo events thought to arise during gametogenesis. Most gCNVs involve several genes, making it difﬁcult to pinpoint speciﬁc causative genes. A notable excep- tion is deletion of NRXN1, which encodes a presynaptic adhe- sion protein and has been suggested to have a role in SCZ along with other synaptic genes.2 Somatic copy-number variants (sCNVs) present in only a frac- tion of cells in the body, are increasingly implicated in neuropsy- chiatric disease.3–8 For example, a recurrent, large sCNV of chromosome 1q has been repeatedly observed in focal epileptic brain malformations,9–11 while blood samples from autism Cell Genomics 3, 100356, August 9, 2023 ª 2023 The Author(s). 1 This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). ll OPEN ACCESS A B Article C D E F G H 2 Cell Genomics 3, 100356, August 9, 2023 (legend on next page) Article ll OPEN ACCESS spectrum disorder (ASD)3 showed enrichment of large (>4 Mb) sCNVs, with sCNV size positively correlated with phenotypic severity. The overlap in the genetic architecture of ASD and SCZ12 suggests the hypothesis that sCNVs may have similar roles in SCZ liability. Since sCNVs are less common than germline gCNVs, large da- tasets must be analyzed to assess their contribution to disease, but such large genotyping datasets are generally only available from blood-derived single-nucleotide polymorphism (SNP)-array data created for genome-wide association studies (GWASs), which creates two challenges. The ﬁrst challenge is that these ar- rays only capture the earliest developmental events, present in a relatively large fraction of cells13,14 and hence also likely to be shared in brain cells and other tissues. Prior studies have shown that non-oncological somatic variants present in more than (cid:3)1%–3% of cells in a tissue are typically shared in all develop- mental lineages in a mosaic fashion.14–16 The mosaic fraction of variants in blood exhibited a linear relationship with the mosaic fraction in other tissues,17 suggesting that studying highly mosaic variants in blood might reﬂect, to an extent, somatic vari- ation in other tissues such as brain. The second challenge in assessing sCNVs in blood is the increasing recognition that aging and environmental exposures are correlated with sCNVs that are restricted to blood, which are associated with leukemia or pre-cancerous conditions such as clonal hematopoiesis of indeterminate potential (CHIP).5,18,19 However, CHIP-related sCNVs have now been extensively characterized in dozens of studies in terms of size and mosaic fraction and found to occur at recurrent chromo- somal locations that disrupt speciﬁc driver genes,18–22 allowing sCNVs at these loci to be ﬁltered to identify non-CHIP, early- developmental sCNVs that may be associated with SCZ. In this study, we analyzed SNP-array data from 12,834 cases and 11,648 controls from the Psychiatric Genomic Consortium (PGC) SCZ cohort using a widely utilized, highly sensitive algo- rithm that leverages haplotype information to detect sCNVs in blood.3,18,19 We additionally used recent knowledge of the genomic loci of blood events22 to rigorously ﬁlter candidate var- iants that likely originated from CHIP. We observed an excess of non-CHIP-related sCNVs in SCZ compared with controls and discovered recurrent sCNVs, including recurrent NRXN1 so- matic deletions of exons 1–5 and recurrent intragenic events at ABCB11 gene as well. Taken together, these data suggest that potential roles of sCNVs in the genetic architecture of SCZ merit further study. RESULTS Potential enrichment of non-CHIP sCNV in SCZ cases sCNVs were identiﬁed using the MoChA18,19 software on 26,186 blood-derived SNP arrays from the PGC2 SCZ cohort23 (Fig- ure 1A). We removed gCNVs previously identiﬁed in subjects of this cohort.23 Samples that showed signs of contamination, or sCNVs whose copy-number state was not conﬁdently deter- mined, were excluded (STAR Methods). This quality control (QC) led to the identiﬁcation of 1,341 candidate sCNV, including many presumably related to CHIP, and a subset that may poten- tially be associated with SCZ. We identiﬁed 1,143 events likely to have arisen from CHIP, based on their chromosomal location at recurrent CHIP regions and resemblance to known CHIP events. We used these CHIP events, which typically have low cell fraction,18,22 to compare the performance of MoChA in our dataset with prior studies. The events identiﬁed as CHIP in our initial call set followed a similar distribution of cell fraction (CF) and length compared with well-known CHIP events from the UK Biobank19,22 (Fig- ure 1B, left panel). This similarity suggests that our pipeline iden- tiﬁes sCNVs in varied patient datasets with high conﬁdence. While there is variation across cohorts for the number of CHIP events identiﬁed, on average the rates of CHIP events were similar in cases compared with controls. Pooling all the CHIP events did not show a signiﬁcant difference in CHIP events in SCZ compared with controls (Fisher’s exact test odds ratio [OR], 1.08; 95% conﬁdence interval [CI] [0.81–1.46], p = 0.618; Figure 1C). Performing meta-analysis to account for potential batch heterogeneity, similarly, revealed no signiﬁcant decrease in CHIP events across cases and controls (one-sided Fisher’s exact test, Liptak’s combined p value = 0.9; Figure S1A). The mean number of CHIP events across cohorts for SCZ samples was 0.046 (SE = 0.009), similar for controls with 0.041 (SE = 0.008). Since we do not have age information on all samples, it is possible that any difference in CHIP burden in SCZ and con- trols might be masked by differential age distribution or other environmental factors. Nevertheless, this result suggests similar sCNV detection sensitivity in cases compared with controls in our dataset. We next ﬁltered likely CHIP-related events and identiﬁed a sub- set of early-developmental sCNVs, most present in a high CF. Speciﬁcally, we removed all copy-neutral loss of heterozygosity (CN-LOH), loci commonly altered in the immune system (e.g., major histocompatibility locus [MHC]) and other known common Figure 1. Somatic CNV burden in SCZ (A) Schematic of sCNV calling and ﬁltering. (B) Left: scatterplot and marginal distributions of length and CF of sCNVs identiﬁed as CHIP vs. non-CHIP. Middle: distribution of canonical CHIP events in sCNVs identiﬁed as CHIP in our call set compared with CHIP events identiﬁed in the UK Biobank.19 Right: cumulative distributions of CF of CHIP vs. non-CHIP events; p value from Kolmologorov-Smirnov test. (C) Odds ratio plots comparing sCNV burden across different CHIP ﬁltering stages. Odds ratios and 95% CI were derived from Fisher’s exact test. CHIP variants were deﬁned as those overlapping canonical CHIP events.22 (D) Trident plot of ﬁnal call set. Each point represents an event, with colors and shapes indicating subject’s diagnoses and array type. (E) Percentage of individuals with R1 sCNV in cases and controls across different minimum CF thresholds. Dots represent mean fraction and lines represent 95% CI from the binomial distribution using Wilson’s score interval with Newcombe modiﬁcation; p values calculated with two-sided Fisher’s exact test. (F) Histogram of sCNV size (log10 scale) in cases and controls. (G) Boxplots of sCNV CFs in cases vs. controls. (H) Boxplots of the number genes per megabase of sCNVs in cases and controls. Cell Genomics 3, 100356, August 9, 2023 3 ll OPEN ACCESS CHIP loci19,21 and ﬁltered outlier samples with multiple events (>5 sCNVs) (Figure 1A) (STAR Methods). Our non-CHIP events show a different distribution of CF and length compared with events identiﬁed as CHIP (Figure 1B, middle panel). sCNVs that occur early in development are clonally shared across multiple tissues and are thus expected to be present at larger CF than those occurring through CHIP alone. Reassuringly, variants ﬁltered as potential CHIP exhibited signiﬁcantly lower CF than non-CHIP events (Wilcoxon Rank-Sum test p = 6.4e(cid:1)11) (Figure 1B). This difference suggests that our ﬁltering reliably removes likely CHIP events, although some bona ﬁde early-developmental sCNVs may be ﬁltered out as well, especially those coming from CN-LOH events. While we found equivalent rates of CHIP in SCZ samples and controls (Figure 1C), stepwise removal of likely CHIP variants showed increasing enrichment of the remain- ing sCNVs in SCZ compared with controls, with the highest effect size once all the CHIP-related events were removed (Figure 1C). While non-recurrent or small events may be difﬁcult to detect, Loh et al.18 demonstrated detection sensitivity for events as small as 100 kb. There is an inverse-square relationship between event size and CF, such that, at a CF of (cid:3)10%, events > 1 Mb are detectable, while sCNVs > 100 Mb need to be present in (cid:3)1% of cells to be detected.18 Since CHIP events tend to occur at lower CF,18,19,21,22 consequently MoChA would have sensitivity to detect only larger events. On the other hand, it is expected from the modeling of MoChA18 and the inverse-square relation- ship that early-developmental events, such as those predicted to have occurred in some individuals in our study, have a biological higher CF, and hence that MoChA should have sensitivity to detect them even if they are smaller. Whole-genome sequencing supports presence of sCNV likely non-CHIP sCNV using 40–603 We conﬁrmed several whole-genome sequencing (WGS) in ﬁve individuals. We de- tected a probable 650-kb somatic deletion in one individual with a predicted CF of 52%, which was supported by simple in- spection of the Integrative Genomics Viewer (IGV) read pileup (Figure S2). The CF estimated from WGS was (cid:3)50%, closely matching the MoChA estimate. We also called a large 26-Mb somatic deletion in 9q21.11-9q22.2 (hg19 coordinates, chr9:71033538-97246817) with an estimated CF of 43% in another individual. Running MoChA on WGS also supported a similar somatic deletion overlapping the original event in the same individual (hg19 coordinates chr9, 38767760–97259994) with a CF of 46% and with a size larger than seen on the SNP array, attributable to the event extending into the centromeric re- gion that is not well represented on SNP arrays.24 WGS also conﬁrmed both CFs and breakpoints for three of three much smaller NRXN1 deletions presented in detail below. Since the data for this study was generated across different countries, pre- cluding access to DNA for more widespread validation of MoChA calls, we applied MoChA conservatively, only calling variants (for both cases and controls) in a range of size and CFs that have been shown in prior papers with this algorithm to have a negli- gible false-discovery rate.3,18–20,25 However, further studies will be required to provide more precise estimates of prevalence of sCNV in SCZ, and rate comparisons in our study should be inter- preted with caution. 4 Cell Genomics 3, 100356, August 9, 2023 Article Analysis of putative early sCNVs in SCZ and controls sCNVs not related to CHIP occurred in a small but signiﬁcant fraction of SCZ cases. From the initial 13,464 SCZ cases and 12,722 controls, a total of 12,834 cases and 11,648 controls re- mained after QC. The ﬁnal non-CHIP sCNV call set consisted of 198 events in 178 individuals, made up of 127 losses and 70 gains (Tables S1and S2; and Figure 1D). These events ranged in CF from 1.10% to 63.8% (median, 21.1%), and ranged in size from 10.7 kb to 95.3 Mb (median, 686.0 kb). The high CF of events in samples without a blood cancer diagnosis suggests that these somatic variants might have arisen during early-devel- opmental stages.13,14,17 The percentage of individuals with at least one sCNV was 0.91% in SCZ and 0.51% in controls (OR, 1.78; 95% CI, 1.29–2.47; two-sided Fisher’s exact test, p = 2.68e(cid:1)4) (Figure 1E). The sCNV incidence in controls was com- parable with unaffected siblings in an earlier study (0.51% vs. 0.54%),3 while our estimates in SCZ were higher compared with the ASD cases from the same study (0.91% vs. 0.58%).3 This higher rate most likely reﬂects sensitivity improvement in the pipeline since the earlier study, although we cannot rule out differential sources of artifact or biological effect (STAR Methods). Prior analyses with MoChA have estimated burdens of sCNV (with CF > 10%) in blood samples of individuals with no history of hematologic cancer of 3.2% in the UK Biobank, 5.2% in the Mass General Brigham Biobank, 5.9% in FinnGen, and 1.3% in Biobank Japan.20 These numbers are all larger than what we observe in our SCZ cohort and likely reﬂect our ﬁltering of CHIP variants and/or differential environmental exposures. To rule out potential residual CHIP events in our call set contributing to the difference in prevalence of sCNVs, we per- formed the burden test using different minimum CF cutoffs, with higher CF cutoff being less likely to be CHIP events and more likely to be early-developmental events. There remained a statistically signiﬁcant enrichment in SCZ through several ranges, even when events were split into losses and gains (Fig- ure 1E). We further accounted for potential batch heterogeneity (Figure S1A) using meta-analysis across each study batch con- taining both cases and controls, obtaining a Liptak’s combined p value of 0.032 using a one-sided Fisher’s exact test. To further conﬁrm that the enrichment observed was not driven by CHIP events, we removed all variants from samples with suspected CHIP variants (30 variants removed). With this smaller call set, we still obtained a signiﬁcant enrichment in SCZ cases for sCNV compared with controls (Fisher’s exact test, OR, 1.64; 95% CI [1.16:2.33]; p = 0.0041). In contrast with previous ﬁndings in ASD,3 sCNVs in SCZ cases were of similar size compared with control after account- ing for different arrays/cohorts by mixed-effect modeling (p = 0.26) (Figure 1F). These events were also present at similar CF in cases compared with controls (p = 0.986; Figure 1G). There was also no detectable difference in gene density (p = 0.08; Fig- ure 1H). These trends were observed across the different batches as well (Figure S1B–D). In contrast to gCNV,1,23 sCNV did not show overall gene-set enrichment for the top 20% ex- pressed brain genes (p = 0.14), synaptic genes (p = 0.12), or hap- loinsufﬁcient genes as measured by a probability of of being loss-of-function intolerant (pLI) score >0.9026 (p = 0.54). We Article ll OPEN ACCESS A B C Figure 2. Somatic CNVs differ in size, gene content, and location from gCNVs in SCZ (A) Boxplot of event length in SCZ in somatic and germline state. (B) Plot of number of genes affected per megabase; p values for (A) and (B) were calculated using mixed-effect model log-normal and negative binomial regression, respectively, with batch as a random effect. (C) Bar plots showing percentage of CNVs in each category that overlapped recurrent germline rare CNV regions in SCZ across three different minimum recurrence thresholds. did not detect events in the top 10 genes related to SCZ by the SCHEMA consortium27 or the presence of two-hit events (germ- line + sCNVs) in our dataset. Some sCNV overlapped cytobands previously implicated in SCZ but showed distinctive features. While one SCZ case had a 4.1-Mb somatic deletion in cytoband 16p11.2, it was not only signiﬁcantly larger than the canonical germline 16p11.2 deletions (<600 kb) observed in SCZ and ASD23,28 but also the mosaic deletion did not overlap the canonical proximal or distal events (Figure S3A). We also observed one SCZ case with a somatic deletion in the 22q11.21 locus that was signiﬁcantly smaller (686 kb) than the recurrent germline 22q11.21 deletions observed in SCZ (2.35 Mb) (Figure S3B). The mosaic 22q11 deletion we observed, however, overlapped the genes TBX1 and COMT, which have been suggested as key genes driving some of the phenotypic effects and SCZ risk of germline 22q11 deletion.29,30 Predicted sCNV are larger and affect more gene-dense regions compared with gCNVs Comparison of the genomic features of sCNVs with rare (minor population allele frequency <0.5%) gCNVs calls of SCZ cases from the arrays used in our current study23 showed that sCNVs were larger (fold change, 4.57; 95% CI, 3.76–5.48; mixed-effect log-normal regression p < 2e(cid:1)16) and involved more genes (fold change, 1.27; 95% CI, 1.03–2.56; mixed-effect negative bino- mial regression p = 0.027) (Figures 2A and 2B). We observed that genomic regions affected by rare gCNVs present in at least ﬁve SCZ cases overlapped 43.6% of all the gCNVs, whereas these same regions overlapped only 4.48% of SCZ sCNVs (Fig- ure 2C). This difference in genomic regions persisted throughout for rare gCNVs present at different minimum recurrence cutoffs (Figure 2C). These ﬁndings suggest that, with sufﬁcient statistical power, mosaic events might offer additional new insights into different risk regions of the genome. Recurrent, intragenic deletions in NRXN1 observed in SCZ Six individuals showed somatic deletions in cytoband 2p16.3 affecting only the NRXN1 gene, at remarkably stereotyped and distinctive regions of the gene. The size of these events ranged from 105 to 534 kb, with CF ranging from 13.8% to 43.1%, sug- gesting that they occurred early in development. One deletion was limited to intron 5 (Figure 3A) and is of uncertain disease sig- niﬁcance since multiple germline deletions of this intron have been reported in control individuals.23 In contrast, the remaining ﬁve 2p16.3 deletions consistently removed exons 1–5 of NRXN1a while leaving exon 6 and the rest of the gene intact. This stereotyped ﬁve-exon deletion contrasts with germline de- letions in NRXN1, previously implicated in SCZ,23,31 which show highly variable breakpoints and relationships to NRXN1 exons.23,32,33 Therefore, the recurrent, mosaic deletion of the same exons 1–5 in all ﬁve exonic deletions would seem to de- mand a speciﬁc mechanistic explanation. To further assess the prevalence of somatic NRXN1 deletions, we re-ran MoChA with a more lenient threshold and checked whether NRXN1 copy-number variants (CNVs) identiﬁed in the original PGC study23 as germline might in fact be somatic. This strategy re- vealed an NRXN1 deletion previously identiﬁed as germline, Cell Genomics 3, 100356, August 9, 2023 5 ll OPEN ACCESS Article A B F D E C G Figure 3. Somatic deletions of NRXN1 exons 1–5 (A) Adapted GenomeBrowser view of seven somatic deletions of NRXN1. The alpha promoter and in-frame ATG/methionine sites on exons are annotated for NRXN1. Histone marks were obtained from Roadmap epigenomics tracks.34 Potential cryptic promoter/enhancer is marked by a red box. Gray horizontal bar indicates CNV previously called germline that was found to be somatic. (B) Prevalence of somatic deletions of NRXN1 exons 1–5 in SCZ, controls, and UK Biobank; p values were estimated using two-sided Fisher’s exact test, and 95% CIs were obtained using the Wilson’s score interval with Newcombe modiﬁcation. (C) Histogram of the distribution of number of overlaps of NRXN1 exons 1-5 from randomly shufﬂing the discovered NRXN1 sCNVs across the NRXN1 locus. The blue dashed line is the observed number of overlaps, which is equal to six. (D) IGV plots of the deletions of three SCZ subjects with somatic deletions in NRXN1 exons 1–5 from WGS. For clarity, not all the reads are shown. (legend continued on next page) 6 Cell Genomics 3, 100356, August 9, 2023 Article ll OPEN ACCESS with an estimated CF of 41%, consistent with being somatic. This variant appeared to overlap exons 4–5 for NRXN1 (Fig- ure 3A), although its exact boundaries are uncertain. Comparing the burden of NRXN1 somatic deletions in SCZ cases vs. controls revealed signiﬁcant enrichment in cases (two-sided Fisher’s exact test p = 0.032, exonic only; p = 0.016, exonic + intronic; Figure 3B). Previously generated sCNV calls from the UK Biobank18,19 identiﬁed two persons without history of psychiatric disorder (of (cid:3)500,000 individuals) with similar sCNV breakpoints affecting exons 1–5 in NRXN1. Although the arrays used in the UK Biobank have different sensitivity compared with arrays used in this study, they should have comparable sensitivity to detect these large events at CF > 10%.18 Consequently, while we cannot fully rule out batch effect bias, combining our results with the UK Biobank suggest an enrichment of exon 1–5 NRXN1 deletions in the somatic state in SCZ samples (OR, 117.08; 95% CI, 20.91–1165.84; Fisher’s exact test p = 6.57e(cid:1)9; Figure 3B). To further assess whether we could have observed ﬁve overlaps of exons 1–5 by chance, we performed a bootstrap test by randomly shufﬂing regions of length equal to the seven NRXN1 sCNV that we discovered across the NRXN1 locus and computed the number of random overlaps with exons 1–5. We observed that ﬁve random overlaps of exactly exons 1–5 was a highly unlikely event (p < 0.0001; see STAR Methods; Figure 3C). A similar study with a similar pipeline and dataset to this study3 on ASD and control samples did not detect somatic deletions in NRXN1 overlapping exons 1–5, sug- gesting relative speciﬁcity of this event to SCZ. We were able to obtain 403 WGS from three NRXN1a deletion cases processed at the Broad Institute, conﬁrming that each event removed exons 1–5 of the gene with estimated CFs of 42.4%, 33.3%, and 32.4%, as expected (Figure 3D), and deﬁning their breakpoints at base-pair resolution. WGS analysis showed that none of the exact NRXN1 sCNVs breakpoints were recurrent or overlapped known interspersed repeats or low- complexity DNA sequences, which could have predisposed this genomic region to genomic instability. Further breakpoint analysis of these NRXN1 sCNVs using previ- ously established classiﬁcation criteria35,36 (Figure 3E) suggested diverse mechanisms of formation. One event had only 1 bp of mi- crohomology (MH), suggesting that this event arose via non-ho- mologous end-joining repair (NHEJ). Another event had a 3-bp MH, implicating an alternative end-joining repair mechanism (alt- EJ). The last event had no MH but revealed 8 bp inserted at the breakpoint, small enough to have occurred by non-template directed repair associated with NHEJ, although it is also possible that a fork-stalling template switching mechanism might have occurred,37 but this mechanism tends to produce insertions >10 bp and usually occurs where some microhomology exists.35 Taken together, these results suggest that the somatic deletions of NRXN1 that we observed do not disrupt recurrent exons due to instability of the genomic region around the events. NRXN1 deletions suggest a potential cryptic promoter in human induced neurons The absence of a genomic mechanism for the recurrent somatic deletions in NRXN1a suggests an alternative hypothesis, that the recurrence reﬂects an unknown but speciﬁc effect of these deletions on NRXN1 gene function. These sCNVs overlap the NRXN1a promoter and the ﬁrst in-frame ATG transcription start site, which would be expected to disrupt transcription of the full alpha isoform (Figure 3A) while leaving downstream beta and gamma isoforms intact. Intriguingly, the somatic deletions leave intact H3K4Me1 histone marks that lie just 50 from exon 6, which contains an in-frame ATG (Figure 3A). These features might indi- cate a cryptic promoter or enhancer adjacent to the in-frame ATG in exon 6, potentially producing an N-terminal truncated NRXN1a. This truncated protein would lack the signal peptide required for shuttling to the cell surface, potentially causing abnormal trafﬁcking. Similar germline NRXN1 deletions have been shown to cause accumulation of the NRXN1 intracellular binding protein CASK in human induced pluripotent cells (iPSCs) from SCZ patients.38 To further explore functional effects of somatic deletions in the 50 end of NRXN1, we generated Hi-C data from neurons differen- tiated from human iPSCs (hiPSCs) containing heterozygous germline deletions in the 50 end (exons 1–2) and compared them with an iPSC line that had no germline deletion in NRXN1 (STAR Methods). Unphased Hi-C heatmaps in iPSC neurons showed that somatic deletions affecting exons 1–5 all fully over- lap the topologically associating domain (TAD) boundary co- localized with the alpha promoter (Figure 3F). Recently, disrup- tion of TAD boundaries by germline structural variants have been associated with developmental disorders as well as SCZ.39,40 These observations together suggest that 50 NRXN1 deletions might disrupt the structural integrity of the TAD bound- ary in SCZ and could result in ectopic enhancer-promoter mis- wiring and dysregulated gene expression. To investigate possible 3D genome miswiring due to NRXN1 de- letions, we generated allele-speciﬁc, phased Hi-C maps in both control as well as deletion-carrying SCZ iPSC-neurons (STAR Methods). Surprisingly, we observed the de novo formation of an ectopic looping interaction (Figure 3G, green circle) between exon 6 of NRXN1 (Figure 3G, blue star) and a putative non-coding cis-regulatory element upstream of the NRXN1 alpha promoter (Figure 3G, purple star). This ectopic loop appeared to be speciﬁc to the deletion-harboring allele of the sample bearing a heterozy- gous deletion spanning the alpha promoter at the 50 end of NRXN1 (973FB) and was not observed on either allele in samples that lacked the deletion (2607FB). Because the interaction spans the deleted region, we hypothesize that the deleted region contains an element with some degree of boundary function pre- venting this loop from forming normally. Consistent with our hy- pothesis, the frequency of non-speciﬁc interactions increased across the boundary only on the NRXN1-deleted allele, suggesting (E) Breakpoint analysis schematic showing observed insertions and microhomology at breakpoints of NRXN1 sCNVs along with event length. NHEJ, non-ho- mologous end-joining repair; Alt-EJ, alternative end joining. (F) Unphased Hi-C heatmap for hiPSC-derived neurons with and without 50 (exon 1and 2) deletions. Black bars indicate regions of somatic NRXN1 deletions. (G) Phased Hi-C heatmaps for hiPSC-derived neurons. Green circles indicate areas of higher signal with 50 deletion of NRXN1 in the affected allele. Black bar indicates germline NRXN1 deletion of exons 1and 2. RE, regulatory element. Cell Genomics 3, 100356, August 9, 2023 7 ll OPEN ACCESS Article A B C D Figure 4. Somatic CNVs in treatment-resistant SCZ subjects overlap the ABCB11 gene (A) Adapted GenomeBrowser view of ﬁve somatic deletions and one somatic duplication of ABCB11. Protein domains of interest overlapped by the sCNVs have orange font. (B) PyMOL schematic of the ABCB11 protein shows HAX1 protein interaction region and the ABC transporter 1 domain, which are affected by somatic deletions of ABCB11. The protein is on an ‘‘inner-open’’ conformation, not bound to ATP. (C) Prevalence of intragenic sCNV in ABCB11 in SCZ and controls. (D) Prevalence of intragenic sCNV in ABCB11 in CLOZUK cohort samples. For (C) and (D), p values were estimated using two-sided Fisher’s exact test, and 95% CIs were obtained using the Wilson’s score interval with Newcombe modiﬁcation. allele-speciﬁc compromise of TAD structural integrity in SCZ (Fig- ure 3G). Together, a working model is that de novo looping interac- tion in 50 NRXN1 deletions in SCZ connecting exon 6 to a putatively regulatory element could promote spurious pathological tran- scripts initiating at exon 6, although other alternative explanations remain as well. Recurrent sCNVs in the ABCB11 gene observed in treatment-resistant SCZ cases We identiﬁed six SCZ cases with focal sCNVs within the ABCB11 gene (ﬁve deletions and one gain; Figure 4A), which has previ- ously been associated with anti-psychotic response.41,42 These sCNVs were all smaller than average, from 10.5 to 35.4 kb, but also with high CFs (18.3%–26.8%), suggesting that they also occurred early in development. ABCB11 encodes a member of the ATP-binding cassette (ABC) transporter superfamily and has a key role in transporting bile acids across the cell mem- brane42 in hepatocytes, the cells involved in a wide range anti- psychotic metabolism. Biallelic loss-of-function variants in ABCB11 result in severe pediatric-onset liver disease, with many patients developing malignancies or pathological compli- cations within the ﬁrst decade of life.43–46 All the ABCB11 sCNVs overlapped the ABC transporter 1 domain and the domain responsible for interaction with the HAX1 protein (Figure 4B), the latter facilitating internalization of ABCB11 via clathrin-medi- ated endocytosis.47,48 Consequently, deletions might not only alter the protein’s function by altering the transporter domains but also prevent removal of ABCB11 from the cell surface, perhaps leading to a dominant-negative loss of function. Since the sCNVs in ABCB11 do not overlap the gene’s promoter and there are in-frame ATG sites in downstream exons 19 and 20, a truncated protein could be produced. The consequences of the somatic duplication event are less clear. We also note that four out of ﬁve deletions and the duplication overlap one of the transmembrane domains, further supporting the idea that these sCNVs might have a detrimental effect on ABCB11 function. The case-control enrichment of ABCB11 sCNVs was statistically sig- niﬁcant (two-sided Fisher’s exact test, p = 0.03; Figure 4C). All six cases with ABCB11 sCNV came from batches of CLOZUK,49 a treatment-resistant SCZ (TRS) cohort. These 8 Cell Genomics 3, 100356, August 9, 2023 Article A B ll OPEN ACCESS C D (legend on next page) Cell Genomics 3, 100356, August 9, 2023 9 ll OPEN ACCESS samples were obtained from individuals with TRS, taking cloza- pine, and thus subject to standard blood monitoring for this drug.50 Even though the CLOZUK samples constituted a signiﬁ- cant portion of our study, observing six cases from only this cohort represents a statistically signiﬁcant enrichment (two- sided Fisher’s exact test, p = 0.00079, and p = 0.015 for SCZ only; Figure 4D). ABCB11 sCNVs were not found in any previous analyses of healthy individuals from the UK Biobank and Biobank Japan.19,21 Thus, these variants might plausibly regulate either SCZ liability or the samples with ABCB11 sCNV, only two (one gain and one loss) were available in repetitive regions for WGS. The predicted breakpoints fall (short interspersed nuclear elements [SINE]) (Figure S4), making it difﬁcult to identify exact breakpoints, although the presence of these repetitive sequences suggests a potential mechanism of somatic deletion through microhomology. It is also possible that this part of the genome is unstable, since the ABCB11 gene has a signiﬁcant burden of Alu-family members ﬂanking exons as quantiﬁed by the AluAluCNV predictor score of 0.46, potentially implicating Alu-Alu-mediated rearrangement (AAMR).51 response. Of treatment Combining the ABCB11 somatic deletions we observed in our SCZ cases with germline deletions identiﬁed as part of the phase 2 PGC gCNV dataset revealed robust overlap between the mosaic deletions we detected and those present in separate SCZ cases in the germline state. There were ﬁve SCZ cases with gCNVs at the ABCB11 locus, with three of them coming from the CLOZUK cohort (Figure S5). We were not able to obtain clinical data to determine whether the remaining two cases had TRS. Although six controls showed germline ABCB11 deletions, these events tended to cluster downstream of the SCZ gCNV and sCNV variants (Figure S4). SCZ risk association analyses combining germline and somatic deletions of ABCB11 revealed a nominally statistically signiﬁcant association of sCNV at the HAX1 interaction site and ABC transporter 1 site (peak associa- tion, p = 1.4e(cid:1)4), although this did not meet the threshold (p = 8.3e(cid:1)8) for genome-wide signiﬁcance. ABCB11 is enriched in human dopaminergic neurons residing in the dorsal tier of the substantia nigra pars compacta While ABCB11 has been primarily studied in hepatocytes, we explored whether it might show expression in human brain. In publicly available single-nuclei RNA-sequencing data from three brain regions—cortex, caudate nucleus, and substantia nigra Article pars compacta (SNpc)52,53— across the 151 cell types surveyed from these regions, we found that two dopaminergic (DA) popu- lations in the SNpc showed the strongest expression of ABCB11, along with expression in subpopulations of layer 5 excitatory neurons in motor cortex (Figure 5A). The two ABCB11-express- ing substantia nigra (SN) DA populations also showed strong expression of CALB1 in a recent survey of human midbrain DA neurons52 (Figure 5B). Interestingly, calbindin-positive DA neu- rons reside in the dorsal tier of the SNpc, which projects to the ventral striatum, amygdala, as well as to cortical areas through the mesolimbic and, more preferentially, the mesocortical path- ways (Figure 5C).52,54,55 These projections have been repeatedly implicated in SCZ pathology and treatment response.56 We validated the expression of ABCB11 in human DA neurons of the SNpc with single-molecule ﬂuorescence in situ hybridiza- tion (smFISH) across the midbrain of a postmortem neurotypical control. In support of the small nuclear RNA sequencing (snRNA- seq) data, we found multiple TH+ (tyrosine hydroxylase, the gene encoding the rate-limiting enzyme for dopamine production)/ CALB1+/ABCB11+ cells residing in the midbrain pars compacta region (Figure 5D). We also noted ABCB11 expression in EXC L5 FEZF2 CSN1S1 and EXC L3-5 FEZF2 ASGR2 cells (Figure 5B), which correspond to Betz cells in the primary motor cortex53; however, the relationship of this cell type to SCZ is less clear and we cannot rule out that ABCB11 might be present in other related excitatory layer 5 neurons. DISCUSSION We show that somatic CNVs may contribute a small but signiﬁ- cant part of the genetic architecture of SCZ, mirroring previous ﬁndings of rare germline and de novo CNVs,1,23 but involving a much more modest proportion of cases. The estimated excess burden of sCNV in SCZ would be 0.4%, which represents a pre- liminary estimate; we are limited to detecting events with large enough CFs to be present as mosaics in different tissues such as blood and are not able to assess events that might be restricted to brain, in addition to limitations of sequencing valida- tion to better characterize potential sources of artifact. Future studies with additional orthogonal validation, accounting for he- reditary stratiﬁcation and germline background risks, might pro- vide more accurate estimates of the risk carried by sCNVs in SCZ. In this study, we also report the discovery of ﬁve SCZ cases with mosaic deletions of exons 1–5 that also cover the promoter of NRXN1a. Deletions of these exons were present in only two Figure 5. Expression of ABCB11 in human brain DA neurons (A) Boxplot of log-normalized ABCB11 expression across three brain regions. Each point indicates an individual sample. Cell type annotations obtained from Kamath et al. (for SN and dorsal striatum samples) and Bakken et al. (for M1 motor cortex samples). Ex/Exc, excitatory neurons; Inh, inhibitory interneurons; Olig, oligodendrocytes; MG, microglia/macrophages; Endoﬁbro, endothelial cells/pericytes; DRD1, direct spiny projection neurons; DRD2, indirect spiny projection neurons; Astro, astrocytes; OPC, oligodendrocyte precursor cells. (B) Left: uniform manifold approximation projection (UMAP) of low-dimensional embedding of 15,684 DA neurons from eight neurotypical donors. Points are colored by clusters obtained from Kamath et al.52 Right: dot plot of normalized ABCB11 expression across 10 DA subtypes. (C) Schematic of major DA projections from dorsal and ventral streams of SN pars compacta to cortical areas associated with SCZ. (D) Top row: tiled image of a postmortem midbrain tissue section with increasing magniﬁcation. Right: white dashed box corresponds to approximate location of middle image and similarly, for the middle image, white boxed arrow with the right image. Red outline indicates approximate ventral tier and blue is approximate dorsal tier. Bottom row: representative image of smFISH of human DA neurons. Scale bar, 15 mm. Colors are DAPI (gray), TH (green), CALB1 (yellow), and ABCB11 (magenta). Outline indicates approximate boundary of DA neuron as identiﬁable by TH. RN, red nucleus; CP, cerebral peduncles; cartesian arrow labels are D, dorsal, V, ventral; M, medial; L, lateral. 10 Cell Genomics 3, 100356, August 9, 2023 Article out (cid:3)500,000 individuals in the UK Biobank, which has an ascer- tainment bias for healthy individuals, and were absent from our control cohort. This high prevalence in our SCZ cohort of rela- tively large (cid:3)100–500-kb deletions, and the known involvement of germline NRXN1 mutations in SCZ and other neurodevelop- mental disorders, suggests that mosaic deletions of exons 1–5 might also contribute to SCZ risk. A study characterizing germline NRXN1 deletions from 19,263 clinical arrays in individuals with neurodevelopmental disease found that most of these events were present in the 50 end of NRXN1 and covered exons 1–5.32 In a case series of germline deletions in NRXN1 in individuals with more severe develop- mental disorders,57 two subjects with severe developmental delay had inherited deletions of exons 1–5. In contrast, germline deletions of NRXN1 in SCZ are widely distributed throughout the gene23 rather than being concentrated in the ﬁrst few exons as in neurodevelopmental disorders.32,33 This contrast might indicate that germline deletions of exons 1–5 result in more severe devel- opmental phenotypes but, if present in only a fraction of cells, could result in a milder phenotype resembling SCZ. While the most parsimonious model of pathogenicity of so- matic deletions in NRXN1 exons 1–5 is simple loss of function through deletion of the alpha promoter, the vast diversity of NRXN1 isoforms warrants further exploration of alternative mechanisms. Our analysis of Hi-C data using hiPSC neurons suggests a potential formation of a cryptic promoter once the NRXN1 alpha promoter is deleted, potentially forming an N-ter- minal truncated form of NRXN1, leading to a novel dominant- negative mechanism by trapping NRXN1a in the cytoplasm. This mechanism is consistent with higher intracellular protein levels of a NRXN1-binding protein CASK in hiPSC lines from SCZ patients with 50 NRXN1 deletions.38 However, further tran- scriptional and functional experiments could better validate the presence and role of this putative cryptic promoter in NRXN1 and SCZ biology. In this study, we also found ﬁve early-developmental recurrent somatic deletions in the ABCB11 transporter gene. These dele- tions were present only in the SCZ cases diagnosed with TRS, which is deﬁned as nonresponse to at least two anti-psychotic medications58 and affects (cid:3)30% of individuals with SCZ.59 Genes in this transporter family, including ABCB11, have previously been associated with differential response to anti-psychotics.41 How- ever, the exact mechanism by which mutations in these genes might lead to poor response to anti-psychotics remains unknown. We show that ABCB11 is strongly expressed in human DA neurons, speciﬁcally within the dorsal stream of the SN. Most anti-psychotic medications used to treat SCZ target DA signaling in the brain, but how DA pathways become abnormal in SCZ remains unclear. Disruption of ABCB11 could alter the function of this key neuronal circuitry in a relatively cell-type-spe- ciﬁc manner. While the exact role of ABCB11 on DA neuron physiology or excitatory layer 5 neurons is yet unknown, our re- sults suggest this as an area for further inquiry with potential dis- ease relevance. Limitations of the study This study has several limitations that represent further areas of research. While we were able to validate most of the variants ll OPEN ACCESS from DNA we were able to procure, further orthogonal validation is warranted to provide more accurate estimates of the burden of sCNV in SCZ and further characterize sources of potential arti- fact. The main limitation to validation was obtaining DNA from samples processed at institutions across the world with diverse data-sharing protocols. Our study was also limited to studying variants present in blood at a high cellular fraction, restricting var- iants characterized to those that might have arisen during early development, which are predicted to also be mosaic in the brain. While this experimental setup provided large sample sizes to test whether SCZ-associated sCNVs were present, future studies us- ing brain-derived tissue might allow further characterization of the potential risk of sCNVs in SCZ. We were also limited in detect- ing chimeric fusion genes since the SNP density of the array plat- forms used in this study ((cid:3)1/3 of SNPs being heterozygous) pre- vents enough resolution to call these events conﬁdently without a more dedicated method. In addition, the sparsity of clinical and environmental information in our dataset limited our ability to measure interactions of these factors with the burden of sCNV in SCZ, which suggests a potential area of future research. Finally, studying the functional role of the sCNVs in NRXN1 and ABCB11, and somatic variants in general, will require novel mosaic models such as organoids, or animal models, where spe- ciﬁc fractions of cells carry the desired events. The development of these models was outside the scope of the current study but presents an exciting future direction. The data presented here represent an initial and preliminary study that is potentially of in- terest to the ﬁeld as the role of somatic mutations in general, and sCNV speciﬁcally, in disease comes into focus. CONSORTIA Brain Somatic Mosaicism Network: Peter J. Park, Daniel Wein- berger, John V. Moran, Fred H. Gage, Flora M. Vaccarino, Joseph Gleeson, Gary Mathern, Eric Courchesne, Subhojit Roy, Sara Bizzotto, Michael Coulter, Caroline Dias, Alissa D’Gama, Javier Ganz, Robert Hill, August Yue Huang, Sattar Khoshkhoo, Sonia Kim, Michael Lodato,Michael Miller, Rebeca Borges-Monroy, Rachel Rodin, Zinan Zhou, Craig Bohrson, Chong Chu, Isidro Cortes-Ciriano, Yanmei Dou, Alon Galor, Doga Gulhan, Minseok Kwon, Joe Luquette, Vinay Viswanadham, Attila Jones,Chaggai Rosenbluh, Sean Cho, Ben Langmead, Jeremy Thorpe, Jennifer Erwin, Andrew Jaffe, Michael McConnell,Rujuta Narurkar, Apua Paquola, Jooheon Shin, Richard Straub, Alexej Abyzov, Tae- jeong Bae, Yeongjun Jang, Yifan Wang, Fred Gage, Sara Linker, Patrick Reed, Meiyan Wang, Alexander Urban, Bo Zhou, Xiaowei Zhu, Reenal Pattni, Aitor Serres Amero, David Juan, Irene Lobon, Tomas Marques-Bonet, Manuel Solis Moruno, Raquel Garcia Perez, Inna Povolotskaya, Eduardo Soriano, Danny Antaki, Dan Averbuj, Laurel Ball, Martin Breuss, Xiaoxu Yang, Changuk Chung, Sarah B. Emery, Diane A. Flasch, Jeffrey M. Kidd, Huira C. Kopera, Kenneth Y. Kwan, Ryan E. Mills, John B. Moldovan, Chen Sun, Xuefang Zhao, Weichen Zhou, Trenton J. Frisbie, Adri- ana Cherskov, Liana Fasching, Alexandre Jourdon, Sirisha Po- chareddy, Soraya Scuderi, Nenad Sestan. Psychiatric Genomic Consortium: Christian R. Marshall, Dan- iele Merico, Bhooma Thiruvahindrapuram, Zhouzhi Wang, Ste- phen W. Scherer, Daniel P Howrigan, Stephan Ripke, Brendan Cell Genomics 3, 100356, August 9, 2023 11 ll OPEN ACCESS Bulik-Sullivan, Kai-How Farh, Menachem Fromer, Jacqueline I. Goldstein, Hailiang Huang, Phil Lee, Mark J. Daly, Benjamin M. Neale, Richard A. Belliveau Jr, Sarah E. Bergen, Elizabeth Bevi- lacqua, Kimberley D. Chambert, Colm O’Dushlaine, Edward M. Scolnick, Jordan W. Smoller, Jennifer L. Moran, Aarno Palotie, Tracey L. Petryshen, Wenting Wu, Douglas S. Greer, Danny An- taki, Aniket Shetty, Madhusudan Gujral, William M. Brandler, Dheeraj Malhotra, Karin V. Fuentes Fajarado, Michelle S. Maile, Peter A. Holmans, Noa Carrera, Nick Craddock, Valentina Escott-Price, Lyudmila Georgieva, Marian L. Hamshere, David Kavanagh, Sophie E. Legge, Andrew J. Pocklington, Alexander L. Richards, Douglas M. Ruderfer, Nigel M. Williams, George Kirov, Michael J. Owen, Dalila Pinto, Guiqing Cai, Kenneth L. Da- vis, Elodie Drapeau, Joseph I Friedman, Vahram Haroutunian, Elena Parkhomenko, Abraham Reichenberg, Jeremy M. Silver- man, Joseph D. Buxbaum, Enrico Domenici, Ingrid Agartz, Srdjan Djurovic, Morten Mattingsdal, Ingrid Melle, Ole A. Andreassen, Erik G. Jo¨ nsson, Erik So¨ derman, Margot Albus, Madeline Alex- ander, Claudine Laurent, Douglas F. Levinson, Farooq Amin, Joshua Atkins, Murray J. Cairns, Rodney J. Scott, Paul A. Too- ney, Jing Qin Wu, Silviu A. Bacanu, Tim B. Bigdeli, Mark A. Reim- ers, Bradley T. Webb, Aaron R. Wolen, Brandon K. Wormley, Kenneth S. Kendler, Brien P. Riley, Anna K. Ka¨ hler, Patrik K. E. Magnusson, Christina M. Hultman, Marcelo Bertalan, Thomas Hansen, Line Olsen, Henrik B. Rasmussen, Thomas Werge, Man- uel Mattheisen, Donald W. Black, Richard Bruggeman, Nancy G. Buccola, Randy L. Buckner, Joshua L. Roffman, William Byerley, Wiepke Cahn, Rene´ S Kahn, Eric Strengman, Roel A. Ophoff, Vaughan J. Carr, Stanley V. Catts, Frans A. Henskens, Carmel M. Loughland, Patricia T. Michie, Christos Pantelis, Ulrich Schall, Assen V. Jablensky, Brian J. Kelly, Dominique Campion, Rita M. Cantor, Wei Cheng, C. Robert Cloninger, Dragan M Svrakic, Da- vid Cohen, Paul Cormican, Gary Donohoe, Derek W. Morris, Ai- den Corvin, Michael Gill, Benedicto Crespo-Facorro, James J. Crowley, Martilias S. Farrell, Paola Giusti-Rodrı´guez, Yunjung Kim, Jin P. Szatkiewicz, Stephanie Williams, David Curtis, Jona- than Pimm, Hugh Gurling, Andrew McQuillin, Michael Davidson, Mark Weiser, Franziska Degenhardt, Andreas J. Forstner, Stefan Herms, Per Hoffmann, Andrea Hofman, Sven Cichon, Markus M. No¨ then, Jurgen Del Favero, Lynn E. DeLisi, Robert W. McCarley, Deborah L. Levy, Raquelle I. Mesholam-Gately, Larry J. Seidman, Dimitris Dikeos, George N. Papadimitriou, Timothy Dinan, Jubao Duan, Alan R. Sanders, Pablo V. Gejman, Elliot S. Gershon, Frank Dudbridge, Peter Eichhammer, Johan Eriksson, Veikko Salomaa, Laurent Essioux, Ayman H. Fanous, James A. Knowles, Michele T. Pato, Carlos N. Pato, Josef Frank, Sandra Meier, Thomas G. Schulze, Jana Strohmaier, Stephanie H. Witt, Marcella Rietschel, Lude Franke, Juha Karjalainen, Robert Freedman, Ann Olincy, Nelson B. Freimer, Shaun M. Purcell, Panos Roussos, Eli A. Stahl, Pamela Sklar, Jordan W. Smoller, Ina Giegling, Annette M. Hart- mann, Bettina Konte, Dan Rujescu, Stephanie Godard, Joel N. Hirschhorn, Tune H. Pers, Alkes Price, To˜ nu Esko, Jacob Gratten, S. Hong Lee, Peter M. Visscher, Naomi R. Wray, Bryan J. Mowry, Lieuwe de Haan, Carin J. Meijer, Mark Hansen, Masashi Ikeda, Nakao Iwata, Inge Joa, Luba Kalaydjieva, Matthew C. Keller, James L. Kennedy, Clement C. Zai, Jo Knight, Bernard Lerer, Kung-Yee Liang, Jeffrey Lieberman, T. Scott Stroup, Jouko Lo¨ nnqvist, Jaana Suvisaari, Brion S. Maher, Wolfgang Maier, 12 Cell Genomics 3, 100356, August 9, 2023 Article Jacques Mallet, Colm McDonald, Andrew M. McIntosh, Douglas H. R. Blackwood, Andres Metspalu, Lili Milani, Vihra Milanova, Younes Mokrab, David A. Collier, Bertram M€uller-Myhsok, Kieran C. Murphy, Robin M. Murray, John Powell, Inez Myin-Germeys, Jim Van Os, Igor Nenadic, Deborah A. Nertney, Gerald Nestadt, Ann E. Pulver, Kristin K. Nicodemus, Laura Nisenbaum, Annelie Nordin, Rolf Adolfsson, Eadbhard O’Callaghan, Sang-Yun Oh, F. Anthony O’Neill, Tiina Paunio, Olli Pietila¨ inen, Diana O. Perkins, Digby Quested, Adam Savitz, Qingqin S. Li, Sibylle G. Schwab, Jianxin Shi, Chris C. A. Spencer, Srinivas Thirumalai, Juha Veijola, John Waddington, Dermot Walsh, Dieter B. Wildenauer, Elvira Bramon, Ariel Darvasi, Danielle Posthuma, David St. Clair, Omar Shanta, Marieke Klein. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d KEY RESOURCES TABLE d RESOURCE AVAILABILITY B Lead contact B Materials availability B Data and code availability d EXPERIMENTAL MODEL AND SUBJECT DETAILS B SNP array data acquisition B Data processing d METHOD DETAILS B Variant level quality control B Sample-level quality control B Event type classiﬁcation B Filtration of mosaic CNV calls B Statistical analysis B Cell fraction, gene-set, length, and gene number burden analysis B Gene set enrichment analysis B Permutation test for enrichment of sCNV overlapping exons 1–5 of NRXN1 B Germline CNV analyses B Breakpoint microhomology analysis B In situ Hi-C from hiPSC-derived neurons B Hi-C read alignment B Preprocessing for variant calling B Variant calling for Hi-C analysis B Haplotype phasing for Hi-C analysis B Hi-C matrix construction and visualization B Analysis of human postmortem snRNA-seq datasets B Single-molecule in situ hybridization (smFISH) and im- aging of postmortem human nigra SUPPLEMENTAL INFORMATION Supplemental xgen.2023.100356. information can be found online at https://doi.org/10.1016/j. ACKNOWLEDGMENTS E.A.M. is supported by the Harvard/MIT MD-PhD program (T32GM007753), Training Program the Biomedical and Data Science Informatics Article (T15LM007092), and the Ruth L. Kirschstein NRSA F31 Fellowship (F31MH124292). G.G. is supported by NIH grant R01HG006855, NIH grant R01MH104964, and the Stanley Center for Psychiatric Research. S.A. was supported by NIH grant R01MH106056 and S.A., A.C., and C.A.W. were sup- ported by the NIMH grant (U01MH106883) through the Brain Somatic Mosai- cism Network (BSMN). C.A.W. is an investigator of the Howard Hughes Med- ical Institute. C.A.W. and E.A.L. are supported by the Allen Frontiers Program through the Allen Discovery Center for Human Brain Evolution. E.A.L. is sup- ported by NIH grants (K01 AG051791, DP2 AG072437, and R01AG070921) and the SUHF foundation. J.S. is supported by NIH grants (MH113715, MH119746, MH109501, and MH119746). J.E.P.-C. and K.J.B. are supported by a Chan Zuckerberg Initiative grant (2020-221479). J.E.P.-C. is supported by NIH grants (DP1OD031253, R01NS-114226, R01MH12026, and U01DK127405). P.-R.L. is supported by NIH grant DP2 ES030554 and a Bur- roughs Wellcome Fund Career Award at the Scientiﬁc Interfaces. T.K. is sup- ported by F30AG069446-01. E.Z.M. is supported by DP2AG058488, U01MH124602, and Chan Zuckerberg Initiative (no. 2017-175259). AUTHOR CONTRIBUTIONS E.A.M. and C.A.W. conceived and designed the study. E.A.M. designed and implemented the statistical methods. E.A.M. performed computational ana- lyses, with assistance from M.A.S. and G.G. J.S. curated the data and facili- tated access. S.M. and A.C. facilitated acquisition of samples for whole- genome sequencing validation. J.T.R.W., M.O., and P.S. facilitated clinical and genomic data procurement for validation and interpretation. P.R., S.A., and K.J.B. generated the Hi-C data. T.G.G. and J.E.P.-C. analyzed and inter- preted the Hi-C data. E.F. and K.J.B. generated and characterized the hiPSCs/ neurons. T.K., S.B., and E.Z.M. contributed and analyzed the ABCB11 snRNA- seq data. E.A.L., P.-R.L., S.A.M., and J.S. provided comments and guidance throughout. E.A.M., E.A.L., and C.A.W. wrote the manuscript. DECLARATION OF INTERESTS The authors declare no competing interests. INCLUSION AND DIVERSITY We support inclusive, diverse, and equitable conduct of research. Received: November 30, 2021 Revised: February 21, 2022 Accepted: June 9, 2023 Published: July 6, 2023 REFERENCES 1. Kirov, G., Pocklington, A.J., Holmans, P., Ivanov, D., Ikeda, M., Ruderfer, D., Moran, J., Chambert, K., Toncheva, D., Georgieva, L., et al. (2012). De novo CNV analysis implicates speciﬁc abnormalities of postsynaptic sig- nalling complexes in the pathogenesis of schizophrenia. Mol. Psychiatry 17, 142–153. https://doi.org/10.1038/mp.2011.154. 2. Kirov, G., Rujescu, D., Ingason, A., Collier, D.A., O’Donovan, M.C., and Owen, M.J. (2009). Neurexin 1 (NRXN1) Deletions in Schizophrenia. Schiz- ophr. Bull. 35, 851–854. https://doi.org/10.1093/schbul/sbp079. 3. Sherman, M.A., Rodin, R.E., Genovese, G., Dias, C., Barton, A.R., Muka- mel, R.E., Berger, B., Park, P.J., Walsh, C.A., and Loh, P.R. (2021). Large mosaic copy number variations confer autism risk. Nat. Neurosci. 24, 197–203. https://doi.org/10.1038/s41593-020-00766-5. 4. Maury, E.A., and Walsh, C.A. (2021). Somatic copy number variants in neuropsychiatric disorders. Curr. Opin. Genet. Dev. 68, 9–17. https:// doi.org/10.1016/j.gde.2020.12.013. 5. Ruderfer, D.M., Chambert, K., Moran, J., Talkowski, M., Chen, E.S., Gigek, C., Gusella, J.F., Blackwood, D.H., Corvin, A., Gurling, H.M., et al. (2013). Mosaic copy number variation in schizophrenia. Eur. J. Hum. Genet. 21, 1007–1011. https://doi.org/10.1038/ejhg.2012.287. ll OPEN ACCESS 6. King, D.A., Jones, W.D., Crow, Y.J., Dominiczak, A.F., Foster, N.A., Gaunt, T.R., Harris, J., Hellens, S.W., Homfray, T., Innes, J., et al. (2015). Mosaic structural variation in children with developmental disorders. Hum. Mol. Genet. 24, 2733–2745. https://doi.org/10.1093/hmg/ddv033. 7. Bae, T., Fasching, L., Wang, Y., Shin, J.H., Suvakov, M., Jang, Y., Norton, S., Dias, C., Mariani, J., Jourdon, A., et al. (2022). Analysis of somatic mu- tations in 131 human brains reveals aging-associated hypermutability. Science 377, 511–517. https://doi.org/10.1126/SCIENCE.ABM6222/ SUPPL_FILE/SCIENCE.ABM6222_MDAR_REPRODUCIBILITY_CHECK- LIST.PDF. 8. Bae, T., Wang, Y., Vaccarino, F.M., and Abyzov, A. (2022). Somatic genomic mosaicism in the brain during aging: Scratching the surface. Clin. Transl. Med. 12, e1138. https://doi.org/10.1002/CTM2.1138. 9. Poduri, A., Evrony, G.D., Cai, X., Elhosary, P.C., Beroukhim, R., Lehtinen, M.K., Hills, L.B., Heinzen, E.L., Hill, A., Hill, R.S., et al. (2012). Somatic Acti- vation of AKT3 Causes Hemispheric Developmental Brain Malformations. Neuron 74, 41–48. https://doi.org/10.1016/j.neuron.2012.03.010. 10. Conti, V., Pantaleo, M., Barba, C., Baroni, G., Mei, D., Buccoliero, A.M., Gi- glio, S., Giordano, F., Baek, S.T., Gleeson, J.G., and Guerrini, R. (2015). Focal dysplasia of the cerebral cortex and infantile spasms associated with somatic 1q21.1-q44 duplication including the AKT3 gene. Clin. Genet. 88, 241–247. https://doi.org/10.1111/cge.12476. 11. Lo´ pez-Rivera, J.A., Leu, C., Macnee, M., Khoury, J., Hoffmann, L., Coras, R., Kobow, K., Bhattarai, N., Pe´ rez-Palma, E., Hamer, H., et al. (2022). The genomic landscape across 474 surgically accessible epileptogenic human brain lesions. Brain. https://doi.org/10.1093/BRAIN/AWAC376. 12. Kushima, I., Aleksic, B., Nakatochi, M., Mori, D., Iwata, N., and Ozaki, N. (2018). Comparative Analyses of Copy-Number Variation in Autism Spec- trum Disorder and Schizophrenia Reveal Etiological Overlap and Biolog- ical Insights. Cell Rep. 24, 2838–2856, Etiological overlap Orange: Intellec- tual disability + Patients without pathogenic CNVs. https://doi.org/10. 1016/j.celrep.2018.08.022. 13. Lodato, M.A., Woodworth, M.B., Lee, S., Evrony, G.D., Mehta, B.K., Karger, A., Lee, S., Chittenden, T.W., D’Gama, A.M., Cai, X., et al. (2015). Somatic mutation in single human neurons tracks developmental and transcriptional history. Science 350, 94–98. https://doi.org/10.1126/ science.aab1785. 14. Bizzotto, S., Dou, Y., Ganz, J., Doan, R.N., Kwon, M., Bohrson, C.L., Kim, S.N., Bae, T., Abyzov, A., et al.; NIMH Brain Somatic Mosaicism Network (2021). Landmarks of human embryonic development inscribed in somatic mutations. Science 371, 1249–1253. https://doi.org/10.1126/SCIENCE. ABE1544/SUPPL_FILE/ABE1544-BIZZOTTO-SM.PDF. 15. Bae, T., Tomasini, L., Mariani, J., Zhou, B., Roychowdhury, T., Franjic, D., Pletikos, M., Pattni, R., Chen, B.-J., Venturini, E., et al. (2018). Different mutational rates and mechanisms in human cells at pregastrulation and neurogenesis. Science 359, 550–555. https://doi.org/10.1126/SCIENCE. AAN8690. 16. Fasching, L., Jang, Y., Tomasi, S., Schreiner, J., Tomasini, L., Brady, M.V., Bae, T., Sarangi, V., Vasmatzis, N., Wang, Y., et al. (2021). Early develop- mental asymmetries in cell lineage trees in living individuals. Science 371, 1245–1248. https://doi.org/10.1126/SCIENCE.ABE0981/SUPPL_FILE/ ABE0981_TABLE_S3.XLSX. 17. Ju, Y.S., Martincorena, I., Gerstung, M., Petljak, M., Alexandrov, L.B., Rahbari, R., Wedge, D.C., Davies, H.R., Ramakrishna, M., Fullam, A., et al. (2017). Somatic mutations reveal asymmetric cellular dynamics in the early human embryo. Nature 543, 714–718. https://doi.org/10.1038/ nature21703. 18. Loh, P.-R., Genovese, G., Handsaker, R.E., Finucane, H.K., Reshef, Y.A., Palamara, P.F., Birmann, B.M., Talkowski, M.E., Bakhoum, S.F., McCar- roll, S.A., and Price, A.L. (2018). Insights into clonal haematopoiesis from 8,342 mosaic chromosomal alterations. Nature 559, 350–355. https://doi.org/10.1038/s41586-018-0321-x. Cell Genomics 3, 100356, August 9, 2023 13 ll OPEN ACCESS Article 19. Loh, P.R., Genovese, G., and McCarroll, S.A. (2020). Monogenic and poly- genic inheritance become instruments for clonal selection. Nature 584, 136–141. https://doi.org/10.1038/s41586-020-2430-6. (2020). The clinical relevance of et al. intragenic NRXN1 deletions. J. Med. Genet. 57, 347–355. https://doi.org/10.1136/jmedgenet-2019- 106448. 20. Zekavat, S.M., Lin, S.H., Bick, A.G., Liu, A., Paruchuri, K., Wang, C., Uddin, M.M., Ye, Y., Yu, Z., Liu, X., et al. (2021). Hematopoietic mosaic chromo- somal alterations increase the risk for diverse types of infection. Nat. Med. 27, 1012–1024. https://doi.org/10.1038/S41591-021-01371-0. 34. Roadmap Epigenomics Consortium; Kundaje, A., Meuleman, W., Ernst, J., Bilenky, M., Yen, A., Heravi-Moussavi, A., Kheradpour, P., Zhang, Z., Wang, J., et al. (2015). Integrative analysis of 111 reference human epige- nomes. Nature 518, 317–330. https://doi.org/10.1038/nature14248. 21. Terao, C., Suzuki, A., Momozawa, Y., Akiyama, M., Ishigaki, K., Yama- moto, K., Matsuda, K., Murakami, Y., McCarroll, S.A., Kubo, M., et al. (2020). Chromosomal alterations among age-related haematopoietic clones in Japan. Nature 584, 130–135. https://doi.org/10.1038/s41586- 020-2426-2. 22. Saiki, R., Momozawa, Y., Nannya, Y., Nakagawa, M.M., Ochi, Y., Yoshi- zato, T., Terao, C., Kuroda, Y., Shiraishi, Y., Chiba, K., et al. (2021). Com- bined landscape of single-nucleotide variants and copy number alter- ations in clonal hematopoiesis. Nat. Med. 27, 1239–1249. https://doi. org/10.1038/S41591-021-01411-9. 23. Marshall, C.R., Howrigan, D.P., Merico, D., Thiruvahindrapuram, B., Wu, W., Greer, D.S., Antaki, D., Shetty, A., Holmans, P.A., Pinto, D., et al. (2017). Contribution of copy number variants to schizophrenia from a genome-wide study of 41,321 subjects. Nat. Genet. 49, 27–35. https:// doi.org/10.1038/ng.3725. 24. Qin, N., Wang, C., Chen, C., Yang, L., Liu, S., Xiang, J., Xie, Y., Liang, S., Zhou, J., Xu, X., et al. (2022). Association of the interaction between mosaic chromosomal alterations and polygenic risk score with the risk of lung cancer: an array-based case-control association and prospective cohort study. Lancet Oncol. 23, 1465–1474. https://doi.org/10.1016/ S1470-2045(22)00600-3. 25. Brown, D.W., Lin, S.H., Loh, P.R., Chanock, S.J., Savage, S.A., and Ma- chiela, M.J. (2020). Genetically predicted telomere length is associated with clonal somatic copy number alterations in peripheral leukocytes. PLoS Genet. 16, e1009078. https://doi.org/10.1371/JOURNAL.PGEN. 1009078. 26. Lek, M., Karczewski, K.J., Minikel, E.V., Samocha, K.E., Banks, E., Fen- nell, T., O’Donnell-Luria, A.H., Ware, J.S., Hill, A.J., Cummings, B.B., et al. (2016). Analysis of protein-coding genetic variation in 60,706 hu- mans. Nature 536, 285–291. https://doi.org/10.1038/nature19057. 27. Singh, T., Poterba, T., Curtis, D., Akil, H., Al Eissa, M., Barchas, J.D., Bass, N., Bigdeli, T.B., Breen, G., Bromet, E.J., et al. (2022). Rare coding variants in ten genes confer substantial risk for schizophrenia. Nat 604, 509–516. https://doi.org/10.1038/s41586-022-04556-w. 28. Weiss, L.A., Shen, Y., Korn, J.M., Arking, D.E., Miller, D.T., Fossdal, R., Saemundsen, E., Stefansson, H., Ferreira, M.A.R., Green, T., et al. (2008). Association between Microdeletion and Microduplication at 16p11.2 and Autism. N. Engl. J. Med. 358, 667–675. 29. Arinami, T. (2006). Analyses of the associations between the genes of 22q11 deletion syndrome and schizophrenia. J. Hum. Genet. 51, 1037– 1045. https://doi.org/10.1007/s10038-006-0058-5. 30. Gothelf, D., Law, A.J., Frisch, A., Chen, J., Zarchi, O., Michaelovsky, E., Ren-Patterson, R., Lipska, B.K., Carmel, M., Kolachana, B., et al. (2014). Biological effects of COMT haplotypes and psychosis risk in 22q11.2 dele- tion syndrome. Biol. Psychiatry 75, 406–413. https://doi.org/10.1016/j. biopsych.2013.07.021. 31. Flaherty, E., Zhu, S., Barretto, N., Cheng, E., Deans, P.J.M., Fernando, M.B., Schrode, N., Francoeur, N., Antoine, A., Alganem, K., et al. (2019). impact of patient-speciﬁc aberrant NRXN1a splicing. Nat. Neuronal Genet. 51, 1679–1690. https://doi.org/10.1038/s41588-019-0539-z. 35. Yang, L., Luquette, L.J., Gehlenborg, N., Xi, R., Haseley, P.S., Hsieh, C.H., Zhang, C., Ren, X., Protopopov, A., Chin, L., et al. (2013). Diverse mecha- nisms of somatic structural variations in human cancer genomes. Cell 153, 919–929. https://doi.org/10.1016/j.cell.2013.04.010. 36. Kidd, J.M., Graves, T., Newman, T.L., Fulton, R., Hayden, H.S., Malig, M., Kallicki, J., Kaul, R., Wilson, R.K., and Eichler, E.E. (2010). A Human Genome Structural Variation Sequencing Resource Reveals Insights into Mutational Mechanisms. Cell 143, 837–847. https://doi.org/10.1016/j. cell.2010.10.027. 37. Zhang, F., Khajavi, M., Connolly, A.M., Towne, C.F., Batish, S.D., and Lup- ski, J.R. (2009). The DNA replication FoSTeS/MMBIR mechanism can generate genomic, genic and exonic complex rearrangements in humans. Nat. Genet. 41, 849–853. https://doi.org/10.1038/ng.399. 38. Pak, C., Danko, T., Mirabella, V.R., Wang, J., Liu, Y., Vangipuram, M., Grieder, S., Zhang, X., Ward, T., Huang, Y.-W.A., et al. (2021). Cross-plat- form validation of neurotransmitter release impairments in schizophrenia patient-derived NRXN1-mutant neurons. Proc. Natl. Acad. Sci. USA 118, 2025598118. https://doi.org/10.1073/PNAS.2025598118. 39. Bompadre, O., and Andrey, G. (2019). Chromatin topology in development and disease. Curr. Opin. Genet. Dev. 55, 32–38. https://doi.org/10.1016/J. GDE.2019.04.007. 40. Halvorsen, M., Huh, R., Oskolkov, N., Wen, J., Netotea, S., Giusti-Rodri- guez, P., Karlsson, R., Bryois, J., Nystedt, B., Ameur, A., et al. (2020). Increased burden of ultra-rare structural variants localizing to boundaries of topologically associated domains in schizophrenia. Nat. Commun. 11, 1842. https://doi.org/10.1038/s41467-020-15707-w. 41. Vita, A., Minelli, A., Barlati, S., Deste, G., Giacopuzzi, E., Valsecchi, P., Tur- rina, C., and Gennarelli, M. (2019). Treatment-resistant schizophrenia: Ge- netic and neuroimaging correlates. Front. Pharmacol. 10, 402. https://doi. org/10.3389/fphar.2019.00402. 42. Gonzalez-Covarrubias, V., Martı´nez-Magan˜ a, J.J., Coronado-Sosa, R., Villegas-Torres, B., Genis-Mendoza, A.D., Canales-Herrerias, P., Nicolini, H., and Sobero´ n, X. (2016). Exploring Variation in Known Pharmacoge- netic Variants and its Association with Drug Response in Different Mexican Populations. Pharm. Res. (N. Y.) 33, 2644–2652. https://doi.org/10.1007/ s11095-016-1990-5. 43. Strautnieks, S.S., Bull, L.N., Knisely, A.S., Kocoshis, S.A., Dahl, N., Arnell, H., Sokal, E., Dahan, K., Childs, S., Ling, V., et al. (1998). A gene encoding a liver-speciﬁc ABC transporter is mutated in progressive familial intrahe- patic cholestasis. Nat. Genet. 20, 233–238. https://doi.org/10.1038/3034. 44. Knisely, A.S., Strautnieks, S.S., Meier, Y., Stieger, B., Byrne, J.A., Port- mann, B.C., Bull, L.N., Pawlikowska, L., Bilezikc¸ i, B., Ozc¸ ay, F., et al. (2006). Hepatocellular carcinoma in ten children under ﬁve years of age with bile salt export pump deﬁciency. Hepatology 44, 478–486. https:// doi.org/10.1002/HEP.21287. 45. Van Mil, S.W.C., Van Der Woerd, W.L., Van Der Brugge, G., Sturm, E., Jan- sen, P.L.M., Bull, L.N., Van Den Berg, I.E.T., Berger, R., Houwen, R.H.J., and Klomp, L.W.J. (2004). Benign recurrent intrahepatic cholestasis type 2 is caused by mutations in ABCB11. Gastroenterology 127, 379–384. https://doi.org/10.1053/J.GASTRO.2004.04.065. 32. Lowther, C., Speevak, M., Armour, C.M., Goh, E.S., Graham, G.E., Li, C., Zeesman, S., Nowaczyk, M.J.M., Schultz, L.A., Morra, A., et al. (2017). Molecular characterization of NRXN1 deletions from 19,263 clinical micro- array cases identiﬁes exons important for neurodevelopmental disease expression. Genet. Med. 19, 53–61. https://doi.org/10.1038/gim.2016.54. 46. Jansen, P.L., Strautnieks, S.S., Jacquemin, E., Hadchouel, M., Sokal, E.M., Hooiveld, G.J., Koning, J.H., De Jager-Krikken, A., Kuipers, F., Stel- laard, F., et al. (1999). Hepatocanalicular bile salt export pump deﬁciency in patients with progressive familial intrahepatic cholestasis. Gastroenter- ology 117, 1370–1379. https://doi.org/10.1016/S0016-5085(99)70287-8. 33. Cosemans, N., Vandenhove, L., Vogels, A., Devriendt, K., Van Esch, H., Van Buggenhout, G., Olivie´ , H., De Ravel, T., Ortibus, E., Legius, E., 47. Ortiz, D.F., Moseley, J., Calderon, G., Swift, A.L., Li, S., and Arias, I.M. (2004). Identiﬁcation of HAX-1 as a protein that binds bile salt export 14 Cell Genomics 3, 100356, August 9, 2023 Article ll OPEN ACCESS protein and regulates its abundance in the apical membrane of Madin- Darby canine kidney cells. J. Biol. Chem. 279, 32761–32770. https://doi. org/10.1074/jbc.M404337200. 48. Alogaili, F., Chinnarasu, S., Jaeschke, A., Kranias, E.G., Hui, D.Y., and Pessin, J.E. (2020). Hepatic HAX-1 inactivation prevents metabolic dis- eases by enhancing mitochondrial activity and bile salt export. J. Biol. Chem. 295, 4631–4646. https://doi.org/10.1074/jbc.RA119.012361. 49. Schizophrenia Working Group of the Psychiatric Genomics Consortium; Ripke, S., Neale, B.M., Corvin, A., Walters, J.T.R., Farh, K.-H., Holmans, P.A., Lee, P., Bulik-Sullivan, B., Collier, D.A., et al. (2014). Biological in- sights from 108 schizophrenia-associated genetic loci. Nature 511, 421–427. https://doi.org/10.1038/nature13595. 50. Hamshere, M.L., Walters, J.T.R., Smith, R., Richards, A.L., Green, E., Gro- zeva, D., Jones, I., Forty, L., Jones, L., Gordon-Smith, K., et al. (2013). Genome-wide signiﬁcant associations in schizophrenia to ITIH3/4, CACNA1C and SDCCAG8, and extensive replication of associations re- ported by the Schizophrenia PGC. Mol. Psychiatry 18, 708–712. https:// doi.org/10.1038/mp.2012.67. 51. Song, X., Beck, C.R., Du, R., Campbell, I.M., Coban-Akdemir, Z., Gu, S., Breman, A.M., Stankiewicz, P., Ira, G., Shaw, C.A., and Lupski, J.R. (2018). Predicting human genes susceptible to genomic instability associ- ated with Alu/Alu-mediated rearrangements. Genome Res. 28, 1228– 1242. https://doi.org/10.1101/GR.229401.117. 52. Kamath, T., Abdulraouf, A., Burris, S.J., Langlieb, J., Gazestani, V., Nadaf, N.M., Balderrama, K., Vanderburg, C., and Macosko, E.Z. (2022). Single- cell genomic proﬁling of human dopamine neurons identiﬁes a population that selectively degenerates in Parkinson’s disease. Nat. Neurosci. 25, 588–595. https://doi.org/10.1038/s41593-022-01061-1. 53. Bakken, T.E., Jorstad, N.L., Hu, Q., Lake, B.B., Tian, W., Kalmbach, B.E., Crow, M., Hodge, R.D., Krienen, F.M., Sorensen, S.A., et al. (2021). Comparative cellular analysis of motor cortex in human, marmoset and mouse. Nat 598, 111–119. https://doi.org/10.1038/s41586-021-03465-8. lies. J. Neurochem. 113, 287–302. https://doi.org/10.1111/J.1471-4159. 2010.06604.X. 57. Ching, M.S.L., Shen, Y., Tan, W.H., Jeste, S.S., Morrow, E.M., Chen, X., Mukaddes, N.M., Yoo, S.Y., Hanson, E., Hundley, R., et al. (2010). Dele- tions of NRXN1 (neurexin-1) predispose to a wide spectrum of develop- mental disorders. Am. J. Med. Genet. B Neuropsychiatr. Genet. 153B, 937–947. https://doi.org/10.1002/ajmg.b.31063. 58. National Institute of Health and Clinical Excellence (2014). Psychosis and Schizophrenia in Adults (NICE Guidel. treament Manag.), pp. 74–80. 59. Meltzer, H.Y. (1997). Treatment-resistant schizophrenia - The role of clozapine. Curr. Med. Res. Opin. 14, 1–20. https://doi.org/10.1185/ 03007999709113338. 60. Gel, B., Dı´ez-Villanueva, A., Serra, E., Buschbeck, M., Peinado, M.A., and Malinverni, R. (2016). regioneR: an R/Bioconductor package for the asso- ciation analysis of genomic regions based on permutation tests. Bioinfor- matics 32, 289–291. https://doi.org/10.1093/BIOINFORMATICS/BTV562. 61. Thorvaldsdo´ ttir, H., Robinson, J.T., and Mesirov, J.P. (2013). Integrative Genomics Viewer (IGV): High-performance genomics data visualization and exploration. Brief. Bioinform. 14, 178–192. https://doi.org/10.1093/ bib/bbs017. 62. Bates, D.W., Ma¨ chler, M., Bolker, B.M., and Walker, S.C. (2015). Fitting linear mixed-effects models using lme4. BMJ Qual. Saf. 24, 1–3. https:// doi.org/10.18637/jss.v067.i01. 63. Kuznetsova, A., Brockhoff, P.B., and Christensen, R.H.B. (2017). lmerTest Package: Tests in Linear Mixed Effects Models. J. Stat. Softw. 82, 1–26. https://doi.org/10.18637/jss.v082.i13. 64. Loh, P.R., Danecek, P., Palamara, P.F., Fuchsberger, C., A Reshef, Y., K Finucane, H., Schoenherr, S., Forer, L., McCarthy, S., Abecasis, G.R., et al. (2016). Reference-based phasing using the Haplotype Reference Consortium panel. Nat. Genet. 48, 1443–1448. https://doi.org/10.1038/ ng.3679. 54. Haber, S.N. (2014). The place of dopamine in the cortico-basal ganglia cir- cuit. Neuroscience 282, 248–257. https://doi.org/10.1016/J.NEUROSCI- ENCE.2014.10.008. 65. Vattathil, S., and Scheet, P. (2013). Haplotype-based proﬁling of subtle allelic imbalance with SNP arrays. Genome Res. 23, 152–158. https:// doi.org/10.1101/gr.141374.112. 55. Zhang, Y., Larcher, K.M.H., Misic, B., and Dagher, A. (2017). Anatomical and functional organization of the human substantia Nigra and its connec- tions. Elife 6, e26653. https://doi.org/10.7554/ELIFE.26653. 56. Perez-Costas, E., Melendez-Ferro, M., and Roberts, R.C. (2010). Basal ganglia pathology in schizophrenia: dopamine connections and anoma- 66. Raychaudhuri, S., Korn, J.M., McCarroll, S.A., International Schizophrenia Consortium; Altshuler, D., Sklar, P., Purcell, S., and Daly, M.J. (2010). Accurately Assessing the Risk of Schizophrenia Conferred by Rare Copy-Number Variation Affecting Genes with Brain Function. PLoS Genet. 6, e1001097. https://doi.org/10.1371/journal.pgen.1001097. Cell Genomics 3, 100356, August 9, 2023 15 ll OPEN ACCESS STAR+METHODS KEY RESOURCES TABLE REAGENT or RESOURCE Critical commercial assays Hi-C Kit SuperFrost Plus slides Probe hybridization buffer Probe ampliﬁcation buffer 5xSSCT (20% Tween) Hairpins Probes Biological sample data healthy adult postmortem midbrain block Deposited data Individual level SNP-array data Filtered sCNV callset Top 20% brain expressed genes Synaptic genes SOURCE Arima Molecular Instruments Molecular Instruments ThermoFisher Scientiﬁc Molecular Instruments Molecular Instrument Sepulveda Human Brain and Spinal Fluid Resource Center Psychiatric Genomic Consortium This paper GTEx SynaptomeDB gnomAD constrain statistics gnomAD HapMap variants v3.3 1000 Genomes ‘‘Omni’’ platform variants v2.5 Whole Genome Sequencing data Broad Institute Article IDENTIFIER N/A N/A N/A N/A Catalog # 15557044 N/A Costume made based on accession number see STAR Methods http://brainbank.ucla.edu/ https://www.med.unc.edu/pgc/ shared-methods/how-to/ data listed in Filtered sCNV callset is in Table S2 https://www.gtexportal.org/home/datasets http://metamoodics.org/SynaptomeDB/ index.php https://gnomad.broadinstitute.org/ downloads https://www.sanger.ac.uk/resources/ downloads/human/hapmap3.html https://www.internationalgenome.org/ category/omni/ Sequencing data will be uploaded to the NIMH Data Archive after publication. Experimental models: Cell lines hiPSC cell lines: Control NSB2607-2 (2607 clone 1) 50 deletion NSB973-5 (973 clone 1) Software and algorithms MoChA R v 4.0.3 regioneR (R package) IGV Lme4 (R package) lmerTest (R package) Python v3.6.12 BWA mem v0.7.17-r1188 GATK e1 Cell Genomics 3, 100356, August 9, 2023 Flaherty et al.31 N/A Loh et al.,18 Loh et al.19 https://github.com/freeseek/mocha R Core Team Gel et al.60 Thorvaldsdottir et al.61 Bates et al.62 Kuznetsova et al.63 Python Core Team https://www.r-project.org https://bioconductor.org/packages/ release/bioc/html/regioneR.html https://software.broadinstitute.org/ software/igv/download https://cran.r-project.org/web/packages/ lme4/index.html https://cran.r-project.org/web/packages/ lmerTest/index.html https://www.python.org/ https://github.com/lh3/bwa https://gatk.broadinstitute.org/hc/en-us (Continued on next page) Article Continued REAGENT or RESOURCE SOURCE IDENTIFIER ll OPEN ACCESS HapCUT2 PyMOL Other Code for main ﬁgures and analysis This paper; Zenodo PyMOL was used for ABCB11 schematic in Figure 4 using PBID: 6LR0 RESOURCE AVAILABILITY Lead contact https://github.com/vibansal/HapCUT2 https://pymol.org/2/ emauryg/SCZ_sCNV_paper_repo: Publication release (v1.0.0). Zenodo. https://doi.org/10.5281/zenodo.7778664 N/A Further information requests for resources and reagents should be directed to and will be fulﬁlled by lead contact, Christopher A. Walsh ([email protected]). Materials availability All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement. Data and code availability d Individual level SNP-array data is part of the Psychiatric Genomic Consortium with the corresponding privacy agreement. Ac- cess can be provided by applying through this website (https://www.med.unc.edu/pgc/shared-methods/how-to/). Whole genome sequncing data for validation experiments will be uploaded to the NIMH Data Archive after publication NDA: (https://nda.nih.gov/). d Filtered sCNV callset is inTable S2. d Scripts used to generate the main ﬁgures and analyses are available in a frozen Zenodo repository Zenodo: https://doi.org/10. 5281/zenodo.7778664. d PyMOL was used for ABCB11 schematic in Figure 4 using PBID: 6LR0. d Any additional information required to reanalyze the data reported in this paper is available. EXPERIMENTAL MODEL AND SUBJECT DETAILS SNP array data acquisition Allelic intensity data for cases and controls were obtained from the Psychiatric Genomic Consortium (PGC) CNV working group. The exact details of the data generation were previously described,23 removing samples derived from cell lines. SNP array data consisting of 13,464 SCZ cases and 12,722 controls was obtained. These data were proﬁled with the Illumina OmniExpress, OmniExpress plus exome chip, Illum610K, and Affymetrix SNP6.0 arrays. For each determined position the B allele frequency (BAF; proportion of B allele), Log-R ratio (LRR; total genotyping intensity of A and B alleles), and genotype calls, were calculated. Data processing The genotypes from the SNPs from the arrays were phased using the Eagle264 software. Then, the BCFtools plug-in MoChA (2021- 01-20 release) was used to conﬁdently call mosaic CNVs, by taking advantage of long-range haplotype phasing of heterozygous SNP sites and BAF estimates of genotype array data. Genotyping and intensity data from Illumina platforms were distributed by the PGC in the Illumina GenomeStudio Final Report format, with the genomic positions genotyped using the hg18 human reference genome. To convert the Final Report format to VCF format, the rsID numbers were used to liftover coordinates to hg19, discarding positions without rsID, similar to Sherman et al.3 Custom scripts were used to transform Final Reports to binary VCF format, and Illumina’s TOP-BOT format was converted to dbSNP REF-ALT format using a modiﬁed version of BCFtools plug-in ﬁx-ref. MoChA calculates cell fraction from BAF as follows: j0:5 (cid:1) 1 = CNj = DBAF; CF = jCN (cid:1) 2j where CN is the copy number and DBAF is the deviation of B allele fraction compared to 0.50. This equation is valid for gains and losses. Cell Genomics 3, 100356, August 9, 2023 e2 ll OPEN ACCESS METHOD DETAILS Variant level quality control Article In accordance with the suggestions of the MoChA processing pipeline, the following variants were ﬁltered out: more than 2% ge- notypes missing, evidence of excess heterozygosity (p < 1e-6, Hardy-Weinberg equilibrium test), correlation of autosomal geno- types with sex (Fisher exact test comparing number of 0/0 genotypes vs. number of 1/1 genotypes in males and females), variants falling within segmental duplications with low divergence (<2%). This variant-level QC was performed on each separate batch. Sample-level quality control In order to ﬁlter out samples with contamination from another individual two statistics were calculated: BAF concordance and BAF autocorrelation. Brieﬂy, BAF concordance calculates the probability that an adjacent heterozygous SNP has a deviation from a BAF of 0.5 given that the previous heterozygous site had the same deviation from 0.5.65 BAF autocorrelation is the correlation of the BAF statistic at consecutive heterozygous sites once adjusted for the genotype phase. Samples with contamination with DNA from another individual would be expected to have a BAF concordance >0.5 and BAF autocorrelation >0 because of allelic intensities correlated at variants within haplotypes shared between sample DNA and contaminated DNA. Samples with BAF concordance >0.51 or BAF autocorrelation >0.03 were removed. Event type classiﬁcation An Expectation Maximization algorithm was applied to classify events as either a Gain, Loss, or CN-LOH. The algorithm determines the slopes that characterizes the relationship between the deviation of the LRR from 0 jDLRRj, and the BAF deviation from 0.5, jDBAFj. In other words, the events are classiﬁed based on the optimization of linear regression parameters described by jDLRRj = jDBAFjb cÞ is the error for each c event-type clustering. c is the slope for each event type, e (cid:3) Nð0; s2 + e, where c ˛ fGain; Loss; CN (cid:1) LOHg, b To further enhance the robustness of the classiﬁcation method, we used the fact that CN-LOH events are expected to be less com- mon within the chromosomes compared to events that extend to the telomeres. Since CN-LOH events are thought to arise during mitotic recombination, for them to occur within a chromosome would require a double crossover, which is highly unlikely. To incor- porate this information into the classiﬁcation model, we estimated the frequency using the UK Biobank sCNV calls18,19 for of each event type occurring on telomeres and interstitially. These frequencies were used as priors to multiply the likelihoods for each event type, resulting in posterior probabilities. The computation for each event Si is as follows: Let X = jDBAFj and Y = jDLRRj, then PrðSi = c j Li; Xi; YiÞfPrðLiÞ e(cid:1) ðYi (cid:1) Xi bcÞ2=2 s2 c , where Li is an indicator of whether the event involves a telomere, and c is deﬁned as above. This estimation is calculated for each event type and then normalized to sum to one. Filtration of mosaic CNV calls Filtration was focused on removing potential germline events and events likely to arise due to age-related clonal hematopoiesis, as well as artifacts. We required events to have a log10-odds >10 for the model based on BAF and phase, which measures how much more likely the data for a given segment of DNA is consistent with a non-diploid model than a diploid model. Events that were classiﬁed as copy number polymorphism (known CNV polymorphisms in 1000 Genomes Project) by MoChA were ﬁltered out as possible germline events. We further excluded events that had a reciprocal overlap with events from control sam- ples or with any CNVs reported in the 1000 Genomes project by >50%. Events that overlapped >50% with germline events pre- viously identiﬁed in the same sample by the PGC23 were also removed for duplications, since small duplications with high BAF deviations can be mistakenly identiﬁed as somatic variants. Copy number state was taken into consideration when calculating overlaps, i.e. overlap between gains and losses were not considered. Calls with an estimated cell fraction of 1 were also removed. For gains, we further removed any events with a deviation in BAF greater than 0.10 to have a conservative assurance that germline gains were not misclassiﬁed as mosaic, as germline gains tend to be small and produce large deviations from the a BAF of about 1/6.18 Finally, since most of our datasets did not include age information for individuals besides the broad estimate of being younger than 40, we used a conservative approach to remove events that could have risen from clonal hematopoiesis. CN-LOH events were fully excluded from any downstream analysis as these events have been shown to be largely enriched in clonal hematopoiesis events.18 We also removed sCNVs that contained loci commonly altered within the immune system, speciﬁcally IGH (chr14:105,000,000– 108,000,000) and IGL (chr22:22,000,000–40,000,000). We also excluded CNVs within the extended MHC region (chr6:19,000,000– 40,000,000). In addition, we removed deletion involving the following loci that are frequently affected by clonal hematopoiesis: 20q11, DNMT3A, TET2, 13q14, 17p, 5q14, ATM. We removed duplications in 15q. We also removed any sCNVs in 7q34 and 14q11.2, as well as trisomy 12 events. We also removed events whose copy-number state could not be determined. Statistical analysis Overall burden analysis To test the hypothesis of whether more individuals with at least one sCNV of cell fraction greater than a given cell fraction cut-off in cases vs. controls, the two-sided Fisher’s Exact test was used.3 The 95% conﬁdence intervals were calculated using Wilson’s score e3 Cell Genomics 3, 100356, August 9, 2023 Article ll OPEN ACCESS interval. For the meta-analysis using each batch separately we used a one-sided Fisher’s Exact test. The p values were combined using the Tippet’s (minimum p value), and the Liptk’s (weighted sum of p values) approaches. Cell fraction, gene-set, length, and gene number burden analysis To calculate the contribution of the features of gene, length, and gene number burden, we ﬁt a mixed effect logistic regression on the case-control phenotype as the outcome variable. Let yi ˛ f0; 1g be an indicator of whether the subject is diagnosed with SCZ or a control respectively. We modeled the burden as follows: logitðPrðyi = 1ÞÞ = b 0 + b sexXi;sex + b LENGTHXi;LENGTH logitðPrðyi = 1ÞÞ = b 0 + b sexXi;sex + b LENGTHXi;LENGTH + b meanCF Xi;meanCF logitðPrðyi = 1ÞÞ = b 0 + b sexXi;sex + b LENGTHXi;LENGTH + b#genesXi;#genes where XLENGTH and X#genes are the sum of the length and number of genes overlapped by events of individual i, and XmeanCF is the mean cell fraction of the events of individual i. Inference was not altered by the sufﬁcient statistic used to summarize cell fraction (i.e. min, max, median). In the models above we were interested on testing whether bs0 for the feature of interest. The models were ﬁt using a generalized mixed-effect model as implemented by the R package lme462 to account for the sample collection batches of the PGC. Statistical signiﬁcance was assessed using the Satterwhite approximation to the t-test as implemented in the package lmerTest.63 Gene set enrichment analysis We used a similar approach as recommended by Raychaudhri et al.66 to control for event length and rate, which might result in false positive associations with neuronal genes. Namely, we ﬁt the following model logitðPrðyi = 1ÞÞ = b 0 + b sexXi;sex + b LENGTHXi;LENGTH + b#sCNVsXi;#sCNVs + b genesetXgeneset where the parameters are as deﬁned the section above, but with X#sCNVs is the number of sCNVs in that individual, and Xgeneset is the number of genes in an event that intersect a gene-set of interest. We then used the likelihood ratio test to test whether bgenesets 0. We used 3 gene-sets: (1) Brain expressed genes: deﬁned as the top 20% of brain expressed genes from the GTEx GTEx_Analysis_2017- 06-05_v8_RNASeQCv1.1.9_gene_median_tpm.gct.gz (https://www.gtexportal.org/home/datasets). (2) Synaptic genes obtained from SynaptomeDB (http://metamoodics.org/SynaptomeDB/index.php). (3) High pLI genes, i.e. pLI >0.90, obtained from ExAC (ﬁle: fordist_cleaned_nonpsych_z_pli_rec_null_data.txt) (https://gnomad.broadinstitute.org/downloads). Permutation test for enrichment of sCNV overlapping exons 1–5 of NRXN1 We used the R package regioneR60 to randomly shufﬂe the 7 sCNV that overlapped NRXN1 across the NRXN1 locus using the ran- domizeRegions function, to generate a null distribution of overlaps to perform a boostrap test. We added a padding of 1Mb to the 50 and 30 ends of the NRXN1 locus. After randomly shufﬂing the sCNV we counted how many segments overlapped exons 1–5. We repeated this procedure 10,000 times. The p value was calculated empirically by the fraction of overlaps greater than the observed 5. Since we performed 10,000 iterations our smaller possible p value was 0.0001. Germline CNV analyses We obtained gCNV ﬁnal calls from the SCZ Phase 2 study by the PGC CNV working group.23 We narrowed down the gCNV calls to those that were identiﬁed in the same genotype arrays that were analyzed for sCNVs. To further control for sensitivity between the methods used to call sCNVs and gCNVs we focused on gCNV events with size >100Kb. Length analysis were performed using a log- normal mixed effect model framework using sample batch as the random effect. Gene burden analysis was done with a negative binomial mixed effect model using batch as a random effect, and log(event length) as a covariate. Breakpoint microhomology analysis For the NRXN1 somatic deletions, we identiﬁed the breakpoints at the single base resolution by looking for clipped reads with IGV61 in the vicinity of discordant paired reads mapping to genomic locations that implied a larger insert size than expected. Microhomology was identiﬁed by looking at the surrounding bases of the clipped reads covering the breakpoint and looking for corresponding iden- tical basepairs. Characterization of the mechanism of origin was identiﬁed using the strategy described in Yang et al.35 In brief, if there was no microhomology nor insertions >10 bp, the event was predicted to be created by non-homologous end-joining repair (NHEJ). If there was a microhomology >2 bp but <100 bp, the event was classiﬁed as alternative end joining (alt-EJ). If the microhomology was >100bp, which was not observed in this study, the event was classiﬁed as non-allelic homologous repair (NHAR). Cell Genomics 3, 100356, August 9, 2023 e4 ll OPEN ACCESS Article The cell fraction of the events was estimated by identifying the breakpoints as above, and counting the number of clipped reads supporting the breakpoints from IGV images. Speciﬁcally, the number of clipped reads was divided by the sequencing depth at that site and multiplied by 2. For each event, the estimate of the cell fraction was obtained from the breakpoint with the highest coverage. In situ Hi-C from hiPSC-derived neurons Forebrain neurons were generated as previously described.31 Brieﬂy, neural precursor cells (NPCs) derived from hiPSCs with het- erozygous germline deletions in the 50-end (exons 1–2), 30-end (exons 21–23) and from an hiPSC line with no germline deletion in NRXN1 were seeded at low density and cultured in neural differentiation medium (DMEM/F12, 1xN2, 1xB27-RA, 20 ng mL(cid:1)1 BDNF (Peprotech), 20 ng mL(cid:1)1 GDNF (Peprotech), 1mM dibutyryl-cyclic AMP (Sigma), 200nM ascorbic acid (Sigma) and 1 mgml(cid:1)1 laminin (ThermoFisher Scientiﬁc) 1–2 days later. Cells were maintained in differentiation medium for 7.5 weeks before harvesting. In situ Hi-C libraries were generated from 500K to 1 million cultured hiPSC-derived neurons using the Arima Hi-C kit (Arima Geno- mics, San Diego) per manufacturer’s instructions without modiﬁcations. Brieﬂy, in situ Hi-C consists of 7 steps: (1) crosslinking cells with formaldehyde, (2) digestion of the DNA using a proprietary restriction enzyme cocktail within intact nuclei, (3) ﬁlling and bio- tinylation of the resulting 50-overhangs, (4) ligation of blunt ends, (5) shearing of the DNA, (6) pull down of the biotinylated ligation junc- tions with streptavidin beads, and (7) analyzing these fragments using paired end sequencing. The resulting Hi-C libraries were sequenced on the Illumina HiSeq1000 platform (125bp paired-end) (New York Genome Center). Hi-C read alignment Hi-C reads were aligned to the hg19 reference genome using bwa mem (v0.7.17-r1188) using the ﬂags ‘‘-SP5M’’ (‘‘-SP’’ for aligning each end of the paired end reads separately, ‘‘-5’’ to force always reporting the 50 part of a chimeric read as primary). Aligned reads were subsequently used for two different tasks: 1) variant calling with the GATK pipeline followed by HapCUT2 phasing, and 2) Hi-C matrix construction via pairtools. Preprocessing for variant calling Duplicate Hi-C reads were marked using Picard’s MarkDuplicates (via GATK, v4.0.12.0). Bamﬁles were recalibrated using the GATK BQSR (base quality score recalibration) procedure. Brieﬂy, BaseRecalibrator was run using dbSNP build 138, the Mills +1000 Ge- nomes gold standard indels, and the 1000 Genomes Phase I gold standard indels as reference variants. The recalibration adjustment was then applied with ApplyBQSR. Variant calling for Hi-C analysis Deduplicated and recalibrated Hi-C reads were then processed using the GATK (v4.0.12.0) germline short-read variant discovery pipeline. Brieﬂy, HaplotypeCaller was run in gVCF mode (ﬂags ‘‘-ERC GVCF’’) using dbSNP build 138 as a reference. Merged gVCFs then were converted to genomicsDB format with GenomicsDBImport and genotypes were called against this genomicsDB with GenotypeVCFs. Variant quality scores were separately recalibrated for SNVs and indels via the GATK VQSR (variant quality score recalibration) procedure. Brieﬂy, separate VQSR models were built for SNVs and indels using VariantRecalibrator, run in SNP or INDEL mode, respectively. The reference variants used for SNV quality recalibration were: HapMap variants (v3.3): training and truth, prior of 15. 1000 Genomes "Omni" platform variants (v2.5): training and truth, prior of 12. 1000 Genomes Phase I gold standard SNPs: training only, prior of 10 dbSNP variants without 1000 Genomes (build 138, excluding sites after build 129): known, prior of 2. The reference variants used for indel quality recalibration were: Mills +1000 Genomes gold standard indels: training and truth, prior of 12. The ﬂags ‘‘–max-gaussians 2 -an QD -an MQ -an ReadPosRankSum -an FS -an SOR -an DP’’ were used when building the SNV recalibration model, and the ﬂags ‘‘–max-gaussians 4 -an QD -an DP -an FS -an SOR -an ReadPosRankSum’’ were used when build- ing the indel recalibration model. The VQSR models for SNVs and indels were then applied using ApplyVQSR in SNP or INDEL mode, respectively, with a truth sensi- tivity ﬁlter level of 99. Haplotype phasing for Hi-C analysis Haplotypes were phased using HapCUT2. Brieﬂy, recalibrated and ﬁltered variants were separated for each sample, then HAIRS were extracted with extractHAIRS with ﬂags ‘‘–hic 1 –indels 1’’. HAPCUT2 was then run with ﬂag ‘‘–hic 1’’. Each Hi-C read was then assigned to one of the two haplotype blocks called by HapCUT2 by counting how many variants that overlapped the read were part of each haplotype block. If a read overlapped multiple variants that were phased to different haplotype blocks, a majority voting system was used to assign those reads to the haplotype block that had more variants overlapping that read. If an equal number of variants from each haplotype block overlapped the read, the read was discarded from the phasing process. e5 Cell Genomics 3, 100356, August 9, 2023 Article ll OPEN ACCESS Hi-C matrix construction and visualization Hi-C matrices were constructed from mapped reads using the pairtools pipeline. Brieﬂy, Hi-C read pairs were parsed, sorted, merged, and deduplicated. Restriction fragments were assigned to read pairs by using ‘‘pairtools restrict’’ with a restriction fragment bedﬁle generated using the ‘‘digest_genome.py’’ script from HiC-Pro. Phased pairsﬁles were generated by subsetting the unphased pairsﬁle to only those reads that were phased to a speciﬁc haplotype block. Phased and unphased pairsﬁles were used to assemble contact matrices using the ‘‘juicer pre’’ command in juicer_tools (v1.8.9), using a MAPQ threshold of 10. Phased matrices were assembled at 40 kKb resolution, while unphased matrices were assembled at 10 kKb resolution. Unphased matrices were balanced using the KR (Knight-Ruiz) normalization implemented in juicer_tools and visualized in balanced form. Phased matrices were visualized in unbalanced form. H3K27ac ChIP-seq tracks from ENCODE (H1 neurons, Bern- stein Lab, ENCODE ID ENCFF516KKW) were overlaid on the heatmaps. Analysis of human postmortem snRNA-seq datasets We gathered three publicly available postmortem snRNA-seq datasets from two studies.52,53 We used publicly available annotations from both studies to identify cell types. To determine the log-normalized expression of ABCB11 across these datasets, we normal- ized gene expression to the total number of transcripts sampled per cell, multiplied by 10000, added a pseudocount of 1, and log- transformed the data. We then averaged expression for each cell type for each cell type for each donor from the studies (e.g. 2 donors from the Bakken et al. and 8 neurotypical controls from Kamath et al.) in order to account for intra-individual variation. The uniform manifold approximation (UMAP) low-dimension embedding shown is taken from a previous analysis of the SN dataset.52 Single-molecule in situ hybridization (smFISH) and imaging of postmortem human nigra Postmortem human midbrain tissues ﬂash frozen in (cid:1)80(cid:4)C were cryosectioned at (cid:1)15 to (cid:1)20(cid:4)C to make 12-micron sections on SuperFrost Plus slides. The slides were then allowed to warm up to room temperature (RT) before being placed in 4% PFA for 15 min at RT. Slides were next washed three times with 70% ethanol for 5 min followed by a 2-h 70% ethanol wash at RT. Subse- quently, slides were incubated at 37(cid:4)C in the Probe Hybridization buffer (Molecular Instruments) for 10 min in a humidiﬁed chamber to pre-hybridize. At this time, the probe solution was prepared by adding 0.4 pmol of each probe set (Molecular Instruments) per 100 mL of Probe Hybridization buffer and vortexed to ensure proper mixing. The Probe Hybridization buffer was then replaced by the probe solution and the slides were incubated overnight at 37(cid:4)C in a humidiﬁed chamber. After 18–24 h, sections were sequentially washed for 15 min each in the following solutions at 37(cid:4)C in a humidiﬁed chamber: (1) 75% Probe Wash buffer (Molecular Instruments) and 25% 5x SSCT (SSC +10% Tween 20), (2) 50% probe wash buffer and 50% 5x SSCT, (3) 25% probe wash buffer and 75% 5X SSCT, and (4) 100% 5x SSCT. The slides were then washed for 5 min at room temperature in 5x SSCT. Slides were then allowed to pre- amplify in the Probe Ampliﬁcation buffer (Molecular Instruments) for 30+ minutes at RT. During this time, the hairpins (Molecular Instruments) are prepared. Approximately, 1uL of hairpin for every 100uL of ﬁnal ampliﬁcation solution were snap-cooled in a PCR thermocycler with the following settings: 95(cid:4) for 90 s, cool to room temperature (20(cid:4)C) at a rate of 3(cid:4) per minute. After snap- cooling, hairpins were added to the desired volume of the ampliﬁcation buffer. Slides were incubated overnight at RT in a humidiﬁed chamber. After overnight incubation, the slides are washed twice for 30 min at room temperature with 5x SSCT. An appropriate amount of Fluoromount Gold with NucBlue (Thermo Fisher) was added to the slides which then are coverslipped. Slides were stored at 4(cid:4)C until imaging. We used the following probe accession numbers: TH (NM_000360.4), CALB1 (NM_001366795), ABCB11 (NM_003742.4). Imaging was performed with either a: DragonFly confocal scanner unit with an Andor Zyla 4.2 Plus camera (for high resolution im- ages of DA neurons) or a Keyence BZ800XE microscope (for tiled image of overview SN). Images were acquired using either a Nikon Apo 10x objection (for the overview SN) or Nikon Apo 40x/1.15 WI objective for the (high resolution images of DA neurons). Cell Genomics 3, 100356, August 9, 2023 e6
10.1021_acscentsci.3c00160	http://pubs.acs.org/journal/acscii Research Article Macrocyclization and Backbone Rearrangement During RiPP Biosynthesis by a SAM-Dependent Domain-of-Unknown-Function 692 Richard S. Ayikpoe, Lingyang Zhu, Jeff Y. Chen, Chi P. Ting, and Wilfred A. van der Donk* Cite This: ACS Cent. Sci. 2023, 9, 1008−1018 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: The domain of unknown function 692 (DUF692) is an emerging family of post-translational modification enzymes involved in the biosynthesis of ribosomally synthesized and post- translationally modified peptide (RiPP) natural products. Members of this family are multinuclear iron-containing enzymes, and only two members have been functionally characterized to date: MbnB and TglH. Here, we used bioinformatics to select another member of the DUF692 family, ChrH, that is encoded in the genomes of the Chryseobacterium genus along with a partner protein ChrI. We structurally characterized the ChrH reaction product and show that the enzyme complex catalyzes an unprecedented chemical transformation that results in the formation of a macrocycle, an imidazolidinedione heterocycle, two thioaminals, and a thiomethyl group. Based on isotopic labeling studies, we propose a mechanism for the four-electron oxidation and methylation of the substrate peptide. This work identifies the first SAM-dependent reaction catalyzed by a DUF692 enzyme complex, further expanding the repertoire of remarkable reactions catalyzed by these enzymes. Based on the three currently characterized DUF692 family members, we suggest the family be called multinuclear non-heme iron dependent oxidative enzymes (MNIOs). ■ INTRODUCTION Recent advances in bioinformatic tools and an explosion in the publicly available genomic data have led to the identification of many new peptide-based natural products.1−6 Often, these compounds possess antimicrobial, antiviral, antifungal, herbi- cidal, or cytotoxic properties.7−12 Ribosomally synthesized and post-translationally modified peptides (RiPPs) have garnered significant attention over the past decade owing to their remarkable structural diversity and biological activities.10,12−14 RiPPs are produced from genetically encoded precursor peptides that are modified by post-translational modification (PTM) enzymes encoded in the same biosynthetic gene clusters (BGCs). Many of the post-translational modification reactions are unique chemical transformations resulting in macrocycles and heterocycles, structures that are privileged in bioactive scaffolds. Furthermore, exciting recent developments in biocatalysis have increasingly shown the value of enzymatic transformations in preparation of compounds that are useful to human society, with many of the enzymes coming from biosynthetic pathways to natural products.15−18 Therefore, discovery of new chemical reactions catalyzed by enzymes and assignment of function to poorly characterized enzyme families is an important goal. A common biosynthetic feature of most RiPPs is the ribosomal production of a precursor peptide consisting of a leader and a core sequence. The leader peptide facilitates recognition and processing by the PTM enzymes,19 with modifications imparted to the core peptide. The modified peptide undergoes proteolytic removal of the leader peptide, and the matured peptide is in most cases exported from the cell to exert its biological function. More than 40 families of RiPPs have been classified based on the chemical transformations made to the precursor peptide.13,14 One emerging and underexplored group of RiPPs is generated by multinuclear iron-dependent enzymes belonging to the DUF692 family. Members of this enzyme family are structurally related to the triose-phosphate isomerase family,20 and only two members of the DUF692 enzyme family have been functionally charac- terized thus far, MbnB21−23 and TglH.24,25 MbnB forms a heterodimeric complex with MbnC to catalyze a central step in the biosynthesis of methanobactin, a copper-chelating peptidic compound produced by methano- trophic bacteria under copper limiting conditions. During the maturation of methanobactin, MbnBC catalyzes the four- electron oxidation of Cys residues in its precursor peptide, MbnA, to an oxazolone and an adjacent thioamide group (Figure 1A).21−23 The second characterized member, TglH, is Received: February 7, 2023 Published: April 24, 2023 © 2023 The Authors. Published by American Chemical Society 1008 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 1. Bioinformatic analysis of the DUF692 family. (A) Sequence similarity network of DUF692 enzymes using an E-value of 50. The orthologs of MbnB are colored magenta and the orthologs of TglH in green. The group of BGCs investigated in this work are colored orange. The cytoscape file for the SSN is provided in the SI. (B) Representative gene organization of the Chryseobacterium biosynthetic gene clusters. (C) Sequence logo of the precursor peptides from 115 BGCs showing conservation of a terminal CPACGMG motif. An Excel file containing precursor peptide sequences used to generate the sequence logo is provided in the SI. involved in the maturation of 3-thiaglutamate, an amino acid- derived natural product biosynthesized at the C-terminus of a carrier peptide.24 TglH forms a complex with a second protein TglI to catalyze the β-carbon excision of a C-terminal Cys residue to generate a 2-mercaptoglycine residue (Figure 1A).24−26 Although they belong to the same enzyme family, TglH and MbnB catalyze completely different but equally remarkable chemical substrate peptides. This precedent motivated us to investigate additional the DUF692 family. Herein, we used bio- members of transformations on their informatics to uncover a new RiPP BGC that encodes a DUF692 homologue. This BGC is present predominantly in the genomes of several members of the Chryseobacterium genus. We therefore propose the name chryseobasin for the natural product of the pathway. Structural investigations of the product of the DUF692 enzyme revealed an unprecedented chemical transformation distinct from those catalyzed by the two previously characterized members. 1009 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 2. Modification of ChrA by ChrHI in E. coli. (A) Sequence of ChrA showing the C-terminal fragment after digestion with endoproteinase LysC. (B) MALDI-TOF MS spectra of the unmodified and modified full-length ChrA (left) and the corresponding LysC-digested peptides (right). ■ RESULTS AND DISCUSSION RiPP Biosynthetic Pathways That Encode Uncharac- terized DUF692 Enzymes. To identify additional members of the DUF692 family, we generated a sequence similarity network (SSN) using the Enzyme Function Initiative Enzyme Similarity Tool (EFI-EST).27 A total of 13,108 sequences of the PF05114 family were used in the initial SSN analysis. For the SSN shown in Figure 1, an E-value of 10−50 and a sequence alignment score threshold of 70 were used resulting in more than 100 groups. These groups were colored based on the gene context of each cluster, and analyzed using the EFI’s genome neighborhood network (GNN) tool.28 The analysis identified clusters that are associated with the two characterized members TglH and MbnB (Figure 1A). In addition, the analyses also revealed additional clustered family members including a group of enzymes predominantly found in the Chryseobacterium genus. We selected this group of BGCs for further investigation because of the gene organization and the sequences of the putative precursor peptides (Figure 1B and C), which suggested products distinct from methanobactin and 3-thiaglutamate. ID: ID: This group of BGCs encodes proteins annotated as a precursor peptide (ChrA; Uniprot IW22_14840), a IW22_14845), a DUF692 enzyme (ChrH; Uniprot DUF1772-containing predicted integral membrane protein (ChrI; Uniprot IDs: IW22_14850), a ribulose phosphate 3- epimerase (ChrE), and an M42 peptidase (ChrP). A sequence logo of the precursor peptides from 115 BGCs revealed a highly conserved CPACGMG motif at the C-terminus (Figure 1C). While our bioinformatic discovery was based on an enzyme SSN followed by GNN analysis, similar BGCs were also recently identified using a co-occurrence based ap- proach.29 ChrH Requires ChrI to Modify the CPACGMG Motif of ChrA. The two previously characterized members of the DUF692 family both modify cysteine residues on their substrates. We therefore anticipated that the conserved Cys residues in ChrA (Figure 2A) might be modified by ChrH. To probe this hypothesis, we employed a heterologous coex- pression approach. The precursor peptide ChrA was expressed in Escherichia coli with an N-terminal His6-tag. Purification of ChrA by immobilized metal affinity chromatography (IMAC) and subsequent analysis by matrix-assisted laser desorption/ ionization time-of-flight mass spectrometry (MALDI-TOF MS) gave the expected mass for the unmodified linear peptide (Figure 2B). When ChrA was coexpressed with ChrH in lysogeny broth (LB) supplemented with 50 μM iron(II) citrate, no change in the mass of the purified peptide was observed. Both MbnB and TglH require a second protein for activity.21,24 MbnB forms a heterodimer with MbnC,22,23 and TglH forms a predicted heterodimer with TglI.25 Both MbnC their cognate and TglI recognize the leader peptides of substrates by formation of an antiparallel β-sheet,22 with TglI having the typical fold of a RiPP precursor peptide Recognition Element (RRE).30 The integral membrane protein ChrI is not homologous in sequence to MbnC or TglI and was not bioinformatically predicted to contain an RRE,31 nor do any of the other enzymes in the chr BGC contain a predicted RRE. ChrI does contain a DUF1772 that has been speculated to play a role in protein−protein interactions.32 We therefore next included ChrI in the expression system. Following coex- pression of all three proteins, a new peak consistent with a 10 Da increase in mass of ChrA was observed by MALDI-TOF MS (Figure 2B). Digestion of isolated modified and unmodified peptide with endoproteinase LysC suggested the modification is localized to the C-terminus of ChrA (Figure 2B). The 10 Da mass increase was confirmed by high- resolution mass spectrometric (HRMS) analysis (Figures S1 and S2). We designate this LysC-digested peptide ChrA. To determine if the conserved Cys residues were modified, we first derivatized ChrA and ChrA* with N-ethylmaleimide (NEM) to probe for free thiols. Analysis of the derivatized peptides by MALDI-TOF MS showed that ChrHI-processed ChrA did not react with NEM (Figure S3), suggesting that the thiols have been modified. To corroborate this conclusion, the 1010 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 3. NMR analysis of ChrA. (A) TOCSY spectrum showing the loss of amide NH signals of Gly9 and Met10 and a change in the chemical shift of the βH of Cys8 (blue box). (B) NOESY spectrum focusing on important NOE correlations in the macrocycle and heterocycle of ChrA. (C) HMBC spectrum highlighting the through-bond correlations of Cys8-Cα-S-CH3, and the correlations that support the macrocycle and the heterocycle of ChrA. (D) HMBC spectrum of 1-13C-Cys labeled ChrA showing the connectivities of the former Cys8 carbonyl carbon with resonances derived from Cys8, Gly9, and Met10 (blue box) consistent with an imidazolidinedione. (E) HMBC spectrum of 3-13C-Cys labeled ChrA* highlighting the connectivity between the βH of Cys5 and the βC of Cys8 as well as an important connectivity between the macrocycle and heterocycle (blue boxes). (F) Proposed structure of ChrA* showing all important correlations identified from the NMR spectral analysis. (G) Proposed reaction catalyzed by ChrHI. two conserved Cys residues in ChrA, Cys63 and Cys66, were mutated to Ser residues. Neither single nor double mutants resulted in formation of the product that had increased by 10 Da, suggesting that both Cys residues are modified by ChrH (Figure S4). The ChrA-C63S variant resulted in partial formation of a product with a decrease in mass of 36 Da that will be further discussed below. Taken together, these results suggest that ChrH, like TglH and MbnB, requires a complex with a second protein to install post-translational modifications on ChrA and that this transformation involves the thiols of the two conserved Cys residues of ChrA. Structure Elucidation of ChrA. To gain more detailed structural information regarding the reaction product of ChrHI, both unmodified and modified ChrA were produced on a larger scale and subjected to detailed one-dimensional and two-dimensional nuclear magnetic resonance (NMR) analysis after endoproteinase LysC digestion to reduce the size to an 11mer peptide. Analysis of TOCSY data for LysC-digested ChrA identified all 11 amino acids including nine amide protons (Figure S5 and Table S1), whereas only seven spin systems were identified for ChrA associated with just seven amide protons (Figures 3A, S6, and Table S2). Significantly, the amide protons of Gly9 and Met10 (Gly67 and Met68 of full-length peptide) were missing in ChrA, suggesting that these two amides were modified (Figures 3A and S6). Subsequent 1H−13C-HSQC analysis of substrate and product also revealed a new cross peak with chemical shifts at 1.96 (1H) and 10 ppm (13C) (Figure S6). Interestingly, this new peak integrated in the 1H spectrum to three protons and appeared as a singlet suggesting the possible introduction of a methyl group. The 1H and 13C chemical shifts of this methyl group suggest attachment to either carbon or a heteroatom like sulfur that is not highly electronegative. In addition, the HSQC data identified the disappearance of the CH2 group at the β position of the former Cys8. Instead, a new CH cross peak was observed in ChrA at 6.04 ppm, a significant change in the chemical shift compared to the original β-protons of Cys8 of ChrA (3.36 and 2.96 ppm; Figure S6 and Tables S1 and S2). As we will show, this new signal arises from one of the β-protons of the former Cys8, and we will refer to it as such in the remainder of the discussion. The chemical shift of the αH of the former Cys8 also changed from 4.44 ppm in ChrA to 4.94 ppm in ChrA* (Figures S6 and Tables S1 and S2). The assignments of the new signals originating from the α and β protons of the former Cys8 in ChrA* were supported by the 1H−1H dqCOSY spectrum, which showed cross-peaks between the proton at 4.94 ppm and the NH of the former Cys8, and between the two protons at 4.94 and 6.04 ppm (Figure S6). To probe the chemical environment of the new signals further, 1H−13C-HMBC and 1H−1H NOESY experiments were carried out on ChrA. The NOESY spectra showed direct 1011 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 4. HR-ESI tandem mass spectrometry analysis of ChrA digested with LysC. (A) Sequence of ChrA* and observed fragment ions. (B) Tandem mass spectrum showing ions and their corresponding fragments. (C) A graph of the ppm errors for each identified ion in panel B.33 correlations between the new methyl group and the αH of the former Cys8. In addition, a second NOE correlation was observed from the βH of Cys5 to the signal at 6.04 ppm associated with the βH of the former Cys8 (Figures 3B and S6), as well as a weaker NOE between the βH of Cys5 and the signal at 4.94 ppm associated with the αH of the former Cys8. The HMBC data also showed connectivity (via 3-bond correlations) between the new methyl group and the αC of the former Cys8 and vice versa. Another connectivity between the βC of Cys5 and the 1H peak at 6.04 ppm (former βH of Cys8) was observed, suggesting the formation of a macrocycle. The HMBC spectrum also revealed direct connectivities between the carbonyl of the former Cys8 (156.5 ppm) and the αH of Gly9 and αH of Met10, indicating a significant rearrangement in ChrA* (Figure 3C). These preliminary experiments suggested that Cys8 is significantly altered. To probe the fate of the individual carbon atoms of Cys5 and Cys8, we first prepared ChrA* generated from ChrA that was selectively labeled with 1-13C-Cys (label at the carbonyl carbon) by heterologous expression in E. coli for HMBC NMR analysis. As expected, two carbon peaks, one at 170.0 ppm for Cys5 and the other at 156.5 ppm for the former Cys8, were observed in the 13C NMR spectrum (Figure S7), corroborating a significant rearrangement of the position of the carbonyl carbon of the former Cys8. The HMBC spectrum also revealed direct correlations between the βH of the former Cys8 to the carbonyl of the former Cys8, the αH of Met10 to the carbonyl carbons of Gly9 as well as the former Cys8, and from the αH of Gly9 to the carbonyl carbons of Gly9 and the former Cys8 (Figures 3D and S7). When ChrA* was labeled with 3-13C-Cys (label at the β-carbon), the acquired HMBC data showed through-bond connectivities from the βC of Cys5 to the βH of the former Cys8 and vice versa, suggesting that the sulfur of Cys5 is directly connected to the Cβ of the former Cys8 (Figures 3E and S8). Collectively, all collected NMR data are consistent with the structure shown in Figure 3F. The stereochemistry at the two new stereogenic centers of the former Cys8 is currently not known. However, the 3JH−H coupling constant between the protons at 4.94 and 6.04 ppm is 11.6 Hz, suggesting that the relative stereochemistry involves a trans arrangement. To corroborate the findings from the NMR data, ChrA* was next analyzed by high-resolution electrospray ionization tandem mass spectrometry (HR-ESI MS/MS). The HR-ESI 1012 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 5. Determination of the methyl source and in vitro reconstitution of ChrHI activity. (A) MALDI-TOF MS data of ChrA* produced by heterologous coexpression of ChrA with ChrHI in M9 media supplemented with either 13CD3-Met or selenomethionine and treatment of the purified peptide with endoproteinase LysC. (B) MALDI-TOF MS of LysC-digested in vitro reaction products of ChrHI in the presence and absence of SAM. The dashed lines indicate the expected m/z of modified and unmodified peptides. MS/MS spectrum localized the modifications to the C- terminal CGMG peptide as no fragments corresponding to b-ions could be observed beyond this point, whereas all amino acids in the N-terminal segment were unmodified (Figure 4). The observation of fragment ions indicated as “y4” and “b7” (since the structure is no longer an α-amino acid peptide these fragments are not true y and b ions) is explained by cleavage of the thioaminal linkages in the proposed structure (Table S3). In addition, various internal fragment ions consistent with the proposed structure of ChrA* were observed, especially fragmentation in the two thioether bonds as well as the amide bonds of the imidazolidinedione. Taken together, the HR-ESI MS/MS and the NMR data suggest that ChrHI catalyzes an unprecedented multistep chemical transformation on ChrA: macrocyclization to form a cross-link between the thiol of Cys5 and the βC of the former Cys8, migration of the carbonyl carbon of the former Cys8 to become inserted between the amide nitrogens of Gly9 and Met10 to form an imidazolidinedione, and methylation and migration of the thiol from Cβ to Cα of the former Cys8. Thus, the net +10 Da transformation from ChrA observed by MS involves a −4 Da change resulting from two oxidative ring formations in which the amide protons of Gly9 and Met10 are removed as well as the thiol proton of Cys5 and one of the β-protons of the former Cys8, and a +14 Da change from the methylation event. In Vitro Reconstitution of ChrHI Activity and Identification of S-Adenosylmethionine as the Methyl Donor. The proposed structure of ChrA* generated in E. coli contains a new methyl group that was not present in the ribosomally generated precursor peptide. This finding raised a question regarding the source of the methyl group. To determine its origin, isotope feeding experiments were carried out. ChrA was coexpressed with ChrHI in E. coli in M9 minimal media supplemented with 13CD3-Met.34 Isolation of the modified peptide and subsequent MALDI-TOF MS analysis of the full-length and LysC-digested peptide revealed the incorporation of two methyl groups from 13CD3-Met into ChrA, one in Met10 and one in the new thiomethyl group (Figures 5A and S9). When selenomethionine was substituted for 13CD3-methionine, the isolated product contained four selenium atoms in the full-length modified ChrA consistent with the four Met residues in ChrA and only one selenium atom in ChrA as evidenced by the isotope distribution pattern (Figures 5A and S10).35 These results suggest that only the methyl group (and not the thiomethyl group) of methionine is incorporated into ChrA* and suggests that either S-adenosylmethionine (SAM) or possibly a methionine residue of the modifying enzyme ChrHI could be the methyl donor. SAM is the most common methyl donor in biology.36 We next aimed to reconstitute the activity of ChrHI in vitro to probe for the possibility of SAM as the methyl source. ChrH was heterologously expressed in and aerobically purified from E. coli Rosetta-pLysS cells as an N-terminally His6-tagged protein (Figure S11A). The purified protein exhibited a purple color. The amount of iron was quantified using the ferene assay,37 and the as-isolated enzyme contained 1.9 equiv of iron per monomer of ChrH. Since the coexpression studies revealed the requirement of ChrI for ChrH activity, the predicted integral membrane protein ChrI was also heterologously expressed in E. coli as an N-terminally hexahistidine-tagged protein and partially purified in the presence of detergent (0.01% n-dodecyl-β-D-maltopyranoside) (Figure S11A). Re- actions were carried out aerobically by incubating ChrA with ChrH, ChrI, and DTT in the presence and absence of SAM. DTT was included because the two Cys residues in the substrate peptide readily form a disulfide (Figures S1 and S2). When the reaction without SAM was analyzed by MALDI- TOF MS, the +10 Da product was not observed, but instead, an ion indicating the loss of −36 Da from ChrA was observed 1013 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article Figure 6. Proposed mechanism for ChrHI catalysis. The order of some steps can be inverted (e.g., the two C−S bond forming steps), and the timing of methylation is not known and could happen much earlier (e.g., in intermediate II or III). For some alternative mechanisms, see Figure S14. (Figures 5B and S12). Such an ion was also observed in small amounts when ChrA was heterologously coexpressed with ChrHI in E. coli (Figures 2B and S3). The difference in mass between the −36 Da species and +10 Da product is 46 Da, consistent with a thiomethyl substituent. Further attempts to isolate the −36 species for detailed structural characterization proved unsuccessful due to degradation under the acidic conditions required for ChrA purification. We next analyzed the in vitro reaction in the presence of SAM resulting in the formation of the +10 Da peptide as the main product together with the −36 Da species (Figures 5B and S12). These results suggest that SAM is indeed the source of the methyl group and show that in vitro and in E. coli the same product is formed, strongly suggesting that the observed transformation is the native function of ChrHI. We next synthesized 13CD3-SAM enzymatically38−40 and used the isotopically labeled product in 1014 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article the in vitro reaction with ChrA and ChrHI. Analysis of the full- length and LysC-digested reaction products revealed the incorporation of one 13CD3-methyl group from 13CD3-SAM into ChrA* but no label incorporation in the −36 Da product. Taken together, these results suggest that SAM is the methyl donor in the post-translational modification of ChrA by ChrHI, thus representing the first example of a DUF692- mediated reaction to use SAM and iron as cofactors. Possible Mechanism of ChrHI Catalysis. With the structure of ChrA* as well as the source of the methyl group established, a proposed mechanism for the ChrHI-catalyzed transformation of ChrA is shown in Figure 6. DUF692 enzymes have been shown to contain two or three iron ions in their active sites (Protein Database accession numbers 3BWW, 7DZ9, 7FC0, 7TCR, 7TCX, 7TCU, and 7TCW).21−24 In the structure of MbnB bound to its partner MbnC and its substrate MbnA, the side chain sulfur atom of a Cys in the substrate is liganded to one of the iron atoms.22 The protein ligands to bind two or three iron ions are conserved in ChrH and present in the active site of an AlphaFold model of the enzyme (Figure S13). Akin to TglH24 and MbnB,22,23 we suggest that the active form of ChrH requires at least two iron ions, one of which is in the Fe(II) form that is used for catalysis and a second (or third) ion that is in the Fe(III) oxidation state that is important for substrate binding and positioning. Such use of iron ions in different oxidation states for different roles is an emerging feature of multinuclear mixed-valent non-heme iron enzymes.41−45 It is possible that the two Cys thiols of ChrA coordinate to different irons in the active site, facilitated by the Pro-Ala sequence that separates the two Cys residues that is an inducer of β-turn formation. Because we currently do not have any structural data on the coordination of the substrate to the metal ions, the mechanism in Figure 6 shows just one of the Cys side chains ligated to Fe as in the MbnABC cocrystal structure.22 The FeII ion in the active site of ChrH is proposed to react with molecular oxygen to form a superoxo-FeIII intermediate I,23,46,47 which initiates the reaction by abstracting a hydrogen atom from the β-carbon of Cys8 as in the proposed mechanisms for TglH and MbnB.21−24 The highly reducing thioketyl radical formed during this abstraction is expected to transfer an electron to one of the Fe(III) ions to generate a thioaldehyde and Fe(II). As drawn in Figure 6, this is a different iron than the iron that activates O2, but it could be the same iron (e.g., some of the mechanisms shown in Figure S14). Subsequent attack of the amide nitrogen of Gly9 onto the thioaldehyde forms a β-lactam II, possibly facilitated by the iron ion functioning as a Lewis acid. Formation of a similar β- lactam was also proposed for MbnBC catalysis as well as other mononuclear non-heme iron enzymes such as isopenicillin N synthase.21−23,48−50 At the reaction of ChrHI this point, diverges from previous enzymes in that the proposed model involves a second nucleophilic attack of the amide nitrogen of the downstream Met10 onto the carbonyl of the β-lactam, which would result in the formation of a bicyclic intermediate III. For MbnBC, the β-lactam is proposed to be attacked by the oxygen atom of the upstream amide,21,22 possibly with the intermediacy of an enzyme bound intermediate.23 The transfer of an electron from Fe(II) in intermediate II to the ferric hydroperoxo intermediate results in generation of a ferryl (FeIV-oxo) species. The exact timing of the formation of this highly reactive intermediate is interesting, as it could be favorable to delay its formation until after (or concomitant with) the formation of the bicycle in intermediate III. One possible means to delay its formation could be by having the reducing equivalent reside on an iron atom that is not bound to the peroxide as shown in Figure 6. Once formed, the ferryl could initiate a proton-coupled electron transfer from the adjacent O−H bond to generate an oxygen radical IV. A well- precedented β-scission reaction51 would result in the opening of the four-membered ring to form a resonance stabilized carbon-based radical V. Although the context and enzyme classes are quite different, β-scission by oxidation of an alcohol by a ferryl was also proposed as a key step in other non-heme iron dependent enzymes with experimental and computational support.42,52,53 We propose that this intermediate radical V could generate the product via formation of an episulfide followed by nucleophilic ring opening by Cys5 and methylation of the thiol (pathway A). Alternatively, oxidation of radical V to a cation VI (pathway B) followed by S- methylation of the thioaminal, migration to the Cβ of the former Cys, and addition of the thiol of Cys5 to the Cα of the former Cys8 would complete the formation of ChrA. We note that many of these steps could occur in a different order, that several steps are expected to require acid−base catalysis, and that other mechanisms can be drawn to arrive at the final product (e.g., Figure S14). We currently do not know whether ChrH or ChrI is responsible for catalyzing the methylation event. An Alpha- Fold-multimer54 model of the complex of ChrHI and ChrA did not predict a Rossmann fold that is typically used to bind SAM (Figure S15). ChrH and its orthologs in the genomes are ∼80−100 amino acids longer than MbnB, which may be the origin of the long unstructured loop in the AlphaFold model. It is possible that the additional amino acids form a SAM binding domain in the presence of substrate. Analysis of ChrI by DeepTMHMM55 suggests the protein is mostly composed of transmembrane helices, and indeed the AlphaFold model predicts a helical bundle for the protein. The predicted interfaces between ChrH and ChrI are quite different from that observed in the structures of MbnBC,22,23 and so is the manner by which ChrA is predicted to engage with ChrHI with the interaction almost entirely between ChrA and ChrH (Figure S16) compared to extensive interactions of MbnA with both MbnB and MbnC.22 Despite the major differences, which may or may not be accurate as they are based on a theoretical model, the Cys residues of ChrA are in close proximity to the iron ligands in the ChrAHI complex (Figure S16). The mechanism in Figure 6 (and alternatives in Figure S14) may also explain the formation of the −36 Da species detected in our assays. Unlike MbnBC and TglHI, ChrHI also uses SAM as a cofactor. In vitro reconstitution of ChrHI revealed in the absence of SAM, only a −36 Da species that accumulates as evidenced by MALDI-TOF MS (Figures 5 and S12). This compound cannot be generated by elimination of methanethiol from ChrA, as that would lead to a product that would be 34 Da lighter than ChrA. We propose that in the absence of SAM, radical V does not form the episulfide (pathway A) or the cation VI (pathway B). Instead, the radical V could be reduced leading to the product shown in pathway C (Figure 6). Regardless of the exact mechanism and the actual structure of the −36 Da product, methylation appears critical to arrive at the structure of ChrA. The mechanisms in Figures 6 and S14 can potentially also explain the outcome with the ChrA variants C63S and C66S. initiated at Cys66, Since the proposed chemistry is 1015 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article residue abolishes all activity. replacement with a Ser In contrast, replacement of Cys63 with Ser can still result in the initial oxidation steps on Cys66. If macrocycle formation involving Cys63 is required for either methylation or thiomethyl migration to the former α-carbon then once again a shunt product can be formed that is 36 Da decreased in mass from the starting peptide (Figure S17). ■ CONCLUSIONS Using bioinformatics, we identified many groups of uncharac- terized enzymes that belong to the DUF692 enzyme family. By spectrometric and combining biochemical assays, mass spectroscopic experiments, we demonstrate that one group of such enzymes catalyzes a peptide backbone rearrangement to form an imidazolidinedione heterocycle adding to the current repertoire of heterocycles formed in RiPPs.13 In addition to heterocycle formation, ChrHI installs a thioether macrocycle from two conserved Cys residues and methylates a thiohemiaminal using SAM. Future spectroscopic studies and possibly structural information on the observed side products the may provide further insights into the mechanism of remarkable overall process. With three different DUF692 reactions now characterized, some commonalities are starting to emerge. All three proteins (MbnB, TglH, and ChrH) catalyze four-electron oxidations of their peptide substrates, apparently without requiring reducing equivalents. All three proteins also act on Cys residues, although whether this will hold for the entire family remains to be established. All three enzymes catalyze rearrangements that result in attachment of the sulfur atom of Cys and an amide nitrogen to the same carbon, the former β-carbon of Cys in the case of MbnBC and the former α-carbon of Cys for TglHI and ChrHI. For MbnB and ChrH, the net transformation involves removal of four hydrogens, likely with the formation of two water molecules (oxidase), whereas for TglH oxygenation chemistry is presumably involved in generating formate; whether formate generation involves monooxygenase or dioxygenase chemistry is at present not known. All three proteins seem to bind at least two iron ions and have the ligand set for binding three irons. Presently only for MbnB has activity been correlated with the amount of Fe(II) and Fe(III) in the protein,23 but it is likely that all three enzymes will be utilizing a mixed-valent state for catalysis. While the reactions catalyzed by the majority of DUF692 enzymes remain to be determined (Figure 1), the characterized transformations are all oxidations of the substrates using O2 catalyzed by multinuclear iron enzymes. Because these enzymes differ in iron-dependent enzymes,42 we fold from other families of suggest to replace the designation domain-of-unknown function 692 with multinuclear non-heme iron dependent oxidative enzymes (MNIOs). The final structure of chryseobasin remains to be determined and will require reconstitution of the remaining two enzymes encoded in the BGC, a protease ChrP and a putative epimerase ChrE, which has not yet been achieved to that of orthologs, would date. Their characterization, or facilitate investigation into the biological function of chryseobasin. Collectively, the data in this study expand the scope of post-translational modifications in RiPP biosynthesis, demonstrate another unexpected and complex reaction catalyzed by a homologue of MbnB and TglH, and lay the foundation toward understanding the chemistry of additional members of the enzyme family formerly known as DUF692. ■ ASSOCIATED CONTENT Data Availability Statement The authors declare that the data supporting the findings of this study are available within the paper and its Supporting Information files, and at Mendeley Data, V1, doi: 10.17632/ dn37cj6z5m.1, as well as from the corresponding author upon reasonable request. sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acscentsci.3c00160. Supporting data (XLSX) Cytoscape file for Figure 1A (ZIP) Materials and Methods, S1−S17 and tables S1−S6 (PDF) Transparent Peer Review report available (PDF) including supporting figures ■ AUTHOR INFORMATION Corresponding Author Wilfred A. van der Donk − Department of Chemistry and Carl R. Woese Institute for Genomic Biology, University of Illinois at Urbana−Champaign, Urbana 61801 Illinois, United States; Howard Hughes Medical Institute at the University of Illinois at Urbana−Champaign, Urbana 61801 Illinois, United States; orcid.org/0000-0002-5467-7071; Email: [email protected] Authors Richard S. Ayikpoe − Department of Chemistry and Carl R. Woese Institute for Genomic Biology, University of Illinois at Urbana−Champaign, Urbana 61801 Illinois, United States; orcid.org/0000-0001-6698-4020 Lingyang Zhu − School of Chemical Sciences NMR Laboratory, University of Illinois at Urbana−Champaign, orcid.org/0000- Urbana 61801 Illinois, United States; 0002-6657-271X Jeff Y. Chen − Department of Chemistry and Carl R. Woese Institute for Genomic Biology, University of Illinois at Urbana−Champaign, Urbana 61801 Illinois, United States; orcid.org/0000-0002-8507-8215 Chi P. Ting − Department of Chemistry and Carl R. Woese Institute for Genomic Biology, University of Illinois at Urbana−Champaign, Urbana 61801 Illinois, United States; orcid.org/0000-0003-1386-1902 Complete contact information is available at: https://pubs.acs.org/10.1021/acscentsci.3c00160 Funding This work was supported by the National Institutes of Health (F32 GM140621 to RSA and R37 GM058822 to WAV). Notes The authors declare no competing financial interest. ■ ACKNOWLEDGMENTS This work is dedicated to the memory of Prof. Christopher T. Walsh whose pioneering studies on enzyme mechanisms and natural product biosynthesis have been an inspiration to the authors. We thank D. T. Nguyen for his assistance with HR- MS/MS experiments. We also thank Dr. P. Yau and Dr. J. Arrington in the Roy J. Carver Biotechnology Center at the University of Illinois at Urbana−Champaign for assistance with MS/MS data acquisition. 1016 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article ■ REFERENCES (1) Hetrick, K. J.; van der Donk, W. A. Ribosomally synthesized and post-translationally modified peptide natural product discovery in the genomic era. Curr. Opin. Chem. Biol. 2017, 38, 36−44. (2) Blin, K.; Shaw, S.; Steinke, K.; Villebro, R.; Ziemert, N.; Lee, S. Y.; Medema, M. H.; Weber, T. antiSMASH 5.0: updates to the secondary metabolite genome mining pipeline. Nucleic Acids Res. 2019, 47, W81−W7. (3) Tietz, J. I.; Schwalen, C. J.; Patel, P. S.; Maxson, T.; Blair, P. M.; Tai, H. C.; Zakai, U. I.; Mitchell, D. A. A new genome-mining tool redefines the lasso peptide biosynthetic landscape. Nat. Chem. Biol. 2017, 13, 470−8. (4) Ren, H.; Shi, C.; Zhao, H. Computational tools for discovering and engineering natural product biosynthetic pathways. iScience 2020, 23, 100795. (5) Kloosterman, A. M.; Medema, M. H.; van Wezel, G. P. Omics- based strategies to discover novel classes of RiPP natural products. Curr. Opin. Biotechnol. 2021, 69, 60−7. (6) Baltz, R. H. Genome mining for drug discovery: progress at the front end. J. Ind. Microbiol. Biotechnol. 2021, 48, kuab044. (7) Wang, G.; Li, X.; Wang, Z. APD3: the antimicrobial peptide database as a tool for research and education. Nucleic Acids Res. 2016, 44, D1087−93. (8) Dang, T.; Süssmuth, R. D. Bioactive peptide natural products as lead structures for medicinal use. Acc. Chem. Res. 2017, 50, 1566−76. (9) Sarkar, T.; Chetia, M.; Chatterjee, S. Antimicrobial peptides and proteins: from nature′s reservoir to the laboratory and beyond. Front. Chem. 2021, 9, 691532. (10) Cao, L.; Do, T.; Link, A. J. Mechanisms of action of ribosomally synthesized and posttranslationally modified peptides (RiPPs). J. Ind. Microbiol. Biotechnol. 2021, 48, kuab005. (11) Lahiri, D.; Nag, M.; Dutta, B.; Sarkar, T.; Pati, S.; Basu, D.; Abdul Kari, Z.; Wei, L. S.; Smaoui, S.; Wen Goh, K.; et al. Bacteriocin: A natural approach for food safety and food security. Front. Bioeng. Biotechnol. 2022, 10, 1005918. (12) Ongpipattanakul, C.; Desormeaux, E. K.; DiCaprio, A.; van der Donk, W. A.; Mitchell, D. A.; Nair, S. K. Mechanism of action of ribosomally synthesized and post-translationally modified peptides. Chem. Rev. 2022, 122, 14722−814. (13) Montalbán-López, M.; Scott, T. A.; Ramesh, S.; Rahman, I. R.; van Heel, A. J.; Viel, J. H.; Bandarian, V.; Dittmann, E.; Genilloud, O.; Goto, Y.; et al. New developments in RiPP discovery, enzymology and engineering. Nat. Prod. Rep. 2021, 38, 130−239. (14) Arnison, P. G.; Bibb, M. J.; Bierbaum, G.; Bowers, A. A.; Bugni, T. S.; Bulaj, G.; Camarero, J. A.; Campopiano, D. J.; Challis, G. L.; Clardy, J.; et al. Ribosomally synthesized and post-translationally modified peptide natural products: overview and recommendations for a universal nomenclature. Nat. Prod. Rep. 2013, 30, 108−60. (15) Rosenthal, K.; Bornscheuer, U. T.; Lütz, S. Cascades of evolved enzymes for the synthesis of complex molecules. Angew. Chem., Int. Ed. 2022, 61, e202208358. (16) Kaspar, F.; Schallmey, A. Chemo-enzymatic synthesis of natural products and their analogs. Curr. Opin. Biotechnol. 2022, 77, 102759. (17) Heath, R. S.; Ruscoe, R. E.; Turner, N. J. The beauty of biocatalysis: sustainable synthesis of ingredients in cosmetics. Nat. Prod. Rep. 2022, 39, 335−88. (18) Benítez-Mateos, A. I.; Roura Padrosa, D.; Paradisi, F. Multistep enzyme cascades as a route towards green and sustainable pharmaceutical syntheses. Nat. Chem. 2022, 14, 489−99. (19) Oman, T. J.; van der Donk, W. A. Follow the leader: the use of leader peptides to guide natural product biosynthesis. Nat. Chem. Biol. 2010, 6, 9−18. (20) Nagano, N.; Orengo, C. A.; Thornton, J. M. One fold with many functions: the evolutionary relationships between TIM barrel families based on their sequences, structures and functions. J. Mol. Biol. 2002, 321, 741−65. (21) Kenney, G. E.; Dassama, L. M. K.; Pandelia, M. E.; Gizzi, A. S.; Martinie, R. J.; Gao, P.; DeHart, C. J.; Schachner, L. F.; Skinner, O. S.; Ro, S. Y.; et al. The biosynthesis of methanobactin. Science 2018, 359, 1411−6. (22) Dou, C.; Long, Z.; Li, S.; Zhou, D.; Jin, Y.; Zhang, L.; Zhang, X.; Zheng, Y.; Li, L.; Zhu, X.; et al. Crystal structure and catalytic mechanism of the MbnBC holoenzyme required for methanobactin biosynthesis. Cell Res. 2022, 32, 302−14. (23) Park, Y. J.; Jodts, R. J.; Slater, J. W.; Reyes, R. M.; Winton, V. J.; Montaser, R. A.; Thomas, P. M.; Dowdle, W. B.; Ruiz, A.; Kelleher, N. L.; et al. A mixed-valent Fe(II)Fe(III) species converts cysteine to an oxazolone/thioamide pair in methanobactin biosynthesis. Proc. Natl. Acad. Sci. U. S. A. 2022, 119, e2123566119. (24) Ting, C. P.; Funk, M. A.; Halaby, S. L.; Zhang, Z.; Gonen, T.; van der Donk, W. A. Use of a scaffold peptide in the biosynthesis of amino acid-derived natural products. Science 2019, 365, 280−4. (25) McLaughlin, M. I.; Yu, Y.; van der Donk, W. A. Substrate recognition by the peptidyl-(S)-2-mercaptoglycine synthase TglHI during 3-thiaglutamate biosynthesis. ACS Chem. Biol. 2022, 17, 930− 40. (26) Yu, Y.; van der Donk, W. A. Biosynthesis of 3-thia-α-amino acids on a carrier peptide. Proc. Natl. Acad. Sci. U. S. A. 2022, 119, e2205285119. (27) Gerlt, J. A.; Bouvier, J. T.; Davidson, D. B.; Imker, H. J.; Sadkhin, B.; Slater, D. R.; Whalen, K. L. Enzyme Function Initiative- Enzyme Similarity Tool (EFI-EST): A web tool for generating protein sequence similarity networks. Biochim. Biophys. Acta 2015, 1854, 1019−37. (28) Zallot, R.; Oberg, N.; Gerlt, J. A. The EFI web resource for genomic enzymology tools: Leveraging protein, genome, and metagenome databases to discover novel enzymes and metabolic pathways. Biochemistry 2019, 58, 4169−82. (29) Clark, K. A.; Seyedsayamdost, M. R. Bioinformatic atlas of radical SAM enzyme-modified RiPP natural products reveals an isoleucine-tryptophan crosslink. J. Am. Chem. Soc. 2022, 144, 17876− 88. (30) Burkhart, B. J.; Hudson, G. A.; Dunbar, K. L.; Mitchell, D. A. A prevalent peptide-binding domain guides ribosomal natural product biosynthesis. Nat. Chem. Biol. 2015, 11, 564−70. (31) Kloosterman, A. M.; Shelton, K. E.; van Wezel, G. P.; Medema, M. H.; Mitchell, D. A. RRE-finder: a genome-mining tool for class- independent RiPP discovery. mSystems 2020, 5, e00267-20. (32) Ehrlich, K. C.; Li, P.; Scharfenstein, L.; Chang, P. K. HypC, the anthrone oxidase involved in aflatoxin biosynthesis. Appl. Environ. Microbiol. 2010, 76, 3374−7. (33) Brademan, D. R.; Riley, N. M.; Kwiecien, N. W.; Coon, J. J. Interactive peptide spectral annotator: A versatile web-based tool for proteomic applications. Mol. Cell. Proteom. 2019, 18, S193−S201. (34) Bumbak, F.; Keen, A. C.; Gunn, N. J.; Gooley, P. R.; Bathgate, R. A. D.; Scott, D. J. Optimization and 13CH3 methionine labeling of a signaling competent neurotensin receptor 1 variant for NMR studies. Biochim. Biophys. Acta. Biomembr. 2018, 1860, 1372−83. (35) Hu, J.; Liu, F.; Feng, N.; Ju, H. Selenium-isotopic signature toward mass spectrometric identification and enzyme activity assay. Anal. Chim. Acta 2019, 1064, 1−10. (36) Walsh, C. Enzymatic reaction mechanisms; WH Freeman & Co: Oxford, 1979. (37) Hennessy, D. J.; Reid, G. R.; Smith, F. E.; Thompson, S. L. Ferene�a new spectrophotometric reagent for iron. Can. J. Chem. 1984, 62, 721−4. (38) Dippe, M.; Brandt, W.; Rost, H.; Porzel, A.; Schmidt, J.; Wessjohann, L. A. Rationally engineered variants of S-adenosylme- thionine (SAM) synthase: reduced product inhibition and synthesis of artificial cofactor homologues. Chem. Commun. 2015, 51, 3637−40. (39) McCarty, R. M.; Krebs, C.; Bandarian, V. Spectroscopic, steady- state kinetic, and mechanistic characterization of the radical SAM enzyme QueE, which catalyzes a complex cyclization reaction in the biosynthesis of 7-deazapurines. Biochemistry 2013, 52, 188−98. (40) Zhong, A.; Lee, Y.-H.; Liu, Y.-n.; Liu, H.-w. Biosynthesis of oxetanocin-A includes a B12-dependent radical SAM enzyme that can 1017 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018 ACS Central Science http://pubs.acs.org/journal/acscii Research Article catalyze both oxidative ring contraction and the demethylation of SAM. Biochemistry 2021, 60, 537−46. (41) Rajakovich, L. J.; Pandelia, M. E.; Mitchell, A. J.; Chang, W. C.; Zhang, B.; Boal, A. K.; Krebs, C.; Bollinger, J. M., Jr. A new microbial pathway for organophosphonate degradation catalyzed by two previously misannotated non-heme-iron oxygenases. Biochemistry 2019, 58, 1627−47. (42) Rajakovich, L. J.; Zhang, B.; McBride, M. J.; Boal, A. K.; Krebs, III: C.; Bollinger, J. M., Jr. In Comprehensive Natural Products Chemistry and Biology, Liu, H.-w.; Begley, T. P., Eds.; Elsevier: Amsterdam, 2020; pp 215−50. (43) Xing, G.; Barr, E. W.; Diao, Y.; Hoffart, L. M.; Prabhu, K. S.; Arner, R. J.; Reddy, C. C.; Krebs, C.; Bollinger, J. M., Jr. Oxygen activation by a mixed-valent, diiron(II/III) cluster in the glycol cleavage reaction catalyzed by myo-inositol oxygenase. Biochemistry 2006, 45, 5402−12. (44) van Staalduinen, L. M.; McSorley, F. R.; Schiessl, K.; Séguin, J.; Wyatt, P. B.; Hammerschmidt, F.; Zechel, D. L.; Jia, Z. Crystal structure of PhnZ in complex with substrate reveals a di-iron oxygenase mechanism for catabolism of organophosphonates. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 5171−6. (45) Wörsdörfer, B.; Lingaraju, M.; Yennawar, N. H.; Boal, A. K.; Krebs, C.; Bollinger, J. M., Jr; Pandelia, M. E. Organophosphonate- degrading PhnZ reveals an emerging family of HD domain mixed- valent diiron oxygenases. Proc. Natl. Acad. Sci. U. S. A. 2013, 110, 18874−9. (46) Brown, P. M.; Caradoc-Davies, T. T.; Dickson, J. M.; Cooper, G. J.; Loomes, K. M.; Baker, E. N. Crystal structure of a substrate complex of myo-inositol oxygenase, a di-iron oxygenase with a key role in inositol metabolism. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 15032−7. (47) Xing, G.; Diao, Y.; Hoffart, L. M.; Barr, E. W.; Prabhu, K. S.; Arner, R. J.; Reddy, C. C.; Krebs, C.; Bollinger, J. M., Jr. Evidence for C-H cleavage by an iron-superoxide complex in the glycol cleavage reaction catalyzed by myo-inositol oxygenase. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 6130−5. (48) Lundberg, M.; Siegbahn, P. E.; Morokuma, K. The mechanism for isopenicillin N synthase from density-functional modeling highlights the similarities with other enzymes in the 2-His-1- carboxylate family. Biochemistry 2008, 47, 1031−42. (49) Burzlaff, N. I.; Rutledge, P. J.; Clifton, I. J.; Hensgens, C. M.; Pickford, M.; Adlington, R. M.; Roach, P. L.; Baldwin, J. E. The reaction cycle of isopenicillin N synthase observed by X-ray diffraction. Nature 1999, 401, 721−4. (50) Tamanaha, E.; Zhang, B.; Guo, Y.; Chang, W. C.; Barr, E. W.; Xing, G.; St Clair, J.; Ye, S.; Neese, F.; Bollinger, J. M., Jr.; et al. Spectroscopic evidence for the two C-H-cleaving intermediates of Aspergillus nidulans isopenicillin N synthase. J. Am. Chem. Soc. 2016, 138, 8862−8874. (51) Headlam, H. A.; Davies, M. J. Beta-scission of side-chain alkoxyl radicals on peptides and proteins results in the loss of side- chains as aldehydes and ketones. Free Radic. Biol. Med. 2002, 32, 1171−84. (52) Peck, S. C.; Wang, C.; Dassama, L. M.; Zhang, B.; Guo, Y.; Rajakovich, L. J.; Bollinger, J. M., Jr; Krebs, C.; van der Donk, W. A. O-H activation by an unexpected ferryl intermediate during catalysis by 2-hydroxyethylphosphonate dioxygenase. J. Am. Chem. Soc. 2017, 139, 2045−52. (53) Born, D. A.; Ulrich, E. C.; Ju, K. S.; Peck, S. C.; van der Donk, W. A.; Drennan, C. L. Structural basis for methylphosphonate biosynthesis. Science 2017, 358, 1336−9. (54) Evans, R.; O’Neill, M.; Pritzel, A.; Antropova, N.; Senior, A.; Green, T.; Žídek, A.; Bates, R.; Blackwell, S.; Yim, J.et al. Protein complex prediction with AlphaFold-Multimer. bioRxiv; 2021.10.04.463034; 2021. (55) Hallgren, J.; Tsirigos, K. D.; Pedersen, M. D.; Almagro Armenteros, J. J.; Marcatili, P.; Nielsen, H.; Krogh, A.; Winther, O. DeepTMHMM predicts alpha and beta transmembrane proteins using deep neural networks. bioRxiv; 2022.04.08.487609; 2022. 1018 https://doi.org/10.1021/acscentsci.3c00160 ACS Cent. Sci. 2023, 9, 1008−1018
10.1021_acschembio.2c00840	pubs.acs.org/acschemicalbiology Articles Trapoxin A Analogue as a Selective Nanomolar Inhibitor of HDAC11 Thanh Tu Ho, Changmin Peng, Edward Seto, and Hening Lin* Cite This: ACS Chem. Biol. 2023, 18, 803−809 Read Online ACCESS Metrics & More Article Recommendations sı Supporting Information ABSTRACT: Histone deacetylases (HDACs) are enzymes that regulate many important biological pathways. There is a need for the development of isoform- selective HDAC inhibitors for further biological applications. Here, we report the development of trapoxin A analogues as potent and selective inhibitors of HDAC11, an enzyme that can efficiently remove long-chain fatty acyl groups from proteins. In particular, we show that one of the trapoxin A analogues, TD034, has nanomolar potency in enzymatic assays. We show that in cells, TD034 is active at low micromolar concentrations and inhibits the defatty acylation of SHMT2, a known HDAC11 substrate. The high potency and selectivity of TD034 would permit further development of HDAC11 inhibitors for biological and therapeutic applications. ■ INTRODUCTION Histone deacetylases (HDACs) were originally described as a class of enzymes that can remove the acetyl group from protein lysine residues.1 In humans, there are 11 HDACs that use a Zn2+-dependent mechanism for substrate deacetylation. HDACs can regulate chromatin structure and transcription through the deacetylation of histones but are also involved in other cellular processes through the regulation of nonhistone substrates.2,3 Histone deacetylase 11 (HDAC11) is the smallest and the last discovered HDAC, and a sole member of class IV HDAC.4 Its biological function is not yet fully elucidated. We and others showed that HDAC11 has a high defatty-acylase activity, while its deacetylase activity is essentially undetectable.5−7 We also found that serine hydroxymethyltransferase 2 (SHMT2) is a physiological substrate of HDAC11.5 The defatty acylation of SHMT2 by HDAC11 leads to increased type I interferon signaling in both cells and mouse models,5 which suggests that the inhibition of HDAC11 has the potential to treat diseases by modulating immune response. There have been other reports suggesting that the inhibition of HDAC11 could be beneficial for treating cancers,8,9 obesity,10 and multiple sclerosis.11 Therefore, there is a need for highly potent and specific HDAC11 inhibitorsto further study its biological function and explore the therapeutic potential of inhibiting HDAC11. The earliest known selective HDAC11 inhibitor is FT895 (Figure 1), which was developed by Forma Therapeutics.12 Based on its efficient catalytic activity in removing long-chain fatty acyl groups, we surmised that HDAC11 contains a hydrophobic pocket close to its Zn2+ catalytic center. Thus, our laboratory developed another HDAC11 inhibitor, SIS17 (Figure 1), which can fit this hydrophobic pocket.13 Both inhibition of FT895 and SIS17 display low micromolar HDAC11 demyristoylation activity in vitro.13 Surprisingly, SAHA (Figure 1), an FDA-approved HDAC inhibitor, cannot efficiently inhibit HDAC11’s demyristoylation activity.13 Meanwhile, trapoxin A (Figure 1), a class I HDAC inhibitor,14 can inhibit HDAC11 in the sub-micromolar range, although its nonselective HDAC inhibition activities limit its usefulness for studying HDAC11. We hypothesized that the modification of trapoxin A to exploit hydrophobic acyl pocket can yield potent, specific inhibitors for HDAC11. ■ RESULTS AND DISCUSSION Design and Synthesis of Trapoxin A Analogues. Schreiber15 and Kazmaier16 developed the only syntheses for trapoxin A and analogues. Despite its promising activity, very few synthetic derivatives of trapoxin A have been reported due to difficulties in modifying its structure. For our synthesis, we started by preparing various epoxyketone motifs containing long hydrocarbon chains at the β-position (Scheme 1) and α- position (Scheme S2, Supporting Information). Oxazolidine sulfur ylide 317 was prepared from (S)-phenylglycinol, and subsequent reactions with aliphatic aldehydes of various lengths afforded glycidyl amides 4 with (S)-configuration. reduction with Red-Al provided unstable epoxy Careful aldehyde 5, which was reacted with vinylmagnesium bromide at −40 °C, followed by oxidation with Dess-Martin period- Received: November 7, 2022 Accepted: February 17, 2023 Published: March 28, 2023 © 2023 The Authors. Published by American Chemical Society 803 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 1. Structure of known inhibitors. Scheme 1. Synthesis of the Epoxyketone Motif Scheme 2. Synthesis of the Cyclic Peptides 804 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 2. Trapoxin A analogues synthesized as HDAC11 inhibitors. Figure 3. TD034 is a potent, selective, reversible HDAC11 inhibitor in vitro. Each measurement was performed in triplicate. (A) TD034 (IC50 = 5.1 ± 1.1 nM) is much more potent than trapoxin A (IC50 = 94.4 ± 22.4 nM). (B) TD034 inhibition of HDAC11 activity is reversible after 40× dilution and ultrafiltration. (C) TD034 is a competitive inhibitor. (D) Morrison curve for the TD034 inhibition of HDAC11: [E]act and Ki were simultaneously determined by two-step nonlinear regression. (E) In enzymatic assays, TD034 does not significantly inhibit other HDACs/SIRTs. inane to afford vinyl ketones 6. The highest overall yield of 6 from 3 is obtained for 6b (24%); other chain lengths lead to poor yield and lengthy purification (6% for 6a and 1.6% for 6c). For the cyclic peptide backbone, we synthesized the unnatural amino acid (Uaa, 9), whose terminal alkene provided the anchor for subsequent olefin metathesis (Scheme 2). Alkylation of (S)-BPB-Ni-Gly complex 7 with 5-bromo-1- pentene under strictly air-free conditions afforded 8 stereo- selectively.18 Acidic methanolysis of 8, ion-exchange purifica- tion followed by Boc protection provided Boc-Uaa-OH 9. Solid-phase peptide synthesis afforded linear tetrapeptide 10, which was cyclized under high dilution conditions using AOP as the coupling reagent to yield cyclic peptide 11. Olefin metathesis of 11 with 6 using Hoveyda Grubbs second- generation catalyst, followed by Pd/C-catalyzed hydrogenation afforded the final inhibitors 12. 805 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Figure 4. TD034 inhibits HDAC11 in HEK293T cells and leads to an elevated fatty acylation level of SHMT2. (A) Representative Western blot images showing the cellular SHMT2 acylation levels with different concentrations of TD034, FT895, and SIS17. (B) Quantification of SHMT2 fatty acylation levels. P < 0.05, P < 0.01, P < 0.001. Figure 5. TD034 selectively inhibits HDAC11 and leads to YAP1 protein level decrease in A549 cells. (A) TD034, but not the less active TD034- (R), decreased the YAP1 protein level. (B) TD034, but not TD034-(R), decreased the mRNA level of YAP1 target genes. YAP1 mRNA level was not affected by TD034. (C) TD034 treatment decreased the YAP1 protein level in WT but not in HDAC11 KO cells. (D) TD034 does not inhibit class I HDACs, HDAC6, SIRT1/2 in cells, as measured by α-tubulin, p53, and histone H3 acetylation levels. P < 0.05, P < 0.01, P < 0.001. In Vitro Testing of the Synthesized Trapoxin A Analogues. The inhibitors synthesized are shown in Figure 2. Alkyl substitution at the α-position of epoxyketone (TD036) abolished HDAC11 inhibition, while substitution at the β- position with a C11-chain led to TD034, which has a nearly 20-fold increase in HDAC11 inhibition potency (IC50 = 5.1 ± 1.1 nM) compared to trapoxin A (IC50 = 94.4 ± 22.4 nM) (Figure 3A). Changing the stereochemistry of the epoxide yielded the diastereomer TD034-(R) with diminished potency, indicating that the orientation of the epoxide is crucial for guiding the hydrophobic chain into the pocket. Varying the aliphatic chain length afforded inhibitors TD037 (C-7 chain) and TD038 (C- 15 chain), which were less potent than TD034, perhaps due to a mismatch in chain length versus hydrophobic pocket depth. We also attempted to replace the Phe residues, but this led to a <5% yield of cyclic peptides due to the unfavorable entropy of head-to-tail tetrapeptide cyclization. These cyclizations were known to be very sensitive to residue interactions and stereochemistry.19 Thus, we decided to use TD034 for further investigation. We next investigated the mode of inhibition of TD034. Previous studies indicated that trapoxin A is either a covalent14 or tight-binding reversible inhibitor of HDACs.20 First, we checked whether HDAC11 inhibition by TD034 is reversible. We incubated HDAC11 (15 nM) with either DMSO (as control) or TD034 (15 nM) for 5 or 15 min. Afterward, the samples were either used directly for activity assay, or diluted 40x with buffer, ultrafiltered with Amicon 30K to remove excess inhibitor, and then subjected to activity assay. The absolute activity of HDAC11 decreased 4-fold after ultra- filtration due to the instability of HDAC11 after prolonged dilution. Regardless, we found that the HDAC11 activity was recovered after dilution and ultrafiltration, and prolonged incubation time with TD034 did not affect the recovered activity of HDAC11 (Figure 3B), confirming that inhibition by 806 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles TD034 is reversible. We then measured IC50 at different ratios of [S]t/Km (Figure 3C). We found that the IC50 displayed a linear correlation with [S]t/Km, consistent with a competitive mechanism.21 Thus, we concluded that TD034 is a high- affinity, reversible, noncovalent inhibitor. Finally, we fitted dose−response data using a two-step nonlinear regression of the Morrison equation22,23 to simultaneously determine the active enzyme concentration ([E]act = 4.7 ± 1.9 nM) and inhibition constant (Ki = 1.5 ± 0.3 nM) for TD034 (Figure 3D). We screened TD034 against several other human HDACs these Interestingly, TD034 did not and sirtuins. HDACs or sirtuins, although it showed some potency against SIRT2 (IC50 ∼ 25 μM) (Figure 3E). SIRT2 is known to have efficient demyristoylase activity and possesses a large hydro- phobic pocket, which could explain the inhibitory activity.24 Nonetheless, TD034 still exhibited >5000× selectivity for HDAC11 versus SIRT2. inhibit TD034 Inhibits HDAC11 Selectively in Cells. With this encouraging data, we then tested TD034 in HEK293T to check whether it could inhibit HDAC11 selectively in cells. SHMT2 was reported as a defatty-acylation substrate of HDAC11.5 Thus, we tested whether TD034 could inhibit HDAC11 and increase the fatty acylation level of endogenous SHMT2. We treated HEK293T cells with an alkyne-tagged myristic acid analog, Alk14, along with the inhibitors (TD034, SIS17, or FT895) for 3 h. Click chemistry was performed on cell lysate with Biotin-azide, followed by Streptavidin pull- down. The amount of labeled SHMT2 was then detected by Western blot (Figure 4A). TD034 significantly increased the fatty acylation level of SHMT2 at 2 μM (Figure 4B). The same effect was observed at 20 μM for SIS17, while FT895 at 4 μM had no statistically significant effect, consistent with a previous report.13 We noted that higher concentrations of TD034 are needed to inhibit HDAC11 in cells than in the in vitro enzymatic assay, likely due to unfavorable membrane partitioning of the alkyl chain. HDAC11 expression is upregulated in lung cancer and is associated with poor prognosis in lung cancer patients. Consistent with the previous finding that depletion of HDAC11 downregulated YAP1 (yes-associated protein 1) protein expression in lung cancer cells,8 A549 cells treated with TD034 resulted in a significant reduction of YAP1 protein levels (Figure 5A) and a decrease in the mRNA levels of two YAP1 target genes, CTGF and CYR61 (Figure 5B). Using TD034-(R), a much less potent analogue, we did not observe such an effect. To confirm whether the downregulation of YAP1 was due to the inhibition of HDAC11 by TD034, we tested TD034 on both wild-type (WT) and HDAC11 knockout (KO) A549 cells. First, we found that without the TD034 treatment, the endogenous protein level of YAP1 in HDAC11 KO cells was lower than that in WT cells. Second, treatment with TD034 led to a reduced YAP1 protein level in WT cells but not in HDAC11 KO cells (Figure 5C). These results confirmed that TD034 decreases the YAP1 level via HDAC11 inhibition and extended the potential of using TD034 to manipulate the HDAC11-mediated hippo-YAP signaling pathway. To demonstrate that TD034 is selective toward HDAC11 in cells, we measured the acetylation levels of α-tubulin, p53, and histone H3 by Western blot. As a positive control, we used trichostatin A (TSA), a nonselective HDAC inhibitor against both class I and class IIb HDACs.25 After 3 h of treatment, the TSA-treated cell had an elevated level of acetylated α-tubulin (HDAC6 and SIRT2 target) and increased levels of acetylated histone H3 and acetylated p53 (class I HDACs and SIRT1 target).26 Meanwhile, the TD034-treated cells did not have such an effect (Figure 5D). Thus, TD034 is selective for HDAC11 and does not the concentration tested, consistent with the in vitro activity assay results. inhibit other HDACs at ■ CONCLUSIONS low nanomolar concentrations In summary, by modifying trapoxin A, we have developed TD034, a highly potent HDAC11 selective inhibitor. TD034 inhibits HDAC11 at in enzymatic assays in vitro and low micromolar concentrations in cells, making it more potent than previously discovered inhibitors. Furthermore, TD034 selectively inhibits HDAC11 in cells. TD034 is a great HDAC11 inhibitor candidate for further optimization and biological applications. ■ METHODS Chemical Syntheses. Detailed synthesis procedures and characterization of compounds 2−12 are provided in the Supporting Information. Data Processing. All quantified measurements were performed in triplicates. Data processing was performed using Graphpad Prism 9.5.0. Repeated measures of one-way ANOVA with Fisher’s LSD test were used to determine the P value. The dose−response data was fitted using a two-step nonlinear regression of the Morrison equation. In the first step, an estimated [E]act = 7.6 nM (determined by linear extrapolation of Zone A)22 was held constant for regression to yield an estimated Ki = 0.85 nM. In the second step, both [E]act and Ki were treated as variables, using previous estimates as initial values. The best-fit curve (R2 = 0.99) yields [E]act = 4.7 ± 1.9 nM and Ki = 1.5 ± 0.3 nM for TD034. HDAC Enzyme Activity Assays. HDACs and SIRTs were expressed and purified as previously described.13 The HDAC11 concentration used in the experiments was estimated to be 74 nM by SDS-PAGE gel; active HDAC11 concentration was estimated by Morrison curve fitting to be 4.7 ± 1.9 nM. Not all HDAC11 enzyme was active due to post-translation modifications, as well as denaturation during purification, storage, and handling. For the HDAC11 activity assay, Myr-H3K9 peptide (25 μM), HDAC11 (4.7 nM), and inhibitors at various concentrations were incubated in 20 μL of assay buffer (50 mM Tris/Cl, pH 8.0, 137 mM NaCl, 2.7 mM KCl, 1 mM MgCl2) at 37 °C. For HDAC4, trifluoroacetyl-H3K9 was used as a substrate. For HDAC1, 6, 8, and SIRT1-2, Ac-H3K9 was used as a substrate. For SIRT1/2, the assay buffer includes 1 mM DTT and 1 mM NAD+.13 TD034 is relatively stable in Tris buffer and DTT (Figure S2, Supporting Information). The reaction was conducted for 15 min (HDAC11), 30 min (HDAC1, 4, 6, 8), and 5 min (SIRT1-2). Then, 20 μL of 0.2% TFA/acetonitrile was added to quench the reaction. The samples were analyzed by HPLC using a Chromolith HighResolution RP-18 end-capped 100 mm × 4.6 mm column (EMD Millipore). Mobile phase A was 0.1% TFA in water and mobile phase B was 0.1% TFA in acetonitrile. The total flow rate was 1 mL/min, and the gradient was 0% B (2 min), 0−60% B (7 min), 100% B (4 min), and 0% B (2 min). The relative ratio of product/substrate in each sample was compared to control sample (no inhibitor) to determine the inhibition level. HDAC11 Enzyme Kinetic Assays. For preincubation assay, TD034 (15 nM) or DMSO (control) was incubated with HDAC11 (15 nM) for 5 or 15 min in 10 μL of assay buffer (50 mM Tris−Cl, pH 8.0, 137 mM NaCl, 2.7 mM KCl, 1 mM MgCl2) at 37 °C. Afterward, either (i) 10 μL of the Myr-H3K9 peptide (50 μM) was added, or (ii) the solution was diluted with 390 μL of assay buffer, concentrated by Amicon 30K filters until ∼20 μL remained. Then, the Myr-H3K9 peptide (25 μM) was added. The samples were incubated 807 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles at 37 °C for 15 min. Each reaction was quenched with 20 μL of 0.2% TFA/acetonitrile, and the samples were analyzed as described above. For IC50 versus [S]t/Km assay, the Myr-H3K9 peptide (200, 100, 50, 25 μM), HDAC11 (4.7 nM), and inhibitors at various concentrations were incubated in 20 μL of assay buffer (50 mM Tris/Cl, pH 8.0, 137 mM NaCl, 2.7 mM KCl, 1 mM MgCl2) at 37 °C for 15 min. Each reaction was quenched with 20 μL of 0.2% TFA/acetonitrile, and the samples were analyzed as described above. HDAC11 in Cell Assay: Defatty Acylation of SHMT2. HEK293T in a six-well plate at 80% confluency was treated with 50 μM Alk14 and inhibitors at various concentrations. The cells were incubated for 3 h. The cells were harvested and lysed in 200 μL of 4% SDS lysis buffer (50 mM triethanolamine, 150 mM NaCl, 4% SDS, pH 7.4) with a 1:100 protease inhibitor cocktail and 1:1000 nuclease for 15 min. The cell lysates were then diluted with 3.8 mL of HEPES buffer (50 mM HEPES, 150 mM NaCl, 1% NP-40, pH 7.4), and then concentrated using Amicon Ultra-4 (30 kDa cutoff) for 45 min at 4000g. The retained samples were then diluted to 0.5 mL with HEPES buffer, followed by the addition of Biotin-N3 (5 μL, 5 mM in DMF), TBTA (5 μL, 2 mM in DMF), CuSO4 (5 μL, 50 mM in water), and TCEP (5 μL, 50 mM in water). The samples were shaken at 37 °C for 1 h, then diluted with 3 mL of HEPES buffer, and concentrated again using Amicon Ultra-4 (30 kDa cutoff) for 45 min at 4000g. The retained samples were then diluted to 0.5 mL with followed by the addition of 20 μg of magnetic HEPES buffer, streptavidin beads (prewashed with the HEPES buffer). The mixture was shaken for 1 h and the supernatant was removed. Hydroxylamine in the HEPES buffer (100 μL, 0.5 M) was added, and the mixture was then shaken for 30 min. The supernatant was removed, and the beads were washed with HEPES buffer (2 × 500 μL). The remaining beads were incubated at 95 °C with 40 μL of 4% SDS lysis buffer and 8 μL of 6× loading buffer for 10 min. The eluants were further analyzed by SDS-PAGE and Western blot for SHMT2. YAP1 Protein Level and Target Genes mRNA Level Determination. A549 cells were treated with TD034 at 5 and 10 μM for 24 hr. Western blot was used to check for YAP1 protein level. qRT-PCR was used to check for YAP1 downstream genes transcription (CTGF and CYR61). Total RNA was extracted using IBI Isolate Total Extraction Reagent Kit (IB47602). Two milligrams of RNA was reverse transcribed using OneScript Plus cDNA Synthesis Kit (ABM G236) according to the manufacturer’s protocol. Real-time PCR was performed using BlasTaq 2X qPCR MasterMix (ABM G892) on a QuantStudio 3 Real-Time PCR System. All qPCR reactions were performed in triplicates. The list of primers is included in the Supporting Information. ■ ASSOCIATED CONTENT sı Supporting Information The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.2c00840. Reagents, instruments, primers for qRT-PCR, activity assay data for TD037, TD038, and TD034(R), synthetic methods, and NMR spectra of important compounds (PDF) ■ AUTHOR INFORMATION Corresponding Author Hening Lin − Department of Chemistry and Chemical Biology, Cornell University, Ithaca, New York 14853, United States; Howard Hughes Medical Institute, Cornell University, orcid.org/0000- Ithaca, New York 14853, United States; 0002-0255-2701; Email: [email protected] Authors Thanh Tu Ho − Department of Chemistry and Chemical Biology, Cornell University, Ithaca, New York 14853, United States; orcid.org/0000-0002-1582-8404 Changmin Peng − Department of Biochemistry & Molecular Medicine, School of Medicine & Health Sciences, George Washington Cancer Center, George Washington University, Washington, District of Columbia 20037, United States Edward Seto − Department of Biochemistry & Molecular Medicine, School of Medicine & Health Sciences, George Washington Cancer Center, George Washington University, Washington, District of Columbia 20037, United States Complete contact information is available at: https://pubs.acs.org/10.1021/acschembio.2c00840 Notes The authors declare the following competing financial interest(s): HL is a founder and consultant for Sedec Therapeutics. ■ ACKNOWLEDGMENTS The work is supported by grants from NIH-NCI and NIH- NIAID: R01CA240529 and R01AI153110. ■ REFERENCES (1) Yang, X.-J.; Seto, E. The Rpd3/Hda1 Family of Lysine Deacetylases: From Bacteria and Yeast to Mice and Men. Nat. Rev. Mol. Cell Biol. 2008, 9, 206−218. (2) Seto, E.; Yoshida, M. Erasers of Histone Acetylation: The Histone Deacetylase Enzymes. Cold Spring Harbor Perspect. Biol. 2014, 6, No. a018713. (3) Li, Y.; Seto, E. HDACs and HDAC Inhibitors in Cancer Development and Therapy. Cold Spring Harbor Perspect. Med. 2016, 6, No. a026831. (4) Gao, L.; Cueto, M. A.; Asselbergs, F.; Atadja, P. Cloning and Functional Characterization of HDAC11, a Novel Member of the Human Histone Deacetylase Family. J. Biol. Chem. 2002, 277, 25748− 25755. (5) Cao, J.; Sun, L.; Aramsangtienchai, P.; Spiegelman, N. A.; Zhang, X.; Huang, W.; Seto, E.; Lin, H. HDAC11 Regulates Type I Interferon Signaling through Defatty-Acylation of SHMT2. Proc. Natl. Acad. Sci. U.S.A. 2019, 116, 5487−5492. J.; (6) Kutil, Z.; Novakova, Z.; Meleshin, M.; Mikesova, Schutkowski, M.; Barinka, C. Histone Deacetylase 11 Is a Fatty- Acid Deacylase. ACS Chem. Biol. 2018, 13, 685−693. (7) Moreno-Yruela, C.; Galleano, I.; Madsen, A. S.; Olsen, C. A. Histone Deacetylase 11 Is an ε-N-Myristoyllysine Hydrolase. Cell Chem. Biol. 2018, 25, 849−856.e8. (8) Bora-Singhal, N.; Mohankumar, D.; Saha, B.; Colin, C. M.; Lee, J. Y.; Martin, M. W.; Zheng, X.; Coppola, D.; Chellappan, S. Novel HDAC11 Inhibitors Suppress Lung Adenocarcinoma Stem Cell Self- Renewal and Overcome Drug Resistance by Suppressing Sox2. Sci. Rep. 2020, 10, No. 4722. (9) Mostofa, A. G. M.; Distler, A.; Meads, M. B.; Sahakian, E.; Powers, J. J.; Achille, A.; Noyes, D.; Wright, G.; Fang, B.; Izumi, V.; Koomen, J.; Rampakrishnan, R.; Nguyen, T. P.; Avila, G. D.; Silva, A. S.; Sudalagunta, P.; Canevarolo, R. R.; Silva, M. D. C. S.; Alugubelli, R. R.; Dai, H. A.; Kulkarni, A.; Dalton, W. S.; Hampton, O. A.; Welsh, E. A.; Teer, J. K.; Tungesvik, A.; Wright, K. L.; Pinilla-Ibarz, J.; Sotomayor, E. M.; Shain, K. H.; Brayer, J. Plasma Cell Dependence on Histone/Protein Deacetylase 11 Reveals a Therapeutic Target in Multiple Myeloma. JCI Insight 2021, 6, No. e151713. (10) Bagchi, R. A.; Ferguson, B. S.; Stratton, M. S.; Hu, T.; Cavasin, M. A.; Sun, L.; Lin, Y.-H.; Liu, D.; Londono, P.; Song, K.; Pino, M. F.; Sparks, L. M.; Smith, S. R.; Scherer, P. E.; Collins, S.; Seto, E.; McKinsey, T. A. HDAC11 Suppresses the Thermogenic Program of Adipose Tissue via BRD2. JCI Insight 2018, 3, No. e120159. (11) Sun, L.; Telles, E.; Karl, M.; Cheng, F.; Luetteke, N.; Sotomayor, E. M.; Miller, R. H.; Seto, E. Loss of HDAC11 808 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809 ACS Chemical Biology pubs.acs.org/acschemicalbiology Articles Ameliorates Clinical Symptoms in a Multiple Sclerosis Mouse Model. Life Sci. Alliance 2018, 1, No. e201800039. (12) Martin, M. W.; Lee, J. Y.; Lancia, D. R.; Ng, P. Y.; Han, B.; Thomason, J. R.; Lynes, M. S.; Marshall, C. G.; Conti, C.; Collis, A.; Morales, M. A.; Doshi, K.; Rudnitskaya, A.; Yao, L.; Zheng, X. Discovery of Novel N-Hydroxy-2-Arylisoindoline-4-Carboxamides as Potent and Selective Inhibitors of HDAC11. Bioorg. Med. Chem. Lett. 2018, 28, 2143−2147. (13) Son, S. I.; Cao, J.; Zhu, C.-L.; Miller, S. P.; Lin, H. Activity- Guided Design of HDAC11-Specific Inhibitors. ACS Chem. Biol. 2019, 14, 1393−1397. (14) Kijima, M.; Yoshida, M.; Sugita, K.; Horinouchi, S.; Beppu, T. Trapoxin, an Antitumor Cyclic Tetrapeptide, Is an Irreversible Inhibitor of Mammalian Histone Deacetylase. J. Biol. Chem. 1993, 268, 22429−22435. (15) Taunton, J.; Collins, J. L.; Schreiber, S. L. Synthesis of Natural and Modified Trapoxins, Useful Reagents for Exploring Histone Deacetylase Function. J. Am. Chem. Soc. 1996, 118, 10412−10422. (16) Servatius, P.; Kazmaier, U. Total Synthesis of Trapoxin A, a Fungal HDAC Inhibitor from Helicoma Ambiens. J. Org. Chem. 2018, 83, 11341−11349. (17) Gordillo, P. G.; Aparicio, D. M.; Flores, M.; Mendoza, A.; Orea, L.; Juárez, J. R.; Huelgas, G.; Gnecco, D.; Terán, J. L. Oxazolidine Sulfur Ylides Derived from Phenylglycinol for the Specific and Highly Diastereoselective Synthesis of Aryl and Alkyl Trans-Epoxyamides. Eur. J. Org. Chem. 2013, 2013, 5561−5565. (18) Zou, Y.; Han, J.; Saghyan, A. S.; Mkrtchyan, A. F.; Konno, H.; Moriwaki, H.; Izawa, K.; Soloshonok, V. A. Asymmetric Synthesis of Tailor-Made Amino Acids Using Chiral Ni(II) Complexes of Schiff Bases. An Update of the Recent Literature. Molecules 2020, 25, 2739. (19) Sarojini, V.; Cameron, A. J.; Varnava, K. G.; Denny, W. A.; Sanjayan, G. Cyclic Tetrapeptides from Nature and Design: A Review of Synthetic Methodologies, Structure, and Function. Chem. Rev. 2019, 119, 10318−10359. (20) Porter, N. J.; Christianson, D. W. Binding of the Microbial Cyclic Tetrapeptide Trapoxin A to the Class I Histone Deacetylase HDAC8. ACS Chem. Biol. 2017, 12, 2281−2286. (21) Strelow, J.; Dewe, W.; Iversen, P. W.; Brooks, H. B.; Radding, J. A.; McGee, J.; Weidner, J. Mechanism of Action Assays for Enzymes. In Assay Guidance Manual; Eli Lilly & Company and the National Center for Advancing Translational Sciences: Bethesda (MD), 2004. (22) Copeland, R. A. Tight Binding Inhibition. In Evaluation of Enzyme Inhibitors in Drug Discovery; John Wiley & Sons, Ltd, 2013; pp 245−285. (23) Kuzmič, P.; Elrod, K. C.; Cregar, L. M.; Sideris, S.; Rai, R.; Janc, J. W. High-Throughput Screening of Enzyme Inhibitors: Simulta- neous Determination of Tight-Binding Inhibition Constants and Enzyme Concentration. Anal. Biochem. 2000, 286, 45−50. (24) Teng, Y.-B.; Jing, H.; Aramsangtienchai, P.; He, B.; Khan, S.; Hu, J.; Lin, H.; Hao, Q. Efficient Demyristoylase Activity of SIRT2 Revealed by Kinetic and Structural Studies. Sci. Rep. 2015, 5, No. 8529. (25) Bradner, J. E.; West, N.; Grachan, M. L.; Greenberg, E. F.; Haggarty, S. J.; Warnow, T.; Mazitschek, R. Chemical Phylogenetics of Histone Deacetylases. Nat. Chem. Biol. 2010, 6, 238−243. (26) Hong, J. Y.; Fernandez, I.; Anmangandla, A.; Lu, X.; Bai, J. J.; Lin, H. Pharmacological Advantage of SIRT2-Selective versus Pan- SIRT1−3 Inhibitors. ACS Chem. Biol. 2021, 16, 1266−1275. 809 https://doi.org/10.1021/acschembio.2c00840 ACS Chem. Biol. 2023, 18, 803−809
10.1093_gbe_evad100	GBE No Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of Cicadas Noah Spencer 1, Piotr Łukasik2,3, Mariah Meyer2,4, Claudio Veloso5, and John P. McCutcheon1,2,6,* 1Biodesign Center for Mechanisms of Evolution and School of Life Sciences, Arizona State University, Tempe, Arizona, USA 2Division of Biological Sciences, University of Montana, Missoula, Montana, USA 3Institute of Environmental Sciences, Jagiellonian University, Kraków, Poland 4Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus, Aurora, Colorado, USA 5Department of Ecological Sciences, Science Faculty, University of Chile, Santiago, Chile 6Howard Hughes Medical Institute, Chevy Chase, Maryland, USA *Corresponding author: E-mail: [email protected]. Accepted: 26 May 2023 Abstract Bacteria that form long-term intracellular associations with host cells lose many genes, a process that often results in tiny, gene-dense, and stable genomes. Paradoxically, the some of the same evolutionary processes that drive genome reduction and simplification may also cause genome expansion and complexification. A bacterial endosymbiont of cicadas, Hodgkinia cicadicola, exemplifies this paradox. In many cicada species, a single Hodgkinia lineage with a tiny, gene-dense genome has split into several interdependent cell and genome lineages. Each new Hodgkinia lineage encodes a unique subset of the an- cestral unsplit genome in a complementary way, such that the collective gene contents of all lineages match the total found in the ancestral single genome. This splitting creates genetically distinct Hodgkinia cells that must function together to carry out basic cellular processes. It also creates a gene dosage problem where some genes are encoded by only a small fraction of cells while others are much more abundant. Here, by sequencing DNA and RNA of Hodgkinia from different cicada species with different amounts of splitting—along with its structurally stable, unsplit partner endosymbiont Sulcia muelleri—we show that Hodgkinia does not transcriptionally compensate to rescue the wildly unbalanced gene and genome ratios that result from lineage splitting. We also find that Hodgkinia has a reduced capacity for basic transcriptional control independent of the split- ting process. Our findings reveal another layer of degeneration further pushing the limits of canonical molecular and cell biol- ogy in Hodgkinia and may partially explain its propensity to go extinct through symbiont replacement. Key words: endosymbionts, transcriptomics, gene dosage, cicadas, genome evolution, nonadaptive evolution. Significance Many cicadas host two bacterial endosymbionts, Hodgkinia and Sulcia, which produce essential amino acids missing from the insect’s xylem sap diet. Following 100+ million years of strict host-association, both bacteria have lost many genes and possess extremely tiny genomes. In some cicadas, Hodgkinia has split into multiple cell lineages, distributing its genes, with little respect to their function, among separate lineages present at (sometimes wildly) different abundances. We find no transcriptional response to genome fragmentation in Hodgkinia: mRNA abundance reflects gene abundance. We also find less overall control of transcription in Hodgkinia compared to Sulcia. Hodgkinia’s transcriptome seems to reflect a bacter- ium on the edge of existence, and raises questions about how multilineage Hodgkinia remain functional. © The Author(s) 2023. Published by Oxford University Press on behalf of Society for Molecular Biology and Evolution. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 1 Spencer et al. GBE Introduction Vertically transmitted bacterial endosymbionts that form very stable and long-term association with host cells, in- cluding the ancestors of mitochondria and plastids, can lose most of the genes originally encoded by their free- living ancestors (Andersson and Kurland 1998; Green 2011; Gray 2012). Endosymbiont genomes are often small in size, stable in structure, and densely packed with a core set of functional genes (Boore 1999; Tamas et al. 2002; McCutcheon and Moran 2011; Graf et al. 2021). While such tiny, stable, and gene-dense endosymbiont genomes have evolved again and again in diverse host lineages, some endosymbiont and organelle genomes have second- arily become unstable, expanding in size through the accu- mulation or proliferation of non-coding and nonfunctional DNA. The cicada endosymbiont Candidatus Hodgkinia cica- dicola (hereafter, Hodgkinia) and the mitochondria of some sucking lice and flowering plants have all evolved multi- chromosomal genomes several times larger than those of closely related lineages despite virtually no change to their overall gene repertoire (Shao et al. 2012; Sloan et al. 2012; Campbell et al. 2015; Campbell et al. 2017). In the case of Hodgkinia—and in contrast to mitochondria, where differ- ent chromosomes are mixed together throughout the mito- chondrial compartments of a cell—genome fragmentation occurs in parallel with cellular diversification such that the total gene set is divided among distinct Hodgkinia cell po- pulations which are present at different relative abun- dances in the host (Van Leuven et al. 2014; Łukasik et al. 2018). As a result, genes critical both to Hodgkinia’s symbi- otic role in nutrient biosynthesis along with genes central to basic bacterial cell function can differ in abundance by or- ders of magnitude within the same insect. This gene dosage problem raises the question of whether complex Hodgkinia can correct for large differences in gene abundance in some way, for example through transcriptional up-regulation of lowly abundant genes (Campbell et al. 2015; Łukasik et al. 2018). The Hodgkinia genome has the expected single circular- mapping chromosome in many cicadas structure (McCutcheon et al. 2009b; Van Leuven et al. 2014; Łukasik et al. 2018). In some cicadas, however, Hodgkinia has independently undergone varying degrees of genome fragmentation via cell lineage splitting (Łukasik et al. 2018; Campbell et al. 2017). Compared to the unsplit an- cestral genome, individual split genomic lineages lack func- tional copies of many essential genes, but these losses occur in a complementary fashion such that the unsplit gene set is maintained at the level of the total Hodgkinia population in each cicada (Van Leuven et al. 2014; Łukasik et al. 2018). The complementary genome erosion of each lineage enforces transmission of all Hodgkinia gen- omes to the subsequent host generation, resulting in an relative expansion of the total Hodgkinia genome from the perspective of the host (Campbell et al. 2015). In extreme cases, this splitting process results in genome complexes consisting of at least a dozen lineages and totaling over 1.5 Mb in length, a more than tenfold increase in genome size to single-lineage Hodgkinia genomes (Campbell et al. 2017). Importantly, comparisons between these largest Hodgkinia complexes show extreme variation in splitting outcomes with respect to the size and gene con- tent of their constituent genomes, which suggests that splitting does not converge on a particular endpoint or op- timum (Campbell et al. 2017). While splitting results in an expansion of the total unique Hodgkinia genome found in each cicada, each individual genome lineage experiences only gene loss and genome re- duction. Lineage splitting can therefore only decrease the overall abundance of functional Hodgkinia genes in the sys- tem, dependent on how many and which genome(s) a gi- ven gene resides. Importantly, gene products of even the mostly lowly abundant Hodgkinia genes must be shared among all lineages in a given host to preserve their collect- ive function. While we have shown by in situ hybridization that Hodgkinia genomes and ribosomal RNAs are con- tained by their respective cell boundaries, the biochemistry of these cells must somehow behave as though these boundaries do not exist or are easily crossed (Campbell et al. 2015; Łukasik et al. 2018). Genomics shows that many Hodgkinia genes within the same biochemical path- way have differential gene dosages that would result in 10- or 100-fold disruptions in pathway stoichiometry if left uncorrected. These differences are well in excess of those associated with dosage sensitivity and haploinsuffi- ciency in eukaryotes and could introduce choke points in the enzyme kinetics of essential processes like nutrient bio- synthesis (Papp et al. 2003; Morril and Amon 2019). These dosage disruptions also far exceed the modest several-fold capacity for gene-specific transcriptional tuning exhibited by some insect endosymbionts in response to changes in their hosts’ nutrition or developmental stage (Moran et al. 2005a, Stoll et al. 2009; Wilcox et al. 2003). Nevertheless, endosymbionts such as Hodgkinia remain able to some- what regulate the expression of RNAs and proteins at a rela- tive level within a genome, because transcripts such as those from rRNA, tRNA, RNase PRNA, and protein chaper- ones are more abundant than most other transcripts (Wilcox et al. 2003; Van Leuven et al. 2019; Husnik et al. 2020). It is therefore possible that some baseline level of constitutive gene expression control remains in Hodgkinia. Hodgkinia’s predisposition to splitting may owe in part to its high rate of sequence evolution, a feature also ob- served in the huge, fragmented mitochondrial genomes of the angiosperms Silene conica and Silene noctiflora (Sloan et al. 2012; Van Leuven et al. 2014). This is con- trasted by the roughly 50–100 times lower nucleotide 2 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 No Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE substitution rate exhibited by Hodgkinia’s partner endo- symbiont, Candidatus Sulcia muelleri (hereafter, Sulcia) (Van Leuven et al. 2014). While Hodgkinia genomes are structurally unstable and vary widely in size, Sulcia tends to be much more stable. Following at least 250 million years of strict host-association, Sulcia genomes from distantly related hosts show almost perfect gene co-linearity and very similar gene sets (Moran et al. 2005b; Bennett and Moran 2015) [although a broader sampling of Auchenorrhynchan insects shows that several genomic in- versions have occurred in different Sulcia lineages (Deng et al. 2022)]. Likewise, while several cicada groups have replaced Hodgkinia with fungal endosymbionts, Sulcia is re- tained in every cicada species examined to date (Matsuura et al. 2018; Wang et al. 2022b). Given the diversity of Hodgkinia genome size and organ- ization and the relative structural stasis of Sulcia genomes in cicadas, this system constitutes an elegant natural experi- ment for evaluating the downstream transcriptional conse- quences of wild swings in gene dosage resulting from endosymbiont genome instability. To characterize the tran- scriptional activity of Hodgkinia and Sulcia genomes relative to their genomic abundance, we sequenced DNA and RNA from the symbiotic organs of 18 cicadas representing six species encompassing a spectrum of Hodgkinia complexity. We find that Hodgkinia exerts limited transcriptional con- trol compared to Sulcia and is unable to transcriptionally compensate for the massive effect of gene dosage imbal- ance that is produced by lineage splitting. Results In the absence of lineage splitting, we assume each Hodgkinia cell contributes equally to the total abundance of each Hodgkinia transcript (fig. 1A). Following splitting and differential gene loss, some transcripts can only be pro- duced by a (sometimes very small) subset of Hodgkinia cells. We evaluated four different hypotheses, two adaptive and two nonadaptive, for how Hodgkinia may or may not com- pensate at the transcriptional level for the gene dosage im- balances that result from splitting and differential gene loss: an adaptive response of nonspecific, constitutive transcriptional up-regulation to bring transcripts of low- abundance genes to some threshold level (“overcompensa- tion,” fig. 1B); a specific adaptive response where lowly abundant genes are upregulated to rescue presplitting transcript abundances (“complementation,” fig. 1C); a nonresponse, where each gene is transcribed at its presplit- ting levels in each cell irrespective of its relative abundance (“subdivision,” fig. 1C); and a response of further regula- tory decay, where noncompensatory changes to transcrip- tion are introduced as a side-effect of splitting and genome erosion (“disruption,” fig. 1D). To look for signatures of these outcomes across the spectrum of Hodgkinia complexity, we collected three individuals from single populations representing each of six different cicada species (table 1) and sequenced the metagenomes and their dissected bacteriomes metatranscriptomes of (endosymbiont-housing organs). The multilineage Hodgkinia studied here all originated from independent splitting events (Campbell et al. 2017; Łukasik et al. 2018). Closed genomes are available for all relevant Hodgkinia lineages except in Magicicada septen- decim, in which the Hodgkinia genome has been as- sembled into 39 circular molecules and 124 additional contigs. Similarly, closed genomes of all relevant Sulcia lineages are available, with the exception of Sulcia from M. septendecim. In this case, we used the genome of Sulcia from Magicicada tredecim, which is completely co- linear with and over 99% identical to its counterpart in M. septendecim. We obtained between 18.7 and 44.4 million paired-end reads from the bacteriome metagenome libraries and between 43.6 and 181.7 million reads from the corresponding metatranscriptome libraries. Each of from these Hodgkinia, Sulcia, and the cicada host. sequences derived contains libraries Cicada Endosymbionts Retain Different Degrees of Transcriptional Control to We began by examining transcription in the cicada endo- symbionts in general. Relatively little work has characterized transcription in endosymbionts with extremely reduced genomes (Bennett and Chong 2017; Van Leuven et al. 2019; Wang et al. 2022a), but a comparative analysis of RNA polymerase genes suggests that some endosymbionts, including Hodgkinia, may have a limited capacity for pro- moter recognition (Rangel-Chávez et al. 2021). To compare the degree to which Hodgkinia and Sulcia can specifically transcribe coding DNA and can transcribe genes at different levels in line with biological expectations, we aligned stranded bacteriome mRNA-seq reads from each cicada species (fig. 2A–B; the corresponding Sulcia supplementary fig. S1, Supplementary Material online) and Hodgkinia (fig. 2C–D; supplementary figs. S2–S5, Supplementary Material online) reference genomes and vi- sualized per-base coverage across each chromosome. We obtained >1X coverage of the vast majority of genomic re- gions, except for some of the small, unplaced Hodgkinia contigs in M. septendecim. In both endosymbionts and across host species, patterns of coverage were highly con- sistent among biological replicates (fig. 2; supplementary figs. S1–S5, Supplementary Material online). In the case of Sulcia, we saw clear similarities between species in the rela- tive (fig. 2A–B; supplementary fig. S1, Supplementary Material online), while Hodgkinia was much more variable (fig. 2C–D; supplementary figs. S2–S5, Supplementary Material online). transcription of different genes Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 3 Spencer et al. GBE FIG. 1.—Schematic of a Hodgkinia cell lineage split and its possible transcriptional outcomes. (A) In the absence of cell lineage splitting, Hodgkinia cells all contain the same genes (here x, y, and z) and contribute to their respective transcript abundances. Lineage splitting and complementary gene loss decrease the relative dosage (total supply) of genes that are lost in some cells. In this abstracted example, genes x, y, and z have relative postsplitting dosages of 1.0 (full dosage), 0.1, and 0.9, respectively. (B) If Hodgkinia cells increase transcription genome-wide, the transcript abundances will remain imbalanced but could reach some required threshold level for dosage-depleted genes. (C) If Hodgkinia transcription is regulated to transcribe dosage-depleted genes at higher levels, presplitting transcript abundances could be rescued. (D) If Hodgkinia cells do not change transcription in response to changes in gene dosage (i.e., each gene is transcribed at roughly its original level in each cell), transcript abundance of genes with reduced dosage will decrease. (E) If the processes of cell lineage splitting and/or gene loss intrinsically affect the transcription of certain genes, transcript abundances could change in unpredictable ways. Table 1 Hodgkinia Genome Complexity and Abundance Ratios in all Cicada Species Sampled Cicada Species Number of Hodgkinia Lineages Diceroprocta near semicincta Tettigades ulnaria Tettigades undata Okanagana oregona Tettigades limbata Magicicada septendecim 1 1 2 4 5 12+ Approximate Genome Abundance Distribution 100 100 60:40 45:35:18:2 75:10:8:5:2 9:5:4:3:1:1:1 … Compared with Hodgkinia, Sulcia exhibited patterns of transcription that indicate a greater ability to terminate transcription. Sulcia exhibited clearly distinguishable peaks of high RNA coverage overlapping with annotated genes (inset of fig. 2A). In Hodgkinia, RNA coverage was often (but less consistently) high along annotated genes. However, rather than producing symmetrical peaks of coverage centered on annotated genes, transcription in Hodgkinia frequently continued past the ends of genes, gradually decreasing until another peak of RNA coverage began. For example, the RNase P RNA gene is transcribed at high levels in Hodgkinia from both Diceroprocta near semicincta and Tettigades ulnaria (fig. 2C and D), but the corresponding peak in RNA-seq coverage continues for 4 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 ABCDENo Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE FIG. 2.—Strand-specific, per-base RNA-seq coverage along the chromosomes of Sulcia (A–B) and Hodgkinia (C–D) from D. near semicincta and T. ulnaria. Rectangles in the central track of each plot represent annotated genes and are colored according to functional categories. Positive and negative Y axes cor- respond to coverage of unfiltered RNA-seq reads derived from the plus and minus strands of each chromosome, respectively. For each plot, coverage profiles from each biological replicate (translucent gray) are overlaid along the same axes. Panels A and B represent alignments downsampled to approximately 3500X mean coverage of the Sulcia genome and are cropped at approximately y = ±20,000. Panels C and D represent alignments downsampled to approximately 450X mean coverage of the Hodgkinia genome and are cropped at approximately y = ±2000. Antisense counts as a percentage of sense + antisense counts are shown for each genome. Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 5 ABCDSpencer et al. GBE several times the length of the functional RNA and into an intergenic region (inset of fig. 2C). Unlike in protein-coding genes, several of which show similar “run-on” transcrip- tional profiles in Hodgkinia, the transcriptional read- through at this locus cannot be explained or resolved by termination at the level of translation. The Hodgkinia and Sulcia lineages sampled often ap- peared to highly transcribe chaperone genes (groS, groL, dnaJ, and dnaK; fig. 2, gold stars). This is consistent with published endosymbiont transcriptomes (Stoll et al. 2009; Luck et al. 2015; Medina Munoz et al. 2017) and with prote- omic data showing that chaperones are among the most abundant proteins in endosymbionts (Charles et al. 1997; McCutcheon et al. 2009a, 2009b; Poliakov et al. 2011). To evaluate transcriptional control more quantitatively, we calculated levels of antisense transcription in Hodgkinia and Sulcia from each cicada species. To do this, we first used the transcript quantification tool FADU (Feature Aggregate Depth Utility) to obtain transcript counts for functional genes in Sulcia and Hodgkinia (excluding tRNA and rRNA genes) (Chung et al. 2021). We then repeated this step for antisense transcription by deliberately specifying the opposite strand orientation for our libraries such that FADU output counted alignments to the strand opposite each open reading frame (Srinivasan et al. 2020). Hodgkinia had a higher proportion of antisense counts than its coresident Sulcia in every bio- logical replicate from each cicada species (supplementary table S2, Supplementary Material online, supplementary figs. S1–S5, Supplementary Material online). Hodgkinia from D. near semicincta, which has experienced no genome fragmentation, stood out in this regard with between 44% and 49% antisense transcripts compared to just 13–16% in its coresident Sulcia (fig. 2A and C). Taken together, these data show an overall loss of tran- scriptional control in Hodgkinia compared to Sulcia (Supplementary figs. S1–S5, Supplementary Material on- line), and provide the first indirect hint that dosage com- pensation at the level of Hodgkinia transcription is unlikely to be occurring in these symbioses. Complex Hodgkinia Produce RNA in Proportion to Their Cell Abundance Under our first hypothesized adaptive scenario, complex Hodgkinia could rescue the transcript abundances of low- copy genes through a general increase in mRNA synthesis (overcompensation, fig. 1B), potentially resembling the high transcript abundances observed in plant organelles (Forsythe et al. 2022) or in the reduced nucleomorph genomes of certain green algae (Tanifuji et al. 2014). We determined the relative contributions of Hodgkinia and Sulcia-derived DNA and mRNA to the sequencing libraries from each specimen by filtering out any remaining riboso- mal RNA sequences, mapping each set of filtered reads to the corresponding endosymbiont genomes, and calculat- ing the coverage of each genome as a proportion of all fil- tered reads (fig. 3). We have already shown that Hodgkinia genome coverage is a good proxy for cell abundance (Van Leuven et al. 2014; Campbell et al. 2018; Łukasik et al. 2018). The DNA abundance of Sulcia was consistently high- er than that of Hodgkinia except in the M. septendecim samples, which also had the highest overall Hodgkinia DNA abundance. Compared to the DNA libraries, the RNA libraries gener- ally contained more endosymbiont-derived reads. In an overcompensation scenario, complex Hodgkinia would be expected to produce a greater ratio of RNA:DNA coverage than their coresident Sulcia. While Hodgkinia from Tettigades undata showed patterns of coverage potentially consistent with overcompensation in all three biological re- plicates (e.g., specimen A showed 3% Hodgkinia coverage in DNA reads but 23% coverage in RNA reads), we did not observe this pattern in any of the other multilineage Hodgkinia examined. In the samples representing the most extreme level of splitting, Hodgkinia from M. septen- decim, where overcompensation might be expected to be the most obvious, RNA coverage was actually underrepre- sented relative to its DNA abundance (e.g., specimen C showed 21% Hodgkinia coverage in DNA reads but only 10% coverage in RNA reads). This decrease in relative RNA abundance was not an artifact of rRNA depletion being less effective in certain RNA-seq libraries, as the trends we observed in total RNA coverage fractions hold even when rRNA is not removed bioinformatically (supplementary fig. S6, Supplementary Material online). One sample from T. ulnaria contained very little Hodgkinia material and was excluded from any other Hodgkinia-based analysis (marked with an asterisk in fig. 3). While Hodgkinia and Sulcia are spatially separated within the bacteriome, it is unlikely that Hodgkinia was excluded due to a dissection er- ror (Van Leuven et al. 2014; Campbell et al. 2015; Łukasik et al. 2018). The lack of Hodgkinia material might instead suggest that this sample originated from a slightly older, sen- escent individual (Kono et al. 2008; Vigneron et al. 2014; Simonet et al. 2018). A sample from M. septendecim pro- duced very little endosymbiont-derived RNA coverage in general, with Hodgkinia and Sulcia collectively contributing less than 0.1%, and was excluded from all other analysis (marked with two asterisks in fig. 3). Because cicadas live underground for most of their lives, only emerge once a year (at most), and are difficult to catch, we were unable to add new samples to replace these lost data points. Gene Dosage Depletion Reshapes the Hodgkinia Transcriptome Under a complementation scenario (fig. 1C), the distribu- tion of transcript abundances in the total Hodgkinia 6 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 No Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE FIG. 3.—Proportional contributions of Hodgkinia and Sulcia to the total (A) DNA and (B) RNA sequencing coverage, shown for triplicate biological repli- cates of each cicada species examined. Reads mapping to the Hodgkinia and Sulcia rRNA genes for each cicada were removed before the calculation of RNA coverage. T. ulnaria specimen C (one asterisk) lacked Hodgkinia DNA and RNA and was excluded from subsequent Hodgkinia-based analyses. M. septendecim specimen B (two asterisks) lacked RNA coverage from either endosymbiont and was excluded from all further analyses. population (i.e., from the perspective of the insect host) would be similar between single-lineage and complex Hodgkinia. This could occur if, for example, Hodgkinia transcriptional machinery had evolved a greater affinity for sequences associated with lowly abundant genes or chromosomes (Veita et al. 2013). Conversely, in the ab- sence of complementation, genes that are present in fewer copies in the Hodgkinia population following splitting would be represented by fewer transcripts than more abundant genes. To distinguish these two outcomes, we compared relative gene dosage with total transcript abun- dance in each Hodgkinia system. In each biological repli- cate, we defined the dosage or abundance of a gene as the percentage of Hodgkinia DNA sequencing coverage contributed by Hodgkinia contigs that contain the gene (Supplementary table S1, Supplementary Material online). Similarly, we measured the total transcript abundance of a gene as the summed transcripts per million (TPM) of each distinct copy of the gene in a Hodgkinia complex (Li and Dewey 2011; Wagner et al. 2012). The TPM distributions from single-lineage Hodgkinia sys- tems (D. near semicincta and T. ulnaria, fig. 4A and B) showed relatively consistent shapes among biological repli- cates but differed slightly in the spread between the two host species, comparable to the between host variation in Sulcia TPM distributions (fig. 4G–L). The TPM distributions from T. undata, hosting two Hodgkinia lineages, were simi- lar in shape but showed a clear biforcation based on gene dosage with genes at full dosage corresponding to the upper half of the distributions and genes at 40–60% rela- tive dosage corresponding to the lower half (fig. 4C). Compared to these three species, the TPM distributions from the more fragmented Hodgkinia of Okanagana ore- gona and Tettigades limbata showed relatively more genes at their extreme low ends (fig. 4D–E), and most of these genes had relative dosages of less than 20% in these cica- das. This was not the case in the highly fragmented Hodgkinia of M. septendecim in which all genes are far from maximal dosage. Instead, its TPM distributions showed uniquely high dispersion with relatively few values Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 7 ABSpencer et al. GBE FIG. 4.—Endosymbiont gene dosage and gene expression from the host perspective in symbiotic systems of varying complexity. (A–F) Distribution of log10 TPM expression levels of Hodgkinia genes from each cicada specimen examined. Points represent the summed TPM of all copies of a gene. Points are colored according to their relative dosage. To make differences in gene dosage visible in M. septendecim, gene dosages for all specimens were converted to standard deviations above or below the specimen’s average Hodgkinia gene dosage (Z transformation). (G–L) Distribution of log10 TPM expression levels of Sulcia genes from each cicada specimen examined. (M–P) Total log10 TPM expression levels (y-axis) of Hodgkinia genes of different relative dosage (x-axis, log10 scale) in Hodgkinia systems of varying complexity. TPM and relative dosage values are averages of values from each biological replicate. For each comparison, Kendall’s rank correlation coefficient τ as well as the test statistic (Z) and P-value for hypothesis tests of rank correlation are given, showing a significant positive cor- relation between relative Hodgkinia gene dosage and transcript abundance in all four species. 8 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 ABCDEFGMNOPHIJKLNo Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE concentrated around the average (fig. 4F). Such differences in the shape of the TPM distribution were specific to Hodgkinia: they were not observed in their coresident Sulcia lineages (fig. 4G–L). In all cicada species with split Hodgkinia, we found a sig- nificant positive correlation between relative gene dosage and total TPM at a significance threshold of α = 0.05 (Kendall’s rank correlation, P < 0.0001 for all four compar- isons, fig. 4M–P). In other words, Hodgkinia genes that had greater dosage at the genomic level in each cicada tended to be represented by a greater number of transcripts. The strength of the relationship, given by Kendall’s τ, was simi- lar across host species, ranging from 0.44 in M. septende- cim to 0.58 in O. oregona. This correspondence between total gene and transcript abundances in multilineage Hodgkinia is inconsistent with complementation. However, complementary changes in transcription could still exist, even if they do not overcome the influence of gene dosage altogether. We tested for cell- level complementary responses to differences in total gene rank correlations. supply using semipartial Kendall Semipartial correlations allowed us to characterize the rela- tionship between the TPM abundance of all distinct Hodgkinia gene copies and their relative dosage from the host perspective while controlling for the effect of each gene copy’s DNA abundance on its measured transcript abundance. Per-cell transcription was expected to be nega- tively correlated with relative gene dosage under a comple- mentation scenario (fig. 1C) and not correlated under a subdivision scenario (fig. 1D). We found no significant cor- relation between cell abundance-controlled TPM and rela- tive gene dosage in O. oregona (τ= 0.018, Z = 0.44, P = 0.66), in T. limbata (τ = 0.053, Z = 1.372, P = 0.17), or in M. septendecim (τ = 0.054, Z = 1.268, P = 0.205) at a sig- nificance threshold of α = 0.05. In T. undata, abundance- controlled TPM and relative gene dosage were significantly positively correlated, indicating that genes encoded in only one of T. undata’s two Hodgkinia cell lineages actually tended to have a lower per-cell expression (τ = 0.209, Z = 5.071, P < 0.0001). Hodgkinia Transcription Profiles Are Not Conserved Across Host Species Having found no evidence for transcriptional compensation for the gene dosage outcomes resulting from Hodgkinia lineage splitting and reciprocal gene loss, we next asked whether gene expression patterns in the nonfragmented ancestral Hodgkinia transcriptome are conserved between species. We compared the log10 TPM expression of hom- ologous, protein-coding Hodgkinia and Sulcia genes be- tween D. near semicincta and T. ulnaria, which both host a single Hodgkinia lineage (fig. 5A and B). Expression of Sulcia genes showed a strong linear correlation between the two host species (Pearson correlation: r = 0.932, t = 36.275, ν = 198, P < 0.0001) while Hodgkinia gene expression was only weakly correlated (r = 0.186, t = 2.087, ν = 121, P = 0.039). In addition to the strong bio- logical contrast between these outcomes, the consistency of Sulcia transcriptional profiles across relatively distantly related host species gives us confidence that our RNA-seq data are of good overall quality and that the relatively noisy nature of the Hodgkinia data is not the result of technical artifacts. Given the incongruence between transcript abundances in phylogenetically distant single-lineage Hodgkinia, we di- rected our focus to the genus Tettigades, from which we had sampled three different species, reasoning that Hodgkinia transcript abundances in T. ulnaria (which hosts a single Hodgkinia lineage) may approximate a presplitting “starting point” for this group and that some semblance of transcriptional control may persist in phylogenetically re- lated lineages. Total TPM transcript abundances in multili- neage Hodgkinia from Tettigades cicadas appear to deviate from this hypothetical starting point, even in the case of T. undata, which hosts only two different Hodgkinia lineages (fig. 5C). Discussion Hodgkinia Does Not Compensate for Transcriptional Consequences of Gene Dosage Imbalance The process of splitting into multiple interdependent cell lineages combined with complementary gene loss has re- sulted in varied and sometimes extreme gene dosage out- comes for the Hodgkinia populations contained in each cicada (Campbell et al. 2017; Łukasik et al. 2018). We con- sidered two possible outcomes that would reflect compen- sation for this change at the level of transcription: widespread overproduction of mRNA to guarantee suffi- cient transcript abundance (overcompensation, fig. 1B) and fine-tuned compensatory regulation to rescue the tran- scription levels of dosage-depleted genes (complementa- tion, fig. 1C). Our analysis of genome relative abundance and transcription in Hodgkinia of multiple complexity levels shows that neither of these adaptive responses occurs. Rather, in Hodgkinia composed of 2, 4, 5, and 12+ cell lineages con- sistently, strongly, and simply reflect the gene dosage of corresponding genes on their genomes. transcriptional changes that occur the Some form of compensation could, in principle, occur at the level of translation. This would presumably rely on fac- tors external to Hodgkinia, particularly since the least abun- dant Hodgkinia cell lineages in the species examined here tend to encode relatively limited complements of translation-related genes compared to more abundant lineages (Campbell et al. 2017; Łukasik et al. 2018). Our Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 9 Spencer et al. GBE FIG. 5.—Differences in homologous gene expression between endosymbionts of different host species. Scatter plots show the correlation of relative ex- pression levels (log10 TPM) for homologous protein-coding genes in two symbionts, (A) single-lineage Hodgkinia and (B) Sulcia, between two distantly related cicadas: D. near semicincta and T. ulnaria. The Pearson correlation coefficient r and the P-value for a test of correlation are given. (C) Expression levels of hom- ologous Hodgkinia genes in T. ulnaria, T. undata, and T. limbata are shown colored according to their TPM expression percentile in T. ulnaria, highlighting variation in Hodgkinia transcript abundances within the genus Tettigades. All TPM values are averages across biological replicates. TPM values in T. undata and T. limbata represent summed values from all copies of a given gene. observations could also be affected by the age of the cica- das sampled, which, as fully grown adults nearing the ends of their lives, may no longer be as reliant on the proper functioning of their nutritional endosymbionts in one or both sexes. However, given the lack of conservation in Hodgkinia transcript abundance across cicada species, and the relative conservation we see in Sulcia transcription, we favor the idea that Hodgkinia simply tolerates the tran- scriptional consequences of gene dosage changes, even quite extreme ones. This is not to say that such changes are always selectively neutral in Hodgkinia. Given that our results are most consistent with a lack of transcriptional response in Hodgkinia after splitting, our finding that genes that had been lost in one of T. undata’s two Hodgkinia lineages had lower per-cell transcription could suggest that lineage- specific gene losses are more likely to be fixed when they occur in lowly-expressed genes. Additionally, while the dos- age outcomes in highly complex Hodgkinia are not deter- ministic, some mechanism seems to favor the retention of 10 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 ACBNo Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE certain Hodgkinia genes at a greater total abundance, and this is reflected in their transcript abundance (Campbell et al. 2017; Łukasik et al. 2018). Basic Transcriptional Control Shows Signs of Erosion in Hodgkinia The transcriptional machinery encoded by the bacterial en- dosymbionts with the tiniest genomes is extremely rudi- mentary (McCutcheon and Moran 2011). Despite this, we found evidence for at least some degree of transcriptional control in two such endosymbionts. All Sulcia and Hodgkinia transcriptomes produced more alignments to genes in the sense orientation than in the antisense orien- tation, suggesting that open reading frames are preferably transcribed over random positions on the opposite strand of DNA. We also observed consistently high chaperone gene expression in both endosymbionts, a recurring feature of endosymbiont transcriptomes thought to be of function- al importance to the endosymbiotic lifestyle (Fares et al. 2002; Stoll et al. 2009; McCutcheon and Moran 2011; Luck et al. 2015; Medina Munoz et al. 2017). This occurs despite the loss of the rpoH-encoded σ32 heat shock sigma factor, which modulates the expression of chaperone genes in free-living bacteria (Neidhardt and VanBogelen 1981; Yamamori and Yura 1982; Grossman et al. 1987). Across all six host species examined, Hodgkinia and Sulcia differed in two potential indicators of transcriptional control. First, we found that RNA-seq coverage declined predictably at gene ends in Sulcia while the high coverage typical of transcriptional start sites frequently extended past annotated genes in Hodgkinia, possibly indicating transcriptional read-through. The gene contents of these two endosymbionts point to a potential mechanistic ex- planation: Sulcia, unlike Hodgkinia, retains rho and its co- factor nusA, which have well-characterized roles in transcription termination (Schmidt and Chamberlin 1984; Richardson 2002; McCutcheon et al. 2009a, 2009b; Łukasik et al. 2018). Second, Hodgkinia endosymbionts showed consistently higher levels of antisense transcription than their coresident Sulcia, although this may simply be a reflection of increased transcriptional read-through at genes located adjacent to a gene on the opposite strand. Surprisingly, the single-lineage Hodgkinia endosymbiont of D. near semicincta stood out in its apparent loss of transcriptional control, exhibiting a considerably higher proportion of antisense transcription than any other endo- symbiont lineage we examined. The fact that antisense transcription in Sulcia from the D. near semicincta samples was not correspondingly high suggests that this effect is not a technical artifact. A previous comparative genomic analysis of endosymbiont RNA polymerases identified a de- letion of seven amino acid residues long in the σ3 subunit from Hodgkinia in a very closely related cicada species, D. semicincta, and predicted that this loss could impede recognition of an extended −10 box promoter element (Rangel-Chávez et al. 2021). Promoter elements have not been characterized in Hodgkinia or in other endosymbionts with tiny genomes, and we have similarly found no recognizable sequence motifs upstream of Hodgkinia or Sulcia start codons regardless of transcription level (supplementary figs. S7–S8, Supplementary Material on- line). However, we note that the rpoD gene in Hodgkinia from D. near semicincta, like D. semicincta, lacks this por- tion of the σ3 subunit found in most other Hodgkinia gen- omes (supplementary fig. S9, Supplementary Material online), although similar deletions in rpoD genes in Hodgkinia from M. septendecim are evidently not accom- panied by a correspondingly high level of antisense tran- scription (supplementary figs. S5 and S9, Supplementary Material online). We also found that Hodgkinia transcript abundances in D. near semicincta were weakly correlated with their homo- logs’ relative abundances in the single-lineage Hodgkinia of T. ulnaria in contrast to the strong correlation observed in the Sulcia transcriptomes of those cicadas. It is unclear to what extent host-specific losses in Hodgkinia transcription- al control may have contributed to this lack of conservation versus the 50+ million years of evolutionary divergence be- tween these Hodgkinia lineages (Marshall et al. 2018; Wang et al. 2022b). Gene Products May Be Spread Extremely Thin in Complex Hodgkinia On one hand, the unresponsiveness of Hodgkinia transcrip- tion to extreme gene dosage outcomes is unsurprising gi- ven that Hodgkinia encodes no transcription factors or alternative sigma factors and has even accumulated func- tionally important losses to basic transcriptional machinery (McCutcheon et al. 2009b; Galán-Vásquez et al. 2016; Łukasik et al. 2018; Rangel-Chávez et al. 2021). Even in un- split Hodgkinia lineages with uniform gene dosage, precise ratios of relative transcript abundance do not appear to be conserved. On the other hand, Hodgkinia’s unresponsive- ness is surprising because of what it implies about its biol- ogy, specifically its apparent tolerance for extreme unbalancing of essential transcripts’ absolute abundance. In M. septendecim, where many genes may be present in fewer than ten percent of cells, we found no evidence for a generalized up-regulation of Hodgkinia transcription. In fact, Hodgkinia’s relative contributions to DNA and RNA coverage in this system imply an overall reduced transcrip- tional activity. While in situ hybridization has shown that rRNA and genomic DNA are not shared among cells in complex Hodgkinia, the endosymbiont’s continued existence neces- sarily implies the movement of either mRNA, protein, Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 11 Spencer et al. GBE metabolites, or some combination of these between cells by an unknown mechanism (Campbell et al. 2015; Łukasik et al. 2018). The likelihood of a biologically import- ant encounter between two Hodgkinia proteins could therefore be limited not just by the abundance of the genes by which they are encoded but also by those genes’ spatial distribution within the cicada bacteriome. In the absence of massive complementation or overcompensation at the level of protein synthesis, it is conceivable that the biochemistry of the most complex Hodgkinia occurs slowly or inefficient- ly relative to their single-lineage counterparts. Conclusions The transcriptomes of cicadas’ bacterial endosymbionts, like their genomes, embody two opposite extremes. Sulcia exhibits highly conserved transcript abundance ratios and patterns of RNA-seq coverage that line up with bio- logical expectations. Hodgkinia, meanwhile, shows dimin- ished transcriptional control and transcribes genes in proportion to their sometimes wildly imbalanced DNA abundance. In either case, it is difficult to quantify the fit- ness consequences of these transcriptional outcomes. We expect that at least some of the outcomes we observe in Hodgkinia, such as widespread antisense transcription in D. near semicincta and failure to compensate for massive gene dilution in M. septendecim, are costly. The magni- tudes of these costs are dependent on translational com- pensatory changes—if any occur—and, in the latter case, gene product transport. Both of these processes have yet to be characterized in Hodgkinia. As with the reproductive burden cicadas experience in order to transmit a complete Hodgkinia gene complement to their eggs following exten- sive lineage splitting, we speculate that these events are costly for the symbiosis and may tip the scales in favor of Hodgkinia extinction and replacement with a new endo- symbiont (Campbell et al. 2018; Matsuura et al. 2018; Wang et al. 2022b). Materials and Methods Insect Collection Adult cicadas, a mixture of males and females, were col- lected in their natural habitat using insect nets and dis- sected in the field, with abdomens torn open and placed in 7 mL tubes with RNAlater. They were kept refrigerated initially, and, after arrival in the laboratory, stored at −8 °C until processing. We preliminarily identified specimens based on morpho- logical characters and later confirmed identifications using marker gene sequences. In the case of Diceroprocta, we collected multiple individuals that we could not distin- guish based on morphology, but that represented two gen- otypes divergent by about 3% within the mitochondrial cytochrome C oxidase I (COI) gene, one of which matched the previously characterized D. semicincta (Van Leuven and McCutcheon 2012). Since all individuals represented the other COI genotype, we decided to refer to them as D. near semicincta. An additional specimen of O. oregona collected previ- ously was used for the assembly of its respective Hodgkinia and Sulcia genomes. This specimen was col- lected and dissected in the same manner, placed in 90% EtOH, and stored at −2 °C until processing. Collection Details Diceroprocta near semicincta (two males + female) University of Arizona campus, Tucson, AZ, USA, 32.23, −110.95, July 2017. Tettigades ulnaria (three males) Side of the road near Putaendo, Valparaíso Region, Chile, −32.588, −70.715, January 2017. Tettigades undata (three males) Side of the road to Termas de Chillan, Bio Bio Region, Chile, −36.903, −71.537, January 6, 2017. Okanagana oregona (three males) Mt. Sentinel, Missoula, MT, USA, 46.86, −113.98, 30 Jun 2017. Okanagana oregona (one male specimen used for Hodgkinia and Sulcia genome assemblies) Mt. Sentinel, Missoula, MT, USA, 46.86, −113.98, June 13, 2016. Tettigades limbata (male + two females) Hills South of Sierra de Bellavista, O’Higgins Region, Chile, −34.826, −70.742, December 13, 2014. Magicicada septendecim (three females) Washington, PA, USA, 40.171, −80.221, 2017. DNA and RNA Extraction, Library Preparation, and Sequencing DNA was extracted from carefully dissected bacteriome tis- sue using the Qiagen DNeasy Blood and Tissue kit (Hilden, Germany) except in the case of the M. septendecim sam- ples. For these samples, DNA libraries were prepared from co-extracted DNA obtained during RNA isolation (see RNA work details below). Illumina libraries for all samples were prepared using the Illumina Truseq PCR-free kit (San Diego, CA, USA). RNA was also extracted from each bacteriome tissue sample using the Qiagen RNeasy Mini kit (Hilden, Germany) according to the included protocol for animal tis- sue and then DNase-treated using the Invitrogen TURBO DNA-free kit (Waltham, MA, USA). Ribosomal RNA was de- pleted using the Ribo-Zero Epidemiology Kit from Illumina (San Diego, CA, USA) followed by cleanup with the RNeasy MinElute Cleanup Kit (Hilden, Germany). The DNA and RNA libraries were sequenced in four batches across a total of six lanes on a HiSeq X instrument in 2 × 150 bp mode at Novogene (Sacramento, CA, USA). 12 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 No Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE One additional DNA library from an O. oregona specimen used for genome assembly of its endosymbionts was sequenced on a MiSeq v3 instrument in 2 × 300 bp mode. Endosymbiont Genome Assemblies and Annotation The genomes of Sulcia from T. ulnaria and T. limbata were assembled from previously published cicada bacteriome metagenomes deposited under BioProject accessions PRJNA246493 and PRJNA385844 (Van Leuven et al. 2014; Łukasik et al. 2018). These, as well as Sulcia and Hodgkinia genomes from D. near semicincta and O. orego- na were assembled as follows: reads were trimmed of low- quality ends and adapters using Trimmomatic or Trim Galore! (Bolger et al. 2014) and then merged using Pear (Zhang et al. 2014) or bbmerge (Bushnell et al. 2017). Bacteriome metagenomes were initially assembled using custom installations of SPAdes 3.7.1, 3.11.0, or 3.12.0 (Prjibelski et al. 2020) which were compiled with an in- creased k-mer length limit of 249 bp. Scaffolds from this as- sembly were used for blastx searches against a custom database comprising the six frame-translated genomes of several Hodgkinia lineages and protein-coding genes from Sulcia, 12 other insect-associated and free-living bacteria, cicada mitochondria, and the planthopper lugens. Quality-filtered reads were then Nilaparvata remapped to scaffolds with top matches (evalue < 1e–10) to Hodgkinia and Sulcia references, respectively, using ei- ther qualimap (García-Alcalde et al. 2012) or bbmap (Bushnell 2014), and the mapped reads were used for final SPAdes assemblies. For Hodgkinia from O. oregona, poly- merase chain reaction was used to close gaps and verify rRNA operon sequences. Hodgkinia and Sulcia genomes were annotated using a custom pipeline described previously (Łukasik et al. 2018), including curated sets of reference genes extracted from published genomes and tRNA annotation using tRNAscan-SE v.1.23 (Chan and Lowe 2019). DNA Coverage and Genome Abundance Analyses Raw reads from the bacteriome metagenome sequencing libraries were inspected for quality using FastQC version 0.11.7 (https://www.bioinformatics.babraham.ac.uk/projects/ fastqc/) and trimmed using Trim Galore! version 0.6.1 (https://www.bioinformatics.babraham.ac.uk/projects/trim_ galore/) to remove Illumina adapters and low-quality bases (Phred scores <10) from read ends, retaining read pairs in which each read has a post-trimming length of at least 20 bp (Martin 2011). For comparative analysis of Hodgkinia and Sulcia DNA coverage, BowTie 2 indexes were built from the Hodgkinia and Sulcia genomes of each cicada species using BowTie 2 version 2.4.1 (Langmead and Salzberg 2012). Trimmed DNA reads from each sample was aligned to the corresponding indexes. This and all subsequent DNA and RNA alignments were carried out in BowTie 2’s – very-sensitive mode, and this and all output alignment files were binary-compressed and sorted by chromosome position using SamTools version 1.12.0 (Li et al. 2009). MetaBAT adjusted coverage for contigs representing Hodgkinia and Sulcia was obtained using the jgi_summarize_bam_contig_depths function in MetaBAT2 (Kang et al. 2019). Using contig lengths and adjusted coverage values reads representing Hodgkinia and Sulcia coverage were calculated and converted to per- centages of total processed reads for each library. To determine the relative abundance of each Hodgkinia genome in cicadas hosting multiple Hodgkinia lineages, trimmed DNA reads from each sample were aligned to the corresponding Hodgkinia genomes, and the coverage proportions were calculated exactly as in the comparison of Hodgkinia and Sulcia coverage, this time giving the pro- portion of Hodgkinia DNA coverage contributed by each Hodgkinia genome hosted by a given cicada. RNA Coverage Analyses and rRNA Sequence Removal For visualization of strand-specific RNA-seq coverage of the Hodgkinia and Sulcia genomes, reads from the metatran- scriptome sequencing libraries were quality checked and trimmed according to the same parameters as the DNA reads (see section “DNA Coverage and Genome Abundance Analyses”). Reads from each library were aligned separately to the appropriate Hodgkinia and Sulcia genomes. The RNA alignments were then separated according to the DNA strand from which the alignments originated. Briefly, reads which were the second in a pair and which aligned to the forward strand were written to a separate file (samtools view -f 128 -F 16). This was repeated for reads which were the first in a pair and aligned to the reverse strand (samtools view -f 80). (samtools These alignment files were combined merge), collectively representing RNA coverage of the minus corresponding genome(s). Alignments representing RNA coverage of the plus strands of these genomes were separated with a similar set of com- mands (samtools view -f 144, samtools view -f 64 -F 16, samtools merge). strand of the From these alignment files, per-base coverage values were obtained for each strand using the genomecov -d command in BEDTools v.2.24.0 (Quinlan and Hall 2010). Stranded per-base coverage values, along with annotation information extracted from the GFF annotation files using custom Python scripts were used to generate coverage plots with processing v.3.5.4 in Python mode. Custom Python and processing scripts can be accessed from the fol- lowing GitHub repository: https://github.com/noah- spencer/Supplement-for-Spencer2023. For final coverage Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 13 Spencer et al. GBE plots shown in fig. 3 and supplementary figure S1, Supplementary Material online, these steps were repeated using alignment files randomly downsampled with SAMTools to achieve approximately 450X and 3500X coverage of Hodgkinia and Sulcia genomes of interest, respectively. Since substantial rRNA sequence coverage was detected in some libraries, the processed reads were subject to bio- informatic rRNA depletion before determining transcript counts. Briefly, trimmed reads were mapped to all of the corresponding endosymbiont rRNA sequences. Mapped reads were removed from the resulting alignment files using SamTools (samtools view -f 4). Unmapped reads were converted back to paired-end FASTQ files using the SamToFastq function in Picard Toolkit v.2.23.7 (2020). Transcript Abundance and Antisense Transcription Analyses The rRNA-depleted reads were aligned to the correspond- ing Hodgkinia or Sulcia genome(s). Transcript counts for Hodgkinia and Sulcia genes were obtained using FADU, a drop-in replacement for transcript quantification tools like htseq-count that uses partial counts and expectation maxi- mization algorithms to more accurately assign reads de- rived from polycistronic transcripts (as produced by operons) and gene-dense coding regions (Chung et al. 2021). FADU was run in -s “reverse” mode to accurate- ly quantify these stranded RNA-seq data. A second run in -s “yes” mode was performed to quantify transcription on the opposite strand relative to annotated open reading frames (i.e., to quantify antisense transcription). Percent antisense transcription for each biological replicate was es- timated as the total number of counts output by FADU is -s “yes” mode divided by the summed counts from both runs. The percentages reported represent averages across all biological replicates included for a given cicada species. Counts output by FADU for putatively functional genes (excluding tRNA and rRNA genes) were converted to TPM (Wagner et al. 2012) using a custom Python script by Arkadiy (https://github.com/Arkadiy-Garber/ BagOfTricks/blob/main/count-to-tpm.py). Statistical ana- lysis of these TPM expression data was performed in R v.4.0.4. Semipartial correlation analysis of TPM expression data, relative gene dosage, and gene copy DNA abundance was performed using the R package ppcor v.1.1 (Kim 2015). Garber Genome Sequence-Based Analyses Sequences spanning 50 bp upstream of start codons in (A) all protein-coding genes, (B) the 15 protein-coding genes with the highest average TPM, and (C) the 15 protein- coding genes with the lowest average TPM were extracted from the Hodgkinia and Sulcia genomes from T. ulnaria and used to make six logo plots with WebLogo (Crooks et al. 2004). the from free-living Protein alignments of all copies of RpoD represented in our data, as well as RpoD from Hodgkinia in D. semicincta and alphaproteobacterium Methylobacterium oxalidis (retrieved from NCBI, protein ac- cessions ACT34206 and GEP04622.1, respectively) were performed using MUSCLE algorithm (Edgar 2004) imple- mented through the M-Coffee web server (Moretti et al. 2007) and then visualized using NCBI’s Multiple Sequence Alignment Viewer v.1.2.0. Supplementary Material Supplementary data are available at Genome Biology and Evolution online (http://www.gbe.oxfordjournals.org/). Acknowledgments We thank DeAnna Bublitz and Katherine Nazario for their help with specimen collection and Arkadiy Garber for bio- informatics and programming assistance. We also thank the two anonymous reviewers for their helpful feedback. This work was supported by the National Science Foundation (IOS-1553529 to J.P.M. and 026257-001 to N.J.S.); the National Geographic Society (9760-15 to P.Ł.); and Foundation (GBMF5602 to J.P.M.). the Gordon and Betty Moore Data Availability All data described here are available from the NCBI Umbrella BioProject PRJNA386376. All cicada bacteriome metatranscritpome and metagenome sequencing libraries were deposited in the Sequence Read Archive (SRA) data- base under BioProject PRJNA923375. Newly generated genome assemblies for endosymbionts of D. near semicinc- ta, T. ulnaria, T. limbata, and O. oregona, as well as the cor- responding SRA experiments (containing the raw reads), PRJNA923375, are available PRJNA512238, PRJNA385844, respectively. under PRJNA246493, BioProjects and Literature Cited Andersson SG, Kurland CG. 1998. Reductive evolution of resident gen- omes. Trends Microbiol. 6:263–268. Bennett GM, Chong RA. 2017. Genome-Wide transcriptional dynam- ics in the companion bacterial symbionts of the glassy-winged sharpshooter (Cicadellidae: Homalodisca vitripennis) reveal differ- ential gene expression in Bacteria occupying multiple host organs. G3 (Bethesda). 7:3073–3082. Bennett GM, Moran NA. 2015. Heritable symbiosis: the advantages and perils of an evolutionary rabbit hole. Proc Natl Acad Sci USA. 112:10169–10176. Bolger AM, Lohse M, Usadel B. 2014. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30(15):2114–2120. 14 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 No Transcriptional Compensation for Extreme Gene Dosage Imbalance in Fragmented Bacterial Endosymbionts of CicadasGBE Boore JL. 1999. Animal mitochondrial genomes. Nucleic Acids Res. 27: 1767–1780. Bushnell B. 2014. BBMap: A Fast, Accurate, Splice-Aware Aligner. Lawrence Berkeley National Laboratory. LBNL Report #: LBNL-7065E. endosymbiotic system of mealybugs. Appl Environ Microbiol. 74(13):4175–4184. Langmead B, Salzberg SL. 2012. Fast gapped-read alignment with bowtie 2. Nat Methods 9:357–359. Li H, et al. 2009. The sequence alignment/map format and SAMtools. Bushnell B, Rood J, Singer E. 2017. BBMerge—accurate paired shot- Bioinformatics 25:2078–2079. gun read merging via overlap. PLoS One. 12:e0185056. Campbell MA, et al. 2015. Genome expansion via lineage splitting and genome reduction in the cicada endosymbiont hodgkinia. Proc Natl Acad Sci U S A. 112:10192–10199. Campbell MA, et al. 2018. Changes in endosymbiont complexity drive host-level compensatory adaptations in cicadas. MBio 9(6): e02104-e2118. Campbell MA, Łukasik P, Simon C, McCutcheon JP. 2017. Idiosyncratic genome degradation in a bacterial endosymbiont of periodical ci- cadas. Curr Biol. 27:3568–3575.e3. Chan PP, Lowe TM. 2019. tRNAscan-SE: searching for tRNA genes in genomic sequences. Methods Mol Biol. 1962:1–14. Charles H, Heddi A, Guillaud J, Nardon C, Nardon P. 1997. A molecular aspect of symbiotic interactions between the weevil Sitophilus or- yzae and its endosymbiotic bacteria: over-expression of a chaper- onin. Biochem Biophys Res Commun. 239:769–774. Chung M, et al. 2021. FADU: a quantification tool for prokaryotic tran- scriptomic analyses. mSystems 6:e00917–20. Crooks GE, Hon G, Chandonia J, Brenner SE. 2004. Weblogo: a se- quence logo generator. Genome Res. 14:1188–1190. Deng J, et al. 2022. Genome comparison reveals inversions and alter- native evolutionary history of nutritional endosymbionts in planthoppers (Hemiptera: Fulgoromorpha). bioRxiv. https://doi. org/10.1101/2022.12.07.519479 Edgar RC. 2004. MUSCLE: a multiple sequence alignment method with reduced time and space complexity. BMC Bioinformatics 5: 113. Fares MA, Ruiz-González MX, Moya A, Elena SF, Barrio E. 2002. Endosymbiotic bacteria: groEL buffers against deleterious muta- tions. Nature 417:398. Forsythe ES, et al. 2022. Organellar transcripts dominate the cellular mRNA pool across plants of varying ploidy levels. Proc Natl Acad Sci U S A. 119(30):e2204187119. Galán-Vásquez E, Sánchez-Osorio I, Martínez-Antonio A. 2016. Transcription factors exhibit differential conservation in Bacteria with reduced genomes. PLoS One. 11:e0146901. García-Alcalde F, et al. 2012. Qualimap: evaluating next-generation sequencing alignment data. Bioinformatics 28:2678–2679. Li B, Dewey CN. 2011. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics 12:323. Luck AN, et al. 2015. Tissue-specific transcriptomics and proteomics of a filarial nematode and its Wolbachia endosymbiont. BMC Genomics 16:920. Łukasik P, et al. 2018. Multiple origins of interdependent endosymbi- otic complexes in a genus of cicadas. Proc Natl Acad Sci U S A. 115: E226–E235. Marshall DC, et al. 2018. A molecular phylogeny of the cicadas (Hemiptera: cicadidae) with a review of tribe and subfamily classi- fication. Zootaxa 4424:1–64. Martin M. 2011. Cutadapt removes adapter sequences from high- throughput sequencing reads. EMBnet J 17:10–12. Matsuura Y, et al. 2018. Recurrent symbiont recruitment from fungal parasites in cicadas. Proc Natl Acad Sci U S A. 115:E5970–E5979. McCutcheon JP, McDonald BR, Moran NA. 2009a. Convergent evolu- tion of metabolic roles in bacterial co-symbionts of insects. Proc Natl Acad Sci U S A. 106:15394–15399. McCutcheon JP, McDonald BR, Moran NA. 2009b. Origin of an alter- native genetic code in the extremely small and GC-rich genome of a bacterial symbiont. PLoS Genet. 5:e1000565. McCutcheon JP, Moran NA. 2011. Extreme genome reduction in sym- biotic bacteria. Nat Rev Microbiol. 10:13–26. Medina Munoz M, Pollio AR, White HL, Rio RVM. 2017. Into the wild: parallel transcriptomics of the tsetse-Wigglesworthia mutualism within Kenyan populations. Genome Biol Evol. 9:2276–2291. Moran NA, Dunbar HE, Wilcox JL. 2005a. Regulation of transcription in a reduced bacterial genome: nutrient-provisioning genes of the obligate symbiont Buchnera aphidicola. J Bacteriol. 187(12): 4229–4237. Moran NA, Tran P, Gerardo NM. 2005b. Symbiosis and insect diversi- fication: an ancient symbiont of sap-feeding insects from the bac- terial phylum Bacteroidetes. Appl Environ Microbiol. 71: 8802–8810. Moretti S, et al. 2007. The M-coffee web server: a meta-method for computing multiple sequence alignments by combining alternative alignment methods. Nucleic Acids Res. 35:W645–W648. Graf JS, et al. 2021. Anaerobic endosymbiont generates energy for cili- Morril SA, Amon A. 2019. Why haploinsufficiency persists. Proc Natl ate host by denitrification. Nature 591:445–450. Acad Sci U S A. 116(24):11866–11871. Gray MW. 2012. Mitochondrial evolution. Cold Spring Harb Perspect Biol. 4:a011403. Green BR. 2011. Chloroplast genomes of photosynthetic eukaryotes. Plant J. 66:34–44. Grossman AD, Straus DB, Walter WA, Gross CA. 1987. Sigma 32 syn- thesis can regulate the synthesis of heat shock proteins in Escherichia coli. Genes Dev. 1:179–184. Husnik F, Vaclav H, Darby A. 2020. Symbiont gene expression in the midgut bacteriocytes of a blood-sucking parasite. Genome Biol Evol. 12(4):429–442. Kang DD, et al. 2019. MetaBAT 2: an adaptive binning algorithm for robust and efficient genome reconstruction from metagenome as- semblies. PeerJ 7:e7359. Kim S. 2015. . Ppcor: an R package for a fast calculation to semi- partial correlation coefficients. Commun Stat Appl Methods 22: 665–674. Kono M, Koga R, Shimada M, Fukatsu T. 2008. Infection dynamics of in the nested coexisting Beta- and Gammaproteobacteria Neidhardt FC, VanBogelen RA. 1981. Positive regulatory gene for temperature-controlled proteins in Escherichia coli. Biochem Biophys Res Commun. 100:894–900. Papp B, Pál C, Hurst L. 2003. Dosage sensitivity and the evolution of gene families in yeast. Nature 424:194–197. Picard Toolkit. 2020. Broad Institute, GitHub repository. https:// broadinstitute.github.io/picard/ Poliakov A, et al. 2011. Large-scale label-free quantitative proteomics of the pea aphid-Buchnera symbiosis. Mol Cell Proteomics 10: M110.007039. Prjibelski A, Antipov D, Meleshko D, Lapidus A, Korobeynikov A. 2020. Using SPAdes De Novo assembler. Curr Protoc Bioinformatics 70: e102. Quinlan AR, Hall IM. 2010. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26:841–842. Rangel-Chávez CP, Galán-Vásquez E, Pescador-Tapia A, Delaye L, Martínez-Antonio A. 2021. RNA Polymerases in strict endosymbi- ont bacteria with extreme genome reduction show distinct Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023 15 Spencer et al. GBE erosions that might result in limited and differential promoter rec- ognition. PLoS One 16:e0239350. Richardson JP. 2002. Rho-dependent termination and ATPases in tran- script termination. Biochem Biophys Acta 1577:251–260. Schmidt MC, Chamberlin MJ. 1984. Binding of rho factor to Escherichia coli RNA polymerase mediated by nusA protein. J Biol Chem. 259:15000–15002. Shao R, Zhu X-Q, Barker SC, Herd K. 2012. Evolution of extensively fragmented mitochondrial genomes in the lice of humans. Genome Biol Evol. 4:1088–1101. Simonet P, et al. 2018. Bacteriocyte cell death in the pea aphid/Buchnera symbiotic system. Proc Natl Acad Sci U S A. 115(8):E1819–E1828. Sloan DB, et al. 2012. Rapid evolution of enormous, multichromoso- mal genomes in flowering plant mitochondria with exceptionally high mutation rates. PLoS Biol. 10:e1001241. Srinivasan KA, Virdee SK, McArthur AG. 2020. Strandedness during cDNA synthesis, the stranded parameter in htseq-count and ana- lysis of RNA-Seq data. Brief Funct Genomics 19:339–342. Stoll S, Feldhaar H, Gross R. 2009. Transcriptional profiling of the endosym- biont Blochmannia floridanus during different developmental stages of its holometabolous ant host. Environ Microbiol. 11:877–888. Tamas I, et al. 2002. 50 Million years of genomic stasis in endosymbi- otic bacteria. Science 296:2376–2379. Tanifuji G, Onodera NT, Moore CE, Archibald JM. 2014. Reduced nu- clear genomes maintain high gene transcription levels. Mol Biol Evol. 31(3):625–635. Van Leuven JT, Mao M, Xing DD, Bennett GM, McCutcheon JP. 2019. Cicada endosymbionts have tRNAs that are correctly processed despite having genomes that do not encode all of the tRNA pro- cessing machinery. MBio 10:e01950-18. Van Leuven JT, McCutcheon JP. 2012. An AT Mutational Bias in the Tiny GC-Rich Endosymbiont Genome of Hodgkinia. Genome Biol Evol. 4(1):24–27. Van Leuven JT, Meister RC, Simon C, McCutcheon JP. 2014. Sympatric speciation in a bacterial endosymbiont results in two genomes with the functionality of one. Cell 158:1270–1280. Veita RA, Bottani S, Birchler JA. 2013. Gene dosage effects: nonlinea- rities, genetic interactions, and dosage compensation. Trends Genet. 29(7):385–393. Vigneron A, et al. 2014. Insects recycle endosymbionts when the bene- fit is over. Curr Biol. 24(19):2267–2273. Wagner GP, Kin K, Lynch VJ. 2012. Measurement of mRNA abundance using RNA-Seq data: rPKM measure is inconsistent among sam- ples. Theory Biosci. 131:281–285. Wang D, et al. 2022b. Complex co-evolutionary relationships be- tween cicadas and their symbionts. Environ Microbiol. 24: 195–211. Wang D, Hong G, Wei C. 2022a. Cellular and potential molecular me- chanisms underlying transovarial transmission of the obligate sym- biont sulcia in cicadas. Environ Microbiol. 25:836–852. Wilcox JL, Dunbar HE, Wolfinger RD, Moran NA. 2003. Consequences of reductive evolution for gene expression in an obligate endosym- biont. Mol Microbiol. 48:1491–1500. Yamamori T, Yura T. 1982. Genetic control of heat-shock protein syn- thesis and its bearing on growth and thermal resistance in Escherichia coli K-12. Proc Natl Acad Sci U S A. 79:860–864. Zhang J, Kobert K, Flouri T, Stamatakis A. 2014. PEAR: a fast and ac- curate illumina paired-End reAd mergeR. Bioinformatics 30: 614–620. Associate editor: Dr. Howard Ochman 16 Genome Biol. Evol. 15(6) https://doi.org/10.1093/gbe/evad100 Advance Access publication 2 June 2023

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.